linux/fs/nfs/super.c

1940 lines
46 KiB
C
Raw Normal View History

/*
* linux/fs/nfs/super.c
*
* Copyright (C) 1992 Rick Sladkey
*
* nfs superblock handling functions
*
* Modularised by Alan Cox <Alan.Cox@linux.org>, while hacking some
* experimental NFS changes. Modularisation taken straight from SYS5 fs.
*
* Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
* J.S.Peatfield@damtp.cam.ac.uk
*
* Split from inode.c by David Howells <dhowells@redhat.com>
*
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
* - superblocks are indexed on server only - all inodes, dentries, etc. associated with a
* particular server are held in the same superblock
* - NFS superblocks can have several effective roots to the dentry tree
* - directory type roots are spliced into the tree when a path from one root reaches the root
* of another (see nfs_lookup())
*/
#include <linux/module.h>
#include <linux/init.h>
#include <linux/time.h>
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/string.h>
#include <linux/stat.h>
#include <linux/errno.h>
#include <linux/unistd.h>
#include <linux/sunrpc/clnt.h>
#include <linux/sunrpc/stats.h>
#include <linux/sunrpc/metrics.h>
#include <linux/nfs_fs.h>
#include <linux/nfs_mount.h>
#include <linux/nfs4_mount.h>
#include <linux/lockd/bind.h>
#include <linux/smp_lock.h>
#include <linux/seq_file.h>
#include <linux/mount.h>
#include <linux/nfs_idmap.h>
#include <linux/vfs.h>
#include <linux/inet.h>
#include <linux/nfs_xdr.h>
#include <linux/magic.h>
#include <linux/parser.h>
#include <asm/system.h>
#include <asm/uaccess.h>
#include "nfs4_fs.h"
#include "callback.h"
#include "delegation.h"
#include "iostat.h"
#include "internal.h"
#define NFSDBG_FACILITY NFSDBG_VFS
struct nfs_parsed_mount_data {
int flags;
int rsize, wsize;
int timeo, retrans;
int acregmin, acregmax,
acdirmin, acdirmax;
int namlen;
unsigned int bsize;
unsigned int auth_flavor_len;
rpc_authflavor_t auth_flavors[1];
char *client_address;
struct {
struct sockaddr_in address;
unsigned int program;
unsigned int version;
unsigned short port;
int protocol;
} mount_server;
struct {
struct sockaddr_in address;
char *hostname;
char *export_path;
unsigned int program;
int protocol;
} nfs_server;
};
enum {
/* Mount options that take no arguments */
Opt_soft, Opt_hard,
Opt_intr, Opt_nointr,
Opt_posix, Opt_noposix,
Opt_cto, Opt_nocto,
Opt_ac, Opt_noac,
Opt_lock, Opt_nolock,
Opt_v2, Opt_v3,
Opt_udp, Opt_tcp,
Opt_acl, Opt_noacl,
Opt_rdirplus, Opt_nordirplus,
Opt_sharecache, Opt_nosharecache,
/* Mount options that take integer arguments */
Opt_port,
Opt_rsize, Opt_wsize, Opt_bsize,
Opt_timeo, Opt_retrans,
Opt_acregmin, Opt_acregmax,
Opt_acdirmin, Opt_acdirmax,
Opt_actimeo,
Opt_namelen,
Opt_mountport,
Opt_mountprog, Opt_mountvers,
Opt_nfsprog, Opt_nfsvers,
/* Mount options that take string arguments */
Opt_sec, Opt_proto, Opt_mountproto,
Opt_addr, Opt_mounthost, Opt_clientaddr,
/* Mount options that are ignored */
Opt_userspace, Opt_deprecated,
Opt_err
};
static match_table_t nfs_mount_option_tokens = {
{ Opt_userspace, "bg" },
{ Opt_userspace, "fg" },
{ Opt_soft, "soft" },
{ Opt_hard, "hard" },
{ Opt_intr, "intr" },
{ Opt_nointr, "nointr" },
{ Opt_posix, "posix" },
{ Opt_noposix, "noposix" },
{ Opt_cto, "cto" },
{ Opt_nocto, "nocto" },
{ Opt_ac, "ac" },
{ Opt_noac, "noac" },
{ Opt_lock, "lock" },
{ Opt_nolock, "nolock" },
{ Opt_v2, "v2" },
{ Opt_v3, "v3" },
{ Opt_udp, "udp" },
{ Opt_tcp, "tcp" },
{ Opt_acl, "acl" },
{ Opt_noacl, "noacl" },
{ Opt_rdirplus, "rdirplus" },
{ Opt_nordirplus, "nordirplus" },
{ Opt_sharecache, "sharecache" },
{ Opt_nosharecache, "nosharecache" },
{ Opt_port, "port=%u" },
{ Opt_rsize, "rsize=%u" },
{ Opt_wsize, "wsize=%u" },
{ Opt_bsize, "bsize=%u" },
{ Opt_timeo, "timeo=%u" },
{ Opt_retrans, "retrans=%u" },
{ Opt_acregmin, "acregmin=%u" },
{ Opt_acregmax, "acregmax=%u" },
{ Opt_acdirmin, "acdirmin=%u" },
{ Opt_acdirmax, "acdirmax=%u" },
{ Opt_actimeo, "actimeo=%u" },
{ Opt_userspace, "retry=%u" },
{ Opt_namelen, "namlen=%u" },
{ Opt_mountport, "mountport=%u" },
{ Opt_mountprog, "mountprog=%u" },
{ Opt_mountvers, "mountvers=%u" },
{ Opt_nfsprog, "nfsprog=%u" },
{ Opt_nfsvers, "nfsvers=%u" },
{ Opt_nfsvers, "vers=%u" },
{ Opt_sec, "sec=%s" },
{ Opt_proto, "proto=%s" },
{ Opt_mountproto, "mountproto=%s" },
{ Opt_addr, "addr=%s" },
{ Opt_clientaddr, "clientaddr=%s" },
{ Opt_mounthost, "mounthost=%s" },
{ Opt_err, NULL }
};
enum {
Opt_xprt_udp, Opt_xprt_tcp,
Opt_xprt_err
};
static match_table_t nfs_xprt_protocol_tokens = {
{ Opt_xprt_udp, "udp" },
{ Opt_xprt_tcp, "tcp" },
{ Opt_xprt_err, NULL }
};
enum {
Opt_sec_none, Opt_sec_sys,
Opt_sec_krb5, Opt_sec_krb5i, Opt_sec_krb5p,
Opt_sec_lkey, Opt_sec_lkeyi, Opt_sec_lkeyp,
Opt_sec_spkm, Opt_sec_spkmi, Opt_sec_spkmp,
Opt_sec_err
};
static match_table_t nfs_secflavor_tokens = {
{ Opt_sec_none, "none" },
{ Opt_sec_none, "null" },
{ Opt_sec_sys, "sys" },
{ Opt_sec_krb5, "krb5" },
{ Opt_sec_krb5i, "krb5i" },
{ Opt_sec_krb5p, "krb5p" },
{ Opt_sec_lkey, "lkey" },
{ Opt_sec_lkeyi, "lkeyi" },
{ Opt_sec_lkeyp, "lkeyp" },
{ Opt_sec_err, NULL }
};
static void nfs_umount_begin(struct vfsmount *, int);
static int nfs_statfs(struct dentry *, struct kstatfs *);
static int nfs_show_options(struct seq_file *, struct vfsmount *);
static int nfs_show_stats(struct seq_file *, struct vfsmount *);
static int nfs_get_sb(struct file_system_type *, int, const char *, void *, struct vfsmount *);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
static int nfs_xdev_get_sb(struct file_system_type *fs_type,
int flags, const char *dev_name, void *raw_data, struct vfsmount *mnt);
static void nfs_kill_super(struct super_block *);
static struct file_system_type nfs_fs_type = {
.owner = THIS_MODULE,
.name = "nfs",
.get_sb = nfs_get_sb,
.kill_sb = nfs_kill_super,
.fs_flags = FS_RENAME_DOES_D_MOVE|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
};
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct file_system_type nfs_xdev_fs_type = {
.owner = THIS_MODULE,
.name = "nfs",
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
.get_sb = nfs_xdev_get_sb,
.kill_sb = nfs_kill_super,
.fs_flags = FS_RENAME_DOES_D_MOVE|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
};
static const struct super_operations nfs_sops = {
.alloc_inode = nfs_alloc_inode,
.destroy_inode = nfs_destroy_inode,
.write_inode = nfs_write_inode,
.statfs = nfs_statfs,
.clear_inode = nfs_clear_inode,
.umount_begin = nfs_umount_begin,
.show_options = nfs_show_options,
.show_stats = nfs_show_stats,
};
#ifdef CONFIG_NFS_V4
static int nfs4_get_sb(struct file_system_type *fs_type,
int flags, const char *dev_name, void *raw_data, struct vfsmount *mnt);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
static int nfs4_xdev_get_sb(struct file_system_type *fs_type,
int flags, const char *dev_name, void *raw_data, struct vfsmount *mnt);
static int nfs4_referral_get_sb(struct file_system_type *fs_type,
int flags, const char *dev_name, void *raw_data, struct vfsmount *mnt);
static void nfs4_kill_super(struct super_block *sb);
static struct file_system_type nfs4_fs_type = {
.owner = THIS_MODULE,
.name = "nfs4",
.get_sb = nfs4_get_sb,
.kill_sb = nfs4_kill_super,
.fs_flags = FS_RENAME_DOES_D_MOVE|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
};
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct file_system_type nfs4_xdev_fs_type = {
.owner = THIS_MODULE,
.name = "nfs4",
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
.get_sb = nfs4_xdev_get_sb,
.kill_sb = nfs4_kill_super,
.fs_flags = FS_RENAME_DOES_D_MOVE|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
};
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct file_system_type nfs4_referral_fs_type = {
.owner = THIS_MODULE,
.name = "nfs4",
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
.get_sb = nfs4_referral_get_sb,
.kill_sb = nfs4_kill_super,
.fs_flags = FS_RENAME_DOES_D_MOVE|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
};
static const struct super_operations nfs4_sops = {
.alloc_inode = nfs_alloc_inode,
.destroy_inode = nfs_destroy_inode,
.write_inode = nfs_write_inode,
.statfs = nfs_statfs,
.clear_inode = nfs4_clear_inode,
.umount_begin = nfs_umount_begin,
.show_options = nfs_show_options,
.show_stats = nfs_show_stats,
};
#endif
static struct shrinker *acl_shrinker;
/*
* Register the NFS filesystems
*/
int __init register_nfs_fs(void)
{
int ret;
ret = register_filesystem(&nfs_fs_type);
if (ret < 0)
goto error_0;
ret = nfs_register_sysctl();
if (ret < 0)
goto error_1;
#ifdef CONFIG_NFS_V4
ret = register_filesystem(&nfs4_fs_type);
if (ret < 0)
goto error_2;
#endif
acl_shrinker = set_shrinker(DEFAULT_SEEKS, nfs_access_cache_shrinker);
return 0;
#ifdef CONFIG_NFS_V4
error_2:
nfs_unregister_sysctl();
#endif
error_1:
unregister_filesystem(&nfs_fs_type);
error_0:
return ret;
}
/*
* Unregister the NFS filesystems
*/
void __exit unregister_nfs_fs(void)
{
if (acl_shrinker != NULL)
remove_shrinker(acl_shrinker);
#ifdef CONFIG_NFS_V4
unregister_filesystem(&nfs4_fs_type);
nfs_unregister_sysctl();
#endif
unregister_filesystem(&nfs_fs_type);
}
/*
* Deliver file system statistics to userspace
*/
static int nfs_statfs(struct dentry *dentry, struct kstatfs *buf)
{
struct nfs_server *server = NFS_SB(dentry->d_sb);
unsigned char blockbits;
unsigned long blockres;
struct nfs_fh *fh = NFS_FH(dentry->d_inode);
struct nfs_fattr fattr;
struct nfs_fsstat res = {
.fattr = &fattr,
};
int error;
lock_kernel();
error = server->nfs_client->rpc_ops->statfs(server, fh, &res);
if (error < 0)
goto out_err;
buf->f_type = NFS_SUPER_MAGIC;
/*
* Current versions of glibc do not correctly handle the
* case where f_frsize != f_bsize. Eventually we want to
* report the value of wtmult in this field.
*/
buf->f_frsize = dentry->d_sb->s_blocksize;
/*
* On most *nix systems, f_blocks, f_bfree, and f_bavail
* are reported in units of f_frsize. Linux hasn't had
* an f_frsize field in its statfs struct until recently,
* thus historically Linux's sys_statfs reports these
* fields in units of f_bsize.
*/
buf->f_bsize = dentry->d_sb->s_blocksize;
blockbits = dentry->d_sb->s_blocksize_bits;
blockres = (1 << blockbits) - 1;
buf->f_blocks = (res.tbytes + blockres) >> blockbits;
buf->f_bfree = (res.fbytes + blockres) >> blockbits;
buf->f_bavail = (res.abytes + blockres) >> blockbits;
buf->f_files = res.tfiles;
buf->f_ffree = res.afiles;
buf->f_namelen = server->namelen;
unlock_kernel();
return 0;
out_err:
dprintk("%s: statfs error = %d\n", __FUNCTION__, -error);
unlock_kernel();
return error;
}
/*
* Map the security flavour number to a name
*/
static const char *nfs_pseudoflavour_to_name(rpc_authflavor_t flavour)
{
static const struct {
rpc_authflavor_t flavour;
const char *str;
} sec_flavours[] = {
{ RPC_AUTH_NULL, "null" },
{ RPC_AUTH_UNIX, "sys" },
{ RPC_AUTH_GSS_KRB5, "krb5" },
{ RPC_AUTH_GSS_KRB5I, "krb5i" },
{ RPC_AUTH_GSS_KRB5P, "krb5p" },
{ RPC_AUTH_GSS_LKEY, "lkey" },
{ RPC_AUTH_GSS_LKEYI, "lkeyi" },
{ RPC_AUTH_GSS_LKEYP, "lkeyp" },
{ RPC_AUTH_GSS_SPKM, "spkm" },
{ RPC_AUTH_GSS_SPKMI, "spkmi" },
{ RPC_AUTH_GSS_SPKMP, "spkmp" },
{ UINT_MAX, "unknown" }
};
int i;
for (i = 0; sec_flavours[i].flavour != UINT_MAX; i++) {
if (sec_flavours[i].flavour == flavour)
break;
}
return sec_flavours[i].str;
}
/*
* Describe the mount options in force on this server representation
*/
static void nfs_show_mount_options(struct seq_file *m, struct nfs_server *nfss, int showdefaults)
{
static const struct proc_nfs_info {
int flag;
const char *str;
const char *nostr;
} nfs_info[] = {
{ NFS_MOUNT_SOFT, ",soft", ",hard" },
{ NFS_MOUNT_INTR, ",intr", "" },
{ NFS_MOUNT_NOCTO, ",nocto", "" },
{ NFS_MOUNT_NOAC, ",noac", "" },
{ NFS_MOUNT_NONLM, ",nolock", "" },
{ NFS_MOUNT_NOACL, ",noacl", "" },
{ NFS_MOUNT_NORDIRPLUS, ",nordirplus", "" },
{ NFS_MOUNT_UNSHARED, ",nosharecache", ""},
{ 0, NULL, NULL }
};
const struct proc_nfs_info *nfs_infop;
struct nfs_client *clp = nfss->nfs_client;
char buf[12];
const char *proto;
seq_printf(m, ",vers=%d", clp->rpc_ops->version);
seq_printf(m, ",rsize=%d", nfss->rsize);
seq_printf(m, ",wsize=%d", nfss->wsize);
if (nfss->acregmin != 3*HZ || showdefaults)
seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
if (nfss->acregmax != 60*HZ || showdefaults)
seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
if (nfss->acdirmin != 30*HZ || showdefaults)
seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
if (nfss->acdirmax != 60*HZ || showdefaults)
seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
if (nfss->flags & nfs_infop->flag)
seq_puts(m, nfs_infop->str);
else
seq_puts(m, nfs_infop->nostr);
}
switch (nfss->client->cl_xprt->prot) {
case IPPROTO_TCP:
proto = "tcp";
break;
case IPPROTO_UDP:
proto = "udp";
break;
default:
snprintf(buf, sizeof(buf), "%u", nfss->client->cl_xprt->prot);
proto = buf;
}
seq_printf(m, ",proto=%s", proto);
seq_printf(m, ",timeo=%lu", 10U * clp->retrans_timeo / HZ);
seq_printf(m, ",retrans=%u", clp->retrans_count);
seq_printf(m, ",sec=%s", nfs_pseudoflavour_to_name(nfss->client->cl_auth->au_flavor));
}
/*
* Describe the mount options on this VFS mountpoint
*/
static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
{
struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
nfs_show_mount_options(m, nfss, 0);
seq_puts(m, ",addr=");
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
seq_escape(m, nfss->nfs_client->cl_hostname, " \t\n\\");
return 0;
}
/*
* Present statistical information for this VFS mountpoint
*/
static int nfs_show_stats(struct seq_file *m, struct vfsmount *mnt)
{
int i, cpu;
struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
struct rpc_auth *auth = nfss->client->cl_auth;
struct nfs_iostats totals = { };
seq_printf(m, "statvers=%s", NFS_IOSTAT_VERS);
/*
* Display all mount option settings
*/
seq_printf(m, "\n\topts:\t");
seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? "ro" : "rw");
seq_puts(m, mnt->mnt_sb->s_flags & MS_SYNCHRONOUS ? ",sync" : "");
seq_puts(m, mnt->mnt_sb->s_flags & MS_NOATIME ? ",noatime" : "");
seq_puts(m, mnt->mnt_sb->s_flags & MS_NODIRATIME ? ",nodiratime" : "");
nfs_show_mount_options(m, nfss, 1);
seq_printf(m, "\n\tage:\t%lu", (jiffies - nfss->mount_time) / HZ);
seq_printf(m, "\n\tcaps:\t");
seq_printf(m, "caps=0x%x", nfss->caps);
seq_printf(m, ",wtmult=%d", nfss->wtmult);
seq_printf(m, ",dtsize=%d", nfss->dtsize);
seq_printf(m, ",bsize=%d", nfss->bsize);
seq_printf(m, ",namelen=%d", nfss->namelen);
#ifdef CONFIG_NFS_V4
if (nfss->nfs_client->cl_nfsversion == 4) {
seq_printf(m, "\n\tnfsv4:\t");
seq_printf(m, "bm0=0x%x", nfss->attr_bitmask[0]);
seq_printf(m, ",bm1=0x%x", nfss->attr_bitmask[1]);
seq_printf(m, ",acl=0x%x", nfss->acl_bitmask);
}
#endif
/*
* Display security flavor in effect for this mount
*/
seq_printf(m, "\n\tsec:\tflavor=%d", auth->au_ops->au_flavor);
if (auth->au_flavor)
seq_printf(m, ",pseudoflavor=%d", auth->au_flavor);
/*
* Display superblock I/O counters
*/
for_each_possible_cpu(cpu) {
struct nfs_iostats *stats;
preempt_disable();
stats = per_cpu_ptr(nfss->io_stats, cpu);
for (i = 0; i < __NFSIOS_COUNTSMAX; i++)
totals.events[i] += stats->events[i];
for (i = 0; i < __NFSIOS_BYTESMAX; i++)
totals.bytes[i] += stats->bytes[i];
preempt_enable();
}
seq_printf(m, "\n\tevents:\t");
for (i = 0; i < __NFSIOS_COUNTSMAX; i++)
seq_printf(m, "%lu ", totals.events[i]);
seq_printf(m, "\n\tbytes:\t");
for (i = 0; i < __NFSIOS_BYTESMAX; i++)
seq_printf(m, "%Lu ", totals.bytes[i]);
seq_printf(m, "\n");
rpc_print_iostats(m, nfss->client);
return 0;
}
/*
* Begin unmount by attempting to remove all automounted mountpoints we added
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
* in response to xdev traversals and referrals
*/
static void nfs_umount_begin(struct vfsmount *vfsmnt, int flags)
{
struct nfs_server *server = NFS_SB(vfsmnt->mnt_sb);
struct rpc_clnt *rpc;
shrink_submounts(vfsmnt, &nfs_automount_list);
if (!(flags & MNT_FORCE))
return;
/* -EIO all pending I/O */
rpc = server->client_acl;
if (!IS_ERR(rpc))
rpc_killall_tasks(rpc);
rpc = server->client;
if (!IS_ERR(rpc))
rpc_killall_tasks(rpc);
}
/*
* Sanity-check a server address provided by the mount command
*/
static int nfs_verify_server_address(struct sockaddr *addr)
{
switch (addr->sa_family) {
case AF_INET: {
struct sockaddr_in *sa = (struct sockaddr_in *) addr;
if (sa->sin_addr.s_addr != INADDR_ANY)
return 1;
break;
}
}
return 0;
}
/*
* Error-check and convert a string of mount options from user space into
* a data structure
*/
static int nfs_parse_mount_options(char *raw,
struct nfs_parsed_mount_data *mnt)
{
char *p, *string;
if (!raw) {
dfprintk(MOUNT, "NFS: mount options string was NULL.\n");
return 1;
}
dfprintk(MOUNT, "NFS: nfs mount opts='%s'\n", raw);
while ((p = strsep(&raw, ",")) != NULL) {
substring_t args[MAX_OPT_ARGS];
int option, token;
if (!*p)
continue;
dfprintk(MOUNT, "NFS: parsing nfs mount option '%s'\n", p);
token = match_token(p, nfs_mount_option_tokens, args);
switch (token) {
case Opt_soft:
mnt->flags |= NFS_MOUNT_SOFT;
break;
case Opt_hard:
mnt->flags &= ~NFS_MOUNT_SOFT;
break;
case Opt_intr:
mnt->flags |= NFS_MOUNT_INTR;
break;
case Opt_nointr:
mnt->flags &= ~NFS_MOUNT_INTR;
break;
case Opt_posix:
mnt->flags |= NFS_MOUNT_POSIX;
break;
case Opt_noposix:
mnt->flags &= ~NFS_MOUNT_POSIX;
break;
case Opt_cto:
mnt->flags &= ~NFS_MOUNT_NOCTO;
break;
case Opt_nocto:
mnt->flags |= NFS_MOUNT_NOCTO;
break;
case Opt_ac:
mnt->flags &= ~NFS_MOUNT_NOAC;
break;
case Opt_noac:
mnt->flags |= NFS_MOUNT_NOAC;
break;
case Opt_lock:
mnt->flags &= ~NFS_MOUNT_NONLM;
break;
case Opt_nolock:
mnt->flags |= NFS_MOUNT_NONLM;
break;
case Opt_v2:
mnt->flags &= ~NFS_MOUNT_VER3;
break;
case Opt_v3:
mnt->flags |= NFS_MOUNT_VER3;
break;
case Opt_udp:
mnt->flags &= ~NFS_MOUNT_TCP;
mnt->nfs_server.protocol = IPPROTO_UDP;
mnt->timeo = 7;
mnt->retrans = 5;
break;
case Opt_tcp:
mnt->flags |= NFS_MOUNT_TCP;
mnt->nfs_server.protocol = IPPROTO_TCP;
mnt->timeo = 600;
mnt->retrans = 2;
break;
case Opt_acl:
mnt->flags &= ~NFS_MOUNT_NOACL;
break;
case Opt_noacl:
mnt->flags |= NFS_MOUNT_NOACL;
break;
case Opt_rdirplus:
mnt->flags &= ~NFS_MOUNT_NORDIRPLUS;
break;
case Opt_nordirplus:
mnt->flags |= NFS_MOUNT_NORDIRPLUS;
break;
case Opt_sharecache:
mnt->flags &= ~NFS_MOUNT_UNSHARED;
break;
case Opt_nosharecache:
mnt->flags |= NFS_MOUNT_UNSHARED;
break;
case Opt_port:
if (match_int(args, &option))
return 0;
if (option < 0 || option > 65535)
return 0;
mnt->nfs_server.address.sin_port = htonl(option);
break;
case Opt_rsize:
if (match_int(args, &mnt->rsize))
return 0;
break;
case Opt_wsize:
if (match_int(args, &mnt->wsize))
return 0;
break;
case Opt_bsize:
if (match_int(args, &option))
return 0;
if (option < 0)
return 0;
mnt->bsize = option;
break;
case Opt_timeo:
if (match_int(args, &mnt->timeo))
return 0;
break;
case Opt_retrans:
if (match_int(args, &mnt->retrans))
return 0;
break;
case Opt_acregmin:
if (match_int(args, &mnt->acregmin))
return 0;
break;
case Opt_acregmax:
if (match_int(args, &mnt->acregmax))
return 0;
break;
case Opt_acdirmin:
if (match_int(args, &mnt->acdirmin))
return 0;
break;
case Opt_acdirmax:
if (match_int(args, &mnt->acdirmax))
return 0;
break;
case Opt_actimeo:
if (match_int(args, &option))
return 0;
if (option < 0)
return 0;
mnt->acregmin =
mnt->acregmax =
mnt->acdirmin =
mnt->acdirmax = option;
break;
case Opt_namelen:
if (match_int(args, &mnt->namlen))
return 0;
break;
case Opt_mountport:
if (match_int(args, &option))
return 0;
if (option < 0 || option > 65535)
return 0;
mnt->mount_server.port = option;
break;
case Opt_mountprog:
if (match_int(args, &option))
return 0;
if (option < 0)
return 0;
mnt->mount_server.program = option;
break;
case Opt_mountvers:
if (match_int(args, &option))
return 0;
if (option < 0)
return 0;
mnt->mount_server.version = option;
break;
case Opt_nfsprog:
if (match_int(args, &option))
return 0;
if (option < 0)
return 0;
mnt->nfs_server.program = option;
break;
case Opt_nfsvers:
if (match_int(args, &option))
return 0;
switch (option) {
case 2:
mnt->flags &= ~NFS_MOUNT_VER3;
break;
case 3:
mnt->flags |= NFS_MOUNT_VER3;
break;
default:
goto out_unrec_vers;
}
break;
case Opt_sec:
string = match_strdup(args);
if (string == NULL)
goto out_nomem;
token = match_token(string, nfs_secflavor_tokens, args);
kfree(string);
/*
* The flags setting is for v2/v3. The flavor_len
* setting is for v4. v2/v3 also need to know the
* difference between NULL and UNIX.
*/
switch (token) {
case Opt_sec_none:
mnt->flags &= ~NFS_MOUNT_SECFLAVOUR;
mnt->auth_flavor_len = 0;
mnt->auth_flavors[0] = RPC_AUTH_NULL;
break;
case Opt_sec_sys:
mnt->flags &= ~NFS_MOUNT_SECFLAVOUR;
mnt->auth_flavor_len = 0;
mnt->auth_flavors[0] = RPC_AUTH_UNIX;
break;
case Opt_sec_krb5:
mnt->flags |= NFS_MOUNT_SECFLAVOUR;
mnt->auth_flavor_len = 1;
mnt->auth_flavors[0] = RPC_AUTH_GSS_KRB5;
break;
case Opt_sec_krb5i:
mnt->flags |= NFS_MOUNT_SECFLAVOUR;
mnt->auth_flavor_len = 1;
mnt->auth_flavors[0] = RPC_AUTH_GSS_KRB5I;
break;
case Opt_sec_krb5p:
mnt->flags |= NFS_MOUNT_SECFLAVOUR;
mnt->auth_flavor_len = 1;
mnt->auth_flavors[0] = RPC_AUTH_GSS_KRB5P;
break;
case Opt_sec_lkey:
mnt->flags |= NFS_MOUNT_SECFLAVOUR;
mnt->auth_flavor_len = 1;
mnt->auth_flavors[0] = RPC_AUTH_GSS_LKEY;
break;
case Opt_sec_lkeyi:
mnt->flags |= NFS_MOUNT_SECFLAVOUR;
mnt->auth_flavor_len = 1;
mnt->auth_flavors[0] = RPC_AUTH_GSS_LKEYI;
break;
case Opt_sec_lkeyp:
mnt->flags |= NFS_MOUNT_SECFLAVOUR;
mnt->auth_flavor_len = 1;
mnt->auth_flavors[0] = RPC_AUTH_GSS_LKEYP;
break;
case Opt_sec_spkm:
mnt->flags |= NFS_MOUNT_SECFLAVOUR;
mnt->auth_flavor_len = 1;
mnt->auth_flavors[0] = RPC_AUTH_GSS_SPKM;
break;
case Opt_sec_spkmi:
mnt->flags |= NFS_MOUNT_SECFLAVOUR;
mnt->auth_flavor_len = 1;
mnt->auth_flavors[0] = RPC_AUTH_GSS_SPKMI;
break;
case Opt_sec_spkmp:
mnt->flags |= NFS_MOUNT_SECFLAVOUR;
mnt->auth_flavor_len = 1;
mnt->auth_flavors[0] = RPC_AUTH_GSS_SPKMP;
break;
default:
goto out_unrec_sec;
}
break;
case Opt_proto:
string = match_strdup(args);
if (string == NULL)
goto out_nomem;
token = match_token(string,
nfs_xprt_protocol_tokens, args);
kfree(string);
switch (token) {
case Opt_udp:
mnt->flags &= ~NFS_MOUNT_TCP;
mnt->nfs_server.protocol = IPPROTO_UDP;
mnt->timeo = 7;
mnt->retrans = 5;
break;
case Opt_tcp:
mnt->flags |= NFS_MOUNT_TCP;
mnt->nfs_server.protocol = IPPROTO_TCP;
mnt->timeo = 600;
mnt->retrans = 2;
break;
default:
goto out_unrec_xprt;
}
break;
case Opt_mountproto:
string = match_strdup(args);
if (string == NULL)
goto out_nomem;
token = match_token(string,
nfs_xprt_protocol_tokens, args);
kfree(string);
switch (token) {
case Opt_udp:
mnt->mount_server.protocol = IPPROTO_UDP;
break;
case Opt_tcp:
mnt->mount_server.protocol = IPPROTO_TCP;
break;
default:
goto out_unrec_xprt;
}
break;
case Opt_addr:
string = match_strdup(args);
if (string == NULL)
goto out_nomem;
mnt->nfs_server.address.sin_family = AF_INET;
mnt->nfs_server.address.sin_addr.s_addr =
in_aton(string);
kfree(string);
break;
case Opt_clientaddr:
string = match_strdup(args);
if (string == NULL)
goto out_nomem;
mnt->client_address = string;
break;
case Opt_mounthost:
string = match_strdup(args);
if (string == NULL)
goto out_nomem;
mnt->mount_server.address.sin_family = AF_INET;
mnt->mount_server.address.sin_addr.s_addr =
in_aton(string);
kfree(string);
break;
case Opt_userspace:
case Opt_deprecated:
break;
default:
goto out_unknown;
}
}
return 1;
out_nomem:
printk(KERN_INFO "NFS: not enough memory to parse option\n");
return 0;
out_unrec_vers:
printk(KERN_INFO "NFS: unrecognized NFS version number\n");
return 0;
out_unrec_xprt:
printk(KERN_INFO "NFS: unrecognized transport protocol\n");
return 0;
out_unrec_sec:
printk(KERN_INFO "NFS: unrecognized security flavor\n");
return 0;
out_unknown:
printk(KERN_INFO "NFS: unknown mount option: %s\n", p);
return 0;
}
/*
* Use the remote server's MOUNT service to request the NFS file handle
* corresponding to the provided path.
*/
static int nfs_try_mount(struct nfs_parsed_mount_data *args,
struct nfs_fh *root_fh)
{
struct sockaddr_in sin;
int status;
if (args->mount_server.version == 0) {
if (args->flags & NFS_MOUNT_VER3)
args->mount_server.version = NFS_MNT3_VERSION;
else
args->mount_server.version = NFS_MNT_VERSION;
}
/*
* Construct the mount server's address.
*/
if (args->mount_server.address.sin_addr.s_addr != INADDR_ANY)
sin = args->mount_server.address;
else
sin = args->nfs_server.address;
if (args->mount_server.port == 0) {
status = rpcb_getport_sync(&sin,
args->mount_server.program,
args->mount_server.version,
args->mount_server.protocol);
if (status < 0)
goto out_err;
sin.sin_port = htons(status);
} else
sin.sin_port = htons(args->mount_server.port);
/*
* Now ask the mount server to map our export path
* to a file handle.
*/
status = nfs_mount((struct sockaddr *) &sin,
sizeof(sin),
args->nfs_server.hostname,
args->nfs_server.export_path,
args->mount_server.version,
args->mount_server.protocol,
root_fh);
if (status < 0)
goto out_err;
return status;
out_err:
dfprintk(MOUNT, "NFS: unable to contact server on host "
NIPQUAD_FMT "\n", NIPQUAD(sin.sin_addr.s_addr));
return status;
}
/*
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
* Validate the NFS2/NFS3 mount data
* - fills in the mount root filehandle
*
* For option strings, user space handles the following behaviors:
*
* + DNS: mapping server host name to IP address ("addr=" option)
*
* + failure mode: how to behave if a mount request can't be handled
* immediately ("fg/bg" option)
*
* + retry: how often to retry a mount request ("retry=" option)
*
* + breaking back: trying proto=udp after proto=tcp, v2 after v3,
* mountproto=tcp after mountproto=udp, and so on
*
* XXX: as far as I can tell, changing the NFS program number is not
* supported in the NFS client.
*/
static int nfs_validate_mount_data(struct nfs_mount_data **options,
struct nfs_fh *mntfh,
const char *dev_name)
{
struct nfs_mount_data *data = *options;
if (data == NULL)
goto out_no_data;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
switch (data->version) {
case 1:
data->namlen = 0;
case 2:
data->bsize = 0;
case 3:
if (data->flags & NFS_MOUNT_VER3)
goto out_no_v3;
data->root.size = NFS2_FHSIZE;
memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
case 4:
if (data->flags & NFS_MOUNT_SECFLAVOUR)
goto out_no_sec;
case 5:
memset(data->context, 0, sizeof(data->context));
case 6:
if (data->flags & NFS_MOUNT_VER3)
mntfh->size = data->root.size;
else
mntfh->size = NFS2_FHSIZE;
if (mntfh->size > sizeof(mntfh->data))
goto out_invalid_fh;
memcpy(mntfh->data, data->root.data, mntfh->size);
if (mntfh->size < sizeof(mntfh->data))
memset(mntfh->data + mntfh->size, 0,
sizeof(mntfh->data) - mntfh->size);
break;
default: {
unsigned int len;
char *c;
int status;
struct nfs_parsed_mount_data args = {
.flags = (NFS_MOUNT_VER3 | NFS_MOUNT_TCP),
.rsize = NFS_MAX_FILE_IO_SIZE,
.wsize = NFS_MAX_FILE_IO_SIZE,
.timeo = 600,
.retrans = 2,
.acregmin = 3,
.acregmax = 60,
.acdirmin = 30,
.acdirmax = 60,
.mount_server.protocol = IPPROTO_UDP,
.mount_server.program = NFS_MNT_PROGRAM,
.nfs_server.protocol = IPPROTO_TCP,
.nfs_server.program = NFS_PROGRAM,
};
if (nfs_parse_mount_options((char *) *options, &args) == 0)
return -EINVAL;
data = kzalloc(sizeof(*data), GFP_KERNEL);
if (data == NULL)
return -ENOMEM;
/*
* NB: after this point, caller will free "data"
* if we return an error
*/
*options = data;
c = strchr(dev_name, ':');
if (c == NULL)
return -EINVAL;
len = c - dev_name - 1;
if (len > sizeof(data->hostname))
return -EINVAL;
strncpy(data->hostname, dev_name, len);
args.nfs_server.hostname = data->hostname;
c++;
if (strlen(c) > NFS_MAXPATHLEN)
return -EINVAL;
args.nfs_server.export_path = c;
status = nfs_try_mount(&args, mntfh);
if (status)
return -EINVAL;
/*
* Translate to nfs_mount_data, which nfs_fill_super
* can deal with.
*/
data->version = 6;
data->flags = args.flags;
data->rsize = args.rsize;
data->wsize = args.wsize;
data->timeo = args.timeo;
data->retrans = args.retrans;
data->acregmin = args.acregmin;
data->acregmax = args.acregmax;
data->acdirmin = args.acdirmin;
data->acdirmax = args.acdirmax;
data->addr = args.nfs_server.address;
data->namlen = args.namlen;
data->bsize = args.bsize;
data->pseudoflavor = args.auth_flavors[0];
break;
}
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
data->pseudoflavor = RPC_AUTH_UNIX;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
#ifndef CONFIG_NFS_V3
if (data->flags & NFS_MOUNT_VER3)
goto out_v3_not_compiled;
#endif /* !CONFIG_NFS_V3 */
if (!nfs_verify_server_address((struct sockaddr *) &data->addr))
goto out_no_address;
return 0;
out_no_data:
dfprintk(MOUNT, "NFS: mount program didn't pass any mount data\n");
return -EINVAL;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
out_no_v3:
dfprintk(MOUNT, "NFS: nfs_mount_data version %d does not support v3\n",
data->version);
return -EINVAL;
out_no_sec:
dfprintk(MOUNT, "NFS: nfs_mount_data version supports only AUTH_SYS\n");
return -EINVAL;
#ifndef CONFIG_NFS_V3
out_v3_not_compiled:
dfprintk(MOUNT, "NFS: NFSv3 is not compiled into kernel\n");
return -EPROTONOSUPPORT;
#endif /* !CONFIG_NFS_V3 */
out_no_address:
dfprintk(MOUNT, "NFS: mount program didn't pass remote address\n");
return -EINVAL;
out_invalid_fh:
dfprintk(MOUNT, "NFS: invalid root filehandle\n");
return -EINVAL;
}
/*
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
* Initialise the common bits of the superblock
*/
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
static inline void nfs_initialise_sb(struct super_block *sb)
{
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct nfs_server *server = NFS_SB(sb);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
sb->s_magic = NFS_SUPER_MAGIC;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/* We probably want something more informative here */
snprintf(sb->s_id, sizeof(sb->s_id),
"%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (sb->s_blocksize == 0)
sb->s_blocksize = nfs_block_bits(server->wsize,
&sb->s_blocksize_bits);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (server->flags & NFS_MOUNT_NOAC)
sb->s_flags |= MS_SYNCHRONOUS;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
nfs_super_set_maxbytes(sb, server->maxfilesize);
}
/*
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
* Finish setting up an NFS2/3 superblock
*/
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
static void nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data)
{
struct nfs_server *server = NFS_SB(sb);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
sb->s_blocksize_bits = 0;
sb->s_blocksize = 0;
if (data->bsize)
sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (server->flags & NFS_MOUNT_VER3) {
/* The VFS shouldn't apply the umask to mode bits. We will do
* so ourselves when necessary.
*/
sb->s_flags |= MS_POSIXACL;
sb->s_time_gran = 1;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
sb->s_op = &nfs_sops;
nfs_initialise_sb(sb);
}
/*
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
* Finish setting up a cloned NFS2/3 superblock
*/
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
static void nfs_clone_super(struct super_block *sb,
const struct super_block *old_sb)
{
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct nfs_server *server = NFS_SB(sb);
sb->s_blocksize_bits = old_sb->s_blocksize_bits;
sb->s_blocksize = old_sb->s_blocksize;
sb->s_maxbytes = old_sb->s_maxbytes;
if (server->flags & NFS_MOUNT_VER3) {
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/* The VFS shouldn't apply the umask to mode bits. We will do
* so ourselves when necessary.
*/
sb->s_flags |= MS_POSIXACL;
sb->s_time_gran = 1;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
sb->s_op = old_sb->s_op;
nfs_initialise_sb(sb);
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
static int nfs_set_super(struct super_block *s, void *_server)
{
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct nfs_server *server = _server;
int ret;
s->s_fs_info = server;
ret = set_anon_super(s, server);
if (ret == 0)
server->s_dev = s->s_dev;
return ret;
}
static int nfs_compare_super(struct super_block *sb, void *data)
{
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct nfs_server *server = data, *old = NFS_SB(sb);
if (memcmp(&old->nfs_client->cl_addr,
&server->nfs_client->cl_addr,
sizeof(old->nfs_client->cl_addr)) != 0)
return 0;
/* Note: NFS_MOUNT_UNSHARED == NFS4_MOUNT_UNSHARED */
if (old->flags & NFS_MOUNT_UNSHARED)
return 0;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (memcmp(&old->fsid, &server->fsid, sizeof(old->fsid)) != 0)
return 0;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
return 1;
}
#define NFS_MS_MASK (MS_RDONLY|MS_NOSUID|MS_NODEV|MS_NOEXEC|MS_SYNCHRONOUS)
static int nfs_compare_mount_options(const struct super_block *s, const struct nfs_server *b, int flags)
{
const struct nfs_server *a = s->s_fs_info;
const struct rpc_clnt *clnt_a = a->client;
const struct rpc_clnt *clnt_b = b->client;
if ((s->s_flags & NFS_MS_MASK) != (flags & NFS_MS_MASK))
goto Ebusy;
if (a->nfs_client != b->nfs_client)
goto Ebusy;
if (a->flags != b->flags)
goto Ebusy;
if (a->wsize != b->wsize)
goto Ebusy;
if (a->rsize != b->rsize)
goto Ebusy;
if (a->acregmin != b->acregmin)
goto Ebusy;
if (a->acregmax != b->acregmax)
goto Ebusy;
if (a->acdirmin != b->acdirmin)
goto Ebusy;
if (a->acdirmax != b->acdirmax)
goto Ebusy;
if (clnt_a->cl_auth->au_flavor != clnt_b->cl_auth->au_flavor)
goto Ebusy;
return 0;
Ebusy:
return -EBUSY;
}
static int nfs_get_sb(struct file_system_type *fs_type,
int flags, const char *dev_name, void *raw_data, struct vfsmount *mnt)
{
struct nfs_server *server = NULL;
struct super_block *s;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct nfs_fh mntfh;
struct nfs_mount_data *data = raw_data;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct dentry *mntroot;
int (*compare_super)(struct super_block *, void *) = nfs_compare_super;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
int error;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/* Validate the mount data */
error = nfs_validate_mount_data(&data, &mntfh, dev_name);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (error < 0)
goto out;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/* Get a volume representation */
server = nfs_create_server(data, &mntfh);
if (IS_ERR(server)) {
error = PTR_ERR(server);
goto out;
}
if (server->flags & NFS_MOUNT_UNSHARED)
compare_super = NULL;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/* Get a superblock - note that we may end up sharing one that already exists */
s = sget(fs_type, compare_super, nfs_set_super, server);
if (IS_ERR(s)) {
error = PTR_ERR(s);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
goto out_err_nosb;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (s->s_fs_info != server) {
error = nfs_compare_mount_options(s, server, flags);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
nfs_free_server(server);
server = NULL;
if (error < 0)
goto error_splat_super;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (!s->s_root) {
/* initial superblock/root creation */
s->s_flags = flags;
nfs_fill_super(s, data);
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
mntroot = nfs_get_root(s, &mntfh);
if (IS_ERR(mntroot)) {
error = PTR_ERR(mntroot);
goto error_splat_super;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
s->s_flags |= MS_ACTIVE;
mnt->mnt_sb = s;
mnt->mnt_root = mntroot;
error = 0;
out:
if (data != raw_data)
kfree(data);
return error;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
out_err_nosb:
nfs_free_server(server);
goto out;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
error_splat_super:
up_write(&s->s_umount);
deactivate_super(s);
goto out;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/*
* Destroy an NFS2/3 superblock
*/
static void nfs_kill_super(struct super_block *s)
{
struct nfs_server *server = NFS_SB(s);
kill_anon_super(s);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
nfs_free_server(server);
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/*
* Clone an NFS2/3 server record on xdev traversal (FSID-change)
*/
static int nfs_xdev_get_sb(struct file_system_type *fs_type, int flags,
const char *dev_name, void *raw_data,
struct vfsmount *mnt)
{
struct nfs_clone_mount *data = raw_data;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct super_block *s;
struct nfs_server *server;
struct dentry *mntroot;
int (*compare_super)(struct super_block *, void *) = nfs_compare_super;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
int error;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
dprintk("--> nfs_xdev_get_sb()\n");
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/* create a new volume representation */
server = nfs_clone_server(NFS_SB(data->sb), data->fh, data->fattr);
if (IS_ERR(server)) {
error = PTR_ERR(server);
goto out_err_noserver;
}
if (server->flags & NFS_MOUNT_UNSHARED)
compare_super = NULL;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/* Get a superblock - note that we may end up sharing one that already exists */
s = sget(&nfs_fs_type, compare_super, nfs_set_super, server);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (IS_ERR(s)) {
error = PTR_ERR(s);
goto out_err_nosb;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (s->s_fs_info != server) {
error = nfs_compare_mount_options(s, server, flags);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
nfs_free_server(server);
server = NULL;
if (error < 0)
goto error_splat_super;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (!s->s_root) {
/* initial superblock/root creation */
s->s_flags = flags;
nfs_clone_super(s, data->sb);
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
mntroot = nfs_get_root(s, data->fh);
if (IS_ERR(mntroot)) {
error = PTR_ERR(mntroot);
goto error_splat_super;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
s->s_flags |= MS_ACTIVE;
mnt->mnt_sb = s;
mnt->mnt_root = mntroot;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
dprintk("<-- nfs_xdev_get_sb() = 0\n");
return 0;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
out_err_nosb:
nfs_free_server(server);
out_err_noserver:
dprintk("<-- nfs_xdev_get_sb() = %d [error]\n", error);
return error;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
error_splat_super:
up_write(&s->s_umount);
deactivate_super(s);
dprintk("<-- nfs_xdev_get_sb() = %d [splat]\n", error);
return error;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
#ifdef CONFIG_NFS_V4
/*
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
* Finish setting up a cloned NFS4 superblock
*/
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
static void nfs4_clone_super(struct super_block *sb,
const struct super_block *old_sb)
{
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
sb->s_blocksize_bits = old_sb->s_blocksize_bits;
sb->s_blocksize = old_sb->s_blocksize;
sb->s_maxbytes = old_sb->s_maxbytes;
sb->s_time_gran = 1;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
sb->s_op = old_sb->s_op;
nfs_initialise_sb(sb);
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/*
* Set up an NFS4 superblock
*/
static void nfs4_fill_super(struct super_block *sb)
{
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
sb->s_time_gran = 1;
sb->s_op = &nfs4_sops;
nfs_initialise_sb(sb);
}
/*
* Validate NFSv4 mount options
*/
static int nfs4_validate_mount_data(struct nfs4_mount_data **options,
const char *dev_name,
struct sockaddr_in *addr,
rpc_authflavor_t *authflavour,
char **hostname,
char **mntpath,
char **ip_addr)
{
struct nfs4_mount_data *data = *options;
char *c;
if (data == NULL)
goto out_no_data;
switch (data->version) {
case 1:
if (data->host_addrlen != sizeof(*addr))
goto out_no_address;
if (copy_from_user(addr, data->host_addr, sizeof(*addr)))
return -EFAULT;
if (addr->sin_port == 0)
addr->sin_port = htons(NFS_PORT);
if (!nfs_verify_server_address((struct sockaddr *) addr))
goto out_no_address;
switch (data->auth_flavourlen) {
case 0:
*authflavour = RPC_AUTH_UNIX;
break;
case 1:
if (copy_from_user(authflavour, data->auth_flavours,
sizeof(*authflavour)))
return -EFAULT;
break;
default:
goto out_inval_auth;
}
c = strndup_user(data->hostname.data, NFS4_MAXNAMLEN);
if (IS_ERR(c))
return PTR_ERR(c);
*hostname = c;
c = strndup_user(data->mnt_path.data, NFS4_MAXPATHLEN);
if (IS_ERR(c))
return PTR_ERR(c);
*mntpath = c;
dfprintk(MOUNT, "NFS: MNTPATH: '%s'\n", *mntpath);
c = strndup_user(data->client_addr.data, 16);
if (IS_ERR(c))
return PTR_ERR(c);
*ip_addr = c;
break;
default: {
unsigned int len;
struct nfs_parsed_mount_data args = {
.rsize = NFS_MAX_FILE_IO_SIZE,
.wsize = NFS_MAX_FILE_IO_SIZE,
.timeo = 600,
.retrans = 2,
.acregmin = 3,
.acregmax = 60,
.acdirmin = 30,
.acdirmax = 60,
.nfs_server.protocol = IPPROTO_TCP,
};
if (nfs_parse_mount_options((char *) *options, &args) == 0)
return -EINVAL;
if (!nfs_verify_server_address((struct sockaddr *)
&args.nfs_server.address))
return -EINVAL;
*addr = args.nfs_server.address;
switch (args.auth_flavor_len) {
case 0:
*authflavour = RPC_AUTH_UNIX;
break;
case 1:
*authflavour = (rpc_authflavor_t) args.auth_flavors[0];
break;
default:
goto out_inval_auth;
}
/*
* Translate to nfs4_mount_data, which nfs4_fill_super
* can deal with.
*/
data = kzalloc(sizeof(*data), GFP_KERNEL);
if (data == NULL)
return -ENOMEM;
*options = data;
data->version = 1;
data->flags = args.flags & NFS4_MOUNT_FLAGMASK;
data->rsize = args.rsize;
data->wsize = args.wsize;
data->timeo = args.timeo;
data->retrans = args.retrans;
data->acregmin = args.acregmin;
data->acregmax = args.acregmax;
data->acdirmin = args.acdirmin;
data->acdirmax = args.acdirmax;
data->proto = args.nfs_server.protocol;
/*
* Split "dev_name" into "hostname:mntpath".
*/
c = strchr(dev_name, ':');
if (c == NULL)
return -EINVAL;
/* while calculating len, pretend ':' is '\0' */
len = c - dev_name;
if (len > NFS4_MAXNAMLEN)
return -EINVAL;
*hostname = kzalloc(len, GFP_KERNEL);
if (*hostname == NULL)
return -ENOMEM;
strncpy(*hostname, dev_name, len - 1);
c++; /* step over the ':' */
len = strlen(c);
if (len > NFS4_MAXPATHLEN)
return -EINVAL;
*mntpath = kzalloc(len + 1, GFP_KERNEL);
if (*mntpath == NULL)
return -ENOMEM;
strncpy(*mntpath, c, len);
dprintk("MNTPATH: %s\n", *mntpath);
*ip_addr = args.client_address;
break;
}
}
return 0;
out_no_data:
dfprintk(MOUNT, "NFS4: mount program didn't pass any mount data\n");
return -EINVAL;
out_inval_auth:
dfprintk(MOUNT, "NFS4: Invalid number of RPC auth flavours %d\n",
data->auth_flavourlen);
return -EINVAL;
out_no_address:
dfprintk(MOUNT, "NFS4: mount program didn't pass remote address\n");
return -EINVAL;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/*
* Get the superblock for an NFS4 mountpoint
*/
static int nfs4_get_sb(struct file_system_type *fs_type,
int flags, const char *dev_name, void *raw_data, struct vfsmount *mnt)
{
struct nfs4_mount_data *data = raw_data;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct super_block *s;
struct nfs_server *server;
struct sockaddr_in addr;
rpc_authflavor_t authflavour;
struct nfs_fh mntfh;
struct dentry *mntroot;
char *mntpath = NULL, *hostname = NULL, *ip_addr = NULL;
int (*compare_super)(struct super_block *, void *) = nfs_compare_super;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
int error;
/* Validate the mount data */
error = nfs4_validate_mount_data(&data, dev_name, &addr, &authflavour,
&hostname, &mntpath, &ip_addr);
if (error < 0)
goto out;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/* Get a volume representation */
server = nfs4_create_server(data, hostname, &addr, mntpath, ip_addr,
authflavour, &mntfh);
if (IS_ERR(server)) {
error = PTR_ERR(server);
goto out;
}
if (server->flags & NFS4_MOUNT_UNSHARED)
compare_super = NULL;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/* Get a superblock - note that we may end up sharing one that already exists */
s = sget(fs_type, compare_super, nfs_set_super, server);
if (IS_ERR(s)) {
error = PTR_ERR(s);
goto out_free;
}
if (s->s_fs_info != server) {
nfs_free_server(server);
server = NULL;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (!s->s_root) {
/* initial superblock/root creation */
s->s_flags = flags;
nfs4_fill_super(s);
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
mntroot = nfs4_get_root(s, &mntfh);
if (IS_ERR(mntroot)) {
error = PTR_ERR(mntroot);
goto error_splat_super;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
s->s_flags |= MS_ACTIVE;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
mnt->mnt_sb = s;
mnt->mnt_root = mntroot;
error = 0;
out:
kfree(ip_addr);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
kfree(mntpath);
kfree(hostname);
return error;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
out_free:
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
nfs_free_server(server);
goto out;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
error_splat_super:
up_write(&s->s_umount);
deactivate_super(s);
goto out;
}
static void nfs4_kill_super(struct super_block *sb)
{
struct nfs_server *server = NFS_SB(sb);
nfs_return_all_delegations(sb);
kill_anon_super(sb);
nfs4_renewd_prepare_shutdown(server);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
nfs_free_server(server);
}
/*
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
* Clone an NFS4 server record on xdev traversal (FSID-change)
*/
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
static int nfs4_xdev_get_sb(struct file_system_type *fs_type, int flags,
const char *dev_name, void *raw_data,
struct vfsmount *mnt)
{
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct nfs_clone_mount *data = raw_data;
struct super_block *s;
struct nfs_server *server;
struct dentry *mntroot;
int (*compare_super)(struct super_block *, void *) = nfs_compare_super;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
int error;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
dprintk("--> nfs4_xdev_get_sb()\n");
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/* create a new volume representation */
server = nfs_clone_server(NFS_SB(data->sb), data->fh, data->fattr);
if (IS_ERR(server)) {
error = PTR_ERR(server);
goto out_err_noserver;
}
if (server->flags & NFS4_MOUNT_UNSHARED)
compare_super = NULL;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/* Get a superblock - note that we may end up sharing one that already exists */
s = sget(&nfs_fs_type, compare_super, nfs_set_super, server);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (IS_ERR(s)) {
error = PTR_ERR(s);
goto out_err_nosb;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (s->s_fs_info != server) {
nfs_free_server(server);
server = NULL;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (!s->s_root) {
/* initial superblock/root creation */
s->s_flags = flags;
nfs4_clone_super(s, data->sb);
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
mntroot = nfs4_get_root(s, data->fh);
if (IS_ERR(mntroot)) {
error = PTR_ERR(mntroot);
goto error_splat_super;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
s->s_flags |= MS_ACTIVE;
mnt->mnt_sb = s;
mnt->mnt_root = mntroot;
dprintk("<-- nfs4_xdev_get_sb() = 0\n");
return 0;
out_err_nosb:
nfs_free_server(server);
out_err_noserver:
dprintk("<-- nfs4_xdev_get_sb() = %d [error]\n", error);
return error;
error_splat_super:
up_write(&s->s_umount);
deactivate_super(s);
dprintk("<-- nfs4_xdev_get_sb() = %d [splat]\n", error);
return error;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/*
* Create an NFS4 server record on referral traversal
*/
static int nfs4_referral_get_sb(struct file_system_type *fs_type, int flags,
const char *dev_name, void *raw_data,
struct vfsmount *mnt)
{
struct nfs_clone_mount *data = raw_data;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct super_block *s;
struct nfs_server *server;
struct dentry *mntroot;
struct nfs_fh mntfh;
int (*compare_super)(struct super_block *, void *) = nfs_compare_super;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
int error;
dprintk("--> nfs4_referral_get_sb()\n");
/* create a new volume representation */
server = nfs4_create_referral_server(data, &mntfh);
if (IS_ERR(server)) {
error = PTR_ERR(server);
goto out_err_noserver;
}
if (server->flags & NFS4_MOUNT_UNSHARED)
compare_super = NULL;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/* Get a superblock - note that we may end up sharing one that already exists */
s = sget(&nfs_fs_type, compare_super, nfs_set_super, server);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (IS_ERR(s)) {
error = PTR_ERR(s);
goto out_err_nosb;
}
if (s->s_fs_info != server) {
nfs_free_server(server);
server = NULL;
}
if (!s->s_root) {
/* initial superblock/root creation */
s->s_flags = flags;
nfs4_fill_super(s);
}
mntroot = nfs4_get_root(s, &mntfh);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (IS_ERR(mntroot)) {
error = PTR_ERR(mntroot);
goto error_splat_super;
}
s->s_flags |= MS_ACTIVE;
mnt->mnt_sb = s;
mnt->mnt_root = mntroot;
dprintk("<-- nfs4_referral_get_sb() = 0\n");
return 0;
out_err_nosb:
nfs_free_server(server);
out_err_noserver:
dprintk("<-- nfs4_referral_get_sb() = %d [error]\n", error);
return error;
error_splat_super:
up_write(&s->s_umount);
deactivate_super(s);
dprintk("<-- nfs4_referral_get_sb() = %d [splat]\n", error);
return error;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
#endif /* CONFIG_NFS_V4 */