2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* fs/dcache.c
|
|
|
|
*
|
|
|
|
* Complete reimplementation
|
|
|
|
* (C) 1997 Thomas Schoebel-Theuer,
|
|
|
|
* with heavy changes by Linus Torvalds
|
|
|
|
*/
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Notes on the allocation strategy:
|
|
|
|
*
|
|
|
|
* The dcache is a master of the icache - whenever a dcache entry
|
|
|
|
* exists, the inode will always exist. "iput()" is done either when
|
|
|
|
* the dcache entry is deleted or garbage collected.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/syscalls.h>
|
|
|
|
#include <linux/string.h>
|
|
|
|
#include <linux/mm.h>
|
|
|
|
#include <linux/fs.h>
|
2005-08-09 01:52:16 +08:00
|
|
|
#include <linux/fsnotify.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
#include <linux/slab.h>
|
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/hash.h>
|
|
|
|
#include <linux/cache.h>
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/mount.h>
|
|
|
|
#include <linux/file.h>
|
|
|
|
#include <asm/uaccess.h>
|
|
|
|
#include <linux/security.h>
|
|
|
|
#include <linux/seqlock.h>
|
|
|
|
#include <linux/swap.h>
|
|
|
|
#include <linux/bootmem.h>
|
2009-03-30 07:50:06 +08:00
|
|
|
#include <linux/fs_struct.h>
|
2009-07-16 21:44:29 +08:00
|
|
|
#include <linux/hardirq.h>
|
2006-10-01 02:52:18 +08:00
|
|
|
#include "internal.h"
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2006-03-26 17:37:24 +08:00
|
|
|
int sysctl_vfs_cache_pressure __read_mostly = 100;
|
2005-04-17 06:20:36 +08:00
|
|
|
EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
|
|
|
|
|
|
|
|
__cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock);
|
[PATCH] audit: watching subtrees
New kind of audit rule predicates: "object is visible in given subtree".
The part that can be sanely implemented, that is. Limitations:
* if you have hardlink from outside of tree, you'd better watch
it too (or just watch the object itself, obviously)
* if you mount something under a watched tree, tell audit
that new chunk should be added to watched subtrees
* if you umount something in a watched tree and it's still mounted
elsewhere, you will get matches on events happening there. New command
tells audit to recalculate the trees, trimming such sources of false
positives.
Note that it's _not_ about path - if something mounted in several places
(multiple mount, bindings, different namespaces, etc.), the match does
_not_ depend on which one we are using for access.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2007-07-22 20:04:18 +08:00
|
|
|
__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
EXPORT_SYMBOL(dcache_lock);
|
|
|
|
|
2006-12-07 12:33:20 +08:00
|
|
|
static struct kmem_cache *dentry_cache __read_mostly;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
#define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This is the single most critical data structure when it comes
|
|
|
|
* to the dcache: the hashtable for lookups. Somebody should try
|
|
|
|
* to make this good - I've just made it work.
|
|
|
|
*
|
|
|
|
* This hash-function tries to avoid losing too many bits of hash
|
|
|
|
* information, yet avoid using a prime hash-size or similar.
|
|
|
|
*/
|
|
|
|
#define D_HASHBITS d_hash_shift
|
|
|
|
#define D_HASHMASK d_hash_mask
|
|
|
|
|
2006-03-26 17:37:24 +08:00
|
|
|
static unsigned int d_hash_mask __read_mostly;
|
|
|
|
static unsigned int d_hash_shift __read_mostly;
|
|
|
|
static struct hlist_head *dentry_hashtable __read_mostly;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* Statistics gathering. */
|
|
|
|
struct dentry_stat_t dentry_stat = {
|
|
|
|
.age_limit = 45,
|
|
|
|
};
|
|
|
|
|
fs: use fast counters for vfs caches
percpu_counter library generates quite nasty code, so unless you need
to dynamically allocate counters or take fast approximate value, a
simple per cpu set of counters is much better.
The percpu_counter can never be made to work as well, because it has an
indirection from pointer to percpu memory, and it can't use direct
this_cpu_inc interfaces because it doesn't use static PER_CPU data, so
code will always be worse.
In the fastpath, it is the difference between this:
incl %gs:nr_dentry # nr_dentry
and this:
movl percpu_counter_batch(%rip), %edx # percpu_counter_batch,
movl $1, %esi #,
movq $nr_dentry, %rdi #,
call __percpu_counter_add # (plus I clobber registers)
__percpu_counter_add:
pushq %rbp #
movq %rsp, %rbp #,
subq $32, %rsp #,
movq %rbx, -24(%rbp) #,
movq %r12, -16(%rbp) #,
movq %r13, -8(%rbp) #,
movq %rdi, %rbx # fbc, fbc
#APP
# 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1
movq %gs:kernel_stack,%rax #, pfo_ret__
# 0 "" 2
#NO_APP
incl -8124(%rax) # <variable>.preempt_count
movq 32(%rdi), %r12 # <variable>.counters, tcp_ptr__
#APP
# 78 "lib/percpu_counter.c" 1
add %gs:this_cpu_off, %r12 # this_cpu_off, tcp_ptr__
# 0 "" 2
#NO_APP
movslq (%r12),%r13 #* tcp_ptr__, tmp73
movslq %edx,%rax # batch, batch
addq %rsi, %r13 # amount, count
cmpq %rax, %r13 # batch, count
jge .L27 #,
negl %edx # tmp76
movslq %edx,%rdx # tmp76, tmp77
cmpq %rdx, %r13 # tmp77, count
jg .L28 #,
.L27:
movq %rbx, %rdi # fbc,
call _raw_spin_lock #
addq %r13, 8(%rbx) # count, <variable>.count
movq %rbx, %rdi # fbc,
movl $0, (%r12) #,* tcp_ptr__
call _raw_spin_unlock #
.L29:
#APP
# 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1
movq %gs:kernel_stack,%rax #, pfo_ret__
# 0 "" 2
#NO_APP
decl -8124(%rax) # <variable>.preempt_count
movq -8136(%rax), %rax #, D.14625
testb $8, %al #, D.14625
jne .L32 #,
.L31:
movq -24(%rbp), %rbx #,
movq -16(%rbp), %r12 #,
movq -8(%rbp), %r13 #,
leave
ret
.p2align 4,,10
.p2align 3
.L28:
movl %r13d, (%r12) # count,*
jmp .L29 #
.L32:
call preempt_schedule #
.p2align 4,,6
jmp .L31 #
.size __percpu_counter_add, .-__percpu_counter_add
.p2align 4,,15
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:19 +08:00
|
|
|
static DEFINE_PER_CPU(unsigned int, nr_dentry);
|
2010-10-10 17:36:23 +08:00
|
|
|
|
|
|
|
#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
|
fs: use fast counters for vfs caches
percpu_counter library generates quite nasty code, so unless you need
to dynamically allocate counters or take fast approximate value, a
simple per cpu set of counters is much better.
The percpu_counter can never be made to work as well, because it has an
indirection from pointer to percpu memory, and it can't use direct
this_cpu_inc interfaces because it doesn't use static PER_CPU data, so
code will always be worse.
In the fastpath, it is the difference between this:
incl %gs:nr_dentry # nr_dentry
and this:
movl percpu_counter_batch(%rip), %edx # percpu_counter_batch,
movl $1, %esi #,
movq $nr_dentry, %rdi #,
call __percpu_counter_add # (plus I clobber registers)
__percpu_counter_add:
pushq %rbp #
movq %rsp, %rbp #,
subq $32, %rsp #,
movq %rbx, -24(%rbp) #,
movq %r12, -16(%rbp) #,
movq %r13, -8(%rbp) #,
movq %rdi, %rbx # fbc, fbc
#APP
# 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1
movq %gs:kernel_stack,%rax #, pfo_ret__
# 0 "" 2
#NO_APP
incl -8124(%rax) # <variable>.preempt_count
movq 32(%rdi), %r12 # <variable>.counters, tcp_ptr__
#APP
# 78 "lib/percpu_counter.c" 1
add %gs:this_cpu_off, %r12 # this_cpu_off, tcp_ptr__
# 0 "" 2
#NO_APP
movslq (%r12),%r13 #* tcp_ptr__, tmp73
movslq %edx,%rax # batch, batch
addq %rsi, %r13 # amount, count
cmpq %rax, %r13 # batch, count
jge .L27 #,
negl %edx # tmp76
movslq %edx,%rdx # tmp76, tmp77
cmpq %rdx, %r13 # tmp77, count
jg .L28 #,
.L27:
movq %rbx, %rdi # fbc,
call _raw_spin_lock #
addq %r13, 8(%rbx) # count, <variable>.count
movq %rbx, %rdi # fbc,
movl $0, (%r12) #,* tcp_ptr__
call _raw_spin_unlock #
.L29:
#APP
# 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1
movq %gs:kernel_stack,%rax #, pfo_ret__
# 0 "" 2
#NO_APP
decl -8124(%rax) # <variable>.preempt_count
movq -8136(%rax), %rax #, D.14625
testb $8, %al #, D.14625
jne .L32 #,
.L31:
movq -24(%rbp), %rbx #,
movq -16(%rbp), %r12 #,
movq -8(%rbp), %r13 #,
leave
ret
.p2align 4,,10
.p2align 3
.L28:
movl %r13d, (%r12) # count,*
jmp .L29 #
.L32:
call preempt_schedule #
.p2align 4,,6
jmp .L31 #
.size __percpu_counter_add, .-__percpu_counter_add
.p2align 4,,15
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:19 +08:00
|
|
|
static int get_nr_dentry(void)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
int sum = 0;
|
|
|
|
for_each_possible_cpu(i)
|
|
|
|
sum += per_cpu(nr_dentry, i);
|
|
|
|
return sum < 0 ? 0 : sum;
|
|
|
|
}
|
|
|
|
|
2010-10-10 17:36:23 +08:00
|
|
|
int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
|
|
|
|
size_t *lenp, loff_t *ppos)
|
|
|
|
{
|
fs: use fast counters for vfs caches
percpu_counter library generates quite nasty code, so unless you need
to dynamically allocate counters or take fast approximate value, a
simple per cpu set of counters is much better.
The percpu_counter can never be made to work as well, because it has an
indirection from pointer to percpu memory, and it can't use direct
this_cpu_inc interfaces because it doesn't use static PER_CPU data, so
code will always be worse.
In the fastpath, it is the difference between this:
incl %gs:nr_dentry # nr_dentry
and this:
movl percpu_counter_batch(%rip), %edx # percpu_counter_batch,
movl $1, %esi #,
movq $nr_dentry, %rdi #,
call __percpu_counter_add # (plus I clobber registers)
__percpu_counter_add:
pushq %rbp #
movq %rsp, %rbp #,
subq $32, %rsp #,
movq %rbx, -24(%rbp) #,
movq %r12, -16(%rbp) #,
movq %r13, -8(%rbp) #,
movq %rdi, %rbx # fbc, fbc
#APP
# 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1
movq %gs:kernel_stack,%rax #, pfo_ret__
# 0 "" 2
#NO_APP
incl -8124(%rax) # <variable>.preempt_count
movq 32(%rdi), %r12 # <variable>.counters, tcp_ptr__
#APP
# 78 "lib/percpu_counter.c" 1
add %gs:this_cpu_off, %r12 # this_cpu_off, tcp_ptr__
# 0 "" 2
#NO_APP
movslq (%r12),%r13 #* tcp_ptr__, tmp73
movslq %edx,%rax # batch, batch
addq %rsi, %r13 # amount, count
cmpq %rax, %r13 # batch, count
jge .L27 #,
negl %edx # tmp76
movslq %edx,%rdx # tmp76, tmp77
cmpq %rdx, %r13 # tmp77, count
jg .L28 #,
.L27:
movq %rbx, %rdi # fbc,
call _raw_spin_lock #
addq %r13, 8(%rbx) # count, <variable>.count
movq %rbx, %rdi # fbc,
movl $0, (%r12) #,* tcp_ptr__
call _raw_spin_unlock #
.L29:
#APP
# 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1
movq %gs:kernel_stack,%rax #, pfo_ret__
# 0 "" 2
#NO_APP
decl -8124(%rax) # <variable>.preempt_count
movq -8136(%rax), %rax #, D.14625
testb $8, %al #, D.14625
jne .L32 #,
.L31:
movq -24(%rbp), %rbx #,
movq -16(%rbp), %r12 #,
movq -8(%rbp), %r13 #,
leave
ret
.p2align 4,,10
.p2align 3
.L28:
movl %r13d, (%r12) # count,*
jmp .L29 #
.L32:
call preempt_schedule #
.p2align 4,,6
jmp .L31 #
.size __percpu_counter_add, .-__percpu_counter_add
.p2align 4,,15
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:19 +08:00
|
|
|
dentry_stat.nr_dentry = get_nr_dentry();
|
2010-10-10 17:36:23 +08:00
|
|
|
return proc_dointvec(table, write, buffer, lenp, ppos);
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2010-10-10 17:36:22 +08:00
|
|
|
static void __d_free(struct rcu_head *head)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2010-10-10 17:36:22 +08:00
|
|
|
struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
|
|
|
|
|
2008-10-21 21:47:33 +08:00
|
|
|
WARN_ON(!list_empty(&dentry->d_alias));
|
2005-04-17 06:20:36 +08:00
|
|
|
if (dname_external(dentry))
|
|
|
|
kfree(dentry->d_name.name);
|
|
|
|
kmem_cache_free(dentry_cache, dentry);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2010-10-10 17:36:23 +08:00
|
|
|
* no dcache_lock, please.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
|
|
|
static void d_free(struct dentry *dentry)
|
|
|
|
{
|
fs: use fast counters for vfs caches
percpu_counter library generates quite nasty code, so unless you need
to dynamically allocate counters or take fast approximate value, a
simple per cpu set of counters is much better.
The percpu_counter can never be made to work as well, because it has an
indirection from pointer to percpu memory, and it can't use direct
this_cpu_inc interfaces because it doesn't use static PER_CPU data, so
code will always be worse.
In the fastpath, it is the difference between this:
incl %gs:nr_dentry # nr_dentry
and this:
movl percpu_counter_batch(%rip), %edx # percpu_counter_batch,
movl $1, %esi #,
movq $nr_dentry, %rdi #,
call __percpu_counter_add # (plus I clobber registers)
__percpu_counter_add:
pushq %rbp #
movq %rsp, %rbp #,
subq $32, %rsp #,
movq %rbx, -24(%rbp) #,
movq %r12, -16(%rbp) #,
movq %r13, -8(%rbp) #,
movq %rdi, %rbx # fbc, fbc
#APP
# 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1
movq %gs:kernel_stack,%rax #, pfo_ret__
# 0 "" 2
#NO_APP
incl -8124(%rax) # <variable>.preempt_count
movq 32(%rdi), %r12 # <variable>.counters, tcp_ptr__
#APP
# 78 "lib/percpu_counter.c" 1
add %gs:this_cpu_off, %r12 # this_cpu_off, tcp_ptr__
# 0 "" 2
#NO_APP
movslq (%r12),%r13 #* tcp_ptr__, tmp73
movslq %edx,%rax # batch, batch
addq %rsi, %r13 # amount, count
cmpq %rax, %r13 # batch, count
jge .L27 #,
negl %edx # tmp76
movslq %edx,%rdx # tmp76, tmp77
cmpq %rdx, %r13 # tmp77, count
jg .L28 #,
.L27:
movq %rbx, %rdi # fbc,
call _raw_spin_lock #
addq %r13, 8(%rbx) # count, <variable>.count
movq %rbx, %rdi # fbc,
movl $0, (%r12) #,* tcp_ptr__
call _raw_spin_unlock #
.L29:
#APP
# 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1
movq %gs:kernel_stack,%rax #, pfo_ret__
# 0 "" 2
#NO_APP
decl -8124(%rax) # <variable>.preempt_count
movq -8136(%rax), %rax #, D.14625
testb $8, %al #, D.14625
jne .L32 #,
.L31:
movq -24(%rbp), %rbx #,
movq -16(%rbp), %r12 #,
movq -8(%rbp), %r13 #,
leave
ret
.p2align 4,,10
.p2align 3
.L28:
movl %r13d, (%r12) # count,*
jmp .L29 #
.L32:
call preempt_schedule #
.p2align 4,,6
jmp .L31 #
.size __percpu_counter_add, .-__percpu_counter_add
.p2align 4,,15
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:19 +08:00
|
|
|
this_cpu_dec(nr_dentry);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (dentry->d_op && dentry->d_op->d_release)
|
|
|
|
dentry->d_op->d_release(dentry);
|
2010-10-10 17:36:23 +08:00
|
|
|
|
2006-12-07 12:38:48 +08:00
|
|
|
/* if dentry was never inserted into hash, immediate free is OK */
|
2008-02-06 17:37:07 +08:00
|
|
|
if (hlist_unhashed(&dentry->d_hash))
|
2010-10-10 17:36:22 +08:00
|
|
|
__d_free(&dentry->d_u.d_rcu);
|
2006-12-07 12:38:48 +08:00
|
|
|
else
|
2010-10-10 17:36:22 +08:00
|
|
|
call_rcu(&dentry->d_u.d_rcu, __d_free);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Release the dentry's inode, using the filesystem
|
|
|
|
* d_iput() operation if defined.
|
|
|
|
*/
|
2006-01-15 05:20:43 +08:00
|
|
|
static void dentry_iput(struct dentry * dentry)
|
2008-06-24 00:11:52 +08:00
|
|
|
__releases(dentry->d_lock)
|
|
|
|
__releases(dcache_lock)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct inode *inode = dentry->d_inode;
|
|
|
|
if (inode) {
|
|
|
|
dentry->d_inode = NULL;
|
|
|
|
list_del_init(&dentry->d_alias);
|
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
spin_unlock(&dcache_lock);
|
2005-09-20 10:54:29 +08:00
|
|
|
if (!inode->i_nlink)
|
|
|
|
fsnotify_inoderemove(inode);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (dentry->d_op && dentry->d_op->d_iput)
|
|
|
|
dentry->d_op->d_iput(dentry, inode);
|
|
|
|
else
|
|
|
|
iput(inode);
|
|
|
|
} else {
|
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
/*
|
2010-10-10 17:36:26 +08:00
|
|
|
* dentry_lru_(add|del|move_tail) must be called with dcache_lock held.
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
*/
|
|
|
|
static void dentry_lru_add(struct dentry *dentry)
|
|
|
|
{
|
2010-10-10 17:36:26 +08:00
|
|
|
if (list_empty(&dentry->d_lru)) {
|
|
|
|
list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
|
|
|
|
dentry->d_sb->s_nr_dentry_unused++;
|
2011-01-07 14:49:18 +08:00
|
|
|
dentry_stat.nr_unused++;
|
2010-10-10 17:36:26 +08:00
|
|
|
}
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void dentry_lru_del(struct dentry *dentry)
|
|
|
|
{
|
|
|
|
if (!list_empty(&dentry->d_lru)) {
|
2010-10-10 17:36:26 +08:00
|
|
|
list_del_init(&dentry->d_lru);
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
dentry->d_sb->s_nr_dentry_unused--;
|
2011-01-07 14:49:18 +08:00
|
|
|
dentry_stat.nr_unused--;
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2010-10-10 17:36:26 +08:00
|
|
|
static void dentry_lru_move_tail(struct dentry *dentry)
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
{
|
2010-10-10 17:36:26 +08:00
|
|
|
if (list_empty(&dentry->d_lru)) {
|
|
|
|
list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
|
|
|
|
dentry->d_sb->s_nr_dentry_unused++;
|
2011-01-07 14:49:18 +08:00
|
|
|
dentry_stat.nr_unused++;
|
2010-10-10 17:36:26 +08:00
|
|
|
} else {
|
|
|
|
list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-05-08 15:23:46 +08:00
|
|
|
/**
|
|
|
|
* d_kill - kill dentry and return parent
|
|
|
|
* @dentry: dentry to kill
|
|
|
|
*
|
2008-06-24 00:11:52 +08:00
|
|
|
* The dentry must already be unhashed and removed from the LRU.
|
2007-05-08 15:23:46 +08:00
|
|
|
*
|
|
|
|
* If this is the root of the dentry tree, return NULL.
|
|
|
|
*/
|
|
|
|
static struct dentry *d_kill(struct dentry *dentry)
|
2008-06-24 00:11:52 +08:00
|
|
|
__releases(dentry->d_lock)
|
|
|
|
__releases(dcache_lock)
|
2007-05-08 15:23:46 +08:00
|
|
|
{
|
|
|
|
struct dentry *parent;
|
|
|
|
|
|
|
|
list_del(&dentry->d_u.d_child);
|
|
|
|
/*drops the locks, at that point nobody can reach this dentry */
|
|
|
|
dentry_iput(dentry);
|
2008-10-16 06:50:27 +08:00
|
|
|
if (IS_ROOT(dentry))
|
|
|
|
parent = NULL;
|
|
|
|
else
|
|
|
|
parent = dentry->d_parent;
|
2007-05-08 15:23:46 +08:00
|
|
|
d_free(dentry);
|
2008-10-16 06:50:27 +08:00
|
|
|
return parent;
|
2007-05-08 15:23:46 +08:00
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* This is dput
|
|
|
|
*
|
|
|
|
* This is complicated by the fact that we do not want to put
|
|
|
|
* dentries that are no longer on any hash chain on the unused
|
|
|
|
* list: we'd much rather just get rid of them immediately.
|
|
|
|
*
|
|
|
|
* However, that implies that we have to traverse the dentry
|
|
|
|
* tree upwards to the parents which might _also_ now be
|
|
|
|
* scheduled for deletion (it may have been only waiting for
|
|
|
|
* its last child to go away).
|
|
|
|
*
|
|
|
|
* This tail recursion is done by hand as we don't want to depend
|
|
|
|
* on the compiler to always get this right (gcc generally doesn't).
|
|
|
|
* Real recursion would eat up our stack space.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/*
|
|
|
|
* dput - release a dentry
|
|
|
|
* @dentry: dentry to release
|
|
|
|
*
|
|
|
|
* Release a dentry. This will drop the usage count and if appropriate
|
|
|
|
* call the dentry unlink method as well as removing it from the queues and
|
|
|
|
* releasing its resources. If the parent dentries were scheduled for release
|
|
|
|
* they too may now get deleted.
|
|
|
|
*
|
|
|
|
* no dcache lock, please.
|
|
|
|
*/
|
|
|
|
|
|
|
|
void dput(struct dentry *dentry)
|
|
|
|
{
|
|
|
|
if (!dentry)
|
|
|
|
return;
|
|
|
|
|
|
|
|
repeat:
|
|
|
|
if (atomic_read(&dentry->d_count) == 1)
|
|
|
|
might_sleep();
|
|
|
|
if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
|
|
|
|
return;
|
|
|
|
|
|
|
|
spin_lock(&dentry->d_lock);
|
|
|
|
if (atomic_read(&dentry->d_count)) {
|
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* AV: ->d_delete() is _NOT_ allowed to block now.
|
|
|
|
*/
|
|
|
|
if (dentry->d_op && dentry->d_op->d_delete) {
|
|
|
|
if (dentry->d_op->d_delete(dentry))
|
|
|
|
goto unhash_it;
|
|
|
|
}
|
2010-10-10 17:36:24 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/* Unreachable? Get rid of it */
|
|
|
|
if (d_unhashed(dentry))
|
|
|
|
goto kill_it;
|
2010-10-10 17:36:24 +08:00
|
|
|
|
|
|
|
/* Otherwise leave it cached and ensure it's on the LRU */
|
|
|
|
dentry->d_flags |= DCACHE_REFERENCED;
|
2010-10-10 17:36:26 +08:00
|
|
|
dentry_lru_add(dentry);
|
2010-10-10 17:36:24 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
return;
|
|
|
|
|
|
|
|
unhash_it:
|
|
|
|
__d_drop(dentry);
|
2007-05-08 15:23:46 +08:00
|
|
|
kill_it:
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
/* if dentry was on the d_lru list delete it from there */
|
|
|
|
dentry_lru_del(dentry);
|
2007-05-08 15:23:46 +08:00
|
|
|
dentry = d_kill(dentry);
|
|
|
|
if (dentry)
|
|
|
|
goto repeat;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(dput);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/**
|
|
|
|
* d_invalidate - invalidate a dentry
|
|
|
|
* @dentry: dentry to invalidate
|
|
|
|
*
|
|
|
|
* Try to invalidate the dentry if it turns out to be
|
|
|
|
* possible. If there are other dentries that can be
|
|
|
|
* reached through this one we can't delete it and we
|
|
|
|
* return -EBUSY. On success we return 0.
|
|
|
|
*
|
|
|
|
* no dcache lock.
|
|
|
|
*/
|
|
|
|
|
|
|
|
int d_invalidate(struct dentry * dentry)
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
* If it's already been dropped, return OK.
|
|
|
|
*/
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
if (d_unhashed(dentry)) {
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* Check whether to do a partial shrink_dcache
|
|
|
|
* to get rid of unused child entries.
|
|
|
|
*/
|
|
|
|
if (!list_empty(&dentry->d_subdirs)) {
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
shrink_dcache_parent(dentry);
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Somebody else still using it?
|
|
|
|
*
|
|
|
|
* If it's a directory, we can't drop it
|
|
|
|
* for fear of somebody re-populating it
|
|
|
|
* with children (even though dropping it
|
|
|
|
* would make it unreachable from the root,
|
|
|
|
* we might still populate it if it was a
|
|
|
|
* working directory or similar).
|
|
|
|
*/
|
|
|
|
spin_lock(&dentry->d_lock);
|
|
|
|
if (atomic_read(&dentry->d_count) > 1) {
|
|
|
|
if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
|
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
return -EBUSY;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
__d_drop(dentry);
|
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
return 0;
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(d_invalidate);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* This should be called _only_ with dcache_lock held */
|
|
|
|
static inline struct dentry * __dget_locked(struct dentry *dentry)
|
|
|
|
{
|
|
|
|
atomic_inc(&dentry->d_count);
|
2010-10-10 17:36:26 +08:00
|
|
|
dentry_lru_del(dentry);
|
2005-04-17 06:20:36 +08:00
|
|
|
return dentry;
|
|
|
|
}
|
|
|
|
|
|
|
|
struct dentry * dget_locked(struct dentry *dentry)
|
|
|
|
{
|
|
|
|
return __dget_locked(dentry);
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(dget_locked);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/**
|
|
|
|
* d_find_alias - grab a hashed alias of inode
|
|
|
|
* @inode: inode in question
|
|
|
|
* @want_discon: flag, used by d_splice_alias, to request
|
|
|
|
* that only a DISCONNECTED alias be returned.
|
|
|
|
*
|
|
|
|
* If inode has a hashed alias, or is a directory and has any alias,
|
|
|
|
* acquire the reference to alias and return it. Otherwise return NULL.
|
|
|
|
* Notice that if inode is a directory there can be only one alias and
|
|
|
|
* it can be unhashed only if it has no children, or if it is the root
|
|
|
|
* of a filesystem.
|
|
|
|
*
|
[PATCH] knfsd: close a race-opportunity in d_splice_alias
There is a possible race in d_splice_alias. Though __d_find_alias(inode, 1)
will only return a dentry with DCACHE_DISCONNECTED set, it is possible for it
to get cleared before the BUG_ON, and it is is not possible to lock against
that.
There are a couple of problems here. Firstly, the code doesn't match the
comment. The comment describes a 'disconnected' dentry as being IS_ROOT as
well as DCACHE_DISCONNECTED, however there is not testing of IS_ROOT anythere.
A dentry is marked DCACHE_DISCONNECTED when allocated with d_alloc_anon, and
remains DCACHE_DISCONNECTED while a path is built up towards the root. So a
dentry can have a valid name and a valid parent and even grandparent, but will
still be DCACHE_DISCONNECTED until a path to the root is created. Once the
path to the root is complete, everything in the path gets DCACHE_DISCONNECTED
cleared. So the fact that DCACHE_DISCONNECTED isn't enough to say that a
dentry is free to be spliced in with a given name. This can only be allowed
if the dentry does not yet have a name, so the IS_ROOT test is needed too.
However even adding that test to __d_find_alias isn't enough. As
d_splice_alias drops dcache_lock before calling d_move to perform the splice,
it could race with another thread calling d_splice_alias to splice the inode
in with a different name in a different part of the tree (in the case where a
file has hard links). So that splicing code is only really safe for
directories (as we know that directories only have one link). For
directories, the caller of d_splice_alias will be holding i_mutex on the
(unique) parent so there is no room for a race.
A consequence of this is that a non-directory will never benefit from being
spliced into a pre-exisiting dentry, but that isn't a problem. It is
perfectly OK for a non-directory to have multiple dentries, some anonymous,
some not. And the comment for d_splice_alias says that it only happens for
directories anyway.
Signed-off-by: Neil Brown <neilb@suse.de>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-04 17:16:16 +08:00
|
|
|
* If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
|
2005-04-17 06:20:36 +08:00
|
|
|
* any other hashed alias over that one unless @want_discon is set,
|
[PATCH] knfsd: close a race-opportunity in d_splice_alias
There is a possible race in d_splice_alias. Though __d_find_alias(inode, 1)
will only return a dentry with DCACHE_DISCONNECTED set, it is possible for it
to get cleared before the BUG_ON, and it is is not possible to lock against
that.
There are a couple of problems here. Firstly, the code doesn't match the
comment. The comment describes a 'disconnected' dentry as being IS_ROOT as
well as DCACHE_DISCONNECTED, however there is not testing of IS_ROOT anythere.
A dentry is marked DCACHE_DISCONNECTED when allocated with d_alloc_anon, and
remains DCACHE_DISCONNECTED while a path is built up towards the root. So a
dentry can have a valid name and a valid parent and even grandparent, but will
still be DCACHE_DISCONNECTED until a path to the root is created. Once the
path to the root is complete, everything in the path gets DCACHE_DISCONNECTED
cleared. So the fact that DCACHE_DISCONNECTED isn't enough to say that a
dentry is free to be spliced in with a given name. This can only be allowed
if the dentry does not yet have a name, so the IS_ROOT test is needed too.
However even adding that test to __d_find_alias isn't enough. As
d_splice_alias drops dcache_lock before calling d_move to perform the splice,
it could race with another thread calling d_splice_alias to splice the inode
in with a different name in a different part of the tree (in the case where a
file has hard links). So that splicing code is only really safe for
directories (as we know that directories only have one link). For
directories, the caller of d_splice_alias will be holding i_mutex on the
(unique) parent so there is no room for a race.
A consequence of this is that a non-directory will never benefit from being
spliced into a pre-exisiting dentry, but that isn't a problem. It is
perfectly OK for a non-directory to have multiple dentries, some anonymous,
some not. And the comment for d_splice_alias says that it only happens for
directories anyway.
Signed-off-by: Neil Brown <neilb@suse.de>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-04 17:16:16 +08:00
|
|
|
* in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
|
|
|
|
|
|
|
static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
|
|
|
|
{
|
|
|
|
struct list_head *head, *next, *tmp;
|
|
|
|
struct dentry *alias, *discon_alias=NULL;
|
|
|
|
|
|
|
|
head = &inode->i_dentry;
|
|
|
|
next = inode->i_dentry.next;
|
|
|
|
while (next != head) {
|
|
|
|
tmp = next;
|
|
|
|
next = tmp->next;
|
|
|
|
prefetch(next);
|
|
|
|
alias = list_entry(tmp, struct dentry, d_alias);
|
|
|
|
if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
|
[PATCH] knfsd: close a race-opportunity in d_splice_alias
There is a possible race in d_splice_alias. Though __d_find_alias(inode, 1)
will only return a dentry with DCACHE_DISCONNECTED set, it is possible for it
to get cleared before the BUG_ON, and it is is not possible to lock against
that.
There are a couple of problems here. Firstly, the code doesn't match the
comment. The comment describes a 'disconnected' dentry as being IS_ROOT as
well as DCACHE_DISCONNECTED, however there is not testing of IS_ROOT anythere.
A dentry is marked DCACHE_DISCONNECTED when allocated with d_alloc_anon, and
remains DCACHE_DISCONNECTED while a path is built up towards the root. So a
dentry can have a valid name and a valid parent and even grandparent, but will
still be DCACHE_DISCONNECTED until a path to the root is created. Once the
path to the root is complete, everything in the path gets DCACHE_DISCONNECTED
cleared. So the fact that DCACHE_DISCONNECTED isn't enough to say that a
dentry is free to be spliced in with a given name. This can only be allowed
if the dentry does not yet have a name, so the IS_ROOT test is needed too.
However even adding that test to __d_find_alias isn't enough. As
d_splice_alias drops dcache_lock before calling d_move to perform the splice,
it could race with another thread calling d_splice_alias to splice the inode
in with a different name in a different part of the tree (in the case where a
file has hard links). So that splicing code is only really safe for
directories (as we know that directories only have one link). For
directories, the caller of d_splice_alias will be holding i_mutex on the
(unique) parent so there is no room for a race.
A consequence of this is that a non-directory will never benefit from being
spliced into a pre-exisiting dentry, but that isn't a problem. It is
perfectly OK for a non-directory to have multiple dentries, some anonymous,
some not. And the comment for d_splice_alias says that it only happens for
directories anyway.
Signed-off-by: Neil Brown <neilb@suse.de>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-04 17:16:16 +08:00
|
|
|
if (IS_ROOT(alias) &&
|
|
|
|
(alias->d_flags & DCACHE_DISCONNECTED))
|
2005-04-17 06:20:36 +08:00
|
|
|
discon_alias = alias;
|
|
|
|
else if (!want_discon) {
|
|
|
|
__dget_locked(alias);
|
|
|
|
return alias;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (discon_alias)
|
|
|
|
__dget_locked(discon_alias);
|
|
|
|
return discon_alias;
|
|
|
|
}
|
|
|
|
|
|
|
|
struct dentry * d_find_alias(struct inode *inode)
|
|
|
|
{
|
2006-03-25 19:06:36 +08:00
|
|
|
struct dentry *de = NULL;
|
|
|
|
|
|
|
|
if (!list_empty(&inode->i_dentry)) {
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
de = __d_find_alias(inode, 0);
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
return de;
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(d_find_alias);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Try to kill dentries associated with this inode.
|
|
|
|
* WARNING: you must own a reference to inode.
|
|
|
|
*/
|
|
|
|
void d_prune_aliases(struct inode *inode)
|
|
|
|
{
|
2005-09-10 15:27:07 +08:00
|
|
|
struct dentry *dentry;
|
2005-04-17 06:20:36 +08:00
|
|
|
restart:
|
|
|
|
spin_lock(&dcache_lock);
|
2005-09-10 15:27:07 +08:00
|
|
|
list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_lock(&dentry->d_lock);
|
|
|
|
if (!atomic_read(&dentry->d_count)) {
|
|
|
|
__dget_locked(dentry);
|
|
|
|
__d_drop(dentry);
|
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
dput(dentry);
|
|
|
|
goto restart;
|
|
|
|
}
|
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
}
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(d_prune_aliases);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/*
|
2006-06-23 05:47:31 +08:00
|
|
|
* Throw away a dentry - free the inode, dput the parent. This requires that
|
|
|
|
* the LRU list has already been removed.
|
|
|
|
*
|
2007-10-17 14:27:09 +08:00
|
|
|
* Try to prune ancestors as well. This is necessary to prevent
|
|
|
|
* quadratic behavior of shrink_dcache_parent(), but is also expected
|
|
|
|
* to be beneficial in reducing dentry cache fragmentation.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2007-10-17 14:27:09 +08:00
|
|
|
static void prune_one_dentry(struct dentry * dentry)
|
2008-06-24 00:11:52 +08:00
|
|
|
__releases(dentry->d_lock)
|
|
|
|
__releases(dcache_lock)
|
|
|
|
__acquires(dcache_lock)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
__d_drop(dentry);
|
2007-05-08 15:23:46 +08:00
|
|
|
dentry = d_kill(dentry);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Prune ancestors. Locking is simpler than in dput(),
|
|
|
|
* because dcache_lock needs to be taken anyway.
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_lock(&dcache_lock);
|
2007-05-08 15:23:46 +08:00
|
|
|
while (dentry) {
|
|
|
|
if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock))
|
|
|
|
return;
|
|
|
|
|
2010-10-10 17:36:26 +08:00
|
|
|
dentry_lru_del(dentry);
|
2007-05-08 15:23:46 +08:00
|
|
|
__d_drop(dentry);
|
|
|
|
dentry = d_kill(dentry);
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2010-10-10 17:36:25 +08:00
|
|
|
static void shrink_dentry_list(struct list_head *list)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
struct dentry *dentry;
|
|
|
|
|
2010-10-10 17:36:25 +08:00
|
|
|
while (!list_empty(list)) {
|
|
|
|
dentry = list_entry(list->prev, struct dentry, d_lru);
|
2010-10-10 17:36:26 +08:00
|
|
|
dentry_lru_del(dentry);
|
2010-10-10 17:36:25 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* We found an inuse dentry which was not removed from
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
* the LRU because of laziness during lookup. Do not free
|
|
|
|
* it - just keep it off the LRU list.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2010-10-10 17:36:25 +08:00
|
|
|
spin_lock(&dentry->d_lock);
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
if (atomic_read(&dentry->d_count)) {
|
|
|
|
spin_unlock(&dentry->d_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
continue;
|
|
|
|
}
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
prune_one_dentry(dentry);
|
|
|
|
/* dentry->d_lock was dropped in prune_one_dentry() */
|
|
|
|
cond_resched_lock(&dcache_lock);
|
|
|
|
}
|
2010-10-10 17:36:25 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __shrink_dcache_sb - shrink the dentry LRU on a given superblock
|
|
|
|
* @sb: superblock to shrink dentry LRU.
|
|
|
|
* @count: number of entries to prune
|
|
|
|
* @flags: flags to control the dentry processing
|
|
|
|
*
|
|
|
|
* If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
|
|
|
|
*/
|
|
|
|
static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
|
|
|
|
{
|
|
|
|
/* called from prune_dcache() and shrink_dcache_parent() */
|
|
|
|
struct dentry *dentry;
|
|
|
|
LIST_HEAD(referenced);
|
|
|
|
LIST_HEAD(tmp);
|
|
|
|
int cnt = *count;
|
|
|
|
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
while (!list_empty(&sb->s_dentry_lru)) {
|
|
|
|
dentry = list_entry(sb->s_dentry_lru.prev,
|
|
|
|
struct dentry, d_lru);
|
|
|
|
BUG_ON(dentry->d_sb != sb);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If we are honouring the DCACHE_REFERENCED flag and the
|
|
|
|
* dentry has this flag set, don't free it. Clear the flag
|
|
|
|
* and put it back on the LRU.
|
|
|
|
*/
|
|
|
|
if (flags & DCACHE_REFERENCED) {
|
|
|
|
spin_lock(&dentry->d_lock);
|
|
|
|
if (dentry->d_flags & DCACHE_REFERENCED) {
|
|
|
|
dentry->d_flags &= ~DCACHE_REFERENCED;
|
|
|
|
list_move(&dentry->d_lru, &referenced);
|
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
cond_resched_lock(&dcache_lock);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
list_move_tail(&dentry->d_lru, &tmp);
|
|
|
|
if (!--cnt)
|
|
|
|
break;
|
|
|
|
cond_resched_lock(&dcache_lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
*count = cnt;
|
|
|
|
shrink_dentry_list(&tmp);
|
|
|
|
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
if (!list_empty(&referenced))
|
|
|
|
list_splice(&referenced, &sb->s_dentry_lru);
|
|
|
|
spin_unlock(&dcache_lock);
|
2010-10-10 17:36:25 +08:00
|
|
|
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* prune_dcache - shrink the dcache
|
|
|
|
* @count: number of entries to try to free
|
|
|
|
*
|
|
|
|
* Shrink the dcache. This is done when we need more memory, or simply when we
|
|
|
|
* need to unmount something (at which point we need to unuse all dentries).
|
|
|
|
*
|
|
|
|
* This function may fail to free any resources if all the dentries are in use.
|
|
|
|
*/
|
|
|
|
static void prune_dcache(int count)
|
|
|
|
{
|
2010-07-25 06:31:46 +08:00
|
|
|
struct super_block *sb, *p = NULL;
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
int w_count;
|
2011-01-07 14:49:18 +08:00
|
|
|
int unused = dentry_stat.nr_unused;
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
int prune_ratio;
|
|
|
|
int pruned;
|
|
|
|
|
|
|
|
if (unused == 0 || count == 0)
|
|
|
|
return;
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
if (count >= unused)
|
|
|
|
prune_ratio = 1;
|
|
|
|
else
|
|
|
|
prune_ratio = unused / count;
|
|
|
|
spin_lock(&sb_lock);
|
2010-07-25 06:31:46 +08:00
|
|
|
list_for_each_entry(sb, &super_blocks, s_list) {
|
2010-03-23 07:36:35 +08:00
|
|
|
if (list_empty(&sb->s_instances))
|
|
|
|
continue;
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
if (sb->s_nr_dentry_unused == 0)
|
2005-04-17 06:20:36 +08:00
|
|
|
continue;
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
sb->s_count++;
|
|
|
|
/* Now, we reclaim unused dentrins with fairness.
|
|
|
|
* We reclaim them same percentage from each superblock.
|
|
|
|
* We calculate number of dentries to scan on this sb
|
|
|
|
* as follows, but the implementation is arranged to avoid
|
|
|
|
* overflows:
|
|
|
|
* number of dentries to scan on this sb =
|
|
|
|
* count * (number of dentries on this sb /
|
|
|
|
* number of dentries in the machine)
|
[PATCH] Fix dcache race during umount
The race is that the shrink_dcache_memory shrinker could get called while a
filesystem is being unmounted, and could try to prune a dentry belonging to
that filesystem.
If it does, then it will call in to iput on the inode while the dentry is
no longer able to be found by the umounting process. If iput takes a
while, generic_shutdown_super could get all the way though
shrink_dcache_parent and shrink_dcache_anon and invalidate_inodes without
ever waiting on this particular inode.
Eventually the superblock gets freed anyway and if the iput tried to touch
it (which some filesystems certainly do), it will lose. The promised
"Self-destruct in 5 seconds" doesn't lead to a nice day.
The race is closed by holding s_umount while calling prune_one_dentry on
someone else's dentry. As a down_read_trylock is used,
shrink_dcache_memory will no longer try to prune the dentry of a filesystem
that is being unmounted, and unmount will not be able to start until any
such active prune_one_dentry completes.
This requires that prune_dcache *knows* which filesystem (if any) it is
doing the prune on behalf of so that it can be careful of other
filesystems. shrink_dcache_memory isn't called it on behalf of any
filesystem, and so is careful of everything.
shrink_dcache_anon is now passed a super_block rather than the s_anon list
out of the superblock, so it can get the s_anon list itself, and can pass
the superblock down to prune_dcache.
If prune_dcache finds a dentry that it cannot free, it leaves it where it
is (at the tail of the list) and exits, on the assumption that some other
thread will be removing that dentry soon. To try to make sure that some
work gets done, a limited number of dnetries which are untouchable are
skipped over while choosing the dentry to work on.
I believe this race was first found by Kirill Korotaev.
Cc: Jan Blunck <jblunck@suse.de>
Acked-by: Kirill Korotaev <dev@openvz.org>
Cc: Olaf Hering <olh@suse.de>
Acked-by: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Balbir Singh <balbir@in.ibm.com>
Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 05:47:28 +08:00
|
|
|
*/
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
spin_unlock(&sb_lock);
|
|
|
|
if (prune_ratio != 1)
|
|
|
|
w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
|
|
|
|
else
|
|
|
|
w_count = sb->s_nr_dentry_unused;
|
|
|
|
pruned = w_count;
|
[PATCH] Fix dcache race during umount
The race is that the shrink_dcache_memory shrinker could get called while a
filesystem is being unmounted, and could try to prune a dentry belonging to
that filesystem.
If it does, then it will call in to iput on the inode while the dentry is
no longer able to be found by the umounting process. If iput takes a
while, generic_shutdown_super could get all the way though
shrink_dcache_parent and shrink_dcache_anon and invalidate_inodes without
ever waiting on this particular inode.
Eventually the superblock gets freed anyway and if the iput tried to touch
it (which some filesystems certainly do), it will lose. The promised
"Self-destruct in 5 seconds" doesn't lead to a nice day.
The race is closed by holding s_umount while calling prune_one_dentry on
someone else's dentry. As a down_read_trylock is used,
shrink_dcache_memory will no longer try to prune the dentry of a filesystem
that is being unmounted, and unmount will not be able to start until any
such active prune_one_dentry completes.
This requires that prune_dcache *knows* which filesystem (if any) it is
doing the prune on behalf of so that it can be careful of other
filesystems. shrink_dcache_memory isn't called it on behalf of any
filesystem, and so is careful of everything.
shrink_dcache_anon is now passed a super_block rather than the s_anon list
out of the superblock, so it can get the s_anon list itself, and can pass
the superblock down to prune_dcache.
If prune_dcache finds a dentry that it cannot free, it leaves it where it
is (at the tail of the list) and exits, on the assumption that some other
thread will be removing that dentry soon. To try to make sure that some
work gets done, a limited number of dnetries which are untouchable are
skipped over while choosing the dentry to work on.
I believe this race was first found by Kirill Korotaev.
Cc: Jan Blunck <jblunck@suse.de>
Acked-by: Kirill Korotaev <dev@openvz.org>
Cc: Olaf Hering <olh@suse.de>
Acked-by: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Balbir Singh <balbir@in.ibm.com>
Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 05:47:28 +08:00
|
|
|
/*
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
* We need to be sure this filesystem isn't being unmounted,
|
|
|
|
* otherwise we could race with generic_shutdown_super(), and
|
|
|
|
* end up holding a reference to an inode while the filesystem
|
|
|
|
* is unmounted. So we try to get s_umount, and make sure
|
|
|
|
* s_root isn't NULL.
|
[PATCH] Fix dcache race during umount
The race is that the shrink_dcache_memory shrinker could get called while a
filesystem is being unmounted, and could try to prune a dentry belonging to
that filesystem.
If it does, then it will call in to iput on the inode while the dentry is
no longer able to be found by the umounting process. If iput takes a
while, generic_shutdown_super could get all the way though
shrink_dcache_parent and shrink_dcache_anon and invalidate_inodes without
ever waiting on this particular inode.
Eventually the superblock gets freed anyway and if the iput tried to touch
it (which some filesystems certainly do), it will lose. The promised
"Self-destruct in 5 seconds" doesn't lead to a nice day.
The race is closed by holding s_umount while calling prune_one_dentry on
someone else's dentry. As a down_read_trylock is used,
shrink_dcache_memory will no longer try to prune the dentry of a filesystem
that is being unmounted, and unmount will not be able to start until any
such active prune_one_dentry completes.
This requires that prune_dcache *knows* which filesystem (if any) it is
doing the prune on behalf of so that it can be careful of other
filesystems. shrink_dcache_memory isn't called it on behalf of any
filesystem, and so is careful of everything.
shrink_dcache_anon is now passed a super_block rather than the s_anon list
out of the superblock, so it can get the s_anon list itself, and can pass
the superblock down to prune_dcache.
If prune_dcache finds a dentry that it cannot free, it leaves it where it
is (at the tail of the list) and exits, on the assumption that some other
thread will be removing that dentry soon. To try to make sure that some
work gets done, a limited number of dnetries which are untouchable are
skipped over while choosing the dentry to work on.
I believe this race was first found by Kirill Korotaev.
Cc: Jan Blunck <jblunck@suse.de>
Acked-by: Kirill Korotaev <dev@openvz.org>
Cc: Olaf Hering <olh@suse.de>
Acked-by: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Balbir Singh <balbir@in.ibm.com>
Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 05:47:28 +08:00
|
|
|
*/
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
if (down_read_trylock(&sb->s_umount)) {
|
|
|
|
if ((sb->s_root != NULL) &&
|
|
|
|
(!list_empty(&sb->s_dentry_lru))) {
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
__shrink_dcache_sb(sb, &w_count,
|
|
|
|
DCACHE_REFERENCED);
|
|
|
|
pruned -= w_count;
|
|
|
|
spin_lock(&dcache_lock);
|
[PATCH] Fix dcache race during umount
The race is that the shrink_dcache_memory shrinker could get called while a
filesystem is being unmounted, and could try to prune a dentry belonging to
that filesystem.
If it does, then it will call in to iput on the inode while the dentry is
no longer able to be found by the umounting process. If iput takes a
while, generic_shutdown_super could get all the way though
shrink_dcache_parent and shrink_dcache_anon and invalidate_inodes without
ever waiting on this particular inode.
Eventually the superblock gets freed anyway and if the iput tried to touch
it (which some filesystems certainly do), it will lose. The promised
"Self-destruct in 5 seconds" doesn't lead to a nice day.
The race is closed by holding s_umount while calling prune_one_dentry on
someone else's dentry. As a down_read_trylock is used,
shrink_dcache_memory will no longer try to prune the dentry of a filesystem
that is being unmounted, and unmount will not be able to start until any
such active prune_one_dentry completes.
This requires that prune_dcache *knows* which filesystem (if any) it is
doing the prune on behalf of so that it can be careful of other
filesystems. shrink_dcache_memory isn't called it on behalf of any
filesystem, and so is careful of everything.
shrink_dcache_anon is now passed a super_block rather than the s_anon list
out of the superblock, so it can get the s_anon list itself, and can pass
the superblock down to prune_dcache.
If prune_dcache finds a dentry that it cannot free, it leaves it where it
is (at the tail of the list) and exits, on the assumption that some other
thread will be removing that dentry soon. To try to make sure that some
work gets done, a limited number of dnetries which are untouchable are
skipped over while choosing the dentry to work on.
I believe this race was first found by Kirill Korotaev.
Cc: Jan Blunck <jblunck@suse.de>
Acked-by: Kirill Korotaev <dev@openvz.org>
Cc: Olaf Hering <olh@suse.de>
Acked-by: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Balbir Singh <balbir@in.ibm.com>
Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 05:47:28 +08:00
|
|
|
}
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
up_read(&sb->s_umount);
|
[PATCH] Fix dcache race during umount
The race is that the shrink_dcache_memory shrinker could get called while a
filesystem is being unmounted, and could try to prune a dentry belonging to
that filesystem.
If it does, then it will call in to iput on the inode while the dentry is
no longer able to be found by the umounting process. If iput takes a
while, generic_shutdown_super could get all the way though
shrink_dcache_parent and shrink_dcache_anon and invalidate_inodes without
ever waiting on this particular inode.
Eventually the superblock gets freed anyway and if the iput tried to touch
it (which some filesystems certainly do), it will lose. The promised
"Self-destruct in 5 seconds" doesn't lead to a nice day.
The race is closed by holding s_umount while calling prune_one_dentry on
someone else's dentry. As a down_read_trylock is used,
shrink_dcache_memory will no longer try to prune the dentry of a filesystem
that is being unmounted, and unmount will not be able to start until any
such active prune_one_dentry completes.
This requires that prune_dcache *knows* which filesystem (if any) it is
doing the prune on behalf of so that it can be careful of other
filesystems. shrink_dcache_memory isn't called it on behalf of any
filesystem, and so is careful of everything.
shrink_dcache_anon is now passed a super_block rather than the s_anon list
out of the superblock, so it can get the s_anon list itself, and can pass
the superblock down to prune_dcache.
If prune_dcache finds a dentry that it cannot free, it leaves it where it
is (at the tail of the list) and exits, on the assumption that some other
thread will be removing that dentry soon. To try to make sure that some
work gets done, a limited number of dnetries which are untouchable are
skipped over while choosing the dentry to work on.
I believe this race was first found by Kirill Korotaev.
Cc: Jan Blunck <jblunck@suse.de>
Acked-by: Kirill Korotaev <dev@openvz.org>
Cc: Olaf Hering <olh@suse.de>
Acked-by: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Balbir Singh <balbir@in.ibm.com>
Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 05:47:28 +08:00
|
|
|
}
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
spin_lock(&sb_lock);
|
2010-07-25 06:31:46 +08:00
|
|
|
if (p)
|
|
|
|
__put_super(p);
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
count -= pruned;
|
2010-07-25 06:31:46 +08:00
|
|
|
p = sb;
|
2010-03-23 08:27:55 +08:00
|
|
|
/* more work left to do? */
|
|
|
|
if (count <= 0)
|
|
|
|
break;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2010-07-25 06:31:46 +08:00
|
|
|
if (p)
|
|
|
|
__put_super(p);
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
spin_unlock(&sb_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* shrink_dcache_sb - shrink dcache for a superblock
|
|
|
|
* @sb: superblock
|
|
|
|
*
|
2010-10-10 17:36:25 +08:00
|
|
|
* Shrink the dcache for the specified super block. This is used to free
|
|
|
|
* the dcache before unmounting a file system.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2010-10-10 17:36:25 +08:00
|
|
|
void shrink_dcache_sb(struct super_block *sb)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2010-10-10 17:36:25 +08:00
|
|
|
LIST_HEAD(tmp);
|
|
|
|
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
while (!list_empty(&sb->s_dentry_lru)) {
|
|
|
|
list_splice_init(&sb->s_dentry_lru, &tmp);
|
|
|
|
shrink_dentry_list(&tmp);
|
|
|
|
}
|
|
|
|
spin_unlock(&dcache_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(shrink_dcache_sb);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2006-10-11 16:22:19 +08:00
|
|
|
/*
|
|
|
|
* destroy a single subtree of dentries for unmount
|
|
|
|
* - see the comments on shrink_dcache_for_umount() for a description of the
|
|
|
|
* locking
|
|
|
|
*/
|
|
|
|
static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
|
|
|
|
{
|
|
|
|
struct dentry *parent;
|
2006-10-29 01:38:46 +08:00
|
|
|
unsigned detached = 0;
|
2006-10-11 16:22:19 +08:00
|
|
|
|
|
|
|
BUG_ON(!IS_ROOT(dentry));
|
|
|
|
|
|
|
|
/* detach this root from the system */
|
|
|
|
spin_lock(&dcache_lock);
|
2010-10-10 17:36:26 +08:00
|
|
|
dentry_lru_del(dentry);
|
2006-10-11 16:22:19 +08:00
|
|
|
__d_drop(dentry);
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
|
|
|
|
for (;;) {
|
|
|
|
/* descend to the first leaf in the current subtree */
|
|
|
|
while (!list_empty(&dentry->d_subdirs)) {
|
|
|
|
struct dentry *loop;
|
|
|
|
|
|
|
|
/* this is a branch with children - detach all of them
|
|
|
|
* from the system in one go */
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
list_for_each_entry(loop, &dentry->d_subdirs,
|
|
|
|
d_u.d_child) {
|
2010-10-10 17:36:26 +08:00
|
|
|
dentry_lru_del(loop);
|
2006-10-11 16:22:19 +08:00
|
|
|
__d_drop(loop);
|
|
|
|
cond_resched_lock(&dcache_lock);
|
|
|
|
}
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
|
|
|
|
/* move to the first child */
|
|
|
|
dentry = list_entry(dentry->d_subdirs.next,
|
|
|
|
struct dentry, d_u.d_child);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* consume the dentries from this leaf up through its parents
|
|
|
|
* until we find one with children or run out altogether */
|
|
|
|
do {
|
|
|
|
struct inode *inode;
|
|
|
|
|
|
|
|
if (atomic_read(&dentry->d_count) != 0) {
|
|
|
|
printk(KERN_ERR
|
|
|
|
"BUG: Dentry %p{i=%lx,n=%s}"
|
|
|
|
" still in use (%d)"
|
|
|
|
" [unmount of %s %s]\n",
|
|
|
|
dentry,
|
|
|
|
dentry->d_inode ?
|
|
|
|
dentry->d_inode->i_ino : 0UL,
|
|
|
|
dentry->d_name.name,
|
|
|
|
atomic_read(&dentry->d_count),
|
|
|
|
dentry->d_sb->s_type->name,
|
|
|
|
dentry->d_sb->s_id);
|
|
|
|
BUG();
|
|
|
|
}
|
|
|
|
|
2008-10-16 06:50:27 +08:00
|
|
|
if (IS_ROOT(dentry))
|
2006-10-11 16:22:19 +08:00
|
|
|
parent = NULL;
|
2008-10-16 06:50:27 +08:00
|
|
|
else {
|
|
|
|
parent = dentry->d_parent;
|
2006-10-11 16:22:19 +08:00
|
|
|
atomic_dec(&parent->d_count);
|
2008-10-16 06:50:27 +08:00
|
|
|
}
|
2006-10-11 16:22:19 +08:00
|
|
|
|
|
|
|
list_del(&dentry->d_u.d_child);
|
2006-10-29 01:38:46 +08:00
|
|
|
detached++;
|
2006-10-11 16:22:19 +08:00
|
|
|
|
|
|
|
inode = dentry->d_inode;
|
|
|
|
if (inode) {
|
|
|
|
dentry->d_inode = NULL;
|
|
|
|
list_del_init(&dentry->d_alias);
|
|
|
|
if (dentry->d_op && dentry->d_op->d_iput)
|
|
|
|
dentry->d_op->d_iput(dentry, inode);
|
|
|
|
else
|
|
|
|
iput(inode);
|
|
|
|
}
|
|
|
|
|
|
|
|
d_free(dentry);
|
|
|
|
|
|
|
|
/* finished when we fall off the top of the tree,
|
|
|
|
* otherwise we ascend to the parent and move to the
|
|
|
|
* next sibling if there is one */
|
|
|
|
if (!parent)
|
2010-10-10 17:36:23 +08:00
|
|
|
return;
|
2006-10-11 16:22:19 +08:00
|
|
|
dentry = parent;
|
|
|
|
} while (list_empty(&dentry->d_subdirs));
|
|
|
|
|
|
|
|
dentry = list_entry(dentry->d_subdirs.next,
|
|
|
|
struct dentry, d_u.d_child);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* destroy the dentries attached to a superblock on unmounting
|
|
|
|
* - we don't need to use dentry->d_lock, and only need dcache_lock when
|
|
|
|
* removing the dentry from the system lists and hashes because:
|
|
|
|
* - the superblock is detached from all mountings and open files, so the
|
|
|
|
* dentry trees will not be rearranged by the VFS
|
|
|
|
* - s_umount is write-locked, so the memory pressure shrinker will ignore
|
|
|
|
* any dentries belonging to this superblock that it comes across
|
|
|
|
* - the filesystem itself is no longer permitted to rearrange the dentries
|
|
|
|
* in this superblock
|
|
|
|
*/
|
|
|
|
void shrink_dcache_for_umount(struct super_block *sb)
|
|
|
|
{
|
|
|
|
struct dentry *dentry;
|
|
|
|
|
|
|
|
if (down_read_trylock(&sb->s_umount))
|
|
|
|
BUG();
|
|
|
|
|
|
|
|
dentry = sb->s_root;
|
|
|
|
sb->s_root = NULL;
|
|
|
|
atomic_dec(&dentry->d_count);
|
|
|
|
shrink_dcache_for_umount_subtree(dentry);
|
|
|
|
|
|
|
|
while (!hlist_empty(&sb->s_anon)) {
|
|
|
|
dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash);
|
|
|
|
shrink_dcache_for_umount_subtree(dentry);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* Search for at least 1 mount point in the dentry's subdirs.
|
|
|
|
* We descend to the next level whenever the d_subdirs
|
|
|
|
* list is non-empty and continue searching.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/**
|
|
|
|
* have_submounts - check for mounts over a dentry
|
|
|
|
* @parent: dentry to check.
|
|
|
|
*
|
|
|
|
* Return true if the parent or its subdirectories contain
|
|
|
|
* a mount point
|
|
|
|
*/
|
|
|
|
|
|
|
|
int have_submounts(struct dentry *parent)
|
|
|
|
{
|
|
|
|
struct dentry *this_parent = parent;
|
|
|
|
struct list_head *next;
|
|
|
|
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
if (d_mountpoint(parent))
|
|
|
|
goto positive;
|
|
|
|
repeat:
|
|
|
|
next = this_parent->d_subdirs.next;
|
|
|
|
resume:
|
|
|
|
while (next != &this_parent->d_subdirs) {
|
|
|
|
struct list_head *tmp = next;
|
[PATCH] shrink dentry struct
Some long time ago, dentry struct was carefully tuned so that on 32 bits
UP, sizeof(struct dentry) was exactly 128, ie a power of 2, and a multiple
of memory cache lines.
Then RCU was added and dentry struct enlarged by two pointers, with nice
results for SMP, but not so good on UP, because breaking the above tuning
(128 + 8 = 136 bytes)
This patch reverts this unwanted side effect, by using an union (d_u),
where d_rcu and d_child are placed so that these two fields can share their
memory needs.
At the time d_free() is called (and d_rcu is really used), d_child is known
to be empty and not touched by the dentry freeing.
Lockless lookups only access d_name, d_parent, d_lock, d_op, d_flags (so
the previous content of d_child is not needed if said dentry was unhashed
but still accessed by a CPU because of RCU constraints)
As dentry cache easily contains millions of entries, a size reduction is
worth the extra complexity of the ugly C union.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Maneesh Soni <maneesh@in.ibm.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Ian Kent <raven@themaw.net>
Cc: Paul Jackson <pj@sgi.com>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: James Morris <jmorris@namei.org>
Cc: Stephen Smalley <sds@epoch.ncsc.mil>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 17:03:32 +08:00
|
|
|
struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
|
2005-04-17 06:20:36 +08:00
|
|
|
next = tmp->next;
|
|
|
|
/* Have we found a mount point ? */
|
|
|
|
if (d_mountpoint(dentry))
|
|
|
|
goto positive;
|
|
|
|
if (!list_empty(&dentry->d_subdirs)) {
|
|
|
|
this_parent = dentry;
|
|
|
|
goto repeat;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* All done at this level ... ascend and resume the search.
|
|
|
|
*/
|
|
|
|
if (this_parent != parent) {
|
[PATCH] shrink dentry struct
Some long time ago, dentry struct was carefully tuned so that on 32 bits
UP, sizeof(struct dentry) was exactly 128, ie a power of 2, and a multiple
of memory cache lines.
Then RCU was added and dentry struct enlarged by two pointers, with nice
results for SMP, but not so good on UP, because breaking the above tuning
(128 + 8 = 136 bytes)
This patch reverts this unwanted side effect, by using an union (d_u),
where d_rcu and d_child are placed so that these two fields can share their
memory needs.
At the time d_free() is called (and d_rcu is really used), d_child is known
to be empty and not touched by the dentry freeing.
Lockless lookups only access d_name, d_parent, d_lock, d_op, d_flags (so
the previous content of d_child is not needed if said dentry was unhashed
but still accessed by a CPU because of RCU constraints)
As dentry cache easily contains millions of entries, a size reduction is
worth the extra complexity of the ugly C union.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Maneesh Soni <maneesh@in.ibm.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Ian Kent <raven@themaw.net>
Cc: Paul Jackson <pj@sgi.com>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: James Morris <jmorris@namei.org>
Cc: Stephen Smalley <sds@epoch.ncsc.mil>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 17:03:32 +08:00
|
|
|
next = this_parent->d_u.d_child.next;
|
2005-04-17 06:20:36 +08:00
|
|
|
this_parent = this_parent->d_parent;
|
|
|
|
goto resume;
|
|
|
|
}
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
return 0; /* No mount points found in tree */
|
|
|
|
positive:
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
return 1;
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(have_submounts);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Search the dentry child list for the specified parent,
|
|
|
|
* and move any unused dentries to the end of the unused
|
|
|
|
* list for prune_dcache(). We descend to the next level
|
|
|
|
* whenever the d_subdirs list is non-empty and continue
|
|
|
|
* searching.
|
|
|
|
*
|
|
|
|
* It returns zero iff there are no unused children,
|
|
|
|
* otherwise it returns the number of children moved to
|
|
|
|
* the end of the unused list. This may not be the total
|
|
|
|
* number of unused children, because select_parent can
|
|
|
|
* drop the lock and return early due to latency
|
|
|
|
* constraints.
|
|
|
|
*/
|
|
|
|
static int select_parent(struct dentry * parent)
|
|
|
|
{
|
|
|
|
struct dentry *this_parent = parent;
|
|
|
|
struct list_head *next;
|
|
|
|
int found = 0;
|
|
|
|
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
repeat:
|
|
|
|
next = this_parent->d_subdirs.next;
|
|
|
|
resume:
|
|
|
|
while (next != &this_parent->d_subdirs) {
|
|
|
|
struct list_head *tmp = next;
|
[PATCH] shrink dentry struct
Some long time ago, dentry struct was carefully tuned so that on 32 bits
UP, sizeof(struct dentry) was exactly 128, ie a power of 2, and a multiple
of memory cache lines.
Then RCU was added and dentry struct enlarged by two pointers, with nice
results for SMP, but not so good on UP, because breaking the above tuning
(128 + 8 = 136 bytes)
This patch reverts this unwanted side effect, by using an union (d_u),
where d_rcu and d_child are placed so that these two fields can share their
memory needs.
At the time d_free() is called (and d_rcu is really used), d_child is known
to be empty and not touched by the dentry freeing.
Lockless lookups only access d_name, d_parent, d_lock, d_op, d_flags (so
the previous content of d_child is not needed if said dentry was unhashed
but still accessed by a CPU because of RCU constraints)
As dentry cache easily contains millions of entries, a size reduction is
worth the extra complexity of the ugly C union.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Maneesh Soni <maneesh@in.ibm.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Ian Kent <raven@themaw.net>
Cc: Paul Jackson <pj@sgi.com>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: James Morris <jmorris@namei.org>
Cc: Stephen Smalley <sds@epoch.ncsc.mil>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 17:03:32 +08:00
|
|
|
struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
|
2005-04-17 06:20:36 +08:00
|
|
|
next = tmp->next;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* move only zero ref count dentries to the end
|
|
|
|
* of the unused list for prune_dcache
|
|
|
|
*/
|
|
|
|
if (!atomic_read(&dentry->d_count)) {
|
2010-10-10 17:36:26 +08:00
|
|
|
dentry_lru_move_tail(dentry);
|
2005-04-17 06:20:36 +08:00
|
|
|
found++;
|
2010-10-10 17:36:26 +08:00
|
|
|
} else {
|
|
|
|
dentry_lru_del(dentry);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We can return to the caller if we have found some (this
|
|
|
|
* ensures forward progress). We'll be coming back to find
|
|
|
|
* the rest.
|
|
|
|
*/
|
|
|
|
if (found && need_resched())
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Descend a level if the d_subdirs list is non-empty.
|
|
|
|
*/
|
|
|
|
if (!list_empty(&dentry->d_subdirs)) {
|
|
|
|
this_parent = dentry;
|
|
|
|
goto repeat;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* All done at this level ... ascend and resume the search.
|
|
|
|
*/
|
|
|
|
if (this_parent != parent) {
|
[PATCH] shrink dentry struct
Some long time ago, dentry struct was carefully tuned so that on 32 bits
UP, sizeof(struct dentry) was exactly 128, ie a power of 2, and a multiple
of memory cache lines.
Then RCU was added and dentry struct enlarged by two pointers, with nice
results for SMP, but not so good on UP, because breaking the above tuning
(128 + 8 = 136 bytes)
This patch reverts this unwanted side effect, by using an union (d_u),
where d_rcu and d_child are placed so that these two fields can share their
memory needs.
At the time d_free() is called (and d_rcu is really used), d_child is known
to be empty and not touched by the dentry freeing.
Lockless lookups only access d_name, d_parent, d_lock, d_op, d_flags (so
the previous content of d_child is not needed if said dentry was unhashed
but still accessed by a CPU because of RCU constraints)
As dentry cache easily contains millions of entries, a size reduction is
worth the extra complexity of the ugly C union.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Maneesh Soni <maneesh@in.ibm.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Ian Kent <raven@themaw.net>
Cc: Paul Jackson <pj@sgi.com>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: James Morris <jmorris@namei.org>
Cc: Stephen Smalley <sds@epoch.ncsc.mil>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 17:03:32 +08:00
|
|
|
next = this_parent->d_u.d_child.next;
|
2005-04-17 06:20:36 +08:00
|
|
|
this_parent = this_parent->d_parent;
|
|
|
|
goto resume;
|
|
|
|
}
|
|
|
|
out:
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
return found;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* shrink_dcache_parent - prune dcache
|
|
|
|
* @parent: parent of entries to prune
|
|
|
|
*
|
|
|
|
* Prune the dcache to remove unused children of the parent dentry.
|
|
|
|
*/
|
|
|
|
|
|
|
|
void shrink_dcache_parent(struct dentry * parent)
|
|
|
|
{
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
struct super_block *sb = parent->d_sb;
|
2005-04-17 06:20:36 +08:00
|
|
|
int found;
|
|
|
|
|
|
|
|
while ((found = select_parent(parent)) != 0)
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
__shrink_dcache_sb(sb, &found, 0);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(shrink_dcache_parent);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Scan `nr' dentries and return the number which remain.
|
|
|
|
*
|
|
|
|
* We need to avoid reentering the filesystem if the caller is performing a
|
|
|
|
* GFP_NOFS allocation attempt. One example deadlock is:
|
|
|
|
*
|
|
|
|
* ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
|
|
|
|
* prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
|
|
|
|
* ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
|
|
|
|
*
|
|
|
|
* In this case we return -1 to tell the caller that we baled.
|
|
|
|
*/
|
2010-07-19 12:56:17 +08:00
|
|
|
static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
if (nr) {
|
|
|
|
if (!(gfp_mask & __GFP_FS))
|
|
|
|
return -1;
|
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries
[Summary]
Split LRU-list of unused dentries to one per superblock to avoid soft
lock up during NFS mounts and remounting of any filesystem.
Previously I posted here:
http://lkml.org/lkml/2008/3/5/590
[Descriptions]
- background
dentry_unused is a list of dentries which are not referenced.
dentry_unused grows up when references on directories or files are
released. This list can be very long if there is huge free memory.
- the problem
When shrink_dcache_sb() is called, it scans all dentry_unused linearly
under spin_lock(), and if dentry->d_sb is differnt from given
superblock, scan next dentry. This scan costs very much if there are
many entries, and very ineffective if there are many superblocks.
IOW, When we need to shrink unused dentries on one dentry, but scans
unused dentries on all superblocks in the system. For example, we scan
500 dentries to unmount a filesystem, but scans 1,000,000 or more unused
dentries on other superblocks.
In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink
unused dentries on NFS, but scans 100,000,000 unused dentries on
superblocks in the system such as local ext3 filesystems. I hear NFS
mounting took 1 min on some system in use.
* : NFS uses virtual filesystem in rpc layer, so NFS is affected by
this problem.
100,000,000 is possible number on large systems.
Per-superblock LRU of unused dentried can reduce the cost in
reasonable manner.
- How to fix
I found this problem is solved by David Chinner's "Per-superblock
unused dentry LRU lists V3"(1), so I rebase it and add some fix to
reclaim with fairness, which is in Andrew Morton's comments(2).
1) http://lkml.org/lkml/2006/5/25/318
2) http://lkml.org/lkml/2006/5/25/320
Split LRU-list of unused dentries to each superblocks. Then, NFS
mounting will check dentries under a superblock instead of all. But
this spliting will break LRU of dentry-unused. So, I've attempted to
make reclaim unused dentrins with fairness by calculate number of
dentries to scan on this sb based on following way
number of dentries to scan on this sb =
count * (number of dentries on this sb / number of dentries in the machine)
- ToDo
- I have to measuring performance number and do stress tests.
- When unmount occurs during prune_dcache(), scanning on same
superblock, It is unable to reach next superblock because it is gone
away. We restart scannig superblock from first one, it causes
unfairness of reclaim unused dentries on first superblock. But I think
this happens very rarely.
- Test Results
Result on 6GB boxes with excessive unused dentries.
Without patch:
$ cat /proc/sys/fs/dentry-state
10181835 10180203 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m1.830s
user 0m0.001s
sys 0m1.653s
With this patch:
$ cat /proc/sys/fs/dentry-state
10236610 10234751 45 0 0 0
# mount -t nfs 10.124.60.70:/work/kernel-src nfs
real 0m0.106s
user 0m0.002s
sys 0m0.032s
[akpm@linux-foundation.org: fix comments]
Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: David Chinner <dgc@sgi.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
|
|
|
prune_dcache(nr);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2010-10-10 17:36:23 +08:00
|
|
|
|
2011-01-07 14:49:18 +08:00
|
|
|
return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2007-07-17 19:03:17 +08:00
|
|
|
static struct shrinker dcache_shrinker = {
|
|
|
|
.shrink = shrink_dcache_memory,
|
|
|
|
.seeks = DEFAULT_SEEKS,
|
|
|
|
};
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
|
|
|
* d_alloc - allocate a dcache entry
|
|
|
|
* @parent: parent of entry to allocate
|
|
|
|
* @name: qstr of the name
|
|
|
|
*
|
|
|
|
* Allocates a dentry. It returns %NULL if there is insufficient memory
|
|
|
|
* available. On a success the dentry is returned. The name passed in is
|
|
|
|
* copied and the copy passed in may be reused after this call.
|
|
|
|
*/
|
|
|
|
|
|
|
|
struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
|
|
|
|
{
|
|
|
|
struct dentry *dentry;
|
|
|
|
char *dname;
|
|
|
|
|
2007-10-16 16:25:52 +08:00
|
|
|
dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!dentry)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
if (name->len > DNAME_INLINE_LEN-1) {
|
|
|
|
dname = kmalloc(name->len + 1, GFP_KERNEL);
|
|
|
|
if (!dname) {
|
|
|
|
kmem_cache_free(dentry_cache, dentry);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
dname = dentry->d_iname;
|
|
|
|
}
|
|
|
|
dentry->d_name.name = dname;
|
|
|
|
|
|
|
|
dentry->d_name.len = name->len;
|
|
|
|
dentry->d_name.hash = name->hash;
|
|
|
|
memcpy(dname, name->name, name->len);
|
|
|
|
dname[name->len] = 0;
|
|
|
|
|
|
|
|
atomic_set(&dentry->d_count, 1);
|
|
|
|
dentry->d_flags = DCACHE_UNHASHED;
|
|
|
|
spin_lock_init(&dentry->d_lock);
|
|
|
|
dentry->d_inode = NULL;
|
|
|
|
dentry->d_parent = NULL;
|
|
|
|
dentry->d_sb = NULL;
|
|
|
|
dentry->d_op = NULL;
|
|
|
|
dentry->d_fsdata = NULL;
|
|
|
|
dentry->d_mounted = 0;
|
|
|
|
INIT_HLIST_NODE(&dentry->d_hash);
|
|
|
|
INIT_LIST_HEAD(&dentry->d_lru);
|
|
|
|
INIT_LIST_HEAD(&dentry->d_subdirs);
|
|
|
|
INIT_LIST_HEAD(&dentry->d_alias);
|
|
|
|
|
|
|
|
if (parent) {
|
|
|
|
dentry->d_parent = dget(parent);
|
|
|
|
dentry->d_sb = parent->d_sb;
|
|
|
|
} else {
|
[PATCH] shrink dentry struct
Some long time ago, dentry struct was carefully tuned so that on 32 bits
UP, sizeof(struct dentry) was exactly 128, ie a power of 2, and a multiple
of memory cache lines.
Then RCU was added and dentry struct enlarged by two pointers, with nice
results for SMP, but not so good on UP, because breaking the above tuning
(128 + 8 = 136 bytes)
This patch reverts this unwanted side effect, by using an union (d_u),
where d_rcu and d_child are placed so that these two fields can share their
memory needs.
At the time d_free() is called (and d_rcu is really used), d_child is known
to be empty and not touched by the dentry freeing.
Lockless lookups only access d_name, d_parent, d_lock, d_op, d_flags (so
the previous content of d_child is not needed if said dentry was unhashed
but still accessed by a CPU because of RCU constraints)
As dentry cache easily contains millions of entries, a size reduction is
worth the extra complexity of the ugly C union.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Maneesh Soni <maneesh@in.ibm.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Ian Kent <raven@themaw.net>
Cc: Paul Jackson <pj@sgi.com>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: James Morris <jmorris@namei.org>
Cc: Stephen Smalley <sds@epoch.ncsc.mil>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 17:03:32 +08:00
|
|
|
INIT_LIST_HEAD(&dentry->d_u.d_child);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
if (parent)
|
[PATCH] shrink dentry struct
Some long time ago, dentry struct was carefully tuned so that on 32 bits
UP, sizeof(struct dentry) was exactly 128, ie a power of 2, and a multiple
of memory cache lines.
Then RCU was added and dentry struct enlarged by two pointers, with nice
results for SMP, but not so good on UP, because breaking the above tuning
(128 + 8 = 136 bytes)
This patch reverts this unwanted side effect, by using an union (d_u),
where d_rcu and d_child are placed so that these two fields can share their
memory needs.
At the time d_free() is called (and d_rcu is really used), d_child is known
to be empty and not touched by the dentry freeing.
Lockless lookups only access d_name, d_parent, d_lock, d_op, d_flags (so
the previous content of d_child is not needed if said dentry was unhashed
but still accessed by a CPU because of RCU constraints)
As dentry cache easily contains millions of entries, a size reduction is
worth the extra complexity of the ugly C union.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Maneesh Soni <maneesh@in.ibm.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Ian Kent <raven@themaw.net>
Cc: Paul Jackson <pj@sgi.com>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: James Morris <jmorris@namei.org>
Cc: Stephen Smalley <sds@epoch.ncsc.mil>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 17:03:32 +08:00
|
|
|
list_add(&dentry->d_u.d_child, &parent->d_subdirs);
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
|
fs: use fast counters for vfs caches
percpu_counter library generates quite nasty code, so unless you need
to dynamically allocate counters or take fast approximate value, a
simple per cpu set of counters is much better.
The percpu_counter can never be made to work as well, because it has an
indirection from pointer to percpu memory, and it can't use direct
this_cpu_inc interfaces because it doesn't use static PER_CPU data, so
code will always be worse.
In the fastpath, it is the difference between this:
incl %gs:nr_dentry # nr_dentry
and this:
movl percpu_counter_batch(%rip), %edx # percpu_counter_batch,
movl $1, %esi #,
movq $nr_dentry, %rdi #,
call __percpu_counter_add # (plus I clobber registers)
__percpu_counter_add:
pushq %rbp #
movq %rsp, %rbp #,
subq $32, %rsp #,
movq %rbx, -24(%rbp) #,
movq %r12, -16(%rbp) #,
movq %r13, -8(%rbp) #,
movq %rdi, %rbx # fbc, fbc
#APP
# 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1
movq %gs:kernel_stack,%rax #, pfo_ret__
# 0 "" 2
#NO_APP
incl -8124(%rax) # <variable>.preempt_count
movq 32(%rdi), %r12 # <variable>.counters, tcp_ptr__
#APP
# 78 "lib/percpu_counter.c" 1
add %gs:this_cpu_off, %r12 # this_cpu_off, tcp_ptr__
# 0 "" 2
#NO_APP
movslq (%r12),%r13 #* tcp_ptr__, tmp73
movslq %edx,%rax # batch, batch
addq %rsi, %r13 # amount, count
cmpq %rax, %r13 # batch, count
jge .L27 #,
negl %edx # tmp76
movslq %edx,%rdx # tmp76, tmp77
cmpq %rdx, %r13 # tmp77, count
jg .L28 #,
.L27:
movq %rbx, %rdi # fbc,
call _raw_spin_lock #
addq %r13, 8(%rbx) # count, <variable>.count
movq %rbx, %rdi # fbc,
movl $0, (%r12) #,* tcp_ptr__
call _raw_spin_unlock #
.L29:
#APP
# 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1
movq %gs:kernel_stack,%rax #, pfo_ret__
# 0 "" 2
#NO_APP
decl -8124(%rax) # <variable>.preempt_count
movq -8136(%rax), %rax #, D.14625
testb $8, %al #, D.14625
jne .L32 #,
.L31:
movq -24(%rbp), %rbx #,
movq -16(%rbp), %r12 #,
movq -8(%rbp), %r13 #,
leave
ret
.p2align 4,,10
.p2align 3
.L28:
movl %r13d, (%r12) # count,*
jmp .L29 #
.L32:
call preempt_schedule #
.p2align 4,,6
jmp .L31 #
.size __percpu_counter_add, .-__percpu_counter_add
.p2align 4,,15
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:19 +08:00
|
|
|
this_cpu_inc(nr_dentry);
|
2010-10-10 17:36:23 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
return dentry;
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(d_alloc);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
struct dentry *d_alloc_name(struct dentry *parent, const char *name)
|
|
|
|
{
|
|
|
|
struct qstr q;
|
|
|
|
|
|
|
|
q.name = name;
|
|
|
|
q.len = strlen(name);
|
|
|
|
q.hash = full_name_hash(q.name, q.len);
|
|
|
|
return d_alloc(parent, &q);
|
|
|
|
}
|
2009-09-30 08:09:42 +08:00
|
|
|
EXPORT_SYMBOL(d_alloc_name);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2008-10-16 06:50:28 +08:00
|
|
|
/* the caller must hold dcache_lock */
|
|
|
|
static void __d_instantiate(struct dentry *dentry, struct inode *inode)
|
|
|
|
{
|
|
|
|
if (inode)
|
|
|
|
list_add(&dentry->d_alias, &inode->i_dentry);
|
|
|
|
dentry->d_inode = inode;
|
|
|
|
fsnotify_d_instantiate(dentry, inode);
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
|
|
|
* d_instantiate - fill in inode information for a dentry
|
|
|
|
* @entry: dentry to complete
|
|
|
|
* @inode: inode to attach to this dentry
|
|
|
|
*
|
|
|
|
* Fill in inode information in the entry.
|
|
|
|
*
|
|
|
|
* This turns negative dentries into productive full members
|
|
|
|
* of society.
|
|
|
|
*
|
|
|
|
* NOTE! This assumes that the inode count has been incremented
|
|
|
|
* (or otherwise set) by the caller to indicate that it is now
|
|
|
|
* in use by the dcache.
|
|
|
|
*/
|
|
|
|
|
|
|
|
void d_instantiate(struct dentry *entry, struct inode * inode)
|
|
|
|
{
|
2006-03-27 00:25:39 +08:00
|
|
|
BUG_ON(!list_empty(&entry->d_alias));
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_lock(&dcache_lock);
|
2008-10-16 06:50:28 +08:00
|
|
|
__d_instantiate(entry, inode);
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
security_d_instantiate(entry, inode);
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(d_instantiate);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/**
|
|
|
|
* d_instantiate_unique - instantiate a non-aliased dentry
|
|
|
|
* @entry: dentry to instantiate
|
|
|
|
* @inode: inode to attach to this dentry
|
|
|
|
*
|
|
|
|
* Fill in inode information in the entry. On success, it returns NULL.
|
|
|
|
* If an unhashed alias of "entry" already exists, then we return the
|
2006-01-10 12:52:51 +08:00
|
|
|
* aliased dentry instead and drop one reference to inode.
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
|
|
|
* Note that in order to avoid conflicts with rename() etc, the caller
|
|
|
|
* had better be holding the parent directory semaphore.
|
2006-01-10 12:52:51 +08:00
|
|
|
*
|
|
|
|
* This also assumes that the inode count has been incremented
|
|
|
|
* (or otherwise set) by the caller to indicate that it is now
|
|
|
|
* in use by the dcache.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2006-08-23 08:06:07 +08:00
|
|
|
static struct dentry *__d_instantiate_unique(struct dentry *entry,
|
|
|
|
struct inode *inode)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct dentry *alias;
|
|
|
|
int len = entry->d_name.len;
|
|
|
|
const char *name = entry->d_name.name;
|
|
|
|
unsigned int hash = entry->d_name.hash;
|
|
|
|
|
2006-08-23 08:06:07 +08:00
|
|
|
if (!inode) {
|
2008-10-16 06:50:28 +08:00
|
|
|
__d_instantiate(entry, NULL);
|
2006-08-23 08:06:07 +08:00
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
list_for_each_entry(alias, &inode->i_dentry, d_alias) {
|
|
|
|
struct qstr *qstr = &alias->d_name;
|
|
|
|
|
|
|
|
if (qstr->hash != hash)
|
|
|
|
continue;
|
|
|
|
if (alias->d_parent != entry->d_parent)
|
|
|
|
continue;
|
|
|
|
if (qstr->len != len)
|
|
|
|
continue;
|
|
|
|
if (memcmp(qstr->name, name, len))
|
|
|
|
continue;
|
|
|
|
dget_locked(alias);
|
|
|
|
return alias;
|
|
|
|
}
|
2006-08-23 08:06:07 +08:00
|
|
|
|
2008-10-16 06:50:28 +08:00
|
|
|
__d_instantiate(entry, inode);
|
2005-04-17 06:20:36 +08:00
|
|
|
return NULL;
|
|
|
|
}
|
2006-08-23 08:06:07 +08:00
|
|
|
|
|
|
|
struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
|
|
|
|
{
|
|
|
|
struct dentry *result;
|
|
|
|
|
|
|
|
BUG_ON(!list_empty(&entry->d_alias));
|
|
|
|
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
result = __d_instantiate_unique(entry, inode);
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
|
|
|
|
if (!result) {
|
|
|
|
security_d_instantiate(entry, inode);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
BUG_ON(!d_unhashed(result));
|
|
|
|
iput(inode);
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
EXPORT_SYMBOL(d_instantiate_unique);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* d_alloc_root - allocate root dentry
|
|
|
|
* @root_inode: inode to allocate the root for
|
|
|
|
*
|
|
|
|
* Allocate a root ("/") dentry for the inode given. The inode is
|
|
|
|
* instantiated and returned. %NULL is returned if there is insufficient
|
|
|
|
* memory or the inode passed is %NULL.
|
|
|
|
*/
|
|
|
|
|
|
|
|
struct dentry * d_alloc_root(struct inode * root_inode)
|
|
|
|
{
|
|
|
|
struct dentry *res = NULL;
|
|
|
|
|
|
|
|
if (root_inode) {
|
|
|
|
static const struct qstr name = { .name = "/", .len = 1 };
|
|
|
|
|
|
|
|
res = d_alloc(NULL, &name);
|
|
|
|
if (res) {
|
|
|
|
res->d_sb = root_inode->i_sb;
|
|
|
|
res->d_parent = res;
|
|
|
|
d_instantiate(res, root_inode);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return res;
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(d_alloc_root);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
static inline struct hlist_head *d_hash(struct dentry *parent,
|
|
|
|
unsigned long hash)
|
|
|
|
{
|
|
|
|
hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
|
|
|
|
hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
|
|
|
|
return dentry_hashtable + (hash & D_HASHMASK);
|
|
|
|
}
|
|
|
|
|
2008-08-11 21:48:57 +08:00
|
|
|
/**
|
|
|
|
* d_obtain_alias - find or allocate a dentry for a given inode
|
|
|
|
* @inode: inode to allocate the dentry for
|
|
|
|
*
|
|
|
|
* Obtain a dentry for an inode resulting from NFS filehandle conversion or
|
|
|
|
* similar open by handle operations. The returned dentry may be anonymous,
|
|
|
|
* or may have a full name (if the inode was already in the cache).
|
|
|
|
*
|
|
|
|
* When called on a directory inode, we must ensure that the inode only ever
|
|
|
|
* has one dentry. If a dentry is found, that is returned instead of
|
|
|
|
* allocating a new one.
|
|
|
|
*
|
|
|
|
* On successful return, the reference to the inode has been transferred
|
2008-08-11 21:49:04 +08:00
|
|
|
* to the dentry. In case of an error the reference on the inode is released.
|
|
|
|
* To make it easier to use in export operations a %NULL or IS_ERR inode may
|
|
|
|
* be passed in and will be the error will be propagate to the return value,
|
|
|
|
* with a %NULL @inode replaced by ERR_PTR(-ESTALE).
|
2008-08-11 21:48:57 +08:00
|
|
|
*/
|
|
|
|
struct dentry *d_obtain_alias(struct inode *inode)
|
|
|
|
{
|
2008-08-11 21:49:12 +08:00
|
|
|
static const struct qstr anonstring = { .name = "" };
|
|
|
|
struct dentry *tmp;
|
|
|
|
struct dentry *res;
|
2008-08-11 21:48:57 +08:00
|
|
|
|
|
|
|
if (!inode)
|
2008-08-11 21:49:04 +08:00
|
|
|
return ERR_PTR(-ESTALE);
|
2008-08-11 21:48:57 +08:00
|
|
|
if (IS_ERR(inode))
|
|
|
|
return ERR_CAST(inode);
|
|
|
|
|
2008-08-11 21:49:12 +08:00
|
|
|
res = d_find_alias(inode);
|
|
|
|
if (res)
|
|
|
|
goto out_iput;
|
|
|
|
|
|
|
|
tmp = d_alloc(NULL, &anonstring);
|
|
|
|
if (!tmp) {
|
|
|
|
res = ERR_PTR(-ENOMEM);
|
|
|
|
goto out_iput;
|
2008-08-11 21:48:57 +08:00
|
|
|
}
|
2008-08-11 21:49:12 +08:00
|
|
|
tmp->d_parent = tmp; /* make sure dput doesn't croak */
|
|
|
|
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
res = __d_find_alias(inode, 0);
|
|
|
|
if (res) {
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
dput(tmp);
|
|
|
|
goto out_iput;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* attach a disconnected dentry */
|
|
|
|
spin_lock(&tmp->d_lock);
|
|
|
|
tmp->d_sb = inode->i_sb;
|
|
|
|
tmp->d_inode = inode;
|
|
|
|
tmp->d_flags |= DCACHE_DISCONNECTED;
|
|
|
|
tmp->d_flags &= ~DCACHE_UNHASHED;
|
|
|
|
list_add(&tmp->d_alias, &inode->i_dentry);
|
|
|
|
hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon);
|
|
|
|
spin_unlock(&tmp->d_lock);
|
|
|
|
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
return tmp;
|
|
|
|
|
|
|
|
out_iput:
|
|
|
|
iput(inode);
|
|
|
|
return res;
|
2008-08-11 21:48:57 +08:00
|
|
|
}
|
2009-02-28 06:02:59 +08:00
|
|
|
EXPORT_SYMBOL(d_obtain_alias);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/**
|
|
|
|
* d_splice_alias - splice a disconnected dentry into the tree if one exists
|
|
|
|
* @inode: the inode which may have a disconnected dentry
|
|
|
|
* @dentry: a negative dentry which we want to point to the inode.
|
|
|
|
*
|
|
|
|
* If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
|
|
|
|
* DCACHE_DISCONNECTED), then d_move that in place of the given dentry
|
|
|
|
* and return it, else simply d_add the inode to the dentry and return NULL.
|
|
|
|
*
|
|
|
|
* This is needed in the lookup routine of any filesystem that is exportable
|
|
|
|
* (via knfsd) so that we can build dcache paths to directories effectively.
|
|
|
|
*
|
|
|
|
* If a dentry was found and moved, then it is returned. Otherwise NULL
|
|
|
|
* is returned. This matches the expected return value of ->lookup.
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
|
|
|
|
{
|
|
|
|
struct dentry *new = NULL;
|
|
|
|
|
[PATCH] knfsd: close a race-opportunity in d_splice_alias
There is a possible race in d_splice_alias. Though __d_find_alias(inode, 1)
will only return a dentry with DCACHE_DISCONNECTED set, it is possible for it
to get cleared before the BUG_ON, and it is is not possible to lock against
that.
There are a couple of problems here. Firstly, the code doesn't match the
comment. The comment describes a 'disconnected' dentry as being IS_ROOT as
well as DCACHE_DISCONNECTED, however there is not testing of IS_ROOT anythere.
A dentry is marked DCACHE_DISCONNECTED when allocated with d_alloc_anon, and
remains DCACHE_DISCONNECTED while a path is built up towards the root. So a
dentry can have a valid name and a valid parent and even grandparent, but will
still be DCACHE_DISCONNECTED until a path to the root is created. Once the
path to the root is complete, everything in the path gets DCACHE_DISCONNECTED
cleared. So the fact that DCACHE_DISCONNECTED isn't enough to say that a
dentry is free to be spliced in with a given name. This can only be allowed
if the dentry does not yet have a name, so the IS_ROOT test is needed too.
However even adding that test to __d_find_alias isn't enough. As
d_splice_alias drops dcache_lock before calling d_move to perform the splice,
it could race with another thread calling d_splice_alias to splice the inode
in with a different name in a different part of the tree (in the case where a
file has hard links). So that splicing code is only really safe for
directories (as we know that directories only have one link). For
directories, the caller of d_splice_alias will be holding i_mutex on the
(unique) parent so there is no room for a race.
A consequence of this is that a non-directory will never benefit from being
spliced into a pre-exisiting dentry, but that isn't a problem. It is
perfectly OK for a non-directory to have multiple dentries, some anonymous,
some not. And the comment for d_splice_alias says that it only happens for
directories anyway.
Signed-off-by: Neil Brown <neilb@suse.de>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-04 17:16:16 +08:00
|
|
|
if (inode && S_ISDIR(inode->i_mode)) {
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
new = __d_find_alias(inode, 1);
|
|
|
|
if (new) {
|
|
|
|
BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
security_d_instantiate(new, inode);
|
|
|
|
d_move(new, dentry);
|
|
|
|
iput(inode);
|
|
|
|
} else {
|
2008-10-16 06:50:28 +08:00
|
|
|
/* already taking dcache_lock, so d_add() by hand */
|
|
|
|
__d_instantiate(dentry, inode);
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
security_d_instantiate(dentry, inode);
|
|
|
|
d_rehash(dentry);
|
|
|
|
}
|
|
|
|
} else
|
|
|
|
d_add(dentry, inode);
|
|
|
|
return new;
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(d_splice_alias);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2008-05-21 14:50:46 +08:00
|
|
|
/**
|
|
|
|
* d_add_ci - lookup or allocate new dentry with case-exact name
|
|
|
|
* @inode: the inode case-insensitive lookup has found
|
|
|
|
* @dentry: the negative dentry that was passed to the parent's lookup func
|
|
|
|
* @name: the case-exact name to be associated with the returned dentry
|
|
|
|
*
|
|
|
|
* This is to avoid filling the dcache with case-insensitive names to the
|
|
|
|
* same inode, only the actual correct case is stored in the dcache for
|
|
|
|
* case-insensitive filesystems.
|
|
|
|
*
|
|
|
|
* For a case-insensitive lookup match and if the the case-exact dentry
|
|
|
|
* already exists in in the dcache, use it and return it.
|
|
|
|
*
|
|
|
|
* If no entry exists with the exact case name, allocate new dentry with
|
|
|
|
* the exact case, and return the spliced entry.
|
|
|
|
*/
|
2008-08-08 05:49:07 +08:00
|
|
|
struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
|
2008-05-21 14:50:46 +08:00
|
|
|
struct qstr *name)
|
|
|
|
{
|
|
|
|
int error;
|
|
|
|
struct dentry *found;
|
|
|
|
struct dentry *new;
|
|
|
|
|
2009-01-06 02:10:37 +08:00
|
|
|
/*
|
|
|
|
* First check if a dentry matching the name already exists,
|
|
|
|
* if not go ahead and create it now.
|
|
|
|
*/
|
2008-05-21 14:50:46 +08:00
|
|
|
found = d_hash_and_lookup(dentry->d_parent, name);
|
|
|
|
if (!found) {
|
|
|
|
new = d_alloc(dentry->d_parent, name);
|
|
|
|
if (!new) {
|
|
|
|
error = -ENOMEM;
|
|
|
|
goto err_out;
|
|
|
|
}
|
2009-01-06 02:10:37 +08:00
|
|
|
|
2008-05-21 14:50:46 +08:00
|
|
|
found = d_splice_alias(inode, new);
|
|
|
|
if (found) {
|
|
|
|
dput(new);
|
|
|
|
return found;
|
|
|
|
}
|
|
|
|
return new;
|
|
|
|
}
|
2009-01-06 02:10:37 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* If a matching dentry exists, and it's not negative use it.
|
|
|
|
*
|
|
|
|
* Decrement the reference count to balance the iget() done
|
|
|
|
* earlier on.
|
|
|
|
*/
|
2008-05-21 14:50:46 +08:00
|
|
|
if (found->d_inode) {
|
|
|
|
if (unlikely(found->d_inode != inode)) {
|
|
|
|
/* This can't happen because bad inodes are unhashed. */
|
|
|
|
BUG_ON(!is_bad_inode(inode));
|
|
|
|
BUG_ON(!is_bad_inode(found->d_inode));
|
|
|
|
}
|
|
|
|
iput(inode);
|
|
|
|
return found;
|
|
|
|
}
|
2009-01-06 02:10:37 +08:00
|
|
|
|
2008-05-21 14:50:46 +08:00
|
|
|
/*
|
|
|
|
* Negative dentry: instantiate it unless the inode is a directory and
|
2009-01-06 02:10:37 +08:00
|
|
|
* already has a dentry.
|
2008-05-21 14:50:46 +08:00
|
|
|
*/
|
|
|
|
spin_lock(&dcache_lock);
|
2009-01-06 02:10:37 +08:00
|
|
|
if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) {
|
2008-10-16 06:50:28 +08:00
|
|
|
__d_instantiate(found, inode);
|
2008-05-21 14:50:46 +08:00
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
security_d_instantiate(found, inode);
|
|
|
|
return found;
|
|
|
|
}
|
2009-01-06 02:10:37 +08:00
|
|
|
|
2008-05-21 14:50:46 +08:00
|
|
|
/*
|
2009-01-06 02:10:37 +08:00
|
|
|
* In case a directory already has a (disconnected) entry grab a
|
|
|
|
* reference to it, move it in place and use it.
|
2008-05-21 14:50:46 +08:00
|
|
|
*/
|
|
|
|
new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
|
|
|
|
dget_locked(new);
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
security_d_instantiate(found, inode);
|
|
|
|
d_move(new, found);
|
|
|
|
iput(inode);
|
|
|
|
dput(found);
|
|
|
|
return new;
|
|
|
|
|
|
|
|
err_out:
|
|
|
|
iput(inode);
|
|
|
|
return ERR_PTR(error);
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(d_add_ci);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/**
|
|
|
|
* d_lookup - search for a dentry
|
|
|
|
* @parent: parent dentry
|
|
|
|
* @name: qstr of name we wish to find
|
fs: remove extra lookup in __lookup_hash
fs: remove extra lookup in __lookup_hash
Optimize lookup for create operations, where no dentry should often be
common-case. In cases where it is not, such as unlink, the added overhead
is much smaller than the removed.
Also, move comments about __d_lookup racyness to the __d_lookup call site.
d_lookup is intuitive; __d_lookup is what needs commenting. So in that same
vein, add kerneldoc comments to __d_lookup and clean up some of the comments:
- We are interested in how the RCU lookup works here, particularly with
renames. Make that explicit, and point to the document where it is explained
in more detail.
- RCU is pretty standard now, and macros make implementations pretty mindless.
If we want to know about RCU barrier details, we look in RCU code.
- Delete some boring legacy comments because we don't care much about how the
code used to work, more about the interesting parts of how it works now. So
comments about lazy LRU may be interesting, but would better be done in the
LRU or refcount management code.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2010-08-18 02:37:34 +08:00
|
|
|
* Returns: dentry, or NULL
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
fs: remove extra lookup in __lookup_hash
fs: remove extra lookup in __lookup_hash
Optimize lookup for create operations, where no dentry should often be
common-case. In cases where it is not, such as unlink, the added overhead
is much smaller than the removed.
Also, move comments about __d_lookup racyness to the __d_lookup call site.
d_lookup is intuitive; __d_lookup is what needs commenting. So in that same
vein, add kerneldoc comments to __d_lookup and clean up some of the comments:
- We are interested in how the RCU lookup works here, particularly with
renames. Make that explicit, and point to the document where it is explained
in more detail.
- RCU is pretty standard now, and macros make implementations pretty mindless.
If we want to know about RCU barrier details, we look in RCU code.
- Delete some boring legacy comments because we don't care much about how the
code used to work, more about the interesting parts of how it works now. So
comments about lazy LRU may be interesting, but would better be done in the
LRU or refcount management code.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2010-08-18 02:37:34 +08:00
|
|
|
* d_lookup searches the children of the parent dentry for the name in
|
|
|
|
* question. If the dentry is found its reference count is incremented and the
|
|
|
|
* dentry is returned. The caller must use dput to free the entry when it has
|
|
|
|
* finished using it. %NULL is returned if the dentry does not exist.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
|
|
|
struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
|
|
|
|
{
|
|
|
|
struct dentry * dentry = NULL;
|
|
|
|
unsigned long seq;
|
|
|
|
|
|
|
|
do {
|
|
|
|
seq = read_seqbegin(&rename_lock);
|
|
|
|
dentry = __d_lookup(parent, name);
|
|
|
|
if (dentry)
|
|
|
|
break;
|
|
|
|
} while (read_seqretry(&rename_lock, seq));
|
|
|
|
return dentry;
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(d_lookup);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
fs: remove extra lookup in __lookup_hash
fs: remove extra lookup in __lookup_hash
Optimize lookup for create operations, where no dentry should often be
common-case. In cases where it is not, such as unlink, the added overhead
is much smaller than the removed.
Also, move comments about __d_lookup racyness to the __d_lookup call site.
d_lookup is intuitive; __d_lookup is what needs commenting. So in that same
vein, add kerneldoc comments to __d_lookup and clean up some of the comments:
- We are interested in how the RCU lookup works here, particularly with
renames. Make that explicit, and point to the document where it is explained
in more detail.
- RCU is pretty standard now, and macros make implementations pretty mindless.
If we want to know about RCU barrier details, we look in RCU code.
- Delete some boring legacy comments because we don't care much about how the
code used to work, more about the interesting parts of how it works now. So
comments about lazy LRU may be interesting, but would better be done in the
LRU or refcount management code.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2010-08-18 02:37:34 +08:00
|
|
|
/*
|
|
|
|
* __d_lookup - search for a dentry (racy)
|
|
|
|
* @parent: parent dentry
|
|
|
|
* @name: qstr of name we wish to find
|
|
|
|
* Returns: dentry, or NULL
|
|
|
|
*
|
|
|
|
* __d_lookup is like d_lookup, however it may (rarely) return a
|
|
|
|
* false-negative result due to unrelated rename activity.
|
|
|
|
*
|
|
|
|
* __d_lookup is slightly faster by avoiding rename_lock read seqlock,
|
|
|
|
* however it must be used carefully, eg. with a following d_lookup in
|
|
|
|
* the case of failure.
|
|
|
|
*
|
|
|
|
* __d_lookup callers must be commented.
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
|
|
|
|
{
|
|
|
|
unsigned int len = name->len;
|
|
|
|
unsigned int hash = name->hash;
|
|
|
|
const unsigned char *str = name->name;
|
|
|
|
struct hlist_head *head = d_hash(parent,hash);
|
|
|
|
struct dentry *found = NULL;
|
|
|
|
struct hlist_node *node;
|
2005-11-07 16:59:17 +08:00
|
|
|
struct dentry *dentry;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
fs: remove extra lookup in __lookup_hash
fs: remove extra lookup in __lookup_hash
Optimize lookup for create operations, where no dentry should often be
common-case. In cases where it is not, such as unlink, the added overhead
is much smaller than the removed.
Also, move comments about __d_lookup racyness to the __d_lookup call site.
d_lookup is intuitive; __d_lookup is what needs commenting. So in that same
vein, add kerneldoc comments to __d_lookup and clean up some of the comments:
- We are interested in how the RCU lookup works here, particularly with
renames. Make that explicit, and point to the document where it is explained
in more detail.
- RCU is pretty standard now, and macros make implementations pretty mindless.
If we want to know about RCU barrier details, we look in RCU code.
- Delete some boring legacy comments because we don't care much about how the
code used to work, more about the interesting parts of how it works now. So
comments about lazy LRU may be interesting, but would better be done in the
LRU or refcount management code.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2010-08-18 02:37:34 +08:00
|
|
|
/*
|
|
|
|
* The hash list is protected using RCU.
|
|
|
|
*
|
|
|
|
* Take d_lock when comparing a candidate dentry, to avoid races
|
|
|
|
* with d_move().
|
|
|
|
*
|
|
|
|
* It is possible that concurrent renames can mess up our list
|
|
|
|
* walk here and result in missing our dentry, resulting in the
|
|
|
|
* false-negative result. d_lookup() protects against concurrent
|
|
|
|
* renames using rename_lock seqlock.
|
|
|
|
*
|
|
|
|
* See Documentation/vfs/dcache-locking.txt for more details.
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
rcu_read_lock();
|
|
|
|
|
2005-11-07 16:59:17 +08:00
|
|
|
hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
|
2005-04-17 06:20:36 +08:00
|
|
|
struct qstr *qstr;
|
|
|
|
|
|
|
|
if (dentry->d_name.hash != hash)
|
|
|
|
continue;
|
|
|
|
if (dentry->d_parent != parent)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
spin_lock(&dentry->d_lock);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Recheck the dentry after taking the lock - d_move may have
|
fs: remove extra lookup in __lookup_hash
fs: remove extra lookup in __lookup_hash
Optimize lookup for create operations, where no dentry should often be
common-case. In cases where it is not, such as unlink, the added overhead
is much smaller than the removed.
Also, move comments about __d_lookup racyness to the __d_lookup call site.
d_lookup is intuitive; __d_lookup is what needs commenting. So in that same
vein, add kerneldoc comments to __d_lookup and clean up some of the comments:
- We are interested in how the RCU lookup works here, particularly with
renames. Make that explicit, and point to the document where it is explained
in more detail.
- RCU is pretty standard now, and macros make implementations pretty mindless.
If we want to know about RCU barrier details, we look in RCU code.
- Delete some boring legacy comments because we don't care much about how the
code used to work, more about the interesting parts of how it works now. So
comments about lazy LRU may be interesting, but would better be done in the
LRU or refcount management code.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2010-08-18 02:37:34 +08:00
|
|
|
* changed things. Don't bother checking the hash because
|
|
|
|
* we're about to compare the whole name anyway.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
|
|
|
if (dentry->d_parent != parent)
|
|
|
|
goto next;
|
|
|
|
|
Fix NULL pointer dereference in proc_sys_compare
The VFS interface for the 'd_compare()' is a bit special (read: 'odd'),
because it really just essentially replaces a memcmp(). The filesystem
is supposed to just compare the two names with whatever case-independent
or other function.
And when I say 'is supposed to', I obviously mean that 'procfs does odd
things, and actually looks at the dentry that we don't even pass down,
rather than just the name'. Which results in problems, because we
actually call d_compare before we have even verified that the dentry is
still hashed at all.
And that causes a problm since the inode that procfs looks at may have
been free'd and the d_inode pointer is NULL. procfs just assumes that
all dentries are positive, since procfs itself never generates a
negative one. But memory pressure will still result in the dentry
getting torn down, and as it is removed by RCU, it still remains visible
on some lists - and to d_compare.
If the filesystem just did a name comparison, we wouldn't care. And we
could just fix procfs to know about negative dentries too. But rather
than have the low-level filesystems know about internal VFS details,
just move the check for a unhashed dentry up a bit, so that we will only
call d_compare on dentries that are still active.
The actual oops this caused didn't look like a NULL pointer dereference
because procfs did a 'container_of(inode, struct proc_inode, vfs_inode)'
to get at its internal proc_inode information from the inode pointer,
and accessed a field below the inode. So the oops would look something
like
BUG: unable to handle kernel paging request at fffffffffffffff0
IP: [<ffffffff802bc6c6>] proc_sys_compare+0x36/0x50
and was seen on both x86-64 (Alexey Dobriyan and Hugh Dickins) and
ppc64 (Hugh Dickins).
Reported-by: Alexey Dobriyan <adobriyan@gmail.com>
Acked-by: Hugh Dickins <hugh@veritas.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Reviewed-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-of-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-09-29 22:42:57 +08:00
|
|
|
/* non-existing due to RCU? */
|
|
|
|
if (d_unhashed(dentry))
|
|
|
|
goto next;
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* It is safe to compare names since d_move() cannot
|
|
|
|
* change the qstr (protected by d_lock).
|
|
|
|
*/
|
|
|
|
qstr = &dentry->d_name;
|
|
|
|
if (parent->d_op && parent->d_op->d_compare) {
|
2011-01-07 14:49:27 +08:00
|
|
|
if (parent->d_op->d_compare(parent, parent->d_inode,
|
|
|
|
dentry, dentry->d_inode,
|
|
|
|
qstr->len, qstr->name, name))
|
2005-04-17 06:20:36 +08:00
|
|
|
goto next;
|
|
|
|
} else {
|
|
|
|
if (qstr->len != len)
|
|
|
|
goto next;
|
|
|
|
if (memcmp(qstr->name, str, len))
|
|
|
|
goto next;
|
|
|
|
}
|
|
|
|
|
Fix NULL pointer dereference in proc_sys_compare
The VFS interface for the 'd_compare()' is a bit special (read: 'odd'),
because it really just essentially replaces a memcmp(). The filesystem
is supposed to just compare the two names with whatever case-independent
or other function.
And when I say 'is supposed to', I obviously mean that 'procfs does odd
things, and actually looks at the dentry that we don't even pass down,
rather than just the name'. Which results in problems, because we
actually call d_compare before we have even verified that the dentry is
still hashed at all.
And that causes a problm since the inode that procfs looks at may have
been free'd and the d_inode pointer is NULL. procfs just assumes that
all dentries are positive, since procfs itself never generates a
negative one. But memory pressure will still result in the dentry
getting torn down, and as it is removed by RCU, it still remains visible
on some lists - and to d_compare.
If the filesystem just did a name comparison, we wouldn't care. And we
could just fix procfs to know about negative dentries too. But rather
than have the low-level filesystems know about internal VFS details,
just move the check for a unhashed dentry up a bit, so that we will only
call d_compare on dentries that are still active.
The actual oops this caused didn't look like a NULL pointer dereference
because procfs did a 'container_of(inode, struct proc_inode, vfs_inode)'
to get at its internal proc_inode information from the inode pointer,
and accessed a field below the inode. So the oops would look something
like
BUG: unable to handle kernel paging request at fffffffffffffff0
IP: [<ffffffff802bc6c6>] proc_sys_compare+0x36/0x50
and was seen on both x86-64 (Alexey Dobriyan and Hugh Dickins) and
ppc64 (Hugh Dickins).
Reported-by: Alexey Dobriyan <adobriyan@gmail.com>
Acked-by: Hugh Dickins <hugh@veritas.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Reviewed-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-of-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-09-29 22:42:57 +08:00
|
|
|
atomic_inc(&dentry->d_count);
|
|
|
|
found = dentry;
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
break;
|
|
|
|
next:
|
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
}
|
|
|
|
rcu_read_unlock();
|
|
|
|
|
|
|
|
return found;
|
|
|
|
}
|
|
|
|
|
2006-03-31 18:31:43 +08:00
|
|
|
/**
|
|
|
|
* d_hash_and_lookup - hash the qstr then search for a dentry
|
|
|
|
* @dir: Directory to search in
|
|
|
|
* @name: qstr of name we wish to find
|
|
|
|
*
|
|
|
|
* On hash failure or on lookup failure NULL is returned.
|
|
|
|
*/
|
|
|
|
struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
|
|
|
|
{
|
|
|
|
struct dentry *dentry = NULL;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Check for a fs-specific hash function. Note that we must
|
|
|
|
* calculate the standard hash first, as the d_op->d_hash()
|
|
|
|
* routine may choose to leave the hash value unchanged.
|
|
|
|
*/
|
|
|
|
name->hash = full_name_hash(name->name, name->len);
|
|
|
|
if (dir->d_op && dir->d_op->d_hash) {
|
2011-01-07 14:49:28 +08:00
|
|
|
if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
|
2006-03-31 18:31:43 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
dentry = d_lookup(dir, name);
|
|
|
|
out:
|
|
|
|
return dentry;
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
2011-01-07 14:49:16 +08:00
|
|
|
* d_validate - verify dentry provided from insecure source (deprecated)
|
2005-04-17 06:20:36 +08:00
|
|
|
* @dentry: The dentry alleged to be valid child of @dparent
|
|
|
|
* @dparent: The parent dentry (known to be valid)
|
|
|
|
*
|
|
|
|
* An insecure source has sent us a dentry, here we verify it and dget() it.
|
|
|
|
* This is used by ncpfs in its readdir implementation.
|
|
|
|
* Zero is returned in the dentry is invalid.
|
2011-01-07 14:49:16 +08:00
|
|
|
*
|
|
|
|
* This function is slow for big directories, and deprecated, do not use it.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2011-01-05 17:01:21 +08:00
|
|
|
int d_validate(struct dentry *dentry, struct dentry *dparent)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2011-01-07 14:49:16 +08:00
|
|
|
struct dentry *child;
|
2011-01-05 17:01:21 +08:00
|
|
|
|
|
|
|
spin_lock(&dcache_lock);
|
2011-01-07 14:49:16 +08:00
|
|
|
list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
|
|
|
|
if (dentry == child) {
|
2011-01-05 17:01:21 +08:00
|
|
|
__dget_locked(dentry);
|
|
|
|
spin_unlock(&dcache_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
}
|
2011-01-05 17:01:21 +08:00
|
|
|
spin_unlock(&dcache_lock);
|
2011-01-07 14:49:16 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
return 0;
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(d_validate);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* When a file is deleted, we have two options:
|
|
|
|
* - turn this dentry into a negative dentry
|
|
|
|
* - unhash this dentry and free it.
|
|
|
|
*
|
|
|
|
* Usually, we want to just turn this into
|
|
|
|
* a negative dentry, but if anybody else is
|
|
|
|
* currently using the dentry or the inode
|
|
|
|
* we can't do that and we fall back on removing
|
|
|
|
* it from the hash queues and waiting for
|
|
|
|
* it to be deleted later when it has no users
|
|
|
|
*/
|
|
|
|
|
|
|
|
/**
|
|
|
|
* d_delete - delete a dentry
|
|
|
|
* @dentry: The dentry to delete
|
|
|
|
*
|
|
|
|
* Turn the dentry into a negative dentry if possible, otherwise
|
|
|
|
* remove it from the hash queues so it can be deleted later
|
|
|
|
*/
|
|
|
|
|
|
|
|
void d_delete(struct dentry * dentry)
|
|
|
|
{
|
2005-08-09 01:52:16 +08:00
|
|
|
int isdir = 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* Are we the only user?
|
|
|
|
*/
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
spin_lock(&dentry->d_lock);
|
2005-08-09 01:52:16 +08:00
|
|
|
isdir = S_ISDIR(dentry->d_inode->i_mode);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (atomic_read(&dentry->d_count) == 1) {
|
2010-05-22 04:11:04 +08:00
|
|
|
dentry->d_flags &= ~DCACHE_CANT_MOUNT;
|
2005-04-17 06:20:36 +08:00
|
|
|
dentry_iput(dentry);
|
2005-08-09 01:52:16 +08:00
|
|
|
fsnotify_nameremove(dentry, isdir);
|
2005-04-17 06:20:36 +08:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!d_unhashed(dentry))
|
|
|
|
__d_drop(dentry);
|
|
|
|
|
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
spin_unlock(&dcache_lock);
|
2005-08-09 01:52:16 +08:00
|
|
|
|
|
|
|
fsnotify_nameremove(dentry, isdir);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(d_delete);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
static void __d_rehash(struct dentry * entry, struct hlist_head *list)
|
|
|
|
{
|
|
|
|
|
|
|
|
entry->d_flags &= ~DCACHE_UNHASHED;
|
|
|
|
hlist_add_head_rcu(&entry->d_hash, list);
|
|
|
|
}
|
|
|
|
|
2006-08-23 08:06:07 +08:00
|
|
|
static void _d_rehash(struct dentry * entry)
|
|
|
|
{
|
|
|
|
__d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
|
|
|
* d_rehash - add an entry back to the hash
|
|
|
|
* @entry: dentry to add to the hash
|
|
|
|
*
|
|
|
|
* Adds a dentry to the hash according to its name.
|
|
|
|
*/
|
|
|
|
|
|
|
|
void d_rehash(struct dentry * entry)
|
|
|
|
{
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
spin_lock(&entry->d_lock);
|
2006-08-23 08:06:07 +08:00
|
|
|
_d_rehash(entry);
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_unlock(&entry->d_lock);
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(d_rehash);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2011-01-07 14:49:26 +08:00
|
|
|
/**
|
|
|
|
* dentry_update_name_case - update case insensitive dentry with a new name
|
|
|
|
* @dentry: dentry to be updated
|
|
|
|
* @name: new name
|
|
|
|
*
|
|
|
|
* Update a case insensitive dentry with new case of name.
|
|
|
|
*
|
|
|
|
* dentry must have been returned by d_lookup with name @name. Old and new
|
|
|
|
* name lengths must match (ie. no d_compare which allows mismatched name
|
|
|
|
* lengths).
|
|
|
|
*
|
|
|
|
* Parent inode i_mutex must be held over d_lookup and into this call (to
|
|
|
|
* keep renames and concurrent inserts, and readdir(2) away).
|
|
|
|
*/
|
|
|
|
void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
|
|
|
|
{
|
|
|
|
BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
|
|
|
|
BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
|
|
|
|
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
spin_lock(&dentry->d_lock);
|
|
|
|
memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
|
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(dentry_update_name_case);
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* When switching names, the actual string doesn't strictly have to
|
|
|
|
* be preserved in the target - because we're dropping the target
|
|
|
|
* anyway. As such, we can just do a simple memcpy() to copy over
|
|
|
|
* the new name before we switch.
|
|
|
|
*
|
|
|
|
* Note that we have to be a lot more careful about getting the hash
|
|
|
|
* switched - we have to switch the hash value properly even if it
|
|
|
|
* then no longer matches the actual (corrupted) string of the target.
|
|
|
|
* The hash value has to match the hash queue that the dentry is on..
|
|
|
|
*/
|
|
|
|
static void switch_names(struct dentry *dentry, struct dentry *target)
|
|
|
|
{
|
|
|
|
if (dname_external(target)) {
|
|
|
|
if (dname_external(dentry)) {
|
|
|
|
/*
|
|
|
|
* Both external: swap the pointers
|
|
|
|
*/
|
2009-01-08 10:09:14 +08:00
|
|
|
swap(target->d_name.name, dentry->d_name.name);
|
2005-04-17 06:20:36 +08:00
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* dentry:internal, target:external. Steal target's
|
|
|
|
* storage and make target internal.
|
|
|
|
*/
|
2007-10-22 07:41:38 +08:00
|
|
|
memcpy(target->d_iname, dentry->d_name.name,
|
|
|
|
dentry->d_name.len + 1);
|
2005-04-17 06:20:36 +08:00
|
|
|
dentry->d_name.name = target->d_name.name;
|
|
|
|
target->d_name.name = target->d_iname;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if (dname_external(dentry)) {
|
|
|
|
/*
|
|
|
|
* dentry:external, target:internal. Give dentry's
|
|
|
|
* storage to target and make dentry internal
|
|
|
|
*/
|
|
|
|
memcpy(dentry->d_iname, target->d_name.name,
|
|
|
|
target->d_name.len + 1);
|
|
|
|
target->d_name.name = dentry->d_name.name;
|
|
|
|
dentry->d_name.name = dentry->d_iname;
|
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* Both are internal. Just copy target to dentry
|
|
|
|
*/
|
|
|
|
memcpy(dentry->d_iname, target->d_name.name,
|
|
|
|
target->d_name.len + 1);
|
2008-11-04 04:03:50 +08:00
|
|
|
dentry->d_name.len = target->d_name.len;
|
|
|
|
return;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
2009-01-08 10:09:14 +08:00
|
|
|
swap(dentry->d_name.len, target->d_name.len);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We cannibalize "target" when moving dentry on top of it,
|
|
|
|
* because it's going to be thrown away anyway. We could be more
|
|
|
|
* polite about it, though.
|
|
|
|
*
|
|
|
|
* This forceful removal will result in ugly /proc output if
|
|
|
|
* somebody holds a file open that got deleted due to a rename.
|
|
|
|
* We could be nicer about the deleted file, and let it show
|
2007-10-17 14:29:42 +08:00
|
|
|
* up under the name it had before it was deleted rather than
|
|
|
|
* under the original name of the file that was moved on top of it.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
|
|
|
|
2006-10-22 01:24:20 +08:00
|
|
|
/*
|
|
|
|
* d_move_locked - move a dentry
|
2005-04-17 06:20:36 +08:00
|
|
|
* @dentry: entry to move
|
|
|
|
* @target: new dentry
|
|
|
|
*
|
|
|
|
* Update the dcache to reflect the move of a file name. Negative
|
|
|
|
* dcache entries should not be moved in this way.
|
|
|
|
*/
|
2006-10-22 01:24:20 +08:00
|
|
|
static void d_move_locked(struct dentry * dentry, struct dentry * target)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct hlist_head *list;
|
|
|
|
|
|
|
|
if (!dentry->d_inode)
|
|
|
|
printk(KERN_WARNING "VFS: moving negative dcache entry\n");
|
|
|
|
|
|
|
|
write_seqlock(&rename_lock);
|
|
|
|
/*
|
|
|
|
* XXXX: do we really need to take target->d_lock?
|
|
|
|
*/
|
|
|
|
if (target < dentry) {
|
|
|
|
spin_lock(&target->d_lock);
|
2006-07-03 15:25:04 +08:00
|
|
|
spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
|
2005-04-17 06:20:36 +08:00
|
|
|
} else {
|
|
|
|
spin_lock(&dentry->d_lock);
|
2006-07-03 15:25:04 +08:00
|
|
|
spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Move the dentry to the target hash queue, if on different bucket */
|
2007-10-17 14:30:11 +08:00
|
|
|
if (d_unhashed(dentry))
|
2005-04-17 06:20:36 +08:00
|
|
|
goto already_unhashed;
|
|
|
|
|
|
|
|
hlist_del_rcu(&dentry->d_hash);
|
|
|
|
|
|
|
|
already_unhashed:
|
|
|
|
list = d_hash(target->d_parent, target->d_name.hash);
|
|
|
|
__d_rehash(dentry, list);
|
|
|
|
|
|
|
|
/* Unhash the target: dput() will then get rid of it */
|
|
|
|
__d_drop(target);
|
|
|
|
|
[PATCH] shrink dentry struct
Some long time ago, dentry struct was carefully tuned so that on 32 bits
UP, sizeof(struct dentry) was exactly 128, ie a power of 2, and a multiple
of memory cache lines.
Then RCU was added and dentry struct enlarged by two pointers, with nice
results for SMP, but not so good on UP, because breaking the above tuning
(128 + 8 = 136 bytes)
This patch reverts this unwanted side effect, by using an union (d_u),
where d_rcu and d_child are placed so that these two fields can share their
memory needs.
At the time d_free() is called (and d_rcu is really used), d_child is known
to be empty and not touched by the dentry freeing.
Lockless lookups only access d_name, d_parent, d_lock, d_op, d_flags (so
the previous content of d_child is not needed if said dentry was unhashed
but still accessed by a CPU because of RCU constraints)
As dentry cache easily contains millions of entries, a size reduction is
worth the extra complexity of the ugly C union.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Maneesh Soni <maneesh@in.ibm.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Ian Kent <raven@themaw.net>
Cc: Paul Jackson <pj@sgi.com>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: James Morris <jmorris@namei.org>
Cc: Stephen Smalley <sds@epoch.ncsc.mil>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 17:03:32 +08:00
|
|
|
list_del(&dentry->d_u.d_child);
|
|
|
|
list_del(&target->d_u.d_child);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* Switch the names.. */
|
|
|
|
switch_names(dentry, target);
|
2009-01-08 10:09:14 +08:00
|
|
|
swap(dentry->d_name.hash, target->d_name.hash);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* ... and switch the parents */
|
|
|
|
if (IS_ROOT(dentry)) {
|
|
|
|
dentry->d_parent = target->d_parent;
|
|
|
|
target->d_parent = target;
|
[PATCH] shrink dentry struct
Some long time ago, dentry struct was carefully tuned so that on 32 bits
UP, sizeof(struct dentry) was exactly 128, ie a power of 2, and a multiple
of memory cache lines.
Then RCU was added and dentry struct enlarged by two pointers, with nice
results for SMP, but not so good on UP, because breaking the above tuning
(128 + 8 = 136 bytes)
This patch reverts this unwanted side effect, by using an union (d_u),
where d_rcu and d_child are placed so that these two fields can share their
memory needs.
At the time d_free() is called (and d_rcu is really used), d_child is known
to be empty and not touched by the dentry freeing.
Lockless lookups only access d_name, d_parent, d_lock, d_op, d_flags (so
the previous content of d_child is not needed if said dentry was unhashed
but still accessed by a CPU because of RCU constraints)
As dentry cache easily contains millions of entries, a size reduction is
worth the extra complexity of the ugly C union.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Maneesh Soni <maneesh@in.ibm.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Ian Kent <raven@themaw.net>
Cc: Paul Jackson <pj@sgi.com>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: James Morris <jmorris@namei.org>
Cc: Stephen Smalley <sds@epoch.ncsc.mil>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 17:03:32 +08:00
|
|
|
INIT_LIST_HEAD(&target->d_u.d_child);
|
2005-04-17 06:20:36 +08:00
|
|
|
} else {
|
2009-01-08 10:09:14 +08:00
|
|
|
swap(dentry->d_parent, target->d_parent);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* And add them back to the (new) parent lists */
|
[PATCH] shrink dentry struct
Some long time ago, dentry struct was carefully tuned so that on 32 bits
UP, sizeof(struct dentry) was exactly 128, ie a power of 2, and a multiple
of memory cache lines.
Then RCU was added and dentry struct enlarged by two pointers, with nice
results for SMP, but not so good on UP, because breaking the above tuning
(128 + 8 = 136 bytes)
This patch reverts this unwanted side effect, by using an union (d_u),
where d_rcu and d_child are placed so that these two fields can share their
memory needs.
At the time d_free() is called (and d_rcu is really used), d_child is known
to be empty and not touched by the dentry freeing.
Lockless lookups only access d_name, d_parent, d_lock, d_op, d_flags (so
the previous content of d_child is not needed if said dentry was unhashed
but still accessed by a CPU because of RCU constraints)
As dentry cache easily contains millions of entries, a size reduction is
worth the extra complexity of the ugly C union.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Maneesh Soni <maneesh@in.ibm.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Ian Kent <raven@themaw.net>
Cc: Paul Jackson <pj@sgi.com>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: James Morris <jmorris@namei.org>
Cc: Stephen Smalley <sds@epoch.ncsc.mil>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 17:03:32 +08:00
|
|
|
list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
[PATCH] shrink dentry struct
Some long time ago, dentry struct was carefully tuned so that on 32 bits
UP, sizeof(struct dentry) was exactly 128, ie a power of 2, and a multiple
of memory cache lines.
Then RCU was added and dentry struct enlarged by two pointers, with nice
results for SMP, but not so good on UP, because breaking the above tuning
(128 + 8 = 136 bytes)
This patch reverts this unwanted side effect, by using an union (d_u),
where d_rcu and d_child are placed so that these two fields can share their
memory needs.
At the time d_free() is called (and d_rcu is really used), d_child is known
to be empty and not touched by the dentry freeing.
Lockless lookups only access d_name, d_parent, d_lock, d_op, d_flags (so
the previous content of d_child is not needed if said dentry was unhashed
but still accessed by a CPU because of RCU constraints)
As dentry cache easily contains millions of entries, a size reduction is
worth the extra complexity of the ugly C union.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Maneesh Soni <maneesh@in.ibm.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Ian Kent <raven@themaw.net>
Cc: Paul Jackson <pj@sgi.com>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: James Morris <jmorris@namei.org>
Cc: Stephen Smalley <sds@epoch.ncsc.mil>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 17:03:32 +08:00
|
|
|
list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_unlock(&target->d_lock);
|
2006-03-25 19:07:09 +08:00
|
|
|
fsnotify_d_move(dentry);
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
write_sequnlock(&rename_lock);
|
2006-10-22 01:24:20 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* d_move - move a dentry
|
|
|
|
* @dentry: entry to move
|
|
|
|
* @target: new dentry
|
|
|
|
*
|
|
|
|
* Update the dcache to reflect the move of a file name. Negative
|
|
|
|
* dcache entries should not be moved in this way.
|
|
|
|
*/
|
|
|
|
|
|
|
|
void d_move(struct dentry * dentry, struct dentry * target)
|
|
|
|
{
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
d_move_locked(dentry, target);
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(d_move);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2008-10-16 06:50:28 +08:00
|
|
|
/**
|
|
|
|
* d_ancestor - search for an ancestor
|
|
|
|
* @p1: ancestor dentry
|
|
|
|
* @p2: child dentry
|
|
|
|
*
|
|
|
|
* Returns the ancestor dentry of p2 which is a child of p1, if p1 is
|
|
|
|
* an ancestor of p2, else NULL.
|
2006-10-22 01:24:20 +08:00
|
|
|
*/
|
2008-10-16 06:50:28 +08:00
|
|
|
struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
|
2006-10-22 01:24:20 +08:00
|
|
|
{
|
|
|
|
struct dentry *p;
|
|
|
|
|
2008-10-16 06:50:27 +08:00
|
|
|
for (p = p2; !IS_ROOT(p); p = p->d_parent) {
|
2006-10-22 01:24:20 +08:00
|
|
|
if (p->d_parent == p1)
|
2008-10-16 06:50:28 +08:00
|
|
|
return p;
|
2006-10-22 01:24:20 +08:00
|
|
|
}
|
2008-10-16 06:50:28 +08:00
|
|
|
return NULL;
|
2006-10-22 01:24:20 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This helper attempts to cope with remotely renamed directories
|
|
|
|
*
|
|
|
|
* It assumes that the caller is already holding
|
|
|
|
* dentry->d_parent->d_inode->i_mutex and the dcache_lock
|
|
|
|
*
|
|
|
|
* Note: If ever the locking in lock_rename() changes, then please
|
|
|
|
* remember to update this too...
|
|
|
|
*/
|
|
|
|
static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias)
|
2008-06-24 00:11:52 +08:00
|
|
|
__releases(dcache_lock)
|
2006-10-22 01:24:20 +08:00
|
|
|
{
|
|
|
|
struct mutex *m1 = NULL, *m2 = NULL;
|
|
|
|
struct dentry *ret;
|
|
|
|
|
|
|
|
/* If alias and dentry share a parent, then no extra locks required */
|
|
|
|
if (alias->d_parent == dentry->d_parent)
|
|
|
|
goto out_unalias;
|
|
|
|
|
|
|
|
/* Check for loops */
|
|
|
|
ret = ERR_PTR(-ELOOP);
|
2008-10-16 06:50:28 +08:00
|
|
|
if (d_ancestor(alias, dentry))
|
2006-10-22 01:24:20 +08:00
|
|
|
goto out_err;
|
|
|
|
|
|
|
|
/* See lock_rename() */
|
|
|
|
ret = ERR_PTR(-EBUSY);
|
|
|
|
if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
|
|
|
|
goto out_err;
|
|
|
|
m1 = &dentry->d_sb->s_vfs_rename_mutex;
|
|
|
|
if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
|
|
|
|
goto out_err;
|
|
|
|
m2 = &alias->d_parent->d_inode->i_mutex;
|
|
|
|
out_unalias:
|
|
|
|
d_move_locked(alias, dentry);
|
|
|
|
ret = alias;
|
|
|
|
out_err:
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
if (m2)
|
|
|
|
mutex_unlock(m2);
|
|
|
|
if (m1)
|
|
|
|
mutex_unlock(m1);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2006-08-23 08:06:07 +08:00
|
|
|
/*
|
|
|
|
* Prepare an anonymous dentry for life in the superblock's dentry tree as a
|
|
|
|
* named dentry in place of the dentry to be replaced.
|
|
|
|
*/
|
|
|
|
static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
|
|
|
|
{
|
|
|
|
struct dentry *dparent, *aparent;
|
|
|
|
|
|
|
|
switch_names(dentry, anon);
|
2009-01-08 10:09:14 +08:00
|
|
|
swap(dentry->d_name.hash, anon->d_name.hash);
|
2006-08-23 08:06:07 +08:00
|
|
|
|
|
|
|
dparent = dentry->d_parent;
|
|
|
|
aparent = anon->d_parent;
|
|
|
|
|
|
|
|
dentry->d_parent = (aparent == anon) ? dentry : aparent;
|
|
|
|
list_del(&dentry->d_u.d_child);
|
|
|
|
if (!IS_ROOT(dentry))
|
|
|
|
list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
|
|
|
|
else
|
|
|
|
INIT_LIST_HEAD(&dentry->d_u.d_child);
|
|
|
|
|
|
|
|
anon->d_parent = (dparent == dentry) ? anon : dparent;
|
|
|
|
list_del(&anon->d_u.d_child);
|
|
|
|
if (!IS_ROOT(anon))
|
|
|
|
list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
|
|
|
|
else
|
|
|
|
INIT_LIST_HEAD(&anon->d_u.d_child);
|
|
|
|
|
|
|
|
anon->d_flags &= ~DCACHE_DISCONNECTED;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* d_materialise_unique - introduce an inode into the tree
|
|
|
|
* @dentry: candidate dentry
|
|
|
|
* @inode: inode to bind to the dentry, to which aliases may be attached
|
|
|
|
*
|
|
|
|
* Introduces an dentry into the tree, substituting an extant disconnected
|
|
|
|
* root directory alias in its place if there is one
|
|
|
|
*/
|
|
|
|
struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
|
|
|
|
{
|
2006-10-22 01:24:20 +08:00
|
|
|
struct dentry *actual;
|
2006-08-23 08:06:07 +08:00
|
|
|
|
|
|
|
BUG_ON(!d_unhashed(dentry));
|
|
|
|
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
|
|
|
|
if (!inode) {
|
|
|
|
actual = dentry;
|
2008-10-16 06:50:28 +08:00
|
|
|
__d_instantiate(dentry, NULL);
|
2006-08-23 08:06:07 +08:00
|
|
|
goto found_lock;
|
|
|
|
}
|
|
|
|
|
2006-10-22 01:24:20 +08:00
|
|
|
if (S_ISDIR(inode->i_mode)) {
|
|
|
|
struct dentry *alias;
|
|
|
|
|
|
|
|
/* Does an aliased dentry already exist? */
|
|
|
|
alias = __d_find_alias(inode, 0);
|
|
|
|
if (alias) {
|
|
|
|
actual = alias;
|
|
|
|
/* Is this an anonymous mountpoint that we could splice
|
|
|
|
* into our tree? */
|
|
|
|
if (IS_ROOT(alias)) {
|
|
|
|
spin_lock(&alias->d_lock);
|
|
|
|
__d_materialise_dentry(dentry, alias);
|
|
|
|
__d_drop(alias);
|
|
|
|
goto found;
|
|
|
|
}
|
|
|
|
/* Nope, but we must(!) avoid directory aliasing */
|
|
|
|
actual = __d_unalias(dentry, alias);
|
|
|
|
if (IS_ERR(actual))
|
|
|
|
dput(alias);
|
|
|
|
goto out_nolock;
|
|
|
|
}
|
2006-08-23 08:06:07 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Add a unique reference */
|
|
|
|
actual = __d_instantiate_unique(dentry, inode);
|
|
|
|
if (!actual)
|
|
|
|
actual = dentry;
|
|
|
|
else if (unlikely(!d_unhashed(actual)))
|
|
|
|
goto shouldnt_be_hashed;
|
|
|
|
|
|
|
|
found_lock:
|
|
|
|
spin_lock(&actual->d_lock);
|
|
|
|
found:
|
|
|
|
_d_rehash(actual);
|
|
|
|
spin_unlock(&actual->d_lock);
|
|
|
|
spin_unlock(&dcache_lock);
|
2006-10-22 01:24:20 +08:00
|
|
|
out_nolock:
|
2006-08-23 08:06:07 +08:00
|
|
|
if (actual == dentry) {
|
|
|
|
security_d_instantiate(dentry, inode);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
iput(inode);
|
|
|
|
return actual;
|
|
|
|
|
|
|
|
shouldnt_be_hashed:
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
BUG();
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL_GPL(d_materialise_unique);
|
2006-08-23 08:06:07 +08:00
|
|
|
|
2008-06-24 00:11:53 +08:00
|
|
|
static int prepend(char **buffer, int *buflen, const char *str, int namelen)
|
2008-03-27 20:06:20 +08:00
|
|
|
{
|
|
|
|
*buflen -= namelen;
|
|
|
|
if (*buflen < 0)
|
|
|
|
return -ENAMETOOLONG;
|
|
|
|
*buffer -= namelen;
|
|
|
|
memcpy(*buffer, str, namelen);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-06-24 00:11:53 +08:00
|
|
|
static int prepend_name(char **buffer, int *buflen, struct qstr *name)
|
|
|
|
{
|
|
|
|
return prepend(buffer, buflen, name->name, name->len);
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
2010-08-10 17:41:39 +08:00
|
|
|
* Prepend path string to a buffer
|
|
|
|
*
|
2008-03-27 20:06:21 +08:00
|
|
|
* @path: the dentry/vfsmount to report
|
|
|
|
* @root: root vfsmnt/dentry (may be modified by this function)
|
2010-08-10 17:41:39 +08:00
|
|
|
* @buffer: pointer to the end of the buffer
|
|
|
|
* @buflen: pointer to buffer length
|
2007-02-14 04:08:18 +08:00
|
|
|
*
|
2010-08-10 17:41:39 +08:00
|
|
|
* Caller holds the dcache_lock.
|
2008-03-27 20:06:21 +08:00
|
|
|
*
|
|
|
|
* If path is not reachable from the supplied root, then the value of
|
|
|
|
* root is changed (without modifying refcounts).
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2010-08-10 17:41:39 +08:00
|
|
|
static int prepend_path(const struct path *path, struct path *root,
|
|
|
|
char **buffer, int *buflen)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2008-03-27 20:06:21 +08:00
|
|
|
struct dentry *dentry = path->dentry;
|
|
|
|
struct vfsmount *vfsmnt = path->mnt;
|
2010-08-10 17:41:39 +08:00
|
|
|
bool slash = false;
|
|
|
|
int error = 0;
|
2008-03-27 20:06:20 +08:00
|
|
|
|
fs: brlock vfsmount_lock
fs: brlock vfsmount_lock
Use a brlock for the vfsmount lock. It must be taken for write whenever
modifying the mount hash or associated fields, and may be taken for read when
performing mount hash lookups.
A new lock is added for the mnt-id allocator, so it doesn't need to take
the heavy vfsmount write-lock.
The number of atomics should remain the same for fastpath rlock cases, though
code would be slightly slower due to per-cpu access. Scalability is not not be
much improved in common cases yet, due to other locks (ie. dcache_lock) getting
in the way. However path lookups crossing mountpoints should be one case where
scalability is improved (currently requiring the global lock).
The slowpath is slower due to use of brlock. On a 64 core, 64 socket, 32 node
Altix system (high latency to remote nodes), a simple umount microbenchmark
(mount --bind mnt mnt2 ; umount mnt2 loop 1000 times), before this patch it
took 6.8s, afterwards took 7.1s, about 5% slower.
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2010-08-18 02:37:39 +08:00
|
|
|
br_read_lock(vfsmount_lock);
|
2010-08-10 17:41:39 +08:00
|
|
|
while (dentry != root->dentry || vfsmnt != root->mnt) {
|
2005-04-17 06:20:36 +08:00
|
|
|
struct dentry * parent;
|
|
|
|
|
|
|
|
if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
|
2007-02-14 04:08:18 +08:00
|
|
|
/* Global root? */
|
2005-04-17 06:20:36 +08:00
|
|
|
if (vfsmnt->mnt_parent == vfsmnt) {
|
|
|
|
goto global_root;
|
|
|
|
}
|
|
|
|
dentry = vfsmnt->mnt_mountpoint;
|
|
|
|
vfsmnt = vfsmnt->mnt_parent;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
parent = dentry->d_parent;
|
|
|
|
prefetch(parent);
|
2010-08-10 17:41:39 +08:00
|
|
|
error = prepend_name(buffer, buflen, &dentry->d_name);
|
|
|
|
if (!error)
|
|
|
|
error = prepend(buffer, buflen, "/", 1);
|
|
|
|
if (error)
|
|
|
|
break;
|
|
|
|
|
|
|
|
slash = true;
|
2005-04-17 06:20:36 +08:00
|
|
|
dentry = parent;
|
|
|
|
}
|
|
|
|
|
2008-06-16 19:28:07 +08:00
|
|
|
out:
|
2010-08-10 17:41:39 +08:00
|
|
|
if (!error && !slash)
|
|
|
|
error = prepend(buffer, buflen, "/", 1);
|
|
|
|
|
fs: brlock vfsmount_lock
fs: brlock vfsmount_lock
Use a brlock for the vfsmount lock. It must be taken for write whenever
modifying the mount hash or associated fields, and may be taken for read when
performing mount hash lookups.
A new lock is added for the mnt-id allocator, so it doesn't need to take
the heavy vfsmount write-lock.
The number of atomics should remain the same for fastpath rlock cases, though
code would be slightly slower due to per-cpu access. Scalability is not not be
much improved in common cases yet, due to other locks (ie. dcache_lock) getting
in the way. However path lookups crossing mountpoints should be one case where
scalability is improved (currently requiring the global lock).
The slowpath is slower due to use of brlock. On a 64 core, 64 socket, 32 node
Altix system (high latency to remote nodes), a simple umount microbenchmark
(mount --bind mnt mnt2 ; umount mnt2 loop 1000 times), before this patch it
took 6.8s, afterwards took 7.1s, about 5% slower.
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2010-08-18 02:37:39 +08:00
|
|
|
br_read_unlock(vfsmount_lock);
|
2010-08-10 17:41:39 +08:00
|
|
|
return error;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
global_root:
|
2010-08-10 17:41:38 +08:00
|
|
|
/*
|
|
|
|
* Filesystems needing to implement special "root names"
|
|
|
|
* should do so with ->d_dname()
|
|
|
|
*/
|
|
|
|
if (IS_ROOT(dentry) &&
|
|
|
|
(dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
|
|
|
|
WARN(1, "Root dentry has weird name <%.*s>\n",
|
|
|
|
(int) dentry->d_name.len, dentry->d_name.name);
|
|
|
|
}
|
2008-03-27 20:06:21 +08:00
|
|
|
root->mnt = vfsmnt;
|
|
|
|
root->dentry = dentry;
|
2008-06-16 19:28:07 +08:00
|
|
|
goto out;
|
2010-08-10 17:41:39 +08:00
|
|
|
}
|
2008-06-16 19:28:07 +08:00
|
|
|
|
2010-08-10 17:41:39 +08:00
|
|
|
/**
|
|
|
|
* __d_path - return the path of a dentry
|
|
|
|
* @path: the dentry/vfsmount to report
|
|
|
|
* @root: root vfsmnt/dentry (may be modified by this function)
|
2010-08-15 04:05:31 +08:00
|
|
|
* @buf: buffer to return value in
|
2010-08-10 17:41:39 +08:00
|
|
|
* @buflen: buffer length
|
|
|
|
*
|
2010-08-10 17:41:40 +08:00
|
|
|
* Convert a dentry into an ASCII path name.
|
2010-08-10 17:41:39 +08:00
|
|
|
*
|
|
|
|
* Returns a pointer into the buffer or an error code if the
|
|
|
|
* path was too long.
|
|
|
|
*
|
2010-10-10 17:36:21 +08:00
|
|
|
* "buflen" should be positive.
|
2010-08-10 17:41:39 +08:00
|
|
|
*
|
|
|
|
* If path is not reachable from the supplied root, then the value of
|
|
|
|
* root is changed (without modifying refcounts).
|
|
|
|
*/
|
|
|
|
char *__d_path(const struct path *path, struct path *root,
|
|
|
|
char *buf, int buflen)
|
|
|
|
{
|
|
|
|
char *res = buf + buflen;
|
|
|
|
int error;
|
|
|
|
|
|
|
|
prepend(&res, &buflen, "\0", 1);
|
2010-10-10 17:36:21 +08:00
|
|
|
spin_lock(&dcache_lock);
|
2010-08-10 17:41:39 +08:00
|
|
|
error = prepend_path(path, root, &res, &buflen);
|
2010-10-10 17:36:21 +08:00
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
|
2010-08-10 17:41:39 +08:00
|
|
|
if (error)
|
|
|
|
return ERR_PTR(error);
|
|
|
|
return res;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2010-08-10 17:41:40 +08:00
|
|
|
/*
|
|
|
|
* same as __d_path but appends "(deleted)" for unlinked files.
|
|
|
|
*/
|
|
|
|
static int path_with_deleted(const struct path *path, struct path *root,
|
|
|
|
char **buf, int *buflen)
|
|
|
|
{
|
|
|
|
prepend(buf, buflen, "\0", 1);
|
|
|
|
if (d_unlinked(path->dentry)) {
|
|
|
|
int error = prepend(buf, buflen, " (deleted)", 10);
|
|
|
|
if (error)
|
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
|
|
|
return prepend_path(path, root, buf, buflen);
|
|
|
|
}
|
|
|
|
|
2010-08-10 17:41:41 +08:00
|
|
|
static int prepend_unreachable(char **buffer, int *buflen)
|
|
|
|
{
|
|
|
|
return prepend(buffer, buflen, "(unreachable)", 13);
|
|
|
|
}
|
|
|
|
|
2008-02-15 11:38:32 +08:00
|
|
|
/**
|
|
|
|
* d_path - return the path of a dentry
|
2008-02-15 11:38:44 +08:00
|
|
|
* @path: path to report
|
2008-02-15 11:38:32 +08:00
|
|
|
* @buf: buffer to return value in
|
|
|
|
* @buflen: buffer length
|
|
|
|
*
|
|
|
|
* Convert a dentry into an ASCII path name. If the entry has been deleted
|
|
|
|
* the string " (deleted)" is appended. Note that this is ambiguous.
|
|
|
|
*
|
2008-12-02 06:35:00 +08:00
|
|
|
* Returns a pointer into the buffer or an error code if the path was
|
|
|
|
* too long. Note: Callers should use the returned pointer, not the passed
|
|
|
|
* in buffer, to use the name! The implementation often starts at an offset
|
|
|
|
* into the buffer, and may leave 0 bytes at the start.
|
2008-02-15 11:38:32 +08:00
|
|
|
*
|
2008-06-24 00:11:52 +08:00
|
|
|
* "buflen" should be positive.
|
2008-02-15 11:38:32 +08:00
|
|
|
*/
|
2008-06-10 07:40:36 +08:00
|
|
|
char *d_path(const struct path *path, char *buf, int buflen)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2010-08-10 17:41:40 +08:00
|
|
|
char *res = buf + buflen;
|
2008-02-15 11:34:38 +08:00
|
|
|
struct path root;
|
2008-03-27 20:06:21 +08:00
|
|
|
struct path tmp;
|
2010-08-10 17:41:40 +08:00
|
|
|
int error;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-05-08 15:26:18 +08:00
|
|
|
/*
|
|
|
|
* We have various synthetic filesystems that never get mounted. On
|
|
|
|
* these filesystems dentries are never used for lookup purposes, and
|
|
|
|
* thus don't need to be hashed. They also don't need a name until a
|
|
|
|
* user wants to identify the object in /proc/pid/fd/. The little hack
|
|
|
|
* below allows us to generate a name for these objects on demand:
|
|
|
|
*/
|
2008-02-15 11:38:44 +08:00
|
|
|
if (path->dentry->d_op && path->dentry->d_op->d_dname)
|
|
|
|
return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
|
2007-05-08 15:26:18 +08:00
|
|
|
|
2010-08-10 17:41:36 +08:00
|
|
|
get_fs_root(current->fs, &root);
|
2007-02-14 04:08:18 +08:00
|
|
|
spin_lock(&dcache_lock);
|
2008-03-27 20:06:21 +08:00
|
|
|
tmp = root;
|
2010-08-10 17:41:40 +08:00
|
|
|
error = path_with_deleted(path, &tmp, &res, &buflen);
|
|
|
|
if (error)
|
|
|
|
res = ERR_PTR(error);
|
2007-02-14 04:08:18 +08:00
|
|
|
spin_unlock(&dcache_lock);
|
2008-02-15 11:34:38 +08:00
|
|
|
path_put(&root);
|
2005-04-17 06:20:36 +08:00
|
|
|
return res;
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(d_path);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2010-08-10 17:41:41 +08:00
|
|
|
/**
|
|
|
|
* d_path_with_unreachable - return the path of a dentry
|
|
|
|
* @path: path to report
|
|
|
|
* @buf: buffer to return value in
|
|
|
|
* @buflen: buffer length
|
|
|
|
*
|
|
|
|
* The difference from d_path() is that this prepends "(unreachable)"
|
|
|
|
* to paths which are unreachable from the current process' root.
|
|
|
|
*/
|
|
|
|
char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
|
|
|
|
{
|
|
|
|
char *res = buf + buflen;
|
|
|
|
struct path root;
|
|
|
|
struct path tmp;
|
|
|
|
int error;
|
|
|
|
|
|
|
|
if (path->dentry->d_op && path->dentry->d_op->d_dname)
|
|
|
|
return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
|
|
|
|
|
|
|
|
get_fs_root(current->fs, &root);
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
tmp = root;
|
|
|
|
error = path_with_deleted(path, &tmp, &res, &buflen);
|
|
|
|
if (!error && !path_equal(&tmp, &root))
|
|
|
|
error = prepend_unreachable(&res, &buflen);
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
path_put(&root);
|
|
|
|
if (error)
|
|
|
|
res = ERR_PTR(error);
|
|
|
|
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
2007-05-08 15:26:18 +08:00
|
|
|
/*
|
|
|
|
* Helper function for dentry_operations.d_dname() members
|
|
|
|
*/
|
|
|
|
char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
|
|
|
|
const char *fmt, ...)
|
|
|
|
{
|
|
|
|
va_list args;
|
|
|
|
char temp[64];
|
|
|
|
int sz;
|
|
|
|
|
|
|
|
va_start(args, fmt);
|
|
|
|
sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
|
|
|
|
va_end(args);
|
|
|
|
|
|
|
|
if (sz > sizeof(temp) || sz > buflen)
|
|
|
|
return ERR_PTR(-ENAMETOOLONG);
|
|
|
|
|
|
|
|
buffer += buflen - sz;
|
|
|
|
return memcpy(buffer, temp, sz);
|
|
|
|
}
|
|
|
|
|
2008-03-27 20:06:20 +08:00
|
|
|
/*
|
|
|
|
* Write full pathname from the root of the filesystem into the buffer.
|
|
|
|
*/
|
2011-01-07 14:49:29 +08:00
|
|
|
static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
|
2008-03-27 20:06:20 +08:00
|
|
|
{
|
|
|
|
char *end = buf + buflen;
|
|
|
|
char *retval;
|
|
|
|
|
|
|
|
prepend(&end, &buflen, "\0", 1);
|
|
|
|
if (buflen < 1)
|
|
|
|
goto Elong;
|
|
|
|
/* Get '/' right */
|
|
|
|
retval = end-1;
|
|
|
|
*retval = '/';
|
|
|
|
|
2008-06-24 00:11:53 +08:00
|
|
|
while (!IS_ROOT(dentry)) {
|
|
|
|
struct dentry *parent = dentry->d_parent;
|
2008-03-27 20:06:20 +08:00
|
|
|
|
|
|
|
prefetch(parent);
|
2008-06-24 00:11:53 +08:00
|
|
|
if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
|
2008-03-27 20:06:20 +08:00
|
|
|
(prepend(&end, &buflen, "/", 1) != 0))
|
|
|
|
goto Elong;
|
|
|
|
|
|
|
|
retval = end;
|
|
|
|
dentry = parent;
|
|
|
|
}
|
2010-06-07 10:31:14 +08:00
|
|
|
return retval;
|
|
|
|
Elong:
|
|
|
|
return ERR_PTR(-ENAMETOOLONG);
|
|
|
|
}
|
2011-01-07 14:49:29 +08:00
|
|
|
|
|
|
|
char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
|
|
|
|
{
|
|
|
|
char *retval;
|
|
|
|
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
retval = __dentry_path(dentry, buf, buflen);
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
|
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(dentry_path_raw);
|
2010-06-07 10:31:14 +08:00
|
|
|
|
|
|
|
char *dentry_path(struct dentry *dentry, char *buf, int buflen)
|
|
|
|
{
|
|
|
|
char *p = NULL;
|
|
|
|
char *retval;
|
|
|
|
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
if (d_unlinked(dentry)) {
|
|
|
|
p = buf + buflen;
|
|
|
|
if (prepend(&p, &buflen, "//deleted", 10) != 0)
|
|
|
|
goto Elong;
|
|
|
|
buflen++;
|
|
|
|
}
|
|
|
|
retval = __dentry_path(dentry, buf, buflen);
|
2008-03-27 20:06:20 +08:00
|
|
|
spin_unlock(&dcache_lock);
|
2010-06-07 10:31:14 +08:00
|
|
|
if (!IS_ERR(retval) && p)
|
|
|
|
*p = '/'; /* restore '/' overriden with '\0' */
|
2008-03-27 20:06:20 +08:00
|
|
|
return retval;
|
|
|
|
Elong:
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
return ERR_PTR(-ENAMETOOLONG);
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* NOTE! The user-level library version returns a
|
|
|
|
* character pointer. The kernel system call just
|
|
|
|
* returns the length of the buffer filled (which
|
|
|
|
* includes the ending '\0' character), or a negative
|
|
|
|
* error value. So libc would do something like
|
|
|
|
*
|
|
|
|
* char *getcwd(char * buf, size_t size)
|
|
|
|
* {
|
|
|
|
* int retval;
|
|
|
|
*
|
|
|
|
* retval = sys_getcwd(buf, size);
|
|
|
|
* if (retval >= 0)
|
|
|
|
* return buf;
|
|
|
|
* errno = -retval;
|
|
|
|
* return NULL;
|
|
|
|
* }
|
|
|
|
*/
|
2009-01-14 21:14:22 +08:00
|
|
|
SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2007-02-14 04:08:18 +08:00
|
|
|
int error;
|
2008-02-15 11:34:38 +08:00
|
|
|
struct path pwd, root;
|
2007-02-14 04:08:18 +08:00
|
|
|
char *page = (char *) __get_free_page(GFP_USER);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
if (!page)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2010-08-10 17:41:36 +08:00
|
|
|
get_fs_root_and_pwd(current->fs, &root, &pwd);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-02-14 04:08:18 +08:00
|
|
|
error = -ENOENT;
|
|
|
|
spin_lock(&dcache_lock);
|
2009-05-04 07:32:03 +08:00
|
|
|
if (!d_unlinked(pwd.dentry)) {
|
2007-02-14 04:08:18 +08:00
|
|
|
unsigned long len;
|
2008-03-27 20:06:21 +08:00
|
|
|
struct path tmp = root;
|
2010-08-10 17:41:41 +08:00
|
|
|
char *cwd = page + PAGE_SIZE;
|
|
|
|
int buflen = PAGE_SIZE;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2010-08-10 17:41:41 +08:00
|
|
|
prepend(&cwd, &buflen, "\0", 1);
|
|
|
|
error = prepend_path(&pwd, &tmp, &cwd, &buflen);
|
2007-02-14 04:08:18 +08:00
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
|
2010-08-10 17:41:41 +08:00
|
|
|
if (error)
|
2007-02-14 04:08:18 +08:00
|
|
|
goto out;
|
|
|
|
|
2010-08-10 17:41:41 +08:00
|
|
|
/* Unreachable from current root */
|
|
|
|
if (!path_equal(&tmp, &root)) {
|
|
|
|
error = prepend_unreachable(&cwd, &buflen);
|
|
|
|
if (error)
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2007-02-14 04:08:18 +08:00
|
|
|
error = -ERANGE;
|
|
|
|
len = PAGE_SIZE + page - cwd;
|
|
|
|
if (len <= size) {
|
|
|
|
error = len;
|
|
|
|
if (copy_to_user(buf, cwd, len))
|
|
|
|
error = -EFAULT;
|
|
|
|
}
|
|
|
|
} else
|
|
|
|
spin_unlock(&dcache_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
out:
|
2008-02-15 11:34:38 +08:00
|
|
|
path_put(&pwd);
|
|
|
|
path_put(&root);
|
2005-04-17 06:20:36 +08:00
|
|
|
free_page((unsigned long) page);
|
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Test whether new_dentry is a subdirectory of old_dentry.
|
|
|
|
*
|
|
|
|
* Trivially implemented using the dcache structure
|
|
|
|
*/
|
|
|
|
|
|
|
|
/**
|
|
|
|
* is_subdir - is new dentry a subdirectory of old_dentry
|
|
|
|
* @new_dentry: new dentry
|
|
|
|
* @old_dentry: old dentry
|
|
|
|
*
|
|
|
|
* Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
|
|
|
|
* Returns 0 otherwise.
|
|
|
|
* Caller must ensure that "new_dentry" is pinned before calling is_subdir()
|
|
|
|
*/
|
|
|
|
|
2008-10-16 06:50:28 +08:00
|
|
|
int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
int result;
|
|
|
|
unsigned long seq;
|
|
|
|
|
2008-10-16 06:50:28 +08:00
|
|
|
if (new_dentry == old_dentry)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Need rcu_readlock to protect against the d_parent trashing
|
|
|
|
* due to d_move
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
|
|
|
rcu_read_lock();
|
2008-10-16 06:50:28 +08:00
|
|
|
do {
|
2005-04-17 06:20:36 +08:00
|
|
|
/* for restarting inner loop in case of seq retry */
|
|
|
|
seq = read_seqbegin(&rename_lock);
|
2008-10-16 06:50:28 +08:00
|
|
|
if (d_ancestor(old_dentry, new_dentry))
|
2005-04-17 06:20:36 +08:00
|
|
|
result = 1;
|
2008-10-16 06:50:28 +08:00
|
|
|
else
|
|
|
|
result = 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
} while (read_seqretry(&rename_lock, seq));
|
|
|
|
rcu_read_unlock();
|
|
|
|
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
2010-01-31 02:16:21 +08:00
|
|
|
int path_is_under(struct path *path1, struct path *path2)
|
|
|
|
{
|
|
|
|
struct vfsmount *mnt = path1->mnt;
|
|
|
|
struct dentry *dentry = path1->dentry;
|
|
|
|
int res;
|
fs: brlock vfsmount_lock
fs: brlock vfsmount_lock
Use a brlock for the vfsmount lock. It must be taken for write whenever
modifying the mount hash or associated fields, and may be taken for read when
performing mount hash lookups.
A new lock is added for the mnt-id allocator, so it doesn't need to take
the heavy vfsmount write-lock.
The number of atomics should remain the same for fastpath rlock cases, though
code would be slightly slower due to per-cpu access. Scalability is not not be
much improved in common cases yet, due to other locks (ie. dcache_lock) getting
in the way. However path lookups crossing mountpoints should be one case where
scalability is improved (currently requiring the global lock).
The slowpath is slower due to use of brlock. On a 64 core, 64 socket, 32 node
Altix system (high latency to remote nodes), a simple umount microbenchmark
(mount --bind mnt mnt2 ; umount mnt2 loop 1000 times), before this patch it
took 6.8s, afterwards took 7.1s, about 5% slower.
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2010-08-18 02:37:39 +08:00
|
|
|
|
|
|
|
br_read_lock(vfsmount_lock);
|
2010-01-31 02:16:21 +08:00
|
|
|
if (mnt != path2->mnt) {
|
|
|
|
for (;;) {
|
|
|
|
if (mnt->mnt_parent == mnt) {
|
fs: brlock vfsmount_lock
fs: brlock vfsmount_lock
Use a brlock for the vfsmount lock. It must be taken for write whenever
modifying the mount hash or associated fields, and may be taken for read when
performing mount hash lookups.
A new lock is added for the mnt-id allocator, so it doesn't need to take
the heavy vfsmount write-lock.
The number of atomics should remain the same for fastpath rlock cases, though
code would be slightly slower due to per-cpu access. Scalability is not not be
much improved in common cases yet, due to other locks (ie. dcache_lock) getting
in the way. However path lookups crossing mountpoints should be one case where
scalability is improved (currently requiring the global lock).
The slowpath is slower due to use of brlock. On a 64 core, 64 socket, 32 node
Altix system (high latency to remote nodes), a simple umount microbenchmark
(mount --bind mnt mnt2 ; umount mnt2 loop 1000 times), before this patch it
took 6.8s, afterwards took 7.1s, about 5% slower.
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2010-08-18 02:37:39 +08:00
|
|
|
br_read_unlock(vfsmount_lock);
|
2010-01-31 02:16:21 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
if (mnt->mnt_parent == path2->mnt)
|
|
|
|
break;
|
|
|
|
mnt = mnt->mnt_parent;
|
|
|
|
}
|
|
|
|
dentry = mnt->mnt_mountpoint;
|
|
|
|
}
|
|
|
|
res = is_subdir(dentry, path2->dentry);
|
fs: brlock vfsmount_lock
fs: brlock vfsmount_lock
Use a brlock for the vfsmount lock. It must be taken for write whenever
modifying the mount hash or associated fields, and may be taken for read when
performing mount hash lookups.
A new lock is added for the mnt-id allocator, so it doesn't need to take
the heavy vfsmount write-lock.
The number of atomics should remain the same for fastpath rlock cases, though
code would be slightly slower due to per-cpu access. Scalability is not not be
much improved in common cases yet, due to other locks (ie. dcache_lock) getting
in the way. However path lookups crossing mountpoints should be one case where
scalability is improved (currently requiring the global lock).
The slowpath is slower due to use of brlock. On a 64 core, 64 socket, 32 node
Altix system (high latency to remote nodes), a simple umount microbenchmark
(mount --bind mnt mnt2 ; umount mnt2 loop 1000 times), before this patch it
took 6.8s, afterwards took 7.1s, about 5% slower.
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2010-08-18 02:37:39 +08:00
|
|
|
br_read_unlock(vfsmount_lock);
|
2010-01-31 02:16:21 +08:00
|
|
|
return res;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(path_is_under);
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
void d_genocide(struct dentry *root)
|
|
|
|
{
|
|
|
|
struct dentry *this_parent = root;
|
|
|
|
struct list_head *next;
|
|
|
|
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
repeat:
|
|
|
|
next = this_parent->d_subdirs.next;
|
|
|
|
resume:
|
|
|
|
while (next != &this_parent->d_subdirs) {
|
|
|
|
struct list_head *tmp = next;
|
[PATCH] shrink dentry struct
Some long time ago, dentry struct was carefully tuned so that on 32 bits
UP, sizeof(struct dentry) was exactly 128, ie a power of 2, and a multiple
of memory cache lines.
Then RCU was added and dentry struct enlarged by two pointers, with nice
results for SMP, but not so good on UP, because breaking the above tuning
(128 + 8 = 136 bytes)
This patch reverts this unwanted side effect, by using an union (d_u),
where d_rcu and d_child are placed so that these two fields can share their
memory needs.
At the time d_free() is called (and d_rcu is really used), d_child is known
to be empty and not touched by the dentry freeing.
Lockless lookups only access d_name, d_parent, d_lock, d_op, d_flags (so
the previous content of d_child is not needed if said dentry was unhashed
but still accessed by a CPU because of RCU constraints)
As dentry cache easily contains millions of entries, a size reduction is
worth the extra complexity of the ugly C union.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Maneesh Soni <maneesh@in.ibm.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Ian Kent <raven@themaw.net>
Cc: Paul Jackson <pj@sgi.com>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: James Morris <jmorris@namei.org>
Cc: Stephen Smalley <sds@epoch.ncsc.mil>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 17:03:32 +08:00
|
|
|
struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
|
2005-04-17 06:20:36 +08:00
|
|
|
next = tmp->next;
|
|
|
|
if (d_unhashed(dentry)||!dentry->d_inode)
|
|
|
|
continue;
|
|
|
|
if (!list_empty(&dentry->d_subdirs)) {
|
|
|
|
this_parent = dentry;
|
|
|
|
goto repeat;
|
|
|
|
}
|
|
|
|
atomic_dec(&dentry->d_count);
|
|
|
|
}
|
|
|
|
if (this_parent != root) {
|
[PATCH] shrink dentry struct
Some long time ago, dentry struct was carefully tuned so that on 32 bits
UP, sizeof(struct dentry) was exactly 128, ie a power of 2, and a multiple
of memory cache lines.
Then RCU was added and dentry struct enlarged by two pointers, with nice
results for SMP, but not so good on UP, because breaking the above tuning
(128 + 8 = 136 bytes)
This patch reverts this unwanted side effect, by using an union (d_u),
where d_rcu and d_child are placed so that these two fields can share their
memory needs.
At the time d_free() is called (and d_rcu is really used), d_child is known
to be empty and not touched by the dentry freeing.
Lockless lookups only access d_name, d_parent, d_lock, d_op, d_flags (so
the previous content of d_child is not needed if said dentry was unhashed
but still accessed by a CPU because of RCU constraints)
As dentry cache easily contains millions of entries, a size reduction is
worth the extra complexity of the ugly C union.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Maneesh Soni <maneesh@in.ibm.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Ian Kent <raven@themaw.net>
Cc: Paul Jackson <pj@sgi.com>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: James Morris <jmorris@namei.org>
Cc: Stephen Smalley <sds@epoch.ncsc.mil>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 17:03:32 +08:00
|
|
|
next = this_parent->d_u.d_child.next;
|
2005-04-17 06:20:36 +08:00
|
|
|
atomic_dec(&this_parent->d_count);
|
|
|
|
this_parent = this_parent->d_parent;
|
|
|
|
goto resume;
|
|
|
|
}
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* find_inode_number - check for dentry with name
|
|
|
|
* @dir: directory to check
|
|
|
|
* @name: Name to find.
|
|
|
|
*
|
|
|
|
* Check whether a dentry already exists for the given name,
|
|
|
|
* and return the inode number if it has an inode. Otherwise
|
|
|
|
* 0 is returned.
|
|
|
|
*
|
|
|
|
* This routine is used to post-process directory listings for
|
|
|
|
* filesystems using synthetic inode numbers, and is necessary
|
|
|
|
* to keep getcwd() working.
|
|
|
|
*/
|
|
|
|
|
|
|
|
ino_t find_inode_number(struct dentry *dir, struct qstr *name)
|
|
|
|
{
|
|
|
|
struct dentry * dentry;
|
|
|
|
ino_t ino = 0;
|
|
|
|
|
2006-03-31 18:31:43 +08:00
|
|
|
dentry = d_hash_and_lookup(dir, name);
|
|
|
|
if (dentry) {
|
2005-04-17 06:20:36 +08:00
|
|
|
if (dentry->d_inode)
|
|
|
|
ino = dentry->d_inode->i_ino;
|
|
|
|
dput(dentry);
|
|
|
|
}
|
|
|
|
return ino;
|
|
|
|
}
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(find_inode_number);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
static __initdata unsigned long dhash_entries;
|
|
|
|
static int __init set_dhash_entries(char *str)
|
|
|
|
{
|
|
|
|
if (!str)
|
|
|
|
return 0;
|
|
|
|
dhash_entries = simple_strtoul(str, &str, 0);
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
__setup("dhash_entries=", set_dhash_entries);
|
|
|
|
|
|
|
|
static void __init dcache_init_early(void)
|
|
|
|
{
|
|
|
|
int loop;
|
|
|
|
|
|
|
|
/* If hashes are distributed across NUMA nodes, defer
|
|
|
|
* hash allocation until vmalloc space is available.
|
|
|
|
*/
|
|
|
|
if (hashdist)
|
|
|
|
return;
|
|
|
|
|
|
|
|
dentry_hashtable =
|
|
|
|
alloc_large_system_hash("Dentry cache",
|
|
|
|
sizeof(struct hlist_head),
|
|
|
|
dhash_entries,
|
|
|
|
13,
|
|
|
|
HASH_EARLY,
|
|
|
|
&d_hash_shift,
|
|
|
|
&d_hash_mask,
|
|
|
|
0);
|
|
|
|
|
|
|
|
for (loop = 0; loop < (1 << d_hash_shift); loop++)
|
|
|
|
INIT_HLIST_HEAD(&dentry_hashtable[loop]);
|
|
|
|
}
|
|
|
|
|
2007-10-17 14:26:30 +08:00
|
|
|
static void __init dcache_init(void)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
int loop;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* A constructor could be added for stable state like the lists,
|
|
|
|
* but it is probably not worth it because of the cache nature
|
|
|
|
* of the dcache.
|
|
|
|
*/
|
2007-05-07 05:49:57 +08:00
|
|
|
dentry_cache = KMEM_CACHE(dentry,
|
|
|
|
SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-07-17 19:03:17 +08:00
|
|
|
register_shrinker(&dcache_shrinker);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* Hash may have been set up in dcache_init_early */
|
|
|
|
if (!hashdist)
|
|
|
|
return;
|
|
|
|
|
|
|
|
dentry_hashtable =
|
|
|
|
alloc_large_system_hash("Dentry cache",
|
|
|
|
sizeof(struct hlist_head),
|
|
|
|
dhash_entries,
|
|
|
|
13,
|
|
|
|
0,
|
|
|
|
&d_hash_shift,
|
|
|
|
&d_hash_mask,
|
|
|
|
0);
|
|
|
|
|
|
|
|
for (loop = 0; loop < (1 << d_hash_shift); loop++)
|
|
|
|
INIT_HLIST_HEAD(&dentry_hashtable[loop]);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* SLAB cache for __getname() consumers */
|
2006-12-07 12:33:20 +08:00
|
|
|
struct kmem_cache *names_cachep __read_mostly;
|
2010-01-06 04:45:18 +08:00
|
|
|
EXPORT_SYMBOL(names_cachep);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
EXPORT_SYMBOL(d_genocide);
|
|
|
|
|
|
|
|
void __init vfs_caches_init_early(void)
|
|
|
|
{
|
|
|
|
dcache_init_early();
|
|
|
|
inode_init_early();
|
|
|
|
}
|
|
|
|
|
|
|
|
void __init vfs_caches_init(unsigned long mempages)
|
|
|
|
{
|
|
|
|
unsigned long reserve;
|
|
|
|
|
|
|
|
/* Base hash sizes on available memory, with a reserve equal to
|
|
|
|
150% of current kernel size */
|
|
|
|
|
|
|
|
reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
|
|
|
|
mempages -= reserve;
|
|
|
|
|
|
|
|
names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
|
2007-07-20 09:11:58 +08:00
|
|
|
SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-10-17 14:26:30 +08:00
|
|
|
dcache_init();
|
|
|
|
inode_init();
|
2005-04-17 06:20:36 +08:00
|
|
|
files_init(mempages);
|
2007-10-17 14:26:30 +08:00
|
|
|
mnt_init();
|
2005-04-17 06:20:36 +08:00
|
|
|
bdev_cache_init();
|
|
|
|
chrdev_init();
|
|
|
|
}
|