mirror of https://gitee.com/openkylin/linux.git
d634cc15e8
The recent change to locks_remove_flock code in fs/locks.c changes how byte range locks are removed from closing files, which shows up a bug in cifs. The assumption in the cifs code was that the close call sent to the server would remove any pending locks on the server on this file, but that is no longer safe as the fs/locks.c code on the client wants unlock of 0 to PATH_MAX to remove all locks (at least from this client, it is not possible AFAIK to remove all locks from other clients made to the server copy of the file). Note that cifs locks are different from posix locks - and it is not possible to map posix locks perfectly on the wire yet, due to restrictions of the cifs network protocol, even to Samba without adding a new request type to the network protocol (which we plan to do for Samba 3.0.21 within a few months), but the local client will have the correct, posix view, of the lock in most cases. The correct fix for cifs for this would involve a bigger change than I would like to do this late in the 2.6.13-rc cycle - and would involve cifs keeping track of all unmerged (uncoalesced) byte range locks for each remote inode and scanning that list to remove locks that intersect or fall wholly within the range - locks that intersect may have to be reaquired with the smaller, remaining range. Signed-off-by: Steve French <sfrench@us.ibm.com> Signed-off-by: Dave Kleikamp <shaggy@austin.ibm.com> Signed-off-by: Linus Torvalds <torvalds@osdl.org> |
||
---|---|---|
.. | ||
AUTHORS | ||
CHANGES | ||
Makefile | ||
README | ||
TODO | ||
asn1.c | ||
cifs_debug.c | ||
cifs_debug.h | ||
cifs_fs_sb.h | ||
cifs_unicode.c | ||
cifs_unicode.h | ||
cifs_uniupr.h | ||
cifsencrypt.c | ||
cifsencrypt.h | ||
cifsfs.c | ||
cifsfs.h | ||
cifsglob.h | ||
cifspdu.h | ||
cifsproto.h | ||
cifssmb.c | ||
connect.c | ||
dir.c | ||
fcntl.c | ||
file.c | ||
inode.c | ||
ioctl.c | ||
link.c | ||
md4.c | ||
md5.c | ||
md5.h | ||
misc.c | ||
netmisc.c | ||
nterr.c | ||
nterr.h | ||
ntlmssp.h | ||
readdir.c | ||
rfc1002pdu.h | ||
smbdes.c | ||
smbencrypt.c | ||
smberr.h | ||
transport.c | ||
xattr.c |
README
The CIFS VFS support for Linux supports many advanced network filesystem features such as heirarchical dfs like namespace, hardlinks, locking and more. It was designed to comply with the SNIA CIFS Technical Reference (which supersedes the 1992 X/Open SMB Standard) as well as to perform best practice practical interoperability with Windows 2000, Windows XP, Samba and equivalent servers. For questions or bug reports please contact: sfrench@samba.org (sfrench@us.ibm.com) Build instructions: ================== For Linux 2.4: 1) Get the kernel source (e.g.from http://www.kernel.org) and download the cifs vfs source (see the project page at http://us1.samba.org/samba/Linux_CIFS_client.html) and change directory into the top of the kernel directory then patch the kernel (e.g. "patch -p1 < cifs_24.patch") to add the cifs vfs to your kernel configure options if it has not already been added (e.g. current SuSE and UL users do not need to apply the cifs_24.patch since the cifs vfs is already in the kernel configure menu) and then mkdir linux/fs/cifs and then copy the current cifs vfs files from the cifs download to your kernel build directory e.g. cp <cifs_download_dir>/fs/cifs/* to <kernel_download_dir>/fs/cifs 2) make menuconfig (or make xconfig) 3) select cifs from within the network filesystem choices 4) save and exit 5) make dep 6) make modules (or "make" if CIFS VFS not to be built as a module) For Linux 2.6: 1) Download the kernel (e.g. from http://www.kernel.org) and change directory into the top of the kernel directory tree (e.g. /usr/src/linux-2.5.73) 2) make menuconfig (or make xconfig) 3) select cifs from within the network filesystem choices 4) save and exit 5) make Installation instructions: ========================= If you have built the CIFS vfs as module (successfully) simply type "make modules_install" (or if you prefer, manually copy the file to the modules directory e.g. /lib/modules/2.4.10-4GB/kernel/fs/cifs/cifs.o). If you have built the CIFS vfs into the kernel itself, follow the instructions for your distribution on how to install a new kernel (usually you would simply type "make install"). If you do not have the utility mount.cifs (in the Samba 3.0 source tree and on the CIFS VFS web site) copy it to the same directory in which mount.smbfs and similar files reside (usually /sbin). Although the helper software is not required, mount.cifs is recommended. Eventually the Samba 3.0 utility program "net" may also be helpful since it may someday provide easier mount syntax for users who are used to Windows e.g. net use <mount point> <UNC name or cifs URL> Note that running the Winbind pam/nss module (logon service) on all of your Linux clients is useful in mapping Uids and Gids consistently across the domain to the proper network user. The mount.cifs mount helper can be trivially built from Samba 3.0 or later source e.g. by executing: gcc samba/source/client/mount.cifs.c -o mount.cifs If cifs is built as a module, then the size and number of network buffers and maximum number of simultaneous requests to one server can be configured. Changing these from their defaults is not recommended. By executing modinfo modinfo kernel/fs/cifs/cifs.ko on kernel/fs/cifs/cifs.ko the list of configuration changes that can be made at module initialization time (by running insmod cifs.ko) can be seen. Allowing User Mounts ==================== To permit users to mount and unmount over directories they own is possible with the cifs vfs. A way to enable such mounting is to mark the mount.cifs utility as suid (e.g. "chmod +s /sbin/mount.cifs). To enable users to umount shares they mount requires 1) mount.cifs version 1.4 or later 2) an entry for the share in /etc/fstab indicating that a user may unmount it e.g. //server/usersharename /mnt/username cifs user 0 0 Note that when the mount.cifs utility is run suid (allowing user mounts), in order to reduce risks, the "nosuid" mount flag is passed in on mount to disallow execution of an suid program mounted on the remote target. When mount is executed as root, nosuid is not passed in by default, and execution of suid programs on the remote target would be enabled by default. This can be changed, as with nfs and other filesystems, by simply specifying "nosuid" among the mount options. For user mounts though to be able to pass the suid flag to mount requires rebuilding mount.cifs with the following flag: gcc samba/source/client/mount.cifs.c -DCIFS_ALLOW_USR_SUID -o mount.cifs There is a corresponding manual page for cifs mounting in the Samba 3.0 and later source tree in docs/manpages/mount.cifs.8 Allowing User Unmounts ====================== To permit users to ummount directories that they have user mounted (see above), the utility umount.cifs may be used. It may be invoked directly, or if umount.cifs is placed in /sbin, umount can invoke the cifs umount helper (at least for most versions of the umount utility) for umount of cifs mounts, unless umount is invoked with -i (which will avoid invoking a umount helper). As with mount.cifs, to enable user unmounts umount.cifs must be marked as suid (e.g. "chmod +s /sbin/umount.cifs") or equivalent (some distributions allow adding entries to a file to the /etc/permissions file to achieve the equivalent suid effect). For this utility to succeed the target path must be a cifs mount, and the uid of the current user must match the uid of the user who mounted the resource. Also note that the customary way of allowing user mounts and unmounts is (instead of using mount.cifs and unmount.cifs as suid) to add a line to the file /etc/fstab for each //server/share you wish to mount, but this can become unwieldy when potential mount targets include many or unpredictable UNC names. Samba Considerations ==================== To get the maximum benefit from the CIFS VFS, we recommend using a server that supports the SNIA CIFS Unix Extensions standard (e.g. Samba 2.2.5 or later or Samba 3.0) but the CIFS vfs works fine with a wide variety of CIFS servers. Note that uid, gid and file permissions will display default values if you do not have a server that supports the Unix extensions for CIFS (such as Samba 2.2.5 or later). To enable the Unix CIFS Extensions in the Samba server, add the line: unix extensions = yes to your smb.conf file on the server. Note that the following smb.conf settings are also useful (on the Samba server) when the majority of clients are Unix or Linux: case sensitive = yes delete readonly = yes ea support = yes Note that server ea support is required for supporting xattrs from the Linux cifs client, and that EA support is present in later versions of Samba (e.g. 3.0.6 and later (also EA support works in all versions of Windows, at least to shares on NTFS filesystems). Extended Attribute (xattr) support is an optional feature of most Linux filesystems which may require enabling via make menuconfig. Client support for extended attributes (user xattr) can be disabled on a per-mount basis by specifying "nouser_xattr" on mount. The CIFS client can get and set POSIX ACLs (getfacl, setfacl) to Samba servers version 3.10 and later. Setting POSIX ACLs requires enabling both XATTR and then POSIX support in the CIFS configuration options when building the cifs module. POSIX ACL support can be disabled on a per mount basic by specifying "noacl" on mount. Some administrators may want to change Samba's smb.conf "map archive" and "create mask" parameters from the default. Unless the create mask is changed newly created files can end up with an unnecessarily restrictive default mode, which may not be what you want, although if the CIFS Unix extensions are enabled on the server and client, subsequent setattr calls (e.g. chmod) can fix the mode. Note that creating special devices (mknod) remotely may require specifying a mkdev function to Samba if you are not using Samba 3.0.6 or later. For more information on these see the manual pages ("man smb.conf") on the Samba server system. Note that the cifs vfs, unlike the smbfs vfs, does not read the smb.conf on the client system (the few optional settings are passed in on mount via -o parameters instead). Note that Samba 2.2.7 or later includes a fix that allows the CIFS VFS to delete open files (required for strict POSIX compliance). Windows Servers already supported this feature. Samba server does not allow symlinks that refer to files outside of the share, so in Samba versions prior to 3.0.6, most symlinks to files with absolute paths (ie beginning with slash) such as: ln -s /mnt/foo bar would be forbidden. Samba 3.0.6 server or later includes the ability to create such symlinks safely by converting unsafe symlinks (ie symlinks to server files that are outside of the share) to a samba specific format on the server that is ignored by local server applications and non-cifs clients and that will not be traversed by the Samba server). This is opaque to the Linux client application using the cifs vfs. Absolute symlinks will work to Samba 3.0.5 or later, but only for remote clients using the CIFS Unix extensions, and will be invisbile to Windows clients and typically will not affect local applications running on the same server as Samba. Use instructions: ================ Once the CIFS VFS support is built into the kernel or installed as a module (cifs.o), you can use mount syntax like the following to access Samba or Windows servers: mount -t cifs //9.53.216.11/e$ /mnt -o user=myname,pass=mypassword Before -o the option -v may be specified to make the mount.cifs mount helper display the mount steps more verbosely. After -o the following commonly used cifs vfs specific options are supported: user=<username> pass=<password> domain=<domain name> Other cifs mount options are described below. Use of TCP names (in addition to ip addresses) is available if the mount helper (mount.cifs) is installed. If you do not trust the server to which are mounted, or if you do not have cifs signing enabled (and the physical network is insecure), consider use of the standard mount options "noexec" and "nosuid" to reduce the risk of running an altered binary on your local system (downloaded from a hostile server or altered by a hostile router). Although mounting using format corresponding to the CIFS URL specification is not possible in mount.cifs yet, it is possible to use an alternate format for the server and sharename (which is somewhat similar to NFS style mount syntax) instead of the more widely used UNC format (i.e. \\server\share): mount -t cifs tcp_name_of_server:share_name /mnt -o user=myname,pass=mypasswd When using the mount helper mount.cifs, passwords may be specified via alternate mechanisms, instead of specifying it after -o using the normal "pass=" syntax on the command line: 1) By including it in a credential file. Specify credentials=filename as one of the mount options. Credential files contain two lines username=someuser password=your_password 2) By specifying the password in the PASSWD environment variable (similarly the user name can be taken from the USER environment variable). 3) By specifying the password in a file by name via PASSWD_FILE 4) By specifying the password in a file by file descriptor via PASSWD_FD If no password is provided, mount.cifs will prompt for password entry Restrictions ============ Servers must support the NTLM SMB dialect (which is the most recent, supported by Samba and Windows NT version 4, 2000 and XP and many other SMB/CIFS servers) Servers must support either "pure-TCP" (port 445 TCP/IP CIFS connections) or RFC 1001/1002 support for "Netbios-Over-TCP/IP." Neither of these is likely to be a problem as most servers support this. IPv6 support is planned for the future, and is almost complete. Valid filenames differ between Windows and Linux. Windows typically restricts filenames which contain certain reserved characters (e.g.the character : which is used to delimit the beginning of a stream name by Windows), while Linux allows a slightly wider set of valid characters in filenames. Windows servers can remap such characters when an explicit mapping is specified in the Server's registry. Samba starting with version 3.10 will allow such filenames (ie those which contain valid Linux characters, which normally would be forbidden for Windows/CIFS semantics) as long as the server is configured for Unix Extensions (and the client has not disabled /proc/fs/cifs/LinuxExtensionsEnabled). CIFS VFS Mount Options ====================== A partial list of the supported mount options follows: user The user name to use when trying to establish the CIFS session. password The user password. If the mount helper is installed, the user will be prompted for password if it is not supplied. ip The ip address of the target server unc The target server Universal Network Name (export) to mount. domain Set the SMB/CIFS workgroup name prepended to the username during CIFS session establishment uid If CIFS Unix extensions are not supported by the server this overrides the default uid for inodes. For mounts to servers which do support the CIFS Unix extensions, such as a properly configured Samba server, the server provides the uid, gid and mode. For servers which do not support the Unix extensions, the default uid (and gid) returned on lookup of existing files is the uid (gid) of the person who executed the mount (root, except when mount.cifs is configured setuid for user mounts) unless the "uid=" (gid) mount option is specified. For the uid (gid) of newly created files and directories, ie files created since the last mount of the server share, the expected uid (gid) is cached as as long as the inode remains in memory on the client. Also note that permission checks (authorization checks) on accesses to a file occur at the server, but there are cases in which an administrator may want to restrict at the client as well. For those servers which do not report a uid/gid owner (such as Windows), permissions can also be checked at the client, and a crude form of client side permission checking can be enabled by specifying file_mode and dir_mode on the client gid If CIFS Unix extensions are not supported by the server this overrides the default gid for inodes. file_mode If CIFS Unix extensions are not supported by the server this overrides the default mode for file inodes. dir_mode If CIFS Unix extensions are not supported by the server this overrides the default mode for directory inodes. port attempt to contact the server on this tcp port, before trying the usual ports (port 445, then 139). iocharset Codepage used to convert local path names to and from Unicode. Unicode is used by default for network path names if the server supports it. If iocharset is not specified then the nls_default specified during the local client kernel build will be used. If server does not support Unicode, this parameter is unused. rsize default read size wsize default write size rw mount the network share read-write (note that the server may still consider the share read-only) ro mount network share read-only version used to distinguish different versions of the mount helper utility (not typically needed) sep if first mount option (after the -o), overrides the comma as the separator between the mount parms. e.g. -o user=myname,password=mypassword,domain=mydom could be passed instead with period as the separator by -o sep=.user=myname.password=mypassword.domain=mydom this might be useful when comma is contained within username or password or domain. This option is less important when the cifs mount helper cifs.mount (version 1.1 or later) is used. nosuid Do not allow remote executables with the suid bit program to be executed. This is only meaningful for mounts to servers such as Samba which support the CIFS Unix Extensions. If you do not trust the servers in your network (your mount targets) it is recommended that you specify this option for greater security. exec Permit execution of binaries on the mount. noexec Do not permit execution of binaries on the mount. dev Recognize block devices on the remote mount. nodev Do not recognize devices on the remote mount. suid Allow remote files on this mountpoint with suid enabled to be executed (default for mounts when executed as root, nosuid is default for user mounts). credentials Although ignored by the cifs kernel component, it is used by the mount helper, mount.cifs. When mount.cifs is installed it opens and reads the credential file specified in order to obtain the userid and password arguments which are passed to the cifs vfs. guest Although ignored by the kernel component, the mount.cifs mount helper will not prompt the user for a password if guest is specified on the mount options. If no password is specified a null password will be used. perm Client does permission checks (vfs_permission check of uid and gid of the file against the mode and desired operation), Note that this is in addition to the normal ACL check on the target machine done by the server software. Client permission checking is enabled by default. noperm Client does not do permission checks. This can expose files on this mount to access by other users on the local client system. It is typically only needed when the server supports the CIFS Unix Extensions but the UIDs/GIDs on the client and server system do not match closely enough to allow access by the user doing the mount. Note that this does not affect the normal ACL check on the target machine done by the server software (of the server ACL against the user name provided at mount time). serverino Use servers inode numbers instead of generating automatically incrementing inode numbers on the client. Although this will make it easier to spot hardlinked files (as they will have the same inode numbers) and inode numbers may be persistent, note that the server does not guarantee that the inode numbers are unique if multiple server side mounts are exported under a single share (since inode numbers on the servers might not be unique if multiple filesystems are mounted under the same shared higher level directory). Note that this requires that the server support the CIFS Unix Extensions as other servers do not return a unique IndexNumber on SMB FindFirst (most servers return zero as the IndexNumber). Parameter has no effect to Windows servers and others which do not support the CIFS Unix Extensions. noserverino Client generates inode numbers (rather than using the actual one from the server) by default. setuids If the CIFS Unix extensions are negotiated with the server the client will attempt to set the effective uid and gid of the local process on newly created files, directories, and devices (create, mkdir, mknod). nosetuids The client will not attempt to set the uid and gid on on newly created files, directories, and devices (create, mkdir, mknod) which will result in the server setting the uid and gid to the default (usually the server uid of the user who mounted the share). Letting the server (rather than the client) set the uid and gid is the default. This parameter has no effect if the CIFS Unix Extensions are not negotiated. netbiosname When mounting to servers via port 139, specifies the RFC1001 source name to use to represent the client netbios machine name when doing the RFC1001 netbios session initialize. direct Do not do inode data caching on files opened on this mount. This precludes mmaping files on this mount. In some cases with fast networks and little or no caching benefits on the client (e.g. when the application is doing large sequential reads bigger than page size without rereading the same data) this can provide better performance than the default behavior which caches reads (readahead) and writes (writebehind) through the local Linux client pagecache if oplock (caching token) is granted and held. Note that direct allows write operations larger than page size to be sent to the server. acl Allow setfacl and getfacl to manage posix ACLs if server supports them. (default) noacl Do not allow setfacl and getfacl calls on this mount user_xattr Allow getting and setting user xattrs as OS/2 EAs (extended attributes) to the server (default) e.g. via setfattr and getfattr utilities. nouser_xattr Do not allow getfattr/setfattr to get/set xattrs mapchars Translate six of the seven reserved characters (not backslash) *?<>|: to the remap range (above 0xF000), which also allows the CIFS client to recognize files created with such characters by Windows's POSIX emulation. This can also be useful when mounting to most versions of Samba (which also forbids creating and opening files whose names contain any of these seven characters). This has no effect if the server does not support Unicode on the wire. nomapchars Do not translate any of these seven characters (default). remount remount the share (often used to change from ro to rw mounts or vice versa) The mount.cifs mount helper also accepts a few mount options before -o including: -S take password from stdin (equivalent to setting the environment variable "PASSWD_FD=0" -V print mount.cifs version -? display simple usage information With recent 2.6 kernel versions of modutils, the version of the cifs kernel module can be displayed via modinfo. Misc /proc/fs/cifs Flags and Debug Info ======================================= Informational pseudo-files: DebugData Displays information about active CIFS sessions and shares, as well as the cifs.ko version. Stats Lists summary resource usage information as well as per share statistics, if CONFIG_CIFS_STATS in enabled in the kernel configuration. Configuration pseudo-files: MultiuserMount If set to one, more than one CIFS session to the same server ip address can be established if more than one uid accesses the same mount point and if the uids user/password mapping information is available. (default is 0) PacketSigningEnabled If set to one, cifs packet signing is enabled and will be used if the server requires it. If set to two, cifs packet signing is required even if the server considers packet signing optional. (default 1) cifsFYI If set to one, additional debug information is logged to the system error log. (default 0) ExtendedSecurity If set to one, SPNEGO session establishment is allowed which enables more advanced secure CIFS session establishment (default 0) NTLMV2Enabled If set to one, more secure password hashes are used when the server supports them and when kerberos is not negotiated (default 0) traceSMB If set to one, debug information is logged to the system error log with the start of smb requests and responses (default 0) LookupCacheEnable If set to one, inode information is kept cached for one second improving performance of lookups (default 1) OplockEnabled If set to one, safe distributed caching enabled. (default 1) LinuxExtensionsEnabled If set to one then the client will attempt to use the CIFS "UNIX" extensions which are optional protocol enhancements that allow CIFS servers to return accurate UID/GID information as well as support symbolic links. If you use servers such as Samba that support the CIFS Unix extensions but do not want to use symbolic link support and want to map the uid and gid fields to values supplied at mount (rather than the actual values, then set this to zero. (default 1) These experimental features and tracing can be enabled by changing flags in /proc/fs/cifs (after the cifs module has been installed or built into the kernel, e.g. insmod cifs). To enable a feature set it to 1 e.g. to enable tracing to the kernel message log type: echo 1 > /proc/fs/cifs/cifsFYI and for more extensive tracing including the start of smb requests and responses echo 1 > /proc/fs/cifs/traceSMB Two other experimental features are under development and to test require enabling CONFIG_CIFS_EXPERIMENTAL More efficient write operations DNOTIFY fcntl: needed for support of directory change notification and perhaps later for file leases) Per share (per client mount) statistics are available in /proc/fs/cifs/Stats if the kernel was configured with cifs statistics enabled. The statistics represent the number of successful (ie non-zero return code from the server) SMB responses to some of the more common commands (open, delete, mkdir etc.). Also recorded is the total bytes read and bytes written to the server for that share. Note that due to client caching effects this can be less than the number of bytes read and written by the application running on the client. The statistics for the number of total SMBs and oplock breaks are different in that they represent all for that share, not just those for which the server returned success. Also note that "cat /proc/fs/cifs/DebugData" will display information about the active sessions and the shares that are mounted. Note: NTLMv2 enablement will not work since its implementation is not quite complete yet. Do not alter the ExtendedSecurity configuration value unless you are doing specific testing. Enabling extended security works to Windows 2000 Workstations and XP but not to Windows 2000 server or Samba since it does not usually send "raw NTLMSSP" (instead it sends NTLMSSP encapsulated in SPNEGO/GSSAPI, which support is not complete in the CIFS VFS yet).