xfs_repair — repair an XFS filesystem
xfs_repair [ -dfLPv ] [ -n | -e ] [ -m maxmem ] [ -c subopt=value ] [ -o subopt[=value] ] [ -t interval ] [ -l logdev ] [ -r rtdev ] device
xfs_repair repairs corrupt or damaged XFS filesystems (see xfs(5)). The filesystem is specified using the device argument which should be the device name of the disk partition or volume containing the filesystem. If given the name of a block device, xfs_repair will attempt to find the raw device associated with the specified block device and will use the raw device instead.
Regardless, the filesystem to be repaired must be unmounted, otherwise, the resulting filesystem may be inconsistent or corrupt.
Specifies that the filesystem image to be processed is stored in a regular file at device (see the mkfs.xfs -d file option). This might happen if an image copy of a filesystem has been copied or written into an ordinary file. This option implies that any external log or realtime section is also in an ordinary file.
Force Log Zeroing. Forces xfs_repair to zero the log even if it is dirty (contains metadata changes). When using this option the filesystem will likely appear to be corrupt, and can cause the loss of user files and/or data. See the Dirty Logs section for more information.
- -l logdev
Specifies the device special file where the filesystem's external log resides. Only for those filesystems which use an external log. See the mkfs.xfs -l option, and refer to xfs(5) for a detailed description of the XFS log.
- -r rtdev
Specifies the device special file where the filesystem's realtime section resides. Only for those filesystems which use a realtime section. See the mkfs.xfs -r option, and refer to xfs(5) for a detailed description of the XFS realtime section.
No modify mode. Specifies that xfs_repair should not modify the filesystem but should only scan the filesystem and indicate what repairs would have been made. This option cannot be used together with -e.
Disable prefetching of inode and directory blocks. Use this option if you find xfs_repair gets stuck and stops proceeding. Interrupting a stuck xfs_repair is safe.
- -m maxmem
Specifies the approximate maximum amount of memory, in megabytes, to use for xfs_repair. xfs_repair has its own internal block cache which will scale out up to the lesser of the process's virtual address limit or about 75% of the system's physical RAM. This option overrides these limits.
NOTE: These memory limits are only approximate and may use more than the specified limit.
- -c subopt=value
Change filesystem parameters. Refer to xfs_admin(8) for information on changing filesystem parameters.
Override what the program might conclude about the filesystem if left to its own devices.
- The suboptions supported are:
overrides the default buffer cache hash size. The total number of buffer cache entries are limited to 8 times this amount. The default size is set to use up the remainder of 75% of the system's physical RAM size.
This creates additional processing threads to parallel process AGs that span multiple concat units. This can significantly reduce repair times on concat based filesystems.
Check the filesystem even if geometry information could not be validated. Geometry information can not be validated if only a single allocation group exists and thus we do not have a backup superblock available, or if there are two allocation groups and the two superblocks do not agree on the filesystem geometry. Only use this option if you validated the geometry yourself and know what you are doing. If In doubt run in no modify mode first.
- -t interval
Modify reporting interval, specified in seconds. During long runs xfs_repair outputs its progress every 15 minutes. Reporting is only activated when ag_stride is enabled.
Verbose output. May be specified multiple times to increase verbosity.
Repair dangerously. Allow xfs_repair to repair an XFS filesystem mounted read only. This is typically done on a root filesystem from single user mode, immediately followed by a reboot.
If any metadata corruption was repaired, the status returned is 4 instead of the usual 0. This option cannot be used together with -n.
Prints the version number and exits.
Inconsistencies corrected include the following:
- Inode and inode blockmap (addressing) checks: bad magic number in inode, bad magic numbers in inode blockmap blocks, extents out of order, incorrect number of records in inode blockmap blocks, blocks claimed that are not in a legal data area of the filesystem, blocks that are claimed by more than one inode.
- Inode allocation map checks: bad magic number in inode map blocks, inode state as indicated by map (free or in-use) inconsistent with state indicated by the inode, inodes referenced by the filesystem that do not appear in the inode allocation map, inode allocation map referencing blocks that do not appear to contain inodes.
- Size checks: number of blocks claimed by inode inconsistent with inode size, directory size not block aligned, inode size not consistent with inode format.
- Directory checks: bad magic numbers in directory blocks, incorrect number of entries in a directory block, bad freespace information in a directory leaf block, entry pointing to an unallocated (free) or out of range inode, overlapping entries, missing or incorrect dot and dotdot entries, entries out of hashvalue order, incorrect internal directory pointers, directory type not consistent with inode format and size.
- Pathname checks: files or directories not referenced by a pathname starting from the filesystem root, illegal pathname components.
- Link count checks: link counts that do not agree with the number of directory references to the inode.
- Freemap checks: blocks claimed free by the freemap but also claimed by an inode, blocks unclaimed by any inode but not appearing in the freemap.
- Super Block checks: total free block and/or free i-node count incorrect, filesystem geometry inconsistent, secondary and primary superblocks contradictory.
Orphaned files and directories (allocated, in-use but unreferenced) are reconnected by placing them in the lost+found directory. The name assigned is the inode number.
xfs_repair aborts on most disk I/O errors. Therefore, if you are trying to repair a filesystem that was damaged due to a disk drive failure, steps should be taken to ensure that all blocks in the filesystem are readable and writable before attempting to use xfs_repair to repair the filesystem. A possible method is using dd(8) to copy the data onto a good disk.
The directory lost+found does not have to already exist in the filesystem being repaired. If the directory does not exist, it is automatically created if required. If it already exists, it will be checked for consistency and if valid will be used for additional orphaned files. Invalid lost+found directories are removed and recreated. Existing files in a valid lost+found are not removed or renamed.
XFS has both primary and secondary superblocks. xfs_repair uses information in the primary superblock to automatically find and validate the primary superblock against the secondary superblocks before proceeding. Should the primary be too corrupted to be useful in locating the secondary superblocks, the program scans the filesystem until it finds and validates some secondary superblocks. At that point, it generates a primary superblock.
If quotas are in use, it is possible that xfs_repair will clear some or all of the filesystem quota information. If so, the program issues a warning just before it terminates. If all quota information is lost, quotas are disabled and the program issues a warning to that effect.
Note that xfs_repair does not check the validity of quota limits. It is recommended that you check the quota limit information manually after xfs_repair. Also, space usage information is automatically regenerated the next time the filesystem is mounted with quotas turned on, so the next quota mount of the filesystem may take some time.
xfs_repair issues informative messages as it proceeds indicating what it has found that is abnormal or any corrective action that it has taken. Most of the messages are completely understandable only to those who are knowledgeable about the structure of the filesystem. Some of the more common messages are explained here. Note that the language of the messages is slightly different if xfs_repair is run in no-modify mode because the program is not changing anything on disk. No-modify mode indicates what it would do to repair the filesystem if run without the no-modify flag.
disconnected inode ino, moving to lost+found
An inode numbered ino was not connected to the filesystem directory tree and was reconnected to the lost+found directory. The inode is assigned the name of its inode number (ino). If a lost+found directory does not exist, it is automatically created.
disconnected dir inode ino, moving to lost+found
As above only the inode is a directory inode. If a directory inode is attached to lost+found, all of its children (if any) stay attached to the directory and therefore get automatically reconnected when the directory is reconnected.
imap claims in-use inode ino is free, correcting imap
The inode allocation map thinks that inode ino is free whereas examination of the inode indicates that the inode may be in use (although it may be disconnected). The program updates the inode allocation map.
imap claims free inode ino is in use, correcting imap
The inode allocation map thinks that inode ino is in use whereas examination of the inode indicates that the inode is not in use and therefore is free. The program updates the inode allocation map.
resetting inode ino nlinks from x to y
The program detected a mismatch between the number of valid directory entries referencing inode ino and the number of references recorded in the inode and corrected the the number in the inode.
fork-type fork in ino ino claims used block bno
Inode ino claims a block bno that is used (claimed) by either another inode or the filesystem itself for metadata storage. The fork-type is either data or attr indicating whether the problem lies in the portion of the inode that tracks regular data or the portion of the inode that stores XFS attributes. If the inode is a real-time (rt) inode, the message says so. Any inode that claims blocks used by the filesystem is deleted. If two or more inodes claim the same block, they are both deleted.
fork-type fork in ino ino claims dup extent ...
Inode ino claims a block in an extent known to be claimed more than once. The offset in the inode, start and length of the extent is given. The message is slightly different if the inode is a real-time (rt) inode and the extent is therefore a real-time (rt) extent.
inode ino - bad extent ...
An extent record in the blockmap of inode ino claims blocks that are out of the legal range of the filesystem. The message supplies the start, end, and file offset of the extent. The message is slightly different if the extent is a real-time (rt) extent.
bad fork-type fork in inode ino
There was something structurally wrong or inconsistent with the data structures that map offsets to filesystem blocks.
cleared inode ino
There was something wrong with the inode that was uncorrectable so the program freed the inode. This usually happens because the inode claims blocks that are used by something else or the inode itself is badly corrupted. Typically, this message is preceded by one or more messages indicating why the inode needed to be cleared.
bad attribute fork in inode ino, clearing attr fork
There was something wrong with the portion of the inode that stores XFS attributes (the attribute fork) so the program reset the attribute fork. As a result of this, all attributes on that inode are lost.
correcting nextents for inode ino, was x - counted y
The program found that the number of extents used to store the data in the inode is wrong and corrected the number. The message refers to nextents if the count is wrong on the number of extents used to store attribute information.
entry name in dir dir_ino not consistent with .. value (xxxx) in dir ino ino, junking entry name in directory inode dir_ino
The entry name in directory inode dir_ino references a directory inode ino. However, the .. entry in directory ino does not point back to directory dir_ino, so the program deletes the entry name in directory inode dir_ino. If the directory inode ino winds up becoming a disconnected inode as a result of this, it is moved to lost+found later.
entry name in dir dir_ino references already connected dir ino ino, junking entry name in directory inode dir_ino
The entry name in directory inode dir_ino points to a directory inode ino that is known to be a child of another directory. Therefore, the entry is invalid and is deleted. This message refers to an entry in a small directory. If this were a large directory, the last phrase would read "will clear entry".
entry references free inode ino in directory dir_ino, will clear entry
An entry in directory inode dir_ino references an inode ino that is known to be free. The entry is therefore invalid and is deleted. This message refers to a large directory. If the directory were small, the message would read "junking entry ...".
xfs_repair -n (no modify mode) will return a status of 1 if filesystem corruption was detected and 0 if no filesystem corruption was detected. xfs_repair run without the -n option will always return a status code of 0 if it completes without problems, unless the flag -e is used. If it is used, then status 4 is reported when any issue with the filesystem was found, but could be fixed. If a runtime error is encountered during operation, it will return a status of 1. In this case, xfs_repair should be restarted. If xfs_repair is unable to proceed due to a dirty log, it will return a status of 2. See below.
Due to the design of the XFS log, a dirty log can only be replayed by the kernel, on a machine having the same CPU architecture as the machine which was writing to the log. xfs_repair cannot replay a dirty log and will exit with a status code of 2 when it detects a dirty log.
In this situation, the log can be replayed by mounting and immediately unmounting the filesystem on the same class of machine that crashed. Please make sure that the machine's hardware is reliable before replaying to avoid compounding the problems.
If mounting fails, the log can be erased by running xfs_repair with the -L option. All metadata updates in progress at the time of the crash will be lost, which may cause significant filesystem damage. This should only be used as a last resort.
The filesystem to be checked and repaired must have been unmounted cleanly using normal system administration procedures (the umount(8) command or system shutdown), not as a result of a crash or system reset. If the filesystem has not been unmounted cleanly, mount it and unmount it cleanly before running xfs_repair.
xfs_repair does not do a thorough job on XFS extended attributes. The structure of the attribute fork will be consistent, but only the contents of attribute forks that will fit into an inode are checked. This limitation will be fixed in the future.
The no-modify mode (-n option) is not completely accurate. It does not catch inconsistencies in the freespace and inode maps, particularly lost blocks or subtly corrupted maps (trees).
The no-modify mode can generate repeated warnings about the same problems because it cannot fix the problems as they are encountered.
If a filesystem fails to be repaired, a metadump image can be generated with xfs_metadump(8) and be sent to an XFS maintainer to be analysed and xfs_repair fixed and/or improved.
dd(1), mkfs.xfs(8), umount(8), xfs_admin(8), xfs_metadump(8), xfs(5).
fsck.xfs(8), ioctl_xfs_scrub_metadata(2), xfs_admin(8), xfs_db(8), xfs_mdrestore(8), xfs_metadump(8), xfs_scrub(8).