XFS metadata consistency limits after LVM snapshot restoration

Answer to the Question

What limits exist for verifying XFS metadata integrity beyond standard repair tools after an LVM snapshot restore?

• Metadata area size: XFS stores its metadata in a dedicated block region. If a snapshot is taken while the filesystem is active, the copy may be incomplete if the region is larger than the snapshot size, leading to unrecoverable corruption.
• Journal presence: The journal must be fully captured in the snapshot. Without the complete log, xfs_repair cannot recover uncommitted changes, and the filesystem may refuse to mount.
• Atomicity of the snapshot: LVM snapshots copy the entire logical volume atomically, but if the volume is not fully synced (e.g., during heavy I/O), the metadata may be partially written, creating inconsistencies that only a forced repair can fix.
• Repair limits: xfs_repair with the -L option truncates the journal; any uncommitted data is lost. Use it only when you have a full backup of the data or are willing to accept that loss.

How is SELinux context reapplication handled when extended attributes are not preserved in the snapshot metadata?

• SELinux contexts are stored as extended attributes (xattrs). If a snapshot does not preserve these xattrs, the restored filesystem will have missing or default contexts.
• Recovery: Run restorecon -Rv / to walk the tree and reapply the correct contexts according to the policy. This will set the proper SELinux labels based on file names and paths.
• Audit: After restorecon, check semanage fcontext -l to ensure all file types have the expected contexts.
• Note: If the snapshot omitted the xattrs, the contexts are not recoverable from the snapshot itself; they must be regenerated from policy defaults.

Likely Explanation (Based on Common Practices)

When a snapshot is taken while the filesystem is mounted, the kernel flushes dirty pages to disk, which reduces the chance of incomplete metadata. However, if the snapshot is taken on an unmounted or partially mounted volume, the metadata area may be left in a half‑written state. In such cases, xfs_repair with the -n flag will often report inconsistencies that require a full repair. If the snapshot was created with lvcreate --snapshot while the filesystem was busy, the journal may not have been fully written, and the repair process will truncate it if you use -L.

Confirmed Steps for Your Scenario

1. Verify the snapshot contains the entire metadata region:

   xfs_info /dev/<device> | grep "metadata size"
   

   Compare the reported size to the snapshot size. If the snapshot is smaller, you risk unrecoverable corruption.

2. Unmount the target logical volume (if not already unmounted):

   umount /dev/<device>
   

3. Run a dry‑run repair to detect inconsistencies:

   xfs_repair -n /dev/<device>
   

   Review the output. If no errors are reported, proceed to remount.

4. If errors are found, perform a full repair. Use -L only if you are prepared to lose uncommitted data:

   xfs_repair -L /dev/<device>
   

5. Remount the filesystem:

   mount /dev/<device> /mount/point
   

6. Reapply SELinux contexts if extended attributes were not preserved:

   restorecon -Rv /mount/point
   

7. Verify context restoration:

   ls -Z /mount/point | head
   

Diagnostic Question

Was the snapshot taken while the XFS filesystem was mounted?

Answering this will help determine whether a full xfs_repair -L is necessary or if a simple dry‑run will suffice.