xfs: recovery should not clear di_flushiter unconditionally
authorDave Chinner <dchinner@redhat.com>
Fri, 10 Nov 2023 04:33:14 +0000 (15:33 +1100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 28 Nov 2023 17:07:15 +0000 (17:07 +0000)
commit 7930d9e103700cde15833638855b750715c12091 upstream.

Because on v3 inodes, di_flushiter doesn't exist. It overlaps with
zero padding in the inode, except when NREXT64=1 configurations are
in use and the zero padding is no longer padding but holds the 64
bit extent counter.

This manifests obviously on big endian platforms (e.g. s390) because
the log dinode is in host order and the overlap is the LSBs of the
extent count field. It is not noticed on little endian machines
because the overlap is at the MSB end of the extent count field and
we need to get more than 2^^48 extents in the inode before it
manifests. i.e. the heat death of the universe will occur before we
see the problem in little endian machines.

This is a zero-day issue for NREXT64=1 configuraitons on big endian
machines. Fix it by only clearing di_flushiter on v2 inodes during
recovery.

Fixes: 9b7d16e34bbe ("xfs: Introduce XFS_DIFLAG2_NREXT64 and associated helpers")
cc: stable@kernel.org # 5.19+
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/xfs/xfs_inode_item_recover.c

index 0e5dba2343ea13e2e0451178097f6ad247371724..e6609067ef26109017ae4889c2fd6026d1a761c9 100644 (file)
@@ -369,24 +369,26 @@ xlog_recover_inode_commit_pass2(
         * superblock flag to determine whether we need to look at di_flushiter
         * to skip replay when the on disk inode is newer than the log one
         */
-       if (!xfs_has_v3inodes(mp) &&
-           ldip->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
-               /*
-                * Deal with the wrap case, DI_MAX_FLUSH is less
-                * than smaller numbers
-                */
-               if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
-                   ldip->di_flushiter < (DI_MAX_FLUSH >> 1)) {
-                       /* do nothing */
-               } else {
-                       trace_xfs_log_recover_inode_skip(log, in_f);
-                       error = 0;
-                       goto out_release;
+       if (!xfs_has_v3inodes(mp)) {
+               if (ldip->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
+                       /*
+                        * Deal with the wrap case, DI_MAX_FLUSH is less
+                        * than smaller numbers
+                        */
+                       if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
+                           ldip->di_flushiter < (DI_MAX_FLUSH >> 1)) {
+                               /* do nothing */
+                       } else {
+                               trace_xfs_log_recover_inode_skip(log, in_f);
+                               error = 0;
+                               goto out_release;
+                       }
                }
+
+               /* Take the opportunity to reset the flush iteration count */
+               ldip->di_flushiter = 0;
        }
 
-       /* Take the opportunity to reset the flush iteration count */
-       ldip->di_flushiter = 0;
 
        if (unlikely(S_ISREG(ldip->di_mode))) {
                if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) &&