xfs: be more forgiving of a v4 secondary sb w/ junk in v5 fields
authorEric Sandeen <sandeen@sandeen.net>
Mon, 9 Sep 2013 20:33:29 +0000 (15:33 -0500)
committerBen Myers <bpm@sgi.com>
Wed, 30 Oct 2013 21:38:29 +0000 (16:38 -0500)
Today, if xfs_sb_read_verify encounters a v4 superblock
with junk past v4 fields which includes data in sb_crc,
it will be treated as a failing checksum and a significant
corruption.

There are known prior bugs which leave junk at the end
of the V4 superblock; we don't need to actually fail the
verification in this case if other checks pan out ok.

So if this is a secondary superblock, and the primary
superblock doesn't indicate that this is a V5 filesystem,
don't treat this as an actual checksum failure.

We should probably check the garbage condition as
we do in xfs_repair, and possibly warn about it
or self-heal, but that's a different scope of work.

Stable folks: This can go back to v3.10, which is what
introduced the sb CRC checking that is tripped up by old,
stale, incorrect V4 superblocks w/ unzeroed bits.

Cc: stable@vger.kernel.org
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Acked-by: Dave Chinner <david@fromorbit.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
fs/xfs/xfs_sb.c

index 13c7834b0e6552b88980f455954c1b0b262dc137..b7c9aea77f8fd2e7ca88537df77536eca8148418 100644 (file)
@@ -588,6 +588,11 @@ xfs_sb_verify(
  * single bit error could clear the feature bit and unused parts of the
  * superblock are supposed to be zero. Hence a non-null crc field indicates that
  * we've potentially lost a feature bit and we should check it anyway.
+ *
+ * However, past bugs (i.e. in growfs) left non-zeroed regions beyond the
+ * last field in V4 secondary superblocks.  So for secondary superblocks,
+ * we are more forgiving, and ignore CRC failures if the primary doesn't
+ * indicate that the fs version is V5.
  */
 static void
 xfs_sb_read_verify(
@@ -608,8 +613,12 @@ xfs_sb_read_verify(
 
                if (!xfs_verify_cksum(bp->b_addr, be16_to_cpu(dsb->sb_sectsize),
                                      offsetof(struct xfs_sb, sb_crc))) {
-                       error = EFSCORRUPTED;
-                       goto out_error;
+                       /* Only fail bad secondaries on a known V5 filesystem */
+                       if (bp->b_bn != XFS_SB_DADDR &&
+                           xfs_sb_version_hascrc(&mp->m_sb)) {
+                               error = EFSCORRUPTED;
+                               goto out_error;
+                       }
                }
        }
        error = xfs_sb_verify(bp, true);