xfs: fix endian conversion issue in discard code
Dave Chinner [Thu, 26 Jan 2012 19:47:42 +0000 (13:47 -0600)]
commit b1c770c273a4787069306fc82aab245e9ac72e9d upstream

When finding the longest extent in an AG, we read the value directly
out of the AGF buffer without endian conversion. This will give an
incorrect length, resulting in FITRIM operations potentially not
trimming everything that it should.

Note, for 3.0-stable this has been modified to apply to
fs/xfs/linux-2.6/xfs_discard.c instead of fs/xfs/xfs_discard.c.  -bpm

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Varun Wadekar <vwadekar@nvidia.com>
Change-Id: Iad84b2bb82ee0d410cdb64eeea3a8431c5a0a7ad
Reviewed-on: http://git-master/r/79660
Reviewed-by: Automatic_Commit_Validation_User

fs/xfs/xfs_discard.c

index 244e797..572494f 100644 (file)
@@ -68,7 +68,7 @@ xfs_trim_extents(
         * Look up the longest btree in the AGF and start with it.
         */
        error = xfs_alloc_lookup_le(cur, 0,
-                                   XFS_BUF_TO_AGF(agbp)->agf_longest, &i);
+                           be32_to_cpu(XFS_BUF_TO_AGF(agbp)->agf_longest), &i);
        if (error)
                goto out_del_cursor;
 
@@ -84,7 +84,7 @@ xfs_trim_extents(
                if (error)
                        goto out_del_cursor;
                XFS_WANT_CORRUPTED_GOTO(i == 1, out_del_cursor);
-               ASSERT(flen <= XFS_BUF_TO_AGF(agbp)->agf_longest);
+               ASSERT(flen <= be32_to_cpu(XFS_BUF_TO_AGF(agbp)->agf_longest));
 
                /*
                 * Too small?  Give up.