xfs: sanity check attr fork size
authorChristoph Hellwig <hch@lst.de>
Mon, 19 Jan 2009 01:04:16 +0000 (02:04 +0100)
committerLachlan McIlroy <lachlan@sgi.com>
Mon, 19 Jan 2009 03:45:11 +0000 (14:45 +1100)
Recently we have quite a few kerneloops reports about dereferencing a NULL
if_data in the attribute fork.  From looking over the code this can only
happen if we pass a 0 size argument to xfs_iformat_local.  This implies some
sort of corruption and in fact the only mailinglist report about this from
earlier this year was after a powerfail presumably on a system with write
cache and without barriers.

Add a quick sanity check for the attr fork size in xfs_iformat to catch
these early and without an oops.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <david@fromorbit.com>
fs/xfs/xfs_inode.c

index 5a5e035e5d388a0fa844c9c66a47cf92dd2db1e7..323ecd76a12c0fed6773d200cab4ed3f5e80d1ac 100644 (file)
@@ -424,6 +424,19 @@ xfs_iformat(
        case XFS_DINODE_FMT_LOCAL:
                atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
                size = be16_to_cpu(atp->hdr.totsize);
+
+               if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
+                       xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
+                               "corrupt inode %Lu "
+                               "(bad attr fork size %Ld).",
+                               (unsigned long long) ip->i_ino,
+                               (long long) size);
+                       XFS_CORRUPTION_ERROR("xfs_iformat(8)",
+                                            XFS_ERRLEVEL_LOW,
+                                            ip->i_mount, dip);
+                       return XFS_ERROR(EFSCORRUPTED);
+               }
+
                error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
                break;
        case XFS_DINODE_FMT_EXTENTS: