xfs: Fix possible memory corruption in xfs_readlink
commitc38aeb8cd119fddc37d7cd648d6c43e782711247
authorCarlos Maiolino <cmaiolino@redhat.com>
Thu, 1 Dec 2011 23:27:43 +0000 (1 17:27 -0600)
committerGreg Kroah-Hartman <gregkh@suse.de>
Fri, 9 Dec 2011 16:54:30 +0000 (9 08:54 -0800)
tree4b5f2283dd7bc6bdb00c78e3b7eea7f0f61c3aca
parent7f9fae139e1b34b0ecb0ff8ab051c4497d68db2c
xfs: Fix possible memory corruption in xfs_readlink

commit b52a360b2aa1c59ba9970fb0f52bbb093fcc7a24 upstream.

Fixes a possible memory corruption when the link is larger than
MAXPATHLEN and XFS_DEBUG is not enabled. This also remove the
S_ISLNK assert, since the inode mode is checked previously in
xfs_readlink_by_handle() and via VFS.

Updated to address concerns raised by Ben Hutchings about the loose
attention paid to 32- vs 64-bit values, and the lack of handling a
potentially negative pathlen value:
 - Changed type of "pathlen" to be xfs_fsize_t, to match that of
   ip->i_d.di_size
 - Added checking for a negative pathlen to the too-long pathlen
   test, and generalized the message that gets reported in that case
   to reflect the change
As a result, if a negative pathlen were encountered, this function
would return EFSCORRUPTED (and would fail an assertion for a debug
build)--just as would a too-long pathlen.

Signed-off-by: Alex Elder <aelder@sgi.com>
Signed-off-by: Carlos Maiolino <cmaiolino@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Ben Myers <bpm@sgi.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
fs/xfs/xfs_vnodeops.c