gfs2: Don't deref jdesc in evict
authorBob Peterson <rpeterso@redhat.com>
Fri, 28 Apr 2023 16:07:46 +0000 (12:07 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 9 Jun 2023 08:34:08 +0000 (10:34 +0200)
commit5ae4a618a1558d2b536fdd5d42e53d3e2d73870c
tree2e64d98d80695349011fed2efbe6be1b3295bffe
parent61c3962ab1f9647998bab2aa47fb29bcc214d41a
gfs2: Don't deref jdesc in evict

[ Upstream commit 504a10d9e46bc37b23d0a1ae2f28973c8516e636 ]

On corrupt gfs2 file systems the evict code can try to reference the
journal descriptor structure, jdesc, after it has been freed and set to
NULL. The sequence of events is:

init_journal()
...
fail_jindex:
   gfs2_jindex_free(sdp); <------frees journals, sets jdesc = NULL
      if (gfs2_holder_initialized(&ji_gh))
         gfs2_glock_dq_uninit(&ji_gh);
fail:
   iput(sdp->sd_jindex); <--references jdesc in evict_linked_inode
      evict()
         gfs2_evict_inode()
            evict_linked_inode()
               ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
<------references the now freed/zeroed sd_jdesc pointer.

The call to gfs2_trans_begin is done because the truncate_inode_pages
call can cause gfs2 events that require a transaction, such as removing
journaled data (jdata) blocks from the journal.

This patch fixes the problem by adding a check for sdp->sd_jdesc to
function gfs2_evict_inode. In theory, this should only happen to corrupt
gfs2 file systems, when gfs2 detects the problem, reports it, then tries
to evict all the system inodes it has read in up to that point.

Reported-by: Yang Lan <lanyang0908@gmail.com>
Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
fs/gfs2/super.c