Thanks for your feedback. It sounds good to put an "if (m->m_info)" around
the for loop to fix this bug, as well as avoid other potential bugs
caused by calling
ceph_mdsmap_destroy(). I think we can prepare a V2 patch according to
your advice.
On 2021/8/5 20:29, Jeff Layton wrote:
On Thu, 2021-08-05 at 05:20 -0700, Tuo Li wrote:
kcalloc() is called to allocate memory for m->m_info, and if it fails,
ceph_mdsmap_destroy() behind the label out_err will be called:
ceph_mdsmap_destroy(m);
In ceph_mdsmap_destroy(), m->m_info is dereferenced through:
kfree(m->m_info[i].export_targets);
To fix this possible null-pointer dereference, if memory allocation
for m->m_info fails, free m and return -ENOMEM.
Reported-by: TOTE Robot <oslab@xxxxxxxxxxxxxxx>
Signed-off-by: Tuo Li <islituo@xxxxxxxxx>
---
fs/ceph/mdsmap.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/fs/ceph/mdsmap.c b/fs/ceph/mdsmap.c
index abd9af7727ad..7d73e4b64b12 100644
--- a/fs/ceph/mdsmap.c
+++ b/fs/ceph/mdsmap.c
@@ -166,8 +166,10 @@ struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end, bool msgr2)
m->possible_max_rank = max(m->m_num_active_mds, m->m_max_mds);
m->m_info = kcalloc(m->possible_max_rank, sizeof(*m->m_info), GFP_NOFS);
- if (!m->m_info)
- goto nomem;
+ if (!m->m_info) {
+ kfree(m);
+ return ERR_PTR(-ENOMEM);
+ }
/* pick out active nodes from mds_info (state > 0) */
for (i = 0; i < n; i++) {
Good catch. This function is already pretty complex. How about we
instead fix this in ceph_mdsmap_destroy and make it safe to call that
with the mdsmap in this state?
Basically, just put an "if (m->m_info)" around the for loop in that
function. Sound ok?
