mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Max Kellermann <max.kellermann@ionos.com>
To: idryomov@gmail.com, amarkuze@redhat.com, xiubo.li@clyso.com,
	ceph-devel@vger.kernel.org, linux-kernel@vger.kernel.org
Cc: Max Kellermann <max.kellermann@ionos.com>, stable@vger.kernel.org
Subject: [PATCH v4 1/3] ceph: fix use-after-free in check_new_map() after early session put
Date: Fri,  4 Sep 2026 17:03:02 +0200	[thread overview]
Message-ID: <20260904150304.49104-2-max.kellermann@ionos.com> (raw)
In-Reply-To: <20260904150304.49104-1-max.kellermann@ionos.com>

check_new_map() drops mdsc->mutex while it locks a session, prepares a
reconnect, or kicks flushing caps.  A concurrent teardown can call
__unregister_session() in that window and drop the sessions[]
reference. When check_new_map() reacquires mdsc->mutex, its temporary
reference may therefore be the only reference keeping the local
variable `s` alive.

Commit ee611a750955 ("ceph: fix UAF in check_new_map() on session freed
during unlock") addressed this by taking a temporary reference around
each unlock window, but it drops that reference as soon as mdsc->mutex
is reacquired, while `s` is still in use:

  ceph_get_mds_session(s);
  mutex_unlock(&mdsc->mutex);
  mutex_lock(&s->s_mutex);
  mutex_lock(&mdsc->mutex);
  ceph_put_mds_session(s);    /* may drop the last reference */
  ceph_con_close(&s->s_con);  /* use-after-free */
  mutex_unlock(&s->s_mutex);  /* use-after-free */
  s->s_state = CEPH_MDS_SESSION_RESTARTING;

If the session was unregistered during the window, the sessions[]
reference is already gone, so this ceph_put_mds_session() frees it.

Additionally, the session could be freed while its s_mutex is locked,
which trips the WARN_ON(mutex_is_locked(&s->s_mutex)) in
ceph_put_mds_session() and then unlocks freed memory.

Fix this by holding a single reference for the whole loop iteration:
look the session up with __ceph_lookup_mds_session(), which returns it
with a reference held, and release that reference on every exit from
the loop body.  This subsumes the per-window get/put pairs, so remove
them.

Fixes: ee611a750955 ("ceph: fix UAF in check_new_map() on session freed during unlock")
Cc: stable@vger.kernel.org
Signed-off-by: Max Kellermann <max.kellermann@ionos.com>
---
Note the stable maintainers: this is a fixup for ee611a750955, but the
bug has existed before; see
https://lore.kernel.org/ceph-devel/20260828174504.1247038-2-max.kellermann@ionos.com/
for a patch that applies to pre-7.2 kernel versions.
---
 fs/ceph/mds_client.c | 13 ++++---------
 1 file changed, 4 insertions(+), 9 deletions(-)

diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
index a091f77cedaf..d36a114747ae 100644
--- a/fs/ceph/mds_client.c
+++ b/fs/ceph/mds_client.c
@@ -5860,9 +5860,9 @@ static void check_new_map(struct ceph_mds_client *mdsc,
 	}
 
 	for (i = 0; i < oldmap->possible_max_rank && i < mdsc->max_sessions; i++) {
-		if (!mdsc->sessions[i])
+		s = __ceph_lookup_mds_session(mdsc, i);
+		if (!s)
 			continue;
-		s = mdsc->sessions[i];
 		oldstate = ceph_mdsmap_get_state(oldmap, i);
 		newstate = ceph_mdsmap_get_state(newmap, i);
 
@@ -5875,7 +5875,6 @@ static void check_new_map(struct ceph_mds_client *mdsc,
 
 		if (i >= newmap->possible_max_rank) {
 			/* force close session for stopped mds */
-			ceph_get_mds_session(s);
 			__unregister_session(mdsc, s);
 			__wake_requests(mdsc, &s->s_waiting);
 			mutex_unlock(&mdsc->mutex);
@@ -5896,15 +5895,14 @@ static void check_new_map(struct ceph_mds_client *mdsc,
 			   ceph_mdsmap_get_addr(newmap, i),
 			   sizeof(struct ceph_entity_addr))) {
 			/* just close it */
-			ceph_get_mds_session(s);
 			mutex_unlock(&mdsc->mutex);
 			mutex_lock(&s->s_mutex);
 			mutex_lock(&mdsc->mutex);
-			ceph_put_mds_session(s);
 			ceph_con_close(&s->s_con);
 			mutex_unlock(&s->s_mutex);
 			s->s_state = CEPH_MDS_SESSION_RESTARTING;
 		} else if (oldstate == newstate) {
+			ceph_put_mds_session(s);
 			continue;  /* nothing new with this mds */
 		}
 
@@ -5915,7 +5913,6 @@ static void check_new_map(struct ceph_mds_client *mdsc,
 		    newstate >= CEPH_MDS_STATE_RECONNECT) {
 			int rc;
 
-			ceph_get_mds_session(s);
 			mutex_unlock(&mdsc->mutex);
 			clear_bit(i, targets);
 			rc = send_mds_reconnect(mdsc, s);
@@ -5924,7 +5921,6 @@ static void check_new_map(struct ceph_mds_client *mdsc,
 					       "mds%d reconnect failed: %d\n",
 					       i, rc);
 			mutex_lock(&mdsc->mutex);
-			ceph_put_mds_session(s);
 		}
 
 		/*
@@ -5937,15 +5933,14 @@ static void check_new_map(struct ceph_mds_client *mdsc,
 				pr_info_client(cl, "mds%d recovery completed\n",
 					       s->s_mds);
 			kick_requests(mdsc, i);
-			ceph_get_mds_session(s);
 			mutex_unlock(&mdsc->mutex);
 			mutex_lock(&s->s_mutex);
 			mutex_lock(&mdsc->mutex);
-			ceph_put_mds_session(s);
 			ceph_kick_flushing_caps(mdsc, s);
 			mutex_unlock(&s->s_mutex);
 			wake_up_session_caps(s, RECONNECT);
 		}
+		ceph_put_mds_session(s);
 	}
 
 	/*
-- 
2.47.3


  reply	other threads:[~2026-09-04 15:03 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-09-04 15:03 [PATCH v4 0/3] ceph: don't unregister an MDS session before removing its caps Max Kellermann
2026-09-04 15:03 ` Max Kellermann [this message]
2026-09-05  1:52   ` [PATCH v4 1/3] ceph: fix use-after-free in check_new_map() after early session put Xiubo Li
2026-09-04 15:03 ` [PATCH v4 2/3] ceph: stop checking a stale MDS map after dropping mutex Max Kellermann
2026-09-05  1:52   ` Xiubo Li
2026-09-04 15:03 ` [PATCH v4 3/3] ceph: don't unregister an MDS session before removing its caps Max Kellermann
2026-09-05  1:56   ` Xiubo Li

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20260904150304.49104-2-max.kellermann@ionos.com \
    --to=max.kellermann@ionos.com \
    --cc=amarkuze@redhat.com \
    --cc=ceph-devel@vger.kernel.org \
    --cc=idryomov@gmail.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=stable@vger.kernel.org \
    --cc=xiubo.li@clyso.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®