]> git.ipfire.org Git - thirdparty/linux.git/commitdiff
l2tp: use refcount_inc_not_zero in l2tp_session_get_by_ifname
authorMichael Bommarito <michael.bommarito@gmail.com>
Sat, 23 May 2026 02:34:23 +0000 (22:34 -0400)
committerJakub Kicinski <kuba@kernel.org>
Wed, 27 May 2026 00:35:12 +0000 (17:35 -0700)
A reader in l2tp_session_get_by_ifname() can return a pointer to a
session whose refcount has reached zero. The getter takes its
reference with plain refcount_inc(), but every other session getter
in the same file (l2tp_v2_session_get, l2tp_v3_session_get, and the
corresponding _get_next variants) uses refcount_inc_not_zero()
because the IDR/RCU lookup can race with refcount_dec_and_test() ->
l2tp_session_free() -> kfree_rcu(). The ifname getter is the only
outlier; the inconsistency was raised on-list after 979c017803c4
("l2tp: use list_del_rcu in l2tp_session_unhash").

A reader inside rcu_read_lock_bh() that matches session->ifname can
be preempted between the strcmp() and the refcount_inc(). If the
last reference drops on another CPU in that window, the reader's
refcount_inc() runs on a counter that has reached zero. refcount_t
catches the addition-on-zero, prints "refcount_t: addition on 0;
use-after-free", saturates the counter, and returns the saturated
pointer to the caller. Session memory is held live by the in-flight
RCU read section, but the kfree_rcu() callback queued from
l2tp_session_free() will free it once the grace period closes; a
caller that dereferences the returned session past that point hits
a slab-use-after-free. On PREEMPT_RT local_bh_disable() is a per-CPU
sleeping lock and the preemption window is real; on stock PREEMPT
kernels local_bh_disable() is a preempt_count increment that closes
the cross-CPU race in practice (see below).

Use refcount_inc_not_zero() and continue the list walk on failure,
matching the other session getters in the file. The ifname getter
is the only session getter in net/l2tp/ that still uses the bare
refcount_inc() pattern; this change restores file-internal
consistency. The success path is unchanged.

Fixes: abe7a1a7d0b6 ("l2tp: improve tunnel/session refcount helpers")
Cc: stable@vger.kernel.org
Signed-off-by: Michael Bommarito <michael.bommarito@gmail.com>
Reviewed-by: James Chapman <jchapman@katalix.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260523023423.2568972-1-michael.bommarito@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
net/l2tp/l2tp_core.c

index 1455f67e01ddb83321ce801de37ee590942cc7dd..9419c8555d2290f2df9450f30220b508a8df6613 100644 (file)
@@ -441,12 +441,13 @@ struct l2tp_session *l2tp_session_get_by_ifname(const struct net *net,
        idr_for_each_entry_ul(&pn->l2tp_tunnel_idr, tunnel, tmp, tunnel_id) {
                if (tunnel) {
                        list_for_each_entry_rcu(session, &tunnel->session_list, list) {
-                               if (!strcmp(session->ifname, ifname)) {
-                                       refcount_inc(&session->ref_count);
-                                       rcu_read_unlock_bh();
+                               if (strcmp(session->ifname, ifname))
+                                       continue;
+                               if (!refcount_inc_not_zero(&session->ref_count))
+                                       continue;
+                               rcu_read_unlock_bh();
 
-                                       return session;
-                               }
+                               return session;
                        }
                }
        }