CVE-2026-53357

*** Pendiente de traducción *** In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> Bluetooth: fix UAF in l2cap_sock_cleanup_listen() vs l2cap_conn_del()<br /> <br /> bt_accept_dequeue() unlinks a not-yet-accepted child from the parent<br /> accept queue and release_sock()s it before returning, so the returned<br /> sk has no caller reference and is unlocked.<br /> <br /> l2cap_sock_cleanup_listen() walks these children on listening-socket<br /> close. A concurrent HCI disconnect drives hci_rx_work -&gt;<br /> l2cap_conn_del() which runs l2cap_chan_del() + l2cap_sock_kill() and<br /> frees the child sk and its l2cap_chan; cleanup_listen() then uses both:<br /> <br /> BUG: KASAN: slab-use-after-free in l2cap_sock_kill<br /> l2cap_sock_kill / l2cap_sock_cleanup_listen / __x64_sys_close<br /> Freed by: l2cap_conn_del -&gt; l2cap_sock_close_cb -&gt; l2cap_sock_kill<br /> <br /> This is distinct from the two fixes already in this area: commit<br /> e83f5e24da741 ("Bluetooth: serialize accept_q access") serialises the<br /> accept_q list/poll and takes temporary refs inside bt_accept_dequeue(),<br /> and CVE-2025-39860 serialises the userspace close()/accept() race by<br /> calling cleanup_listen() under lock_sock() in l2cap_sock_release().<br /> Neither covers l2cap_conn_del() running from hci_rx_work, so this UAF<br /> still reproduces on current bluetooth/master.<br /> <br /> Take the reference at the source: bt_accept_dequeue() does sock_hold()<br /> while sk is still locked, before release_sock(); callers sock_put().<br /> cleanup_listen() pins the chan with l2cap_chan_hold_unless_zero() under<br /> a brief child sk lock (serialising vs l2cap_sock_teardown_cb()), drops<br /> it before l2cap_chan_lock(), and skips a duplicate l2cap_sock_kill() on<br /> SOCK_DEAD. conn-&gt;lock is not taken here: cleanup_listen() runs under<br /> the parent sk lock and that would invert<br /> conn-&gt;lock -&gt; chan-&gt;lock -&gt; sk_lock (lockdep).<br /> <br /> KASAN/SMP: an unprivileged listen/close vs HCI-disconnect race produced<br /> 12 use-after-free reports per run before this change; 0, and no lockdep<br /> report, over 1600+ raced iterations after it on bluetooth/master.