[PATCH] nfc: llcp: walk the accept queue under the socket lock

From: Henry Martin

Date: Wed Nov 18 2026 - 03:38:30 EST


llcp_accept_poll() walks parent->accept_queue and reads each child's
sk_state without holding the socket lock. nfc_llcp_accept_dequeue()
unlinks closed children and drops their creation reference under that
lock, and the last sock_put() frees the child.

A poll() racing with accept() therefore reads a freed child:

BUG: KASAN: slab-use-after-free in llcp_sock_poll+0x440/0x4e0
Read of size 1 at addr ffff888112387812 by task poc.bin/528
Allocated by task 198:
nfc_llcp_sock_alloc
nfc_llcp_rx_skb
Freed by task 529:
__sk_destruct
nfc_llcp_accept_dequeue

Take the listener lock around the accept queue walk.

This issue was discovered by Tencent CodeBuddy Security.

Fixes: d646960f7986 ("NFC: Initial LLCP support")
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Henry Martin <bsdhenrymartin@xxxxxxxxx>
---
net/nfc/llcp_sock.c | 12 ++++++++++--
1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/net/nfc/llcp_sock.c b/net/nfc/llcp_sock.c
index 1e5ee4bcde68..3503f1a7a7cc 100644
--- a/net/nfc/llcp_sock.c
+++ b/net/nfc/llcp_sock.c
@@ -584,8 +584,16 @@ static __poll_t llcp_sock_poll(struct file *file, struct socket *sock,

sock_poll_wait(file, sock, wait);

- if (sk->sk_state == LLCP_LISTEN)
- return llcp_accept_poll(sk);
+ if (sk->sk_state == LLCP_LISTEN) {
+ /* nfc_llcp_accept_dequeue() unlinks and frees closed children
+ * under the listener lock, so the queue can only be walked
+ * with that lock held.
+ */
+ lock_sock(sk);
+ mask = llcp_accept_poll(sk);
+ release_sock(sk);
+ return mask;
+ }

if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue))
mask |= EPOLLERR |
--
2.43.7