Re: [PATCH 1/1] sctp: revalidate the association socket in timer callbacks
From: Xin Long
Date: Thu Nov 12 2026 - 18:43:28 EST
On Tue, Nov 10, 2026 at 7:03 AM liuyong <pkfxxxing@xxxxxxxxx> wrote:
>
> SCTP timer callbacks cache asoc->base.sk before acquiring its BH lock.
> An accept or peeloff can move the association to a new socket in between.
> The callback can then run the state machine under the old socket lock,
> while an abort on the new socket tears down the association's queues.
> The timer's transport and association references do not keep those
> queues alive. With delays inserted to widen the race, KASAN reported a
> slab-use-after-free in sctp_chunk_abandoned() from the T3 retransmit path.
>
> Recheck the association's socket after acquiring its BH lock. Treat a
> mismatch like a busy socket and use the existing reference-counted timer
> rescheduling path. The next callback reads the current socket again,
> preventing state-machine execution under the old socket's lock. Retain
> the user-ownership check to handle a migration already in progress.
>
> Apply the check to the T3, association, heartbeat, protocol-unreachable,
> reconfiguration and probe timers. Their BH context and the endpoint's
> RCU-deferred sock_put() protect the cached socket during the callback.
> Pair the socket pointer reads with WRITE_ONCE() in sctp_assoc_migrate(),
> taking the new socket reference before updating the pointer.
>
> The issue was discovered and fixed by Qihoo 360's Tulongfeng multi-agent
> system.
>
> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: liuyong <pkfxxxing@xxxxxxxxx>
> ---
> net/sctp/associola.c | 4 ++--
> net/sctp/sm_sideeffect.c | 41 ++++++++++++++++++++--------------------
> 2 files changed, 22 insertions(+), 23 deletions(-)
>
> diff --git a/net/sctp/associola.c b/net/sctp/associola.c
> index 4521be3bd8..c5d520c8d7 100644
> --- a/net/sctp/associola.c
> +++ b/net/sctp/associola.c
> @@ -1088,8 +1088,8 @@ void sctp_assoc_migrate(struct sctp_association *assoc, struct sock *newsk)
> sctp_endpoint_hold(assoc->ep);
>
> /* Get a reference to the new sock. */
> - assoc->base.sk = newsk;
> - sock_hold(assoc->base.sk);
> + sock_hold(newsk);
> + WRITE_ONCE(assoc->base.sk, newsk);
>
> /* Add the association to the new endpoint's list of associations. */
> sctp_endpoint_add_asoc(newsp->ep, assoc);
> diff --git a/net/sctp/sm_sideeffect.c b/net/sctp/sm_sideeffect.c
> index 35f540fb15..bd6b23369f 100644
> --- a/net/sctp/sm_sideeffect.c
> +++ b/net/sctp/sm_sideeffect.c
> @@ -233,15 +233,14 @@ void sctp_generate_t3_rtx_event(struct timer_list *t)
> struct sctp_transport *transport =
> timer_container_of(transport, t, T3_rtx_timer);
> struct sctp_association *asoc = transport->asoc;
> - struct sock *sk = asoc->base.sk;
> + struct sock *sk = READ_ONCE(asoc->base.sk);
> struct net *net = sock_net(sk);
> int error;
>
> - /* Check whether a task is in the sock. */
> -
> + /* Defer if the socket is busy or the association has migrated. */
> bh_lock_sock(sk);
> - if (sock_owned_by_user(sk)) {
> - pr_debug("%s: sock is busy\n", __func__);
> + if (sock_owned_by_user(sk) || sk != READ_ONCE(asoc->base.sk)) {
> + pr_debug("%s: sock is busy or moved\n", __func__);
>
> /* Try again later. */
> sctp_transport_hold(transport);
> @@ -271,13 +270,13 @@ void sctp_generate_t3_rtx_event(struct timer_list *t)
> static void sctp_generate_timeout_event(struct sctp_association *asoc,
> enum sctp_event_timeout timeout_type)
> {
> - struct sock *sk = asoc->base.sk;
> + struct sock *sk = READ_ONCE(asoc->base.sk);
> struct net *net = sock_net(sk);
> int error = 0;
>
> bh_lock_sock(sk);
> - if (sock_owned_by_user(sk)) {
> - pr_debug("%s: sock is busy: timer %d\n", __func__,
> + if (sock_owned_by_user(sk) || sk != READ_ONCE(asoc->base.sk)) {
> + pr_debug("%s: sock is busy or moved: timer %d\n", __func__,
> timeout_type);
>
> /* Try again later. */
> @@ -362,7 +361,7 @@ static void sctp_generate_autoclose_event(struct timer_list *t)
> sctp_generate_timeout_event(asoc, SCTP_EVENT_TIMEOUT_AUTOCLOSE);
> }
>
> -/* Generate a heart beat event. If the sock is busy, reschedule. Make
> +/* Generate a heart beat event. If the sock is busy or moved, reschedule. Make
> * sure that the transport is still valid.
> */
> void sctp_generate_heartbeat_event(struct timer_list *t)
> @@ -370,14 +369,14 @@ void sctp_generate_heartbeat_event(struct timer_list *t)
> struct sctp_transport *transport = timer_container_of(transport, t,
> hb_timer);
> struct sctp_association *asoc = transport->asoc;
> - struct sock *sk = asoc->base.sk;
> + struct sock *sk = READ_ONCE(asoc->base.sk);
> struct net *net = sock_net(sk);
> u32 elapsed, timeout;
> int error = 0;
>
> bh_lock_sock(sk);
> - if (sock_owned_by_user(sk)) {
> - pr_debug("%s: sock is busy\n", __func__);
> + if (sock_owned_by_user(sk) || sk != READ_ONCE(asoc->base.sk)) {
> + pr_debug("%s: sock is busy or moved\n", __func__);
>
> /* Try again later. */
> sctp_transport_hold(transport);
> @@ -418,12 +417,12 @@ void sctp_generate_proto_unreach_event(struct timer_list *t)
> struct sctp_transport *transport =
> timer_container_of(transport, t, proto_unreach_timer);
> struct sctp_association *asoc = transport->asoc;
> - struct sock *sk = asoc->base.sk;
> + struct sock *sk = READ_ONCE(asoc->base.sk);
> struct net *net = sock_net(sk);
>
> bh_lock_sock(sk);
> - if (sock_owned_by_user(sk)) {
> - pr_debug("%s: sock is busy\n", __func__);
> + if (sock_owned_by_user(sk) || sk != READ_ONCE(asoc->base.sk)) {
> + pr_debug("%s: sock is busy or moved\n", __func__);
>
> /* Try again later. */
> sctp_transport_hold(transport);
> @@ -454,13 +453,13 @@ void sctp_generate_reconf_event(struct timer_list *t)
> struct sctp_transport *transport =
> timer_container_of(transport, t, reconf_timer);
> struct sctp_association *asoc = transport->asoc;
> - struct sock *sk = asoc->base.sk;
> + struct sock *sk = READ_ONCE(asoc->base.sk);
> struct net *net = sock_net(sk);
> int error = 0;
>
> bh_lock_sock(sk);
> - if (sock_owned_by_user(sk)) {
> - pr_debug("%s: sock is busy\n", __func__);
> + if (sock_owned_by_user(sk) || sk != READ_ONCE(asoc->base.sk)) {
> + pr_debug("%s: sock is busy or moved\n", __func__);
>
> /* Try again later. */
> sctp_transport_hold(transport);
> @@ -492,13 +491,13 @@ void sctp_generate_probe_event(struct timer_list *t)
> struct sctp_transport *transport = timer_container_of(transport, t,
> probe_timer);
> struct sctp_association *asoc = transport->asoc;
> - struct sock *sk = asoc->base.sk;
> + struct sock *sk = READ_ONCE(asoc->base.sk);
> struct net *net = sock_net(sk);
> int error = 0;
>
> bh_lock_sock(sk);
> - if (sock_owned_by_user(sk)) {
> - pr_debug("%s: sock is busy\n", __func__);
> + if (sock_owned_by_user(sk) || sk != READ_ONCE(asoc->base.sk)) {
> + pr_debug("%s: sock is busy or moved\n", __func__);
>
> /* Try again later. */
> sctp_transport_hold(transport);
> --
> 2.43.0
>
Please check the sashiko report:
https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20261110120212.1110283-2-pkfxxxing%40gmail.com
I think you also need to use READ_ONCE() in sctp_err_lookup() and pass sk
to sctp_v4_err_handle() and sctp_v6_err_handle(), and then add
sk != READ_ONCE(asoc->base.sk) check next to sock_owned_by_user(sk) in the
functions it calls like:
sctp_icmp_frag_needed()/sctp_icmp_proto_unreachable()/sctp_icmp_redirect().
Also, use READ_ONCE() in quic_rcv() and sctp_backlog_rcv().
Thanks.