Backlog entries may retain references to a subflow socket through
skb->sk. These references are later cleaned up during subflow teardown.

Currently, backlog cleanup is not synchronized with RX-side backlog
enqueue, which is serialized under mptcp_data_lock(). This allows a
race where teardown misses an skb referencing the subflow socket, and
the stale skb->sk pointer survives after the socket is released.

Later backlog purge may then dereference the stale pointer, leading to
memory accounting warnings in inet_sock_destruct() followed by a
use-after-free in mptcp_backlog_purge().

Prevent this by synchronizing backlog cleanup against RX enqueue while
the subflow is being closed, ensuring no stale skb->sk references
survive teardown.

Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/621
Signed-off-by: Kalpan Jani <kalpan.jani@mpiricsoftware.com>
Suggested-by: Paolo Abeni <pabeni@redhat.com>

---
v1:
- Detect and clear stale skb->sk references from the MPTCP backlog during subflow teardown to prevent use-after-free in backlog purge.
v2:
- Rework backlog cleanup based on Paolo Abeni's suggestion to serialize cleanup with RX-side backlog enqueue under mptcp_data_lock().
- Move cleanup into __mptcp_close_ssk() while holding the subflow socket lock, ensuring no new backlog skb can retain a reference to the closing subflow after cleanup completes.
- Keep the fallback cleanup path for detached subflows after releasing the subflow socket, preserving correctness for that teardown path.

 net/mptcp/protocol.c | 48 +++++++++++++++++++++++++++++++-------------
 1 file changed, 34 insertions(+), 14 deletions(-)

diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index 718e910ff..3dcffe4b5 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -2527,6 +2527,22 @@ static void __mptcp_subflow_disconnect(struct sock *ssk,
 	}
 }
 
+static void mptcp_cleanup_ssk_backlog(struct sock *sk, struct sock *ssk)
+{
+	struct mptcp_sock *msk = mptcp_sk(sk);
+	struct sk_buff *skb;
+
+	mptcp_data_lock(sk);
+	list_for_each_entry(skb, &msk->backlog_list, list) {
+		if (skb->sk != ssk)
+			continue;
+
+		atomic_sub(skb->truesize, &skb->sk->sk_rmem_alloc);
+		skb->sk = NULL;
+	}
+	mptcp_data_unlock(sk);
+}
+
 /* subflow sockets can be either outgoing (connect) or incoming
  * (accept).
  *
@@ -2540,6 +2556,7 @@ static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
 			      unsigned int flags)
 {
 	struct mptcp_sock *msk = mptcp_sk(sk);
+	struct sk_buff *skb;
 	bool dispose_it, need_push = false;
 	int fwd_remaining;
 
@@ -2549,6 +2566,22 @@ static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
 	 */
 	lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
 	subflow->closing = 1;
+
+	/* Remove any reference from the backlog to this ssk; backlog skbs
+	 * consume space in the msk receive queue, no need to touch
+	 * sk->sk_rmem_alloc. Serialize with mptcp_data_ready() under
+	 * mptcp_data_lock() while the ssk lock is still held, so the
+	 * cleanup is exhaustive: no new skb can be enqueued after this point.
+	 */
+	mptcp_data_lock(sk);
+	list_for_each_entry(skb, &msk->backlog_list, list) {
+		if (skb->sk != ssk)
+			continue;
+
+		atomic_sub(skb->truesize, &skb->sk->sk_rmem_alloc);
+		skb->sk = NULL;
+	}
+	mptcp_data_unlock(sk);
 
 	/* Borrow the fwd allocated page left-over; fwd memory for the subflow
 	 * could be negative at this point, but will be reach zero soon - when
@@ -2587,6 +2620,7 @@ static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
 		__mptcp_subflow_disconnect(ssk, subflow, msk->fastclosing);
 		release_sock(ssk);
 
+		mptcp_cleanup_ssk_backlog(sk, ssk);
 		goto out;
 	}
 
@@ -2641,9 +2675,6 @@ static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
 void mptcp_close_ssk(struct sock *sk, struct sock *ssk,
 		     struct mptcp_subflow_context *subflow)
 {
-	struct mptcp_sock *msk = mptcp_sk(sk);
-	struct sk_buff *skb;
-
 	/* The first subflow can already be closed or disconnected */
 	if (subflow->close_event_done || READ_ONCE(subflow->local_id) < 0)
 		return;
@@ -2653,17 +2684,6 @@ void mptcp_close_ssk(struct sock *sk, struct sock *ssk,
 	if (sk->sk_state == TCP_ESTABLISHED)
 		mptcp_event(MPTCP_EVENT_SUB_CLOSED, mptcp_sk(sk), ssk, GFP_KERNEL);
 
-	/* Remove any reference from the backlog to this ssk; backlog skbs consume
-	 * space in the msk receive queue, no need to touch sk->sk_rmem_alloc
-	 */
-	list_for_each_entry(skb, &msk->backlog_list, list) {
-		if (skb->sk != ssk)
-			continue;
-
-		atomic_sub(skb->truesize, &skb->sk->sk_rmem_alloc);
-		skb->sk = NULL;
-	}
-
 	/* subflow aborted before reaching the fully_established status
 	 * attempt the creation of the next subflow
 	 */
-- 
2.43.0

On 5/18/26 7:00 AM, Kalpan Jani wrote:
> @@ -2549,6 +2566,22 @@ static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
>  	 */
>  	lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
>  	subflow->closing = 1;
> +
> +	/* Remove any reference from the backlog to this ssk; backlog skbs
> +	 * consume space in the msk receive queue, no need to touch
> +	 * sk->sk_rmem_alloc. Serialize with mptcp_data_ready() under
> +	 * mptcp_data_lock() while the ssk lock is still held, so the
> +	 * cleanup is exhaustive: no new skb can be enqueued after this point.
> +	 */
> +	mptcp_data_lock(sk);
> +	list_for_each_entry(skb, &msk->backlog_list, list) {
> +		if (skb->sk != ssk)
> +			continue;
> +
> +		atomic_sub(skb->truesize, &skb->sk->sk_rmem_alloc);
> +		skb->sk = NULL;
> +	}
> +	mptcp_data_unlock(sk);

Here you should use the just defined helper, and only when `flags &
MPTCP_CF_PUSH`, otherwise the backlog cleanup is already performed by
the caller.

>  
>  	/* Borrow the fwd allocated page left-over; fwd memory for the subflow
>  	 * could be negative at this point, but will be reach zero soon - when
> @@ -2587,6 +2620,7 @@ static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
>  		__mptcp_subflow_disconnect(ssk, subflow, msk->fastclosing);
>  		release_sock(ssk);
>  
> +		mptcp_cleanup_ssk_backlog(sk, ssk);

Why are you cleaning the backlog again?

Why you did not use the code I suggested?

For the next iteration, please go trough the mptcp ML only, to reduce
traffic on more busy places.

/P
/P