Hi Paolo,

On 3/4/2026 5:07 PM, Paolo Abeni wrote:

On 3/2/26 6:26 AM, Li Xiasong wrote:

quoted

@@ -2343,7 +2347,7 @@ static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
 
 		pr_debug("block timeout %ld\n", timeo);
 		mptcp_cleanup_rbuf(msk, copied);
-		err = sk_wait_data(sk, &timeo, NULL);
+		err = sk_wait_data(sk, &timeo, last);

Isn't
		err = sk_wait_data(sk, &timeo,  skb_peek_tail(&sk->sk_receive_queue));

Enough?

Thanks for the review! In the general case, letting sk_wait_data block is
correct. The exception is within __mptcp_recvmsg_mskq, where a failure in
skb_copy_datagram_msg will cause a break in the consumption loop. This
leaves the actual processing position inconsistent with
skb_peek_tail(&sk->sk_receive_queue), and at that point, we must not wait.

Also, could you please add a paired tests-case? Thanks!

Further to our discussion, a minimal reproduction case for this scenario is
available here:
https://lore.kernel.org/netdev/e91bf909-bf4d-4f90-a370-688a9424478b@huawei.com/ (local)

When running the reproduced program, you can observe the server's sy
(system) CPU usage hitting 100%.

Thanks,
Li Xiasong