mptcp_mp_fail_no_response shouldn't be invoked on each worker run, it
should be invoked only when MP_FAIL response timeout occurs.

This patch refactors the MP_FAIL response timeout logic. Add fail_tout
in mptcp_sock to record the MP_FAIL timestamp. Check it in mptcp_worker()
before invoking mptcp_mp_fail_no_response(). Drop the code to reuse
sk_timer for MP_FAIL response.

Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/281
Fixes: d9fb797046c5 ("mptcp: Do not traverse the subflow connection list without lock")
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Geliang Tang <geliang.tang@suse.com>
---
 net/mptcp/pm.c       |  5 +----
 net/mptcp/protocol.c |  4 +++-
 net/mptcp/protocol.h |  1 +
 net/mptcp/subflow.c  | 10 ++--------
 4 files changed, 7 insertions(+), 13 deletions(-)

diff --git a/net/mptcp/pm.c b/net/mptcp/pm.c
index 59a85220edc9..45a9e02abf24 100644
--- a/net/mptcp/pm.c
+++ b/net/mptcp/pm.c
@@ -299,7 +299,6 @@ void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq)
 {
 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
-	struct sock *s = (struct sock *)msk;
 
 	pr_debug("fail_seq=%llu", fail_seq);
 
@@ -312,10 +311,8 @@ void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq)
 		subflow->send_mp_fail = 1;
 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILTX);
 		subflow->send_infinite_map = 1;
-	} else if (!sock_flag(sk, SOCK_DEAD)) {
+	} else {
 		pr_debug("MP_FAIL response received");
-
-		sk_stop_timer(s, &s->sk_timer);
 	}
 }
 
diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index d6aef4b13b8a..917df5fb9708 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -2562,7 +2562,8 @@ static void mptcp_worker(struct work_struct *work)
 	if (test_and_clear_bit(MPTCP_WORK_RTX, &msk->flags))
 		__mptcp_retrans(sk);
 
-	mptcp_mp_fail_no_response(msk);
+	if (time_after(jiffies, msk->fail_tout))
+		mptcp_mp_fail_no_response(msk);
 
 unlock:
 	release_sock(sk);
@@ -2589,6 +2590,7 @@ static int __mptcp_init_sock(struct sock *sk)
 	WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
 	WRITE_ONCE(msk->allow_infinite_fallback, true);
 	msk->recovery = false;
+	msk->fail_tout = 0;
 
 	mptcp_pm_data_init(msk);
 
diff --git a/net/mptcp/protocol.h b/net/mptcp/protocol.h
index d406b5afbee4..f7b01275af94 100644
--- a/net/mptcp/protocol.h
+++ b/net/mptcp/protocol.h
@@ -306,6 +306,7 @@ struct mptcp_sock {
 
 	u32 setsockopt_seq;
 	char		ca_name[TCP_CA_NAME_MAX];
+	unsigned long	fail_tout;
 };
 
 #define mptcp_data_lock(sk) spin_lock_bh(&(sk)->sk_lock.slock)
diff --git a/net/mptcp/subflow.c b/net/mptcp/subflow.c
index 8841e8cd9ad8..866d54a0e83c 100644
--- a/net/mptcp/subflow.c
+++ b/net/mptcp/subflow.c
@@ -974,7 +974,6 @@ static enum mapping_status get_mapping_status(struct sock *ssk,
 {
 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 	bool csum_reqd = READ_ONCE(msk->csum_enabled);
-	struct sock *sk = (struct sock *)msk;
 	struct mptcp_ext *mpext;
 	struct sk_buff *skb;
 	u16 data_len;
@@ -1016,9 +1015,6 @@ static enum mapping_status get_mapping_status(struct sock *ssk,
 		pr_debug("infinite mapping received");
 		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
 		subflow->map_data_len = 0;
-		if (!sock_flag(ssk, SOCK_DEAD))
-			sk_stop_timer(sk, &sk->sk_timer);
-
 		return MAPPING_INVALID;
 	}
 
@@ -1238,11 +1234,9 @@ static bool subflow_check_data_avail(struct sock *ssk)
 				tcp_send_active_reset(ssk, GFP_ATOMIC);
 				while ((skb = skb_peek(&ssk->sk_receive_queue)))
 					sk_eat_skb(ssk, skb);
-			} else if (!sock_flag(ssk, SOCK_DEAD)) {
+			} else {
 				WRITE_ONCE(subflow->mp_fail_response_expect, true);
-				sk_reset_timer((struct sock *)msk,
-					       &((struct sock *)msk)->sk_timer,
-					       jiffies + TCP_RTO_MAX);
+				msk->fail_tout = jiffies + TCP_RTO_MAX;
 			}
 			WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
 			return true;
-- 
2.35.3

On Thu, 2022-06-09 at 08:00 +0800, Geliang Tang wrote:
> mptcp_mp_fail_no_response shouldn't be invoked on each worker run, it
> should be invoked only when MP_FAIL response timeout occurs.
> 
> This patch refactors the MP_FAIL response timeout logic. Add fail_tout
> in mptcp_sock to record the MP_FAIL timestamp. Check it in mptcp_worker()
> before invoking mptcp_mp_fail_no_response(). Drop the code to reuse
> sk_timer for MP_FAIL response.

I'm sorry for not noticing it on the previous version - likely for the
lack of clarity on my side: we still need to start the timer for
mp_fail response: otherwise we are not assured that the mptcp_worker
will be triggered.

Additionally we need some more care to manipulate the timer reset. I
think it's easier if I send a squash-to patch - assuming I'll have some
time for that early next week.

/P

On Thu, 9 Jun 2022, Paolo Abeni wrote:

> On Thu, 2022-06-09 at 08:00 +0800, Geliang Tang wrote:
>> mptcp_mp_fail_no_response shouldn't be invoked on each worker run, it
>> should be invoked only when MP_FAIL response timeout occurs.
>>
>> This patch refactors the MP_FAIL response timeout logic. Add fail_tout
>> in mptcp_sock to record the MP_FAIL timestamp. Check it in mptcp_worker()
>> before invoking mptcp_mp_fail_no_response(). Drop the code to reuse
>> sk_timer for MP_FAIL response.
>
> I'm sorry for not noticing it on the previous version - likely for the
> lack of clarity on my side: we still need to start the timer for
> mp_fail response: otherwise we are not assured that the mptcp_worker
> will be triggered.
>

(capturing this on the mailing list in addition to the recent meeting)

Yes, I agree. I saw that the timer stop calls were removed but missed the 
timer reset removal. It is imporant for the timer to still be started.

> Additionally we need some more care to manipulate the timer reset. I
> think it's easier if I send a squash-to patch - assuming I'll have some
> time for that early next week.
>

Thanks Paolo!

--
Mat Martineau
Intel