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 adds a new msk flag MPTCP_WORK_TIMEOUT, set it in
mptcp_timeout_timer(). And test it in mptcp_worker() before invoking
mptcp_mp_fail_no_response().

Check the SOCK_DEAD flag on the subflow in mptcp_mp_fail_no_response()
to make sure it's a MP_FAIL timeout, not a MPTCP socket close timeout.

Now mp_fail_response_expect_subflow() helper is only used by
mptcp_mp_fail_no_response(), move the definition of the helper right
before mptcp_mp_fail_no_response().

Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/281
Fixes: d9fb797046c5 ("mptcp: Do not traverse the subflow connection list without lock")
Signed-off-by: Geliang Tang <geliang.tang@suse.com>
---
 net/mptcp/protocol.c | 41 ++++++++++++++++++++++++-----------------
 net/mptcp/protocol.h |  1 +
 2 files changed, 25 insertions(+), 17 deletions(-)

diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index d6aef4b13b8a..123fa2b3f200 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -2167,25 +2167,13 @@ static void mptcp_retransmit_timer(struct timer_list *t)
 	sock_put(sk);
 }
 
-static struct mptcp_subflow_context *
-mp_fail_response_expect_subflow(struct mptcp_sock *msk)
-{
-	struct mptcp_subflow_context *subflow, *ret = NULL;
-
-	mptcp_for_each_subflow(msk, subflow) {
-		if (READ_ONCE(subflow->mp_fail_response_expect)) {
-			ret = subflow;
-			break;
-		}
-	}
-
-	return ret;
-}
-
 static void mptcp_timeout_timer(struct timer_list *t)
 {
 	struct sock *sk = from_timer(sk, t, sk_timer);
 
+	bh_lock_sock(sk);
+	__set_bit(MPTCP_WORK_TIMEOUT, &mptcp_sk(sk)->flags);
+	bh_unlock_sock(sk);
 	mptcp_schedule_work(sk);
 	sock_put(sk);
 }
@@ -2505,6 +2493,21 @@ static void __mptcp_retrans(struct sock *sk)
 		mptcp_reset_timer(sk);
 }
 
+static struct mptcp_subflow_context *
+mp_fail_response_expect_subflow(struct mptcp_sock *msk)
+{
+	struct mptcp_subflow_context *subflow, *ret = NULL;
+
+	mptcp_for_each_subflow(msk, subflow) {
+		if (READ_ONCE(subflow->mp_fail_response_expect)) {
+			ret = subflow;
+			break;
+		}
+	}
+
+	return ret;
+}
+
 static void mptcp_mp_fail_no_response(struct mptcp_sock *msk)
 {
 	struct mptcp_subflow_context *subflow;
@@ -2513,9 +2516,12 @@ static void mptcp_mp_fail_no_response(struct mptcp_sock *msk)
 
 	subflow = mp_fail_response_expect_subflow(msk);
 	if (subflow) {
+		ssk = mptcp_subflow_tcp_sock(subflow);
+		if (sock_flag(ssk, SOCK_DEAD))
+			return;
+
 		pr_debug("MP_FAIL doesn't respond, reset the subflow");
 
-		ssk = mptcp_subflow_tcp_sock(subflow);
 		slow = lock_sock_fast(ssk);
 		mptcp_subflow_reset(ssk);
 		unlock_sock_fast(ssk, slow);
@@ -2562,7 +2568,8 @@ 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 (test_and_clear_bit(MPTCP_WORK_TIMEOUT, &msk->flags))
+		mptcp_mp_fail_no_response(msk);
 
 unlock:
 	release_sock(sk);
diff --git a/net/mptcp/protocol.h b/net/mptcp/protocol.h
index d406b5afbee4..f78caaaaa360 100644
--- a/net/mptcp/protocol.h
+++ b/net/mptcp/protocol.h
@@ -116,6 +116,7 @@
 #define MPTCP_WORK_EOF		3
 #define MPTCP_FALLBACK_DONE	4
 #define MPTCP_WORK_CLOSE_SUBFLOW 5
+#define MPTCP_WORK_TIMEOUT	6
 
 /* MPTCP socket release cb flags */
 #define MPTCP_PUSH_PENDING	1
-- 
2.35.3

On Wed, 2022-06-08 at 17:33 +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 adds a new msk flag MPTCP_WORK_TIMEOUT, set it in
> mptcp_timeout_timer(). And test it in mptcp_worker() before invoking
> mptcp_mp_fail_no_response().
> 
> Check the SOCK_DEAD flag on the subflow in mptcp_mp_fail_no_response()
> to make sure it's a MP_FAIL timeout, not a MPTCP socket close timeout.
> 
> Now mp_fail_response_expect_subflow() helper is only used by
> mptcp_mp_fail_no_response(), move the definition of the helper right
> before mptcp_mp_fail_no_response().
> 
> Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/281
> Fixes: d9fb797046c5 ("mptcp: Do not traverse the subflow connection list without lock")
> Signed-off-by: Geliang Tang <geliang.tang@suse.com>
> ---
>  net/mptcp/protocol.c | 41 ++++++++++++++++++++++++-----------------
>  net/mptcp/protocol.h |  1 +
>  2 files changed, 25 insertions(+), 17 deletions(-)
> 
> diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
> index d6aef4b13b8a..123fa2b3f200 100644
> --- a/net/mptcp/protocol.c
> +++ b/net/mptcp/protocol.c
> @@ -2167,25 +2167,13 @@ static void mptcp_retransmit_timer(struct timer_list *t)
>  	sock_put(sk);
>  }
>  
> -static struct mptcp_subflow_context *
> -mp_fail_response_expect_subflow(struct mptcp_sock *msk)
> -{
> -	struct mptcp_subflow_context *subflow, *ret = NULL;
> -
> -	mptcp_for_each_subflow(msk, subflow) {
> -		if (READ_ONCE(subflow->mp_fail_response_expect)) {
> -			ret = subflow;
> -			break;
> -		}
> -	}
> -
> -	return ret;
> -}

I think you don't need to move mp_fail_response_expect_subflow() around

> -
>  static void mptcp_timeout_timer(struct timer_list *t)
>  {
>  	struct sock *sk = from_timer(sk, t, sk_timer);
>  
> +	bh_lock_sock(sk);
> +	__set_bit(MPTCP_WORK_TIMEOUT, &mptcp_sk(sk)->flags);
> +	bh_unlock_sock(sk);

This is not correct: the caller must always manipulate msk->flags with
atomic operations. No need to acquire the msk spin lock

>  	mptcp_schedule_work(sk);
>  	sock_put(sk);
>  }
> @@ -2505,6 +2493,21 @@ static void __mptcp_retrans(struct sock *sk)
>  		mptcp_reset_timer(sk);
>  }
>  
> +static struct mptcp_subflow_context *
> +mp_fail_response_expect_subflow(struct mptcp_sock *msk)
> +{
> +	struct mptcp_subflow_context *subflow, *ret = NULL;
> +
> +	mptcp_for_each_subflow(msk, subflow) {
> +		if (READ_ONCE(subflow->mp_fail_response_expect)) {
> +			ret = subflow;
> +			break;
> +		}
> +	}
> +
> +	return ret;
> +}
> +
>  static void mptcp_mp_fail_no_response(struct mptcp_sock *msk)
>  {
>  	struct mptcp_subflow_context *subflow;
> @@ -2513,9 +2516,12 @@ static void mptcp_mp_fail_no_response(struct mptcp_sock *msk)
>  
>  	subflow = mp_fail_response_expect_subflow(msk);
>  	if (subflow) {
> +		ssk = mptcp_subflow_tcp_sock(subflow);
> +		if (sock_flag(ssk, SOCK_DEAD))
> +			return;

I *think* /guess it's better to check the msk SOCK_DEAD flag?!?
Additionally, if you do that, you can move the check in mptcp_worker()

Note that if mp_fail timeout overlap with the close timeout things will
be quite fuzzy. Very complex to follow/understand at best.

What about the following?
- subflow_check_data_avail() does:

	msk->mp_fail_subflow = subflow;
	msk->mp_fail_stamp = jiffies;
	sk_reset_timer((struct sock *)msk,
			&((struct sock *)msk)->sk_timer,
			jiffies + TCP_RTO_MAX);

- no changes to mptcp_timeout_timer()

- mptcp_worker does:
	if (msk->mp_fail_subflow &&
	    time_after(jiffies, msk->mp_fail_stamp)) {
		subflow = msk->mp_fail_subflow;
		pr_debug("MP_FAIL doesn't respond, reset the
subflow");

                ssk = mptcp_subflow_tcp_sock(subflow);
                slow = lock_sock_fast(ssk);
                mptcp_subflow_reset(ssk);
                unlock_sock_fast(ssk, slow);
	}
	
No need to traverse the subflow list and mp_fail will co-exist with
close timeout.
Note: mp_fail_stamp and mp_fail_subflow are new fields.

WDYT?

Thanks!

Paolo