[PATCH mptcp-next v9 0/4] refactor push pending

[PATCH mptcp-next v9 0/4] refactor push pending

[Geliang Tang]

v9:
- move "register default scheduler" out of this series.
- add missing "first = false" in patch 4
- patch 2 could be merged before the commit "mptcp: add
struct mptcp_sched_ops" as a non-bpf cleanup.

v8:
- merge "register default scheduler" into this series.
- Move "multi-subflow scheduling" out of this series.
- This series should be merged between the commit "mptcp: add get_subflow
wrappers" and "bpf: Add bpf_mptcp_sched_ops", except the squash-to patches.

v7:
- update delegate sending in patch 10
- rebase to export/20221011T063543.

v6:
- drop all msk->last_snd, add last_snd variable instead in patch 13
- fix lock_sock issue in patch 3
- merge squash-to patches

v5:
- address Mat's comments in v4.

v4:
- update __mptcp_subflow_push_pending as Mat suggested.
- add more patches from "BPF redundant scheduler" series.

v3:
- add a cleanup patch.
- remove msk->last_snd in mptcp_subflow_get_send().
- add the loop that calls the scheduler again in __mptcp_push_pending().

v2:
- add snd_burst check in dfrags loop as Mat suggested.

Refactor __mptcp_push_pending() and __mptcp_subflow_push_pending() to
remove duplicate code and support redundant scheduler more easily in
__mptcp_subflow_push_pending().

Geliang Tang (4):
  Squash to "mptcp: add get_subflow wrappers"
  mptcp: use msk instead of mptcp_sk
  mptcp: change 'first' as a parameter
  mptcp: refactor push_pending logic

 net/mptcp/protocol.c | 183 +++++++++++++++++++------------------------
 net/mptcp/protocol.h |   4 +-
 net/mptcp/sched.c    |  61 +++++++--------
 3 files changed, 111 insertions(+), 137 deletions(-)

-- 
2.35.3

[PATCH mptcp-next v9 1/4] Squash to "mptcp: add get_subflow wrappers"

[Geliang Tang]

Please update the commit log:

'''
This patch defines two new wrappers mptcp_sched_get_send() and
mptcp_sched_get_retrans(), invoke get_subflow() of msk->sched in them.

Set the subflow pointers array in struct mptcp_sched_data before invoking
get_subflow(), then it can be used in get_subflow() in the BPF contexts.

Check the subflow scheduled flags to test which subflow or subflows are
picked by the scheduler.

Move sock_owned_by_me() and the fallback check code from
mptcp_subflow_get_send/retrans() into the wrappers.
'''

Signed-off-by: Geliang Tang <geliang.tang@suse.com>
---
 net/mptcp/protocol.c |  8 +++---
 net/mptcp/protocol.h |  4 +--
 net/mptcp/sched.c    | 61 +++++++++++++++++++++-----------------------
 3 files changed, 35 insertions(+), 38 deletions(-)

diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index 3196d2a350f9..1cf8b1c68e09 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -1547,7 +1547,7 @@ void __mptcp_push_pending(struct sock *sk, unsigned int flags)
 			int ret = 0;
 
 			prev_ssk = ssk;
-			ssk = mptcp_sched_get_send(msk);
+			ssk = mptcp_subflow_get_send(msk);
 
 			/* First check. If the ssk has changed since
 			 * the last round, release prev_ssk
@@ -1616,7 +1616,7 @@ static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk)
 			 * check for a different subflow usage only after
 			 * spooling the first chunk of data
 			 */
-			xmit_ssk = first ? ssk : mptcp_sched_get_send(mptcp_sk(sk));
+			xmit_ssk = first ? ssk : mptcp_subflow_get_send(mptcp_sk(sk));
 			if (!xmit_ssk)
 				goto out;
 			if (xmit_ssk != ssk) {
@@ -2500,7 +2500,7 @@ static void __mptcp_retrans(struct sock *sk)
 	mptcp_clean_una_wakeup(sk);
 
 	/* first check ssk: need to kick "stale" logic */
-	ssk = mptcp_sched_get_retrans(msk);
+	ssk = mptcp_subflow_get_retrans(msk);
 	dfrag = mptcp_rtx_head(sk);
 	if (!dfrag) {
 		if (mptcp_data_fin_enabled(msk)) {
@@ -3218,7 +3218,7 @@ void __mptcp_check_push(struct sock *sk, struct sock *ssk)
 		return;
 
 	if (!sock_owned_by_user(sk)) {
-		struct sock *xmit_ssk = mptcp_sched_get_send(mptcp_sk(sk));
+		struct sock *xmit_ssk = mptcp_subflow_get_send(mptcp_sk(sk));
 
 		if (xmit_ssk == ssk)
 			__mptcp_subflow_push_pending(sk, ssk);
diff --git a/net/mptcp/protocol.h b/net/mptcp/protocol.h
index 2358a4083eb3..8f48f881adf8 100644
--- a/net/mptcp/protocol.h
+++ b/net/mptcp/protocol.h
@@ -643,8 +643,8 @@ void mptcp_subflow_set_scheduled(struct mptcp_subflow_context *subflow,
 				 bool scheduled);
 struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk);
 struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk);
-struct sock *mptcp_sched_get_send(struct mptcp_sock *msk);
-struct sock *mptcp_sched_get_retrans(struct mptcp_sock *msk);
+int mptcp_sched_get_send(struct mptcp_sock *msk);
+int mptcp_sched_get_retrans(struct mptcp_sock *msk);
 
 static inline bool __tcp_can_send(const struct sock *ssk)
 {
diff --git a/net/mptcp/sched.c b/net/mptcp/sched.c
index 044c5ec8bbfb..9b128714055a 100644
--- a/net/mptcp/sched.c
+++ b/net/mptcp/sched.c
@@ -114,67 +114,64 @@ static int mptcp_sched_data_init(struct mptcp_sock *msk, bool reinject,
 	for (; i < MPTCP_SUBFLOWS_MAX; i++)
 		data->contexts[i] = NULL;
 
+	msk->snd_burst = 0;
+
 	return 0;
 }
 
-struct sock *mptcp_sched_get_send(struct mptcp_sock *msk)
+int mptcp_sched_get_send(struct mptcp_sock *msk)
 {
 	struct mptcp_sched_data data;
 	struct sock *ssk = NULL;
-	int i;
 
-	sock_owned_by_me((struct sock *)msk);
+	sock_owned_by_me((const struct sock *)msk);
 
 	/* the following check is moved out of mptcp_subflow_get_send */
 	if (__mptcp_check_fallback(msk)) {
-		if (!msk->first)
-			return NULL;
-		return __tcp_can_send(msk->first) &&
-		       sk_stream_memory_free(msk->first) ? msk->first : NULL;
+		if (msk->first &&
+		    __tcp_can_send(msk->first) &&
+		    sk_stream_memory_free(msk->first)) {
+			mptcp_subflow_set_scheduled(mptcp_subflow_ctx(msk->first), true);
+			return 0;
+		}
+		return -EINVAL;
 	}
 
-	if (!msk->sched)
-		return mptcp_subflow_get_send(msk);
+	if (!msk->sched) {
+		ssk = mptcp_subflow_get_send(msk);
+		if (!ssk)
+			return -EINVAL;
+		mptcp_subflow_set_scheduled(mptcp_subflow_ctx(ssk), true);
+		return 0;
+	}
 
 	mptcp_sched_data_init(msk, false, &data);
 	msk->sched->get_subflow(msk, &data);
 
-	for (i = 0; i < MPTCP_SUBFLOWS_MAX; i++) {
-		if (data.contexts[i] && READ_ONCE(data.contexts[i]->scheduled)) {
-			ssk = data.contexts[i]->tcp_sock;
-			msk->last_snd = ssk;
-			break;
-		}
-	}
-
-	return ssk;
+	return 0;
 }
 
-struct sock *mptcp_sched_get_retrans(struct mptcp_sock *msk)
+int mptcp_sched_get_retrans(struct mptcp_sock *msk)
 {
 	struct mptcp_sched_data data;
 	struct sock *ssk = NULL;
-	int i;
 
 	sock_owned_by_me((const struct sock *)msk);
 
 	/* the following check is moved out of mptcp_subflow_get_retrans */
 	if (__mptcp_check_fallback(msk))
-		return NULL;
+		return -EINVAL;
 
-	if (!msk->sched)
-		return mptcp_subflow_get_retrans(msk);
+	if (!msk->sched) {
+		ssk = mptcp_subflow_get_retrans(msk);
+		if (!ssk)
+			return -EINVAL;
+		mptcp_subflow_set_scheduled(mptcp_subflow_ctx(ssk), true);
+		return 0;
+	}
 
 	mptcp_sched_data_init(msk, true, &data);
 	msk->sched->get_subflow(msk, &data);
 
-	for (i = 0; i < MPTCP_SUBFLOWS_MAX; i++) {
-		if (data.contexts[i] && READ_ONCE(data.contexts[i]->scheduled)) {
-			ssk = data.contexts[i]->tcp_sock;
-			msk->last_snd = ssk;
-			break;
-		}
-	}
-
-	return ssk;
+	return 0;
 }
-- 
2.35.3

[PATCH mptcp-next v9 2/4] mptcp: use msk instead of mptcp_sk

[Geliang Tang]

Use msk instead of mptcp_sk(sk) in the functions where the variable
"msk = mptcp_sk(sk)" has been defined.

Signed-off-by: Geliang Tang <geliang.tang@suse.com>
---
 net/mptcp/protocol.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index 1cf8b1c68e09..11be5b9c379d 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -1616,7 +1616,7 @@ static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk)
 			 * check for a different subflow usage only after
 			 * spooling the first chunk of data
 			 */
-			xmit_ssk = first ? ssk : mptcp_subflow_get_send(mptcp_sk(sk));
+			xmit_ssk = first ? ssk : mptcp_subflow_get_send(msk);
 			if (!xmit_ssk)
 				goto out;
 			if (xmit_ssk != ssk) {
@@ -2261,7 +2261,7 @@ bool __mptcp_retransmit_pending_data(struct sock *sk)
 	struct mptcp_data_frag *cur, *rtx_head;
 	struct mptcp_sock *msk = mptcp_sk(sk);
 
-	if (__mptcp_check_fallback(mptcp_sk(sk)))
+	if (__mptcp_check_fallback(msk))
 		return false;
 
 	if (tcp_rtx_and_write_queues_empty(sk))
@@ -2943,7 +2943,7 @@ bool __mptcp_close(struct sock *sk, long timeout)
 
 	sock_hold(sk);
 	pr_debug("msk=%p state=%d", sk, sk->sk_state);
-	if (mptcp_sk(sk)->token)
+	if (msk->token)
 		mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL);
 
 	if (sk->sk_state == TCP_CLOSE) {
@@ -3002,8 +3002,8 @@ static int mptcp_disconnect(struct sock *sk, int flags)
 	mptcp_stop_timer(sk);
 	sk_stop_timer(sk, &sk->sk_timer);
 
-	if (mptcp_sk(sk)->token)
-		mptcp_event(MPTCP_EVENT_CLOSED, mptcp_sk(sk), NULL, GFP_KERNEL);
+	if (msk->token)
+		mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL);
 
 	/* msk->subflow is still intact, the following will not free the first
 	 * subflow
-- 
2.35.3

[PATCH mptcp-next v9 3/4] mptcp: change 'first' as a parameter

[Geliang Tang]

The function mptcp_subflow_process_delegated() uses the input ssk first,
while __mptcp_check_push() invokes the packet scheduler first.

So this patch adds a new parameter named 'first' for the function
__mptcp_subflow_push_pending() to deal with these two cases separately.

With this change, the code that invokes the packet scheduler in the
function __mptcp_check_push() can be removed, and replaced by invoking
__mptcp_subflow_push_pending() directly.

Signed-off-by: Geliang Tang <geliang.tang@suse.com>
---
 net/mptcp/protocol.c | 20 ++++++--------------
 1 file changed, 6 insertions(+), 14 deletions(-)

diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index 11be5b9c379d..ddeb8b36a677 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -1593,7 +1593,7 @@ void __mptcp_push_pending(struct sock *sk, unsigned int flags)
 		__mptcp_check_send_data_fin(sk);
 }
 
-static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk)
+static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk, bool first)
 {
 	struct mptcp_sock *msk = mptcp_sk(sk);
 	struct mptcp_sendmsg_info info = {
@@ -1602,7 +1602,6 @@ static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk)
 	struct mptcp_data_frag *dfrag;
 	struct sock *xmit_ssk;
 	int len, copied = 0;
-	bool first = true;
 
 	info.flags = 0;
 	while ((dfrag = mptcp_send_head(sk))) {
@@ -1612,8 +1611,7 @@ static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk)
 		while (len > 0) {
 			int ret = 0;
 
-			/* the caller already invoked the packet scheduler,
-			 * check for a different subflow usage only after
+			/* check for a different subflow usage only after
 			 * spooling the first chunk of data
 			 */
 			xmit_ssk = first ? ssk : mptcp_subflow_get_send(msk);
@@ -3217,16 +3215,10 @@ void __mptcp_check_push(struct sock *sk, struct sock *ssk)
 	if (!mptcp_send_head(sk))
 		return;
 
-	if (!sock_owned_by_user(sk)) {
-		struct sock *xmit_ssk = mptcp_subflow_get_send(mptcp_sk(sk));
-
-		if (xmit_ssk == ssk)
-			__mptcp_subflow_push_pending(sk, ssk);
-		else if (xmit_ssk)
-			mptcp_subflow_delegate(mptcp_subflow_ctx(xmit_ssk), MPTCP_DELEGATE_SEND);
-	} else {
+	if (!sock_owned_by_user(sk))
+		__mptcp_subflow_push_pending(sk, ssk, false);
+	else
 		__set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
-	}
 }
 
 #define MPTCP_FLAGS_PROCESS_CTX_NEED (BIT(MPTCP_PUSH_PENDING) | \
@@ -3317,7 +3309,7 @@ void mptcp_subflow_process_delegated(struct sock *ssk)
 	if (test_bit(MPTCP_DELEGATE_SEND, &subflow->delegated_status)) {
 		mptcp_data_lock(sk);
 		if (!sock_owned_by_user(sk))
-			__mptcp_subflow_push_pending(sk, ssk);
+			__mptcp_subflow_push_pending(sk, ssk, true);
 		else
 			__set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
 		mptcp_data_unlock(sk);
-- 
2.35.3

[PATCH mptcp-next v9 4/4] mptcp: refactor push_pending logic

[Geliang Tang]

To support redundant package schedulers more easily, this patch refactors
__mptcp_push_pending() logic from:

For each dfrag:
	While sends succeed:
		Call the scheduler (selects subflow and msk->snd_burst)
		Update subflow locks (push/release/acquire as needed)
		Send the dfrag data with mptcp_sendmsg_frag()
		Update already_sent, snd_nxt, snd_burst
	Update msk->first_pending
Push/release on final subflow

->

While the scheduler selects one subflow:
	Lock the subflow
	For each pending dfrag:
		While sends succeed:
			Send the dfrag data with mptcp_sendmsg_frag()
			Update already_sent, snd_nxt, snd_burst
		Update msk->first_pending
		Break if required by msk->snd_burst / etc
	Push and release the subflow

Refactors __mptcp_subflow_push_pending logic from:

For each dfrag:
	While sends succeed:
		Call the scheduler (selects subflow and msk->snd_burst)
		Send the dfrag data with mptcp_subflow_delegate(), break
		Send the dfrag data with mptcp_sendmsg_frag()
		Update dfrag->already_sent, msk->snd_nxt, msk->snd_burst
	Update msk->first_pending

->

While first_pending isn't empty:
	Call the scheduler (selects subflow and msk->snd_burst)
	Send the dfrag data with mptcp_subflow_delegate(), break
	Send the dfrag data with mptcp_sendmsg_frag()
	For each pending dfrag:
		While sends succeed:
			Send the dfrag data with mptcp_sendmsg_frag()
			Update already_sent, snd_nxt, snd_burst
		Update msk->first_pending
		Break if required by msk->snd_burst / etc

Move the duplicate code from __mptcp_push_pending() and
__mptcp_subflow_push_pending() into a new helper function, named
__subflow_push_pending(). Simplify __mptcp_push_pending() and
__mptcp_subflow_push_pending() by invoking this helper.

Also move the burst check conditions out of the function
mptcp_subflow_get_send(), check them in __mptcp_push_pending() and
__mptcp_subflow_push_pending() in the inner "for each pending dfrag"
loop.

Signed-off-by: Geliang Tang <geliang.tang@suse.com>
---
 net/mptcp/protocol.c | 155 +++++++++++++++++++------------------------
 1 file changed, 70 insertions(+), 85 deletions(-)

diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index ddeb8b36a677..9b73297f6543 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -1417,14 +1417,6 @@ struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)
 	u64 linger_time;
 	long tout = 0;
 
-	/* re-use last subflow, if the burst allow that */
-	if (msk->last_snd && msk->snd_burst > 0 &&
-	    sk_stream_memory_free(msk->last_snd) &&
-	    mptcp_subflow_active(mptcp_subflow_ctx(msk->last_snd))) {
-		mptcp_set_timeout(sk);
-		return msk->last_snd;
-	}
-
 	/* pick the subflow with the lower wmem/wspace ratio */
 	for (i = 0; i < SSK_MODE_MAX; ++i) {
 		send_info[i].ssk = NULL;
@@ -1491,12 +1483,6 @@ struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)
 	return ssk;
 }
 
-static void mptcp_push_release(struct sock *ssk, struct mptcp_sendmsg_info *info)
-{
-	tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle, info->size_goal);
-	release_sock(ssk);
-}
-
 static void mptcp_update_post_push(struct mptcp_sock *msk,
 				   struct mptcp_data_frag *dfrag,
 				   u32 sent)
@@ -1528,68 +1514,80 @@ void mptcp_check_and_set_pending(struct sock *sk)
 		mptcp_sk(sk)->push_pending |= BIT(MPTCP_PUSH_PENDING);
 }
 
-void __mptcp_push_pending(struct sock *sk, unsigned int flags)
+static int __subflow_push_pending(struct sock *sk, struct sock *ssk,
+				  struct mptcp_sendmsg_info *info)
 {
-	struct sock *prev_ssk = NULL, *ssk = NULL;
 	struct mptcp_sock *msk = mptcp_sk(sk);
-	struct mptcp_sendmsg_info info = {
-				.flags = flags,
-	};
-	bool do_check_data_fin = false;
 	struct mptcp_data_frag *dfrag;
-	int len;
+	int len, copied = 0, err = 0;
 
 	while ((dfrag = mptcp_send_head(sk))) {
-		info.sent = dfrag->already_sent;
-		info.limit = dfrag->data_len;
+		info->sent = dfrag->already_sent;
+		info->limit = dfrag->data_len;
 		len = dfrag->data_len - dfrag->already_sent;
 		while (len > 0) {
 			int ret = 0;
 
-			prev_ssk = ssk;
-			ssk = mptcp_subflow_get_send(msk);
-
-			/* First check. If the ssk has changed since
-			 * the last round, release prev_ssk
-			 */
-			if (ssk != prev_ssk && prev_ssk)
-				mptcp_push_release(prev_ssk, &info);
-			if (!ssk)
-				goto out;
-
-			/* Need to lock the new subflow only if different
-			 * from the previous one, otherwise we are still
-			 * helding the relevant lock
-			 */
-			if (ssk != prev_ssk)
-				lock_sock(ssk);
-
-			ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
+			ret = mptcp_sendmsg_frag(sk, ssk, dfrag, info);
 			if (ret <= 0) {
-				if (ret == -EAGAIN)
-					continue;
-				mptcp_push_release(ssk, &info);
+				err = copied ? : ret;
 				goto out;
 			}
 
-			do_check_data_fin = true;
-			info.sent += ret;
+			info->sent += ret;
+			copied += ret;
 			len -= ret;
 
 			mptcp_update_post_push(msk, dfrag, ret);
 		}
 		WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
+
+		if (msk->snd_burst <= 0 ||
+		    !sk_stream_memory_free(ssk) ||
+		    !mptcp_subflow_active(mptcp_subflow_ctx(ssk))) {
+			err = copied ? : -EAGAIN;
+			goto out;
+		}
+		mptcp_set_timeout(sk);
+	}
+	err = copied;
+
+out:
+	if (copied) {
+		tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle,
+			 info->size_goal);
 	}
 
-	/* at this point we held the socket lock for the last subflow we used */
-	if (ssk)
-		mptcp_push_release(ssk, &info);
+	return err;
+}
+
+void __mptcp_push_pending(struct sock *sk, unsigned int flags)
+{
+	struct mptcp_sock *msk = mptcp_sk(sk);
+	struct mptcp_sendmsg_info info = {
+				.flags = flags,
+	};
+	struct sock *ssk;
+	int ret = 0;
+
+again:
+	while (mptcp_send_head(sk) && (ssk = mptcp_subflow_get_send(msk))) {
+		lock_sock(ssk);
+		ret = __subflow_push_pending(sk, ssk, &info);
+		release_sock(ssk);
+
+		if (ret <= 0) {
+			if (ret == -EAGAIN)
+				goto again;
+			goto out;
+		}
+	}
 
 out:
 	/* ensure the rtx timer is running */
 	if (!mptcp_timer_pending(sk))
 		mptcp_reset_timer(sk);
-	if (do_check_data_fin)
+	if (ret > 0)
 		__mptcp_check_send_data_fin(sk);
 }
 
@@ -1599,51 +1597,38 @@ static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk, bool
 	struct mptcp_sendmsg_info info = {
 		.data_lock_held = true,
 	};
-	struct mptcp_data_frag *dfrag;
 	struct sock *xmit_ssk;
-	int len, copied = 0;
+	int ret = 0;
 
 	info.flags = 0;
-	while ((dfrag = mptcp_send_head(sk))) {
-		info.sent = dfrag->already_sent;
-		info.limit = dfrag->data_len;
-		len = dfrag->data_len - dfrag->already_sent;
-		while (len > 0) {
-			int ret = 0;
-
-			/* check for a different subflow usage only after
-			 * spooling the first chunk of data
-			 */
-			xmit_ssk = first ? ssk : mptcp_subflow_get_send(msk);
-			if (!xmit_ssk)
-				goto out;
-			if (xmit_ssk != ssk) {
-				mptcp_subflow_delegate(mptcp_subflow_ctx(xmit_ssk),
-						       MPTCP_DELEGATE_SEND);
-				goto out;
-			}
-
-			ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
-			if (ret <= 0)
-				goto out;
-
-			info.sent += ret;
-			copied += ret;
-			len -= ret;
-			first = false;
+again:
+	while (mptcp_send_head(sk)) {
+		/* check for a different subflow usage only after
+		 * spooling the first chunk of data
+		 */
+		xmit_ssk = first ? ssk : mptcp_subflow_get_send(msk);
+		if (!xmit_ssk)
+			goto out;
+		if (xmit_ssk != ssk) {
+			mptcp_subflow_delegate(mptcp_subflow_ctx(xmit_ssk),
+					       MPTCP_DELEGATE_SEND);
+			goto out;
+		}
 
-			mptcp_update_post_push(msk, dfrag, ret);
+		ret = __subflow_push_pending(sk, ssk, &info);
+		first = false;
+		if (ret <= 0) {
+			if (ret == -EAGAIN)
+				goto again;
+			break;
 		}
-		WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
 	}
 
 out:
 	/* __mptcp_alloc_tx_skb could have released some wmem and we are
 	 * not going to flush it via release_sock()
 	 */
-	if (copied) {
-		tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
-			 info.size_goal);
+	if (ret > 0) {
 		if (!mptcp_timer_pending(sk))
 			mptcp_reset_timer(sk);
 
-- 
2.35.3

Re: mptcp: refactor push_pending logic: Tests Results

[MPTCP CI]

Hi Geliang,

Thank you for your modifications, that's great!

Our CI did some validations and here is its report:

- KVM Validation: normal:
  - Unstable: 1 failed test(s): selftest_simult_flows 🔴:
  - Task: https://cirrus-ci.com/task/5202885733515264
  - Summary: https://api.cirrus-ci.com/v1/artifact/task/5202885733515264/summary/summary.txt

- {"code":404,"message":
  - "Can't find artifacts containing file conclusion.txt"}:
  - Task: https://cirrus-ci.com/task/6328785640357888
  - Summary: https://api.cirrus-ci.com/v1/artifact/task/6328785640357888/summary/summary.txt

Initiator: Patchew Applier
Commits: https://github.com/multipath-tcp/mptcp_net-next/commits/ddd3c597286e


If there are some issues, you can reproduce them using the same environment as
the one used by the CI thanks to a docker image, e.g.:

    $ cd [kernel source code]
    $ docker run -v "${PWD}:${PWD}:rw" -w "${PWD}" --privileged --rm -it \
        --pull always mptcp/mptcp-upstream-virtme-docker:latest \
        auto-debug

For more details:

    https://github.com/multipath-tcp/mptcp-upstream-virtme-docker


Please note that despite all the efforts that have been already done to have a
stable tests suite when executed on a public CI like here, it is possible some
reported issues are not due to your modifications. Still, do not hesitate to
help us improve that ;-)

Cheers,
MPTCP GH Action bot
Bot operated by Matthieu Baerts (Tessares)

Re: [PATCH mptcp-next v9 4/4] mptcp: refactor push_pending logic

[Mat Martineau]

On Tue, 18 Oct 2022, Geliang Tang wrote:

> To support redundant package schedulers more easily, this patch refactors
> __mptcp_push_pending() logic from:
>
> For each dfrag:
> 	While sends succeed:
> 		Call the scheduler (selects subflow and msk->snd_burst)
> 		Update subflow locks (push/release/acquire as needed)
> 		Send the dfrag data with mptcp_sendmsg_frag()
> 		Update already_sent, snd_nxt, snd_burst
> 	Update msk->first_pending
> Push/release on final subflow
>
> ->
>
> While the scheduler selects one subflow:
> 	Lock the subflow
> 	For each pending dfrag:
> 		While sends succeed:
> 			Send the dfrag data with mptcp_sendmsg_frag()
> 			Update already_sent, snd_nxt, snd_burst
> 		Update msk->first_pending
> 		Break if required by msk->snd_burst / etc
> 	Push and release the subflow
>
> Refactors __mptcp_subflow_push_pending logic from:
>
> For each dfrag:
> 	While sends succeed:
> 		Call the scheduler (selects subflow and msk->snd_burst)
> 		Send the dfrag data with mptcp_subflow_delegate(), break
> 		Send the dfrag data with mptcp_sendmsg_frag()
> 		Update dfrag->already_sent, msk->snd_nxt, msk->snd_burst
> 	Update msk->first_pending
>
> ->
>
> While first_pending isn't empty:
> 	Call the scheduler (selects subflow and msk->snd_burst)
> 	Send the dfrag data with mptcp_subflow_delegate(), break
> 	Send the dfrag data with mptcp_sendmsg_frag()
> 	For each pending dfrag:
> 		While sends succeed:
> 			Send the dfrag data with mptcp_sendmsg_frag()
> 			Update already_sent, snd_nxt, snd_burst
> 		Update msk->first_pending
> 		Break if required by msk->snd_burst / etc
>
> Move the duplicate code from __mptcp_push_pending() and
> __mptcp_subflow_push_pending() into a new helper function, named
> __subflow_push_pending(). Simplify __mptcp_push_pending() and
> __mptcp_subflow_push_pending() by invoking this helper.
>
> Also move the burst check conditions out of the function
> mptcp_subflow_get_send(), check them in __mptcp_push_pending() and
> __mptcp_subflow_push_pending() in the inner "for each pending dfrag"
> loop.
>
> Signed-off-by: Geliang Tang <geliang.tang@suse.com>
> ---
> net/mptcp/protocol.c | 155 +++++++++++++++++++------------------------
> 1 file changed, 70 insertions(+), 85 deletions(-)
>
> diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
> index ddeb8b36a677..9b73297f6543 100644
> --- a/net/mptcp/protocol.c
> +++ b/net/mptcp/protocol.c
> @@ -1417,14 +1417,6 @@ struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)
> 	u64 linger_time;
> 	long tout = 0;
>
> -	/* re-use last subflow, if the burst allow that */
> -	if (msk->last_snd && msk->snd_burst > 0 &&
> -	    sk_stream_memory_free(msk->last_snd) &&
> -	    mptcp_subflow_active(mptcp_subflow_ctx(msk->last_snd))) {
> -		mptcp_set_timeout(sk);
> -		return msk->last_snd;
> -	}
> -
> 	/* pick the subflow with the lower wmem/wspace ratio */
> 	for (i = 0; i < SSK_MODE_MAX; ++i) {
> 		send_info[i].ssk = NULL;
> @@ -1491,12 +1483,6 @@ struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)
> 	return ssk;
> }
>
> -static void mptcp_push_release(struct sock *ssk, struct mptcp_sendmsg_info *info)
> -{
> -	tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle, info->size_goal);
> -	release_sock(ssk);
> -}
> -
> static void mptcp_update_post_push(struct mptcp_sock *msk,
> 				   struct mptcp_data_frag *dfrag,
> 				   u32 sent)
> @@ -1528,68 +1514,80 @@ void mptcp_check_and_set_pending(struct sock *sk)
> 		mptcp_sk(sk)->push_pending |= BIT(MPTCP_PUSH_PENDING);
> }
>
> -void __mptcp_push_pending(struct sock *sk, unsigned int flags)
> +static int __subflow_push_pending(struct sock *sk, struct sock *ssk,
> +				  struct mptcp_sendmsg_info *info)
> {
> -	struct sock *prev_ssk = NULL, *ssk = NULL;
> 	struct mptcp_sock *msk = mptcp_sk(sk);
> -	struct mptcp_sendmsg_info info = {
> -				.flags = flags,
> -	};
> -	bool do_check_data_fin = false;
> 	struct mptcp_data_frag *dfrag;
> -	int len;
> +	int len, copied = 0, err = 0;
>
> 	while ((dfrag = mptcp_send_head(sk))) {
> -		info.sent = dfrag->already_sent;
> -		info.limit = dfrag->data_len;
> +		info->sent = dfrag->already_sent;
> +		info->limit = dfrag->data_len;
> 		len = dfrag->data_len - dfrag->already_sent;
> 		while (len > 0) {
> 			int ret = 0;
>
> -			prev_ssk = ssk;
> -			ssk = mptcp_subflow_get_send(msk);
> -
> -			/* First check. If the ssk has changed since
> -			 * the last round, release prev_ssk
> -			 */
> -			if (ssk != prev_ssk && prev_ssk)
> -				mptcp_push_release(prev_ssk, &info);
> -			if (!ssk)
> -				goto out;
> -
> -			/* Need to lock the new subflow only if different
> -			 * from the previous one, otherwise we are still
> -			 * helding the relevant lock
> -			 */
> -			if (ssk != prev_ssk)
> -				lock_sock(ssk);
> -
> -			ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
> +			ret = mptcp_sendmsg_frag(sk, ssk, dfrag, info);
> 			if (ret <= 0) {
> -				if (ret == -EAGAIN)
> -					continue;
> -				mptcp_push_release(ssk, &info);
> +				err = copied ? : ret;
> 				goto out;
> 			}
>
> -			do_check_data_fin = true;
> -			info.sent += ret;
> +			info->sent += ret;
> +			copied += ret;
> 			len -= ret;
>
> 			mptcp_update_post_push(msk, dfrag, ret);
> 		}
> 		WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
> +
> +		if (msk->snd_burst <= 0 ||
> +		    !sk_stream_memory_free(ssk) ||
> +		    !mptcp_subflow_active(mptcp_subflow_ctx(ssk))) {
> +			err = copied ? : -EAGAIN;
> +			goto out;
> +		}
> +		mptcp_set_timeout(sk);
> +	}
> +	err = copied;
> +
> +out:
> +	if (copied) {
> +		tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle,
> +			 info->size_goal);
> 	}
>
> -	/* at this point we held the socket lock for the last subflow we used */
> -	if (ssk)
> -		mptcp_push_release(ssk, &info);
> +	return err;
> +}
> +
> +void __mptcp_push_pending(struct sock *sk, unsigned int flags)
> +{
> +	struct mptcp_sock *msk = mptcp_sk(sk);
> +	struct mptcp_sendmsg_info info = {
> +				.flags = flags,
> +	};
> +	struct sock *ssk;
> +	int ret = 0;
> +
> +again:
> +	while (mptcp_send_head(sk) && (ssk = mptcp_subflow_get_send(msk))) {
> +		lock_sock(ssk);
> +		ret = __subflow_push_pending(sk, ssk, &info);
> +		release_sock(ssk);

Thanks for reworking this patch set, it's easier to see what's going on 
now.

With this refactoring a long send will push, release, and re-lock the 
subflow for every "burst" (as limited by snd_burst or by the available 
buffer space). That seems like it could add noticeable overhead for 
high-throughput connections, compared to the old code that would defer the 
push & unlock. In earlier reviews I didn't consider this carefully enough: 
I thought that __subflow_push_pending() would usually be able to send all 
the pending dfrags, but I didn't consider sendmsg calls with lots of data 
(more than the burst size).

I think it's possible to keep the push/unlock optimization in 
__mptcp_push_pending() if the tcp_push() is handled outside 
__subflow_push_pending().

Would likely need more information returned from the helper to give the 
caller of __subflow_push_pending() enough information, like:

static int __subflow_push_pending(struct sock *sk, struct sock *ssk,
 				  struct mptcp_sendmsg_info *info,
 				  bool *needs_push)

I'll talk to Paolo about this on Thursday too (and he's welcome to chime 
in here :) )

> +
> +		if (ret <= 0) {
> +			if (ret == -EAGAIN)
> +				goto again;

Better to use a 'continue' here.

> +			goto out;
> +		}
> +	}
>
> out:
> 	/* ensure the rtx timer is running */
> 	if (!mptcp_timer_pending(sk))
> 		mptcp_reset_timer(sk);
> -	if (do_check_data_fin)
> +	if (ret > 0)
> 		__mptcp_check_send_data_fin(sk);
> }
>
> @@ -1599,51 +1597,38 @@ static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk, bool
> 	struct mptcp_sendmsg_info info = {
> 		.data_lock_held = true,
> 	};
> -	struct mptcp_data_frag *dfrag;
> 	struct sock *xmit_ssk;
> -	int len, copied = 0;
> +	int ret = 0;
>
> 	info.flags = 0;
> -	while ((dfrag = mptcp_send_head(sk))) {
> -		info.sent = dfrag->already_sent;
> -		info.limit = dfrag->data_len;
> -		len = dfrag->data_len - dfrag->already_sent;
> -		while (len > 0) {
> -			int ret = 0;
> -
> -			/* check for a different subflow usage only after
> -			 * spooling the first chunk of data
> -			 */
> -			xmit_ssk = first ? ssk : mptcp_subflow_get_send(msk);
> -			if (!xmit_ssk)
> -				goto out;
> -			if (xmit_ssk != ssk) {
> -				mptcp_subflow_delegate(mptcp_subflow_ctx(xmit_ssk),
> -						       MPTCP_DELEGATE_SEND);
> -				goto out;
> -			}
> -
> -			ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
> -			if (ret <= 0)
> -				goto out;
> -
> -			info.sent += ret;
> -			copied += ret;
> -			len -= ret;
> -			first = false;
> +again:
> +	while (mptcp_send_head(sk)) {
> +		/* check for a different subflow usage only after
> +		 * spooling the first chunk of data
> +		 */
> +		xmit_ssk = first ? ssk : mptcp_subflow_get_send(msk);
> +		if (!xmit_ssk)
> +			goto out;
> +		if (xmit_ssk != ssk) {
> +			mptcp_subflow_delegate(mptcp_subflow_ctx(xmit_ssk),
> +					       MPTCP_DELEGATE_SEND);
> +			goto out;
> +		}
>
> -			mptcp_update_post_push(msk, dfrag, ret);
> +		ret = __subflow_push_pending(sk, ssk, &info);
> +		first = false;
> +		if (ret <= 0) {
> +			if (ret == -EAGAIN)
> +				goto again;

Another place to use 'continue'.

> +			break;
> 		}
> -		WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
> 	}
>
> out:
> 	/* __mptcp_alloc_tx_skb could have released some wmem and we are
> 	 * not going to flush it via release_sock()
> 	 */
> -	if (copied) {
> -		tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
> -			 info.size_goal);
> +	if (ret > 0) {
> 		if (!mptcp_timer_pending(sk))
> 			mptcp_reset_timer(sk);
>
> -- 
> 2.35.3
>
>
>

--
Mat Martineau
Intel

Re: mptcp: refactor push_pending logic: Tests Results

[MPTCP CI]

Hi Geliang,

Thank you for your modifications, that's great!

Our CI did some validations and here is its report:

- KVM Validation: normal:
  - Unstable: 1 failed test(s): packetdrill_add_addr 🔴:
  - Task: https://cirrus-ci.com/task/4741266072666112
  - Summary: https://api.cirrus-ci.com/v1/artifact/task/4741266072666112/summary/summary.txt

- {"code":404,"message":
  - "Can't find artifacts containing file conclusion.txt"}:
  - Task: https://cirrus-ci.com/task/5867165979508736
  - Summary: https://api.cirrus-ci.com/v1/artifact/task/5867165979508736/summary/summary.txt

Initiator: Patchew Applier
Commits: https://github.com/multipath-tcp/mptcp_net-next/commits/12a4c489939c


If there are some issues, you can reproduce them using the same environment as
the one used by the CI thanks to a docker image, e.g.:

    $ cd [kernel source code]
    $ docker run -v "${PWD}:${PWD}:rw" -w "${PWD}" --privileged --rm -it \
        --pull always mptcp/mptcp-upstream-virtme-docker:latest \
        auto-debug

For more details:

    https://github.com/multipath-tcp/mptcp-upstream-virtme-docker


Please note that despite all the efforts that have been already done to have a
stable tests suite when executed on a public CI like here, it is possible some
reported issues are not due to your modifications. Still, do not hesitate to
help us improve that ;-)

Cheers,
MPTCP GH Action bot
Bot operated by Matthieu Baerts (Tessares)

Re: [PATCH mptcp-next v9 0/4] refactor push pending

[Mat Martineau]

On Tue, 18 Oct 2022, Geliang Tang wrote:

> v9:
> - move "register default scheduler" out of this series.
> - add missing "first = false" in patch 4
> - patch 2 could be merged before the commit "mptcp: add
> struct mptcp_sched_ops" as a non-bpf cleanup.
>

The first 3 patches look good to me:

Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>


I have some other comments for patch 4.


> v8:
> - merge "register default scheduler" into this series.
> - Move "multi-subflow scheduling" out of this series.
> - This series should be merged between the commit "mptcp: add get_subflow
> wrappers" and "bpf: Add bpf_mptcp_sched_ops", except the squash-to patches.
>
> v7:
> - update delegate sending in patch 10
> - rebase to export/20221011T063543.
>
> v6:
> - drop all msk->last_snd, add last_snd variable instead in patch 13
> - fix lock_sock issue in patch 3
> - merge squash-to patches
>
> v5:
> - address Mat's comments in v4.
>
> v4:
> - update __mptcp_subflow_push_pending as Mat suggested.
> - add more patches from "BPF redundant scheduler" series.
>
> v3:
> - add a cleanup patch.
> - remove msk->last_snd in mptcp_subflow_get_send().
> - add the loop that calls the scheduler again in __mptcp_push_pending().
>
> v2:
> - add snd_burst check in dfrags loop as Mat suggested.
>
> Refactor __mptcp_push_pending() and __mptcp_subflow_push_pending() to
> remove duplicate code and support redundant scheduler more easily in
> __mptcp_subflow_push_pending().
>
> Geliang Tang (4):
>  Squash to "mptcp: add get_subflow wrappers"
>  mptcp: use msk instead of mptcp_sk
>  mptcp: change 'first' as a parameter
>  mptcp: refactor push_pending logic
>
> net/mptcp/protocol.c | 183 +++++++++++++++++++------------------------
> net/mptcp/protocol.h |   4 +-
> net/mptcp/sched.c    |  61 +++++++--------
> 3 files changed, 111 insertions(+), 137 deletions(-)
>
> -- 
> 2.35.3
>
>
>

--
Mat Martineau
Intel

Re: [PATCH mptcp-next v9 0/4] refactor push pending

[Matthieu Baerts]

Hi Geliang, Mat,

On 19/10/2022 02:16, Mat Martineau wrote:
> On Tue, 18 Oct 2022, Geliang Tang wrote:
> 
>> v9:
>> - move "register default scheduler" out of this series.
>> - add missing "first = false" in patch 4
>> - patch 2 could be merged before the commit "mptcp: add
>> struct mptcp_sched_ops" as a non-bpf cleanup.
>>
> 
> The first 3 patches look good to me:
> 
> Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>

Thank you for the patches and the reviews!

I just added these patches in our tree (feat. with net-next) with Mat's
RvB tag:

- f7b33377ef44: "squashed" patch 1/4 in "mptcp: add get_subflow wrappers"
- abc20912434f: tg:msg: update commit message
- 810d6dba28d6: mptcp: use msk instead of mptcp_sk
- 3afbefdac9ae: mptcp: change 'first' as a parameter
- Results: e4c052c32f02..1f94545f93dd (export)

Tests are now in progress:

https://cirrus-ci.com/github/multipath-tcp/mptcp_net-next/export/20221020T142332

Cheers,
Matt
-- 
Tessares | Belgium | Hybrid Access Solutions
www.tessares.net