Merge branch 'mptcp-autotune-related-improvement'

Matthieu Baerts says:

====================
mptcp: autotune related improvement

Here are two patches from Paolo that have been crafted a couple of
months ago, but needed more validation because they were indirectly
causing instabilities in the sefltests. The root cause has been fixed in
'net' recently in commit 8c09412e584d ("selftests: mptcp: more stable
simult_flows tests").

These patches refactor the receive space and RTT estimator, overall
making DRS more correct while avoiding receive buffer drifting to
tcp_rmem[2], which in turn makes the throughput more stable and less
bursty, especially with high bandwidth and low delay environments.

Note that the first patch addresses a very old issue. 'net-next' is
targeted because the change is quite invasive and based on a recent
backlog refactor. The 'Fixes' tag is then there more as a FYI, because
backporting this patch will quickly be blocked due to large conflicts.
====================

Link: https://patch.msgid.link/20260407-net-next-mptcp-reduce-rbuf-v2-0-0d1d135bf6f6@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

3 files changed, 77 insertions, 33 deletions

diff --git a/include/trace/events/mptcp.h b/include/trace/events/mptcp.h
index 269d949b2025..04521acba483 100644
--- a/include/trace/events/mptcp.h
+++ b/include/trace/events/mptcp.h
@@ -219,7 +219,7 @@ TRACE_EVENT(mptcp_rcvbuf_grow,
 		__be32 *p32;
 
 		__entry->time = time;
-		__entry->rtt_us = msk->rcvq_space.rtt_us >> 3;
+		__entry->rtt_us = mptcp_rtt_us_est(msk) >> 3;
 		__entry->copied = msk->rcvq_space.copied;
 		__entry->inq = mptcp_inq_hint(sk);
 		__entry->space = msk->rcvq_space.space;
diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index 2f4776a4f06a..cf5747595259 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -879,6 +879,32 @@ static bool move_skbs_to_msk(struct mptcp_sock *msk, struct sock *ssk)
 	return moved;
 }
 
+static void mptcp_rcv_rtt_update(struct mptcp_sock *msk,
+				 struct mptcp_subflow_context *subflow)
+{
+	const struct tcp_sock *tp = tcp_sk(subflow->tcp_sock);
+	u32 rtt_us = tp->rcv_rtt_est.rtt_us;
+	int id;
+
+	/* Update once per subflow per rcvwnd to avoid touching the msk
+	 * too often.
+	 */
+	if (!rtt_us || tp->rcv_rtt_est.seq == subflow->prev_rtt_seq)
+		return;
+
+	subflow->prev_rtt_seq = tp->rcv_rtt_est.seq;
+
+	/* Pairs with READ_ONCE() in mptcp_rtt_us_est(). */
+	id = msk->rcv_rtt_est.next_sample;
+	WRITE_ONCE(msk->rcv_rtt_est.samples[id], rtt_us);
+	if (++msk->rcv_rtt_est.next_sample == MPTCP_RTT_SAMPLES)
+		msk->rcv_rtt_est.next_sample = 0;
+
+	/* EWMA among the incoming subflows */
+	msk->scaling_ratio = ((msk->scaling_ratio << 3) - msk->scaling_ratio +
+			     tp->scaling_ratio) >> 3;
+}
+
 void mptcp_data_ready(struct sock *sk, struct sock *ssk)
 {
 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
@@ -892,6 +918,7 @@ void mptcp_data_ready(struct sock *sk, struct sock *ssk)
 		return;
 
 	mptcp_data_lock(sk);
+	mptcp_rcv_rtt_update(msk, subflow);
 	if (!sock_owned_by_user(sk)) {
 		/* Wake-up the reader only for in-sequence data */
 		if (move_skbs_to_msk(msk, ssk) && mptcp_epollin_ready(sk))
@@ -2095,7 +2122,6 @@ static void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
 
 	msk->rcvspace_init = 1;
 	msk->rcvq_space.copied = 0;
-	msk->rcvq_space.rtt_us = 0;
 
 	/* initial rcv_space offering made to peer */
 	msk->rcvq_space.space = min_t(u32, tp->rcv_wnd,
@@ -2106,15 +2132,15 @@ static void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
 
 /* receive buffer autotuning.  See tcp_rcv_space_adjust for more information.
  *
- * Only difference: Use highest rtt estimate of the subflows in use.
+ * Only difference: Use lowest rtt estimate of the subflows in use, see
+ * mptcp_rcv_rtt_update() and mptcp_rtt_us_est().
  */
 static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
 {
 	struct mptcp_subflow_context *subflow;
 	struct sock *sk = (struct sock *)msk;
-	u8 scaling_ratio = U8_MAX;
-	u32 time, advmss = 1;
-	u64 rtt_us, mstamp;
+	u32 time, rtt_us;
+	u64 mstamp;
 
 	msk_owned_by_me(msk);
 
@@ -2129,36 +2155,17 @@ static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
 	mstamp = mptcp_stamp();
 	time = tcp_stamp_us_delta(mstamp, READ_ONCE(msk->rcvq_space.time));
 
-	rtt_us = msk->rcvq_space.rtt_us;
-	if (rtt_us && time < (rtt_us >> 3))
-		return;
-
-	rtt_us = 0;
-	mptcp_for_each_subflow(msk, subflow) {
-		const struct tcp_sock *tp;
-		u64 sf_rtt_us;
-		u32 sf_advmss;
-
-		tp = tcp_sk(mptcp_subflow_tcp_sock(subflow));
-
-		sf_rtt_us = READ_ONCE(tp->rcv_rtt_est.rtt_us);
-		sf_advmss = READ_ONCE(tp->advmss);
-
-		rtt_us = max(sf_rtt_us, rtt_us);
-		advmss = max(sf_advmss, advmss);
-		scaling_ratio = min(tp->scaling_ratio, scaling_ratio);
-	}
-
-	msk->rcvq_space.rtt_us = rtt_us;
-	msk->scaling_ratio = scaling_ratio;
-	if (time < (rtt_us >> 3) || rtt_us == 0)
+	rtt_us = mptcp_rtt_us_est(msk);
+	if (rtt_us == U32_MAX || time < (rtt_us >> 3))
 		return;
 
-	if (msk->rcvq_space.copied <= msk->rcvq_space.space)
+	copied = msk->rcvq_space.copied;
+	copied -= mptcp_inq_hint(sk);
+	if (copied <= msk->rcvq_space.space)
 		goto new_measure;
 
 	trace_mptcp_rcvbuf_grow(sk, time);
-	if (mptcp_rcvbuf_grow(sk, msk->rcvq_space.copied)) {
+	if (mptcp_rcvbuf_grow(sk, copied)) {
 		/* Make subflows follow along.  If we do not do this, we
 		 * get drops at subflow level if skbs can't be moved to
 		 * the mptcp rx queue fast enough (announced rcv_win can
@@ -2172,7 +2179,7 @@ static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
 			slow = lock_sock_fast(ssk);
 			/* subflows can be added before tcp_init_transfer() */
 			if (tcp_sk(ssk)->rcvq_space.space)
-				tcp_rcvbuf_grow(ssk, msk->rcvq_space.copied);
+				tcp_rcvbuf_grow(ssk, copied);
 			unlock_sock_fast(ssk, slow);
 		}
 	}
@@ -3015,6 +3022,7 @@ static void __mptcp_init_sock(struct sock *sk)
 	msk->timer_ival = TCP_RTO_MIN;
 	msk->scaling_ratio = TCP_DEFAULT_SCALING_RATIO;
 	msk->backlog_len = 0;
+	mptcp_init_rtt_est(msk);
 
 	WRITE_ONCE(msk->first, NULL);
 	inet_csk(sk)->icsk_sync_mss = mptcp_sync_mss;
@@ -3460,6 +3468,7 @@ static int mptcp_disconnect(struct sock *sk, int flags)
 	msk->bytes_retrans = 0;
 	msk->rcvspace_init = 0;
 	msk->fastclosing = 0;
+	mptcp_init_rtt_est(msk);
 
 	/* for fallback's sake */
 	WRITE_ONCE(msk->ack_seq, 0);
diff --git a/net/mptcp/protocol.h b/net/mptcp/protocol.h
index 1208f317ac33..e1d4783db02f 100644
--- a/net/mptcp/protocol.h
+++ b/net/mptcp/protocol.h
@@ -269,6 +269,13 @@ struct mptcp_data_frag {
 	struct page *page;
 };
 
+/* Arbitrary compromise between as low as possible to react timely to subflow
+ * close event and as big as possible to avoid being fouled by biased large
+ * samples due to peer sending data on a different subflow WRT to the incoming
+ * ack.
+ */
+#define MPTCP_RTT_SAMPLES	5
+
 /* MPTCP connection sock */
 struct mptcp_sock {
 	/* inet_connection_sock must be the first member */
@@ -341,11 +348,17 @@ struct mptcp_sock {
 				 */
 	struct mptcp_pm_data	pm;
 	struct mptcp_sched_ops	*sched;
+
+	/* Most recent rtt_us observed by in use incoming subflows. */
+	struct {
+		u32	samples[MPTCP_RTT_SAMPLES];
+		u32	next_sample;
+	} rcv_rtt_est;
+
 	struct {
 		int	space;	/* bytes copied in last measurement window */
 		int	copied; /* bytes copied in this measurement window */
 		u64	time;	/* start time of measurement window */
-		u64	rtt_us; /* last maximum rtt of subflows */
 	} rcvq_space;
 	u8		scaling_ratio;
 	bool		allow_subflows;
@@ -423,6 +436,27 @@ static inline struct mptcp_data_frag *mptcp_send_head(const struct sock *sk)
 	return msk->first_pending;
 }
 
+static inline void mptcp_init_rtt_est(struct mptcp_sock *msk)
+{
+	int i;
+
+	for (i = 0; i < MPTCP_RTT_SAMPLES; ++i)
+		msk->rcv_rtt_est.samples[i] = U32_MAX;
+	msk->rcv_rtt_est.next_sample = 0;
+	msk->scaling_ratio = TCP_DEFAULT_SCALING_RATIO;
+}
+
+static inline u32 mptcp_rtt_us_est(const struct mptcp_sock *msk)
+{
+	u32 rtt_us = READ_ONCE(msk->rcv_rtt_est.samples[0]);
+	int i;
+
+	/* Lockless access of collected samples. */
+	for (i = 1; i < MPTCP_RTT_SAMPLES; ++i)
+		rtt_us = min(rtt_us, READ_ONCE(msk->rcv_rtt_est.samples[i]));
+	return rtt_us;
+}
+
 static inline struct mptcp_data_frag *mptcp_send_next(struct sock *sk)
 {
 	struct mptcp_sock *msk = mptcp_sk(sk);
@@ -524,6 +558,7 @@ struct mptcp_subflow_context {
 	u32	map_data_len;
 	__wsum	map_data_csum;
 	u32	map_csum_len;
+	u32	prev_rtt_seq;
 	u32	request_mptcp : 1,  /* send MP_CAPABLE */
 		request_join : 1,   /* send MP_JOIN */
 		request_bkup : 1,