mptcp: allow picking different xmit subflows

Update the scheduler to less trivial heuristic: cache
the last used subflow, and try to send on it a reasonably
long burst of data.

When the burst or the subflow send space is exhausted, pick
the subflow with the lower ratio between write space and
send buffer - that is, the subflow with the greater relative
amount of free space.

v1 -> v2:
 - fix 32 bit build breakage due to 64bits div
 - fix checkpath issues (uint64_t -> u64)

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 95 insertions, 16 deletions

diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index ec9c38d3acc7..d7af96a900c4 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -1031,41 +1031,105 @@ static void mptcp_nospace(struct mptcp_sock *msk)
 	}
 }
 
+static bool mptcp_subflow_active(struct mptcp_subflow_context *subflow)
+{
+	struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
+
+	/* can't send if JOIN hasn't completed yet (i.e. is usable for mptcp) */
+	if (subflow->request_join && !subflow->fully_established)
+		return false;
+
+	/* only send if our side has not closed yet */
+	return ((1 << ssk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT));
+}
+
+#define MPTCP_SEND_BURST_SIZE		((1 << 16) - \
+					 sizeof(struct tcphdr) - \
+					 MAX_TCP_OPTION_SPACE - \
+					 sizeof(struct ipv6hdr) - \
+					 sizeof(struct frag_hdr))
+
+struct subflow_send_info {
+	struct sock *ssk;
+	u64 ratio;
+};
+
 static struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk,
 					   u32 *sndbuf)
 {
+	struct subflow_send_info send_info[2];
 	struct mptcp_subflow_context *subflow;
-	struct sock *sk = (struct sock *)msk;
-	struct sock *backup = NULL;
-	bool free;
+	int i, nr_active = 0;
+	struct sock *ssk;
+	u64 ratio;
+	u32 pace;
 
-	sock_owned_by_me(sk);
+	sock_owned_by_me((struct sock *)msk);
 
 	*sndbuf = 0;
 	if (!mptcp_ext_cache_refill(msk))
 		return NULL;
 
-	mptcp_for_each_subflow(msk, subflow) {
-		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
-
-		free = sk_stream_is_writeable(subflow->tcp_sock);
-		if (!free) {
-			mptcp_nospace(msk);
+	if (__mptcp_check_fallback(msk)) {
+		if (!msk->first)
 			return NULL;
+		*sndbuf = msk->first->sk_sndbuf;
+		return sk_stream_memory_free(msk->first) ? msk->first : NULL;
+	}
+
+	/* 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_for_each_subflow(msk, subflow) {
+			ssk =  mptcp_subflow_tcp_sock(subflow);
+			*sndbuf = max(tcp_sk(ssk)->snd_wnd, *sndbuf);
 		}
+		return msk->last_snd;
+	}
+
+	/* pick the subflow with the lower wmem/wspace ratio */
+	for (i = 0; i < 2; ++i) {
+		send_info[i].ssk = NULL;
+		send_info[i].ratio = -1;
+	}
+	mptcp_for_each_subflow(msk, subflow) {
+		ssk =  mptcp_subflow_tcp_sock(subflow);
+		if (!mptcp_subflow_active(subflow))
+			continue;
 
+		nr_active += !subflow->backup;
 		*sndbuf = max(tcp_sk(ssk)->snd_wnd, *sndbuf);
-		if (subflow->backup) {
-			if (!backup)
-				backup = ssk;
+		if (!sk_stream_memory_free(subflow->tcp_sock))
+			continue;
 
+		pace = READ_ONCE(ssk->sk_pacing_rate);
+		if (!pace)
 			continue;
-		}
 
-		return ssk;
+		ratio = div_u64((u64)READ_ONCE(ssk->sk_wmem_queued) << 32,
+				pace);
+		if (ratio < send_info[subflow->backup].ratio) {
+			send_info[subflow->backup].ssk = ssk;
+			send_info[subflow->backup].ratio = ratio;
+		}
 	}
 
-	return backup;
+	pr_debug("msk=%p nr_active=%d ssk=%p:%lld backup=%p:%lld",
+		 msk, nr_active, send_info[0].ssk, send_info[0].ratio,
+		 send_info[1].ssk, send_info[1].ratio);
+
+	/* pick the best backup if no other subflow is active */
+	if (!nr_active)
+		send_info[0].ssk = send_info[1].ssk;
+
+	if (send_info[0].ssk) {
+		msk->last_snd = send_info[0].ssk;
+		msk->snd_burst = min_t(int, MPTCP_SEND_BURST_SIZE,
+				       sk_stream_wspace(msk->last_snd));
+		return msk->last_snd;
+	}
+	return NULL;
 }
 
 static void ssk_check_wmem(struct mptcp_sock *msk)
@@ -1160,6 +1224,10 @@ restart:
 			break;
 		}
 
+		/* burst can be negative, we will try move to the next subflow
+		 * at selection time, if possible.
+		 */
+		msk->snd_burst -= ret;
 		copied += ret;
 
 		tx_ok = msg_data_left(msg);
@@ -1375,6 +1443,11 @@ static bool __mptcp_move_skbs(struct mptcp_sock *msk)
 	unsigned int moved = 0;
 	bool done;
 
+	/* avoid looping forever below on racing close */
+	if (((struct sock *)msk)->sk_state == TCP_CLOSE)
+		return false;
+
+	__mptcp_flush_join_list(msk);
 	do {
 		struct sock *ssk = mptcp_subflow_recv_lookup(msk);
 
@@ -1539,9 +1612,15 @@ static struct sock *mptcp_subflow_get_retrans(const struct mptcp_sock *msk)
 
 	sock_owned_by_me((const struct sock *)msk);
 
+	if (__mptcp_check_fallback(msk))
+		return msk->first;
+
 	mptcp_for_each_subflow(msk, subflow) {
 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 
+		if (!mptcp_subflow_active(subflow))
+			continue;
+
 		/* still data outstanding at TCP level?  Don't retransmit. */
 		if (!tcp_write_queue_empty(ssk))
 			return NULL;