X-Git-Url: http://git.meshlink.io/?a=blobdiff_plain;f=utcp.c;h=cb23e1bf3d6413510d8dd729f7dd00e781eb4291;hb=8b12570493ab77960aeac7e3e4a9bc458bb4f1a1;hp=7e28f9e27e3aef2a7f0937597eabbd5bd3c36f7a;hpb=b92868c9f1e00601c1cb5cccc6dc01fdb79de4ca;p=utcp

diff --git a/utcp.c b/utcp.c
index 7e28f9e..cb23e1b 100644
--- a/utcp.c
+++ b/utcp.c
@@ -54,6 +54,10 @@
 	} while (0)
 #endif
 
+static inline size_t min(size_t a, size_t b) {
+	return a < b ? a : b;
+}
+
 static inline size_t max(size_t a, size_t b) {
 	return a > b ? a : b;
 }
@@ -113,9 +117,14 @@ static void print_packet(struct utcp *utcp, const char *dir, const void *pkt, si
 
 	debug("\n");
 }
+
+static void debug_cwnd(struct utcp_connection *c) {
+	debug("snd.cwnd = %u\n", c->snd.cwnd);
+}
 #else
 #define debug(...) do {} while(0)
 #define print_packet(...) do {} while(0)
+#define debug_cwnd(...) do {} while(0)
 #endif
 
 static void set_state(struct utcp_connection *c, enum state state) {
@@ -382,9 +391,10 @@ static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t s
 #endif
 	c->snd.una = c->snd.iss;
 	c->snd.nxt = c->snd.iss + 1;
-	c->rcv.wnd = utcp->mtu;
 	c->snd.last = c->snd.nxt;
-	c->snd.cwnd = utcp->mtu;
+	c->snd.cwnd = (utcp->mtu > 2190 ? 2 : utcp->mtu > 1095 ? 3 : 4) * utcp->mtu;
+	c->snd.ssthresh = ~0;
+	debug_cwnd(c);
 	c->utcp = utcp;
 
 	// Add it to the sorted list of connections
@@ -415,13 +425,13 @@ static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
 	if(!utcp->srtt) {
 		utcp->srtt = rtt;
 		utcp->rttvar = rtt / 2;
-		utcp->rto = rtt + max(2 * rtt, CLOCK_GRANULARITY);
 	} else {
 		utcp->rttvar = (utcp->rttvar * 3 + absdiff(utcp->srtt, rtt)) / 4;
 		utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
-		utcp->rto = utcp->srtt + max(utcp->rttvar, CLOCK_GRANULARITY);
 	}
 
+	utcp->rto = utcp->srtt + max(4 * utcp->rttvar, CLOCK_GRANULARITY);
+
 	if(utcp->rto > MAX_RTO) {
 		utcp->rto = MAX_RTO;
 	}
@@ -468,7 +478,7 @@ struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_re
 	pkt.hdr.dst = c->dst;
 	pkt.hdr.seq = c->snd.iss;
 	pkt.hdr.ack = 0;
-	pkt.hdr.wnd = c->rcv.wnd;
+	pkt.hdr.wnd = c->rcvbuf.maxsize;
 	pkt.hdr.ctl = SYN;
 	pkt.hdr.aux = 0x0101;
 	pkt.init[0] = 1;
@@ -507,19 +517,22 @@ void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
 
 static void ack(struct utcp_connection *c, bool sendatleastone) {
 	int32_t left = seqdiff(c->snd.last, c->snd.nxt);
-	int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
-	debug("cwndleft = %d\n", cwndleft);
+	int32_t cwndleft = min(c->snd.cwnd, c->snd.wnd) - seqdiff(c->snd.nxt, c->snd.una);
 
 	assert(left >= 0);
 
 	if(cwndleft <= 0) {
-		cwndleft = 0;
-	}
-
-	if(cwndleft < left) {
+		left = 0;
+	} else if(cwndleft < left) {
 		left = cwndleft;
+
+		if(!sendatleastone || cwndleft > c->utcp->mtu) {
+			left -= left % c->utcp->mtu;
+		}
 	}
 
+	debug("cwndleft = %d, left = %d\n", cwndleft, left);
+
 	if(!left && !sendatleastone) {
 		return;
 	}
@@ -538,7 +551,7 @@ static void ack(struct utcp_connection *c, bool sendatleastone) {
 	pkt->hdr.src = c->src;
 	pkt->hdr.dst = c->dst;
 	pkt->hdr.ack = c->rcv.nxt;
-	pkt->hdr.wnd = c->snd.wnd;
+	pkt->hdr.wnd = c->rcvbuf.maxsize;
 	pkt->hdr.ctl = ACK;
 	pkt->hdr.aux = 0;
 
@@ -629,6 +642,8 @@ ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 
 	if(is_reliable(c) || (c->state != SYN_SENT && c->state != SYN_RECEIVED)) {
 		len = buffer_put(&c->sndbuf, data, len);
+	} else {
+		return 0;
 	}
 
 	if(len <= 0) {
@@ -695,7 +710,7 @@ static void retransmit(struct utcp_connection *c) {
 
 	pkt->hdr.src = c->src;
 	pkt->hdr.dst = c->dst;
-	pkt->hdr.wnd = c->rcv.wnd;
+	pkt->hdr.wnd = c->rcvbuf.maxsize;
 	pkt->hdr.aux = 0;
 
 	switch(c->state) {
@@ -743,7 +758,12 @@ static void retransmit(struct utcp_connection *c) {
 		}
 
 		c->snd.nxt = c->snd.una + len;
-		c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
+
+		// RFC 5681 slow start after timeout
+		c->snd.ssthresh = max(c->snd.cwnd / 2, utcp->mtu * 2); // eq. 4
+		c->snd.cwnd = utcp->mtu;
+		debug_cwnd(c);
+
 		buffer_copy(&c->sndbuf, pkt->data, 0, len);
 		print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + len);
 		utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
@@ -1011,6 +1031,8 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 		ptr += 2;
 	}
 
+	bool has_data = len || (hdr.ctl & (SYN | FIN));
+
 	// Try to match the packet to an existing connection
 
 	struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
@@ -1053,6 +1075,7 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 				c->flags = UTCP_TCP;
 			}
 
+synack:
 			// Return SYN+ACK, go to SYN_RECEIVED state
 			c->snd.wnd = hdr.wnd;
 			c->rcv.irs = hdr.seq;
@@ -1068,7 +1091,7 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 			pkt.hdr.dst = c->dst;
 			pkt.hdr.ack = c->rcv.irs + 1;
 			pkt.hdr.seq = c->snd.iss;
-			pkt.hdr.wnd = c->rcv.wnd;
+			pkt.hdr.wnd = c->rcvbuf.maxsize;
 			pkt.hdr.ctl = SYN | ACK;
 
 			if(init) {
@@ -1105,8 +1128,6 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 
 	// It is for an existing connection.
 
-	uint32_t prevrcvnxt = c->rcv.nxt;
-
 	// 1. Drop invalid packets.
 
 	// 1a. Drop packets that should not happen in our current state.
@@ -1130,7 +1151,7 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 		break;
 	}
 
-	// 1b. Drop packets with a sequence number not in our receive window.
+	// 1b. Discard data that is not in our receive window.
 
 	if(is_reliable(c)) {
 		bool acceptable;
@@ -1282,7 +1303,6 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 	// 3. Advance snd.una
 
 	advanced = seqdiff(hdr.ack, c->snd.una);
-	prevrcvnxt = c->rcv.nxt;
 
 	if(advanced) {
 		// RTT measurement
@@ -1314,8 +1334,10 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 
 		assert(data_acked >= 0);
 
+#ifndef NDEBUG
 		int32_t bufused = seqdiff(c->snd.last, c->snd.una);
 		assert(data_acked <= bufused);
+#endif
 
 		if(data_acked) {
 			buffer_get(&c->sndbuf, NULL, data_acked);
@@ -1329,12 +1351,20 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 		c->snd.una = hdr.ack;
 
 		c->dupack = 0;
-		c->snd.cwnd += utcp->mtu;
+
+		// Increase the congestion window according to RFC 5681
+		if(c->snd.cwnd < c->snd.ssthresh) {
+			c->snd.cwnd += min(advanced, utcp->mtu); // eq. 2
+		} else {
+			c->snd.cwnd += max(1, (utcp->mtu * utcp->mtu) / c->snd.cwnd); // eq. 3
+		}
 
 		if(c->snd.cwnd > c->sndbuf.maxsize) {
 			c->snd.cwnd = c->sndbuf.maxsize;
 		}
 
+		debug_cwnd(c);
+
 		// Check if we have sent a FIN that is now ACKed.
 		switch(c->state) {
 		case FIN_WAIT_1:
@@ -1347,7 +1377,7 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 		case CLOSING:
 			if(c->snd.una == c->snd.last) {
 				gettimeofday(&c->conn_timeout, NULL);
-				c->conn_timeout.tv_sec += 60;
+				c->conn_timeout.tv_sec += utcp->timeout;
 				set_state(c, TIME_WAIT);
 			}
 
@@ -1367,6 +1397,7 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 				//Reset the congestion window so we wait for ACKs.
 				c->snd.nxt = c->snd.una;
 				c->snd.cwnd = utcp->mtu;
+				debug_cwnd(c);
 				start_retransmit_timer(c);
 			}
 		}
@@ -1411,6 +1442,9 @@ skip_ack:
 			break;
 
 		case SYN_RECEIVED:
+			// This is a retransmit of a SYN, send back the SYNACK.
+			goto synack;
+
 		case ESTABLISHED:
 		case FIN_WAIT_1:
 		case FIN_WAIT_2:
@@ -1506,7 +1540,7 @@ skip_ack:
 
 		case FIN_WAIT_2:
 			gettimeofday(&c->conn_timeout, NULL);
-			c->conn_timeout.tv_sec += 60;
+			c->conn_timeout.tv_sec += utcp->timeout;
 			set_state(c, TIME_WAIT);
 			break;
 
@@ -1536,13 +1570,13 @@ skip_ack:
 	}
 
 	// Now we send something back if:
-	// - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
+	// - we received data, so we have to send back an ACK
 	//   -> sendatleastone = true
 	// - or we got an ack, so we should maybe send a bit more data
 	//   -> sendatleastone = false
 
 	if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
-		ack(c, len || prevrcvnxt != c->rcv.nxt);
+		ack(c, has_data);
 	}
 
 	return 0;
@@ -1919,8 +1953,14 @@ void utcp_reset_timers(struct utcp *utcp) {
 			continue;
 		}
 
-		c->rtrx_timeout = now;
-		c->conn_timeout = then;
+		if(timerisset(&c->rtrx_timeout)) {
+			c->rtrx_timeout = now;
+		}
+
+		if(timerisset(&c->conn_timeout)) {
+			c->conn_timeout = then;
+		}
+
 		c->rtt_start.tv_sec = 0;
 	}
 
@@ -2071,19 +2111,24 @@ void utcp_expect_data(struct utcp_connection *c, bool expect) {
 }
 
 void utcp_offline(struct utcp *utcp, bool offline) {
+	struct timeval now;
+	gettimeofday(&now, NULL);
+
 	for(int i = 0; i < utcp->nconnections; i++) {
 		struct utcp_connection *c = utcp->connections[i];
 
-		if(!c->reapable) {
-			utcp_expect_data(c, offline);
-
-			// If we are online again, reset the retransmission timers, but keep the connection timeout as it is,
-			// to prevent peers toggling online/offline state frequently from keeping connections alive
-			// if there is no progress in sending actual data.
-			if(!offline) {
-				gettimeofday(&utcp->connections[i]->rtrx_timeout, NULL);
-				utcp->connections[i]->rtt_start.tv_sec = 0;
+		if(c->reapable) {
+			continue;
+		}
+
+		utcp_expect_data(c, offline);
+
+		if(!offline) {
+			if(timerisset(&c->rtrx_timeout)) {
+				c->rtrx_timeout = now;
 			}
+
+			utcp->connections[i]->rtt_start.tv_sec = 0;
 		}
 	}