X-Git-Url: http://git.meshlink.io/?p=utcp;a=blobdiff_plain;f=utcp.c;h=c650ab700ad150fa876b0944a6872ea39298f352;hp=f062319fa33a1daa665faaa96f3c567eb488f245;hb=5ded633782709710c42648c71154d6062e83b7a6;hpb=2ccd5cc7da5f6157beca8cfe9dbf115e95a99776

diff --git a/utcp.c b/utcp.c
index f062319..c650ab7 100644
--- a/utcp.c
+++ b/utcp.c
@@ -49,10 +49,14 @@
 	(r)->tv_sec = (a)->tv_sec - (b)->tv_sec;\
 	(r)->tv_usec = (a)->tv_usec - (b)->tv_usec;\
 	if((r)->tv_usec < 0)\
-		(r)->tv_sec--, (r)->tv_usec += 1000000;\
+		(r)->tv_sec--, (r)->tv_usec += USEC_PER_SEC;\
 } while (0)
 #endif
 
+static inline size_t max(size_t a, size_t b) {
+	return a > b ? a : b;
+}
+
 #ifdef UTCP_DEBUG
 #include <stdarg.h>
 
@@ -66,12 +70,12 @@ static void debug(const char *format, ...) {
 static void print_packet(struct utcp *utcp, const char *dir, const void *pkt, size_t len) {
 	struct hdr hdr;
 	if(len < sizeof hdr) {
-		debug("%p %s: short packet (%zu bytes)\n", utcp, dir, len);
+		debug("%p %s: short packet (%lu bytes)\n", utcp, dir, (unsigned long)len);
 		return;
 	}
 
 	memcpy(&hdr, pkt, sizeof hdr);
-	fprintf (stderr, "%p %s: len=%zu, src=%u dst=%u seq=%u ack=%u wnd=%u ctl=", utcp, dir, len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd);
+	debug("%p %s: len=%lu, src=%u dst=%u seq=%u ack=%u wnd=%u ctl=", utcp, dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd);
 	if(hdr.ctl & SYN)
 		debug("SYN");
 	if(hdr.ctl & RST)
@@ -82,11 +86,18 @@ static void print_packet(struct utcp *utcp, const char *dir, const void *pkt, si
 		debug("ACK");
 
 	if(len > sizeof hdr) {
-		debug(" data=");
-		for(int i = sizeof hdr; i < len; i++) {
-			const char *data = pkt;
-			debug("%c", data[i] >= 32 ? data[i] : '.');
+		uint32_t datalen = len - sizeof hdr;
+		const uint8_t *data = (uint8_t *)pkt + sizeof hdr;
+		char str[datalen * 2 + 1];
+		char *p = str;
+
+		for(uint32_t i = 0; i < datalen; i++) {
+			*p++ = "0123456789ABCDEF"[data[i] >> 4];
+			*p++ = "0123456789ABCDEF"[data[i] & 15];
 		}
+		*p = 0;
+
+		debug(" data=%s", str);
 	}
 
 	debug("\n");
@@ -116,6 +127,10 @@ static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
 	}
 }
 
+static bool is_reliable(struct utcp_connection *c) {
+	return c->flags & UTCP_RELIABLE;
+}
+
 static inline void list_connections(struct utcp *utcp) {
 	debug("%p has %d connections:\n", utcp, utcp->nconnections);
 	for(int i = 0; i < utcp->nconnections; i++)
@@ -130,16 +145,26 @@ static int32_t seqdiff(uint32_t a, uint32_t b) {
 // TODO: convert to ringbuffers to avoid memmove() operations.
 
 // Store data into the buffer
-static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
-	if(buf->maxsize <= buf->used)
-		return 0;
-	if(len > buf->maxsize - buf->used)
-		len = buf->maxsize - buf->used;
-	if(len > buf->size - buf->used) {
+static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
+	debug("buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
+
+	size_t required = offset + len;
+	if(required > buf->maxsize) {
+		if(offset >= buf->maxsize)
+			return 0;
+		len = buf->maxsize - offset;
+		required = buf->maxsize;
+	}
+
+	if(required > buf->size) {
 		size_t newsize = buf->size;
-		do {
-			newsize *= 2;
-		} while(newsize < buf->used + len);
+		if(!newsize) {
+			newsize = required;
+		} else {
+			do {
+				newsize *= 2;
+			} while(newsize < required);
+		}
 		if(newsize > buf->maxsize)
 			newsize = buf->maxsize;
 		char *newdata = realloc(buf->data, newsize);
@@ -148,11 +173,17 @@ static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
 		buf->data = newdata;
 		buf->size = newsize;
 	}
-	memcpy(buf->data + buf->used, data, len);
-	buf->used += len;
+
+	memcpy(buf->data + offset, data, len);
+	if(required > buf->used)
+		buf->used = required;
 	return len;
 }
 
+static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
+	return buffer_put_at(buf, buf->used, data, len);
+}
+
 // Get data from the buffer. data can be NULL.
 static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
 	if(len > buf->used)
@@ -177,9 +208,11 @@ static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t
 
 static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
 	memset(buf, 0, sizeof *buf);
-	buf->data = malloc(len);
-	if(!len)
-		return false;
+	if(len) {
+		buf->data = malloc(len);
+		if(!buf->data)
+			return false;
+	}
 	buf->size = len;
 	buf->maxsize = maxlen;
 	return true;
@@ -234,6 +267,7 @@ static void free_connection(struct utcp_connection *c) {
 	memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof *cp);
 	utcp->nconnections--;
 
+	buffer_exit(&c->rcvbuf);
 	buffer_exit(&c->sndbuf);
 	free(c);
 }
@@ -278,11 +312,21 @@ static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t s
 		return NULL;
 	}
 
+	if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
+		buffer_exit(&c->sndbuf);
+		free(c);
+		return NULL;
+	}
+
 	// Fill in the details
 
 	c->src = src;
 	c->dst = dst;
+#ifdef UTCP_DEBUG
+	c->snd.iss = 0;
+#else
 	c->snd.iss = rand();
+#endif
 	c->snd.una = c->snd.iss;
 	c->snd.nxt = c->snd.iss + 1;
 	c->rcv.wnd = utcp->mtu;
@@ -298,28 +342,78 @@ static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t s
 	return c;
 }
 
-struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
+// Update RTT variables. See RFC 6298.
+static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
+	if(!rtt) {
+		debug("invalid rtt\n");
+		return;
+	}
+
+	struct utcp *utcp = c->utcp;
+
+	if(!utcp->srtt) {
+		utcp->srtt = rtt;
+		utcp->rttvar = rtt / 2;
+		utcp->rto = rtt + max(2 * rtt, CLOCK_GRANULARITY);
+	} else {
+		utcp->rttvar = (utcp->rttvar * 3 + abs(utcp->srtt - rtt)) / 4;
+		utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
+		utcp->rto = utcp->srtt + max(utcp->rttvar, CLOCK_GRANULARITY);
+	}
+
+	if(utcp->rto > MAX_RTO)
+		utcp->rto = MAX_RTO;
+
+	debug("rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
+}
+
+static void start_retransmit_timer(struct utcp_connection *c) {
+	gettimeofday(&c->rtrx_timeout, NULL);
+	c->rtrx_timeout.tv_usec += c->utcp->rto;
+	while(c->rtrx_timeout.tv_usec >= 1000000) {
+		c->rtrx_timeout.tv_usec -= 1000000;
+		c->rtrx_timeout.tv_sec++;
+	}
+	debug("timeout set to %lu.%06lu (%u)\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_usec, c->utcp->rto);
+}
+
+static void stop_retransmit_timer(struct utcp_connection *c) {
+	timerclear(&c->rtrx_timeout);
+	debug("timeout cleared\n");
+}
+
+struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
 	struct utcp_connection *c = allocate_connection(utcp, 0, dst);
 	if(!c)
 		return NULL;
 
+	assert((flags & ~0xf) == 0);
+
+	c->flags = flags;
 	c->recv = recv;
 	c->priv = priv;
 
-	struct hdr hdr;
-
-	hdr.src = c->src;
-	hdr.dst = c->dst;
-	hdr.seq = c->snd.iss;
-	hdr.ack = 0;
-	hdr.wnd = c->rcv.wnd;
-	hdr.ctl = SYN;
-	hdr.aux = 0;
+	struct {
+		struct hdr hdr;
+		uint8_t init[4];
+	} pkt;
+
+	pkt.hdr.src = c->src;
+	pkt.hdr.dst = c->dst;
+	pkt.hdr.seq = c->snd.iss;
+	pkt.hdr.ack = 0;
+	pkt.hdr.wnd = c->rcv.wnd;
+	pkt.hdr.ctl = SYN;
+	pkt.hdr.aux = 0x0101;
+	pkt.init[0] = 1;
+	pkt.init[1] = 0;
+	pkt.init[2] = 0;
+	pkt.init[3] = flags & 0x7;
 
 	set_state(c, SYN_SENT);
 
-	print_packet(utcp, "send", &hdr, sizeof hdr);
-	utcp->send(utcp, &hdr, sizeof hdr);
+	print_packet(utcp, "send", &pkt, sizeof pkt);
+	utcp->send(utcp, &pkt, sizeof pkt);
 
 	gettimeofday(&c->conn_timeout, NULL);
 	c->conn_timeout.tv_sec += utcp->timeout;
@@ -327,6 +421,10 @@ struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_
 	return c;
 }
 
+struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
+	return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
+}
+
 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
 	if(c->reapable || c->state != SYN_RECEIVED) {
 		debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
@@ -342,6 +440,7 @@ 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);
 
 	assert(left >= 0);
 
@@ -360,7 +459,7 @@ static void ack(struct utcp_connection *c, bool sendatleastone) {
 	} *pkt;
 
 	pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
-	if(!pkt->data)
+	if(!pkt)
 		return;
 
 	pkt->hdr.src = c->src;
@@ -384,6 +483,13 @@ static void ack(struct utcp_connection *c, bool sendatleastone) {
 			pkt->hdr.ctl |= FIN;
 		}
 
+		if(!c->rtt_start.tv_sec) {
+			// Start RTT measurement
+			gettimeofday(&c->rtt_start, NULL);
+			c->rtt_seq = pkt->hdr.seq + seglen;
+			debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
+		}
+
 		print_packet(c->utcp, "send", pkt, sizeof pkt->hdr + seglen);
 		c->utcp->send(c->utcp, pkt, sizeof pkt->hdr + seglen);
 	} while(left);
@@ -419,7 +525,7 @@ ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 		return -1;
 	}
 
-	// Add data to send buffer
+	// Exit early if we have nothing to send.
 
 	if(!len)
 		return 0;
@@ -429,6 +535,8 @@ ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 		return -1;
 	}
 
+	// Add data to send buffer.
+
 	len = buffer_put(&c->sndbuf, data, len);
 	if(len <= 0) {
 		errno = EWOULDBLOCK;
@@ -437,6 +545,12 @@ ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 
 	c->snd.last += len;
 	ack(c, false);
+	if(!is_reliable(c)) {
+		c->snd.una = c->snd.nxt = c->snd.last;
+		buffer_get(&c->sndbuf, NULL, c->sndbuf.used);
+	}
+	if(is_reliable(c) && !timerisset(&c->rtrx_timeout))
+		start_retransmit_timer(c);
 	return len;
 }
 
@@ -447,8 +561,11 @@ static void swap_ports(struct hdr *hdr) {
 }
 
 static void retransmit(struct utcp_connection *c) {
-	if(c->state == CLOSED || c->snd.nxt == c->snd.una)
+	if(c->state == CLOSED || c->snd.last == c->snd.una) {
+		debug("Retransmit() called but nothing to retransmit!\n");
+		stop_retransmit_timer(c);
 		return;
+	}
 
 	struct utcp *utcp = c->utcp;
 
@@ -463,22 +580,21 @@ 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.aux = 0;
 
 	switch(c->state) {
-		case LISTEN:
-			// TODO: this should not happen
-			break;
-
 		case SYN_SENT:
+			// Send our SYN again
 			pkt->hdr.seq = c->snd.iss;
 			pkt->hdr.ack = 0;
-			pkt->hdr.wnd = c->rcv.wnd;
 			pkt->hdr.ctl = SYN;
 			print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
 			utcp->send(utcp, pkt, sizeof pkt->hdr);
 			break;
 
 		case SYN_RECEIVED:
+			// Send SYNACK again
 			pkt->hdr.seq = c->snd.nxt;
 			pkt->hdr.ack = c->rcv.nxt;
 			pkt->hdr.ctl = SYN | ACK;
@@ -488,34 +604,185 @@ static void retransmit(struct utcp_connection *c) {
 
 		case ESTABLISHED:
 		case FIN_WAIT_1:
+		case CLOSE_WAIT:
+		case CLOSING:
+		case LAST_ACK:
+			// Send unacked data again.
 			pkt->hdr.seq = c->snd.una;
 			pkt->hdr.ack = c->rcv.nxt;
 			pkt->hdr.ctl = ACK;
-			uint32_t len = seqdiff(c->snd.nxt, c->snd.una);
-			if(c->state == FIN_WAIT_1)
-				len--;
+			uint32_t len = seqdiff(c->snd.last, c->snd.una);
 			if(len > utcp->mtu)
 				len = utcp->mtu;
-			else {
-				if(c->state == FIN_WAIT_1)
-					pkt->hdr.ctl |= FIN;
+			if(fin_wanted(c, c->snd.una + len)) {
+				len--;
+				pkt->hdr.ctl |= FIN;
 			}
+			c->snd.nxt = c->snd.una + len;
+			c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
 			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);
 			break;
 
-		default:
-			// TODO: implement
+		case CLOSED:
+		case LISTEN:
+		case TIME_WAIT:
+		case FIN_WAIT_2:
+			// We shouldn't need to retransmit anything in this state.
 #ifdef UTCP_DEBUG
 			abort();
 #endif
-			break;
+			stop_retransmit_timer(c);
+			goto cleanup;
 	}
 
+	start_retransmit_timer(c);
+	utcp->rto *= 2;
+	if(utcp->rto > MAX_RTO)
+		utcp->rto = MAX_RTO;
+	c->rtt_start.tv_sec = 0; // invalidate RTT timer
+
+cleanup:
 	free(pkt);
 }
 
+/* Update receive buffer and SACK entries after consuming data.
+ *
+ * Situation:
+ *
+ * |.....0000..1111111111.....22222......3333|
+ * |---------------^
+ *
+ * 0..3 represent the SACK entries. The ^ indicates up to which point we want
+ * to remove data from the receive buffer. The idea is to substract "len"
+ * from the offset of all the SACK entries, and then remove/cut down entries
+ * that are shifted to before the start of the receive buffer.
+ *
+ * There are three cases:
+ * - the SACK entry is after ^, in that case just change the offset.
+ * - the SACK entry starts before and ends after ^, so we have to
+ *   change both its offset and size.
+ * - the SACK entry is completely before ^, in that case delete it.
+ */
+static void sack_consume(struct utcp_connection *c, size_t len) {
+	debug("sack_consume %lu\n", (unsigned long)len);
+	if(len > c->rcvbuf.used) {
+		debug("All SACK entries consumed");
+		c->sacks[0].len = 0;
+		return;
+	}
+
+	buffer_get(&c->rcvbuf, NULL, len);
+
+	for(int i = 0; i < NSACKS && c->sacks[i].len; ) {
+		if(len < c->sacks[i].offset) {
+			c->sacks[i].offset -= len;
+			i++;
+		} else if(len < c->sacks[i].offset + c->sacks[i].len) {
+			c->sacks[i].len -= len - c->sacks[i].offset;
+			c->sacks[i].offset = 0;
+			i++;
+		} else {
+			if(i < NSACKS - 1) {
+				memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof c->sacks[i]);
+				c->sacks[NSACKS - 1].len = 0;
+			} else {
+				c->sacks[i].len = 0;
+				break;
+			}
+		}
+	}
+
+	for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
+		debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
+}
+
+static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
+	debug("out of order packet, offset %u\n", offset);
+	// Packet loss or reordering occured. Store the data in the buffer.
+	ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
+	if(rxd < len)
+		abort();
+
+	// Make note of where we put it.
+	for(int i = 0; i < NSACKS; i++) {
+		if(!c->sacks[i].len) { // nothing to merge, add new entry
+			debug("New SACK entry %d\n", i);
+			c->sacks[i].offset = offset;
+			c->sacks[i].len = rxd;
+			break;
+		} else if(offset < c->sacks[i].offset) {
+			if(offset + rxd < c->sacks[i].offset) { // insert before
+				if(!c->sacks[NSACKS - 1].len) { // only if room left
+					debug("Insert SACK entry at %d\n", i);
+					memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof c->sacks[i]);
+					c->sacks[i].offset = offset;
+					c->sacks[i].len = rxd;
+				} else {
+					debug("SACK entries full, dropping packet\n");
+				}
+				break;
+			} else { // merge
+				debug("Merge with start of SACK entry at %d\n", i);
+				c->sacks[i].offset = offset;
+				break;
+			}
+		} else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
+			if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
+				debug("Merge with end of SACK entry at %d\n", i);
+				c->sacks[i].len = offset + rxd - c->sacks[i].offset;
+				// TODO: handle potential merge with next entry
+			}
+			break;
+		}
+	}
+
+	for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
+		debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
+}
+
+static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
+	// Check if we can process out-of-order data now.
+	if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
+		debug("incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
+		buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
+		len = max(len, c->sacks[0].offset + c->sacks[0].len);
+		data = c->rcvbuf.data;
+	}
+
+	if(c->recv) {
+		ssize_t rxd = c->recv(c, data, len);
+		if(rxd != len) {
+			// TODO: handle the application not accepting all data.
+			abort();
+		}
+	}
+
+	if(c->rcvbuf.used)
+		sack_consume(c, len);
+
+	c->rcv.nxt += len;
+}
+
+
+static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
+	if(!is_reliable(c)) {
+		c->recv(c, data, len);
+		c->rcv.nxt = seq + len;
+		return;
+	}
+
+	uint32_t offset = seqdiff(seq, c->rcv.nxt);
+	if(offset + len > c->rcvbuf.maxsize)
+		abort();
+
+	if(offset)
+		handle_out_of_order(c, offset, data, len);
+	else
+		handle_in_order(c, data, len);
+}
+
 
 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 	if(!utcp) {
@@ -582,6 +849,19 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 				goto reset;
 			}
 
+			// Parse auxilliary information
+			if(hdr.aux) {
+				if(hdr.aux != 0x0101 || len < 4 || ((uint8_t *)data)[0] != 1) {
+					len = 1;
+					goto reset;
+				}
+				c->flags = ((uint8_t *)data)[3] & 0x7;
+				data += 4;
+				len -= 4;
+			} else {
+				c->flags = UTCP_TCP;
+			}
+
 			// Return SYN+ACK, go to SYN_RECEIVED state
 			c->snd.wnd = hdr.wnd;
 			c->rcv.irs = hdr.seq;
@@ -609,8 +889,10 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 	// In case this is for a CLOSED connection, ignore the packet.
 	// TODO: make it so incoming packets can never match a CLOSED connection.
 
-	if(c->state == CLOSED)
+	if(c->state == CLOSED) {
+		debug("Got packet for closed connection\n");
 		return 0;
+	}
 
 	// It is for an existing connection.
 
@@ -644,45 +926,41 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 
 	if(c->state == SYN_SENT)
 		acceptable = true;
-
-	// TODO: handle packets overlapping c->rcv.nxt.
-#if 0
-	// Only use this when accepting out-of-order packets.
 	else if(len == 0)
-		if(c->rcv.wnd == 0)
-			acceptable = hdr.seq == c->rcv.nxt;
-		else
-			acceptable = (seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt + c->rcv.wnd) < 0);
-	else
-		if(c->rcv.wnd == 0)
-			// We don't accept data when the receive window is zero.
-			acceptable = false;
-		else
-			// Both start and end of packet must be within the receive window
-			acceptable = (seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt + c->rcv.wnd) < 0)
-				|| (seqdiff(hdr.seq + len + 1, c->rcv.nxt) >= 0 && seqdiff(hdr.seq + len - 1, c->rcv.nxt + c->rcv.wnd) < 0);
-#else
-	if(c->state != SYN_SENT)
-		acceptable = hdr.seq == c->rcv.nxt;
-#endif
+		acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
+	else {
+		int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
+
+		// cut already accepted front overlapping
+		if(rcv_offset < 0) {
+			acceptable = len > -rcv_offset;
+			if(acceptable) {
+				data -= rcv_offset;
+				len += rcv_offset;
+				hdr.seq -= rcv_offset;
+			}
+		} else {
+			acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
+		}
+	}
 
 	if(!acceptable) {
-		debug("Packet not acceptable, %u <= %u + %zu < %u\n", c->rcv.nxt, hdr.seq, len, c->rcv.nxt + c->rcv.wnd);
+		debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
 		// Ignore unacceptable RST packets.
 		if(hdr.ctl & RST)
 			return 0;
-		// Otherwise, send an ACK back in the hope things improve.
-		ack(c, true);
-		return 0;
+		// Otherwise, continue processing.
+		len = 0;
 	}
 
 	c->snd.wnd = hdr.wnd; // TODO: move below
 
 	// 1c. Drop packets with an invalid ACK.
-	// ackno should not roll back, and it should also not be bigger than snd.nxt.
+	// ackno should not roll back, and it should also not be bigger than what we ever could have sent
+	// (= snd.una + c->sndbuf.used).
 
-	if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.nxt) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
-		debug("Packet ack seqno out of range, %u %u %u\n", hdr.ack, c->snd.una, c->snd.nxt);
+	if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
+		debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
 		// Ignore unacceptable RST packets.
 		if(hdr.ctl & RST)
 			return 0;
@@ -748,6 +1026,20 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 	prevrcvnxt = c->rcv.nxt;
 
 	if(advanced) {
+		// RTT measurement
+		if(c->rtt_start.tv_sec) {
+			if(c->rtt_seq == hdr.ack) {
+				struct timeval now, diff;
+				gettimeofday(&now, NULL);
+				timersub(&now, &c->rtt_start, &diff);
+				update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
+				c->rtt_start.tv_sec = 0;
+			} else if(c->rtt_seq < hdr.ack) {
+				debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
+				c->rtt_start.tv_sec = 0;
+			}
+		}
+
 		int32_t data_acked = advanced;
 
 		switch(c->state) {
@@ -768,6 +1060,10 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 		if(data_acked)
 			buffer_get(&c->sndbuf, NULL, data_acked);
 
+		// Also advance snd.nxt if possible
+		if(seqdiff(c->snd.nxt, hdr.ack) < 0)
+			c->snd.nxt = hdr.ack;
+
 		c->snd.una = hdr.ack;
 
 		c->dupack = 0;
@@ -792,15 +1088,16 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 			break;
 		}
 	} else {
-		if(!len) {
+		if(!len && is_reliable(c)) {
 			c->dupack++;
 			if(c->dupack == 3) {
 				debug("Triplicate ACK\n");
 				//TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
 				//We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
-				//This will cause us to start retransmitting, but at the same speed as the incoming ACKs arrive,
-				//thus preventing a drop in speed.
+				//Reset the congestion window so we wait for ACKs.
 				c->snd.nxt = c->snd.una;
+				c->snd.cwnd = utcp->mtu;
+				start_retransmit_timer(c);
 			}
 		}
 	}
@@ -809,8 +1106,10 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 
 	if(advanced) {
 		timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
-		if(c->snd.una == c->snd.nxt)
-			timerclear(&c->rtrx_timeout);
+		if(c->snd.una == c->snd.last)
+			stop_retransmit_timer(c);
+		else if(is_reliable(c))
+			start_retransmit_timer(c);
 	}
 
 	// 5. Process SYN stuff
@@ -891,28 +1190,12 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 			return 0;
 		}
 
-		ssize_t rxd;
-
-		if(c->recv) {
-			rxd = c->recv(c, data, len);
-			if(rxd != len) {
-				// TODO: once we have a receive buffer, handle the application not accepting all data.
-				abort();
-			}
-			if(rxd < 0)
-				rxd = 0;
-			else if(rxd > len)
-				rxd = len; // Bad application, bad!
-		} else {
-			rxd = len;
-		}
-
-		c->rcv.nxt += len;
+		handle_incoming_data(c, hdr.seq, data, len);
 	}
 
 	// 7. Process FIN stuff
 
-	if(hdr.ctl & FIN) {
+	if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
 		switch(c->state) {
 		case SYN_SENT:
 		case SYN_RECEIVED:
@@ -962,8 +1245,7 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 	// - or we got an ack, so we should maybe send a bit more data
 	//   -> sendatleastone = false
 
-ack:
-	ack(c, prevrcvnxt != c->rcv.nxt);
+	ack(c, len || prevrcvnxt != c->rcv.nxt);
 	return 0;
 
 reset:
@@ -984,7 +1266,7 @@ reset:
 }
 
 int utcp_shutdown(struct utcp_connection *c, int dir) {
-	debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c->snd.last);
+	debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
 	if(!c) {
 		errno = EFAULT;
 		return -1;
@@ -996,13 +1278,26 @@ int utcp_shutdown(struct utcp_connection *c, int dir) {
 		return -1;
 	}
 
-	// TODO: handle dir
-	// TODO: check that repeated calls with the same parameters should have no effect
+	if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
+		errno = EINVAL;
+		return -1;
+	}
+
+	// TCP does not have a provision for stopping incoming packets.
+	// The best we can do is to just ignore them.
+	if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR)
+		c->recv = NULL;
+
+	// The rest of the code deals with shutting down writes.
+	if(dir == UTCP_SHUT_RD)
+		return 0;
 
 	switch(c->state) {
 	case CLOSED:
-		return 0;
 	case LISTEN:
+		errno = ENOTCONN;
+		return -1;
+
 	case SYN_SENT:
 		set_state(c, CLOSED);
 		return 0;
@@ -1027,11 +1322,13 @@ int utcp_shutdown(struct utcp_connection *c, int dir) {
 	c->snd.last++;
 
 	ack(c, false);
+	if(!timerisset(&c->rtrx_timeout))
+		start_retransmit_timer(c);
 	return 0;
 }
 
 int utcp_close(struct utcp_connection *c) {
-	if(utcp_shutdown(c, SHUT_RDWR))
+	if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN)
 		return -1;
 	c->recv = NULL;
 	c->poll = NULL;
@@ -1107,6 +1404,7 @@ struct timeval utcp_timeout(struct utcp *utcp) {
 		if(!c)
 			continue;
 
+		// delete connections that have been utcp_close()d.
 		if(c->state == CLOSED) {
 			if(c->reapable) {
 				debug("Reaping %p\n", c);
@@ -1125,6 +1423,7 @@ struct timeval utcp_timeout(struct utcp *utcp) {
 		}
 
 		if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
+			debug("retransmit()\n");
 			retransmit(c);
 		}
 
@@ -1134,13 +1433,6 @@ struct timeval utcp_timeout(struct utcp *utcp) {
 		if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
 			next = c->conn_timeout;
 
-		if(c->snd.nxt != c->snd.una) {
-			c->rtrx_timeout = now;
-			c->rtrx_timeout.tv_sec++;
-		} else {
-			timerclear(&c->rtrx_timeout);
-		}
-
 		if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
 			next = c->rtrx_timeout;
 	}
@@ -1150,22 +1442,34 @@ struct timeval utcp_timeout(struct utcp *utcp) {
 	return diff;
 }
 
-struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
-	struct utcp *utcp = calloc(1, sizeof *utcp);
+bool utcp_is_active(struct utcp *utcp) {
 	if(!utcp)
-		return NULL;
+		return false;
 
+	for(int i = 0; i < utcp->nconnections; i++)
+		if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT)
+			return true;
+
+	return false;
+}
+
+struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
 	if(!send) {
 		errno = EFAULT;
 		return NULL;
 	}
 
+	struct utcp *utcp = calloc(1, sizeof *utcp);
+	if(!utcp)
+		return NULL;
+
 	utcp->accept = accept;
 	utcp->pre_accept = pre_accept;
 	utcp->send = send;
 	utcp->priv = priv;
-	utcp->mtu = 1000;
-	utcp->timeout = 60;
+	utcp->mtu = DEFAULT_MTU;
+	utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
+	utcp->rto = START_RTO; // usec
 
 	return utcp;
 }
@@ -1176,6 +1480,7 @@ void utcp_exit(struct utcp *utcp) {
 	for(int i = 0; i < utcp->nconnections; i++) {
 		if(!utcp->connections[i]->reapable)
 			debug("Warning, freeing unclosed connection %p\n", utcp->connections[i]);
+		buffer_exit(&utcp->connections[i]->rcvbuf);
 		buffer_exit(&utcp->connections[i]->sndbuf);
 		free(utcp->connections[i]);
 	}
@@ -1221,6 +1526,25 @@ void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
 		c->sndbuf.maxsize = -1;
 }
 
+size_t utcp_get_rcvbuf(struct utcp_connection *c) {
+	return c ? c->rcvbuf.maxsize : 0;
+}
+
+size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
+	if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
+		return buffer_free(&c->rcvbuf);
+	else
+		return 0;
+}
+
+void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
+	if(!c)
+		return;
+	c->rcvbuf.maxsize = size;
+	if(c->rcvbuf.maxsize != size)
+		c->rcvbuf.maxsize = -1;
+}
+
 bool utcp_get_nodelay(struct utcp_connection *c) {
 	return c ? c->nodelay : false;
 }
@@ -1252,3 +1576,10 @@ void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
 	if(c)
 		c->poll = poll;
 }
+
+void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
+	if(utcp) {
+		utcp->accept = accept;
+		utcp->pre_accept = pre_accept;
+	}
+}