X-Git-Url: http://git.meshlink.io/?a=blobdiff_plain;f=utcp.c;h=10d026efabf2679db6ca419c71131e5fecd7e582;hb=9470c449236afef22210803b359fe44d74e32636;hp=0b36de15263b8b005c0625f866e901054432b5ba;hpb=bf55733700d946f678947a6caaa782b56669a3f1;p=utcp

diff --git a/utcp.c b/utcp.c
index 0b36de1..10d026e 100644
--- a/utcp.c
+++ b/utcp.c
@@ -43,6 +43,14 @@
 #undef poll
 #endif
 
+#ifndef UTCP_CLOCK
+#if defined(CLOCK_MONOTONIC_RAW) && defined(__x86_64__)
+#define UTCP_CLOCK CLOCK_MONOTONIC_RAW
+#else
+#define UTCP_CLOCK CLOCK_MONOTONIC
+#endif
+#endif
+
 static void timespec_sub(const struct timespec *a, const struct timespec *b, struct timespec *r) {
 	r->tv_sec = a->tv_sec - b->tv_sec;
 	r->tv_nsec = a->tv_nsec - b->tv_nsec;
@@ -53,8 +61,7 @@ static void timespec_sub(const struct timespec *a, const struct timespec *b, str
 }
 
 static int32_t timespec_diff_usec(const struct timespec *a, const struct timespec *b) {
-	int64_t diff = (a->tv_sec - b->tv_sec) * 1000000000 + a->tv_sec - b->tv_sec;
-	return diff / 1000;
+	return (a->tv_sec - b->tv_sec) * 1000000 + (a->tv_nsec - b->tv_nsec) / 1000;
 }
 
 static bool timespec_lt(const struct timespec *a, const struct timespec *b) {
@@ -90,14 +97,6 @@ static inline size_t max(size_t a, size_t b) {
 #define UTCP_DEBUG_DATALEN 20
 #endif
 
-#ifndef UTCP_CLOCK
-#if defined(CLOCK_MONOTONIC_RAW) && defined(__x86_64__)
-#define UTCP_CLOCK CLOCK_MONOTONIC_RAW
-#else
-#define UTCP_CLOCK CLOCK_MONOTONIC
-#endif
-#endif
-
 static void debug(struct utcp_connection *c, const char *format, ...) {
 	struct timespec tv;
 	char buf[1024];
@@ -145,12 +144,13 @@ static void print_packet(struct utcp_connection *c, const char *dir, const void
 
 	*p = 0;
 
-	debug(c, "%s: len %lu src %u dst %u seq %u ack %u wnd %u aux %x ctl %s%s%s%s data %s\n",
+	debug(c, "%s: len %lu src %u dst %u seq %u ack %u wnd %u aux %x ctl %s%s%s%s%s data %s\n",
 	      dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux,
 	      hdr.ctl & SYN ? "SYN" : "",
 	      hdr.ctl & RST ? "RST" : "",
 	      hdr.ctl & FIN ? "FIN" : "",
 	      hdr.ctl & ACK ? "ACK" : "",
+	      hdr.ctl & MF ? "MF" : "",
 	      str
 	     );
 }
@@ -220,7 +220,7 @@ static bool buffer_resize(struct buffer *buf, uint32_t newsize) {
 		// [345.........|........012]
 		uint32_t tailsize = buf->size - buf->offset;
 		uint32_t newoffset = newsize - tailsize;
-		memmove(buf + newoffset, buf + buf->offset, tailsize);
+		memmove(buf->data + newoffset, buf->data + buf->offset, tailsize);
 		buf->offset = newoffset;
 	}
 
@@ -320,6 +320,44 @@ static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t
 	return len;
 }
 
+// Copy data from the buffer without removing it.
+static ssize_t buffer_call(struct buffer *buf, utcp_recv_t cb, void *arg, size_t offset, size_t len) {
+	// Ensure we don't copy more than is actually stored in the buffer
+	if(offset >= buf->used) {
+		return 0;
+	}
+
+	if(buf->used - offset < len) {
+		len = buf->used - offset;
+	}
+
+	uint32_t realoffset = buf->offset + offset;
+
+	if(buf->size - buf->offset < offset) {
+		// The offset wrapped
+		realoffset -= buf->size;
+	}
+
+	if(buf->size - realoffset < len) {
+		// The data is wrapped
+		ssize_t rx1 = cb(arg, buf->data + realoffset, buf->size - realoffset);
+
+		if(rx1 < buf->size - realoffset) {
+			return rx1;
+		}
+
+		ssize_t rx2 = cb(arg, buf->data, len - (buf->size - realoffset));
+
+		if(rx2 < 0) {
+			return rx2;
+		} else {
+			return rx1 + rx2;
+		}
+	} else {
+		return cb(arg, buf->data + realoffset, len);
+	}
+}
+
 // Discard data from the buffer.
 static ssize_t buffer_discard(struct buffer *buf, size_t len) {
 	if(buf->used < len) {
@@ -330,12 +368,22 @@ static ssize_t buffer_discard(struct buffer *buf, size_t len) {
 		buf->offset -= buf->size;
 	}
 
-	buf->offset += len;
+	if(buf->used == len) {
+		buf->offset = 0;
+	} else {
+		buf->offset += len;
+	}
+
 	buf->used -= len;
 
 	return len;
 }
 
+static void buffer_clear(struct buffer *buf) {
+	buf->used = 0;
+	buf->offset = 0;
+}
+
 static bool buffer_set_size(struct buffer *buf, uint32_t minsize, uint32_t maxsize) {
 	if(maxsize < minsize) {
 		maxsize = minsize;
@@ -606,7 +654,7 @@ 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 = min(c->snd.cwnd, c->snd.wnd) - seqdiff(c->snd.nxt, c->snd.una);
+	int32_t cwndleft = is_reliable(c) ? min(c->snd.cwnd, c->snd.wnd) - seqdiff(c->snd.nxt, c->snd.una) : MAX_UNRELIABLE_SIZE;
 
 	assert(left >= 0);
 
@@ -634,7 +682,7 @@ 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->rcvbuf.maxsize;
+	pkt->hdr.wnd = is_reliable(c) ? c->rcvbuf.maxsize : 0;
 	pkt->hdr.ctl = ACK;
 	pkt->hdr.aux = 0;
 
@@ -647,6 +695,14 @@ static void ack(struct utcp_connection *c, bool sendatleastone) {
 		c->snd.nxt += seglen;
 		left -= seglen;
 
+		if(!is_reliable(c)) {
+			if(left) {
+				pkt->hdr.ctl |= MF;
+			} else {
+				pkt->hdr.ctl &= ~MF;
+			}
+		}
+
 		if(seglen && fin_wanted(c, c->snd.nxt)) {
 			seglen--;
 			pkt->hdr.ctl |= FIN;
@@ -661,6 +717,10 @@ static void ack(struct utcp_connection *c, bool sendatleastone) {
 
 		print_packet(c, "send", pkt, sizeof(pkt->hdr) + seglen);
 		c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
+
+		if(left && !is_reliable(c)) {
+			pkt->hdr.wnd += seglen;
+		}
 	} while(left);
 }
 
@@ -721,8 +781,13 @@ ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 
 	// Add data to send buffer.
 
-	if(is_reliable(c) || (c->state != SYN_SENT && c->state != SYN_RECEIVED)) {
+	if(is_reliable(c)) {
 		len = buffer_put(&c->sndbuf, data, len);
+	} else if(c->state != SYN_SENT && c->state != SYN_RECEIVED) {
+		if(len > MAX_UNRELIABLE_SIZE || buffer_put(&c->sndbuf, data, len) != (ssize_t)len) {
+			errno = EMSGSIZE;
+			return -1;
+		}
 	} else {
 		return 0;
 	}
@@ -749,6 +814,7 @@ ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 	if(!is_reliable(c)) {
 		c->snd.una = c->snd.nxt = c->snd.last;
 		buffer_discard(&c->sndbuf, c->sndbuf.used);
+		c->do_poll = true;
 	}
 
 	if(is_reliable(c) && !timespec_isset(&c->rtrx_timeout)) {
@@ -1023,23 +1089,32 @@ static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, cons
 }
 
 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(c, "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 < 0 || (size_t)rxd != len) {
+		if(rxd != (ssize_t)len) {
 			// TODO: handle the application not accepting all data.
 			abort();
 		}
 	}
 
+	// Check if we can process out-of-order data now.
+	if(c->sacks[0].len && len >= c->sacks[0].offset) {
+		debug(c, "incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
+
+		if(len < c->sacks[0].offset + c->sacks[0].len) {
+			size_t offset = len;
+			len = c->sacks[0].offset + c->sacks[0].len;
+			size_t remainder = len - offset;
+			ssize_t rxd = buffer_call(&c->rcvbuf, c->recv, c, offset, remainder);
+
+			if(rxd != (ssize_t)remainder) {
+				// TODO: handle the application not accepting all data.
+				abort();
+			}
+		}
+	}
+
 	if(c->rcvbuf.used) {
 		sack_consume(c, len);
 	}
@@ -1047,15 +1122,50 @@ static void handle_in_order(struct utcp_connection *c, const void *data, size_t
 	c->rcv.nxt += len;
 }
 
+static void handle_unreliable(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
+	// Fast path for unfragmented packets
+	if(!hdr->wnd && !(hdr->ctl & MF)) {
+		c->recv(c, data, len);
+		c->rcv.nxt = hdr->seq + len;
+		return;
+	}
+
+	// Ensure reassembled packet are not larger than 64 kiB
+	if(hdr->wnd >= MAX_UNRELIABLE_SIZE || hdr->wnd + len > MAX_UNRELIABLE_SIZE) {
+		return;
+	}
+
+	// Don't accept out of order fragments
+	if(hdr->wnd && hdr->seq != c->rcv.nxt) {
+		return;
+	}
+
+	// Reset the receive buffer for the first fragment
+	if(!hdr->wnd) {
+		buffer_clear(&c->rcvbuf);
+	}
+
+	ssize_t rxd = buffer_put_at(&c->rcvbuf, hdr->wnd, data, len);
 
-static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
+	if(rxd != (ssize_t)len) {
+		return;
+	}
+
+	// Send the packet if it's the final fragment
+	if(!(hdr->ctl & MF)) {
+		buffer_call(&c->rcvbuf, c->recv, c, 0, hdr->wnd + len);
+	}
+
+	c->rcv.nxt = hdr->seq + len;
+}
+
+static void handle_incoming_data(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
 	if(!is_reliable(c)) {
-		c->recv(c, data, len);
-		c->rcv.nxt = seq + len;
+		handle_unreliable(c, hdr, data, len);
 		return;
 	}
 
-	uint32_t offset = seqdiff(seq, c->rcv.nxt);
+	uint32_t offset = seqdiff(hdr->seq, c->rcv.nxt);
 
 	if(offset + len > c->rcvbuf.maxsize) {
 		abort();
@@ -1112,7 +1222,7 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 
 	// Drop packets with an unknown CTL flag
 
-	if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
+	if(hdr.ctl & ~(SYN | ACK | RST | FIN | MF)) {
 		print_packet(NULL, "recv", data, len);
 		errno = EBADMSG;
 		return -1;
@@ -1326,10 +1436,6 @@ synack:
 			debug(c, "packet out of order, offset %u bytes", rcv_offset);
 		}
 
-		if(rcv_offset >= 0) {
-			c->rcv.nxt = hdr.seq + len;
-		}
-
 #endif
 	}
 
@@ -1484,6 +1590,7 @@ synack:
 
 		if(data_acked) {
 			buffer_discard(&c->sndbuf, data_acked);
+			c->do_poll = true;
 		}
 
 		// Also advance snd.nxt if possible
@@ -1684,7 +1791,7 @@ skip_ack:
 			return 0;
 		}
 
-		handle_incoming_data(c, hdr.seq, ptr, len);
+		handle_incoming_data(c, &hdr, ptr, len);
 	}
 
 	// 7. Process FIN stuff
@@ -2001,8 +2108,9 @@ struct timespec utcp_timeout(struct utcp *utcp) {
 		}
 
 		if(c->poll) {
-			if((c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
-				uint32_t len =  buffer_free(&c->sndbuf);
+			if((c->state == ESTABLISHED || c->state == CLOSE_WAIT) && c->do_poll) {
+				c->do_poll = false;
+				uint32_t len = buffer_free(&c->sndbuf);
 
 				if(len) {
 					c->poll(c, len);
@@ -2094,6 +2202,7 @@ void utcp_exit(struct utcp *utcp) {
 	}
 
 	free(utcp->connections);
+	free(utcp->pkt);
 	free(utcp);
 }
 
@@ -2205,6 +2314,8 @@ void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
 	if(c->sndbuf.maxsize != size) {
 		c->sndbuf.maxsize = -1;
 	}
+
+	c->do_poll = buffer_free(&c->sndbuf);
 }
 
 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
@@ -2272,6 +2383,7 @@ void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
 	if(c) {
 		c->poll = poll;
+		c->do_poll = buffer_free(&c->sndbuf);
 	}
 }