+ if(buf->size - buf->offset <= offset) {
+ // The offset wrapped
+ realoffset -= buf->size;
+ }
+
+ if(buf->size - realoffset < len) {
+ // The data is wrapped
+ ssize_t rx1 = c->recv(c, buf->data + realoffset, buf->size - realoffset);
+
+ if(rx1 < buf->size - realoffset) {
+ return rx1;
+ }
+
+ // The channel might have been closed by the previous callback
+ if(!c->recv) {
+ return len;
+ }
+
+ ssize_t rx2 = c->recv(c, buf->data, len - (buf->size - realoffset));
+
+ if(rx2 < 0) {
+ return rx2;
+ } else {
+ return rx1 + rx2;
+ }
+ } else {
+ return c->recv(c, buf->data + realoffset, len);
+ }
+}
+
+// Discard data from the buffer.
+static ssize_t buffer_discard(struct buffer *buf, size_t len) {
+ if(buf->used < len) {
+ len = buf->used;
+ }
+
+ if(buf->size - buf->offset <= len) {
+ buf->offset -= buf->size;
+ }
+
+ 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;
+ }
+
+ buf->maxsize = maxsize;
+
+ return buf->size >= minsize || buffer_resize(buf, minsize);
+}
+
+static void buffer_exit(struct buffer *buf) {
+ free(buf->data);
+ memset(buf, 0, sizeof(*buf));
+}
+
+static uint32_t buffer_free(const struct buffer *buf) {
+ return buf->maxsize > buf->used ? buf->maxsize - buf->used : 0;
+}
+
+// Connections are stored in a sorted list.
+// This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
+
+static int compare(const void *va, const void *vb) {
+ assert(va && vb);
+
+ const struct utcp_connection *a = *(struct utcp_connection **)va;
+ const struct utcp_connection *b = *(struct utcp_connection **)vb;
+
+ assert(a && b);
+ assert(a->src && b->src);
+
+ int c = (int)a->src - (int)b->src;
+
+ if(c) {
+ return c;
+ }
+
+ c = (int)a->dst - (int)b->dst;
+ return c;
+}
+
+static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
+ if(!utcp->nconnections) {
+ return NULL;
+ }
+
+ struct utcp_connection key = {
+ .src = src,
+ .dst = dst,
+ }, *keyp = &key;
+ struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
+ return match ? *match : NULL;
+}
+
+static void free_connection(struct utcp_connection *c) {
+ struct utcp *utcp = c->utcp;
+ struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
+
+ assert(cp);
+
+ int i = cp - utcp->connections;
+ memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
+ utcp->nconnections--;
+
+ buffer_exit(&c->rcvbuf);
+ buffer_exit(&c->sndbuf);
+ free(c);
+}
+
+static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
+ // Check whether this combination of src and dst is free
+
+ if(src) {
+ if(find_connection(utcp, src, dst)) {
+ errno = EADDRINUSE;
+ return NULL;
+ }
+ } else { // If src == 0, generate a random port number with the high bit set
+ if(utcp->nconnections >= 32767) {
+ errno = ENOMEM;
+ return NULL;
+ }
+
+ src = rand() | 0x8000;
+
+ while(find_connection(utcp, src, dst)) {
+ src++;
+ }
+ }
+
+ // Allocate memory for the new connection
+
+ if(utcp->nconnections >= utcp->nallocated) {
+ if(!utcp->nallocated) {
+ utcp->nallocated = 4;
+ } else {
+ utcp->nallocated *= 2;
+ }
+
+ struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
+
+ if(!new_array) {
+ return NULL;
+ }
+
+ utcp->connections = new_array;
+ }
+
+ struct utcp_connection *c = calloc(1, sizeof(*c));
+
+ if(!c) {
+ return NULL;
+ }
+
+ if(!buffer_set_size(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
+ free(c);
+ return NULL;
+ }
+
+ if(!buffer_set_size(&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->snd.last = c->snd.nxt;
+ c->snd.cwnd = (utcp->mss > 2190 ? 2 : utcp->mss > 1095 ? 3 : 4) * utcp->mss;
+ c->snd.ssthresh = ~0;
+ debug_cwnd(c);
+ c->srtt = 0;
+ c->rttvar = 0;
+ c->rto = START_RTO;
+ c->utcp = utcp;
+
+ // Add it to the sorted list of connections
+
+ utcp->connections[utcp->nconnections++] = c;
+ qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
+
+ return c;
+}
+
+static inline uint32_t absdiff(uint32_t a, uint32_t b) {
+ if(a > b) {
+ return a - b;
+ } else {
+ return b - a;
+ }
+}
+
+// Update RTT variables. See RFC 6298.
+static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
+ if(!rtt) {
+ debug(c, "invalid rtt\n");
+ return;
+ }
+
+ if(!c->srtt) {
+ c->srtt = rtt;
+ c->rttvar = rtt / 2;
+ } else {
+ c->rttvar = (c->rttvar * 3 + absdiff(c->srtt, rtt)) / 4;
+ c->srtt = (c->srtt * 7 + rtt) / 8;
+ }
+
+ c->rto = c->srtt + max(4 * c->rttvar, CLOCK_GRANULARITY);
+
+ if(c->rto > MAX_RTO) {
+ c->rto = MAX_RTO;
+ }
+
+ debug(c, "rtt %u srtt %u rttvar %u rto %u\n", rtt, c->srtt, c->rttvar, c->rto);
+}
+
+static void start_retransmit_timer(struct utcp_connection *c) {
+ clock_gettime(UTCP_CLOCK, &c->rtrx_timeout);
+
+ uint32_t rto = c->rto;
+
+ while(rto > USEC_PER_SEC) {
+ c->rtrx_timeout.tv_sec++;
+ rto -= USEC_PER_SEC;
+ }
+
+ c->rtrx_timeout.tv_nsec += rto * 1000;
+
+ if(c->rtrx_timeout.tv_nsec >= NSEC_PER_SEC) {
+ c->rtrx_timeout.tv_nsec -= NSEC_PER_SEC;
+ c->rtrx_timeout.tv_sec++;
+ }
+
+ debug(c, "rtrx_timeout %ld.%06lu\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_nsec);
+}
+
+static void stop_retransmit_timer(struct utcp_connection *c) {
+ timespec_clear(&c->rtrx_timeout);
+ debug(c, "rtrx_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 & ~0x1f) == 0);
+
+ c->flags = flags;
+ c->recv = recv;
+ c->priv = priv;
+
+ 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->rcvbuf.maxsize;
+ 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(c, "send", &pkt, sizeof(pkt));
+ utcp->send(utcp, &pkt, sizeof(pkt));
+
+ clock_gettime(UTCP_CLOCK, &c->conn_timeout);
+ c->conn_timeout.tv_sec += utcp->timeout;
+
+ start_retransmit_timer(c);
+
+ 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(c, "accept() called on invalid connection in state %s\n", c, strstate[c->state]);
+ return;
+ }
+
+ debug(c, "accepted %p %p\n", c, recv, priv);
+ c->recv = recv;
+ c->priv = priv;
+ set_state(c, ESTABLISHED);
+}
+
+static void ack(struct utcp_connection *c, bool sendatleastone) {
+ int32_t left = seqdiff(c->snd.last, c->snd.nxt);
+ 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);
+
+ if(cwndleft <= 0) {
+ left = 0;
+ } else if(cwndleft < left) {
+ left = cwndleft;
+
+ if(!sendatleastone || cwndleft > c->utcp->mss) {
+ left -= left % c->utcp->mss;
+ }
+ }
+
+ debug(c, "cwndleft %d left %d\n", cwndleft, left);
+
+ if(!left && !sendatleastone) {
+ return;
+ }
+
+ struct {
+ struct hdr hdr;
+ uint8_t data[];
+ } *pkt = c->utcp->pkt;
+
+ pkt->hdr.src = c->src;
+ pkt->hdr.dst = c->dst;
+ pkt->hdr.ack = c->rcv.nxt;
+ pkt->hdr.wnd = is_reliable(c) ? c->rcvbuf.maxsize : 0;
+ pkt->hdr.ctl = ACK;
+ pkt->hdr.aux = 0;
+
+ do {
+ uint32_t seglen = left > c->utcp->mss ? c->utcp->mss : left;
+ pkt->hdr.seq = c->snd.nxt;
+
+ buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
+
+ 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;
+ }
+
+ if(!c->rtt_start.tv_sec) {
+ // Start RTT measurement
+ clock_gettime(UTCP_CLOCK, &c->rtt_start);
+ c->rtt_seq = pkt->hdr.seq + seglen;
+ debug(c, "starting RTT measurement, expecting ack %u\n", c->rtt_seq);
+ }
+
+ 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);
+}
+
+ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
+ if(c->reapable) {
+ debug(c, "send() called on closed connection\n");
+ errno = EBADF;
+ return -1;
+ }
+
+ switch(c->state) {
+ case CLOSED:
+ case LISTEN:
+ debug(c, "send() called on unconnected connection\n");
+ errno = ENOTCONN;
+ return -1;
+
+ case SYN_SENT:
+ case SYN_RECEIVED:
+ case ESTABLISHED:
+ case CLOSE_WAIT:
+ break;
+
+ case FIN_WAIT_1:
+ case FIN_WAIT_2:
+ case CLOSING:
+ case LAST_ACK:
+ case TIME_WAIT:
+ debug(c, "send() called on closed connection\n");
+ errno = EPIPE;
+ return -1;
+ }
+
+ // Exit early if we have nothing to send.
+
+ if(!len) {
+ return 0;
+ }
+
+ if(!data) {
+ errno = EFAULT;
+ return -1;
+ }
+
+ // Check if we need to be able to buffer all data
+
+ if(c->flags & UTCP_NO_PARTIAL) {
+ if(len > buffer_free(&c->sndbuf)) {
+ if(len > c->sndbuf.maxsize) {
+ errno = EMSGSIZE;
+ return -1;
+ } else {
+ errno = EWOULDBLOCK;
+ return 0;
+ }
+ }
+ }
+
+ // Add data to send buffer.
+
+ 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;
+ }
+
+ if(len <= 0) {
+ if(is_reliable(c)) {
+ errno = EWOULDBLOCK;
+ return 0;
+ } else {
+ return len;
+ }
+ }
+
+ c->snd.last += len;
+
+ // Don't send anything yet if the connection has not fully established yet
+
+ if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
+ return len;
+ }
+
+ ack(c, false);
+
+ if(!is_reliable(c)) {
+ c->snd.una = c->snd.nxt = c->snd.last;
+ buffer_discard(&c->sndbuf, c->sndbuf.used);
+ }
+
+ if(is_reliable(c) && !timespec_isset(&c->rtrx_timeout)) {
+ start_retransmit_timer(c);
+ }
+
+ if(is_reliable(c) && !timespec_isset(&c->conn_timeout)) {
+ clock_gettime(UTCP_CLOCK, &c->conn_timeout);
+ c->conn_timeout.tv_sec += c->utcp->timeout;
+ }
+
+ return len;
+}
+
+static void swap_ports(struct hdr *hdr) {
+ uint16_t tmp = hdr->src;
+ hdr->src = hdr->dst;
+ hdr->dst = tmp;
+}
+
+static void fast_retransmit(struct utcp_connection *c) {
+ if(c->state == CLOSED || c->snd.last == c->snd.una) {
+ debug(c, "fast_retransmit() called but nothing to retransmit!\n");
+ return;
+ }
+
+ struct utcp *utcp = c->utcp;
+
+ struct {
+ struct hdr hdr;
+ uint8_t data[];
+ } *pkt = c->utcp->pkt;
+
+ pkt->hdr.src = c->src;
+ pkt->hdr.dst = c->dst;
+ pkt->hdr.wnd = c->rcvbuf.maxsize;
+ pkt->hdr.aux = 0;
+
+ switch(c->state) {
+ 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 = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
+
+ if(fin_wanted(c, c->snd.una + len)) {
+ len--;
+ pkt->hdr.ctl |= FIN;
+ }
+
+ buffer_copy(&c->sndbuf, pkt->data, 0, len);
+ print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
+ utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
+ break;
+
+ default:
+ break;
+ }
+}
+
+static void retransmit(struct utcp_connection *c) {
+ if(c->state == CLOSED || c->snd.last == c->snd.una) {
+ debug(c, "retransmit() called but nothing to retransmit!\n");
+ stop_retransmit_timer(c);
+ return;
+ }
+
+ struct utcp *utcp = c->utcp;
+
+ if(utcp->retransmit) {
+ utcp->retransmit(c);
+ }
+
+ struct {
+ struct hdr hdr;
+ uint8_t data[];
+ } *pkt = c->utcp->pkt;
+
+ pkt->hdr.src = c->src;
+ pkt->hdr.dst = c->dst;
+ pkt->hdr.wnd = c->rcvbuf.maxsize;
+ pkt->hdr.aux = 0;
+
+ switch(c->state) {
+ case SYN_SENT:
+ // Send our SYN again
+ pkt->hdr.seq = c->snd.iss;
+ pkt->hdr.ack = 0;
+ pkt->hdr.ctl = SYN;
+ pkt->hdr.aux = 0x0101;
+ pkt->data[0] = 1;
+ pkt->data[1] = 0;
+ pkt->data[2] = 0;
+ pkt->data[3] = c->flags & 0x7;
+ print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + 4);
+ utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
+ break;
+
+ case SYN_RECEIVED:
+ // Send SYNACK again
+ pkt->hdr.seq = c->snd.nxt;
+ pkt->hdr.ack = c->rcv.nxt;
+ pkt->hdr.ctl = SYN | ACK;
+ print_packet(c, "rtrx", pkt, sizeof(pkt->hdr));
+ utcp->send(utcp, pkt, sizeof(pkt->hdr));
+ break;
+
+ 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 = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
+
+ if(fin_wanted(c, c->snd.una + len)) {
+ len--;
+ pkt->hdr.ctl |= FIN;
+ }
+
+ // RFC 5681 slow start after timeout
+ uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
+ c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
+ c->snd.cwnd = utcp->mss;
+ debug_cwnd(c);
+
+ buffer_copy(&c->sndbuf, pkt->data, 0, len);
+ print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
+ utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
+
+ c->snd.nxt = c->snd.una + len;
+ break;
+
+ 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
+ stop_retransmit_timer(c);
+ goto cleanup;
+ }
+
+ start_retransmit_timer(c);
+ c->rto *= 2;
+
+ if(c->rto > MAX_RTO) {
+ c->rto = MAX_RTO;
+ }
+
+ c->rtt_start.tv_sec = 0; // invalidate RTT timer
+ c->dupack = 0; // cancel any ongoing fast recovery
+
+cleanup:
+ return;
+}
+
+/* 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(c, "sack_consume %lu\n", (unsigned long)len);
+
+ if(len > c->rcvbuf.used) {
+ debug(c, "all SACK entries consumed\n");
+ c->sacks[0].len = 0;
+ return;
+ }
+
+ buffer_discard(&c->rcvbuf, 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(c, "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(c, "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 <= 0) {
+ debug(c, "packet outside receive buffer, dropping\n");
+ return;
+ }
+
+ if((size_t)rxd < len) {
+ debug(c, "packet partially outside receive buffer\n");
+ len = rxd;
+ }
+
+ // 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(c, "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(c, "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(c, "SACK entries full, dropping packet\n");
+ }
+
+ break;
+ } else { // merge
+ debug(c, "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(c, "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(c, "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) {
+ if(c->recv) {
+ ssize_t rxd = c->recv(c, data, 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, &c->rcvbuf, offset, remainder);
+
+ if(rxd != (ssize_t)remainder) {
+ // TODO: handle the application not accepting all data.
+ abort();
+ }
+ }
+ }
+
+ if(c->rcvbuf.used) {
+ sack_consume(c, len);
+ }
+
+ 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)) {
+ if(c->recv) {
+ 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);
+
+ if(rxd != (ssize_t)len) {
+ return;
+ }
+
+ // Send the packet if it's the final fragment
+ if(!(hdr->ctl & MF)) {
+ buffer_call(c, &c->rcvbuf, 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)) {
+ handle_unreliable(c, hdr, data, len);
+ return;
+ }
+
+ uint32_t offset = seqdiff(hdr->seq, c->rcv.nxt);
+
+ 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) {
+ const uint8_t *ptr = data;
+
+ if(!utcp) {
+ errno = EFAULT;
+ return -1;
+ }
+
+ if(!len) {
+ return 0;
+ }
+
+ if(!data) {
+ errno = EFAULT;
+ return -1;
+ }
+
+ // Drop packets smaller than the header
+
+ struct hdr hdr;
+
+ if(len < sizeof(hdr)) {
+ print_packet(NULL, "recv", data, len);
+ errno = EBADMSG;
+ return -1;
+ }
+
+ // Make a copy from the potentially unaligned data to a struct hdr
+
+ memcpy(&hdr, ptr, sizeof(hdr));
+
+ // Try to match the packet to an existing connection
+
+ struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
+ print_packet(c, "recv", data, len);
+
+ // Process the header
+
+ ptr += sizeof(hdr);
+ len -= sizeof(hdr);
+
+ // Drop packets with an unknown CTL flag
+
+ if(hdr.ctl & ~(SYN | ACK | RST | FIN | MF)) {
+ print_packet(NULL, "recv", data, len);
+ errno = EBADMSG;
+ return -1;
+ }
+
+ // Check for auxiliary headers
+
+ const uint8_t *init = NULL;
+
+ uint16_t aux = hdr.aux;
+
+ while(aux) {
+ size_t auxlen = 4 * (aux >> 8) & 0xf;
+ uint8_t auxtype = aux & 0xff;
+
+ if(len < auxlen) {
+ errno = EBADMSG;
+ return -1;
+ }
+
+ switch(auxtype) {
+ case AUX_INIT:
+ if(!(hdr.ctl & SYN) || auxlen != 4) {
+ errno = EBADMSG;
+ return -1;
+ }
+
+ init = ptr;
+ break;
+
+ default:
+ errno = EBADMSG;
+ return -1;
+ }
+
+ len -= auxlen;
+ ptr += auxlen;
+
+ if(!(aux & 0x800)) {
+ break;
+ }
+
+ if(len < 2) {
+ errno = EBADMSG;
+ return -1;
+ }
+
+ memcpy(&aux, ptr, 2);
+ len -= 2;
+ ptr += 2;
+ }
+
+ bool has_data = len || (hdr.ctl & (SYN | FIN));
+
+ // Is it for a new connection?
+
+ if(!c) {
+ // Ignore RST packets
+
+ if(hdr.ctl & RST) {
+ return 0;
+ }
+
+ // Is it a SYN packet and are we LISTENing?
+
+ if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
+ // If we don't want to accept it, send a RST back
+ if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
+ len = 1;
+ goto reset;
+ }
+
+ // Try to allocate memory, otherwise send a RST back
+ c = allocate_connection(utcp, hdr.dst, hdr.src);
+
+ if(!c) {
+ len = 1;
+ goto reset;
+ }
+
+ // Parse auxilliary information
+ if(init) {
+ if(init[0] < 1) {
+ len = 1;
+ goto reset;
+ }
+
+ c->flags = init[3] & 0x7;
+ } else {
+ c->flags = UTCP_TCP;
+ }
+
+synack:
+ // Return SYN+ACK, go to SYN_RECEIVED state
+ c->snd.wnd = hdr.wnd;
+ c->rcv.irs = hdr.seq;
+ c->rcv.nxt = c->rcv.irs + 1;
+ set_state(c, SYN_RECEIVED);
+
+ struct {
+ struct hdr hdr;
+ uint8_t data[4];
+ } pkt;
+
+ pkt.hdr.src = c->src;
+ pkt.hdr.dst = c->dst;
+ pkt.hdr.ack = c->rcv.irs + 1;
+ pkt.hdr.seq = c->snd.iss;
+ pkt.hdr.wnd = c->rcvbuf.maxsize;
+ pkt.hdr.ctl = SYN | ACK;
+
+ if(init) {
+ pkt.hdr.aux = 0x0101;
+ pkt.data[0] = 1;
+ pkt.data[1] = 0;
+ pkt.data[2] = 0;
+ pkt.data[3] = c->flags & 0x7;
+ print_packet(c, "send", &pkt, sizeof(hdr) + 4);
+ utcp->send(utcp, &pkt, sizeof(hdr) + 4);
+ } else {
+ pkt.hdr.aux = 0;
+ print_packet(c, "send", &pkt, sizeof(hdr));
+ utcp->send(utcp, &pkt, sizeof(hdr));
+ }
+
+ start_retransmit_timer(c);
+ } else {
+ // No, we don't want your packets, send a RST back
+ len = 1;
+ goto reset;
+ }
+
+ return 0;
+ }
+
+ debug(c, "state %s\n", strstate[c->state]);
+
+ // 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) {
+ debug(c, "got packet for closed connection\n");
+ return 0;
+ }
+
+ // It is for an existing connection.
+
+ // 1. Drop invalid packets.
+
+ // 1a. Drop packets that should not happen in our current state.
+
+ switch(c->state) {
+ case SYN_SENT:
+ case SYN_RECEIVED:
+ case ESTABLISHED:
+ case FIN_WAIT_1:
+ case FIN_WAIT_2:
+ case CLOSE_WAIT:
+ case CLOSING:
+ case LAST_ACK:
+ case TIME_WAIT:
+ break;
+
+ default:
+#ifdef UTCP_DEBUG
+ abort();
+#endif
+ break;
+ }
+
+ // 1b. Discard data that is not in our receive window.
+
+ if(is_reliable(c)) {
+ bool acceptable;
+
+ if(c->state == SYN_SENT) {
+ acceptable = true;
+ } else if(len == 0) {
+ 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 > (size_t) - rcv_offset;
+
+ if(acceptable) {
+ ptr -= 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(c, "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, continue processing.
+ len = 0;
+ }
+ } else {
+#if UTCP_DEBUG
+ int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
+
+ if(rcv_offset) {
+ debug(c, "packet out of order, offset %u bytes", rcv_offset);
+ }
+
+#endif
+ }
+
+ 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 what we ever could have sent
+ // (= snd.una + c->sndbuf.used).
+
+ if(!is_reliable(c)) {
+ if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
+ hdr.ack = c->snd.una;
+ }
+ }
+
+ if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
+ debug(c, "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;
+ }
+
+ goto reset;
+ }
+
+ // 2. Handle RST packets
+
+ if(hdr.ctl & RST) {
+ switch(c->state) {
+ case SYN_SENT:
+ if(!(hdr.ctl & ACK)) {
+ return 0;
+ }
+
+ // The peer has refused our connection.
+ set_state(c, CLOSED);
+ errno = ECONNREFUSED;
+
+ if(c->recv) {
+ c->recv(c, NULL, 0);
+ }
+
+ if(c->poll && !c->reapable) {
+ c->poll(c, 0);
+ }
+
+ return 0;
+
+ case SYN_RECEIVED:
+ if(hdr.ctl & ACK) {
+ return 0;
+ }
+
+ // We haven't told the application about this connection yet. Silently delete.
+ free_connection(c);
+ return 0;
+
+ case ESTABLISHED:
+ case FIN_WAIT_1:
+ case FIN_WAIT_2:
+ case CLOSE_WAIT:
+ if(hdr.ctl & ACK) {
+ return 0;
+ }
+
+ // The peer has aborted our connection.
+ set_state(c, CLOSED);
+ errno = ECONNRESET;
+
+ if(c->recv) {
+ c->recv(c, NULL, 0);
+ }
+
+ if(c->poll && !c->reapable) {
+ c->poll(c, 0);
+ }
+
+ return 0;
+
+ case CLOSING:
+ case LAST_ACK:
+ case TIME_WAIT:
+ if(hdr.ctl & ACK) {
+ return 0;
+ }
+
+ // As far as the application is concerned, the connection has already been closed.
+ // If it has called utcp_close() already, we can immediately free this connection.
+ if(c->reapable) {
+ free_connection(c);
+ return 0;
+ }
+
+ // Otherwise, immediately move to the CLOSED state.
+ set_state(c, CLOSED);
+ return 0;
+
+ default:
+#ifdef UTCP_DEBUG
+ abort();
+#endif
+ break;
+ }
+ }
+
+ uint32_t advanced;
+
+ if(!(hdr.ctl & ACK)) {
+ advanced = 0;
+ goto skip_ack;
+ }
+
+ // 3. Advance snd.una
+
+ advanced = seqdiff(hdr.ack, c->snd.una);
+
+ if(advanced) {
+ // RTT measurement
+ if(c->rtt_start.tv_sec) {
+ if(c->rtt_seq == hdr.ack) {
+ struct timespec now;
+ clock_gettime(UTCP_CLOCK, &now);
+ int32_t diff = timespec_diff_usec(&now, &c->rtt_start);
+ update_rtt(c, diff);
+ c->rtt_start.tv_sec = 0;
+ } else if(c->rtt_seq < hdr.ack) {
+ debug(c, "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) {
+ case SYN_SENT:
+ case SYN_RECEIVED:
+ data_acked--;
+ break;
+
+ // TODO: handle FIN as well.
+ default:
+ break;
+ }
+
+ 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_discard(&c->sndbuf, data_acked);
+
+ if(is_reliable(c)) {
+ c->do_poll = true;
+ }
+ }
+
+ // Also advance snd.nxt if possible
+ if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
+ c->snd.nxt = hdr.ack;
+ }
+
+ c->snd.una = hdr.ack;
+
+ if(c->dupack) {
+ if(c->dupack >= 3) {
+ debug(c, "fast recovery ended\n");
+ c->snd.cwnd = c->snd.ssthresh;
+ }
+
+ c->dupack = 0;
+ }
+
+ // Increase the congestion window according to RFC 5681
+ if(c->snd.cwnd < c->snd.ssthresh) {
+ c->snd.cwnd += min(advanced, utcp->mss); // eq. 2
+ } else {
+ c->snd.cwnd += max(1, (utcp->mss * utcp->mss) / 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:
+ if(c->snd.una == c->snd.last) {
+ set_state(c, FIN_WAIT_2);
+ }
+
+ break;
+
+ case CLOSING:
+ if(c->snd.una == c->snd.last) {
+ clock_gettime(UTCP_CLOCK, &c->conn_timeout);
+ c->conn_timeout.tv_sec += utcp->timeout;
+ set_state(c, TIME_WAIT);
+ }
+
+ break;
+
+ default:
+ break;
+ }
+ } else {
+ if(!len && is_reliable(c) && c->snd.una != c->snd.last) {
+ c->dupack++;
+ debug(c, "duplicate ACK %d\n", c->dupack);
+
+ if(c->dupack == 3) {
+ // RFC 5681 fast recovery
+ debug(c, "fast recovery started\n", c->dupack);
+ uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
+ c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
+ c->snd.cwnd = min(c->snd.ssthresh + 3 * utcp->mss, c->sndbuf.maxsize);
+
+ if(c->snd.cwnd > c->sndbuf.maxsize) {
+ c->snd.cwnd = c->sndbuf.maxsize;
+ }
+
+ debug_cwnd(c);
+
+ fast_retransmit(c);
+ } else if(c->dupack > 3) {
+ c->snd.cwnd += utcp->mss;
+
+ if(c->snd.cwnd > c->sndbuf.maxsize) {
+ c->snd.cwnd = c->sndbuf.maxsize;
+ }
+
+ debug_cwnd(c);
+ }
+
+ // We got an ACK which indicates the other side did get one of our packets.
+ // Reset the retransmission timer to avoid going to slow start,
+ // but don't touch the connection timeout.
+ start_retransmit_timer(c);
+ }
+ }
+
+ // 4. Update timers
+
+ if(advanced) {
+ if(c->snd.una == c->snd.last) {
+ stop_retransmit_timer(c);
+ timespec_clear(&c->conn_timeout);
+ } else if(is_reliable(c)) {
+ start_retransmit_timer(c);
+ clock_gettime(UTCP_CLOCK, &c->conn_timeout);
+ c->conn_timeout.tv_sec += utcp->timeout;
+ }
+ }
+
+skip_ack:
+ // 5. Process SYN stuff
+
+ if(hdr.ctl & SYN) {
+ switch(c->state) {
+ case SYN_SENT:
+
+ // This is a SYNACK. It should always have ACKed the SYN.
+ if(!advanced) {
+ goto reset;
+ }
+
+ c->rcv.irs = hdr.seq;
+ c->rcv.nxt = hdr.seq + 1;
+
+ if(c->shut_wr) {
+ c->snd.last++;
+ set_state(c, FIN_WAIT_1);
+ } else {
+ set_state(c, ESTABLISHED);
+ }
+
+ 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:
+ case CLOSE_WAIT:
+ case CLOSING:
+ case LAST_ACK:
+ case TIME_WAIT:
+ // This could be a retransmission. Ignore the SYN flag, but send an ACK back.
+ break;
+
+ default:
+#ifdef UTCP_DEBUG
+ abort();
+#endif
+ return 0;
+ }
+ }
+
+ // 6. Process new data
+
+ if(c->state == SYN_RECEIVED) {
+ // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
+ if(!advanced) {
+ goto reset;
+ }
+
+ // Are we still LISTENing?
+ if(utcp->accept) {
+ utcp->accept(c, c->src);
+ }
+
+ if(c->state != ESTABLISHED) {
+ set_state(c, CLOSED);
+ c->reapable = true;
+ goto reset;
+ }
+ }
+
+ if(len) {
+ switch(c->state) {
+ case SYN_SENT:
+ case SYN_RECEIVED:
+ // This should never happen.
+#ifdef UTCP_DEBUG
+ abort();
+#endif
+ return 0;
+
+ case ESTABLISHED:
+ case FIN_WAIT_1:
+ case FIN_WAIT_2:
+ break;
+
+ case CLOSE_WAIT:
+ case CLOSING:
+ case LAST_ACK:
+ case TIME_WAIT:
+ // Ehm no, We should never receive more data after a FIN.
+ goto reset;
+
+ default:
+#ifdef UTCP_DEBUG
+ abort();
+#endif
+ return 0;
+ }
+
+ handle_incoming_data(c, &hdr, ptr, len);
+ }
+
+ // 7. Process FIN stuff
+
+ if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
+ switch(c->state) {
+ case SYN_SENT:
+ case SYN_RECEIVED:
+ // This should never happen.
+#ifdef UTCP_DEBUG
+ abort();
+#endif
+ break;
+
+ case ESTABLISHED:
+ set_state(c, CLOSE_WAIT);
+ break;
+
+ case FIN_WAIT_1:
+ set_state(c, CLOSING);
+ break;
+
+ case FIN_WAIT_2:
+ clock_gettime(UTCP_CLOCK, &c->conn_timeout);
+ c->conn_timeout.tv_sec += utcp->timeout;
+ set_state(c, TIME_WAIT);
+ break;
+
+ case CLOSE_WAIT:
+ case CLOSING:
+ case LAST_ACK:
+ case TIME_WAIT:
+ // Ehm, no. We should never receive a second FIN.
+ goto reset;
+
+ default:
+#ifdef UTCP_DEBUG
+ abort();
+#endif
+ break;
+ }
+
+ // FIN counts as one sequence number
+ c->rcv.nxt++;
+ len++;
+
+ // Inform the application that the peer closed its end of the connection.
+ if(c->recv) {
+ errno = 0;
+ c->recv(c, NULL, 0);
+ }