2 utcp.c -- Userspace TCP
3 Copyright (C) 2014 Guus Sliepen <guus@tinc-vpn.org>
5 This program is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 2 of the License, or
8 (at your option) any later version.
10 This program is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU General Public License for more details.
15 You should have received a copy of the GNU General Public License along
16 with this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
31 #include <sys/socket.h>
33 #include "utcp_priv.h"
48 #define timersub(a, b, r) do {\
49 (r)->tv_sec = (a)->tv_sec - (b)->tv_sec;\
50 (r)->tv_usec = (a)->tv_usec - (b)->tv_usec;\
52 (r)->tv_sec--, (r)->tv_usec += USEC_PER_SEC;\
56 static inline size_t max(size_t a, size_t b) {
63 static void debug(const char *format, ...) {
66 vfprintf(stderr, format, ap);
70 static void print_packet(struct utcp *utcp, const char *dir, const void *pkt, size_t len) {
72 if(len < sizeof hdr) {
73 debug("%p %s: short packet (%lu bytes)\n", utcp, dir, (unsigned long)len);
77 memcpy(&hdr, pkt, sizeof hdr);
78 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);
88 if(len > sizeof hdr) {
89 uint32_t datalen = len - sizeof hdr;
90 const uint8_t *data = (uint8_t *)pkt + sizeof hdr;
91 char str[datalen * 2 + 1];
94 for(uint32_t i = 0; i < datalen; i++) {
95 *p++ = "0123456789ABCDEF"[data[i] >> 4];
96 *p++ = "0123456789ABCDEF"[data[i] & 15];
100 debug(" data=%s", str);
107 #define print_packet(...)
110 static void set_state(struct utcp_connection *c, enum state state) {
112 if(state == ESTABLISHED)
113 timerclear(&c->conn_timeout);
114 debug("%p new state: %s\n", c->utcp, strstate[state]);
117 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
118 if(seq != c->snd.last)
130 static inline void list_connections(struct utcp *utcp) {
131 debug("%p has %d connections:\n", utcp, utcp->nconnections);
132 for(int i = 0; i < utcp->nconnections; i++)
133 debug(" %u -> %u state %s\n", utcp->connections[i]->src, utcp->connections[i]->dst, strstate[utcp->connections[i]->state]);
136 static int32_t seqdiff(uint32_t a, uint32_t b) {
141 // TODO: convert to ringbuffers to avoid memmove() operations.
143 // Store data into the buffer
144 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
145 if(buf->maxsize <= buf->used)
148 debug("buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
150 size_t required = offset + len;
151 if(required > buf->maxsize) {
152 if(offset >= buf->maxsize)
154 len = buf->maxsize - offset;
155 required = buf->maxsize;
158 if(required > buf->size) {
159 size_t newsize = buf->size;
165 } while(newsize < required);
167 if(newsize > buf->maxsize)
168 newsize = buf->maxsize;
169 char *newdata = realloc(buf->data, newsize);
176 memcpy(buf->data + offset, data, len);
177 if(required > buf->used)
178 buf->used = required;
182 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
183 return buffer_put_at(buf, buf->used, data, len);
186 // Get data from the buffer. data can be NULL.
187 static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
191 memcpy(data, buf->data, len);
193 memmove(buf->data, buf->data + len, buf->used - len);
198 // Copy data from the buffer without removing it.
199 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
200 if(offset >= buf->used)
202 if(offset + len > buf->used)
203 len = buf->used - offset;
204 memcpy(data, buf->data + offset, len);
208 static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
209 memset(buf, 0, sizeof *buf);
211 buf->data = malloc(len);
216 buf->maxsize = maxlen;
220 static void buffer_exit(struct buffer *buf) {
222 memset(buf, 0, sizeof *buf);
225 static uint32_t buffer_free(const struct buffer *buf) {
226 return buf->maxsize - buf->used;
229 // Connections are stored in a sorted list.
230 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
232 static int compare(const void *va, const void *vb) {
235 const struct utcp_connection *a = *(struct utcp_connection **)va;
236 const struct utcp_connection *b = *(struct utcp_connection **)vb;
239 assert(a->src && b->src);
241 int c = (int)a->src - (int)b->src;
244 c = (int)a->dst - (int)b->dst;
248 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
249 if(!utcp->nconnections)
251 struct utcp_connection key = {
255 struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
256 return match ? *match : NULL;
259 static void free_connection(struct utcp_connection *c) {
260 struct utcp *utcp = c->utcp;
261 struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
265 int i = cp - utcp->connections;
266 memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof *cp);
267 utcp->nconnections--;
269 buffer_exit(&c->rcvbuf);
270 buffer_exit(&c->sndbuf);
274 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
275 // Check whether this combination of src and dst is free
278 if(find_connection(utcp, src, dst)) {
282 } else { // If src == 0, generate a random port number with the high bit set
283 if(utcp->nconnections >= 32767) {
287 src = rand() | 0x8000;
288 while(find_connection(utcp, src, dst))
292 // Allocate memory for the new connection
294 if(utcp->nconnections >= utcp->nallocated) {
295 if(!utcp->nallocated)
296 utcp->nallocated = 4;
298 utcp->nallocated *= 2;
299 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof *utcp->connections);
302 utcp->connections = new_array;
305 struct utcp_connection *c = calloc(1, sizeof *c);
309 if(!buffer_init(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
314 if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
315 buffer_exit(&c->sndbuf);
320 // Fill in the details
329 c->snd.una = c->snd.iss;
330 c->snd.nxt = c->snd.iss + 1;
331 c->rcv.wnd = utcp->mtu;
332 c->snd.last = c->snd.nxt;
333 c->snd.cwnd = utcp->mtu;
336 // Add it to the sorted list of connections
338 utcp->connections[utcp->nconnections++] = c;
339 qsort(utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
344 // Update RTT variables. See RFC 6298.
345 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
347 debug("invalid rtt\n");
351 struct utcp *utcp = c->utcp;
355 utcp->rttvar = rtt / 2;
356 utcp->rto = rtt + max(2 * rtt, CLOCK_GRANULARITY);
358 utcp->rttvar = (utcp->rttvar * 3 + abs(utcp->srtt - rtt)) / 4;
359 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
360 utcp->rto = utcp->srtt + max(utcp->rttvar, CLOCK_GRANULARITY);
363 if(utcp->rto > MAX_RTO)
366 debug("rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
369 static void start_retransmit_timer(struct utcp_connection *c) {
370 gettimeofday(&c->rtrx_timeout, NULL);
371 c->rtrx_timeout.tv_usec += c->utcp->rto;
372 while(c->rtrx_timeout.tv_usec >= 1000000) {
373 c->rtrx_timeout.tv_usec -= 1000000;
374 c->rtrx_timeout.tv_sec++;
376 debug("timeout set to %lu.%06lu (%u)\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_usec, c->utcp->rto);
379 static void stop_retransmit_timer(struct utcp_connection *c) {
380 timerclear(&c->rtrx_timeout);
381 debug("timeout cleared\n");
384 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
385 struct utcp_connection *c = allocate_connection(utcp, 0, dst);
396 hdr.seq = c->snd.iss;
398 hdr.wnd = c->rcv.wnd;
402 set_state(c, SYN_SENT);
404 print_packet(utcp, "send", &hdr, sizeof hdr);
405 utcp->send(utcp, &hdr, sizeof hdr);
407 gettimeofday(&c->conn_timeout, NULL);
408 c->conn_timeout.tv_sec += utcp->timeout;
413 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
414 if(c->reapable || c->state != SYN_RECEIVED) {
415 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
419 debug("%p accepted, %p %p\n", c, recv, priv);
422 set_state(c, ESTABLISHED);
425 static void ack(struct utcp_connection *c, bool sendatleastone) {
426 int32_t left = seqdiff(c->snd.last, c->snd.nxt);
427 int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
428 debug("cwndleft = %d\n", cwndleft);
438 if(!left && !sendatleastone)
446 pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
450 pkt->hdr.src = c->src;
451 pkt->hdr.dst = c->dst;
452 pkt->hdr.ack = c->rcv.nxt;
453 pkt->hdr.wnd = c->snd.wnd;
458 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
459 pkt->hdr.seq = c->snd.nxt;
461 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
463 c->snd.nxt += seglen;
466 if(seglen && fin_wanted(c, c->snd.nxt)) {
471 if(!c->rtt_start.tv_sec) {
472 // Start RTT measurement
473 gettimeofday(&c->rtt_start, NULL);
474 c->rtt_seq = pkt->hdr.seq + seglen;
475 debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
478 print_packet(c->utcp, "send", pkt, sizeof pkt->hdr + seglen);
479 c->utcp->send(c->utcp, pkt, sizeof pkt->hdr + seglen);
485 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
487 debug("Error: send() called on closed connection %p\n", c);
497 debug("Error: send() called on unconnected connection %p\n", c);
508 debug("Error: send() called on closing connection %p\n", c);
513 // Add data to send buffer
523 len = buffer_put(&c->sndbuf, data, len);
531 if(!timerisset(&c->rtrx_timeout))
532 start_retransmit_timer(c);
536 static void swap_ports(struct hdr *hdr) {
537 uint16_t tmp = hdr->src;
542 static void retransmit(struct utcp_connection *c) {
543 if(c->state == CLOSED || c->snd.last == c->snd.una) {
544 debug("Retransmit() called but nothing to retransmit!\n");
545 stop_retransmit_timer(c);
549 struct utcp *utcp = c->utcp;
556 pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
560 pkt->hdr.src = c->src;
561 pkt->hdr.dst = c->dst;
562 pkt->hdr.wnd = c->rcv.wnd;
567 // Send our SYN again
568 pkt->hdr.seq = c->snd.iss;
571 print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
572 utcp->send(utcp, pkt, sizeof pkt->hdr);
577 pkt->hdr.seq = c->snd.nxt;
578 pkt->hdr.ack = c->rcv.nxt;
579 pkt->hdr.ctl = SYN | ACK;
580 print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
581 utcp->send(utcp, pkt, sizeof pkt->hdr);
589 // Send unacked data again.
590 pkt->hdr.seq = c->snd.una;
591 pkt->hdr.ack = c->rcv.nxt;
593 uint32_t len = seqdiff(c->snd.last, c->snd.una);
596 if(fin_wanted(c, c->snd.una + len)) {
600 c->snd.nxt = c->snd.una + len;
601 c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
602 buffer_copy(&c->sndbuf, pkt->data, 0, len);
603 print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr + len);
604 utcp->send(utcp, pkt, sizeof pkt->hdr + len);
611 // We shouldn't need to retransmit anything in this state.
615 stop_retransmit_timer(c);
619 start_retransmit_timer(c);
621 if(utcp->rto > MAX_RTO)
623 c->rtt_start.tv_sec = 0; // invalidate RTT timer
629 /* Update receive buffer and SACK entries after consuming data.
633 * |.....0000..1111111111.....22222......3333|
636 * 0..3 represent the SACK entries. The ^ indicates up to which point we want
637 * to remove data from the receive buffer. The idea is to substract "len"
638 * from the offset of all the SACK entries, and then remove/cut down entries
639 * that are shifted to before the start of the receive buffer.
641 * There are three cases:
642 * - the SACK entry is after ^, in that case just change the offset.
643 * - the SACK entry starts before and ends after ^, so we have to
644 * change both its offset and size.
645 * - the SACK entry is completely before ^, in that case delete it.
647 static void sack_consume(struct utcp_connection *c, size_t len) {
648 debug("sack_consume %lu\n", (unsigned long)len);
649 if(len > c->rcvbuf.used) {
650 debug("All SACK entries consumed");
655 buffer_get(&c->rcvbuf, NULL, len);
657 for(int i = 0; i < NSACKS && c->sacks[i].len; ) {
658 if(len < c->sacks[i].offset) {
659 c->sacks[i].offset -= len;
661 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
662 c->sacks[i].len -= len - c->sacks[i].offset;
663 c->sacks[i].offset = 0;
667 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof c->sacks[i]);
668 c->sacks[NSACKS - 1].len = 0;
676 for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
677 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
680 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
681 debug("out of order packet, offset %u\n", offset);
682 // Packet loss or reordering occured. Store the data in the buffer.
683 ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
687 // Make note of where we put it.
688 for(int i = 0; i < NSACKS; i++) {
689 if(!c->sacks[i].len) { // nothing to merge, add new entry
690 debug("New SACK entry %d\n", i);
691 c->sacks[i].offset = offset;
692 c->sacks[i].len = rxd;
694 } else if(offset < c->sacks[i].offset) {
695 if(offset + rxd < c->sacks[i].offset) { // insert before
696 if(!c->sacks[NSACKS - 1].len) { // only if room left
697 debug("Insert SACK entry at %d\n", i);
698 memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof c->sacks[i]);
699 c->sacks[i].offset = offset;
700 c->sacks[i].len = rxd;
702 debug("SACK entries full, dropping packet\n");
706 debug("Merge with start of SACK entry at %d\n", i);
707 c->sacks[i].offset = offset;
710 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
711 if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
712 debug("Merge with end of SACK entry at %d\n", i);
713 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
714 // TODO: handle potential merge with next entry
720 for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
721 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
724 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
725 // Check if we can process out-of-order data now.
726 if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
727 debug("incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
728 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
729 len = max(len, c->sacks[0].offset + c->sacks[0].len);
730 data = c->rcvbuf.data;
734 ssize_t rxd = c->recv(c, data, len);
736 // TODO: handle the application not accepting all data.
742 sack_consume(c, len);
748 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
749 uint32_t offset = seqdiff(seq, c->rcv.nxt);
750 if(offset + len > c->rcvbuf.maxsize)
754 handle_out_of_order(c, offset, data, len);
756 handle_in_order(c, data, len);
760 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
774 print_packet(utcp, "recv", data, len);
776 // Drop packets smaller than the header
779 if(len < sizeof hdr) {
784 // Make a copy from the potentially unaligned data to a struct hdr
786 memcpy(&hdr, data, sizeof hdr);
790 // Drop packets with an unknown CTL flag
792 if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
797 // Try to match the packet to an existing connection
799 struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
801 // Is it for a new connection?
804 // Ignore RST packets
809 // Is it a SYN packet and are we LISTENing?
811 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
812 // If we don't want to accept it, send a RST back
813 if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
818 // Try to allocate memory, otherwise send a RST back
819 c = allocate_connection(utcp, hdr.dst, hdr.src);
825 // Return SYN+ACK, go to SYN_RECEIVED state
826 c->snd.wnd = hdr.wnd;
827 c->rcv.irs = hdr.seq;
828 c->rcv.nxt = c->rcv.irs + 1;
829 set_state(c, SYN_RECEIVED);
833 hdr.ack = c->rcv.irs + 1;
834 hdr.seq = c->snd.iss;
836 print_packet(c->utcp, "send", &hdr, sizeof hdr);
837 utcp->send(utcp, &hdr, sizeof hdr);
839 // No, we don't want your packets, send a RST back
847 debug("%p state %s\n", c->utcp, strstate[c->state]);
849 // In case this is for a CLOSED connection, ignore the packet.
850 // TODO: make it so incoming packets can never match a CLOSED connection.
852 if(c->state == CLOSED) {
853 debug("Got packet for closed connection\n");
857 // It is for an existing connection.
859 uint32_t prevrcvnxt = c->rcv.nxt;
861 // 1. Drop invalid packets.
863 // 1a. Drop packets that should not happen in our current state.
883 // 1b. Drop packets with a sequence number not in our receive window.
887 if(c->state == SYN_SENT)
890 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
892 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
894 // cut already accepted front overlapping
896 acceptable = rcv_offset + len >= 0;
903 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
907 debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
908 // Ignore unacceptable RST packets.
911 // Otherwise, send an ACK back in the hope things improve.
916 c->snd.wnd = hdr.wnd; // TODO: move below
918 // 1c. Drop packets with an invalid ACK.
919 // ackno should not roll back, and it should also not be bigger than what we ever could have sent
920 // (= snd.una + c->sndbuf.used).
922 if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
923 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
924 // Ignore unacceptable RST packets.
930 // 2. Handle RST packets
937 // The peer has refused our connection.
938 set_state(c, CLOSED);
939 errno = ECONNREFUSED;
946 // We haven't told the application about this connection yet. Silently delete.
955 // The peer has aborted our connection.
956 set_state(c, CLOSED);
966 // As far as the application is concerned, the connection has already been closed.
967 // If it has called utcp_close() already, we can immediately free this connection.
972 // Otherwise, immediately move to the CLOSED state.
973 set_state(c, CLOSED);
983 // 3. Advance snd.una
985 uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
986 prevrcvnxt = c->rcv.nxt;
990 if(c->rtt_start.tv_sec) {
991 if(c->rtt_seq == hdr.ack) {
992 struct timeval now, diff;
993 gettimeofday(&now, NULL);
994 timersub(&now, &c->rtt_start, &diff);
995 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
996 c->rtt_start.tv_sec = 0;
997 } else if(c->rtt_seq < hdr.ack) {
998 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
999 c->rtt_start.tv_sec = 0;
1003 int32_t data_acked = advanced;
1010 // TODO: handle FIN as well.
1015 assert(data_acked >= 0);
1017 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1018 assert(data_acked <= bufused);
1021 buffer_get(&c->sndbuf, NULL, data_acked);
1023 // Also advance snd.nxt if possible
1024 if(seqdiff(c->snd.nxt, hdr.ack) < 0)
1025 c->snd.nxt = hdr.ack;
1027 c->snd.una = hdr.ack;
1030 c->snd.cwnd += utcp->mtu;
1031 if(c->snd.cwnd > c->sndbuf.maxsize)
1032 c->snd.cwnd = c->sndbuf.maxsize;
1034 // Check if we have sent a FIN that is now ACKed.
1037 if(c->snd.una == c->snd.last)
1038 set_state(c, FIN_WAIT_2);
1041 if(c->snd.una == c->snd.last) {
1042 gettimeofday(&c->conn_timeout, NULL);
1043 c->conn_timeout.tv_sec += 60;
1044 set_state(c, TIME_WAIT);
1053 if(c->dupack == 3) {
1054 debug("Triplicate ACK\n");
1055 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1056 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1057 //Reset the congestion window so we wait for ACKs.
1058 c->snd.nxt = c->snd.una;
1059 c->snd.cwnd = utcp->mtu;
1060 start_retransmit_timer(c);
1068 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
1069 if(c->snd.una == c->snd.last)
1070 stop_retransmit_timer(c);
1072 start_retransmit_timer(c);
1075 // 5. Process SYN stuff
1080 // This is a SYNACK. It should always have ACKed the SYN.
1083 c->rcv.irs = hdr.seq;
1084 c->rcv.nxt = hdr.seq;
1085 set_state(c, ESTABLISHED);
1086 // TODO: notify application of this somehow.
1096 // Ehm, no. We should never receive a second SYN.
1105 // SYN counts as one sequence number
1109 // 6. Process new data
1111 if(c->state == SYN_RECEIVED) {
1112 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1116 // Are we still LISTENing?
1118 utcp->accept(c, c->src);
1120 if(c->state != ESTABLISHED) {
1121 set_state(c, CLOSED);
1131 // This should never happen.
1144 // Ehm no, We should never receive more data after a FIN.
1153 handle_incoming_data(c, hdr.seq, data, len);
1156 // 7. Process FIN stuff
1158 if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
1162 // This should never happen.
1168 set_state(c, CLOSE_WAIT);
1171 set_state(c, CLOSING);
1174 gettimeofday(&c->conn_timeout, NULL);
1175 c->conn_timeout.tv_sec += 60;
1176 set_state(c, TIME_WAIT);
1182 // Ehm, no. We should never receive a second FIN.
1191 // FIN counts as one sequence number
1195 // Inform the application that the peer closed the connection.
1198 c->recv(c, NULL, 0);
1202 // Now we send something back if:
1203 // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
1204 // -> sendatleastone = true
1205 // - or we got an ack, so we should maybe send a bit more data
1206 // -> sendatleastone = false
1208 ack(c, len || prevrcvnxt != c->rcv.nxt);
1218 hdr.ack = hdr.seq + len;
1220 hdr.ctl = RST | ACK;
1222 print_packet(utcp, "send", &hdr, sizeof hdr);
1223 utcp->send(utcp, &hdr, sizeof hdr);
1228 int utcp_shutdown(struct utcp_connection *c, int dir) {
1229 debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1236 debug("Error: shutdown() called on closed connection %p\n", c);
1241 if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1246 // TCP does not have a provision for stopping incoming packets.
1247 // The best we can do is to just ignore them.
1248 if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR)
1251 // The rest of the code deals with shutting down writes.
1252 if(dir == UTCP_SHUT_RD)
1262 set_state(c, CLOSED);
1267 set_state(c, FIN_WAIT_1);
1273 set_state(c, CLOSING);
1285 if(!timerisset(&c->rtrx_timeout))
1286 start_retransmit_timer(c);
1290 int utcp_close(struct utcp_connection *c) {
1291 if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN)
1299 int utcp_abort(struct utcp_connection *c) {
1306 debug("Error: abort() called on closed connection %p\n", c);
1323 set_state(c, CLOSED);
1331 set_state(c, CLOSED);
1341 hdr.seq = c->snd.nxt;
1346 print_packet(c->utcp, "send", &hdr, sizeof hdr);
1347 c->utcp->send(c->utcp, &hdr, sizeof hdr);
1352 * One call to this function will loop through all connections,
1353 * checking if something needs to be resent or not.
1354 * The return value is the time to the next timeout in milliseconds,
1355 * or maybe a negative value if the timeout is infinite.
1357 struct timeval utcp_timeout(struct utcp *utcp) {
1359 gettimeofday(&now, NULL);
1360 struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1362 for(int i = 0; i < utcp->nconnections; i++) {
1363 struct utcp_connection *c = utcp->connections[i];
1367 // delete connections that have been utcp_close()d.
1368 if(c->state == CLOSED) {
1370 debug("Reaping %p\n", c);
1377 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1381 c->recv(c, NULL, 0);
1385 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1386 debug("retransmit()\n");
1390 if(c->poll && buffer_free(&c->sndbuf) && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1391 c->poll(c, buffer_free(&c->sndbuf));
1393 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
1394 next = c->conn_timeout;
1396 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
1397 next = c->rtrx_timeout;
1400 struct timeval diff;
1401 timersub(&next, &now, &diff);
1405 bool utcp_is_active(struct utcp *utcp) {
1409 for(int i = 0; i < utcp->nconnections; i++)
1410 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT)
1416 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1422 struct utcp *utcp = calloc(1, sizeof *utcp);
1426 utcp->accept = accept;
1427 utcp->pre_accept = pre_accept;
1430 utcp->mtu = DEFAULT_MTU;
1431 utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1432 utcp->rto = START_RTO; // usec
1437 void utcp_exit(struct utcp *utcp) {
1440 for(int i = 0; i < utcp->nconnections; i++) {
1441 if(!utcp->connections[i]->reapable)
1442 debug("Warning, freeing unclosed connection %p\n", utcp->connections[i]);
1443 buffer_exit(&utcp->connections[i]->rcvbuf);
1444 buffer_exit(&utcp->connections[i]->sndbuf);
1445 free(utcp->connections[i]);
1447 free(utcp->connections);
1451 uint16_t utcp_get_mtu(struct utcp *utcp) {
1452 return utcp ? utcp->mtu : 0;
1455 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1456 // TODO: handle overhead of the header
1461 int utcp_get_user_timeout(struct utcp *u) {
1462 return u ? u->timeout : 0;
1465 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1467 u->timeout = timeout;
1470 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1471 return c ? c->sndbuf.maxsize : 0;
1474 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1475 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1476 return buffer_free(&c->sndbuf);
1481 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1484 c->sndbuf.maxsize = size;
1485 if(c->sndbuf.maxsize != size)
1486 c->sndbuf.maxsize = -1;
1489 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1490 return c ? c->rcvbuf.maxsize : 0;
1493 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1494 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1495 return buffer_free(&c->rcvbuf);
1500 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1503 c->rcvbuf.maxsize = size;
1504 if(c->rcvbuf.maxsize != size)
1505 c->rcvbuf.maxsize = -1;
1508 bool utcp_get_nodelay(struct utcp_connection *c) {
1509 return c ? c->nodelay : false;
1512 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1514 c->nodelay = nodelay;
1517 bool utcp_get_keepalive(struct utcp_connection *c) {
1518 return c ? c->keepalive : false;
1521 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1523 c->keepalive = keepalive;
1526 size_t utcp_get_outq(struct utcp_connection *c) {
1527 return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
1530 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
1535 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
1540 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
1542 utcp->accept = accept;
1543 utcp->pre_accept = pre_accept;