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 aux=%x ctl=", utcp, dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux);
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 bool is_reliable(struct utcp_connection *c) {
131 return c->flags & UTCP_RELIABLE;
134 static inline void list_connections(struct utcp *utcp) {
135 debug("%p has %d connections:\n", utcp, utcp->nconnections);
136 for(int i = 0; i < utcp->nconnections; i++)
137 debug(" %u -> %u state %s\n", utcp->connections[i]->src, utcp->connections[i]->dst, strstate[utcp->connections[i]->state]);
140 static int32_t seqdiff(uint32_t a, uint32_t b) {
145 // TODO: convert to ringbuffers to avoid memmove() operations.
147 // Store data into the buffer
148 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
149 debug("buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
151 size_t required = offset + len;
152 if(required > buf->maxsize) {
153 if(offset >= buf->maxsize)
155 len = buf->maxsize - offset;
156 required = buf->maxsize;
159 if(required > buf->size) {
160 size_t newsize = buf->size;
166 } while(newsize < required);
168 if(newsize > buf->maxsize)
169 newsize = buf->maxsize;
170 char *newdata = realloc(buf->data, newsize);
177 memcpy(buf->data + offset, data, len);
178 if(required > buf->used)
179 buf->used = required;
183 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
184 return buffer_put_at(buf, buf->used, data, len);
187 // Get data from the buffer. data can be NULL.
188 static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
192 memcpy(data, buf->data, len);
194 memmove(buf->data, buf->data + len, buf->used - len);
199 // Copy data from the buffer without removing it.
200 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
201 if(offset >= buf->used)
203 if(offset + len > buf->used)
204 len = buf->used - offset;
205 memcpy(data, buf->data + offset, len);
209 static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
210 memset(buf, 0, sizeof *buf);
212 buf->data = malloc(len);
217 buf->maxsize = maxlen;
221 static void buffer_exit(struct buffer *buf) {
223 memset(buf, 0, sizeof *buf);
226 static uint32_t buffer_free(const struct buffer *buf) {
227 return buf->maxsize - buf->used;
230 // Connections are stored in a sorted list.
231 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
233 static int compare(const void *va, const void *vb) {
236 const struct utcp_connection *a = *(struct utcp_connection **)va;
237 const struct utcp_connection *b = *(struct utcp_connection **)vb;
240 assert(a->src && b->src);
242 int c = (int)a->src - (int)b->src;
245 c = (int)a->dst - (int)b->dst;
249 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
250 if(!utcp->nconnections)
252 struct utcp_connection key = {
256 struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
257 return match ? *match : NULL;
260 static void free_connection(struct utcp_connection *c) {
261 struct utcp *utcp = c->utcp;
262 struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
266 int i = cp - utcp->connections;
267 memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof *cp);
268 utcp->nconnections--;
270 buffer_exit(&c->rcvbuf);
271 buffer_exit(&c->sndbuf);
275 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
276 // Check whether this combination of src and dst is free
279 if(find_connection(utcp, src, dst)) {
283 } else { // If src == 0, generate a random port number with the high bit set
284 if(utcp->nconnections >= 32767) {
288 src = rand() | 0x8000;
289 while(find_connection(utcp, src, dst))
293 // Allocate memory for the new connection
295 if(utcp->nconnections >= utcp->nallocated) {
296 if(!utcp->nallocated)
297 utcp->nallocated = 4;
299 utcp->nallocated *= 2;
300 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof *utcp->connections);
303 utcp->connections = new_array;
306 struct utcp_connection *c = calloc(1, sizeof *c);
310 if(!buffer_init(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
315 if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
316 buffer_exit(&c->sndbuf);
321 // Fill in the details
330 c->snd.una = c->snd.iss;
331 c->snd.nxt = c->snd.iss + 1;
332 c->rcv.wnd = utcp->mtu;
333 c->snd.last = c->snd.nxt;
334 c->snd.cwnd = utcp->mtu;
337 // Add it to the sorted list of connections
339 utcp->connections[utcp->nconnections++] = c;
340 qsort(utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
345 // Update RTT variables. See RFC 6298.
346 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
348 debug("invalid rtt\n");
352 struct utcp *utcp = c->utcp;
356 utcp->rttvar = rtt / 2;
357 utcp->rto = rtt + max(2 * rtt, CLOCK_GRANULARITY);
359 utcp->rttvar = (utcp->rttvar * 3 + abs(utcp->srtt - rtt)) / 4;
360 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
361 utcp->rto = utcp->srtt + max(utcp->rttvar, CLOCK_GRANULARITY);
364 if(utcp->rto > MAX_RTO)
367 debug("rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
370 static void start_retransmit_timer(struct utcp_connection *c) {
371 gettimeofday(&c->rtrx_timeout, NULL);
372 c->rtrx_timeout.tv_usec += c->utcp->rto;
373 while(c->rtrx_timeout.tv_usec >= 1000000) {
374 c->rtrx_timeout.tv_usec -= 1000000;
375 c->rtrx_timeout.tv_sec++;
377 debug("timeout set to %lu.%06lu (%u)\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_usec, c->utcp->rto);
380 static void stop_retransmit_timer(struct utcp_connection *c) {
381 timerclear(&c->rtrx_timeout);
382 debug("timeout cleared\n");
385 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
386 struct utcp_connection *c = allocate_connection(utcp, 0, dst);
390 assert((flags & ~0xf) == 0);
401 pkt.hdr.src = c->src;
402 pkt.hdr.dst = c->dst;
403 pkt.hdr.seq = c->snd.iss;
405 pkt.hdr.wnd = c->rcv.wnd;
407 pkt.hdr.aux = 0x0101;
411 pkt.init[3] = flags & 0x7;
413 set_state(c, SYN_SENT);
415 print_packet(utcp, "send", &pkt, sizeof pkt);
416 utcp->send(utcp, &pkt, sizeof pkt);
418 gettimeofday(&c->conn_timeout, NULL);
419 c->conn_timeout.tv_sec += utcp->timeout;
421 start_retransmit_timer(c);
426 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
427 return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
430 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
431 if(c->reapable || c->state != SYN_RECEIVED) {
432 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
436 debug("%p accepted, %p %p\n", c, recv, priv);
439 set_state(c, ESTABLISHED);
442 static void ack(struct utcp_connection *c, bool sendatleastone) {
443 int32_t left = seqdiff(c->snd.last, c->snd.nxt);
444 int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
445 debug("cwndleft = %d\n", cwndleft);
455 if(!left && !sendatleastone)
463 pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
467 pkt->hdr.src = c->src;
468 pkt->hdr.dst = c->dst;
469 pkt->hdr.ack = c->rcv.nxt;
470 pkt->hdr.wnd = c->snd.wnd;
475 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
476 pkt->hdr.seq = c->snd.nxt;
478 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
480 c->snd.nxt += seglen;
483 if(seglen && fin_wanted(c, c->snd.nxt)) {
488 if(!c->rtt_start.tv_sec) {
489 // Start RTT measurement
490 gettimeofday(&c->rtt_start, NULL);
491 c->rtt_seq = pkt->hdr.seq + seglen;
492 debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
495 print_packet(c->utcp, "send", pkt, sizeof pkt->hdr + seglen);
496 c->utcp->send(c->utcp, pkt, sizeof pkt->hdr + seglen);
502 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
504 debug("Error: send() called on closed connection %p\n", c);
514 debug("Error: send() called on unconnected connection %p\n", c);
525 debug("Error: send() called on closing connection %p\n", c);
530 // Exit early if we have nothing to send.
540 // Add data to send buffer.
542 len = buffer_put(&c->sndbuf, data, len);
550 if(!is_reliable(c)) {
551 c->snd.una = c->snd.nxt = c->snd.last;
552 buffer_get(&c->sndbuf, NULL, c->sndbuf.used);
554 if(is_reliable(c) && !timerisset(&c->rtrx_timeout))
555 start_retransmit_timer(c);
559 static void swap_ports(struct hdr *hdr) {
560 uint16_t tmp = hdr->src;
565 static void retransmit(struct utcp_connection *c) {
566 if(c->state == CLOSED || c->snd.last == c->snd.una) {
567 debug("Retransmit() called but nothing to retransmit!\n");
568 stop_retransmit_timer(c);
572 struct utcp *utcp = c->utcp;
579 pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
583 pkt->hdr.src = c->src;
584 pkt->hdr.dst = c->dst;
585 pkt->hdr.wnd = c->rcv.wnd;
590 // Send our SYN again
591 pkt->hdr.seq = c->snd.iss;
594 pkt->hdr.aux = 0x0101;
598 pkt->data[3] = c->flags & 0x7;
599 print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr + 4);
600 utcp->send(utcp, pkt, sizeof pkt->hdr + 4);
605 pkt->hdr.seq = c->snd.nxt;
606 pkt->hdr.ack = c->rcv.nxt;
607 pkt->hdr.ctl = SYN | ACK;
608 print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
609 utcp->send(utcp, pkt, sizeof pkt->hdr);
617 // Send unacked data again.
618 pkt->hdr.seq = c->snd.una;
619 pkt->hdr.ack = c->rcv.nxt;
621 uint32_t len = seqdiff(c->snd.last, c->snd.una);
624 if(fin_wanted(c, c->snd.una + len)) {
628 c->snd.nxt = c->snd.una + len;
629 c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
630 buffer_copy(&c->sndbuf, pkt->data, 0, len);
631 print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr + len);
632 utcp->send(utcp, pkt, sizeof pkt->hdr + len);
639 // We shouldn't need to retransmit anything in this state.
643 stop_retransmit_timer(c);
647 start_retransmit_timer(c);
649 if(utcp->rto > MAX_RTO)
651 c->rtt_start.tv_sec = 0; // invalidate RTT timer
657 /* Update receive buffer and SACK entries after consuming data.
661 * |.....0000..1111111111.....22222......3333|
664 * 0..3 represent the SACK entries. The ^ indicates up to which point we want
665 * to remove data from the receive buffer. The idea is to substract "len"
666 * from the offset of all the SACK entries, and then remove/cut down entries
667 * that are shifted to before the start of the receive buffer.
669 * There are three cases:
670 * - the SACK entry is after ^, in that case just change the offset.
671 * - the SACK entry starts before and ends after ^, so we have to
672 * change both its offset and size.
673 * - the SACK entry is completely before ^, in that case delete it.
675 static void sack_consume(struct utcp_connection *c, size_t len) {
676 debug("sack_consume %lu\n", (unsigned long)len);
677 if(len > c->rcvbuf.used) {
678 debug("All SACK entries consumed");
683 buffer_get(&c->rcvbuf, NULL, len);
685 for(int i = 0; i < NSACKS && c->sacks[i].len; ) {
686 if(len < c->sacks[i].offset) {
687 c->sacks[i].offset -= len;
689 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
690 c->sacks[i].len -= len - c->sacks[i].offset;
691 c->sacks[i].offset = 0;
695 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof c->sacks[i]);
696 c->sacks[NSACKS - 1].len = 0;
704 for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
705 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
708 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
709 debug("out of order packet, offset %u\n", offset);
710 // Packet loss or reordering occured. Store the data in the buffer.
711 ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
715 // Make note of where we put it.
716 for(int i = 0; i < NSACKS; i++) {
717 if(!c->sacks[i].len) { // nothing to merge, add new entry
718 debug("New SACK entry %d\n", i);
719 c->sacks[i].offset = offset;
720 c->sacks[i].len = rxd;
722 } else if(offset < c->sacks[i].offset) {
723 if(offset + rxd < c->sacks[i].offset) { // insert before
724 if(!c->sacks[NSACKS - 1].len) { // only if room left
725 debug("Insert SACK entry at %d\n", i);
726 memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof c->sacks[i]);
727 c->sacks[i].offset = offset;
728 c->sacks[i].len = rxd;
730 debug("SACK entries full, dropping packet\n");
734 debug("Merge with start of SACK entry at %d\n", i);
735 c->sacks[i].offset = offset;
738 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
739 if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
740 debug("Merge with end of SACK entry at %d\n", i);
741 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
742 // TODO: handle potential merge with next entry
748 for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
749 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
752 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
753 // Check if we can process out-of-order data now.
754 if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
755 debug("incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
756 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
757 len = max(len, c->sacks[0].offset + c->sacks[0].len);
758 data = c->rcvbuf.data;
762 ssize_t rxd = c->recv(c, data, len);
764 // TODO: handle the application not accepting all data.
770 sack_consume(c, len);
776 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
777 if(!is_reliable(c)) {
778 c->recv(c, data, len);
779 c->rcv.nxt = seq + len;
783 uint32_t offset = seqdiff(seq, c->rcv.nxt);
784 if(offset + len > c->rcvbuf.maxsize)
788 handle_out_of_order(c, offset, data, len);
790 handle_in_order(c, data, len);
794 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
808 print_packet(utcp, "recv", data, len);
810 // Drop packets smaller than the header
813 if(len < sizeof hdr) {
818 // Make a copy from the potentially unaligned data to a struct hdr
820 memcpy(&hdr, data, sizeof hdr);
824 // Drop packets with an unknown CTL flag
826 if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
831 // Check for auxiliary headers
833 const uint8_t *init = NULL;
835 uint16_t aux = hdr.aux;
837 size_t auxlen = 4 * (aux >> 8) & 0xf;
838 uint8_t auxtype = aux & 0xff;
847 if(!(hdr.ctl & SYN) || auxlen != 4) {
869 memcpy(&aux, data, 2);
874 // Try to match the packet to an existing connection
876 struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
878 // Is it for a new connection?
881 // Ignore RST packets
886 // Is it a SYN packet and are we LISTENing?
888 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
889 // If we don't want to accept it, send a RST back
890 if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
895 // Try to allocate memory, otherwise send a RST back
896 c = allocate_connection(utcp, hdr.dst, hdr.src);
902 // Parse auxilliary information
908 c->flags = init[3] & 0x7;
913 // Return SYN+ACK, go to SYN_RECEIVED state
914 c->snd.wnd = hdr.wnd;
915 c->rcv.irs = hdr.seq;
916 c->rcv.nxt = c->rcv.irs + 1;
917 set_state(c, SYN_RECEIVED);
924 pkt.hdr.src = c->src;
925 pkt.hdr.dst = c->dst;
926 pkt.hdr.ack = c->rcv.irs + 1;
927 pkt.hdr.seq = c->snd.iss;
928 pkt.hdr.wnd = c->rcv.wnd;
929 pkt.hdr.ctl = SYN | ACK;
931 pkt.hdr.aux = 0x0101;
935 pkt.data[3] = c->flags & 0x7;
936 print_packet(c->utcp, "send", &pkt, sizeof hdr + 4);
937 utcp->send(utcp, &pkt, sizeof hdr + 4);
940 print_packet(c->utcp, "send", &pkt, sizeof hdr);
941 utcp->send(utcp, &pkt, sizeof hdr);
944 // No, we don't want your packets, send a RST back
952 debug("%p state %s\n", c->utcp, strstate[c->state]);
954 // In case this is for a CLOSED connection, ignore the packet.
955 // TODO: make it so incoming packets can never match a CLOSED connection.
957 if(c->state == CLOSED) {
958 debug("Got packet for closed connection\n");
962 // It is for an existing connection.
964 uint32_t prevrcvnxt = c->rcv.nxt;
966 // 1. Drop invalid packets.
968 // 1a. Drop packets that should not happen in our current state.
988 // 1b. Drop packets with a sequence number not in our receive window.
992 if(c->state == SYN_SENT)
995 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
997 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
999 // cut already accepted front overlapping
1000 if(rcv_offset < 0) {
1001 acceptable = len > -rcv_offset;
1005 hdr.seq -= rcv_offset;
1008 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1013 debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1014 // Ignore unacceptable RST packets.
1017 // Otherwise, continue processing.
1021 c->snd.wnd = hdr.wnd; // TODO: move below
1023 // 1c. Drop packets with an invalid ACK.
1024 // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1025 // (= snd.una + c->sndbuf.used).
1027 if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1028 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1029 // Ignore unacceptable RST packets.
1035 // 2. Handle RST packets
1040 if(!(hdr.ctl & ACK))
1042 // The peer has refused our connection.
1043 set_state(c, CLOSED);
1044 errno = ECONNREFUSED;
1046 c->recv(c, NULL, 0);
1051 // We haven't told the application about this connection yet. Silently delete.
1060 // The peer has aborted our connection.
1061 set_state(c, CLOSED);
1064 c->recv(c, NULL, 0);
1071 // As far as the application is concerned, the connection has already been closed.
1072 // If it has called utcp_close() already, we can immediately free this connection.
1077 // Otherwise, immediately move to the CLOSED state.
1078 set_state(c, CLOSED);
1088 if(!(hdr.ctl & ACK))
1091 // 3. Advance snd.una
1093 uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
1094 prevrcvnxt = c->rcv.nxt;
1098 if(c->rtt_start.tv_sec) {
1099 if(c->rtt_seq == hdr.ack) {
1100 struct timeval now, diff;
1101 gettimeofday(&now, NULL);
1102 timersub(&now, &c->rtt_start, &diff);
1103 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
1104 c->rtt_start.tv_sec = 0;
1105 } else if(c->rtt_seq < hdr.ack) {
1106 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1107 c->rtt_start.tv_sec = 0;
1111 int32_t data_acked = advanced;
1118 // TODO: handle FIN as well.
1123 assert(data_acked >= 0);
1125 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1126 assert(data_acked <= bufused);
1129 buffer_get(&c->sndbuf, NULL, data_acked);
1131 // Also advance snd.nxt if possible
1132 if(seqdiff(c->snd.nxt, hdr.ack) < 0)
1133 c->snd.nxt = hdr.ack;
1135 c->snd.una = hdr.ack;
1138 c->snd.cwnd += utcp->mtu;
1139 if(c->snd.cwnd > c->sndbuf.maxsize)
1140 c->snd.cwnd = c->sndbuf.maxsize;
1142 // Check if we have sent a FIN that is now ACKed.
1145 if(c->snd.una == c->snd.last)
1146 set_state(c, FIN_WAIT_2);
1149 if(c->snd.una == c->snd.last) {
1150 gettimeofday(&c->conn_timeout, NULL);
1151 c->conn_timeout.tv_sec += 60;
1152 set_state(c, TIME_WAIT);
1159 if(!len && is_reliable(c)) {
1161 if(c->dupack == 3) {
1162 debug("Triplicate ACK\n");
1163 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1164 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1165 //Reset the congestion window so we wait for ACKs.
1166 c->snd.nxt = c->snd.una;
1167 c->snd.cwnd = utcp->mtu;
1168 start_retransmit_timer(c);
1176 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
1177 if(c->snd.una == c->snd.last)
1178 stop_retransmit_timer(c);
1179 else if(is_reliable(c))
1180 start_retransmit_timer(c);
1184 // 5. Process SYN stuff
1189 // This is a SYNACK. It should always have ACKed the SYN.
1192 c->rcv.irs = hdr.seq;
1193 c->rcv.nxt = hdr.seq;
1194 set_state(c, ESTABLISHED);
1195 // TODO: notify application of this somehow.
1205 // Ehm, no. We should never receive a second SYN.
1214 // SYN counts as one sequence number
1218 // 6. Process new data
1220 if(c->state == SYN_RECEIVED) {
1221 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1225 // Are we still LISTENing?
1227 utcp->accept(c, c->src);
1229 if(c->state != ESTABLISHED) {
1230 set_state(c, CLOSED);
1240 // This should never happen.
1253 // Ehm no, We should never receive more data after a FIN.
1262 handle_incoming_data(c, hdr.seq, data, len);
1265 // 7. Process FIN stuff
1267 if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
1271 // This should never happen.
1277 set_state(c, CLOSE_WAIT);
1280 set_state(c, CLOSING);
1283 gettimeofday(&c->conn_timeout, NULL);
1284 c->conn_timeout.tv_sec += 60;
1285 set_state(c, TIME_WAIT);
1291 // Ehm, no. We should never receive a second FIN.
1300 // FIN counts as one sequence number
1304 // Inform the application that the peer closed the connection.
1307 c->recv(c, NULL, 0);
1311 // Now we send something back if:
1312 // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
1313 // -> sendatleastone = true
1314 // - or we got an ack, so we should maybe send a bit more data
1315 // -> sendatleastone = false
1317 ack(c, len || prevrcvnxt != c->rcv.nxt);
1328 hdr.ack = hdr.seq + len;
1330 hdr.ctl = RST | ACK;
1332 print_packet(utcp, "send", &hdr, sizeof hdr);
1333 utcp->send(utcp, &hdr, sizeof hdr);
1338 int utcp_shutdown(struct utcp_connection *c, int dir) {
1339 debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1346 debug("Error: shutdown() called on closed connection %p\n", c);
1351 if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1356 // TCP does not have a provision for stopping incoming packets.
1357 // The best we can do is to just ignore them.
1358 if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR)
1361 // The rest of the code deals with shutting down writes.
1362 if(dir == UTCP_SHUT_RD)
1372 set_state(c, CLOSED);
1377 set_state(c, FIN_WAIT_1);
1383 set_state(c, CLOSING);
1395 if(!timerisset(&c->rtrx_timeout))
1396 start_retransmit_timer(c);
1400 int utcp_close(struct utcp_connection *c) {
1401 if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN)
1409 int utcp_abort(struct utcp_connection *c) {
1416 debug("Error: abort() called on closed connection %p\n", c);
1433 set_state(c, CLOSED);
1441 set_state(c, CLOSED);
1451 hdr.seq = c->snd.nxt;
1456 print_packet(c->utcp, "send", &hdr, sizeof hdr);
1457 c->utcp->send(c->utcp, &hdr, sizeof hdr);
1462 * One call to this function will loop through all connections,
1463 * checking if something needs to be resent or not.
1464 * The return value is the time to the next timeout in milliseconds,
1465 * or maybe a negative value if the timeout is infinite.
1467 struct timeval utcp_timeout(struct utcp *utcp) {
1469 gettimeofday(&now, NULL);
1470 struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1472 for(int i = 0; i < utcp->nconnections; i++) {
1473 struct utcp_connection *c = utcp->connections[i];
1477 // delete connections that have been utcp_close()d.
1478 if(c->state == CLOSED) {
1480 debug("Reaping %p\n", c);
1487 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1491 c->recv(c, NULL, 0);
1495 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1496 debug("retransmit()\n");
1501 if((c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1502 uint32_t len = buffer_free(&c->sndbuf);
1505 } else if(c->state == CLOSED) {
1510 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
1511 next = c->conn_timeout;
1513 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
1514 next = c->rtrx_timeout;
1517 struct timeval diff;
1518 timersub(&next, &now, &diff);
1522 bool utcp_is_active(struct utcp *utcp) {
1526 for(int i = 0; i < utcp->nconnections; i++)
1527 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT)
1533 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1539 struct utcp *utcp = calloc(1, sizeof *utcp);
1543 utcp->accept = accept;
1544 utcp->pre_accept = pre_accept;
1547 utcp->mtu = DEFAULT_MTU;
1548 utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1549 utcp->rto = START_RTO; // usec
1554 void utcp_exit(struct utcp *utcp) {
1557 for(int i = 0; i < utcp->nconnections; i++) {
1558 struct utcp_connection *c = utcp->connections[i];
1561 c->recv(c, NULL, 0);
1562 buffer_exit(&c->rcvbuf);
1563 buffer_exit(&c->sndbuf);
1566 free(utcp->connections);
1570 uint16_t utcp_get_mtu(struct utcp *utcp) {
1571 return utcp ? utcp->mtu : 0;
1574 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1575 // TODO: handle overhead of the header
1580 void utcp_reset_timers(struct utcp *utcp) {
1583 struct timeval now, then;
1584 gettimeofday(&now, NULL);
1586 then.tv_sec += utcp->timeout;
1587 for(int i = 0; i < utcp->nconnections; i++) {
1588 utcp->connections[i]->rtrx_timeout = now;
1589 utcp->connections[i]->conn_timeout = then;
1590 utcp->connections[i]->rtt_start.tv_sec = 0;
1592 if(utcp->rto > START_RTO)
1593 utcp->rto = START_RTO;
1596 int utcp_get_user_timeout(struct utcp *u) {
1597 return u ? u->timeout : 0;
1600 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1602 u->timeout = timeout;
1605 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1606 return c ? c->sndbuf.maxsize : 0;
1609 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1610 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1611 return buffer_free(&c->sndbuf);
1616 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1619 c->sndbuf.maxsize = size;
1620 if(c->sndbuf.maxsize != size)
1621 c->sndbuf.maxsize = -1;
1624 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1625 return c ? c->rcvbuf.maxsize : 0;
1628 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1629 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1630 return buffer_free(&c->rcvbuf);
1635 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1638 c->rcvbuf.maxsize = size;
1639 if(c->rcvbuf.maxsize != size)
1640 c->rcvbuf.maxsize = -1;
1643 bool utcp_get_nodelay(struct utcp_connection *c) {
1644 return c ? c->nodelay : false;
1647 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1649 c->nodelay = nodelay;
1652 bool utcp_get_keepalive(struct utcp_connection *c) {
1653 return c ? c->keepalive : false;
1656 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1658 c->keepalive = keepalive;
1661 size_t utcp_get_outq(struct utcp_connection *c) {
1662 return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
1665 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
1670 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
1675 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
1677 utcp->accept = accept;
1678 utcp->pre_accept = pre_accept;