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 += 1000000;\
57 #define max(a, b) ((a) > (b) ? (a) : (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 (%zu bytes)\n", utcp, dir, len);
77 memcpy(&hdr, pkt, sizeof hdr);
78 fprintf (stderr, "%p %s: len=%zu, src=%u dst=%u seq=%u ack=%u wnd=%u ctl=", utcp, dir, len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd);
88 if(len > sizeof hdr) {
89 uint32_t datalen = len - sizeof hdr;
90 uint8_t *str = malloc((datalen << 1) + 7);
91 memcpy(str, " data=", 6);
92 uint8_t *strptr = str + 6;
93 const uint8_t *data = pkt;
94 const uint8_t *dataend = data + datalen;
96 while(data != dataend) {
97 *strptr = (*data >> 4) > 9? (*data >> 4) + 55 : (*data >> 4) + 48;
99 *strptr = (*data & 0xf) > 9? (*data & 0xf) + 55 : (*data & 0xf) + 48;
113 #define print_packet(...)
116 static void set_state(struct utcp_connection *c, enum state state) {
118 if(state == ESTABLISHED)
119 timerclear(&c->conn_timeout);
120 debug("%p new state: %s\n", c->utcp, strstate[state]);
123 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
124 if(seq != c->snd.last)
136 static inline void list_connections(struct utcp *utcp) {
137 debug("%p has %d connections:\n", utcp, utcp->nconnections);
138 for(int i = 0; i < utcp->nconnections; i++)
139 debug(" %u -> %u state %s\n", utcp->connections[i]->src, utcp->connections[i]->dst, strstate[utcp->connections[i]->state]);
142 static int32_t seqdiff(uint32_t a, uint32_t b) {
147 // TODO: convert to ringbuffers to avoid memmove() operations.
149 // Store data into the buffer
150 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
151 if(buf->maxsize <= buf->used)
154 debug("buffer_put_at %zu %zu %zu\n", buf->used, offset, len);
156 size_t required = offset + len;
157 if(required > buf->maxsize) {
158 if(offset >= buf->maxsize)
161 len = buf->maxsize - offset;
162 required = buf->maxsize;
165 if(required > buf->size) {
166 size_t newsize = buf->size;
172 } while(newsize < buf->used + len);
174 if(newsize > buf->maxsize)
175 newsize = buf->maxsize;
176 char *newdata = realloc(buf->data, newsize);
183 memcpy(buf->data + offset, data, len);
184 if(required > buf->used)
185 buf->used = required;
189 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
190 return buffer_put_at(buf, buf->used, data, len);
193 // Get data from the buffer. data can be NULL.
194 static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
198 memcpy(data, buf->data, len);
200 memmove(buf->data, buf->data + len, buf->used - len);
205 // Copy data from the buffer without removing it.
206 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
207 if(offset >= buf->used)
209 if(offset + len > buf->used)
210 len = buf->used - offset;
211 memcpy(data, buf->data + offset, len);
215 static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
216 memset(buf, 0, sizeof *buf);
218 buf->data = malloc(len);
223 buf->maxsize = maxlen;
227 static void buffer_exit(struct buffer *buf) {
229 memset(buf, 0, sizeof *buf);
232 static uint32_t buffer_free(const struct buffer *buf) {
233 return buf->maxsize - buf->used;
236 // Connections are stored in a sorted list.
237 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
239 static int compare(const void *va, const void *vb) {
242 const struct utcp_connection *a = *(struct utcp_connection **)va;
243 const struct utcp_connection *b = *(struct utcp_connection **)vb;
246 assert(a->src && b->src);
248 int c = (int)a->src - (int)b->src;
251 c = (int)a->dst - (int)b->dst;
255 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
256 if(!utcp->nconnections)
258 struct utcp_connection key = {
262 struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
263 return match ? *match : NULL;
266 static void free_connection(struct utcp_connection *c) {
267 struct utcp *utcp = c->utcp;
268 struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
272 int i = cp - utcp->connections;
273 memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof *cp);
274 utcp->nconnections--;
276 buffer_exit(&c->rcvbuf);
277 buffer_exit(&c->sndbuf);
281 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
282 // Check whether this combination of src and dst is free
285 if(find_connection(utcp, src, dst)) {
289 } else { // If src == 0, generate a random port number with the high bit set
290 if(utcp->nconnections >= 32767) {
294 src = rand() | 0x8000;
295 while(find_connection(utcp, src, dst))
299 // Allocate memory for the new connection
301 if(utcp->nconnections >= utcp->nallocated) {
302 if(!utcp->nallocated)
303 utcp->nallocated = 4;
305 utcp->nallocated *= 2;
306 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof *utcp->connections);
309 utcp->connections = new_array;
312 struct utcp_connection *c = calloc(1, sizeof *c);
316 if(!buffer_init(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
321 if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
322 buffer_exit(&c->sndbuf);
327 // Fill in the details
336 c->snd.una = c->snd.iss;
337 c->snd.nxt = c->snd.iss + 1;
338 c->rcv.wnd = utcp->mtu;
339 c->snd.last = c->snd.nxt;
340 c->snd.cwnd = utcp->mtu;
343 // Add it to the sorted list of connections
345 utcp->connections[utcp->nconnections++] = c;
346 qsort(utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
351 // Update RTT variables. See RFC 6298.
352 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
354 debug("invalid rtt\n");
358 struct utcp *utcp = c->utcp;
362 utcp->rttvar = rtt / 2;
363 utcp->rto = rtt + max(2 * rtt, CLOCK_GRANULARITY);
365 utcp->rttvar = (utcp->rttvar * 3 + abs(utcp->srtt - rtt)) / 4;
366 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
367 utcp->rto = utcp->srtt + max(utcp->rttvar, CLOCK_GRANULARITY);
370 if(utcp->rto > MAX_RTO)
373 debug("rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
376 static void start_retransmit_timer(struct utcp_connection *c) {
377 gettimeofday(&c->rtrx_timeout, NULL);
378 c->rtrx_timeout.tv_usec += c->utcp->rto;
379 while(c->rtrx_timeout.tv_usec >= 1000000) {
380 c->rtrx_timeout.tv_usec -= 1000000;
381 c->rtrx_timeout.tv_sec++;
383 debug("timeout set to %lu.%06lu (%u)\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_usec, c->utcp->rto);
386 static void stop_retransmit_timer(struct utcp_connection *c) {
387 timerclear(&c->rtrx_timeout);
388 debug("timeout cleared\n");
391 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
392 struct utcp_connection *c = allocate_connection(utcp, 0, dst);
403 hdr.seq = c->snd.iss;
405 hdr.wnd = c->rcv.wnd;
409 set_state(c, SYN_SENT);
411 print_packet(utcp, "send", &hdr, sizeof hdr);
412 utcp->send(utcp, &hdr, sizeof hdr);
414 gettimeofday(&c->conn_timeout, NULL);
415 c->conn_timeout.tv_sec += utcp->timeout;
420 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
421 if(c->reapable || c->state != SYN_RECEIVED) {
422 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
426 debug("%p accepted, %p %p\n", c, recv, priv);
429 set_state(c, ESTABLISHED);
432 static void ack(struct utcp_connection *c, bool sendatleastone) {
433 int32_t left = seqdiff(c->snd.last, c->snd.nxt);
434 int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
435 debug("cwndleft = %d\n", cwndleft);
445 if(!left && !sendatleastone)
453 pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
457 pkt->hdr.src = c->src;
458 pkt->hdr.dst = c->dst;
459 pkt->hdr.ack = c->rcv.nxt;
460 pkt->hdr.wnd = c->snd.wnd;
465 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
466 pkt->hdr.seq = c->snd.nxt;
468 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
470 c->snd.nxt += seglen;
473 if(seglen && fin_wanted(c, c->snd.nxt)) {
478 if(!c->rtt_start.tv_sec) {
479 // Start RTT measurement
480 gettimeofday(&c->rtt_start, NULL);
481 c->rtt_seq = pkt->hdr.seq + seglen;
482 debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
485 print_packet(c->utcp, "send", pkt, sizeof pkt->hdr + seglen);
486 c->utcp->send(c->utcp, pkt, sizeof pkt->hdr + seglen);
492 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
494 debug("Error: send() called on closed connection %p\n", c);
504 debug("Error: send() called on unconnected connection %p\n", c);
515 debug("Error: send() called on closing connection %p\n", c);
520 // Add data to send buffer
530 len = buffer_put(&c->sndbuf, data, len);
538 if(!timerisset(&c->rtrx_timeout))
539 start_retransmit_timer(c);
543 static void swap_ports(struct hdr *hdr) {
544 uint16_t tmp = hdr->src;
549 static void retransmit(struct utcp_connection *c) {
550 if(c->state == CLOSED || c->snd.last == c->snd.una) {
551 debug("Retransmit() called but nothing to retransmit!\n");
552 stop_retransmit_timer(c);
556 struct utcp *utcp = c->utcp;
563 pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
567 pkt->hdr.src = c->src;
568 pkt->hdr.dst = c->dst;
569 pkt->hdr.wnd = c->rcv.wnd;
574 // Send our SYN again
575 pkt->hdr.seq = c->snd.iss;
578 print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
579 utcp->send(utcp, pkt, sizeof pkt->hdr);
584 pkt->hdr.seq = c->snd.nxt;
585 pkt->hdr.ack = c->rcv.nxt;
586 pkt->hdr.ctl = SYN | ACK;
587 print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
588 utcp->send(utcp, pkt, sizeof pkt->hdr);
596 // Send unacked data again.
597 pkt->hdr.seq = c->snd.una;
598 pkt->hdr.ack = c->rcv.nxt;
600 uint32_t len = seqdiff(c->snd.last, c->snd.una);
603 if(fin_wanted(c, c->snd.una + len)) {
607 c->snd.nxt = c->snd.una + len;
608 c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
609 buffer_copy(&c->sndbuf, pkt->data, 0, len);
610 print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr + len);
611 utcp->send(utcp, pkt, sizeof pkt->hdr + len);
618 // We shouldn't need to retransmit anything in this state.
622 stop_retransmit_timer(c);
626 start_retransmit_timer(c);
628 if(utcp->rto > MAX_RTO)
630 c->rtt_start.tv_sec = 0; // invalidate RTT timer
636 // Update receive buffer and SACK entries after consuming data.
637 static void sack_consume(struct utcp_connection *c, size_t len) {
638 debug("sack_consume %zu\n", len);
639 if(len > c->rcvbuf.used)
642 buffer_get(&c->rcvbuf, NULL, len);
644 for(int i = 0; i < NSACKS && c->sacks[i].len; ) {
645 if(len < c->sacks[i].offset) {
646 c->sacks[i].offset -= len;
648 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
649 c->sacks[i].offset = 0;
650 c->sacks[i].len -= len - c->sacks[i].offset;
654 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof c->sacks[i]);
655 c->sacks[i + 1].len = 0;
663 for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
664 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
667 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
668 debug("out of order packet, offset %u\n", offset);
669 // Packet loss or reordering occured. Store the data in the buffer.
670 ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
674 // Make note of where we put it.
675 for(int i = 0; i < NSACKS; i++) {
676 if(!c->sacks[i].len) { // nothing to merge, add new entry
677 debug("New SACK entry %d\n", i);
678 c->sacks[i].offset = offset;
679 c->sacks[i].len = rxd;
681 } else if(offset < c->sacks[i].offset) {
682 if(offset + rxd < c->sacks[i].offset) { // insert before
683 if(!c->sacks[NSACKS - 1].len) { // only if room left
684 debug("Insert SACK entry at %d\n", i);
685 memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof c->sacks[i]);
686 c->sacks[i].offset = offset;
687 c->sacks[i].len = rxd;
691 debug("Merge with start of SACK entry at %d\n", i);
692 c->sacks[i].offset = offset;
695 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
696 if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
697 debug("Merge with end of SACK entry at %d\n", i);
698 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
699 // TODO: handle potential merge with next entry
705 for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
706 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
709 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
710 // Check if we can process out-of-order data now.
711 if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
712 debug("incoming packet len %zu connected with SACK at %u\n", len, c->sacks[0].offset);
713 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
714 len = max(len, c->sacks[0].offset + c->sacks[0].len);
715 data = c->rcvbuf.data;
719 ssize_t rxd = c->recv(c, data, len);
721 // TODO: handle the application not accepting all data.
727 sack_consume(c, len);
733 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
734 uint32_t offset = seqdiff(seq, c->rcv.nxt);
735 if(offset + len > c->rcvbuf.maxsize)
739 handle_out_of_order(c, offset, data, len);
741 handle_in_order(c, data, len);
745 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
759 print_packet(utcp, "recv", data, len);
761 // Drop packets smaller than the header
764 if(len < sizeof hdr) {
769 // Make a copy from the potentially unaligned data to a struct hdr
771 memcpy(&hdr, data, sizeof hdr);
775 // Drop packets with an unknown CTL flag
777 if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
782 // Try to match the packet to an existing connection
784 struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
786 // Is it for a new connection?
789 // Ignore RST packets
794 // Is it a SYN packet and are we LISTENing?
796 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
797 // If we don't want to accept it, send a RST back
798 if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
803 // Try to allocate memory, otherwise send a RST back
804 c = allocate_connection(utcp, hdr.dst, hdr.src);
810 // Return SYN+ACK, go to SYN_RECEIVED state
811 c->snd.wnd = hdr.wnd;
812 c->rcv.irs = hdr.seq;
813 c->rcv.nxt = c->rcv.irs + 1;
814 set_state(c, SYN_RECEIVED);
818 hdr.ack = c->rcv.irs + 1;
819 hdr.seq = c->snd.iss;
821 print_packet(c->utcp, "send", &hdr, sizeof hdr);
822 utcp->send(utcp, &hdr, sizeof hdr);
824 // No, we don't want your packets, send a RST back
832 debug("%p state %s\n", c->utcp, strstate[c->state]);
834 // In case this is for a CLOSED connection, ignore the packet.
835 // TODO: make it so incoming packets can never match a CLOSED connection.
837 if(c->state == CLOSED)
840 // It is for an existing connection.
842 uint32_t prevrcvnxt = c->rcv.nxt;
844 // 1. Drop invalid packets.
846 // 1a. Drop packets that should not happen in our current state.
866 // 1b. Drop packets with a sequence number not in our receive window.
870 if(c->state == SYN_SENT)
873 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
875 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
877 // cut already accepted front overlapping
879 acceptable = rcv_offset + len >= 0;
886 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
890 debug("Packet not acceptable, %u <= %u + %zu < %u\n", c->rcv.nxt, hdr.seq, len, c->rcv.nxt + c->rcvbuf.maxsize);
891 // Ignore unacceptable RST packets.
894 // Otherwise, send an ACK back in the hope things improve.
899 c->snd.wnd = hdr.wnd; // TODO: move below
901 // 1c. Drop packets with an invalid ACK.
902 // ackno should not roll back, and it should also not be bigger than what we ever could have sent
903 // (= snd.una + c->sndbuf.used).
905 if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
906 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
907 // Ignore unacceptable RST packets.
913 // 2. Handle RST packets
920 // The peer has refused our connection.
921 set_state(c, CLOSED);
922 errno = ECONNREFUSED;
929 // We haven't told the application about this connection yet. Silently delete.
938 // The peer has aborted our connection.
939 set_state(c, CLOSED);
949 // As far as the application is concerned, the connection has already been closed.
950 // If it has called utcp_close() already, we can immediately free this connection.
955 // Otherwise, immediately move to the CLOSED state.
956 set_state(c, CLOSED);
966 // 3. Advance snd.una
968 uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
969 prevrcvnxt = c->rcv.nxt;
973 if(c->rtt_start.tv_sec) {
974 if(c->rtt_seq == hdr.ack) {
975 struct timeval now, diff;
976 gettimeofday(&now, NULL);
977 timersub(&now, &c->rtt_start, &diff);
978 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
979 c->rtt_start.tv_sec = 0;
980 } else if(c->rtt_seq < hdr.ack) {
981 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
982 c->rtt_start.tv_sec = 0;
986 int32_t data_acked = advanced;
993 // TODO: handle FIN as well.
998 assert(data_acked >= 0);
1000 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1001 assert(data_acked <= bufused);
1004 buffer_get(&c->sndbuf, NULL, data_acked);
1006 // Also advance snd.nxt if possible
1007 if(seqdiff(c->snd.nxt, hdr.ack) < 0)
1008 c->snd.nxt = hdr.ack;
1010 c->snd.una = hdr.ack;
1013 c->snd.cwnd += utcp->mtu;
1014 if(c->snd.cwnd > c->sndbuf.maxsize)
1015 c->snd.cwnd = c->sndbuf.maxsize;
1017 // Check if we have sent a FIN that is now ACKed.
1020 if(c->snd.una == c->snd.last)
1021 set_state(c, FIN_WAIT_2);
1024 if(c->snd.una == c->snd.last) {
1025 gettimeofday(&c->conn_timeout, NULL);
1026 c->conn_timeout.tv_sec += 60;
1027 set_state(c, TIME_WAIT);
1036 if(c->dupack == 3) {
1037 debug("Triplicate ACK\n");
1038 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1039 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1040 //Reset the congestion window so we wait for ACKs.
1041 c->snd.nxt = c->snd.una;
1042 c->snd.cwnd = utcp->mtu;
1043 start_retransmit_timer(c);
1051 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
1052 if(c->snd.una == c->snd.last)
1053 stop_retransmit_timer(c);
1055 start_retransmit_timer(c);
1058 // 5. Process SYN stuff
1063 // This is a SYNACK. It should always have ACKed the SYN.
1066 c->rcv.irs = hdr.seq;
1067 c->rcv.nxt = hdr.seq;
1068 set_state(c, ESTABLISHED);
1069 // TODO: notify application of this somehow.
1079 // Ehm, no. We should never receive a second SYN.
1088 // SYN counts as one sequence number
1092 // 6. Process new data
1094 if(c->state == SYN_RECEIVED) {
1095 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1099 // Are we still LISTENing?
1101 utcp->accept(c, c->src);
1103 if(c->state != ESTABLISHED) {
1104 set_state(c, CLOSED);
1114 // This should never happen.
1127 // Ehm no, We should never receive more data after a FIN.
1136 handle_incoming_data(c, hdr.seq, data, len);
1139 // 7. Process FIN stuff
1141 if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
1145 // This should never happen.
1151 set_state(c, CLOSE_WAIT);
1154 set_state(c, CLOSING);
1157 gettimeofday(&c->conn_timeout, NULL);
1158 c->conn_timeout.tv_sec += 60;
1159 set_state(c, TIME_WAIT);
1165 // Ehm, no. We should never receive a second FIN.
1174 // FIN counts as one sequence number
1178 // Inform the application that the peer closed the connection.
1181 c->recv(c, NULL, 0);
1185 // Now we send something back if:
1186 // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
1187 // -> sendatleastone = true
1188 // - or we got an ack, so we should maybe send a bit more data
1189 // -> sendatleastone = false
1191 ack(c, len || prevrcvnxt != c->rcv.nxt);
1201 hdr.ack = hdr.seq + len;
1203 hdr.ctl = RST | ACK;
1205 print_packet(utcp, "send", &hdr, sizeof hdr);
1206 utcp->send(utcp, &hdr, sizeof hdr);
1211 int utcp_shutdown(struct utcp_connection *c, int dir) {
1212 debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1219 debug("Error: shutdown() called on closed connection %p\n", c);
1224 if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1229 // TCP does not have a provision for stopping incoming packets.
1230 // The best we can do is to just ignore them.
1231 if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR)
1234 // The rest of the code deals with shutting down writes.
1235 if(dir == UTCP_SHUT_RD)
1245 set_state(c, CLOSED);
1250 set_state(c, FIN_WAIT_1);
1256 set_state(c, CLOSING);
1268 if(!timerisset(&c->rtrx_timeout))
1269 start_retransmit_timer(c);
1273 int utcp_close(struct utcp_connection *c) {
1274 if(utcp_shutdown(c, SHUT_RDWR))
1282 int utcp_abort(struct utcp_connection *c) {
1289 debug("Error: abort() called on closed connection %p\n", c);
1306 set_state(c, CLOSED);
1314 set_state(c, CLOSED);
1324 hdr.seq = c->snd.nxt;
1329 print_packet(c->utcp, "send", &hdr, sizeof hdr);
1330 c->utcp->send(c->utcp, &hdr, sizeof hdr);
1335 * One call to this function will loop through all connections,
1336 * checking if something needs to be resent or not.
1337 * The return value is the time to the next timeout in milliseconds,
1338 * or maybe a negative value if the timeout is infinite.
1340 struct timeval utcp_timeout(struct utcp *utcp) {
1342 gettimeofday(&now, NULL);
1343 struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1345 for(int i = 0; i < utcp->nconnections; i++) {
1346 struct utcp_connection *c = utcp->connections[i];
1350 // delete connections that have been utcp_close()d.
1351 if(c->state == CLOSED) {
1353 debug("Reaping %p\n", c);
1360 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1364 c->recv(c, NULL, 0);
1368 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1369 debug("retransmit()\n");
1373 if(c->poll && buffer_free(&c->sndbuf) && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1374 c->poll(c, buffer_free(&c->sndbuf));
1376 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
1377 next = c->conn_timeout;
1379 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
1380 next = c->rtrx_timeout;
1383 struct timeval diff;
1384 timersub(&next, &now, &diff);
1388 bool utcp_is_active(struct utcp *utcp) {
1392 for(int i = 0; i < utcp->nconnections; i++)
1393 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT)
1399 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1405 struct utcp *utcp = calloc(1, sizeof *utcp);
1409 utcp->accept = accept;
1410 utcp->pre_accept = pre_accept;
1413 utcp->mtu = DEFAULT_MTU;
1414 utcp->timeout = DEFAULT_USER_TIMEOUT; // s
1415 utcp->rto = START_RTO; // us
1420 void utcp_exit(struct utcp *utcp) {
1423 for(int i = 0; i < utcp->nconnections; i++) {
1424 if(!utcp->connections[i]->reapable)
1425 debug("Warning, freeing unclosed connection %p\n", utcp->connections[i]);
1426 buffer_exit(&utcp->connections[i]->rcvbuf);
1427 buffer_exit(&utcp->connections[i]->sndbuf);
1428 free(utcp->connections[i]);
1430 free(utcp->connections);
1434 uint16_t utcp_get_mtu(struct utcp *utcp) {
1435 return utcp ? utcp->mtu : 0;
1438 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1439 // TODO: handle overhead of the header
1444 int utcp_get_user_timeout(struct utcp *u) {
1445 return u ? u->timeout : 0;
1448 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1450 u->timeout = timeout;
1453 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1454 return c ? c->sndbuf.maxsize : 0;
1457 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1458 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1459 return buffer_free(&c->sndbuf);
1464 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1467 c->sndbuf.maxsize = size;
1468 if(c->sndbuf.maxsize != size)
1469 c->sndbuf.maxsize = -1;
1472 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1473 return c ? c->rcvbuf.maxsize : 0;
1476 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1477 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1478 return buffer_free(&c->rcvbuf);
1483 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1486 c->rcvbuf.maxsize = size;
1487 if(c->rcvbuf.maxsize != size)
1488 c->rcvbuf.maxsize = -1;
1491 bool utcp_get_nodelay(struct utcp_connection *c) {
1492 return c ? c->nodelay : false;
1495 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1497 c->nodelay = nodelay;
1500 bool utcp_get_keepalive(struct utcp_connection *c) {
1501 return c ? c->keepalive : false;
1504 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1506 c->keepalive = keepalive;
1509 size_t utcp_get_outq(struct utcp_connection *c) {
1510 return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
1513 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
1518 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
1523 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
1525 utcp->accept = accept;
1526 utcp->pre_accept = pre_accept;