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) {
90 for(int i = sizeof hdr; i < len; i++) {
91 const char *data = pkt;
92 debug("%c", data[i] >= 32 ? data[i] : '.');
100 #define print_packet(...)
103 static void set_state(struct utcp_connection *c, enum state state) {
105 if(state == ESTABLISHED)
106 timerclear(&c->conn_timeout);
107 debug("%p new state: %s\n", c->utcp, strstate[state]);
110 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
111 if(seq != c->snd.last)
123 static inline void list_connections(struct utcp *utcp) {
124 debug("%p has %d connections:\n", utcp, utcp->nconnections);
125 for(int i = 0; i < utcp->nconnections; i++)
126 debug(" %u -> %u state %s\n", utcp->connections[i]->src, utcp->connections[i]->dst, strstate[utcp->connections[i]->state]);
129 static int32_t seqdiff(uint32_t a, uint32_t b) {
134 // TODO: convert to ringbuffers to avoid memmove() operations.
136 // Store data into the buffer
137 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
138 if(buf->maxsize <= buf->used)
141 debug("buffer_put_at %zu %zu %zu\n", buf->used, offset, len);
143 size_t required = offset + len;
144 if(required > buf->maxsize) {
145 if(offset >= buf->maxsize)
148 len = buf->maxsize - offset;
149 required = buf->maxsize;
152 if(required > buf->size) {
153 size_t newsize = buf->size;
159 } while(newsize < buf->used + len);
161 if(newsize > buf->maxsize)
162 newsize = buf->maxsize;
163 char *newdata = realloc(buf->data, newsize);
170 memcpy(buf->data + offset, data, len);
171 if(required > buf->used)
172 buf->used = required;
176 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
177 return buffer_put_at(buf, buf->used, data, len);
180 // Get data from the buffer. data can be NULL.
181 static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
185 memcpy(data, buf->data, len);
187 memmove(buf->data, buf->data + len, buf->used - len);
192 // Copy data from the buffer without removing it.
193 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
194 if(offset >= buf->used)
196 if(offset + len > buf->used)
197 len = buf->used - offset;
198 memcpy(data, buf->data + offset, len);
202 static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
203 memset(buf, 0, sizeof *buf);
205 buf->data = malloc(len);
210 buf->maxsize = maxlen;
214 static void buffer_exit(struct buffer *buf) {
216 memset(buf, 0, sizeof *buf);
219 static uint32_t buffer_free(const struct buffer *buf) {
220 return buf->maxsize - buf->used;
223 // Connections are stored in a sorted list.
224 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
226 static int compare(const void *va, const void *vb) {
229 const struct utcp_connection *a = *(struct utcp_connection **)va;
230 const struct utcp_connection *b = *(struct utcp_connection **)vb;
233 assert(a->src && b->src);
235 int c = (int)a->src - (int)b->src;
238 c = (int)a->dst - (int)b->dst;
242 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
243 if(!utcp->nconnections)
245 struct utcp_connection key = {
249 struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
250 return match ? *match : NULL;
253 static void free_connection(struct utcp_connection *c) {
254 struct utcp *utcp = c->utcp;
255 struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
259 int i = cp - utcp->connections;
260 memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof *cp);
261 utcp->nconnections--;
263 buffer_exit(&c->sndbuf);
267 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
268 // Check whether this combination of src and dst is free
271 if(find_connection(utcp, src, dst)) {
275 } else { // If src == 0, generate a random port number with the high bit set
276 if(utcp->nconnections >= 32767) {
280 src = rand() | 0x8000;
281 while(find_connection(utcp, src, dst))
285 // Allocate memory for the new connection
287 if(utcp->nconnections >= utcp->nallocated) {
288 if(!utcp->nallocated)
289 utcp->nallocated = 4;
291 utcp->nallocated *= 2;
292 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof *utcp->connections);
295 utcp->connections = new_array;
298 struct utcp_connection *c = calloc(1, sizeof *c);
302 if(!buffer_init(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
307 if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
312 // Fill in the details
321 c->snd.una = c->snd.iss;
322 c->snd.nxt = c->snd.iss + 1;
323 c->rcv.wnd = utcp->mtu;
324 c->snd.last = c->snd.nxt;
325 c->snd.cwnd = utcp->mtu;
328 // Add it to the sorted list of connections
330 utcp->connections[utcp->nconnections++] = c;
331 qsort(utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
336 // Update RTT variables. See RFC 6298.
337 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
339 debug("invalid rtt\n");
343 struct utcp *utcp = c->utcp;
347 utcp->rttvar = rtt / 2;
348 utcp->rto = rtt + max(2 * rtt, CLOCK_GRANULARITY);
350 utcp->rttvar = (utcp->rttvar * 3 + abs(utcp->srtt - rtt)) / 4;
351 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
352 utcp->rto = utcp->srtt + max(utcp->rttvar, CLOCK_GRANULARITY);
355 if(utcp->rto > MAX_RTO)
358 debug("rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
361 static void start_retransmit_timer(struct utcp_connection *c) {
362 gettimeofday(&c->rtrx_timeout, NULL);
363 c->rtrx_timeout.tv_usec += c->utcp->rto;
364 while(c->rtrx_timeout.tv_usec >= 1000000) {
365 c->rtrx_timeout.tv_usec -= 1000000;
366 c->rtrx_timeout.tv_sec++;
368 debug("timeout set to %lu.%06lu (%u)\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_usec, c->utcp->rto);
371 static void stop_retransmit_timer(struct utcp_connection *c) {
372 timerclear(&c->rtrx_timeout);
373 debug("timeout cleared\n");
376 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
377 struct utcp_connection *c = allocate_connection(utcp, 0, dst);
388 hdr.seq = c->snd.iss;
390 hdr.wnd = c->rcv.wnd;
394 set_state(c, SYN_SENT);
396 print_packet(utcp, "send", &hdr, sizeof hdr);
397 utcp->send(utcp, &hdr, sizeof hdr);
399 gettimeofday(&c->conn_timeout, NULL);
400 c->conn_timeout.tv_sec += utcp->timeout;
405 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
406 if(c->reapable || c->state != SYN_RECEIVED) {
407 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
411 debug("%p accepted, %p %p\n", c, recv, priv);
414 set_state(c, ESTABLISHED);
417 static void ack(struct utcp_connection *c, bool sendatleastone) {
418 int32_t left = seqdiff(c->snd.last, c->snd.nxt);
419 int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
420 debug("cwndleft = %d\n", cwndleft);
430 if(!left && !sendatleastone)
438 pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
442 pkt->hdr.src = c->src;
443 pkt->hdr.dst = c->dst;
444 pkt->hdr.ack = c->rcv.nxt;
445 pkt->hdr.wnd = c->snd.wnd;
450 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
451 pkt->hdr.seq = c->snd.nxt;
453 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
455 c->snd.nxt += seglen;
458 if(seglen && fin_wanted(c, c->snd.nxt)) {
463 if(!c->rtt_start.tv_sec) {
464 // Start RTT measurement
465 gettimeofday(&c->rtt_start, NULL);
466 c->rtt_seq = pkt->hdr.seq + seglen;
467 debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
470 print_packet(c->utcp, "send", pkt, sizeof pkt->hdr + seglen);
471 c->utcp->send(c->utcp, pkt, sizeof pkt->hdr + seglen);
477 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
479 debug("Error: send() called on closed connection %p\n", c);
489 debug("Error: send() called on unconnected connection %p\n", c);
500 debug("Error: send() called on closing connection %p\n", c);
505 // Add data to send buffer
515 len = buffer_put(&c->sndbuf, data, len);
523 if(!timerisset(&c->rtrx_timeout))
524 start_retransmit_timer(c);
528 static void swap_ports(struct hdr *hdr) {
529 uint16_t tmp = hdr->src;
534 static void retransmit(struct utcp_connection *c) {
535 if(c->state == CLOSED || c->snd.nxt == c->snd.una)
538 struct utcp *utcp = c->utcp;
545 pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
549 pkt->hdr.src = c->src;
550 pkt->hdr.dst = c->dst;
554 // Send our SYN again
555 pkt->hdr.seq = c->snd.iss;
557 pkt->hdr.wnd = c->rcv.wnd;
559 print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
560 utcp->send(utcp, pkt, sizeof pkt->hdr);
565 pkt->hdr.seq = c->snd.nxt;
566 pkt->hdr.ack = c->rcv.nxt;
567 pkt->hdr.ctl = SYN | ACK;
568 print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
569 utcp->send(utcp, pkt, sizeof pkt->hdr);
577 // Send unacked data again.
578 pkt->hdr.seq = c->snd.una;
579 pkt->hdr.ack = c->rcv.nxt;
581 uint32_t len = seqdiff(c->snd.last, c->snd.una);
584 if(fin_wanted(c, c->snd.una + len)) {
588 c->snd.nxt = c->snd.una + len;
589 c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
590 buffer_copy(&c->sndbuf, pkt->data, 0, len);
591 print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr + len);
592 utcp->send(utcp, pkt, sizeof pkt->hdr + len);
599 // We shouldn't need to retransmit anything in this state.
603 stop_retransmit_timer(c);
607 start_retransmit_timer(c);
609 if(utcp->rto > MAX_RTO)
611 c->rtt_start.tv_sec = 0; // invalidate RTT timer
617 // Update receive buffer and SACK entries after consuming data.
618 static void sack_consume(struct utcp_connection *c, size_t len) {
619 debug("sack_consume %zu\n", len);
620 if(len > c->rcvbuf.used)
623 buffer_get(&c->rcvbuf, NULL, len);
625 for(int i = 0; i < NSACKS && c->sacks[i].len; ) {
626 if(len < c->sacks[i].offset) {
627 c->sacks[i].offset -= len;
629 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
630 c->sacks[i].offset = 0;
631 c->sacks[i].len -= len - c->sacks[i].offset;
635 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof c->sacks[i]);
636 c->sacks[i + 1].len = 0;
644 for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
645 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
648 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
649 debug("out of order packet, offset %u\n", offset);
650 // Packet loss or reordering occured. Store the data in the buffer.
651 ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
655 // Make note of where we put it.
656 for(int i = 0; i < NSACKS; i++) {
657 if(!c->sacks[i].len) { // nothing to merge, add new entry
658 debug("New SACK entry %d\n", i);
659 c->sacks[i].offset = offset;
660 c->sacks[i].len = rxd;
662 } else if(offset < c->sacks[i].offset) {
663 if(offset + rxd < c->sacks[i].offset) { // insert before
664 if(!c->sacks[NSACKS - 1].len) { // only if room left
665 debug("Insert SACK entry at %d\n", i);
666 memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof c->sacks[i]);
667 c->sacks[i].offset = offset;
668 c->sacks[i].len = rxd;
672 debug("Merge with start of SACK entry at %d\n", i);
673 c->sacks[i].offset = offset;
676 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
677 if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
678 debug("Merge with end of SACK entry at %d\n", i);
679 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
680 // TODO: handle potential merge with next entry
686 for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
687 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
690 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
691 // Check if we can process out-of-order data now.
692 if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
693 debug("incoming packet len %zu connected with SACK at %u\n", len, c->sacks[0].offset);
694 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
695 len = max(len, c->sacks[0].offset + c->sacks[0].len);
696 data = c->rcvbuf.data;
700 ssize_t rxd = c->recv(c, data, len);
702 // TODO: handle the application not accepting all data.
708 sack_consume(c, len);
714 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
715 uint32_t offset = seqdiff(seq, c->rcv.nxt);
716 if(offset + len > c->rcvbuf.maxsize)
720 handle_out_of_order(c, offset, data, len);
722 handle_in_order(c, data, len);
726 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
740 print_packet(utcp, "recv", data, len);
742 // Drop packets smaller than the header
745 if(len < sizeof hdr) {
750 // Make a copy from the potentially unaligned data to a struct hdr
752 memcpy(&hdr, data, sizeof hdr);
756 // Drop packets with an unknown CTL flag
758 if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
763 // Try to match the packet to an existing connection
765 struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
767 // Is it for a new connection?
770 // Ignore RST packets
775 // Is it a SYN packet and are we LISTENing?
777 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
778 // If we don't want to accept it, send a RST back
779 if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
784 // Try to allocate memory, otherwise send a RST back
785 c = allocate_connection(utcp, hdr.dst, hdr.src);
791 // Return SYN+ACK, go to SYN_RECEIVED state
792 c->snd.wnd = hdr.wnd;
793 c->rcv.irs = hdr.seq;
794 c->rcv.nxt = c->rcv.irs + 1;
795 set_state(c, SYN_RECEIVED);
799 hdr.ack = c->rcv.irs + 1;
800 hdr.seq = c->snd.iss;
802 print_packet(c->utcp, "send", &hdr, sizeof hdr);
803 utcp->send(utcp, &hdr, sizeof hdr);
805 // No, we don't want your packets, send a RST back
813 debug("%p state %s\n", c->utcp, strstate[c->state]);
815 // In case this is for a CLOSED connection, ignore the packet.
816 // TODO: make it so incoming packets can never match a CLOSED connection.
818 if(c->state == CLOSED)
821 // It is for an existing connection.
823 uint32_t prevrcvnxt = c->rcv.nxt;
825 // 1. Drop invalid packets.
827 // 1a. Drop packets that should not happen in our current state.
847 // 1b. Drop packets with a sequence number not in our receive window.
851 if(c->state == SYN_SENT)
854 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
856 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
858 // cut already accepted front overlapping
860 acceptable = rcv_offset + len >= 0;
867 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
871 debug("Packet not acceptable, %u <= %u + %zu < %u\n", c->rcv.nxt, hdr.seq, len, c->rcv.nxt + c->rcvbuf.maxsize);
872 // Ignore unacceptable RST packets.
875 // Otherwise, send an ACK back in the hope things improve.
880 c->snd.wnd = hdr.wnd; // TODO: move below
882 // 1c. Drop packets with an invalid ACK.
883 // ackno should not roll back, and it should also not be bigger than what we ever could have sent
884 // (= snd.una + c->sndbuf.used).
886 if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
887 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
888 // Ignore unacceptable RST packets.
894 // 2. Handle RST packets
901 // The peer has refused our connection.
902 set_state(c, CLOSED);
903 errno = ECONNREFUSED;
910 // We haven't told the application about this connection yet. Silently delete.
919 // The peer has aborted our connection.
920 set_state(c, CLOSED);
930 // As far as the application is concerned, the connection has already been closed.
931 // If it has called utcp_close() already, we can immediately free this connection.
936 // Otherwise, immediately move to the CLOSED state.
937 set_state(c, CLOSED);
947 // 3. Advance snd.una
949 uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
950 prevrcvnxt = c->rcv.nxt;
954 if(c->rtt_start.tv_sec) {
955 if(c->rtt_seq == hdr.ack) {
956 struct timeval now, diff;
957 gettimeofday(&now, NULL);
958 timersub(&now, &c->rtt_start, &diff);
959 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
960 c->rtt_start.tv_sec = 0;
961 } else if(c->rtt_seq < hdr.ack) {
962 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
963 c->rtt_start.tv_sec = 0;
967 int32_t data_acked = advanced;
974 // TODO: handle FIN as well.
979 assert(data_acked >= 0);
981 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
982 assert(data_acked <= bufused);
985 buffer_get(&c->sndbuf, NULL, data_acked);
987 // Also advance snd.nxt if possible
988 if(seqdiff(c->snd.nxt, hdr.ack) < 0)
989 c->snd.nxt = hdr.ack;
991 c->snd.una = hdr.ack;
994 c->snd.cwnd += utcp->mtu;
995 if(c->snd.cwnd > c->sndbuf.maxsize)
996 c->snd.cwnd = c->sndbuf.maxsize;
998 // Check if we have sent a FIN that is now ACKed.
1001 if(c->snd.una == c->snd.last)
1002 set_state(c, FIN_WAIT_2);
1005 if(c->snd.una == c->snd.last) {
1006 gettimeofday(&c->conn_timeout, NULL);
1007 c->conn_timeout.tv_sec += 60;
1008 set_state(c, TIME_WAIT);
1017 if(c->dupack == 3) {
1018 debug("Triplicate ACK\n");
1019 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1020 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1021 //Reset the congestion window so we wait for ACKs.
1022 c->snd.nxt = c->snd.una;
1023 c->snd.cwnd = utcp->mtu;
1024 start_retransmit_timer(c);
1032 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
1033 if(c->snd.una == c->snd.last)
1034 stop_retransmit_timer(c);
1036 start_retransmit_timer(c);
1039 // 5. Process SYN stuff
1044 // This is a SYNACK. It should always have ACKed the SYN.
1047 c->rcv.irs = hdr.seq;
1048 c->rcv.nxt = hdr.seq;
1049 set_state(c, ESTABLISHED);
1050 // TODO: notify application of this somehow.
1060 // Ehm, no. We should never receive a second SYN.
1069 // SYN counts as one sequence number
1073 // 6. Process new data
1075 if(c->state == SYN_RECEIVED) {
1076 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1080 // Are we still LISTENing?
1082 utcp->accept(c, c->src);
1084 if(c->state != ESTABLISHED) {
1085 set_state(c, CLOSED);
1095 // This should never happen.
1108 // Ehm no, We should never receive more data after a FIN.
1117 handle_incoming_data(c, hdr.seq, data, len);
1120 // 7. Process FIN stuff
1122 if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
1126 // This should never happen.
1132 set_state(c, CLOSE_WAIT);
1135 set_state(c, CLOSING);
1138 gettimeofday(&c->conn_timeout, NULL);
1139 c->conn_timeout.tv_sec += 60;
1140 set_state(c, TIME_WAIT);
1146 // Ehm, no. We should never receive a second FIN.
1155 // FIN counts as one sequence number
1159 // Inform the application that the peer closed the connection.
1162 c->recv(c, NULL, 0);
1166 // Now we send something back if:
1167 // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
1168 // -> sendatleastone = true
1169 // - or we got an ack, so we should maybe send a bit more data
1170 // -> sendatleastone = false
1172 ack(c, len || prevrcvnxt != c->rcv.nxt);
1182 hdr.ack = hdr.seq + len;
1184 hdr.ctl = RST | ACK;
1186 print_packet(utcp, "send", &hdr, sizeof hdr);
1187 utcp->send(utcp, &hdr, sizeof hdr);
1192 int utcp_shutdown(struct utcp_connection *c, int dir) {
1193 debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1200 debug("Error: shutdown() called on closed connection %p\n", c);
1205 if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1210 // TCP does not have a provision for stopping incoming packets.
1211 // The best we can do is to just ignore them.
1212 if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR)
1215 // The rest of the code deals with shutting down writes.
1216 if(dir == UTCP_SHUT_RD)
1226 set_state(c, CLOSED);
1231 set_state(c, FIN_WAIT_1);
1237 set_state(c, CLOSING);
1252 int utcp_close(struct utcp_connection *c) {
1253 if(utcp_shutdown(c, SHUT_RDWR))
1261 int utcp_abort(struct utcp_connection *c) {
1268 debug("Error: abort() called on closed connection %p\n", c);
1285 set_state(c, CLOSED);
1293 set_state(c, CLOSED);
1303 hdr.seq = c->snd.nxt;
1308 print_packet(c->utcp, "send", &hdr, sizeof hdr);
1309 c->utcp->send(c->utcp, &hdr, sizeof hdr);
1314 * One call to this function will loop through all connections,
1315 * checking if something needs to be resent or not.
1316 * The return value is the time to the next timeout in milliseconds,
1317 * or maybe a negative value if the timeout is infinite.
1319 struct timeval utcp_timeout(struct utcp *utcp) {
1321 gettimeofday(&now, NULL);
1322 struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1324 for(int i = 0; i < utcp->nconnections; i++) {
1325 struct utcp_connection *c = utcp->connections[i];
1329 // delete connections that have been utcp_close()d.
1330 if(c->state == CLOSED) {
1332 debug("Reaping %p\n", c);
1339 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1343 c->recv(c, NULL, 0);
1347 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1348 debug("retransmit()\n");
1352 if(c->poll && buffer_free(&c->sndbuf) && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1353 c->poll(c, buffer_free(&c->sndbuf));
1355 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
1356 next = c->conn_timeout;
1358 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
1359 next = c->rtrx_timeout;
1362 struct timeval diff;
1363 timersub(&next, &now, &diff);
1367 bool utcp_is_active(struct utcp *utcp) {
1371 for(int i = 0; i < utcp->nconnections; i++)
1372 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT)
1378 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1384 struct utcp *utcp = calloc(1, sizeof *utcp);
1388 utcp->accept = accept;
1389 utcp->pre_accept = pre_accept;
1392 utcp->mtu = DEFAULT_MTU;
1393 utcp->timeout = DEFAULT_USER_TIMEOUT; // s
1394 utcp->rto = START_RTO; // us
1399 void utcp_exit(struct utcp *utcp) {
1402 for(int i = 0; i < utcp->nconnections; i++) {
1403 if(!utcp->connections[i]->reapable)
1404 debug("Warning, freeing unclosed connection %p\n", utcp->connections[i]);
1405 buffer_exit(&utcp->connections[i]->sndbuf);
1406 free(utcp->connections[i]);
1408 free(utcp->connections);
1412 uint16_t utcp_get_mtu(struct utcp *utcp) {
1413 return utcp ? utcp->mtu : 0;
1416 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1417 // TODO: handle overhead of the header
1422 int utcp_get_user_timeout(struct utcp *u) {
1423 return u ? u->timeout : 0;
1426 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1428 u->timeout = timeout;
1431 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1432 return c ? c->sndbuf.maxsize : 0;
1435 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1436 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1437 return buffer_free(&c->sndbuf);
1442 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1445 c->sndbuf.maxsize = size;
1446 if(c->sndbuf.maxsize != size)
1447 c->sndbuf.maxsize = -1;
1450 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1451 return c ? c->rcvbuf.maxsize : 0;
1454 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1455 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1456 return buffer_free(&c->rcvbuf);
1461 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1464 c->rcvbuf.maxsize = size;
1465 if(c->rcvbuf.maxsize != size)
1466 c->rcvbuf.maxsize = -1;
1469 bool utcp_get_nodelay(struct utcp_connection *c) {
1470 return c ? c->nodelay : false;
1473 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1475 c->nodelay = nodelay;
1478 bool utcp_get_keepalive(struct utcp_connection *c) {
1479 return c ? c->keepalive : false;
1482 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1484 c->keepalive = keepalive;
1487 size_t utcp_get_outq(struct utcp_connection *c) {
1488 return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
1491 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
1496 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
1501 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
1503 utcp->accept = accept;
1504 utcp->pre_accept = pre_accept;