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)
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(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
386 struct utcp_connection *c = allocate_connection(utcp, 0, dst);
397 hdr.seq = c->snd.iss;
399 hdr.wnd = c->rcv.wnd;
403 set_state(c, SYN_SENT);
405 print_packet(utcp, "send", &hdr, sizeof hdr);
406 utcp->send(utcp, &hdr, sizeof hdr);
408 gettimeofday(&c->conn_timeout, NULL);
409 c->conn_timeout.tv_sec += utcp->timeout;
414 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
415 if(c->reapable || c->state != SYN_RECEIVED) {
416 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
420 debug("%p accepted, %p %p\n", c, recv, priv);
423 set_state(c, ESTABLISHED);
426 static void ack(struct utcp_connection *c, bool sendatleastone) {
427 int32_t left = seqdiff(c->snd.last, c->snd.nxt);
428 int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
429 debug("cwndleft = %d\n", cwndleft);
439 if(!left && !sendatleastone)
447 pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
451 pkt->hdr.src = c->src;
452 pkt->hdr.dst = c->dst;
453 pkt->hdr.ack = c->rcv.nxt;
454 pkt->hdr.wnd = c->snd.wnd;
459 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
460 pkt->hdr.seq = c->snd.nxt;
462 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
464 c->snd.nxt += seglen;
467 if(seglen && fin_wanted(c, c->snd.nxt)) {
472 if(!c->rtt_start.tv_sec) {
473 // Start RTT measurement
474 gettimeofday(&c->rtt_start, NULL);
475 c->rtt_seq = pkt->hdr.seq + seglen;
476 debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
479 print_packet(c->utcp, "send", pkt, sizeof pkt->hdr + seglen);
480 c->utcp->send(c->utcp, pkt, sizeof pkt->hdr + seglen);
486 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
488 debug("Error: send() called on closed connection %p\n", c);
498 debug("Error: send() called on unconnected connection %p\n", c);
509 debug("Error: send() called on closing connection %p\n", c);
514 // Add data to send buffer
524 len = buffer_put(&c->sndbuf, data, len);
532 if(!timerisset(&c->rtrx_timeout))
533 start_retransmit_timer(c);
537 static void swap_ports(struct hdr *hdr) {
538 uint16_t tmp = hdr->src;
543 static void retransmit(struct utcp_connection *c) {
544 if(c->state == CLOSED || c->snd.last == c->snd.una) {
545 debug("Retransmit() called but nothing to retransmit!\n");
546 stop_retransmit_timer(c);
550 struct utcp *utcp = c->utcp;
557 pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
561 pkt->hdr.src = c->src;
562 pkt->hdr.dst = c->dst;
563 pkt->hdr.wnd = c->rcv.wnd;
568 // Send our SYN again
569 pkt->hdr.seq = c->snd.iss;
572 print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
573 utcp->send(utcp, pkt, sizeof pkt->hdr);
578 pkt->hdr.seq = c->snd.nxt;
579 pkt->hdr.ack = c->rcv.nxt;
580 pkt->hdr.ctl = SYN | ACK;
581 print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
582 utcp->send(utcp, pkt, sizeof pkt->hdr);
590 // Send unacked data again.
591 pkt->hdr.seq = c->snd.una;
592 pkt->hdr.ack = c->rcv.nxt;
594 uint32_t len = seqdiff(c->snd.last, c->snd.una);
597 if(fin_wanted(c, c->snd.una + len)) {
601 c->snd.nxt = c->snd.una + len;
602 c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
603 buffer_copy(&c->sndbuf, pkt->data, 0, len);
604 print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr + len);
605 utcp->send(utcp, pkt, sizeof pkt->hdr + len);
612 // We shouldn't need to retransmit anything in this state.
616 stop_retransmit_timer(c);
620 start_retransmit_timer(c);
622 if(utcp->rto > MAX_RTO)
624 c->rtt_start.tv_sec = 0; // invalidate RTT timer
630 /* Update receive buffer and SACK entries after consuming data.
634 * |.....0000..1111111111.....22222......3333|
637 * 0..3 represent the SACK entries. The ^ indicates up to which point we want
638 * to remove data from the receive buffer. The idea is to substract "len"
639 * from the offset of all the SACK entries, and then remove/cut down entries
640 * that are shifted to before the start of the receive buffer.
642 * There are three cases:
643 * - the SACK entry is after ^, in that case just change the offset.
644 * - the SACK entry starts before and ends after ^, so we have to
645 * change both its offset and size.
646 * - the SACK entry is completely before ^, in that case delete it.
648 static void sack_consume(struct utcp_connection *c, size_t len) {
649 debug("sack_consume %lu\n", (unsigned long)len);
650 if(len > c->rcvbuf.used)
653 buffer_get(&c->rcvbuf, NULL, len);
655 for(int i = 0; i < NSACKS && c->sacks[i].len; ) {
656 if(len < c->sacks[i].offset) {
657 c->sacks[i].offset -= len;
659 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
660 c->sacks[i].len -= len - c->sacks[i].offset;
661 c->sacks[i].offset = 0;
665 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof c->sacks[i]);
666 c->sacks[NSACKS - 1].len = 0;
674 for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
675 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
678 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
679 debug("out of order packet, offset %u\n", offset);
680 // Packet loss or reordering occured. Store the data in the buffer.
681 ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
685 // Make note of where we put it.
686 for(int i = 0; i < NSACKS; i++) {
687 if(!c->sacks[i].len) { // nothing to merge, add new entry
688 debug("New SACK entry %d\n", i);
689 c->sacks[i].offset = offset;
690 c->sacks[i].len = rxd;
692 } else if(offset < c->sacks[i].offset) {
693 if(offset + rxd < c->sacks[i].offset) { // insert before
694 if(!c->sacks[NSACKS - 1].len) { // only if room left
695 debug("Insert SACK entry at %d\n", i);
696 memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof c->sacks[i]);
697 c->sacks[i].offset = offset;
698 c->sacks[i].len = rxd;
700 debug("SACK entries full, dropping packet\n");
704 debug("Merge with start of SACK entry at %d\n", i);
705 c->sacks[i].offset = offset;
708 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
709 if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
710 debug("Merge with end of SACK entry at %d\n", i);
711 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
712 // TODO: handle potential merge with next entry
718 for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
719 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
722 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
723 // Check if we can process out-of-order data now.
724 if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
725 debug("incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
726 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
727 len = max(len, c->sacks[0].offset + c->sacks[0].len);
728 data = c->rcvbuf.data;
732 ssize_t rxd = c->recv(c, data, len);
734 // TODO: handle the application not accepting all data.
740 sack_consume(c, len);
746 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
747 uint32_t offset = seqdiff(seq, c->rcv.nxt);
748 if(offset + len > c->rcvbuf.maxsize)
752 handle_out_of_order(c, offset, data, len);
754 handle_in_order(c, data, len);
758 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
772 print_packet(utcp, "recv", data, len);
774 // Drop packets smaller than the header
777 if(len < sizeof hdr) {
782 // Make a copy from the potentially unaligned data to a struct hdr
784 memcpy(&hdr, data, sizeof hdr);
788 // Drop packets with an unknown CTL flag
790 if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
795 // Try to match the packet to an existing connection
797 struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
799 // Is it for a new connection?
802 // Ignore RST packets
807 // Is it a SYN packet and are we LISTENing?
809 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
810 // If we don't want to accept it, send a RST back
811 if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
816 // Try to allocate memory, otherwise send a RST back
817 c = allocate_connection(utcp, hdr.dst, hdr.src);
823 // Return SYN+ACK, go to SYN_RECEIVED state
824 c->snd.wnd = hdr.wnd;
825 c->rcv.irs = hdr.seq;
826 c->rcv.nxt = c->rcv.irs + 1;
827 set_state(c, SYN_RECEIVED);
831 hdr.ack = c->rcv.irs + 1;
832 hdr.seq = c->snd.iss;
834 print_packet(c->utcp, "send", &hdr, sizeof hdr);
835 utcp->send(utcp, &hdr, sizeof hdr);
837 // No, we don't want your packets, send a RST back
845 debug("%p state %s\n", c->utcp, strstate[c->state]);
847 // In case this is for a CLOSED connection, ignore the packet.
848 // TODO: make it so incoming packets can never match a CLOSED connection.
850 if(c->state == CLOSED) {
851 debug("Got packet for closed connection\n");
855 // It is for an existing connection.
857 uint32_t prevrcvnxt = c->rcv.nxt;
859 // 1. Drop invalid packets.
861 // 1a. Drop packets that should not happen in our current state.
881 // 1b. Drop packets with a sequence number not in our receive window.
885 if(c->state == SYN_SENT)
888 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
890 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
892 // cut already accepted front overlapping
894 acceptable = rcv_offset + len >= 0;
901 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
905 debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
906 // Ignore unacceptable RST packets.
909 // Otherwise, send an ACK back in the hope things improve.
914 c->snd.wnd = hdr.wnd; // TODO: move below
916 // 1c. Drop packets with an invalid ACK.
917 // ackno should not roll back, and it should also not be bigger than what we ever could have sent
918 // (= snd.una + c->sndbuf.used).
920 if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
921 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
922 // Ignore unacceptable RST packets.
928 // 2. Handle RST packets
935 // The peer has refused our connection.
936 set_state(c, CLOSED);
937 errno = ECONNREFUSED;
944 // We haven't told the application about this connection yet. Silently delete.
953 // The peer has aborted our connection.
954 set_state(c, CLOSED);
964 // As far as the application is concerned, the connection has already been closed.
965 // If it has called utcp_close() already, we can immediately free this connection.
970 // Otherwise, immediately move to the CLOSED state.
971 set_state(c, CLOSED);
981 // 3. Advance snd.una
983 uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
984 prevrcvnxt = c->rcv.nxt;
988 if(c->rtt_start.tv_sec) {
989 if(c->rtt_seq == hdr.ack) {
990 struct timeval now, diff;
991 gettimeofday(&now, NULL);
992 timersub(&now, &c->rtt_start, &diff);
993 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
994 c->rtt_start.tv_sec = 0;
995 } else if(c->rtt_seq < hdr.ack) {
996 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
997 c->rtt_start.tv_sec = 0;
1001 int32_t data_acked = advanced;
1008 // TODO: handle FIN as well.
1013 assert(data_acked >= 0);
1015 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1016 assert(data_acked <= bufused);
1019 buffer_get(&c->sndbuf, NULL, data_acked);
1021 // Also advance snd.nxt if possible
1022 if(seqdiff(c->snd.nxt, hdr.ack) < 0)
1023 c->snd.nxt = hdr.ack;
1025 c->snd.una = hdr.ack;
1028 c->snd.cwnd += utcp->mtu;
1029 if(c->snd.cwnd > c->sndbuf.maxsize)
1030 c->snd.cwnd = c->sndbuf.maxsize;
1032 // Check if we have sent a FIN that is now ACKed.
1035 if(c->snd.una == c->snd.last)
1036 set_state(c, FIN_WAIT_2);
1039 if(c->snd.una == c->snd.last) {
1040 gettimeofday(&c->conn_timeout, NULL);
1041 c->conn_timeout.tv_sec += 60;
1042 set_state(c, TIME_WAIT);
1051 if(c->dupack == 3) {
1052 debug("Triplicate ACK\n");
1053 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1054 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1055 //Reset the congestion window so we wait for ACKs.
1056 c->snd.nxt = c->snd.una;
1057 c->snd.cwnd = utcp->mtu;
1058 start_retransmit_timer(c);
1066 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
1067 if(c->snd.una == c->snd.last)
1068 stop_retransmit_timer(c);
1070 start_retransmit_timer(c);
1073 // 5. Process SYN stuff
1078 // This is a SYNACK. It should always have ACKed the SYN.
1081 c->rcv.irs = hdr.seq;
1082 c->rcv.nxt = hdr.seq;
1083 set_state(c, ESTABLISHED);
1084 // TODO: notify application of this somehow.
1094 // Ehm, no. We should never receive a second SYN.
1103 // SYN counts as one sequence number
1107 // 6. Process new data
1109 if(c->state == SYN_RECEIVED) {
1110 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1114 // Are we still LISTENing?
1116 utcp->accept(c, c->src);
1118 if(c->state != ESTABLISHED) {
1119 set_state(c, CLOSED);
1129 // This should never happen.
1142 // Ehm no, We should never receive more data after a FIN.
1151 handle_incoming_data(c, hdr.seq, data, len);
1154 // 7. Process FIN stuff
1156 if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
1160 // This should never happen.
1166 set_state(c, CLOSE_WAIT);
1169 set_state(c, CLOSING);
1172 gettimeofday(&c->conn_timeout, NULL);
1173 c->conn_timeout.tv_sec += 60;
1174 set_state(c, TIME_WAIT);
1180 // Ehm, no. We should never receive a second FIN.
1189 // FIN counts as one sequence number
1193 // Inform the application that the peer closed the connection.
1196 c->recv(c, NULL, 0);
1200 // Now we send something back if:
1201 // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
1202 // -> sendatleastone = true
1203 // - or we got an ack, so we should maybe send a bit more data
1204 // -> sendatleastone = false
1206 ack(c, len || prevrcvnxt != c->rcv.nxt);
1216 hdr.ack = hdr.seq + len;
1218 hdr.ctl = RST | ACK;
1220 print_packet(utcp, "send", &hdr, sizeof hdr);
1221 utcp->send(utcp, &hdr, sizeof hdr);
1226 int utcp_shutdown(struct utcp_connection *c, int dir) {
1227 debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1234 debug("Error: shutdown() called on closed connection %p\n", c);
1239 if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1244 // TCP does not have a provision for stopping incoming packets.
1245 // The best we can do is to just ignore them.
1246 if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR)
1249 // The rest of the code deals with shutting down writes.
1250 if(dir == UTCP_SHUT_RD)
1260 set_state(c, CLOSED);
1265 set_state(c, FIN_WAIT_1);
1271 set_state(c, CLOSING);
1283 if(!timerisset(&c->rtrx_timeout))
1284 start_retransmit_timer(c);
1288 int utcp_close(struct utcp_connection *c) {
1289 if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN)
1297 int utcp_abort(struct utcp_connection *c) {
1304 debug("Error: abort() called on closed connection %p\n", c);
1321 set_state(c, CLOSED);
1329 set_state(c, CLOSED);
1339 hdr.seq = c->snd.nxt;
1344 print_packet(c->utcp, "send", &hdr, sizeof hdr);
1345 c->utcp->send(c->utcp, &hdr, sizeof hdr);
1350 * One call to this function will loop through all connections,
1351 * checking if something needs to be resent or not.
1352 * The return value is the time to the next timeout in milliseconds,
1353 * or maybe a negative value if the timeout is infinite.
1355 struct timeval utcp_timeout(struct utcp *utcp) {
1357 gettimeofday(&now, NULL);
1358 struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1360 for(int i = 0; i < utcp->nconnections; i++) {
1361 struct utcp_connection *c = utcp->connections[i];
1365 // delete connections that have been utcp_close()d.
1366 if(c->state == CLOSED) {
1368 debug("Reaping %p\n", c);
1375 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1379 c->recv(c, NULL, 0);
1383 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1384 debug("retransmit()\n");
1388 if(c->poll && buffer_free(&c->sndbuf) && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1389 c->poll(c, buffer_free(&c->sndbuf));
1391 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
1392 next = c->conn_timeout;
1394 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
1395 next = c->rtrx_timeout;
1398 struct timeval diff;
1399 timersub(&next, &now, &diff);
1403 bool utcp_is_active(struct utcp *utcp) {
1407 for(int i = 0; i < utcp->nconnections; i++)
1408 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT)
1414 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1420 struct utcp *utcp = calloc(1, sizeof *utcp);
1424 utcp->accept = accept;
1425 utcp->pre_accept = pre_accept;
1428 utcp->mtu = DEFAULT_MTU;
1429 utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1430 utcp->rto = START_RTO; // usec
1435 void utcp_exit(struct utcp *utcp) {
1438 for(int i = 0; i < utcp->nconnections; i++) {
1439 if(!utcp->connections[i]->reapable)
1440 debug("Warning, freeing unclosed connection %p\n", utcp->connections[i]);
1441 buffer_exit(&utcp->connections[i]->rcvbuf);
1442 buffer_exit(&utcp->connections[i]->sndbuf);
1443 free(utcp->connections[i]);
1445 free(utcp->connections);
1449 uint16_t utcp_get_mtu(struct utcp *utcp) {
1450 return utcp ? utcp->mtu : 0;
1453 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1454 // TODO: handle overhead of the header
1459 int utcp_get_user_timeout(struct utcp *u) {
1460 return u ? u->timeout : 0;
1463 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1465 u->timeout = timeout;
1468 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1469 return c ? c->sndbuf.maxsize : 0;
1472 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1473 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1474 return buffer_free(&c->sndbuf);
1479 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1482 c->sndbuf.maxsize = size;
1483 if(c->sndbuf.maxsize != size)
1484 c->sndbuf.maxsize = -1;
1487 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1488 return c ? c->rcvbuf.maxsize : 0;
1491 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1492 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1493 return buffer_free(&c->rcvbuf);
1498 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1501 c->rcvbuf.maxsize = size;
1502 if(c->rcvbuf.maxsize != size)
1503 c->rcvbuf.maxsize = -1;
1506 bool utcp_get_nodelay(struct utcp_connection *c) {
1507 return c ? c->nodelay : false;
1510 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1512 c->nodelay = nodelay;
1515 bool utcp_get_keepalive(struct utcp_connection *c) {
1516 return c ? c->keepalive : false;
1519 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1521 c->keepalive = keepalive;
1524 size_t utcp_get_outq(struct utcp_connection *c) {
1525 return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
1528 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
1533 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
1538 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
1540 utcp->accept = accept;
1541 utcp->pre_accept = pre_accept;