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 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 // Try to match the packet to an existing connection
833 struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
835 // Is it for a new connection?
838 // Ignore RST packets
843 // Is it a SYN packet and are we LISTENing?
845 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
846 // If we don't want to accept it, send a RST back
847 if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
852 // Try to allocate memory, otherwise send a RST back
853 c = allocate_connection(utcp, hdr.dst, hdr.src);
859 // Parse auxilliary information
861 if(hdr.aux != 0x0101 || len < 4 || ((uint8_t *)data)[0] != 1) {
865 c->flags = ((uint8_t *)data)[3] & 0x7;
872 // Return SYN+ACK, go to SYN_RECEIVED state
873 c->snd.wnd = hdr.wnd;
874 c->rcv.irs = hdr.seq;
875 c->rcv.nxt = c->rcv.irs + 1;
876 set_state(c, SYN_RECEIVED);
880 hdr.ack = c->rcv.irs + 1;
881 hdr.seq = c->snd.iss;
883 print_packet(c->utcp, "send", &hdr, sizeof hdr);
884 utcp->send(utcp, &hdr, sizeof hdr);
886 // No, we don't want your packets, send a RST back
894 debug("%p state %s\n", c->utcp, strstate[c->state]);
896 // In case this is for a CLOSED connection, ignore the packet.
897 // TODO: make it so incoming packets can never match a CLOSED connection.
899 if(c->state == CLOSED) {
900 debug("Got packet for closed connection\n");
904 // It is for an existing connection.
906 uint32_t prevrcvnxt = c->rcv.nxt;
908 // 1. Drop invalid packets.
910 // 1a. Drop packets that should not happen in our current state.
930 // 1b. Drop packets with a sequence number not in our receive window.
934 if(c->state == SYN_SENT)
937 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
939 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
941 // cut already accepted front overlapping
943 acceptable = len > -rcv_offset;
947 hdr.seq -= rcv_offset;
950 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
955 debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
956 // Ignore unacceptable RST packets.
959 // Otherwise, continue processing.
963 c->snd.wnd = hdr.wnd; // TODO: move below
965 // 1c. Drop packets with an invalid ACK.
966 // ackno should not roll back, and it should also not be bigger than what we ever could have sent
967 // (= snd.una + c->sndbuf.used).
969 if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
970 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
971 // Ignore unacceptable RST packets.
977 // 2. Handle RST packets
984 // The peer has refused our connection.
985 set_state(c, CLOSED);
986 errno = ECONNREFUSED;
993 // We haven't told the application about this connection yet. Silently delete.
1002 // The peer has aborted our connection.
1003 set_state(c, CLOSED);
1006 c->recv(c, NULL, 0);
1013 // As far as the application is concerned, the connection has already been closed.
1014 // If it has called utcp_close() already, we can immediately free this connection.
1019 // Otherwise, immediately move to the CLOSED state.
1020 set_state(c, CLOSED);
1030 // 3. Advance snd.una
1032 uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
1033 prevrcvnxt = c->rcv.nxt;
1037 if(c->rtt_start.tv_sec) {
1038 if(c->rtt_seq == hdr.ack) {
1039 struct timeval now, diff;
1040 gettimeofday(&now, NULL);
1041 timersub(&now, &c->rtt_start, &diff);
1042 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
1043 c->rtt_start.tv_sec = 0;
1044 } else if(c->rtt_seq < hdr.ack) {
1045 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1046 c->rtt_start.tv_sec = 0;
1050 int32_t data_acked = advanced;
1057 // TODO: handle FIN as well.
1062 assert(data_acked >= 0);
1064 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1065 assert(data_acked <= bufused);
1068 buffer_get(&c->sndbuf, NULL, data_acked);
1070 // Also advance snd.nxt if possible
1071 if(seqdiff(c->snd.nxt, hdr.ack) < 0)
1072 c->snd.nxt = hdr.ack;
1074 c->snd.una = hdr.ack;
1077 c->snd.cwnd += utcp->mtu;
1078 if(c->snd.cwnd > c->sndbuf.maxsize)
1079 c->snd.cwnd = c->sndbuf.maxsize;
1081 // Check if we have sent a FIN that is now ACKed.
1084 if(c->snd.una == c->snd.last)
1085 set_state(c, FIN_WAIT_2);
1088 if(c->snd.una == c->snd.last) {
1089 gettimeofday(&c->conn_timeout, NULL);
1090 c->conn_timeout.tv_sec += 60;
1091 set_state(c, TIME_WAIT);
1098 if(!len && is_reliable(c)) {
1100 if(c->dupack == 3) {
1101 debug("Triplicate ACK\n");
1102 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1103 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1104 //Reset the congestion window so we wait for ACKs.
1105 c->snd.nxt = c->snd.una;
1106 c->snd.cwnd = utcp->mtu;
1107 start_retransmit_timer(c);
1115 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
1116 if(c->snd.una == c->snd.last)
1117 stop_retransmit_timer(c);
1118 else if(is_reliable(c))
1119 start_retransmit_timer(c);
1122 // 5. Process SYN stuff
1127 // This is a SYNACK. It should always have ACKed the SYN.
1130 c->rcv.irs = hdr.seq;
1131 c->rcv.nxt = hdr.seq;
1132 set_state(c, ESTABLISHED);
1133 // TODO: notify application of this somehow.
1143 // Ehm, no. We should never receive a second SYN.
1152 // SYN counts as one sequence number
1156 // 6. Process new data
1158 if(c->state == SYN_RECEIVED) {
1159 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1163 // Are we still LISTENing?
1165 utcp->accept(c, c->src);
1167 if(c->state != ESTABLISHED) {
1168 set_state(c, CLOSED);
1178 // This should never happen.
1191 // Ehm no, We should never receive more data after a FIN.
1200 handle_incoming_data(c, hdr.seq, data, len);
1203 // 7. Process FIN stuff
1205 if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
1209 // This should never happen.
1215 set_state(c, CLOSE_WAIT);
1218 set_state(c, CLOSING);
1221 gettimeofday(&c->conn_timeout, NULL);
1222 c->conn_timeout.tv_sec += 60;
1223 set_state(c, TIME_WAIT);
1229 // Ehm, no. We should never receive a second FIN.
1238 // FIN counts as one sequence number
1242 // Inform the application that the peer closed the connection.
1245 c->recv(c, NULL, 0);
1249 // Now we send something back if:
1250 // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
1251 // -> sendatleastone = true
1252 // - or we got an ack, so we should maybe send a bit more data
1253 // -> sendatleastone = false
1255 ack(c, len || prevrcvnxt != c->rcv.nxt);
1265 hdr.ack = hdr.seq + len;
1267 hdr.ctl = RST | ACK;
1269 print_packet(utcp, "send", &hdr, sizeof hdr);
1270 utcp->send(utcp, &hdr, sizeof hdr);
1275 int utcp_shutdown(struct utcp_connection *c, int dir) {
1276 debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1283 debug("Error: shutdown() called on closed connection %p\n", c);
1288 if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1293 // TCP does not have a provision for stopping incoming packets.
1294 // The best we can do is to just ignore them.
1295 if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR)
1298 // The rest of the code deals with shutting down writes.
1299 if(dir == UTCP_SHUT_RD)
1309 set_state(c, CLOSED);
1314 set_state(c, FIN_WAIT_1);
1320 set_state(c, CLOSING);
1332 if(!timerisset(&c->rtrx_timeout))
1333 start_retransmit_timer(c);
1337 int utcp_close(struct utcp_connection *c) {
1338 if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN)
1346 int utcp_abort(struct utcp_connection *c) {
1353 debug("Error: abort() called on closed connection %p\n", c);
1370 set_state(c, CLOSED);
1378 set_state(c, CLOSED);
1388 hdr.seq = c->snd.nxt;
1393 print_packet(c->utcp, "send", &hdr, sizeof hdr);
1394 c->utcp->send(c->utcp, &hdr, sizeof hdr);
1399 * One call to this function will loop through all connections,
1400 * checking if something needs to be resent or not.
1401 * The return value is the time to the next timeout in milliseconds,
1402 * or maybe a negative value if the timeout is infinite.
1404 struct timeval utcp_timeout(struct utcp *utcp) {
1406 gettimeofday(&now, NULL);
1407 struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1409 for(int i = 0; i < utcp->nconnections; i++) {
1410 struct utcp_connection *c = utcp->connections[i];
1414 // delete connections that have been utcp_close()d.
1415 if(c->state == CLOSED) {
1417 debug("Reaping %p\n", c);
1424 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1428 c->recv(c, NULL, 0);
1432 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1433 debug("retransmit()\n");
1438 if((c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1439 uint32_t len = buffer_free(&c->sndbuf);
1442 } else if(c->state == CLOSED) {
1447 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
1448 next = c->conn_timeout;
1450 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
1451 next = c->rtrx_timeout;
1454 struct timeval diff;
1455 timersub(&next, &now, &diff);
1459 bool utcp_is_active(struct utcp *utcp) {
1463 for(int i = 0; i < utcp->nconnections; i++)
1464 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT)
1470 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1476 struct utcp *utcp = calloc(1, sizeof *utcp);
1480 utcp->accept = accept;
1481 utcp->pre_accept = pre_accept;
1484 utcp->mtu = DEFAULT_MTU;
1485 utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1486 utcp->rto = START_RTO; // usec
1491 void utcp_exit(struct utcp *utcp) {
1494 for(int i = 0; i < utcp->nconnections; i++) {
1495 if(!utcp->connections[i]->reapable)
1496 debug("Warning, freeing unclosed connection %p\n", utcp->connections[i]);
1497 buffer_exit(&utcp->connections[i]->rcvbuf);
1498 buffer_exit(&utcp->connections[i]->sndbuf);
1499 free(utcp->connections[i]);
1501 free(utcp->connections);
1505 uint16_t utcp_get_mtu(struct utcp *utcp) {
1506 return utcp ? utcp->mtu : 0;
1509 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1510 // TODO: handle overhead of the header
1515 int utcp_get_user_timeout(struct utcp *u) {
1516 return u ? u->timeout : 0;
1519 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1521 u->timeout = timeout;
1524 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1525 return c ? c->sndbuf.maxsize : 0;
1528 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1529 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1530 return buffer_free(&c->sndbuf);
1535 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1538 c->sndbuf.maxsize = size;
1539 if(c->sndbuf.maxsize != size)
1540 c->sndbuf.maxsize = -1;
1543 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1544 return c ? c->rcvbuf.maxsize : 0;
1547 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1548 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1549 return buffer_free(&c->rcvbuf);
1554 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1557 c->rcvbuf.maxsize = size;
1558 if(c->rcvbuf.maxsize != size)
1559 c->rcvbuf.maxsize = -1;
1562 bool utcp_get_nodelay(struct utcp_connection *c) {
1563 return c ? c->nodelay : false;
1566 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1568 c->nodelay = nodelay;
1571 bool utcp_get_keepalive(struct utcp_connection *c) {
1572 return c ? c->keepalive : false;
1575 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1577 c->keepalive = keepalive;
1580 size_t utcp_get_outq(struct utcp_connection *c) {
1581 return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
1584 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
1589 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
1594 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
1596 utcp->accept = accept;
1597 utcp->pre_accept = pre_accept;