2 utcp.c -- Userspace TCP
3 Copyright (C) 2014-2017 Guus Sliepen <guus@tinc-vpn.org>
5 This program is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 2 of the License, or
8 (at your option) any later version.
10 This program is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU General Public License for more details.
15 You should have received a copy of the GNU General Public License along
16 with this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
31 #include <sys/socket.h>
33 #include "utcp_priv.h"
48 #define timersub(a, b, r) do {\
49 (r)->tv_sec = (a)->tv_sec - (b)->tv_sec;\
50 (r)->tv_usec = (a)->tv_usec - (b)->tv_usec;\
52 (r)->tv_sec--, (r)->tv_usec += USEC_PER_SEC;\
56 static inline size_t max(size_t a, size_t b) {
63 static void debug(const char *format, ...) {
66 vfprintf(stderr, format, ap);
70 static void print_packet(struct utcp *utcp, const char *dir, const void *pkt, size_t len) {
72 if(len < sizeof(hdr)) {
73 debug("%p %s: short packet (%lu bytes)\n", utcp, dir, (unsigned long)len);
77 memcpy(&hdr, pkt, sizeof(hdr));
78 debug("%p %s: len=%lu, src=%u dst=%u seq=%u ack=%u wnd=%u aux=%x ctl=", utcp, dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux);
88 if(len > sizeof(hdr)) {
89 uint32_t datalen = len - sizeof(hdr);
90 const uint8_t *data = (uint8_t *)pkt + sizeof(hdr);
91 char str[datalen * 2 + 1];
94 for(uint32_t i = 0; i < datalen; i++) {
95 *p++ = "0123456789ABCDEF"[data[i] >> 4];
96 *p++ = "0123456789ABCDEF"[data[i] & 15];
100 debug(" data=%s", str);
106 #define debug(...) do {} while(0)
107 #define print_packet(...) do {} while(0)
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 int32_t seqdiff(uint32_t a, uint32_t b) {
139 // TODO: convert to ringbuffers to avoid memmove() operations.
141 // Store data into the buffer
142 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
143 debug("buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
145 size_t required = offset + len;
146 if(required > buf->maxsize) {
147 if(offset >= buf->maxsize)
149 len = buf->maxsize - offset;
150 required = buf->maxsize;
153 if(required > buf->size) {
154 size_t newsize = buf->size;
160 } while(newsize < required);
162 if(newsize > buf->maxsize)
163 newsize = buf->maxsize;
164 char *newdata = realloc(buf->data, newsize);
171 memcpy(buf->data + offset, data, len);
172 if(required > buf->used)
173 buf->used = required;
177 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
178 return buffer_put_at(buf, buf->used, data, len);
181 // Get data from the buffer. data can be NULL.
182 static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
186 memcpy(data, buf->data, len);
188 memmove(buf->data, buf->data + len, buf->used - len);
193 // Copy data from the buffer without removing it.
194 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
195 if(offset >= buf->used)
197 if(offset + len > buf->used)
198 len = buf->used - offset;
199 memcpy(data, buf->data + offset, len);
203 static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
204 memset(buf, 0, sizeof(*buf));
206 buf->data = malloc(len);
211 buf->maxsize = maxlen;
215 static void buffer_exit(struct buffer *buf) {
217 memset(buf, 0, sizeof(*buf));
220 static uint32_t buffer_free(const struct buffer *buf) {
221 return buf->maxsize - buf->used;
224 // Connections are stored in a sorted list.
225 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
227 static int compare(const void *va, const void *vb) {
230 const struct utcp_connection *a = *(struct utcp_connection **)va;
231 const struct utcp_connection *b = *(struct utcp_connection **)vb;
234 assert(a->src && b->src);
236 int c = (int)a->src - (int)b->src;
239 c = (int)a->dst - (int)b->dst;
243 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
244 if(!utcp->nconnections)
246 struct utcp_connection key = {
250 struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
251 return match ? *match : NULL;
254 static void free_connection(struct utcp_connection *c) {
255 struct utcp *utcp = c->utcp;
256 struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
260 int i = cp - utcp->connections;
261 memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
262 utcp->nconnections--;
264 buffer_exit(&c->rcvbuf);
265 buffer_exit(&c->sndbuf);
269 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
270 // Check whether this combination of src and dst is free
273 if(find_connection(utcp, src, dst)) {
277 } else { // If src == 0, generate a random port number with the high bit set
278 if(utcp->nconnections >= 32767) {
282 src = rand() | 0x8000;
283 while(find_connection(utcp, src, dst))
287 // Allocate memory for the new connection
289 if(utcp->nconnections >= utcp->nallocated) {
290 if(!utcp->nallocated)
291 utcp->nallocated = 4;
293 utcp->nallocated *= 2;
294 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
297 utcp->connections = new_array;
300 struct utcp_connection *c = calloc(1, sizeof(*c));
304 if(!buffer_init(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
309 if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
310 buffer_exit(&c->sndbuf);
315 // Fill in the details
324 c->snd.una = c->snd.iss;
325 c->snd.nxt = c->snd.iss + 1;
326 c->rcv.wnd = utcp->mtu;
327 c->snd.last = c->snd.nxt;
328 c->snd.cwnd = utcp->mtu;
331 // Add it to the sorted list of connections
333 utcp->connections[utcp->nconnections++] = c;
334 qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
339 static inline uint32_t absdiff(uint32_t a, uint32_t b) {
346 // Update RTT variables. See RFC 6298.
347 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
349 debug("invalid rtt\n");
353 struct utcp *utcp = c->utcp;
357 utcp->rttvar = rtt / 2;
358 utcp->rto = rtt + max(2 * rtt, CLOCK_GRANULARITY);
360 utcp->rttvar = (utcp->rttvar * 3 + absdiff(utcp->srtt, rtt)) / 4;
361 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
362 utcp->rto = utcp->srtt + max(utcp->rttvar, CLOCK_GRANULARITY);
365 if(utcp->rto > MAX_RTO)
368 debug("rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
371 static void start_retransmit_timer(struct utcp_connection *c) {
372 gettimeofday(&c->rtrx_timeout, NULL);
373 c->rtrx_timeout.tv_usec += c->utcp->rto;
374 while(c->rtrx_timeout.tv_usec >= 1000000) {
375 c->rtrx_timeout.tv_usec -= 1000000;
376 c->rtrx_timeout.tv_sec++;
378 debug("timeout set to %lu.%06lu (%u)\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_usec, c->utcp->rto);
381 static void stop_retransmit_timer(struct utcp_connection *c) {
382 timerclear(&c->rtrx_timeout);
383 debug("timeout cleared\n");
386 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
387 struct utcp_connection *c = allocate_connection(utcp, 0, dst);
391 assert((flags & ~0xf) == 0);
402 pkt.hdr.src = c->src;
403 pkt.hdr.dst = c->dst;
404 pkt.hdr.seq = c->snd.iss;
406 pkt.hdr.wnd = c->rcv.wnd;
408 pkt.hdr.aux = 0x0101;
412 pkt.init[3] = flags & 0x7;
414 set_state(c, SYN_SENT);
416 print_packet(utcp, "send", &pkt, sizeof(pkt));
417 utcp->send(utcp, &pkt, sizeof(pkt));
419 gettimeofday(&c->conn_timeout, NULL);
420 c->conn_timeout.tv_sec += utcp->timeout;
422 start_retransmit_timer(c);
427 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
428 return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
431 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
432 if(c->reapable || c->state != SYN_RECEIVED) {
433 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
437 debug("%p accepted, %p %p\n", c, recv, priv);
440 set_state(c, ESTABLISHED);
443 static void ack(struct utcp_connection *c, bool sendatleastone) {
444 int32_t left = seqdiff(c->snd.last, c->snd.nxt);
445 int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
446 debug("cwndleft = %d\n", cwndleft);
456 if(!left && !sendatleastone)
464 pkt = malloc(sizeof(pkt->hdr) + c->utcp->mtu);
468 pkt->hdr.src = c->src;
469 pkt->hdr.dst = c->dst;
470 pkt->hdr.ack = c->rcv.nxt;
471 pkt->hdr.wnd = c->snd.wnd;
476 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
477 pkt->hdr.seq = c->snd.nxt;
479 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
481 c->snd.nxt += seglen;
484 if(seglen && fin_wanted(c, c->snd.nxt)) {
489 if(!c->rtt_start.tv_sec) {
490 // Start RTT measurement
491 gettimeofday(&c->rtt_start, NULL);
492 c->rtt_seq = pkt->hdr.seq + seglen;
493 debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
496 print_packet(c->utcp, "send", pkt, sizeof(pkt->hdr) + seglen);
497 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
503 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
505 debug("Error: send() called on closed connection %p\n", c);
515 debug("Error: send() called on unconnected connection %p\n", c);
526 debug("Error: send() called on closing connection %p\n", c);
531 // Exit early if we have nothing to send.
541 // Add data to send buffer.
543 len = buffer_put(&c->sndbuf, data, len);
551 if(!is_reliable(c)) {
552 c->snd.una = c->snd.nxt = c->snd.last;
553 buffer_get(&c->sndbuf, NULL, c->sndbuf.used);
555 if(is_reliable(c) && !timerisset(&c->rtrx_timeout))
556 start_retransmit_timer(c);
560 static void swap_ports(struct hdr *hdr) {
561 uint16_t tmp = hdr->src;
566 static void retransmit(struct utcp_connection *c) {
567 if(c->state == CLOSED || c->snd.last == c->snd.una) {
568 debug("Retransmit() called but nothing to retransmit!\n");
569 stop_retransmit_timer(c);
573 struct utcp *utcp = c->utcp;
580 pkt = malloc(sizeof(pkt->hdr) + c->utcp->mtu);
584 pkt->hdr.src = c->src;
585 pkt->hdr.dst = c->dst;
586 pkt->hdr.wnd = c->rcv.wnd;
591 // Send our SYN again
592 pkt->hdr.seq = c->snd.iss;
595 pkt->hdr.aux = 0x0101;
599 pkt->data[3] = c->flags & 0x7;
600 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + 4);
601 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
606 pkt->hdr.seq = c->snd.nxt;
607 pkt->hdr.ack = c->rcv.nxt;
608 pkt->hdr.ctl = SYN | ACK;
609 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr));
610 utcp->send(utcp, pkt, sizeof(pkt->hdr));
618 // Send unacked data again.
619 pkt->hdr.seq = c->snd.una;
620 pkt->hdr.ack = c->rcv.nxt;
622 uint32_t len = seqdiff(c->snd.last, c->snd.una);
625 if(fin_wanted(c, c->snd.una + len)) {
629 c->snd.nxt = c->snd.una + len;
630 c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
631 buffer_copy(&c->sndbuf, pkt->data, 0, len);
632 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + len);
633 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
640 // We shouldn't need to retransmit anything in this state.
644 stop_retransmit_timer(c);
648 start_retransmit_timer(c);
650 if(utcp->rto > MAX_RTO)
652 c->rtt_start.tv_sec = 0; // invalidate RTT timer
658 /* Update receive buffer and SACK entries after consuming data.
662 * |.....0000..1111111111.....22222......3333|
665 * 0..3 represent the SACK entries. The ^ indicates up to which point we want
666 * to remove data from the receive buffer. The idea is to substract "len"
667 * from the offset of all the SACK entries, and then remove/cut down entries
668 * that are shifted to before the start of the receive buffer.
670 * There are three cases:
671 * - the SACK entry is after ^, in that case just change the offset.
672 * - the SACK entry starts before and ends after ^, so we have to
673 * change both its offset and size.
674 * - the SACK entry is completely before ^, in that case delete it.
676 static void sack_consume(struct utcp_connection *c, size_t len) {
677 debug("sack_consume %lu\n", (unsigned long)len);
678 if(len > c->rcvbuf.used) {
679 debug("All SACK entries consumed");
684 buffer_get(&c->rcvbuf, NULL, len);
686 for(int i = 0; i < NSACKS && c->sacks[i].len; ) {
687 if(len < c->sacks[i].offset) {
688 c->sacks[i].offset -= len;
690 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
691 c->sacks[i].len -= len - c->sacks[i].offset;
692 c->sacks[i].offset = 0;
696 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
697 c->sacks[NSACKS - 1].len = 0;
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_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
710 debug("out of order packet, offset %u\n", offset);
711 // Packet loss or reordering occured. Store the data in the buffer.
712 ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
713 if(rxd < 0 || (size_t)rxd < len)
716 // Make note of where we put it.
717 for(int i = 0; i < NSACKS; i++) {
718 if(!c->sacks[i].len) { // nothing to merge, add new entry
719 debug("New SACK entry %d\n", i);
720 c->sacks[i].offset = offset;
721 c->sacks[i].len = rxd;
723 } else if(offset < c->sacks[i].offset) {
724 if(offset + rxd < c->sacks[i].offset) { // insert before
725 if(!c->sacks[NSACKS - 1].len) { // only if room left
726 debug("Insert SACK entry at %d\n", i);
727 memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
728 c->sacks[i].offset = offset;
729 c->sacks[i].len = rxd;
731 debug("SACK entries full, dropping packet\n");
735 debug("Merge with start of SACK entry at %d\n", i);
736 c->sacks[i].offset = offset;
739 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
740 if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
741 debug("Merge with end of SACK entry at %d\n", i);
742 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
743 // TODO: handle potential merge with next entry
749 for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
750 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
753 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
754 // Check if we can process out-of-order data now.
755 if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
756 debug("incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
757 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
758 len = max(len, c->sacks[0].offset + c->sacks[0].len);
759 data = c->rcvbuf.data;
763 ssize_t rxd = c->recv(c, data, len);
764 if(rxd < 0 || (size_t)rxd != len) {
765 // TODO: handle the application not accepting all data.
771 sack_consume(c, len);
777 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
778 if(!is_reliable(c)) {
779 c->recv(c, data, len);
780 c->rcv.nxt = seq + len;
784 uint32_t offset = seqdiff(seq, c->rcv.nxt);
785 if(offset + len > c->rcvbuf.maxsize)
789 handle_out_of_order(c, offset, data, len);
791 handle_in_order(c, data, len);
795 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
809 print_packet(utcp, "recv", data, len);
811 // Drop packets smaller than the header
814 if(len < sizeof(hdr)) {
819 // Make a copy from the potentially unaligned data to a struct hdr
821 memcpy(&hdr, data, sizeof(hdr));
825 // Drop packets with an unknown CTL flag
827 if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
832 // Check for auxiliary headers
834 const uint8_t *init = NULL;
836 uint16_t aux = hdr.aux;
838 size_t auxlen = 4 * (aux >> 8) & 0xf;
839 uint8_t auxtype = aux & 0xff;
848 if(!(hdr.ctl & SYN) || auxlen != 4) {
870 memcpy(&aux, data, 2);
875 // Try to match the packet to an existing connection
877 struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
879 // Is it for a new connection?
882 // Ignore RST packets
887 // Is it a SYN packet and are we LISTENing?
889 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
890 // If we don't want to accept it, send a RST back
891 if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
896 // Try to allocate memory, otherwise send a RST back
897 c = allocate_connection(utcp, hdr.dst, hdr.src);
903 // Parse auxilliary information
909 c->flags = init[3] & 0x7;
914 // Return SYN+ACK, go to SYN_RECEIVED state
915 c->snd.wnd = hdr.wnd;
916 c->rcv.irs = hdr.seq;
917 c->rcv.nxt = c->rcv.irs + 1;
918 set_state(c, SYN_RECEIVED);
925 pkt.hdr.src = c->src;
926 pkt.hdr.dst = c->dst;
927 pkt.hdr.ack = c->rcv.irs + 1;
928 pkt.hdr.seq = c->snd.iss;
929 pkt.hdr.wnd = c->rcv.wnd;
930 pkt.hdr.ctl = SYN | ACK;
932 pkt.hdr.aux = 0x0101;
936 pkt.data[3] = c->flags & 0x7;
937 print_packet(c->utcp, "send", &pkt, sizeof(hdr) + 4);
938 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
941 print_packet(c->utcp, "send", &pkt, sizeof(hdr));
942 utcp->send(utcp, &pkt, sizeof(hdr));
945 // No, we don't want your packets, send a RST back
953 debug("%p state %s\n", c->utcp, strstate[c->state]);
955 // In case this is for a CLOSED connection, ignore the packet.
956 // TODO: make it so incoming packets can never match a CLOSED connection.
958 if(c->state == CLOSED) {
959 debug("Got packet for closed connection\n");
963 // It is for an existing connection.
965 uint32_t prevrcvnxt = c->rcv.nxt;
967 // 1. Drop invalid packets.
969 // 1a. Drop packets that should not happen in our current state.
989 // 1b. Drop packets with a sequence number not in our receive window.
993 if(c->state == SYN_SENT)
996 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
998 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1000 // cut already accepted front overlapping
1001 if(rcv_offset < 0) {
1002 acceptable = len > (size_t)-rcv_offset;
1006 hdr.seq -= rcv_offset;
1009 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1014 debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1015 // Ignore unacceptable RST packets.
1018 // Otherwise, continue processing.
1022 c->snd.wnd = hdr.wnd; // TODO: move below
1024 // 1c. Drop packets with an invalid ACK.
1025 // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1026 // (= snd.una + c->sndbuf.used).
1028 if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1029 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1030 // Ignore unacceptable RST packets.
1036 // 2. Handle RST packets
1041 if(!(hdr.ctl & ACK))
1043 // The peer has refused our connection.
1044 set_state(c, CLOSED);
1045 errno = ECONNREFUSED;
1047 c->recv(c, NULL, 0);
1052 // We haven't told the application about this connection yet. Silently delete.
1061 // The peer has aborted our connection.
1062 set_state(c, CLOSED);
1065 c->recv(c, NULL, 0);
1072 // As far as the application is concerned, the connection has already been closed.
1073 // If it has called utcp_close() already, we can immediately free this connection.
1078 // Otherwise, immediately move to the CLOSED state.
1079 set_state(c, CLOSED);
1089 if(!(hdr.ctl & ACK))
1092 // 3. Advance snd.una
1094 uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
1095 prevrcvnxt = c->rcv.nxt;
1099 if(c->rtt_start.tv_sec) {
1100 if(c->rtt_seq == hdr.ack) {
1101 struct timeval now, diff;
1102 gettimeofday(&now, NULL);
1103 timersub(&now, &c->rtt_start, &diff);
1104 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
1105 c->rtt_start.tv_sec = 0;
1106 } else if(c->rtt_seq < hdr.ack) {
1107 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1108 c->rtt_start.tv_sec = 0;
1112 int32_t data_acked = advanced;
1119 // TODO: handle FIN as well.
1124 assert(data_acked >= 0);
1126 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1127 assert(data_acked <= bufused);
1130 buffer_get(&c->sndbuf, NULL, data_acked);
1132 // Also advance snd.nxt if possible
1133 if(seqdiff(c->snd.nxt, hdr.ack) < 0)
1134 c->snd.nxt = hdr.ack;
1136 c->snd.una = hdr.ack;
1139 c->snd.cwnd += utcp->mtu;
1140 if(c->snd.cwnd > c->sndbuf.maxsize)
1141 c->snd.cwnd = c->sndbuf.maxsize;
1143 // Check if we have sent a FIN that is now ACKed.
1146 if(c->snd.una == c->snd.last)
1147 set_state(c, FIN_WAIT_2);
1150 if(c->snd.una == c->snd.last) {
1151 gettimeofday(&c->conn_timeout, NULL);
1152 c->conn_timeout.tv_sec += 60;
1153 set_state(c, TIME_WAIT);
1160 if(!len && is_reliable(c)) {
1162 if(c->dupack == 3) {
1163 debug("Triplicate ACK\n");
1164 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1165 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1166 //Reset the congestion window so we wait for ACKs.
1167 c->snd.nxt = c->snd.una;
1168 c->snd.cwnd = utcp->mtu;
1169 start_retransmit_timer(c);
1177 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
1178 if(c->snd.una == c->snd.last)
1179 stop_retransmit_timer(c);
1180 else if(is_reliable(c))
1181 start_retransmit_timer(c);
1185 // 5. Process SYN stuff
1190 // This is a SYNACK. It should always have ACKed the SYN.
1193 c->rcv.irs = hdr.seq;
1194 c->rcv.nxt = hdr.seq;
1195 set_state(c, ESTABLISHED);
1196 // TODO: notify application of this somehow.
1206 // Ehm, no. We should never receive a second SYN.
1215 // SYN counts as one sequence number
1219 // 6. Process new data
1221 if(c->state == SYN_RECEIVED) {
1222 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1226 // Are we still LISTENing?
1228 utcp->accept(c, c->src);
1230 if(c->state != ESTABLISHED) {
1231 set_state(c, CLOSED);
1241 // This should never happen.
1254 // Ehm no, We should never receive more data after a FIN.
1263 handle_incoming_data(c, hdr.seq, data, len);
1266 // 7. Process FIN stuff
1268 if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
1272 // This should never happen.
1278 set_state(c, CLOSE_WAIT);
1281 set_state(c, CLOSING);
1284 gettimeofday(&c->conn_timeout, NULL);
1285 c->conn_timeout.tv_sec += 60;
1286 set_state(c, TIME_WAIT);
1292 // Ehm, no. We should never receive a second FIN.
1301 // FIN counts as one sequence number
1305 // Inform the application that the peer closed the connection.
1308 c->recv(c, NULL, 0);
1312 // Now we send something back if:
1313 // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
1314 // -> sendatleastone = true
1315 // - or we got an ack, so we should maybe send a bit more data
1316 // -> sendatleastone = false
1318 ack(c, len || prevrcvnxt != c->rcv.nxt);
1329 hdr.ack = hdr.seq + len;
1331 hdr.ctl = RST | ACK;
1333 print_packet(utcp, "send", &hdr, sizeof(hdr));
1334 utcp->send(utcp, &hdr, sizeof(hdr));
1339 int utcp_shutdown(struct utcp_connection *c, int dir) {
1340 debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1347 debug("Error: shutdown() called on closed connection %p\n", c);
1352 if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1357 // TCP does not have a provision for stopping incoming packets.
1358 // The best we can do is to just ignore them.
1359 if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR)
1362 // The rest of the code deals with shutting down writes.
1363 if(dir == UTCP_SHUT_RD)
1373 set_state(c, CLOSED);
1378 set_state(c, FIN_WAIT_1);
1384 set_state(c, CLOSING);
1396 if(!timerisset(&c->rtrx_timeout))
1397 start_retransmit_timer(c);
1401 int utcp_close(struct utcp_connection *c) {
1402 if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN)
1410 int utcp_abort(struct utcp_connection *c) {
1417 debug("Error: abort() called on closed connection %p\n", c);
1434 set_state(c, CLOSED);
1442 set_state(c, CLOSED);
1452 hdr.seq = c->snd.nxt;
1457 print_packet(c->utcp, "send", &hdr, sizeof(hdr));
1458 c->utcp->send(c->utcp, &hdr, sizeof(hdr));
1463 * One call to this function will loop through all connections,
1464 * checking if something needs to be resent or not.
1465 * The return value is the time to the next timeout in milliseconds,
1466 * or maybe a negative value if the timeout is infinite.
1468 struct timeval utcp_timeout(struct utcp *utcp) {
1470 gettimeofday(&now, NULL);
1471 struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1473 for(int i = 0; i < utcp->nconnections; i++) {
1474 struct utcp_connection *c = utcp->connections[i];
1478 // delete connections that have been utcp_close()d.
1479 if(c->state == CLOSED) {
1481 debug("Reaping %p\n", c);
1488 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1492 c->recv(c, NULL, 0);
1496 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1497 debug("retransmit()\n");
1502 if((c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1503 uint32_t len = buffer_free(&c->sndbuf);
1506 } else if(c->state == CLOSED) {
1511 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
1512 next = c->conn_timeout;
1514 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
1515 next = c->rtrx_timeout;
1518 struct timeval diff;
1519 timersub(&next, &now, &diff);
1523 bool utcp_is_active(struct utcp *utcp) {
1527 for(int i = 0; i < utcp->nconnections; i++)
1528 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT)
1534 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1540 struct utcp *utcp = calloc(1, sizeof(*utcp));
1544 utcp->accept = accept;
1545 utcp->pre_accept = pre_accept;
1548 utcp->mtu = DEFAULT_MTU;
1549 utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1550 utcp->rto = START_RTO; // usec
1555 void utcp_exit(struct utcp *utcp) {
1558 for(int i = 0; i < utcp->nconnections; i++) {
1559 struct utcp_connection *c = utcp->connections[i];
1562 c->recv(c, NULL, 0);
1563 buffer_exit(&c->rcvbuf);
1564 buffer_exit(&c->sndbuf);
1567 free(utcp->connections);
1571 uint16_t utcp_get_mtu(struct utcp *utcp) {
1572 return utcp ? utcp->mtu : 0;
1575 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1576 // TODO: handle overhead of the header
1581 void utcp_reset_timers(struct utcp *utcp) {
1584 struct timeval now, then;
1585 gettimeofday(&now, NULL);
1587 then.tv_sec += utcp->timeout;
1588 for(int i = 0; i < utcp->nconnections; i++) {
1589 utcp->connections[i]->rtrx_timeout = now;
1590 utcp->connections[i]->conn_timeout = then;
1591 utcp->connections[i]->rtt_start.tv_sec = 0;
1593 if(utcp->rto > START_RTO)
1594 utcp->rto = START_RTO;
1597 int utcp_get_user_timeout(struct utcp *u) {
1598 return u ? u->timeout : 0;
1601 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1603 u->timeout = timeout;
1606 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1607 return c ? c->sndbuf.maxsize : 0;
1610 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1611 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1612 return buffer_free(&c->sndbuf);
1617 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1620 c->sndbuf.maxsize = size;
1621 if(c->sndbuf.maxsize != size)
1622 c->sndbuf.maxsize = -1;
1625 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1626 return c ? c->rcvbuf.maxsize : 0;
1629 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1630 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1631 return buffer_free(&c->rcvbuf);
1636 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1639 c->rcvbuf.maxsize = size;
1640 if(c->rcvbuf.maxsize != size)
1641 c->rcvbuf.maxsize = -1;
1644 bool utcp_get_nodelay(struct utcp_connection *c) {
1645 return c ? c->nodelay : false;
1648 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1650 c->nodelay = nodelay;
1653 bool utcp_get_keepalive(struct utcp_connection *c) {
1654 return c ? c->keepalive : false;
1657 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1659 c->keepalive = keepalive;
1662 size_t utcp_get_outq(struct utcp_connection *c) {
1663 return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
1666 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
1671 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
1676 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
1678 utcp->accept = accept;
1679 utcp->pre_accept = pre_accept;