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.
23 #include "utcp_priv.h"
38 #if defined(CLOCK_MONOTONIC_RAW) && defined(__x86_64__)
39 #define UTCP_CLOCK CLOCK_MONOTONIC_RAW
41 #define UTCP_CLOCK CLOCK_MONOTONIC
45 static void timespec_sub(const struct timespec *a, const struct timespec *b, struct timespec *r) {
46 r->tv_sec = a->tv_sec - b->tv_sec;
47 r->tv_nsec = a->tv_nsec - b->tv_nsec;
50 r->tv_sec--, r->tv_nsec += NSEC_PER_SEC;
54 static int32_t timespec_diff_usec(const struct timespec *a, const struct timespec *b) {
55 return (a->tv_sec - b->tv_sec) * 1000000 + (a->tv_nsec - b->tv_nsec) / 1000;
58 static bool timespec_lt(const struct timespec *a, const struct timespec *b) {
59 if(a->tv_sec == b->tv_sec) {
60 return a->tv_nsec < b->tv_nsec;
62 return a->tv_sec < b->tv_sec;
66 static void timespec_clear(struct timespec *a) {
71 static bool timespec_isset(const struct timespec *a) {
75 static long CLOCK_GRANULARITY; // usec
77 static inline size_t min(size_t a, size_t b) {
81 static inline size_t max(size_t a, size_t b) {
88 #ifndef UTCP_DEBUG_DATALEN
89 #define UTCP_DEBUG_DATALEN 20
92 static void debug(struct utcp_connection *c, const char *format, ...) {
97 clock_gettime(CLOCK_REALTIME, &tv);
98 len = snprintf(buf, sizeof(buf), "%ld.%06lu %u:%u ", (long)tv.tv_sec, tv.tv_nsec / 1000, c ? c->src : 0, c ? c->dst : 0);
100 va_start(ap, format);
101 len += vsnprintf(buf + len, sizeof(buf) - len, format, ap);
104 if(len > 0 && (size_t)len < sizeof(buf)) {
105 fwrite(buf, len, 1, stderr);
109 static void print_packet(struct utcp_connection *c, const char *dir, const void *pkt, size_t len) {
112 if(len < sizeof(hdr)) {
113 debug(c, "%s: short packet (%lu bytes)\n", dir, (unsigned long)len);
117 memcpy(&hdr, pkt, sizeof(hdr));
121 if(len > sizeof(hdr)) {
122 datalen = min(len - sizeof(hdr), UTCP_DEBUG_DATALEN);
128 const uint8_t *data = (uint8_t *)pkt + sizeof(hdr);
129 char str[datalen * 2 + 1];
132 for(uint32_t i = 0; i < datalen; i++) {
133 *p++ = "0123456789ABCDEF"[data[i] >> 4];
134 *p++ = "0123456789ABCDEF"[data[i] & 15];
139 debug(c, "%s: len %lu src %u dst %u seq %u ack %u wnd %u aux %x ctl %s%s%s%s%s data %s\n",
140 dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux,
141 hdr.ctl & SYN ? "SYN" : "",
142 hdr.ctl & RST ? "RST" : "",
143 hdr.ctl & FIN ? "FIN" : "",
144 hdr.ctl & ACK ? "ACK" : "",
145 hdr.ctl & MF ? "MF" : "",
150 static void debug_cwnd(struct utcp_connection *c) {
151 debug(c, "snd.cwnd %u snd.ssthresh %u\n", c->snd.cwnd, ~c->snd.ssthresh ? c->snd.ssthresh : 0);
154 #define debug(...) do {} while(0)
155 #define print_packet(...) do {} while(0)
156 #define debug_cwnd(...) do {} while(0)
159 static void set_state(struct utcp_connection *c, enum state state) {
162 if(state == ESTABLISHED) {
163 timespec_clear(&c->conn_timeout);
166 debug(c, "state %s\n", strstate[state]);
169 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
170 if(seq != c->snd.last) {
185 static bool is_reliable(struct utcp_connection *c) {
186 return c->flags & UTCP_RELIABLE;
189 static int32_t seqdiff(uint32_t a, uint32_t b) {
194 static bool buffer_wraps(struct buffer *buf) {
195 return buf->size - buf->offset < buf->used;
198 static bool buffer_resize(struct buffer *buf, uint32_t newsize) {
199 char *newdata = realloc(buf->data, newsize);
207 if(buffer_wraps(buf)) {
208 // Shift the right part of the buffer until it hits the end of the new buffer.
212 // [345.........|........012]
213 uint32_t tailsize = buf->size - buf->offset;
214 uint32_t newoffset = newsize - tailsize;
215 memmove(buf->data + newoffset, buf->data + buf->offset, tailsize);
216 buf->offset = newoffset;
223 // Store data into the buffer
224 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
225 debug(NULL, "buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
227 // Ensure we don't store more than maxsize bytes in total
228 size_t required = offset + len;
230 if(required > buf->maxsize) {
231 if(offset >= buf->maxsize) {
235 len = buf->maxsize - offset;
236 required = buf->maxsize;
239 // Check if we need to resize the buffer
240 if(required > buf->size) {
241 size_t newsize = buf->size;
249 } while(newsize < required);
251 if(newsize > buf->maxsize) {
252 newsize = buf->maxsize;
255 if(!buffer_resize(buf, newsize)) {
260 uint32_t realoffset = buf->offset + offset;
262 if(buf->size - buf->offset <= offset) {
263 // The offset wrapped
264 realoffset -= buf->size;
267 if(buf->size - realoffset < len) {
268 // The new chunk of data must be wrapped
269 memcpy(buf->data + realoffset, data, buf->size - realoffset);
270 memcpy(buf->data, (char *)data + buf->size - realoffset, len - (buf->size - realoffset));
272 memcpy(buf->data + realoffset, data, len);
275 if(required > buf->used) {
276 buf->used = required;
282 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
283 return buffer_put_at(buf, buf->used, data, len);
286 // Copy data from the buffer without removing it.
287 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
288 // Ensure we don't copy more than is actually stored in the buffer
289 if(offset >= buf->used) {
293 if(buf->used - offset < len) {
294 len = buf->used - offset;
297 uint32_t realoffset = buf->offset + offset;
299 if(buf->size - buf->offset <= offset) {
300 // The offset wrapped
301 realoffset -= buf->size;
304 if(buf->size - realoffset < len) {
305 // The data is wrapped
306 memcpy(data, buf->data + realoffset, buf->size - realoffset);
307 memcpy((char *)data + buf->size - realoffset, buf->data, len - (buf->size - realoffset));
309 memcpy(data, buf->data + realoffset, len);
315 // Copy data from the buffer without removing it.
316 static ssize_t buffer_call(struct utcp_connection *c, struct buffer *buf, size_t offset, size_t len) {
321 // Ensure we don't copy more than is actually stored in the buffer
322 if(offset >= buf->used) {
326 if(buf->used - offset < len) {
327 len = buf->used - offset;
330 uint32_t realoffset = buf->offset + offset;
332 if(buf->size - buf->offset <= offset) {
333 // The offset wrapped
334 realoffset -= buf->size;
337 if(buf->size - realoffset < len) {
338 // The data is wrapped
339 ssize_t rx1 = c->recv(c, buf->data + realoffset, buf->size - realoffset);
341 if(rx1 < buf->size - realoffset) {
345 // The channel might have been closed by the previous callback
350 ssize_t rx2 = c->recv(c, buf->data, len - (buf->size - realoffset));
358 return c->recv(c, buf->data + realoffset, len);
362 // Discard data from the buffer.
363 static ssize_t buffer_discard(struct buffer *buf, size_t len) {
364 if(buf->used < len) {
368 if(buf->size - buf->offset <= len) {
369 buf->offset -= buf->size;
372 if(buf->used == len) {
383 static void buffer_clear(struct buffer *buf) {
388 static bool buffer_set_size(struct buffer *buf, uint32_t minsize, uint32_t maxsize) {
389 if(maxsize < minsize) {
393 buf->maxsize = maxsize;
395 return buf->size >= minsize || buffer_resize(buf, minsize);
398 static void buffer_exit(struct buffer *buf) {
403 memset(buf, 0, sizeof(*buf));
406 static uint32_t buffer_free(const struct buffer *buf) {
407 return buf->maxsize > buf->used ? buf->maxsize - buf->used : 0;
410 // Connections are stored in a sorted list.
411 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
413 static int compare(const void *va, const void *vb) {
416 const struct utcp_connection *a = *(struct utcp_connection **)va;
417 const struct utcp_connection *b = *(struct utcp_connection **)vb;
421 int c = (int)a->src - (int)b->src;
427 c = (int)a->dst - (int)b->dst;
431 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
432 if(!utcp->nconnections) {
436 struct utcp_connection key = {
440 struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
441 return match ? *match : NULL;
444 static void free_connection(struct utcp_connection *c) {
445 struct utcp *utcp = c->utcp;
446 struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
450 int i = cp - utcp->connections;
451 memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
452 utcp->nconnections--;
454 buffer_exit(&c->rcvbuf);
455 buffer_exit(&c->sndbuf);
459 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
460 // Check whether this combination of src and dst is free
463 if(find_connection(utcp, src, dst)) {
467 } else { // If src == 0, generate a random port number with the high bit set
468 if(utcp->nconnections >= 32767) {
473 src = rand() | 0x8000;
475 while(find_connection(utcp, src, dst)) {
480 // Allocate memory for the new connection
482 if(utcp->nconnections >= utcp->nallocated) {
483 if(!utcp->nallocated) {
484 utcp->nallocated = 4;
486 utcp->nallocated *= 2;
489 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
495 utcp->connections = new_array;
498 struct utcp_connection *c = calloc(1, sizeof(*c));
504 if(!buffer_set_size(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
509 if(!buffer_set_size(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
510 buffer_exit(&c->sndbuf);
515 // Fill in the details
524 c->snd.una = c->snd.iss;
525 c->snd.nxt = c->snd.iss + 1;
526 c->snd.last = c->snd.nxt;
527 c->snd.cwnd = (utcp->mss > 2190 ? 2 : utcp->mss > 1095 ? 3 : 4) * utcp->mss;
528 c->snd.ssthresh = ~0;
535 // Add it to the sorted list of connections
537 utcp->connections[utcp->nconnections++] = c;
538 qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
543 static inline uint32_t absdiff(uint32_t a, uint32_t b) {
551 // Update RTT variables. See RFC 6298.
552 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
554 debug(c, "invalid rtt\n");
562 c->rttvar = (c->rttvar * 3 + absdiff(c->srtt, rtt)) / 4;
563 c->srtt = (c->srtt * 7 + rtt) / 8;
566 c->rto = c->srtt + max(4 * c->rttvar, CLOCK_GRANULARITY);
568 if(c->rto > MAX_RTO) {
572 debug(c, "rtt %u srtt %u rttvar %u rto %u\n", rtt, c->srtt, c->rttvar, c->rto);
575 static void start_retransmit_timer(struct utcp_connection *c) {
576 clock_gettime(UTCP_CLOCK, &c->rtrx_timeout);
578 uint32_t rto = c->rto;
580 while(rto > USEC_PER_SEC) {
581 c->rtrx_timeout.tv_sec++;
585 c->rtrx_timeout.tv_nsec += rto * 1000;
587 if(c->rtrx_timeout.tv_nsec >= NSEC_PER_SEC) {
588 c->rtrx_timeout.tv_nsec -= NSEC_PER_SEC;
589 c->rtrx_timeout.tv_sec++;
592 debug(c, "rtrx_timeout %ld.%06lu\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_nsec);
595 static void stop_retransmit_timer(struct utcp_connection *c) {
596 timespec_clear(&c->rtrx_timeout);
597 debug(c, "rtrx_timeout cleared\n");
600 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
601 struct utcp_connection *c = allocate_connection(utcp, 0, dst);
607 assert((flags & ~0x1f) == 0);
618 pkt.hdr.src = c->src;
619 pkt.hdr.dst = c->dst;
620 pkt.hdr.seq = c->snd.iss;
622 pkt.hdr.wnd = c->rcvbuf.maxsize;
624 pkt.hdr.aux = 0x0101;
628 pkt.init[3] = flags & 0x7;
630 set_state(c, SYN_SENT);
632 print_packet(c, "send", &pkt, sizeof(pkt));
633 utcp->send(utcp, &pkt, sizeof(pkt));
635 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
636 c->conn_timeout.tv_sec += utcp->timeout;
638 start_retransmit_timer(c);
643 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
644 return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
647 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
648 if(c->reapable || c->state != SYN_RECEIVED) {
649 debug(c, "accept() called on invalid connection in state %s\n", c, strstate[c->state]);
653 debug(c, "accepted %p %p\n", c, recv, priv);
657 set_state(c, ESTABLISHED);
660 static void ack(struct utcp_connection *c, bool sendatleastone) {
661 int32_t left = seqdiff(c->snd.last, c->snd.nxt);
662 int32_t cwndleft = is_reliable(c) ? min(c->snd.cwnd, c->snd.wnd) - seqdiff(c->snd.nxt, c->snd.una) : MAX_UNRELIABLE_SIZE;
668 } else if(cwndleft < left) {
671 if(!sendatleastone || cwndleft > c->utcp->mss) {
672 left -= left % c->utcp->mss;
676 debug(c, "cwndleft %d left %d\n", cwndleft, left);
678 if(!left && !sendatleastone) {
685 } *pkt = c->utcp->pkt;
687 pkt->hdr.src = c->src;
688 pkt->hdr.dst = c->dst;
689 pkt->hdr.ack = c->rcv.nxt;
690 pkt->hdr.wnd = is_reliable(c) ? c->rcvbuf.maxsize : 0;
695 uint32_t seglen = left > c->utcp->mss ? c->utcp->mss : left;
696 pkt->hdr.seq = c->snd.nxt;
698 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
700 c->snd.nxt += seglen;
703 if(!is_reliable(c)) {
711 if(seglen && fin_wanted(c, c->snd.nxt)) {
716 if(!c->rtt_start.tv_sec) {
717 // Start RTT measurement
718 clock_gettime(UTCP_CLOCK, &c->rtt_start);
719 c->rtt_seq = pkt->hdr.seq + seglen;
720 debug(c, "starting RTT measurement, expecting ack %u\n", c->rtt_seq);
723 print_packet(c, "send", pkt, sizeof(pkt->hdr) + seglen);
724 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
726 if(left && !is_reliable(c)) {
727 pkt->hdr.wnd += seglen;
732 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
734 debug(c, "send() called on closed connection\n");
742 debug(c, "send() called on unconnected connection\n");
757 debug(c, "send() called on closed connection\n");
762 // Exit early if we have nothing to send.
773 // Check if we need to be able to buffer all data
775 if(c->flags & UTCP_NO_PARTIAL) {
776 if(len > buffer_free(&c->sndbuf)) {
777 if(len > c->sndbuf.maxsize) {
787 // Add data to send buffer.
790 len = buffer_put(&c->sndbuf, data, len);
791 } else if(c->state != SYN_SENT && c->state != SYN_RECEIVED) {
792 if(len > MAX_UNRELIABLE_SIZE || buffer_put(&c->sndbuf, data, len) != (ssize_t)len) {
811 // Don't send anything yet if the connection has not fully established yet
813 if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
819 if(!is_reliable(c)) {
820 c->snd.una = c->snd.nxt = c->snd.last;
821 buffer_discard(&c->sndbuf, c->sndbuf.used);
824 if(is_reliable(c) && !timespec_isset(&c->rtrx_timeout)) {
825 start_retransmit_timer(c);
828 if(is_reliable(c) && !timespec_isset(&c->conn_timeout)) {
829 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
830 c->conn_timeout.tv_sec += c->utcp->timeout;
836 static void swap_ports(struct hdr *hdr) {
837 uint16_t tmp = hdr->src;
842 static void fast_retransmit(struct utcp_connection *c) {
843 if(c->state == CLOSED || c->snd.last == c->snd.una) {
844 debug(c, "fast_retransmit() called but nothing to retransmit!\n");
848 struct utcp *utcp = c->utcp;
853 } *pkt = c->utcp->pkt;
855 pkt->hdr.src = c->src;
856 pkt->hdr.dst = c->dst;
857 pkt->hdr.wnd = c->rcvbuf.maxsize;
866 // Send unacked data again.
867 pkt->hdr.seq = c->snd.una;
868 pkt->hdr.ack = c->rcv.nxt;
870 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
872 if(fin_wanted(c, c->snd.una + len)) {
877 buffer_copy(&c->sndbuf, pkt->data, 0, len);
878 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
879 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
887 static void retransmit(struct utcp_connection *c) {
888 if(c->state == CLOSED || c->snd.last == c->snd.una) {
889 debug(c, "retransmit() called but nothing to retransmit!\n");
890 stop_retransmit_timer(c);
894 struct utcp *utcp = c->utcp;
896 if(utcp->retransmit) {
903 } *pkt = c->utcp->pkt;
905 pkt->hdr.src = c->src;
906 pkt->hdr.dst = c->dst;
907 pkt->hdr.wnd = c->rcvbuf.maxsize;
912 // Send our SYN again
913 pkt->hdr.seq = c->snd.iss;
916 pkt->hdr.aux = 0x0101;
920 pkt->data[3] = c->flags & 0x7;
921 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + 4);
922 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
927 pkt->hdr.seq = c->snd.nxt;
928 pkt->hdr.ack = c->rcv.nxt;
929 pkt->hdr.ctl = SYN | ACK;
930 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr));
931 utcp->send(utcp, pkt, sizeof(pkt->hdr));
939 // Send unacked data again.
940 pkt->hdr.seq = c->snd.una;
941 pkt->hdr.ack = c->rcv.nxt;
943 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
945 if(fin_wanted(c, c->snd.una + len)) {
950 // RFC 5681 slow start after timeout
951 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
952 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
953 c->snd.cwnd = utcp->mss;
956 buffer_copy(&c->sndbuf, pkt->data, 0, len);
957 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
958 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
960 c->snd.nxt = c->snd.una + len;
967 // We shouldn't need to retransmit anything in this state.
971 stop_retransmit_timer(c);
975 start_retransmit_timer(c);
978 if(c->rto > MAX_RTO) {
982 c->rtt_start.tv_sec = 0; // invalidate RTT timer
983 c->dupack = 0; // cancel any ongoing fast recovery
989 /* Update receive buffer and SACK entries after consuming data.
993 * |.....0000..1111111111.....22222......3333|
996 * 0..3 represent the SACK entries. The ^ indicates up to which point we want
997 * to remove data from the receive buffer. The idea is to substract "len"
998 * from the offset of all the SACK entries, and then remove/cut down entries
999 * that are shifted to before the start of the receive buffer.
1001 * There are three cases:
1002 * - the SACK entry is after ^, in that case just change the offset.
1003 * - the SACK entry starts before and ends after ^, so we have to
1004 * change both its offset and size.
1005 * - the SACK entry is completely before ^, in that case delete it.
1007 static void sack_consume(struct utcp_connection *c, size_t len) {
1008 debug(c, "sack_consume %lu\n", (unsigned long)len);
1010 if(len > c->rcvbuf.used) {
1011 debug(c, "all SACK entries consumed\n");
1012 c->sacks[0].len = 0;
1016 buffer_discard(&c->rcvbuf, len);
1018 for(int i = 0; i < NSACKS && c->sacks[i].len;) {
1019 if(len < c->sacks[i].offset) {
1020 c->sacks[i].offset -= len;
1022 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
1023 c->sacks[i].len -= len - c->sacks[i].offset;
1024 c->sacks[i].offset = 0;
1027 if(i < NSACKS - 1) {
1028 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
1029 c->sacks[NSACKS - 1].len = 0;
1031 c->sacks[i].len = 0;
1037 for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1038 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1042 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
1043 debug(c, "out of order packet, offset %u\n", offset);
1044 // Packet loss or reordering occured. Store the data in the buffer.
1045 ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
1048 debug(c, "packet outside receive buffer, dropping\n");
1052 if((size_t)rxd < len) {
1053 debug(c, "packet partially outside receive buffer\n");
1057 // Make note of where we put it.
1058 for(int i = 0; i < NSACKS; i++) {
1059 if(!c->sacks[i].len) { // nothing to merge, add new entry
1060 debug(c, "new SACK entry %d\n", i);
1061 c->sacks[i].offset = offset;
1062 c->sacks[i].len = rxd;
1064 } else if(offset < c->sacks[i].offset) {
1065 if(offset + rxd < c->sacks[i].offset) { // insert before
1066 if(!c->sacks[NSACKS - 1].len) { // only if room left
1067 debug(c, "insert SACK entry at %d\n", i);
1068 memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
1069 c->sacks[i].offset = offset;
1070 c->sacks[i].len = rxd;
1072 debug(c, "SACK entries full, dropping packet\n");
1077 debug(c, "merge with start of SACK entry at %d\n", i);
1078 c->sacks[i].offset = offset;
1081 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
1082 if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
1083 debug(c, "merge with end of SACK entry at %d\n", i);
1084 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
1085 // TODO: handle potential merge with next entry
1092 for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1093 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1097 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
1099 ssize_t rxd = c->recv(c, data, len);
1101 if(rxd != (ssize_t)len) {
1102 // TODO: handle the application not accepting all data.
1107 // Check if we can process out-of-order data now.
1108 if(c->sacks[0].len && len >= c->sacks[0].offset) {
1109 debug(c, "incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
1111 if(len < c->sacks[0].offset + c->sacks[0].len) {
1112 size_t offset = len;
1113 len = c->sacks[0].offset + c->sacks[0].len;
1114 size_t remainder = len - offset;
1116 ssize_t rxd = buffer_call(c, &c->rcvbuf, offset, remainder);
1118 if(rxd != (ssize_t)remainder) {
1119 // TODO: handle the application not accepting all data.
1125 if(c->rcvbuf.used) {
1126 sack_consume(c, len);
1132 static void handle_unreliable(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1133 // Fast path for unfragmented packets
1134 if(!hdr->wnd && !(hdr->ctl & MF)) {
1136 c->recv(c, data, len);
1139 c->rcv.nxt = hdr->seq + len;
1143 // Ensure reassembled packet are not larger than 64 kiB
1144 if(hdr->wnd >= MAX_UNRELIABLE_SIZE || hdr->wnd + len > MAX_UNRELIABLE_SIZE) {
1148 // Don't accept out of order fragments
1149 if(hdr->wnd && hdr->seq != c->rcv.nxt) {
1153 // Reset the receive buffer for the first fragment
1155 buffer_clear(&c->rcvbuf);
1158 ssize_t rxd = buffer_put_at(&c->rcvbuf, hdr->wnd, data, len);
1160 if(rxd != (ssize_t)len) {
1164 // Send the packet if it's the final fragment
1165 if(!(hdr->ctl & MF)) {
1166 buffer_call(c, &c->rcvbuf, 0, hdr->wnd + len);
1169 c->rcv.nxt = hdr->seq + len;
1172 static void handle_incoming_data(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1173 if(!is_reliable(c)) {
1174 handle_unreliable(c, hdr, data, len);
1178 uint32_t offset = seqdiff(hdr->seq, c->rcv.nxt);
1181 handle_out_of_order(c, offset, data, len);
1183 handle_in_order(c, data, len);
1188 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
1189 const uint8_t *ptr = data;
1205 // Drop packets smaller than the header
1209 if(len < sizeof(hdr)) {
1210 print_packet(NULL, "recv", data, len);
1215 // Make a copy from the potentially unaligned data to a struct hdr
1217 memcpy(&hdr, ptr, sizeof(hdr));
1219 // Try to match the packet to an existing connection
1221 struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1222 print_packet(c, "recv", data, len);
1224 // Process the header
1229 // Drop packets with an unknown CTL flag
1231 if(hdr.ctl & ~(SYN | ACK | RST | FIN | MF)) {
1232 print_packet(NULL, "recv", data, len);
1237 // Check for auxiliary headers
1239 const uint8_t *init = NULL;
1241 uint16_t aux = hdr.aux;
1244 size_t auxlen = 4 * (aux >> 8) & 0xf;
1245 uint8_t auxtype = aux & 0xff;
1254 if(!(hdr.ctl & SYN) || auxlen != 4) {
1270 if(!(aux & 0x800)) {
1279 memcpy(&aux, ptr, 2);
1284 bool has_data = len || (hdr.ctl & (SYN | FIN));
1286 // Is it for a new connection?
1289 // Ignore RST packets
1295 // Is it a SYN packet and are we LISTENing?
1297 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1298 // If we don't want to accept it, send a RST back
1299 if((utcp->listen && !utcp->listen(utcp, hdr.dst))) {
1304 // Try to allocate memory, otherwise send a RST back
1305 c = allocate_connection(utcp, hdr.dst, hdr.src);
1312 // Parse auxilliary information
1319 c->flags = init[3] & 0x7;
1321 c->flags = UTCP_TCP;
1325 // Return SYN+ACK, go to SYN_RECEIVED state
1326 c->snd.wnd = hdr.wnd;
1327 c->rcv.irs = hdr.seq;
1328 c->rcv.nxt = c->rcv.irs + 1;
1329 set_state(c, SYN_RECEIVED);
1336 pkt.hdr.src = c->src;
1337 pkt.hdr.dst = c->dst;
1338 pkt.hdr.ack = c->rcv.irs + 1;
1339 pkt.hdr.seq = c->snd.iss;
1340 pkt.hdr.wnd = c->rcvbuf.maxsize;
1341 pkt.hdr.ctl = SYN | ACK;
1344 pkt.hdr.aux = 0x0101;
1348 pkt.data[3] = c->flags & 0x7;
1349 print_packet(c, "send", &pkt, sizeof(hdr) + 4);
1350 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1353 print_packet(c, "send", &pkt, sizeof(hdr));
1354 utcp->send(utcp, &pkt, sizeof(hdr));
1357 start_retransmit_timer(c);
1359 // No, we don't want your packets, send a RST back
1367 debug(c, "state %s\n", strstate[c->state]);
1369 // In case this is for a CLOSED connection, ignore the packet.
1370 // TODO: make it so incoming packets can never match a CLOSED connection.
1372 if(c->state == CLOSED) {
1373 debug(c, "got packet for closed connection\n");
1377 // It is for an existing connection.
1379 // 1. Drop invalid packets.
1381 // 1a. Drop packets that should not happen in our current state.
1402 // 1b. Discard data that is not in our receive window.
1404 if(is_reliable(c)) {
1407 if(c->state == SYN_SENT) {
1409 } else if(len == 0) {
1410 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1412 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1414 // cut already accepted front overlapping
1415 if(rcv_offset < 0) {
1416 acceptable = len > (size_t) - rcv_offset;
1421 hdr.seq -= rcv_offset;
1424 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1429 debug(c, "packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1431 // Ignore unacceptable RST packets.
1436 // Otherwise, continue processing.
1441 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1444 debug(c, "packet out of order, offset %u bytes", rcv_offset);
1450 c->snd.wnd = hdr.wnd; // TODO: move below
1452 // 1c. Drop packets with an invalid ACK.
1453 // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1454 // (= snd.una + c->sndbuf.used).
1456 if(!is_reliable(c)) {
1457 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1458 hdr.ack = c->snd.una;
1462 // 2. Handle RST packets
1467 if(!(hdr.ctl & ACK)) {
1471 // The peer has refused our connection.
1472 set_state(c, CLOSED);
1473 errno = ECONNREFUSED;
1476 c->recv(c, NULL, 0);
1479 if(c->poll && !c->reapable) {
1490 // We haven't told the application about this connection yet. Silently delete.
1502 // The peer has aborted our connection.
1503 set_state(c, CLOSED);
1507 c->recv(c, NULL, 0);
1510 if(c->poll && !c->reapable) {
1523 // As far as the application is concerned, the connection has already been closed.
1524 // If it has called utcp_close() already, we can immediately free this connection.
1530 // Otherwise, immediately move to the CLOSED state.
1531 set_state(c, CLOSED);
1544 if(!(hdr.ctl & ACK)) {
1549 // 3. Advance snd.una
1551 if(seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0) {
1552 debug(c, "packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1556 advanced = seqdiff(hdr.ack, c->snd.una);
1560 if(c->rtt_start.tv_sec) {
1561 if(c->rtt_seq == hdr.ack) {
1562 struct timespec now;
1563 clock_gettime(UTCP_CLOCK, &now);
1564 int32_t diff = timespec_diff_usec(&now, &c->rtt_start);
1565 update_rtt(c, diff);
1566 c->rtt_start.tv_sec = 0;
1567 } else if(c->rtt_seq < hdr.ack) {
1568 debug(c, "cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1569 c->rtt_start.tv_sec = 0;
1573 int32_t data_acked = advanced;
1581 // TODO: handle FIN as well.
1586 assert(data_acked >= 0);
1589 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1590 assert(data_acked <= bufused);
1594 buffer_discard(&c->sndbuf, data_acked);
1596 if(is_reliable(c)) {
1601 // Also advance snd.nxt if possible
1602 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1603 c->snd.nxt = hdr.ack;
1606 c->snd.una = hdr.ack;
1609 if(c->dupack >= 3) {
1610 debug(c, "fast recovery ended\n");
1611 c->snd.cwnd = c->snd.ssthresh;
1617 // Increase the congestion window according to RFC 5681
1618 if(c->snd.cwnd < c->snd.ssthresh) {
1619 c->snd.cwnd += min(advanced, utcp->mss); // eq. 2
1621 c->snd.cwnd += max(1, (utcp->mss * utcp->mss) / c->snd.cwnd); // eq. 3
1624 if(c->snd.cwnd > c->sndbuf.maxsize) {
1625 c->snd.cwnd = c->sndbuf.maxsize;
1630 // Check if we have sent a FIN that is now ACKed.
1633 if(c->snd.una == c->snd.last) {
1634 set_state(c, FIN_WAIT_2);
1640 if(c->snd.una == c->snd.last) {
1641 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1642 c->conn_timeout.tv_sec += utcp->timeout;
1643 set_state(c, TIME_WAIT);
1652 if(!len && is_reliable(c) && c->snd.una != c->snd.last) {
1654 debug(c, "duplicate ACK %d\n", c->dupack);
1656 if(c->dupack == 3) {
1657 // RFC 5681 fast recovery
1658 debug(c, "fast recovery started\n", c->dupack);
1659 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
1660 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
1661 c->snd.cwnd = min(c->snd.ssthresh + 3 * utcp->mss, c->sndbuf.maxsize);
1663 if(c->snd.cwnd > c->sndbuf.maxsize) {
1664 c->snd.cwnd = c->sndbuf.maxsize;
1670 } else if(c->dupack > 3) {
1671 c->snd.cwnd += utcp->mss;
1673 if(c->snd.cwnd > c->sndbuf.maxsize) {
1674 c->snd.cwnd = c->sndbuf.maxsize;
1680 // We got an ACK which indicates the other side did get one of our packets.
1681 // Reset the retransmission timer to avoid going to slow start,
1682 // but don't touch the connection timeout.
1683 start_retransmit_timer(c);
1690 if(c->snd.una == c->snd.last) {
1691 stop_retransmit_timer(c);
1692 timespec_clear(&c->conn_timeout);
1693 } else if(is_reliable(c)) {
1694 start_retransmit_timer(c);
1695 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1696 c->conn_timeout.tv_sec += utcp->timeout;
1701 // 5. Process SYN stuff
1707 // This is a SYNACK. It should always have ACKed the SYN.
1712 c->rcv.irs = hdr.seq;
1713 c->rcv.nxt = hdr.seq + 1;
1717 set_state(c, FIN_WAIT_1);
1720 set_state(c, ESTABLISHED);
1726 // This is a retransmit of a SYN, send back the SYNACK.
1736 // This could be a retransmission. Ignore the SYN flag, but send an ACK back.
1747 // 6. Process new data
1749 if(c->state == SYN_RECEIVED) {
1750 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1755 // Are we still LISTENing?
1757 utcp->accept(c, c->src);
1760 if(c->state != ESTABLISHED) {
1761 set_state(c, CLOSED);
1771 // This should never happen.
1783 // We already closed the connection and are not interested in more data.
1793 // Ehm no, We should never receive more data after a FIN.
1803 handle_incoming_data(c, &hdr, ptr, len);
1806 // 7. Process FIN stuff
1808 if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1812 // This should never happen.
1819 set_state(c, CLOSE_WAIT);
1823 set_state(c, CLOSING);
1827 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1828 c->conn_timeout.tv_sec += utcp->timeout;
1829 set_state(c, TIME_WAIT);
1836 // Ehm, no. We should never receive a second FIN.
1846 // FIN counts as one sequence number
1850 // Inform the application that the peer closed its end of the connection.
1853 c->recv(c, NULL, 0);
1857 // Now we send something back if:
1858 // - we received data, so we have to send back an ACK
1859 // -> sendatleastone = true
1860 // - or we got an ack, so we should maybe send a bit more data
1861 // -> sendatleastone = false
1863 if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1878 hdr.ack = hdr.seq + len;
1880 hdr.ctl = RST | ACK;
1883 print_packet(c, "send", &hdr, sizeof(hdr));
1884 utcp->send(utcp, &hdr, sizeof(hdr));
1889 int utcp_shutdown(struct utcp_connection *c, int dir) {
1890 debug(c, "shutdown %d at %u\n", dir, c ? c->snd.last : 0);
1898 debug(c, "shutdown() called on closed connection\n");
1903 if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1908 // TCP does not have a provision for stopping incoming packets.
1909 // The best we can do is to just ignore them.
1910 if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1914 // The rest of the code deals with shutting down writes.
1915 if(dir == UTCP_SHUT_RD) {
1919 // Only process shutting down writes once.
1937 set_state(c, FIN_WAIT_1);
1945 set_state(c, CLOSING);
1958 if(!timespec_isset(&c->rtrx_timeout)) {
1959 start_retransmit_timer(c);
1965 static bool reset_connection(struct utcp_connection *c) {
1972 debug(c, "abort() called on closed connection\n");
1989 set_state(c, CLOSED);
1997 set_state(c, CLOSED);
2007 hdr.seq = c->snd.nxt;
2008 hdr.ack = c->rcv.nxt;
2012 print_packet(c, "send", &hdr, sizeof(hdr));
2013 c->utcp->send(c->utcp, &hdr, sizeof(hdr));
2017 // Closes all the opened connections
2018 void utcp_abort_all_connections(struct utcp *utcp) {
2024 for(int i = 0; i < utcp->nconnections; i++) {
2025 struct utcp_connection *c = utcp->connections[i];
2027 if(c->reapable || c->state == CLOSED) {
2031 utcp_recv_t old_recv = c->recv;
2032 utcp_poll_t old_poll = c->poll;
2034 reset_connection(c);
2038 old_recv(c, NULL, 0);
2041 if(old_poll && !c->reapable) {
2050 int utcp_close(struct utcp_connection *c) {
2051 if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
2061 int utcp_abort(struct utcp_connection *c) {
2062 if(!reset_connection(c)) {
2071 * One call to this function will loop through all connections,
2072 * checking if something needs to be resent or not.
2073 * The return value is the time to the next timeout in milliseconds,
2074 * or maybe a negative value if the timeout is infinite.
2076 struct timespec utcp_timeout(struct utcp *utcp) {
2077 struct timespec now;
2078 clock_gettime(UTCP_CLOCK, &now);
2079 struct timespec next = {now.tv_sec + 3600, now.tv_nsec};
2081 for(int i = 0; i < utcp->nconnections; i++) {
2082 struct utcp_connection *c = utcp->connections[i];
2088 // delete connections that have been utcp_close()d.
2089 if(c->state == CLOSED) {
2091 debug(c, "reaping\n");
2099 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &now)) {
2104 c->recv(c, NULL, 0);
2107 if(c->poll && !c->reapable) {
2114 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &now)) {
2115 debug(c, "retransmitting after timeout\n");
2120 if((c->state == ESTABLISHED || c->state == CLOSE_WAIT) && c->do_poll) {
2122 uint32_t len = buffer_free(&c->sndbuf);
2127 } else if(c->state == CLOSED) {
2132 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &next)) {
2133 next = c->conn_timeout;
2136 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &next)) {
2137 next = c->rtrx_timeout;
2141 struct timespec diff;
2143 timespec_sub(&next, &now, &diff);
2148 bool utcp_is_active(struct utcp *utcp) {
2153 for(int i = 0; i < utcp->nconnections; i++)
2154 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
2161 struct utcp *utcp_init(utcp_accept_t accept, utcp_listen_t listen, utcp_send_t send, void *priv) {
2167 struct utcp *utcp = calloc(1, sizeof(*utcp));
2173 utcp_set_mtu(utcp, DEFAULT_MTU);
2180 if(!CLOCK_GRANULARITY) {
2181 struct timespec res;
2182 clock_getres(UTCP_CLOCK, &res);
2183 CLOCK_GRANULARITY = res.tv_sec * USEC_PER_SEC + res.tv_nsec / 1000;
2186 utcp->accept = accept;
2187 utcp->listen = listen;
2190 utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
2195 void utcp_exit(struct utcp *utcp) {
2200 for(int i = 0; i < utcp->nconnections; i++) {
2201 struct utcp_connection *c = utcp->connections[i];
2205 c->recv(c, NULL, 0);
2208 if(c->poll && !c->reapable) {
2213 buffer_exit(&c->rcvbuf);
2214 buffer_exit(&c->sndbuf);
2218 free(utcp->connections);
2223 uint16_t utcp_get_mtu(struct utcp *utcp) {
2224 return utcp ? utcp->mtu : 0;
2227 uint16_t utcp_get_mss(struct utcp *utcp) {
2228 return utcp ? utcp->mss : 0;
2231 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
2236 if(mtu <= sizeof(struct hdr)) {
2240 if(mtu > utcp->mtu) {
2241 char *new = realloc(utcp->pkt, mtu + sizeof(struct hdr));
2251 utcp->mss = mtu - sizeof(struct hdr);
2254 void utcp_reset_timers(struct utcp *utcp) {
2259 struct timespec now, then;
2261 clock_gettime(UTCP_CLOCK, &now);
2265 then.tv_sec += utcp->timeout;
2267 for(int i = 0; i < utcp->nconnections; i++) {
2268 struct utcp_connection *c = utcp->connections[i];
2274 if(timespec_isset(&c->rtrx_timeout)) {
2275 c->rtrx_timeout = now;
2278 if(timespec_isset(&c->conn_timeout)) {
2279 c->conn_timeout = then;
2282 c->rtt_start.tv_sec = 0;
2284 if(c->rto > START_RTO) {
2290 int utcp_get_user_timeout(struct utcp *u) {
2291 return u ? u->timeout : 0;
2294 void utcp_set_user_timeout(struct utcp *u, int timeout) {
2296 u->timeout = timeout;
2300 size_t utcp_get_sndbuf(struct utcp_connection *c) {
2301 return c ? c->sndbuf.maxsize : 0;
2304 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
2314 return buffer_free(&c->sndbuf);
2321 static void buffer_transfer(struct buffer *buf, char *newdata, size_t newsize) {
2322 if(buffer_wraps(buf)) {
2327 uint32_t tailsize = buf->size - buf->offset;
2328 memcpy(newdata, buf->data + buf->offset, tailsize);
2329 memcpy(newdata + tailsize, buf->data, buf->used - buf->offset);
2335 memcpy(newdata, buf->data + buf->offset, buf->used);
2339 buf->size = newsize;
2342 static void set_buffer_storage(struct buffer *buf, char *data, size_t size) {
2343 if(size > UINT32_MAX) {
2347 buf->maxsize = size;
2351 // Don't allow resizing an external buffer
2355 if(size < buf->used) {
2356 // Ignore requests for an external buffer if we are already using more than it can store
2360 // Transition from internal to external buffer
2361 buffer_transfer(buf, data, size);
2364 buf->external = true;
2365 } else if(buf->external) {
2366 // Transition from external to internal buf
2367 size_t minsize = buf->used < DEFAULT_SNDBUFSIZE ? DEFAULT_SNDBUFSIZE : buf->used;
2368 data = malloc(minsize);
2371 // Cannot handle this
2375 buffer_transfer(buf, data, minsize);
2377 buf->external = false;
2381 void utcp_set_sndbuf(struct utcp_connection *c, void *data, size_t size) {
2386 set_buffer_storage(&c->sndbuf, data, size);
2388 c->do_poll = is_reliable(c) && buffer_free(&c->sndbuf);
2391 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
2392 return c ? c->rcvbuf.maxsize : 0;
2395 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
2396 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
2397 return buffer_free(&c->rcvbuf);
2403 void utcp_set_rcvbuf(struct utcp_connection *c, void *data, size_t size) {
2408 set_buffer_storage(&c->rcvbuf, data, size);
2411 size_t utcp_get_sendq(struct utcp_connection *c) {
2412 return c->sndbuf.used;
2415 size_t utcp_get_recvq(struct utcp_connection *c) {
2416 return c->rcvbuf.used;
2419 bool utcp_get_nodelay(struct utcp_connection *c) {
2420 return c ? c->nodelay : false;
2423 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2425 c->nodelay = nodelay;
2429 bool utcp_get_keepalive(struct utcp_connection *c) {
2430 return c ? c->keepalive : false;
2433 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2435 c->keepalive = keepalive;
2439 size_t utcp_get_outq(struct utcp_connection *c) {
2440 return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2443 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2449 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2452 c->do_poll = is_reliable(c) && buffer_free(&c->sndbuf);
2456 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_listen_t listen) {
2458 utcp->accept = accept;
2459 utcp->listen = listen;
2463 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2464 if(!c || c->reapable) {
2468 if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2473 // If we expect data, start the connection timer.
2474 if(!timespec_isset(&c->conn_timeout)) {
2475 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
2476 c->conn_timeout.tv_sec += c->utcp->timeout;
2479 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2480 if(c->snd.una == c->snd.last) {
2481 timespec_clear(&c->conn_timeout);
2486 void utcp_set_flags(struct utcp_connection *c, uint32_t flags) {
2487 c->flags &= ~UTCP_CHANGEABLE_FLAGS;
2488 c->flags |= flags & UTCP_CHANGEABLE_FLAGS;
2491 void utcp_offline(struct utcp *utcp, bool offline) {
2492 struct timespec now;
2493 clock_gettime(UTCP_CLOCK, &now);
2495 for(int i = 0; i < utcp->nconnections; i++) {
2496 struct utcp_connection *c = utcp->connections[i];
2502 utcp_expect_data(c, offline);
2505 if(timespec_isset(&c->rtrx_timeout)) {
2506 c->rtrx_timeout = now;
2509 utcp->connections[i]->rtt_start.tv_sec = 0;
2511 if(c->rto > START_RTO) {
2518 void utcp_set_retransmit_cb(struct utcp *utcp, utcp_retransmit_t cb) {
2519 utcp->retransmit = cb;
2522 void utcp_set_clock_granularity(long granularity) {
2523 CLOCK_GRANULARITY = granularity;