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_transfer(struct buffer *buf, char *newdata, size_t newsize) {
399 if(buffer_wraps(buf)) {
404 uint32_t tailsize = buf->size - buf->offset;
405 memcpy(newdata, buf->data + buf->offset, tailsize);
406 memcpy(newdata + tailsize, buf->data, buf->used - tailsize);
412 memcpy(newdata, buf->data + buf->offset, buf->used);
419 static void set_buffer_storage(struct buffer *buf, char *data, size_t size) {
420 if(size > UINT32_MAX) {
428 // Don't allow resizing an external buffer
432 if(size < buf->used) {
433 // Ignore requests for an external buffer if we are already using more than it can store
437 // Transition from internal to external buffer
438 buffer_transfer(buf, data, size);
441 buf->external = true;
442 } else if(buf->external) {
443 // Transition from external to internal buf
444 size_t minsize = buf->used < DEFAULT_SNDBUFSIZE ? DEFAULT_SNDBUFSIZE : buf->used;
445 data = malloc(minsize);
448 // Cannot handle this
452 buffer_transfer(buf, data, minsize);
454 buf->external = false;
458 static void buffer_exit(struct buffer *buf) {
463 memset(buf, 0, sizeof(*buf));
466 static uint32_t buffer_free(const struct buffer *buf) {
467 return buf->maxsize > buf->used ? buf->maxsize - buf->used : 0;
470 // Connections are stored in a sorted list.
471 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
473 static int compare(const void *va, const void *vb) {
476 const struct utcp_connection *a = *(struct utcp_connection **)va;
477 const struct utcp_connection *b = *(struct utcp_connection **)vb;
481 int c = (int)a->src - (int)b->src;
487 c = (int)a->dst - (int)b->dst;
491 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
492 if(!utcp->nconnections) {
496 struct utcp_connection key = {
500 struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
501 return match ? *match : NULL;
504 static void free_connection(struct utcp_connection *c) {
505 struct utcp *utcp = c->utcp;
506 struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
510 int i = cp - utcp->connections;
511 memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
512 utcp->nconnections--;
514 buffer_exit(&c->rcvbuf);
515 buffer_exit(&c->sndbuf);
519 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
520 // Check whether this combination of src and dst is free
523 if(find_connection(utcp, src, dst)) {
527 } else { // If src == 0, generate a random port number with the high bit set
528 if(utcp->nconnections >= 32767) {
533 src = rand() | 0x8000;
535 while(find_connection(utcp, src, dst)) {
540 // Allocate memory for the new connection
542 if(utcp->nconnections >= utcp->nallocated) {
543 if(!utcp->nallocated) {
544 utcp->nallocated = 4;
546 utcp->nallocated *= 2;
549 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
555 utcp->connections = new_array;
558 struct utcp_connection *c = calloc(1, sizeof(*c));
564 if(!buffer_set_size(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
569 if(!buffer_set_size(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
570 buffer_exit(&c->sndbuf);
575 // Fill in the details
584 c->snd.una = c->snd.iss;
585 c->snd.nxt = c->snd.iss + 1;
586 c->snd.last = c->snd.nxt;
587 c->snd.cwnd = (utcp->mss > 2190 ? 2 : utcp->mss > 1095 ? 3 : 4) * utcp->mss;
588 c->snd.ssthresh = ~0;
595 // Add it to the sorted list of connections
597 utcp->connections[utcp->nconnections++] = c;
598 qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
603 static inline uint32_t absdiff(uint32_t a, uint32_t b) {
611 // Update RTT variables. See RFC 6298.
612 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
614 debug(c, "invalid rtt\n");
622 c->rttvar = (c->rttvar * 3 + absdiff(c->srtt, rtt)) / 4;
623 c->srtt = (c->srtt * 7 + rtt) / 8;
626 c->rto = c->srtt + max(4 * c->rttvar, CLOCK_GRANULARITY);
628 if(c->rto > MAX_RTO) {
632 debug(c, "rtt %u srtt %u rttvar %u rto %u\n", rtt, c->srtt, c->rttvar, c->rto);
635 static void start_retransmit_timer(struct utcp_connection *c) {
636 clock_gettime(UTCP_CLOCK, &c->rtrx_timeout);
638 uint32_t rto = c->rto;
640 while(rto > USEC_PER_SEC) {
641 c->rtrx_timeout.tv_sec++;
645 c->rtrx_timeout.tv_nsec += rto * 1000;
647 if(c->rtrx_timeout.tv_nsec >= NSEC_PER_SEC) {
648 c->rtrx_timeout.tv_nsec -= NSEC_PER_SEC;
649 c->rtrx_timeout.tv_sec++;
652 debug(c, "rtrx_timeout %ld.%06lu\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_nsec);
655 static void stop_retransmit_timer(struct utcp_connection *c) {
656 timespec_clear(&c->rtrx_timeout);
657 debug(c, "rtrx_timeout cleared\n");
660 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
661 struct utcp_connection *c = allocate_connection(utcp, 0, dst);
667 assert((flags & ~0x1f) == 0);
678 pkt.hdr.src = c->src;
679 pkt.hdr.dst = c->dst;
680 pkt.hdr.seq = c->snd.iss;
682 pkt.hdr.wnd = c->rcvbuf.maxsize;
684 pkt.hdr.aux = 0x0101;
688 pkt.init[3] = flags & 0x7;
690 set_state(c, SYN_SENT);
692 print_packet(c, "send", &pkt, sizeof(pkt));
693 utcp->send(utcp, &pkt, sizeof(pkt));
695 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
696 c->conn_timeout.tv_sec += utcp->timeout;
698 start_retransmit_timer(c);
703 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
704 return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
707 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
708 if(c->reapable || c->state != SYN_RECEIVED) {
709 debug(c, "accept() called on invalid connection in state %s\n", c, strstate[c->state]);
713 debug(c, "accepted %p %p\n", c, recv, priv);
717 set_state(c, ESTABLISHED);
720 static void ack(struct utcp_connection *c, bool sendatleastone) {
721 int32_t left = seqdiff(c->snd.last, c->snd.nxt);
722 int32_t cwndleft = is_reliable(c) ? min(c->snd.cwnd, c->snd.wnd) - seqdiff(c->snd.nxt, c->snd.una) : MAX_UNRELIABLE_SIZE;
728 } else if(cwndleft < left) {
731 if(!sendatleastone || cwndleft > c->utcp->mss) {
732 left -= left % c->utcp->mss;
736 debug(c, "cwndleft %d left %d\n", cwndleft, left);
738 if(!left && !sendatleastone) {
745 } *pkt = c->utcp->pkt;
747 pkt->hdr.src = c->src;
748 pkt->hdr.dst = c->dst;
749 pkt->hdr.ack = c->rcv.nxt;
750 pkt->hdr.wnd = is_reliable(c) ? c->rcvbuf.maxsize : 0;
755 uint32_t seglen = left > c->utcp->mss ? c->utcp->mss : left;
756 pkt->hdr.seq = c->snd.nxt;
758 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
760 c->snd.nxt += seglen;
763 if(!is_reliable(c)) {
771 if(seglen && fin_wanted(c, c->snd.nxt)) {
776 if(!c->rtt_start.tv_sec) {
777 // Start RTT measurement
778 clock_gettime(UTCP_CLOCK, &c->rtt_start);
779 c->rtt_seq = pkt->hdr.seq + seglen;
780 debug(c, "starting RTT measurement, expecting ack %u\n", c->rtt_seq);
783 print_packet(c, "send", pkt, sizeof(pkt->hdr) + seglen);
784 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
786 if(left && !is_reliable(c)) {
787 pkt->hdr.wnd += seglen;
792 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
794 debug(c, "send() called on closed connection\n");
802 debug(c, "send() called on unconnected connection\n");
817 debug(c, "send() called on closed connection\n");
822 // Exit early if we have nothing to send.
833 // Check if we need to be able to buffer all data
835 if(c->flags & UTCP_NO_PARTIAL) {
836 if(len > buffer_free(&c->sndbuf)) {
837 if(len > c->sndbuf.maxsize) {
847 // Add data to send buffer.
850 len = buffer_put(&c->sndbuf, data, len);
851 } else if(c->state != SYN_SENT && c->state != SYN_RECEIVED) {
852 if(len > MAX_UNRELIABLE_SIZE || buffer_put(&c->sndbuf, data, len) != (ssize_t)len) {
871 // Don't send anything yet if the connection has not fully established yet
873 if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
879 if(!is_reliable(c)) {
880 c->snd.una = c->snd.nxt = c->snd.last;
881 buffer_discard(&c->sndbuf, c->sndbuf.used);
884 if(is_reliable(c) && !timespec_isset(&c->rtrx_timeout)) {
885 start_retransmit_timer(c);
888 if(is_reliable(c) && !timespec_isset(&c->conn_timeout)) {
889 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
890 c->conn_timeout.tv_sec += c->utcp->timeout;
896 static void swap_ports(struct hdr *hdr) {
897 uint16_t tmp = hdr->src;
902 static void fast_retransmit(struct utcp_connection *c) {
903 if(c->state == CLOSED || c->snd.last == c->snd.una) {
904 debug(c, "fast_retransmit() called but nothing to retransmit!\n");
908 struct utcp *utcp = c->utcp;
913 } *pkt = c->utcp->pkt;
915 pkt->hdr.src = c->src;
916 pkt->hdr.dst = c->dst;
917 pkt->hdr.wnd = c->rcvbuf.maxsize;
926 // Send unacked data again.
927 pkt->hdr.seq = c->snd.una;
928 pkt->hdr.ack = c->rcv.nxt;
930 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
932 if(fin_wanted(c, c->snd.una + len)) {
937 buffer_copy(&c->sndbuf, pkt->data, 0, len);
938 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
939 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
947 static void retransmit(struct utcp_connection *c) {
948 if(c->state == CLOSED || c->snd.last == c->snd.una) {
949 debug(c, "retransmit() called but nothing to retransmit!\n");
950 stop_retransmit_timer(c);
954 struct utcp *utcp = c->utcp;
956 if(utcp->retransmit) {
963 } *pkt = c->utcp->pkt;
965 pkt->hdr.src = c->src;
966 pkt->hdr.dst = c->dst;
967 pkt->hdr.wnd = c->rcvbuf.maxsize;
972 // Send our SYN again
973 pkt->hdr.seq = c->snd.iss;
976 pkt->hdr.aux = 0x0101;
980 pkt->data[3] = c->flags & 0x7;
981 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + 4);
982 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
987 pkt->hdr.seq = c->snd.nxt;
988 pkt->hdr.ack = c->rcv.nxt;
989 pkt->hdr.ctl = SYN | ACK;
990 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr));
991 utcp->send(utcp, pkt, sizeof(pkt->hdr));
999 // Send unacked data again.
1000 pkt->hdr.seq = c->snd.una;
1001 pkt->hdr.ack = c->rcv.nxt;
1003 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
1005 if(fin_wanted(c, c->snd.una + len)) {
1007 pkt->hdr.ctl |= FIN;
1010 // RFC 5681 slow start after timeout
1011 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
1012 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
1013 c->snd.cwnd = utcp->mss;
1016 buffer_copy(&c->sndbuf, pkt->data, 0, len);
1017 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
1018 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
1020 c->snd.nxt = c->snd.una + len;
1027 // We shouldn't need to retransmit anything in this state.
1031 stop_retransmit_timer(c);
1035 start_retransmit_timer(c);
1038 if(c->rto > MAX_RTO) {
1042 c->rtt_start.tv_sec = 0; // invalidate RTT timer
1043 c->dupack = 0; // cancel any ongoing fast recovery
1049 /* Update receive buffer and SACK entries after consuming data.
1053 * |.....0000..1111111111.....22222......3333|
1056 * 0..3 represent the SACK entries. The ^ indicates up to which point we want
1057 * to remove data from the receive buffer. The idea is to substract "len"
1058 * from the offset of all the SACK entries, and then remove/cut down entries
1059 * that are shifted to before the start of the receive buffer.
1061 * There are three cases:
1062 * - the SACK entry is after ^, in that case just change the offset.
1063 * - the SACK entry starts before and ends after ^, so we have to
1064 * change both its offset and size.
1065 * - the SACK entry is completely before ^, in that case delete it.
1067 static void sack_consume(struct utcp_connection *c, size_t len) {
1068 debug(c, "sack_consume %lu\n", (unsigned long)len);
1070 if(len > c->rcvbuf.used) {
1071 debug(c, "all SACK entries consumed\n");
1072 c->sacks[0].len = 0;
1076 buffer_discard(&c->rcvbuf, len);
1078 for(int i = 0; i < NSACKS && c->sacks[i].len;) {
1079 if(len < c->sacks[i].offset) {
1080 c->sacks[i].offset -= len;
1082 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
1083 c->sacks[i].len -= len - c->sacks[i].offset;
1084 c->sacks[i].offset = 0;
1087 if(i < NSACKS - 1) {
1088 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
1089 c->sacks[NSACKS - 1].len = 0;
1091 c->sacks[i].len = 0;
1097 for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1098 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1102 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
1103 debug(c, "out of order packet, offset %u\n", offset);
1104 // Packet loss or reordering occured. Store the data in the buffer.
1105 ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
1108 debug(c, "packet outside receive buffer, dropping\n");
1112 if((size_t)rxd < len) {
1113 debug(c, "packet partially outside receive buffer\n");
1117 // Make note of where we put it.
1118 for(int i = 0; i < NSACKS; i++) {
1119 if(!c->sacks[i].len) { // nothing to merge, add new entry
1120 debug(c, "new SACK entry %d\n", i);
1121 c->sacks[i].offset = offset;
1122 c->sacks[i].len = rxd;
1124 } else if(offset < c->sacks[i].offset) {
1125 if(offset + rxd < c->sacks[i].offset) { // insert before
1126 if(!c->sacks[NSACKS - 1].len) { // only if room left
1127 debug(c, "insert SACK entry at %d\n", i);
1128 memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
1129 c->sacks[i].offset = offset;
1130 c->sacks[i].len = rxd;
1132 debug(c, "SACK entries full, dropping packet\n");
1137 debug(c, "merge with start of SACK entry at %d\n", i);
1138 c->sacks[i].offset = offset;
1141 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
1142 if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
1143 debug(c, "merge with end of SACK entry at %d\n", i);
1144 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
1145 // TODO: handle potential merge with next entry
1152 for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1153 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1157 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
1159 ssize_t rxd = c->recv(c, data, len);
1161 if(rxd != (ssize_t)len) {
1162 // TODO: handle the application not accepting all data.
1167 // Check if we can process out-of-order data now.
1168 if(c->sacks[0].len && len >= c->sacks[0].offset) {
1169 debug(c, "incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
1171 if(len < c->sacks[0].offset + c->sacks[0].len) {
1172 size_t offset = len;
1173 len = c->sacks[0].offset + c->sacks[0].len;
1174 size_t remainder = len - offset;
1176 ssize_t rxd = buffer_call(c, &c->rcvbuf, offset, remainder);
1178 if(rxd != (ssize_t)remainder) {
1179 // TODO: handle the application not accepting all data.
1185 if(c->rcvbuf.used) {
1186 sack_consume(c, len);
1192 static void handle_unreliable(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1193 // Fast path for unfragmented packets
1194 if(!hdr->wnd && !(hdr->ctl & MF)) {
1196 c->recv(c, data, len);
1199 c->rcv.nxt = hdr->seq + len;
1203 // Ensure reassembled packet are not larger than 64 kiB
1204 if(hdr->wnd >= MAX_UNRELIABLE_SIZE || hdr->wnd + len > MAX_UNRELIABLE_SIZE) {
1208 // Don't accept out of order fragments
1209 if(hdr->wnd && hdr->seq != c->rcv.nxt) {
1213 // Reset the receive buffer for the first fragment
1215 buffer_clear(&c->rcvbuf);
1218 ssize_t rxd = buffer_put_at(&c->rcvbuf, hdr->wnd, data, len);
1220 if(rxd != (ssize_t)len) {
1224 // Send the packet if it's the final fragment
1225 if(!(hdr->ctl & MF)) {
1226 buffer_call(c, &c->rcvbuf, 0, hdr->wnd + len);
1229 c->rcv.nxt = hdr->seq + len;
1232 static void handle_incoming_data(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1233 if(!is_reliable(c)) {
1234 handle_unreliable(c, hdr, data, len);
1238 uint32_t offset = seqdiff(hdr->seq, c->rcv.nxt);
1241 handle_out_of_order(c, offset, data, len);
1243 handle_in_order(c, data, len);
1248 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
1249 const uint8_t *ptr = data;
1265 // Drop packets smaller than the header
1269 if(len < sizeof(hdr)) {
1270 print_packet(NULL, "recv", data, len);
1275 // Make a copy from the potentially unaligned data to a struct hdr
1277 memcpy(&hdr, ptr, sizeof(hdr));
1279 // Try to match the packet to an existing connection
1281 struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1282 print_packet(c, "recv", data, len);
1284 // Process the header
1289 // Drop packets with an unknown CTL flag
1291 if(hdr.ctl & ~(SYN | ACK | RST | FIN | MF)) {
1292 print_packet(NULL, "recv", data, len);
1297 // Check for auxiliary headers
1299 const uint8_t *init = NULL;
1301 uint16_t aux = hdr.aux;
1304 size_t auxlen = 4 * (aux >> 8) & 0xf;
1305 uint8_t auxtype = aux & 0xff;
1314 if(!(hdr.ctl & SYN) || auxlen != 4) {
1330 if(!(aux & 0x800)) {
1339 memcpy(&aux, ptr, 2);
1344 bool has_data = len || (hdr.ctl & (SYN | FIN));
1346 // Is it for a new connection?
1349 // Ignore RST packets
1355 // Is it a SYN packet and are we LISTENing?
1357 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1358 // If we don't want to accept it, send a RST back
1359 if((utcp->listen && !utcp->listen(utcp, hdr.dst))) {
1364 // Try to allocate memory, otherwise send a RST back
1365 c = allocate_connection(utcp, hdr.dst, hdr.src);
1372 // Parse auxilliary information
1379 c->flags = init[3] & 0x7;
1381 c->flags = UTCP_TCP;
1385 // Return SYN+ACK, go to SYN_RECEIVED state
1386 c->snd.wnd = hdr.wnd;
1387 c->rcv.irs = hdr.seq;
1388 c->rcv.nxt = c->rcv.irs + 1;
1389 set_state(c, SYN_RECEIVED);
1396 pkt.hdr.src = c->src;
1397 pkt.hdr.dst = c->dst;
1398 pkt.hdr.ack = c->rcv.irs + 1;
1399 pkt.hdr.seq = c->snd.iss;
1400 pkt.hdr.wnd = c->rcvbuf.maxsize;
1401 pkt.hdr.ctl = SYN | ACK;
1404 pkt.hdr.aux = 0x0101;
1408 pkt.data[3] = c->flags & 0x7;
1409 print_packet(c, "send", &pkt, sizeof(hdr) + 4);
1410 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1413 print_packet(c, "send", &pkt, sizeof(hdr));
1414 utcp->send(utcp, &pkt, sizeof(hdr));
1417 start_retransmit_timer(c);
1419 // No, we don't want your packets, send a RST back
1427 debug(c, "state %s\n", strstate[c->state]);
1429 // In case this is for a CLOSED connection, ignore the packet.
1430 // TODO: make it so incoming packets can never match a CLOSED connection.
1432 if(c->state == CLOSED) {
1433 debug(c, "got packet for closed connection\n");
1437 // It is for an existing connection.
1439 // 1. Drop invalid packets.
1441 // 1a. Drop packets that should not happen in our current state.
1462 // 1b. Discard data that is not in our receive window.
1464 if(is_reliable(c)) {
1467 if(c->state == SYN_SENT) {
1469 } else if(len == 0) {
1470 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1472 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1474 // cut already accepted front overlapping
1475 if(rcv_offset < 0) {
1476 acceptable = len > (size_t) - rcv_offset;
1481 hdr.seq -= rcv_offset;
1484 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1489 debug(c, "packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1491 // Ignore unacceptable RST packets.
1496 // Otherwise, continue processing.
1501 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1504 debug(c, "packet out of order, offset %u bytes", rcv_offset);
1510 c->snd.wnd = hdr.wnd; // TODO: move below
1512 // 1c. Drop packets with an invalid ACK.
1513 // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1514 // (= snd.una + c->sndbuf.used).
1516 if(!is_reliable(c)) {
1517 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1518 hdr.ack = c->snd.una;
1522 // 2. Handle RST packets
1527 if(!(hdr.ctl & ACK)) {
1531 // The peer has refused our connection.
1532 set_state(c, CLOSED);
1533 errno = ECONNREFUSED;
1536 c->recv(c, NULL, 0);
1539 if(c->poll && !c->reapable) {
1550 // We haven't told the application about this connection yet. Silently delete.
1562 // The peer has aborted our connection.
1563 set_state(c, CLOSED);
1567 c->recv(c, NULL, 0);
1570 if(c->poll && !c->reapable) {
1583 // As far as the application is concerned, the connection has already been closed.
1584 // If it has called utcp_close() already, we can immediately free this connection.
1590 // Otherwise, immediately move to the CLOSED state.
1591 set_state(c, CLOSED);
1604 if(!(hdr.ctl & ACK)) {
1609 // 3. Advance snd.una
1611 if(seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0) {
1612 debug(c, "packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1616 advanced = seqdiff(hdr.ack, c->snd.una);
1620 if(c->rtt_start.tv_sec) {
1621 if(c->rtt_seq == hdr.ack) {
1622 struct timespec now;
1623 clock_gettime(UTCP_CLOCK, &now);
1624 int32_t diff = timespec_diff_usec(&now, &c->rtt_start);
1625 update_rtt(c, diff);
1626 c->rtt_start.tv_sec = 0;
1627 } else if(c->rtt_seq < hdr.ack) {
1628 debug(c, "cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1629 c->rtt_start.tv_sec = 0;
1633 int32_t data_acked = advanced;
1641 // TODO: handle FIN as well.
1646 assert(data_acked >= 0);
1649 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1650 assert(data_acked <= bufused);
1654 buffer_discard(&c->sndbuf, data_acked);
1656 if(is_reliable(c)) {
1661 // Also advance snd.nxt if possible
1662 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1663 c->snd.nxt = hdr.ack;
1666 c->snd.una = hdr.ack;
1669 if(c->dupack >= 3) {
1670 debug(c, "fast recovery ended\n");
1671 c->snd.cwnd = c->snd.ssthresh;
1677 // Increase the congestion window according to RFC 5681
1678 if(c->snd.cwnd < c->snd.ssthresh) {
1679 c->snd.cwnd += min(advanced, utcp->mss); // eq. 2
1681 c->snd.cwnd += max(1, (utcp->mss * utcp->mss) / c->snd.cwnd); // eq. 3
1684 if(c->snd.cwnd > c->sndbuf.maxsize) {
1685 c->snd.cwnd = c->sndbuf.maxsize;
1690 // Check if we have sent a FIN that is now ACKed.
1693 if(c->snd.una == c->snd.last) {
1694 set_state(c, FIN_WAIT_2);
1700 if(c->snd.una == c->snd.last) {
1701 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1702 c->conn_timeout.tv_sec += utcp->timeout;
1703 set_state(c, TIME_WAIT);
1712 if(!len && is_reliable(c) && c->snd.una != c->snd.last) {
1714 debug(c, "duplicate ACK %d\n", c->dupack);
1716 if(c->dupack == 3) {
1717 // RFC 5681 fast recovery
1718 debug(c, "fast recovery started\n", c->dupack);
1719 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
1720 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
1721 c->snd.cwnd = min(c->snd.ssthresh + 3 * utcp->mss, c->sndbuf.maxsize);
1723 if(c->snd.cwnd > c->sndbuf.maxsize) {
1724 c->snd.cwnd = c->sndbuf.maxsize;
1730 } else if(c->dupack > 3) {
1731 c->snd.cwnd += utcp->mss;
1733 if(c->snd.cwnd > c->sndbuf.maxsize) {
1734 c->snd.cwnd = c->sndbuf.maxsize;
1740 // We got an ACK which indicates the other side did get one of our packets.
1741 // Reset the retransmission timer to avoid going to slow start,
1742 // but don't touch the connection timeout.
1743 start_retransmit_timer(c);
1750 if(c->snd.una == c->snd.last) {
1751 stop_retransmit_timer(c);
1752 timespec_clear(&c->conn_timeout);
1753 } else if(is_reliable(c)) {
1754 start_retransmit_timer(c);
1755 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1756 c->conn_timeout.tv_sec += utcp->timeout;
1761 // 5. Process SYN stuff
1767 // This is a SYNACK. It should always have ACKed the SYN.
1772 c->rcv.irs = hdr.seq;
1773 c->rcv.nxt = hdr.seq + 1;
1777 set_state(c, FIN_WAIT_1);
1780 set_state(c, ESTABLISHED);
1786 // This is a retransmit of a SYN, send back the SYNACK.
1796 // This could be a retransmission. Ignore the SYN flag, but send an ACK back.
1807 // 6. Process new data
1809 if(c->state == SYN_RECEIVED) {
1810 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1815 // Are we still LISTENing?
1817 utcp->accept(c, c->src);
1820 if(c->state != ESTABLISHED) {
1821 set_state(c, CLOSED);
1831 // This should never happen.
1843 // We already closed the connection and are not interested in more data.
1853 // Ehm no, We should never receive more data after a FIN.
1863 handle_incoming_data(c, &hdr, ptr, len);
1866 // 7. Process FIN stuff
1868 if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1872 // This should never happen.
1879 set_state(c, CLOSE_WAIT);
1883 set_state(c, CLOSING);
1887 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1888 c->conn_timeout.tv_sec += utcp->timeout;
1889 set_state(c, TIME_WAIT);
1896 // Ehm, no. We should never receive a second FIN.
1906 // FIN counts as one sequence number
1910 // Inform the application that the peer closed its end of the connection.
1913 c->recv(c, NULL, 0);
1917 // Now we send something back if:
1918 // - we received data, so we have to send back an ACK
1919 // -> sendatleastone = true
1920 // - or we got an ack, so we should maybe send a bit more data
1921 // -> sendatleastone = false
1923 if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1938 hdr.ack = hdr.seq + len;
1940 hdr.ctl = RST | ACK;
1943 print_packet(c, "send", &hdr, sizeof(hdr));
1944 utcp->send(utcp, &hdr, sizeof(hdr));
1949 int utcp_shutdown(struct utcp_connection *c, int dir) {
1950 debug(c, "shutdown %d at %u\n", dir, c ? c->snd.last : 0);
1958 debug(c, "shutdown() called on closed connection\n");
1963 if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1968 // TCP does not have a provision for stopping incoming packets.
1969 // The best we can do is to just ignore them.
1970 if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1974 // The rest of the code deals with shutting down writes.
1975 if(dir == UTCP_SHUT_RD) {
1979 // Only process shutting down writes once.
1997 set_state(c, FIN_WAIT_1);
2005 set_state(c, CLOSING);
2018 if(!timespec_isset(&c->rtrx_timeout)) {
2019 start_retransmit_timer(c);
2025 static bool reset_connection(struct utcp_connection *c) {
2032 debug(c, "abort() called on closed connection\n");
2049 set_state(c, CLOSED);
2057 set_state(c, CLOSED);
2067 hdr.seq = c->snd.nxt;
2068 hdr.ack = c->rcv.nxt;
2072 print_packet(c, "send", &hdr, sizeof(hdr));
2073 c->utcp->send(c->utcp, &hdr, sizeof(hdr));
2077 static void set_reapable(struct utcp_connection *c) {
2078 if(c->sndbuf.external) {
2079 set_buffer_storage(&c->sndbuf, NULL, DEFAULT_MTU);
2082 if(c->rcvbuf.external) {
2083 set_buffer_storage(&c->rcvbuf, NULL, DEFAULT_MTU);
2091 // Closes all the opened connections
2092 void utcp_abort_all_connections(struct utcp *utcp) {
2098 for(int i = 0; i < utcp->nconnections; i++) {
2099 struct utcp_connection *c = utcp->connections[i];
2101 if(c->reapable || c->state == CLOSED) {
2105 utcp_recv_t old_recv = c->recv;
2106 utcp_poll_t old_poll = c->poll;
2112 old_recv(c, NULL, 0);
2115 if(old_poll && !c->reapable) {
2124 int utcp_close(struct utcp_connection *c) {
2125 if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
2133 int utcp_abort(struct utcp_connection *c) {
2134 if(!reset_connection(c)) {
2143 * One call to this function will loop through all connections,
2144 * checking if something needs to be resent or not.
2145 * The return value is the time to the next timeout in milliseconds,
2146 * or maybe a negative value if the timeout is infinite.
2148 struct timespec utcp_timeout(struct utcp *utcp) {
2149 struct timespec now;
2150 clock_gettime(UTCP_CLOCK, &now);
2151 struct timespec next = {now.tv_sec + 3600, now.tv_nsec};
2153 for(int i = 0; i < utcp->nconnections; i++) {
2154 struct utcp_connection *c = utcp->connections[i];
2160 // delete connections that have been utcp_close()d.
2161 if(c->state == CLOSED) {
2163 debug(c, "reaping\n");
2171 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &now)) {
2176 c->recv(c, NULL, 0);
2179 if(c->poll && !c->reapable) {
2186 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &now)) {
2187 debug(c, "retransmitting after timeout\n");
2192 if((c->state == ESTABLISHED || c->state == CLOSE_WAIT) && c->do_poll) {
2194 uint32_t len = buffer_free(&c->sndbuf);
2199 } else if(c->state == CLOSED) {
2204 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &next)) {
2205 next = c->conn_timeout;
2208 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &next)) {
2209 next = c->rtrx_timeout;
2213 struct timespec diff;
2215 timespec_sub(&next, &now, &diff);
2220 bool utcp_is_active(struct utcp *utcp) {
2225 for(int i = 0; i < utcp->nconnections; i++)
2226 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
2233 struct utcp *utcp_init(utcp_accept_t accept, utcp_listen_t listen, utcp_send_t send, void *priv) {
2239 struct utcp *utcp = calloc(1, sizeof(*utcp));
2245 utcp_set_mtu(utcp, DEFAULT_MTU);
2252 if(!CLOCK_GRANULARITY) {
2253 struct timespec res;
2254 clock_getres(UTCP_CLOCK, &res);
2255 CLOCK_GRANULARITY = res.tv_sec * USEC_PER_SEC + res.tv_nsec / 1000;
2258 utcp->accept = accept;
2259 utcp->listen = listen;
2262 utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
2267 void utcp_exit(struct utcp *utcp) {
2272 for(int i = 0; i < utcp->nconnections; i++) {
2273 struct utcp_connection *c = utcp->connections[i];
2277 c->recv(c, NULL, 0);
2280 if(c->poll && !c->reapable) {
2285 buffer_exit(&c->rcvbuf);
2286 buffer_exit(&c->sndbuf);
2290 free(utcp->connections);
2295 uint16_t utcp_get_mtu(struct utcp *utcp) {
2296 return utcp ? utcp->mtu : 0;
2299 uint16_t utcp_get_mss(struct utcp *utcp) {
2300 return utcp ? utcp->mss : 0;
2303 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
2308 if(mtu <= sizeof(struct hdr)) {
2312 if(mtu > utcp->mtu) {
2313 char *new = realloc(utcp->pkt, mtu + sizeof(struct hdr));
2323 utcp->mss = mtu - sizeof(struct hdr);
2326 void utcp_reset_timers(struct utcp *utcp) {
2331 struct timespec now, then;
2333 clock_gettime(UTCP_CLOCK, &now);
2337 then.tv_sec += utcp->timeout;
2339 for(int i = 0; i < utcp->nconnections; i++) {
2340 struct utcp_connection *c = utcp->connections[i];
2346 if(timespec_isset(&c->rtrx_timeout)) {
2347 c->rtrx_timeout = now;
2350 if(timespec_isset(&c->conn_timeout)) {
2351 c->conn_timeout = then;
2354 c->rtt_start.tv_sec = 0;
2356 if(c->rto > START_RTO) {
2362 int utcp_get_user_timeout(struct utcp *u) {
2363 return u ? u->timeout : 0;
2366 void utcp_set_user_timeout(struct utcp *u, int timeout) {
2368 u->timeout = timeout;
2372 size_t utcp_get_sndbuf(struct utcp_connection *c) {
2373 return c ? c->sndbuf.maxsize : 0;
2376 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
2386 return buffer_free(&c->sndbuf);
2393 void utcp_set_sndbuf(struct utcp_connection *c, void *data, size_t size) {
2398 set_buffer_storage(&c->sndbuf, data, size);
2400 c->do_poll = is_reliable(c) && buffer_free(&c->sndbuf);
2403 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
2404 return c ? c->rcvbuf.maxsize : 0;
2407 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
2408 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
2409 return buffer_free(&c->rcvbuf);
2415 void utcp_set_rcvbuf(struct utcp_connection *c, void *data, size_t size) {
2420 set_buffer_storage(&c->rcvbuf, data, size);
2423 size_t utcp_get_sendq(struct utcp_connection *c) {
2424 return c->sndbuf.used;
2427 size_t utcp_get_recvq(struct utcp_connection *c) {
2428 return c->rcvbuf.used;
2431 bool utcp_get_nodelay(struct utcp_connection *c) {
2432 return c ? c->nodelay : false;
2435 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2437 c->nodelay = nodelay;
2441 bool utcp_get_keepalive(struct utcp_connection *c) {
2442 return c ? c->keepalive : false;
2445 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2447 c->keepalive = keepalive;
2451 size_t utcp_get_outq(struct utcp_connection *c) {
2452 return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2455 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2461 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2464 c->do_poll = is_reliable(c) && buffer_free(&c->sndbuf);
2468 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_listen_t listen) {
2470 utcp->accept = accept;
2471 utcp->listen = listen;
2475 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2476 if(!c || c->reapable) {
2480 if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2485 // If we expect data, start the connection timer.
2486 if(!timespec_isset(&c->conn_timeout)) {
2487 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
2488 c->conn_timeout.tv_sec += c->utcp->timeout;
2491 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2492 if(c->snd.una == c->snd.last) {
2493 timespec_clear(&c->conn_timeout);
2498 void utcp_set_flags(struct utcp_connection *c, uint32_t flags) {
2499 c->flags &= ~UTCP_CHANGEABLE_FLAGS;
2500 c->flags |= flags & UTCP_CHANGEABLE_FLAGS;
2503 void utcp_offline(struct utcp *utcp, bool offline) {
2504 struct timespec now;
2505 clock_gettime(UTCP_CLOCK, &now);
2507 for(int i = 0; i < utcp->nconnections; i++) {
2508 struct utcp_connection *c = utcp->connections[i];
2514 utcp_expect_data(c, offline);
2517 if(timespec_isset(&c->rtrx_timeout)) {
2518 c->rtrx_timeout = now;
2521 utcp->connections[i]->rtt_start.tv_sec = 0;
2523 if(c->rto > START_RTO) {
2530 void utcp_set_retransmit_cb(struct utcp *utcp, utcp_retransmit_t cb) {
2531 utcp->retransmit = cb;
2534 void utcp_set_clock_granularity(long granularity) {
2535 CLOCK_GRANULARITY = granularity;