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.
32 #include "utcp_priv.h"
47 #if defined(CLOCK_MONOTONIC_RAW) && defined(__x86_64__)
48 #define UTCP_CLOCK CLOCK_MONOTONIC_RAW
50 #define UTCP_CLOCK CLOCK_MONOTONIC
54 static void timespec_sub(const struct timespec *a, const struct timespec *b, struct timespec *r) {
55 r->tv_sec = a->tv_sec - b->tv_sec;
56 r->tv_nsec = a->tv_nsec - b->tv_nsec;
59 r->tv_sec--, r->tv_nsec += NSEC_PER_SEC;
63 static int32_t timespec_diff_usec(const struct timespec *a, const struct timespec *b) {
64 return (a->tv_sec - b->tv_sec) * 1000000 + (a->tv_nsec - b->tv_nsec) / 1000;
67 static bool timespec_lt(const struct timespec *a, const struct timespec *b) {
68 if(a->tv_sec == b->tv_sec) {
69 return a->tv_nsec < b->tv_nsec;
71 return a->tv_sec < b->tv_sec;
75 static void timespec_clear(struct timespec *a) {
80 static bool timespec_isset(const struct timespec *a) {
84 static long CLOCK_GRANULARITY; // usec
86 static inline size_t min(size_t a, size_t b) {
90 static inline size_t max(size_t a, size_t b) {
97 #ifndef UTCP_DEBUG_DATALEN
98 #define UTCP_DEBUG_DATALEN 20
101 static void debug(struct utcp_connection *c, const char *format, ...) {
106 clock_gettime(CLOCK_REALTIME, &tv);
107 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);
109 va_start(ap, format);
110 len += vsnprintf(buf + len, sizeof(buf) - len, format, ap);
113 if(len > 0 && (size_t)len < sizeof(buf)) {
114 fwrite(buf, len, 1, stderr);
118 static void print_packet(struct utcp_connection *c, const char *dir, const void *pkt, size_t len) {
121 if(len < sizeof(hdr)) {
122 debug(c, "%s: short packet (%lu bytes)\n", dir, (unsigned long)len);
126 memcpy(&hdr, pkt, sizeof(hdr));
130 if(len > sizeof(hdr)) {
131 datalen = min(len - sizeof(hdr), UTCP_DEBUG_DATALEN);
137 const uint8_t *data = (uint8_t *)pkt + sizeof(hdr);
138 char str[datalen * 2 + 1];
141 for(uint32_t i = 0; i < datalen; i++) {
142 *p++ = "0123456789ABCDEF"[data[i] >> 4];
143 *p++ = "0123456789ABCDEF"[data[i] & 15];
148 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",
149 dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux,
150 hdr.ctl & SYN ? "SYN" : "",
151 hdr.ctl & RST ? "RST" : "",
152 hdr.ctl & FIN ? "FIN" : "",
153 hdr.ctl & ACK ? "ACK" : "",
154 hdr.ctl & MF ? "MF" : "",
159 static void debug_cwnd(struct utcp_connection *c) {
160 debug(c, "snd.cwnd %u snd.ssthresh %u\n", c->snd.cwnd, ~c->snd.ssthresh ? c->snd.ssthresh : 0);
163 #define debug(...) do {} while(0)
164 #define print_packet(...) do {} while(0)
165 #define debug_cwnd(...) do {} while(0)
168 static void set_state(struct utcp_connection *c, enum state state) {
171 if(state == ESTABLISHED) {
172 timespec_clear(&c->conn_timeout);
175 debug(c, "state %s\n", strstate[state]);
178 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
179 if(seq != c->snd.last) {
194 static bool is_reliable(struct utcp_connection *c) {
195 return c->flags & UTCP_RELIABLE;
198 static int32_t seqdiff(uint32_t a, uint32_t b) {
203 static bool buffer_wraps(struct buffer *buf) {
204 return buf->size - buf->offset < buf->used;
207 static bool buffer_resize(struct buffer *buf, uint32_t newsize) {
208 char *newdata = realloc(buf->data, newsize);
216 if(buffer_wraps(buf)) {
217 // Shift the right part of the buffer until it hits the end of the new buffer.
221 // [345.........|........012]
222 uint32_t tailsize = buf->size - buf->offset;
223 uint32_t newoffset = newsize - tailsize;
224 memmove(buf->data + newoffset, buf->data + buf->offset, tailsize);
225 buf->offset = newoffset;
232 // Store data into the buffer
233 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
234 debug(NULL, "buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
236 // Ensure we don't store more than maxsize bytes in total
237 size_t required = offset + len;
239 if(required > buf->maxsize) {
240 if(offset >= buf->maxsize) {
244 len = buf->maxsize - offset;
245 required = buf->maxsize;
248 // Check if we need to resize the buffer
249 if(required > buf->size) {
250 size_t newsize = buf->size;
258 } while(newsize < required);
260 if(newsize > buf->maxsize) {
261 newsize = buf->maxsize;
264 if(!buffer_resize(buf, newsize)) {
269 uint32_t realoffset = buf->offset + offset;
271 if(buf->size - buf->offset <= offset) {
272 // The offset wrapped
273 realoffset -= buf->size;
276 if(buf->size - realoffset < len) {
277 // The new chunk of data must be wrapped
278 memcpy(buf->data + realoffset, data, buf->size - realoffset);
279 memcpy(buf->data, (char *)data + buf->size - realoffset, len - (buf->size - realoffset));
281 memcpy(buf->data + realoffset, data, len);
284 if(required > buf->used) {
285 buf->used = required;
291 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
292 return buffer_put_at(buf, buf->used, data, len);
295 // Copy data from the buffer without removing it.
296 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
297 // Ensure we don't copy more than is actually stored in the buffer
298 if(offset >= buf->used) {
302 if(buf->used - offset < len) {
303 len = buf->used - offset;
306 uint32_t realoffset = buf->offset + offset;
308 if(buf->size - buf->offset <= offset) {
309 // The offset wrapped
310 realoffset -= buf->size;
313 if(buf->size - realoffset < len) {
314 // The data is wrapped
315 memcpy(data, buf->data + realoffset, buf->size - realoffset);
316 memcpy((char *)data + buf->size - realoffset, buf->data, len - (buf->size - realoffset));
318 memcpy(data, buf->data + realoffset, len);
324 // Copy data from the buffer without removing it.
325 static ssize_t buffer_call(struct utcp_connection *c, struct buffer *buf, size_t offset, size_t len) {
330 // Ensure we don't copy more than is actually stored in the buffer
331 if(offset >= buf->used) {
335 if(buf->used - offset < len) {
336 len = buf->used - offset;
339 uint32_t realoffset = buf->offset + offset;
341 if(buf->size - buf->offset <= offset) {
342 // The offset wrapped
343 realoffset -= buf->size;
346 if(buf->size - realoffset < len) {
347 // The data is wrapped
348 ssize_t rx1 = c->recv(c, buf->data + realoffset, buf->size - realoffset);
350 if(rx1 < buf->size - realoffset) {
354 // The channel might have been closed by the previous callback
359 ssize_t rx2 = c->recv(c, buf->data, len - (buf->size - realoffset));
367 return c->recv(c, buf->data + realoffset, len);
371 // Discard data from the buffer.
372 static ssize_t buffer_discard(struct buffer *buf, size_t len) {
373 if(buf->used < len) {
377 if(buf->size - buf->offset <= len) {
378 buf->offset -= buf->size;
381 if(buf->used == len) {
392 static void buffer_clear(struct buffer *buf) {
397 static bool buffer_set_size(struct buffer *buf, uint32_t minsize, uint32_t maxsize) {
398 if(maxsize < minsize) {
402 buf->maxsize = maxsize;
404 return buf->size >= minsize || buffer_resize(buf, minsize);
407 static void buffer_exit(struct buffer *buf) {
409 memset(buf, 0, sizeof(*buf));
412 static uint32_t buffer_free(const struct buffer *buf) {
413 return buf->maxsize > buf->used ? buf->maxsize - buf->used : 0;
416 // Connections are stored in a sorted list.
417 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
419 static int compare(const void *va, const void *vb) {
422 const struct utcp_connection *a = *(struct utcp_connection **)va;
423 const struct utcp_connection *b = *(struct utcp_connection **)vb;
427 int c = (int)a->src - (int)b->src;
433 c = (int)a->dst - (int)b->dst;
437 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
438 if(!utcp->nconnections) {
442 struct utcp_connection key = {
446 struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
447 return match ? *match : NULL;
450 static void free_connection(struct utcp_connection *c) {
451 struct utcp *utcp = c->utcp;
452 struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
456 int i = cp - utcp->connections;
457 memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
458 utcp->nconnections--;
460 buffer_exit(&c->rcvbuf);
461 buffer_exit(&c->sndbuf);
465 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
466 // Check whether this combination of src and dst is free
469 if(find_connection(utcp, src, dst)) {
473 } else { // If src == 0, generate a random port number with the high bit set
474 if(utcp->nconnections >= 32767) {
479 src = rand() | 0x8000;
481 while(find_connection(utcp, src, dst)) {
486 // Allocate memory for the new connection
488 if(utcp->nconnections >= utcp->nallocated) {
489 if(!utcp->nallocated) {
490 utcp->nallocated = 4;
492 utcp->nallocated *= 2;
495 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
501 utcp->connections = new_array;
504 struct utcp_connection *c = calloc(1, sizeof(*c));
510 if(!buffer_set_size(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
515 if(!buffer_set_size(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
516 buffer_exit(&c->sndbuf);
521 // Fill in the details
530 c->snd.una = c->snd.iss;
531 c->snd.nxt = c->snd.iss + 1;
532 c->snd.last = c->snd.nxt;
533 c->snd.cwnd = (utcp->mss > 2190 ? 2 : utcp->mss > 1095 ? 3 : 4) * utcp->mss;
534 c->snd.ssthresh = ~0;
541 // Add it to the sorted list of connections
543 utcp->connections[utcp->nconnections++] = c;
544 qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
549 static inline uint32_t absdiff(uint32_t a, uint32_t b) {
557 // Update RTT variables. See RFC 6298.
558 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
560 debug(c, "invalid rtt\n");
568 c->rttvar = (c->rttvar * 3 + absdiff(c->srtt, rtt)) / 4;
569 c->srtt = (c->srtt * 7 + rtt) / 8;
572 c->rto = c->srtt + max(4 * c->rttvar, CLOCK_GRANULARITY);
574 if(c->rto > MAX_RTO) {
578 debug(c, "rtt %u srtt %u rttvar %u rto %u\n", rtt, c->srtt, c->rttvar, c->rto);
581 static void start_retransmit_timer(struct utcp_connection *c) {
582 clock_gettime(UTCP_CLOCK, &c->rtrx_timeout);
584 uint32_t rto = c->rto;
586 while(rto > USEC_PER_SEC) {
587 c->rtrx_timeout.tv_sec++;
591 c->rtrx_timeout.tv_nsec += rto * 1000;
593 if(c->rtrx_timeout.tv_nsec >= NSEC_PER_SEC) {
594 c->rtrx_timeout.tv_nsec -= NSEC_PER_SEC;
595 c->rtrx_timeout.tv_sec++;
598 debug(c, "rtrx_timeout %ld.%06lu\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_nsec);
601 static void stop_retransmit_timer(struct utcp_connection *c) {
602 timespec_clear(&c->rtrx_timeout);
603 debug(c, "rtrx_timeout cleared\n");
606 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
607 struct utcp_connection *c = allocate_connection(utcp, 0, dst);
613 assert((flags & ~0x1f) == 0);
624 pkt.hdr.src = c->src;
625 pkt.hdr.dst = c->dst;
626 pkt.hdr.seq = c->snd.iss;
628 pkt.hdr.wnd = c->rcvbuf.maxsize;
630 pkt.hdr.aux = 0x0101;
634 pkt.init[3] = flags & 0x7;
636 set_state(c, SYN_SENT);
638 print_packet(c, "send", &pkt, sizeof(pkt));
639 utcp->send(utcp, &pkt, sizeof(pkt));
641 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
642 c->conn_timeout.tv_sec += utcp->timeout;
644 start_retransmit_timer(c);
649 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
650 return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
653 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
654 if(c->reapable || c->state != SYN_RECEIVED) {
655 debug(c, "accept() called on invalid connection in state %s\n", c, strstate[c->state]);
659 debug(c, "accepted %p %p\n", c, recv, priv);
662 set_state(c, ESTABLISHED);
665 static void ack(struct utcp_connection *c, bool sendatleastone) {
666 int32_t left = seqdiff(c->snd.last, c->snd.nxt);
667 int32_t cwndleft = is_reliable(c) ? min(c->snd.cwnd, c->snd.wnd) - seqdiff(c->snd.nxt, c->snd.una) : MAX_UNRELIABLE_SIZE;
673 } else if(cwndleft < left) {
676 if(!sendatleastone || cwndleft > c->utcp->mss) {
677 left -= left % c->utcp->mss;
681 debug(c, "cwndleft %d left %d\n", cwndleft, left);
683 if(!left && !sendatleastone) {
690 } *pkt = c->utcp->pkt;
692 pkt->hdr.src = c->src;
693 pkt->hdr.dst = c->dst;
694 pkt->hdr.ack = c->rcv.nxt;
695 pkt->hdr.wnd = is_reliable(c) ? c->rcvbuf.maxsize : 0;
700 uint32_t seglen = left > c->utcp->mss ? c->utcp->mss : left;
701 pkt->hdr.seq = c->snd.nxt;
703 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
705 c->snd.nxt += seglen;
708 if(!is_reliable(c)) {
716 if(seglen && fin_wanted(c, c->snd.nxt)) {
721 if(!c->rtt_start.tv_sec) {
722 // Start RTT measurement
723 clock_gettime(UTCP_CLOCK, &c->rtt_start);
724 c->rtt_seq = pkt->hdr.seq + seglen;
725 debug(c, "starting RTT measurement, expecting ack %u\n", c->rtt_seq);
728 print_packet(c, "send", pkt, sizeof(pkt->hdr) + seglen);
729 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
731 if(left && !is_reliable(c)) {
732 pkt->hdr.wnd += seglen;
737 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
739 debug(c, "send() called on closed connection\n");
747 debug(c, "send() called on unconnected connection\n");
762 debug(c, "send() called on closed connection\n");
767 // Exit early if we have nothing to send.
778 // Check if we need to be able to buffer all data
780 if(c->flags & UTCP_NO_PARTIAL) {
781 if(len > buffer_free(&c->sndbuf)) {
782 if(len > c->sndbuf.maxsize) {
792 // Add data to send buffer.
795 len = buffer_put(&c->sndbuf, data, len);
796 } else if(c->state != SYN_SENT && c->state != SYN_RECEIVED) {
797 if(len > MAX_UNRELIABLE_SIZE || buffer_put(&c->sndbuf, data, len) != (ssize_t)len) {
816 // Don't send anything yet if the connection has not fully established yet
818 if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
824 if(!is_reliable(c)) {
825 c->snd.una = c->snd.nxt = c->snd.last;
826 buffer_discard(&c->sndbuf, c->sndbuf.used);
829 if(is_reliable(c) && !timespec_isset(&c->rtrx_timeout)) {
830 start_retransmit_timer(c);
833 if(is_reliable(c) && !timespec_isset(&c->conn_timeout)) {
834 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
835 c->conn_timeout.tv_sec += c->utcp->timeout;
841 static void swap_ports(struct hdr *hdr) {
842 uint16_t tmp = hdr->src;
847 static void fast_retransmit(struct utcp_connection *c) {
848 if(c->state == CLOSED || c->snd.last == c->snd.una) {
849 debug(c, "fast_retransmit() called but nothing to retransmit!\n");
853 struct utcp *utcp = c->utcp;
858 } *pkt = c->utcp->pkt;
860 pkt->hdr.src = c->src;
861 pkt->hdr.dst = c->dst;
862 pkt->hdr.wnd = c->rcvbuf.maxsize;
871 // Send unacked data again.
872 pkt->hdr.seq = c->snd.una;
873 pkt->hdr.ack = c->rcv.nxt;
875 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
877 if(fin_wanted(c, c->snd.una + len)) {
882 buffer_copy(&c->sndbuf, pkt->data, 0, len);
883 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
884 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
892 static void retransmit(struct utcp_connection *c) {
893 if(c->state == CLOSED || c->snd.last == c->snd.una) {
894 debug(c, "retransmit() called but nothing to retransmit!\n");
895 stop_retransmit_timer(c);
899 struct utcp *utcp = c->utcp;
901 if(utcp->retransmit) {
908 } *pkt = c->utcp->pkt;
910 pkt->hdr.src = c->src;
911 pkt->hdr.dst = c->dst;
912 pkt->hdr.wnd = c->rcvbuf.maxsize;
917 // Send our SYN again
918 pkt->hdr.seq = c->snd.iss;
921 pkt->hdr.aux = 0x0101;
925 pkt->data[3] = c->flags & 0x7;
926 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + 4);
927 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
932 pkt->hdr.seq = c->snd.nxt;
933 pkt->hdr.ack = c->rcv.nxt;
934 pkt->hdr.ctl = SYN | ACK;
935 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr));
936 utcp->send(utcp, pkt, sizeof(pkt->hdr));
944 // Send unacked data again.
945 pkt->hdr.seq = c->snd.una;
946 pkt->hdr.ack = c->rcv.nxt;
948 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
950 if(fin_wanted(c, c->snd.una + len)) {
955 // RFC 5681 slow start after timeout
956 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
957 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
958 c->snd.cwnd = utcp->mss;
961 buffer_copy(&c->sndbuf, pkt->data, 0, len);
962 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
963 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
965 c->snd.nxt = c->snd.una + len;
972 // We shouldn't need to retransmit anything in this state.
976 stop_retransmit_timer(c);
980 start_retransmit_timer(c);
983 if(c->rto > MAX_RTO) {
987 c->rtt_start.tv_sec = 0; // invalidate RTT timer
988 c->dupack = 0; // cancel any ongoing fast recovery
994 /* Update receive buffer and SACK entries after consuming data.
998 * |.....0000..1111111111.....22222......3333|
1001 * 0..3 represent the SACK entries. The ^ indicates up to which point we want
1002 * to remove data from the receive buffer. The idea is to substract "len"
1003 * from the offset of all the SACK entries, and then remove/cut down entries
1004 * that are shifted to before the start of the receive buffer.
1006 * There are three cases:
1007 * - the SACK entry is after ^, in that case just change the offset.
1008 * - the SACK entry starts before and ends after ^, so we have to
1009 * change both its offset and size.
1010 * - the SACK entry is completely before ^, in that case delete it.
1012 static void sack_consume(struct utcp_connection *c, size_t len) {
1013 debug(c, "sack_consume %lu\n", (unsigned long)len);
1015 if(len > c->rcvbuf.used) {
1016 debug(c, "all SACK entries consumed\n");
1017 c->sacks[0].len = 0;
1021 buffer_discard(&c->rcvbuf, len);
1023 for(int i = 0; i < NSACKS && c->sacks[i].len;) {
1024 if(len < c->sacks[i].offset) {
1025 c->sacks[i].offset -= len;
1027 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
1028 c->sacks[i].len -= len - c->sacks[i].offset;
1029 c->sacks[i].offset = 0;
1032 if(i < NSACKS - 1) {
1033 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
1034 c->sacks[NSACKS - 1].len = 0;
1036 c->sacks[i].len = 0;
1042 for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1043 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1047 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
1048 debug(c, "out of order packet, offset %u\n", offset);
1049 // Packet loss or reordering occured. Store the data in the buffer.
1050 ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
1053 debug(c, "packet outside receive buffer, dropping\n");
1057 if((size_t)rxd < len) {
1058 debug(c, "packet partially outside receive buffer\n");
1062 // Make note of where we put it.
1063 for(int i = 0; i < NSACKS; i++) {
1064 if(!c->sacks[i].len) { // nothing to merge, add new entry
1065 debug(c, "new SACK entry %d\n", i);
1066 c->sacks[i].offset = offset;
1067 c->sacks[i].len = rxd;
1069 } else if(offset < c->sacks[i].offset) {
1070 if(offset + rxd < c->sacks[i].offset) { // insert before
1071 if(!c->sacks[NSACKS - 1].len) { // only if room left
1072 debug(c, "insert SACK entry at %d\n", i);
1073 memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
1074 c->sacks[i].offset = offset;
1075 c->sacks[i].len = rxd;
1077 debug(c, "SACK entries full, dropping packet\n");
1082 debug(c, "merge with start of SACK entry at %d\n", i);
1083 c->sacks[i].offset = offset;
1086 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
1087 if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
1088 debug(c, "merge with end of SACK entry at %d\n", i);
1089 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
1090 // TODO: handle potential merge with next entry
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_in_order(struct utcp_connection *c, const void *data, size_t len) {
1104 ssize_t rxd = c->recv(c, data, len);
1106 if(rxd != (ssize_t)len) {
1107 // TODO: handle the application not accepting all data.
1112 // Check if we can process out-of-order data now.
1113 if(c->sacks[0].len && len >= c->sacks[0].offset) {
1114 debug(c, "incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
1116 if(len < c->sacks[0].offset + c->sacks[0].len) {
1117 size_t offset = len;
1118 len = c->sacks[0].offset + c->sacks[0].len;
1119 size_t remainder = len - offset;
1121 ssize_t rxd = buffer_call(c, &c->rcvbuf, offset, remainder);
1123 if(rxd != (ssize_t)remainder) {
1124 // TODO: handle the application not accepting all data.
1130 if(c->rcvbuf.used) {
1131 sack_consume(c, len);
1137 static void handle_unreliable(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1138 // Fast path for unfragmented packets
1139 if(!hdr->wnd && !(hdr->ctl & MF)) {
1141 c->recv(c, data, len);
1144 c->rcv.nxt = hdr->seq + len;
1148 // Ensure reassembled packet are not larger than 64 kiB
1149 if(hdr->wnd >= MAX_UNRELIABLE_SIZE || hdr->wnd + len > MAX_UNRELIABLE_SIZE) {
1153 // Don't accept out of order fragments
1154 if(hdr->wnd && hdr->seq != c->rcv.nxt) {
1158 // Reset the receive buffer for the first fragment
1160 buffer_clear(&c->rcvbuf);
1163 ssize_t rxd = buffer_put_at(&c->rcvbuf, hdr->wnd, data, len);
1165 if(rxd != (ssize_t)len) {
1169 // Send the packet if it's the final fragment
1170 if(!(hdr->ctl & MF)) {
1171 buffer_call(c, &c->rcvbuf, 0, hdr->wnd + len);
1174 c->rcv.nxt = hdr->seq + len;
1177 static void handle_incoming_data(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1178 if(!is_reliable(c)) {
1179 handle_unreliable(c, hdr, data, len);
1183 uint32_t offset = seqdiff(hdr->seq, c->rcv.nxt);
1186 handle_out_of_order(c, offset, data, len);
1188 handle_in_order(c, data, len);
1193 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
1194 const uint8_t *ptr = data;
1210 // Drop packets smaller than the header
1214 if(len < sizeof(hdr)) {
1215 print_packet(NULL, "recv", data, len);
1220 // Make a copy from the potentially unaligned data to a struct hdr
1222 memcpy(&hdr, ptr, sizeof(hdr));
1224 // Try to match the packet to an existing connection
1226 struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1227 print_packet(c, "recv", data, len);
1229 // Process the header
1234 // Drop packets with an unknown CTL flag
1236 if(hdr.ctl & ~(SYN | ACK | RST | FIN | MF)) {
1237 print_packet(NULL, "recv", data, len);
1242 // Check for auxiliary headers
1244 const uint8_t *init = NULL;
1246 uint16_t aux = hdr.aux;
1249 size_t auxlen = 4 * (aux >> 8) & 0xf;
1250 uint8_t auxtype = aux & 0xff;
1259 if(!(hdr.ctl & SYN) || auxlen != 4) {
1275 if(!(aux & 0x800)) {
1284 memcpy(&aux, ptr, 2);
1289 bool has_data = len || (hdr.ctl & (SYN | FIN));
1291 // Is it for a new connection?
1294 // Ignore RST packets
1300 // Is it a SYN packet and are we LISTENing?
1302 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1303 // If we don't want to accept it, send a RST back
1304 if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
1309 // Try to allocate memory, otherwise send a RST back
1310 c = allocate_connection(utcp, hdr.dst, hdr.src);
1317 // Parse auxilliary information
1324 c->flags = init[3] & 0x7;
1326 c->flags = UTCP_TCP;
1330 // Return SYN+ACK, go to SYN_RECEIVED state
1331 c->snd.wnd = hdr.wnd;
1332 c->rcv.irs = hdr.seq;
1333 c->rcv.nxt = c->rcv.irs + 1;
1334 set_state(c, SYN_RECEIVED);
1341 pkt.hdr.src = c->src;
1342 pkt.hdr.dst = c->dst;
1343 pkt.hdr.ack = c->rcv.irs + 1;
1344 pkt.hdr.seq = c->snd.iss;
1345 pkt.hdr.wnd = c->rcvbuf.maxsize;
1346 pkt.hdr.ctl = SYN | ACK;
1349 pkt.hdr.aux = 0x0101;
1353 pkt.data[3] = c->flags & 0x7;
1354 print_packet(c, "send", &pkt, sizeof(hdr) + 4);
1355 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1358 print_packet(c, "send", &pkt, sizeof(hdr));
1359 utcp->send(utcp, &pkt, sizeof(hdr));
1362 start_retransmit_timer(c);
1364 // No, we don't want your packets, send a RST back
1372 debug(c, "state %s\n", strstate[c->state]);
1374 // In case this is for a CLOSED connection, ignore the packet.
1375 // TODO: make it so incoming packets can never match a CLOSED connection.
1377 if(c->state == CLOSED) {
1378 debug(c, "got packet for closed connection\n");
1382 // It is for an existing connection.
1384 // 1. Drop invalid packets.
1386 // 1a. Drop packets that should not happen in our current state.
1407 // 1b. Discard data that is not in our receive window.
1409 if(is_reliable(c)) {
1412 if(c->state == SYN_SENT) {
1414 } else if(len == 0) {
1415 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1417 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1419 // cut already accepted front overlapping
1420 if(rcv_offset < 0) {
1421 acceptable = len > (size_t) - rcv_offset;
1426 hdr.seq -= rcv_offset;
1429 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1434 debug(c, "packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1436 // Ignore unacceptable RST packets.
1441 // Otherwise, continue processing.
1446 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1449 debug(c, "packet out of order, offset %u bytes", rcv_offset);
1455 c->snd.wnd = hdr.wnd; // TODO: move below
1457 // 1c. Drop packets with an invalid ACK.
1458 // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1459 // (= snd.una + c->sndbuf.used).
1461 if(!is_reliable(c)) {
1462 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1463 hdr.ack = c->snd.una;
1467 if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1468 debug(c, "packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1470 // Ignore unacceptable RST packets.
1478 // 2. Handle RST packets
1483 if(!(hdr.ctl & ACK)) {
1487 // The peer has refused our connection.
1488 set_state(c, CLOSED);
1489 errno = ECONNREFUSED;
1492 c->recv(c, NULL, 0);
1495 if(c->poll && !c->reapable) {
1506 // We haven't told the application about this connection yet. Silently delete.
1518 // The peer has aborted our connection.
1519 set_state(c, CLOSED);
1523 c->recv(c, NULL, 0);
1526 if(c->poll && !c->reapable) {
1539 // As far as the application is concerned, the connection has already been closed.
1540 // If it has called utcp_close() already, we can immediately free this connection.
1546 // Otherwise, immediately move to the CLOSED state.
1547 set_state(c, CLOSED);
1560 if(!(hdr.ctl & ACK)) {
1565 // 3. Advance snd.una
1567 advanced = seqdiff(hdr.ack, c->snd.una);
1571 if(c->rtt_start.tv_sec) {
1572 if(c->rtt_seq == hdr.ack) {
1573 struct timespec now;
1574 clock_gettime(UTCP_CLOCK, &now);
1575 int32_t diff = timespec_diff_usec(&now, &c->rtt_start);
1576 update_rtt(c, diff);
1577 c->rtt_start.tv_sec = 0;
1578 } else if(c->rtt_seq < hdr.ack) {
1579 debug(c, "cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1580 c->rtt_start.tv_sec = 0;
1584 int32_t data_acked = advanced;
1592 // TODO: handle FIN as well.
1597 assert(data_acked >= 0);
1600 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1601 assert(data_acked <= bufused);
1605 buffer_discard(&c->sndbuf, data_acked);
1607 if(is_reliable(c)) {
1612 // Also advance snd.nxt if possible
1613 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1614 c->snd.nxt = hdr.ack;
1617 c->snd.una = hdr.ack;
1620 if(c->dupack >= 3) {
1621 debug(c, "fast recovery ended\n");
1622 c->snd.cwnd = c->snd.ssthresh;
1628 // Increase the congestion window according to RFC 5681
1629 if(c->snd.cwnd < c->snd.ssthresh) {
1630 c->snd.cwnd += min(advanced, utcp->mss); // eq. 2
1632 c->snd.cwnd += max(1, (utcp->mss * utcp->mss) / c->snd.cwnd); // eq. 3
1635 if(c->snd.cwnd > c->sndbuf.maxsize) {
1636 c->snd.cwnd = c->sndbuf.maxsize;
1641 // Check if we have sent a FIN that is now ACKed.
1644 if(c->snd.una == c->snd.last) {
1645 set_state(c, FIN_WAIT_2);
1651 if(c->snd.una == c->snd.last) {
1652 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1653 c->conn_timeout.tv_sec += utcp->timeout;
1654 set_state(c, TIME_WAIT);
1663 if(!len && is_reliable(c) && c->snd.una != c->snd.last) {
1665 debug(c, "duplicate ACK %d\n", c->dupack);
1667 if(c->dupack == 3) {
1668 // RFC 5681 fast recovery
1669 debug(c, "fast recovery started\n", c->dupack);
1670 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
1671 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
1672 c->snd.cwnd = min(c->snd.ssthresh + 3 * utcp->mss, c->sndbuf.maxsize);
1674 if(c->snd.cwnd > c->sndbuf.maxsize) {
1675 c->snd.cwnd = c->sndbuf.maxsize;
1681 } else if(c->dupack > 3) {
1682 c->snd.cwnd += utcp->mss;
1684 if(c->snd.cwnd > c->sndbuf.maxsize) {
1685 c->snd.cwnd = c->sndbuf.maxsize;
1691 // We got an ACK which indicates the other side did get one of our packets.
1692 // Reset the retransmission timer to avoid going to slow start,
1693 // but don't touch the connection timeout.
1694 start_retransmit_timer(c);
1701 if(c->snd.una == c->snd.last) {
1702 stop_retransmit_timer(c);
1703 timespec_clear(&c->conn_timeout);
1704 } else if(is_reliable(c)) {
1705 start_retransmit_timer(c);
1706 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1707 c->conn_timeout.tv_sec += utcp->timeout;
1712 // 5. Process SYN stuff
1718 // This is a SYNACK. It should always have ACKed the SYN.
1723 c->rcv.irs = hdr.seq;
1724 c->rcv.nxt = hdr.seq + 1;
1728 set_state(c, FIN_WAIT_1);
1730 set_state(c, ESTABLISHED);
1736 // This is a retransmit of a SYN, send back the SYNACK.
1746 // This could be a retransmission. Ignore the SYN flag, but send an ACK back.
1757 // 6. Process new data
1759 if(c->state == SYN_RECEIVED) {
1760 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1765 // Are we still LISTENing?
1767 utcp->accept(c, c->src);
1770 if(c->state != ESTABLISHED) {
1771 set_state(c, CLOSED);
1781 // This should never happen.
1796 // Ehm no, We should never receive more data after a FIN.
1806 handle_incoming_data(c, &hdr, ptr, len);
1809 // 7. Process FIN stuff
1811 if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1815 // This should never happen.
1822 set_state(c, CLOSE_WAIT);
1826 set_state(c, CLOSING);
1830 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1831 c->conn_timeout.tv_sec += utcp->timeout;
1832 set_state(c, TIME_WAIT);
1839 // Ehm, no. We should never receive a second FIN.
1849 // FIN counts as one sequence number
1853 // Inform the application that the peer closed its end of the connection.
1856 c->recv(c, NULL, 0);
1860 // Now we send something back if:
1861 // - we received data, so we have to send back an ACK
1862 // -> sendatleastone = true
1863 // - or we got an ack, so we should maybe send a bit more data
1864 // -> sendatleastone = false
1866 if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1881 hdr.ack = hdr.seq + len;
1883 hdr.ctl = RST | ACK;
1886 print_packet(c, "send", &hdr, sizeof(hdr));
1887 utcp->send(utcp, &hdr, sizeof(hdr));
1892 int utcp_shutdown(struct utcp_connection *c, int dir) {
1893 debug(c, "shutdown %d at %u\n", dir, c ? c->snd.last : 0);
1901 debug(c, "shutdown() called on closed connection\n");
1906 if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1911 // TCP does not have a provision for stopping incoming packets.
1912 // The best we can do is to just ignore them.
1913 if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1917 // The rest of the code deals with shutting down writes.
1918 if(dir == UTCP_SHUT_RD) {
1922 // Only process shutting down writes once.
1940 set_state(c, FIN_WAIT_1);
1948 set_state(c, CLOSING);
1961 if(!timespec_isset(&c->rtrx_timeout)) {
1962 start_retransmit_timer(c);
1968 static bool reset_connection(struct utcp_connection *c) {
1975 debug(c, "abort() called on closed connection\n");
1992 set_state(c, CLOSED);
2000 set_state(c, CLOSED);
2010 hdr.seq = c->snd.nxt;
2015 print_packet(c, "send", &hdr, sizeof(hdr));
2016 c->utcp->send(c->utcp, &hdr, sizeof(hdr));
2020 // Closes all the opened connections
2021 void utcp_abort_all_connections(struct utcp *utcp) {
2027 for(int i = 0; i < utcp->nconnections; i++) {
2028 struct utcp_connection *c = utcp->connections[i];
2030 if(c->reapable || c->state == CLOSED) {
2034 utcp_recv_t old_recv = c->recv;
2035 utcp_poll_t old_poll = c->poll;
2037 reset_connection(c);
2041 old_recv(c, NULL, 0);
2044 if(old_poll && !c->reapable) {
2053 int utcp_close(struct utcp_connection *c) {
2054 if(c->rcvbuf.used) {
2055 fprintf(stderr, "UTCP channel closed with stuff in receive buffer\n");
2056 return reset_connection(c) ? 0 : -1;
2059 if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
2069 int utcp_abort(struct utcp_connection *c) {
2070 if(!reset_connection(c)) {
2079 * One call to this function will loop through all connections,
2080 * checking if something needs to be resent or not.
2081 * The return value is the time to the next timeout in milliseconds,
2082 * or maybe a negative value if the timeout is infinite.
2084 struct timespec utcp_timeout(struct utcp *utcp) {
2085 struct timespec now;
2086 clock_gettime(UTCP_CLOCK, &now);
2087 struct timespec next = {now.tv_sec + 3600, now.tv_nsec};
2089 for(int i = 0; i < utcp->nconnections; i++) {
2090 struct utcp_connection *c = utcp->connections[i];
2096 // delete connections that have been utcp_close()d.
2097 if(c->state == CLOSED) {
2099 debug(c, "reaping\n");
2107 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &now)) {
2112 c->recv(c, NULL, 0);
2115 if(c->poll && !c->reapable) {
2122 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &now)) {
2123 debug(c, "retransmitting after timeout\n");
2128 if((c->state == ESTABLISHED || c->state == CLOSE_WAIT) && c->do_poll) {
2130 uint32_t len = buffer_free(&c->sndbuf);
2135 } else if(c->state == CLOSED) {
2140 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &next)) {
2141 next = c->conn_timeout;
2144 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &next)) {
2145 next = c->rtrx_timeout;
2149 struct timespec diff;
2151 timespec_sub(&next, &now, &diff);
2156 bool utcp_is_active(struct utcp *utcp) {
2161 for(int i = 0; i < utcp->nconnections; i++)
2162 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
2169 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
2175 struct utcp *utcp = calloc(1, sizeof(*utcp));
2181 utcp_set_mtu(utcp, DEFAULT_MTU);
2188 if(!CLOCK_GRANULARITY) {
2189 struct timespec res;
2190 clock_getres(UTCP_CLOCK, &res);
2191 CLOCK_GRANULARITY = res.tv_sec * USEC_PER_SEC + res.tv_nsec / 1000;
2194 utcp->accept = accept;
2195 utcp->pre_accept = pre_accept;
2198 utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
2203 void utcp_exit(struct utcp *utcp) {
2208 for(int i = 0; i < utcp->nconnections; i++) {
2209 struct utcp_connection *c = utcp->connections[i];
2213 c->recv(c, NULL, 0);
2216 if(c->poll && !c->reapable) {
2221 buffer_exit(&c->rcvbuf);
2222 buffer_exit(&c->sndbuf);
2226 free(utcp->connections);
2231 uint16_t utcp_get_mtu(struct utcp *utcp) {
2232 return utcp ? utcp->mtu : 0;
2235 uint16_t utcp_get_mss(struct utcp *utcp) {
2236 return utcp ? utcp->mss : 0;
2239 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
2244 if(mtu <= sizeof(struct hdr)) {
2248 if(mtu > utcp->mtu) {
2249 char *new = realloc(utcp->pkt, mtu + sizeof(struct hdr));
2259 utcp->mss = mtu - sizeof(struct hdr);
2262 void utcp_reset_timers(struct utcp *utcp) {
2267 struct timespec now, then;
2269 clock_gettime(UTCP_CLOCK, &now);
2273 then.tv_sec += utcp->timeout;
2275 for(int i = 0; i < utcp->nconnections; i++) {
2276 struct utcp_connection *c = utcp->connections[i];
2282 if(timespec_isset(&c->rtrx_timeout)) {
2283 c->rtrx_timeout = now;
2286 if(timespec_isset(&c->conn_timeout)) {
2287 c->conn_timeout = then;
2290 c->rtt_start.tv_sec = 0;
2292 if(c->rto > START_RTO) {
2298 int utcp_get_user_timeout(struct utcp *u) {
2299 return u ? u->timeout : 0;
2302 void utcp_set_user_timeout(struct utcp *u, int timeout) {
2304 u->timeout = timeout;
2308 size_t utcp_get_sndbuf(struct utcp_connection *c) {
2309 return c ? c->sndbuf.maxsize : 0;
2312 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
2322 return buffer_free(&c->sndbuf);
2329 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
2334 c->sndbuf.maxsize = size;
2336 if(c->sndbuf.maxsize != size) {
2337 c->sndbuf.maxsize = -1;
2340 c->do_poll = is_reliable(c) && buffer_free(&c->sndbuf);
2343 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
2344 return c ? c->rcvbuf.maxsize : 0;
2347 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
2348 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
2349 return buffer_free(&c->rcvbuf);
2355 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
2360 c->rcvbuf.maxsize = size;
2362 if(c->rcvbuf.maxsize != size) {
2363 c->rcvbuf.maxsize = -1;
2367 size_t utcp_get_sendq(struct utcp_connection *c) {
2368 return c->sndbuf.used;
2371 size_t utcp_get_recvq(struct utcp_connection *c) {
2372 return c->rcvbuf.used;
2375 bool utcp_get_nodelay(struct utcp_connection *c) {
2376 return c ? c->nodelay : false;
2379 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2381 c->nodelay = nodelay;
2385 bool utcp_get_keepalive(struct utcp_connection *c) {
2386 return c ? c->keepalive : false;
2389 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2391 c->keepalive = keepalive;
2395 size_t utcp_get_outq(struct utcp_connection *c) {
2396 return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2399 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2405 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2408 c->do_poll = is_reliable(c) && buffer_free(&c->sndbuf);
2412 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
2414 utcp->accept = accept;
2415 utcp->pre_accept = pre_accept;
2419 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2420 if(!c || c->reapable) {
2424 if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2429 // If we expect data, start the connection timer.
2430 if(!timespec_isset(&c->conn_timeout)) {
2431 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
2432 c->conn_timeout.tv_sec += c->utcp->timeout;
2435 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2436 if(c->snd.una == c->snd.last) {
2437 timespec_clear(&c->conn_timeout);
2442 void utcp_offline(struct utcp *utcp, bool offline) {
2443 struct timespec now;
2444 clock_gettime(UTCP_CLOCK, &now);
2446 for(int i = 0; i < utcp->nconnections; i++) {
2447 struct utcp_connection *c = utcp->connections[i];
2453 utcp_expect_data(c, offline);
2456 if(timespec_isset(&c->rtrx_timeout)) {
2457 c->rtrx_timeout = now;
2460 utcp->connections[i]->rtt_start.tv_sec = 0;
2462 if(c->rto > START_RTO) {
2469 void utcp_set_retransmit_cb(struct utcp *utcp, utcp_retransmit_t cb) {
2470 utcp->retransmit = cb;
2473 void utcp_set_clock_granularity(long granularity) {
2474 CLOCK_GRANULARITY = granularity;