2 utcp.c -- Userspace TCP
3 Copyright (C) 2014-2017 Guus Sliepen <guus@tinc-vpn.org>
5 This program is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 2 of the License, or
8 (at your option) any later version.
10 This program is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU General Public License for more details.
15 You should have received a copy of the GNU General Public License along
16 with this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
31 #include <sys/socket.h>
33 #include "utcp_priv.h"
48 #define timersub(a, b, r)\
50 (r)->tv_sec = (a)->tv_sec - (b)->tv_sec;\
51 (r)->tv_usec = (a)->tv_usec - (b)->tv_usec;\
53 (r)->tv_sec--, (r)->tv_usec += USEC_PER_SEC;\
57 static inline size_t max(size_t a, size_t b) {
64 static void debug(const char *format, ...) {
67 vfprintf(stderr, format, ap);
71 static void print_packet(struct utcp *utcp, const char *dir, const void *pkt, size_t len) {
74 if(len < sizeof(hdr)) {
75 debug("%p %s: short packet (%lu bytes)\n", utcp, dir, (unsigned long)len);
79 memcpy(&hdr, pkt, sizeof(hdr));
80 debug("%p %s: len=%lu, src=%u dst=%u seq=%u ack=%u wnd=%u aux=%x ctl=", utcp, dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux);
98 if(len > sizeof(hdr)) {
99 uint32_t datalen = len - sizeof(hdr);
100 const uint8_t *data = (uint8_t *)pkt + sizeof(hdr);
101 char str[datalen * 2 + 1];
104 for(uint32_t i = 0; i < datalen; i++) {
105 *p++ = "0123456789ABCDEF"[data[i] >> 4];
106 *p++ = "0123456789ABCDEF"[data[i] & 15];
111 debug(" data=%s", str);
117 #define debug(...) do {} while(0)
118 #define print_packet(...) do {} while(0)
121 static void set_state(struct utcp_connection *c, enum state state) {
124 if(state == ESTABLISHED) {
125 timerclear(&c->conn_timeout);
128 debug("%p new state: %s\n", c->utcp, strstate[state]);
131 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
132 if(seq != c->snd.last) {
147 static bool is_reliable(struct utcp_connection *c) {
148 return c->flags & UTCP_RELIABLE;
151 static int32_t seqdiff(uint32_t a, uint32_t b) {
156 // TODO: convert to ringbuffers to avoid memmove() operations.
158 // Store data into the buffer
159 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
160 debug("buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
162 size_t required = offset + len;
164 if(required > buf->maxsize) {
165 if(offset >= buf->maxsize) {
169 len = buf->maxsize - offset;
170 required = buf->maxsize;
173 if(required > buf->size) {
174 size_t newsize = buf->size;
181 } while(newsize < required);
184 if(newsize > buf->maxsize) {
185 newsize = buf->maxsize;
188 char *newdata = realloc(buf->data, newsize);
198 memcpy(buf->data + offset, data, len);
200 if(required > buf->used) {
201 buf->used = required;
207 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
208 return buffer_put_at(buf, buf->used, data, len);
211 // Get data from the buffer. data can be NULL.
212 static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
213 if(len > buf->used) {
218 memcpy(data, buf->data, len);
221 if(len < buf->used) {
222 memmove(buf->data, buf->data + len, buf->used - len);
229 // Copy data from the buffer without removing it.
230 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
231 if(offset >= buf->used) {
235 if(offset + len > buf->used) {
236 len = buf->used - offset;
239 memcpy(data, buf->data + offset, len);
243 static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
244 memset(buf, 0, sizeof(*buf));
247 buf->data = malloc(len);
255 buf->maxsize = maxlen;
259 static void buffer_exit(struct buffer *buf) {
261 memset(buf, 0, sizeof(*buf));
264 static uint32_t buffer_free(const struct buffer *buf) {
265 return buf->maxsize - buf->used;
268 // Connections are stored in a sorted list.
269 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
271 static int compare(const void *va, const void *vb) {
274 const struct utcp_connection *a = *(struct utcp_connection **)va;
275 const struct utcp_connection *b = *(struct utcp_connection **)vb;
278 assert(a->src && b->src);
280 int c = (int)a->src - (int)b->src;
286 c = (int)a->dst - (int)b->dst;
290 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
291 if(!utcp->nconnections) {
295 struct utcp_connection key = {
299 struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
300 return match ? *match : NULL;
303 static void free_connection(struct utcp_connection *c) {
304 struct utcp *utcp = c->utcp;
305 struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
309 int i = cp - utcp->connections;
310 memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
311 utcp->nconnections--;
313 buffer_exit(&c->rcvbuf);
314 buffer_exit(&c->sndbuf);
318 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
319 // Check whether this combination of src and dst is free
322 if(find_connection(utcp, src, dst)) {
326 } else { // If src == 0, generate a random port number with the high bit set
327 if(utcp->nconnections >= 32767) {
332 src = rand() | 0x8000;
334 while(find_connection(utcp, src, dst)) {
339 // Allocate memory for the new connection
341 if(utcp->nconnections >= utcp->nallocated) {
342 if(!utcp->nallocated) {
343 utcp->nallocated = 4;
345 utcp->nallocated *= 2;
348 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
354 utcp->connections = new_array;
357 struct utcp_connection *c = calloc(1, sizeof(*c));
363 if(!buffer_init(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
368 if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
369 buffer_exit(&c->sndbuf);
374 // Fill in the details
383 c->snd.una = c->snd.iss;
384 c->snd.nxt = c->snd.iss + 1;
385 c->rcv.wnd = utcp->mtu;
386 c->snd.last = c->snd.nxt;
387 c->snd.cwnd = utcp->mtu;
390 // Add it to the sorted list of connections
392 utcp->connections[utcp->nconnections++] = c;
393 qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
398 static inline uint32_t absdiff(uint32_t a, uint32_t b) {
406 // Update RTT variables. See RFC 6298.
407 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
409 debug("invalid rtt\n");
413 struct utcp *utcp = c->utcp;
417 utcp->rttvar = rtt / 2;
418 utcp->rto = rtt + max(2 * rtt, CLOCK_GRANULARITY);
420 utcp->rttvar = (utcp->rttvar * 3 + absdiff(utcp->srtt, rtt)) / 4;
421 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
422 utcp->rto = utcp->srtt + max(utcp->rttvar, CLOCK_GRANULARITY);
425 if(utcp->rto > MAX_RTO) {
429 debug("rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
432 static void start_retransmit_timer(struct utcp_connection *c) {
433 gettimeofday(&c->rtrx_timeout, NULL);
434 c->rtrx_timeout.tv_usec += c->utcp->rto;
436 while(c->rtrx_timeout.tv_usec >= 1000000) {
437 c->rtrx_timeout.tv_usec -= 1000000;
438 c->rtrx_timeout.tv_sec++;
441 debug("timeout set to %lu.%06lu (%u)\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_usec, c->utcp->rto);
444 static void stop_retransmit_timer(struct utcp_connection *c) {
445 timerclear(&c->rtrx_timeout);
446 debug("timeout cleared\n");
449 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
450 struct utcp_connection *c = allocate_connection(utcp, 0, dst);
456 assert((flags & ~0xf) == 0);
467 pkt.hdr.src = c->src;
468 pkt.hdr.dst = c->dst;
469 pkt.hdr.seq = c->snd.iss;
471 pkt.hdr.wnd = c->rcv.wnd;
473 pkt.hdr.aux = 0x0101;
477 pkt.init[3] = flags & 0x7;
479 set_state(c, SYN_SENT);
481 print_packet(utcp, "send", &pkt, sizeof(pkt));
482 utcp->send(utcp, &pkt, sizeof(pkt));
484 gettimeofday(&c->conn_timeout, NULL);
485 c->conn_timeout.tv_sec += utcp->timeout;
487 start_retransmit_timer(c);
492 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
493 return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
496 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
497 if(c->reapable || c->state != SYN_RECEIVED) {
498 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
502 debug("%p accepted, %p %p\n", c, recv, priv);
505 set_state(c, ESTABLISHED);
508 static void ack(struct utcp_connection *c, bool sendatleastone) {
509 int32_t left = seqdiff(c->snd.last, c->snd.nxt);
510 int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
511 debug("cwndleft = %d\n", cwndleft);
519 if(cwndleft < left) {
523 if(!left && !sendatleastone) {
532 pkt = malloc(sizeof(pkt->hdr) + c->utcp->mtu);
538 pkt->hdr.src = c->src;
539 pkt->hdr.dst = c->dst;
540 pkt->hdr.ack = c->rcv.nxt;
541 pkt->hdr.wnd = c->snd.wnd;
546 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
547 pkt->hdr.seq = c->snd.nxt;
549 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
551 c->snd.nxt += seglen;
554 if(seglen && fin_wanted(c, c->snd.nxt)) {
559 if(!c->rtt_start.tv_sec) {
560 // Start RTT measurement
561 gettimeofday(&c->rtt_start, NULL);
562 c->rtt_seq = pkt->hdr.seq + seglen;
563 debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
566 print_packet(c->utcp, "send", pkt, sizeof(pkt->hdr) + seglen);
567 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
573 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
575 debug("Error: send() called on closed connection %p\n", c);
583 debug("Error: send() called on unconnected connection %p\n", c);
598 debug("Error: send() called on closing connection %p\n", c);
603 // Exit early if we have nothing to send.
614 // Add data to send buffer.
616 len = buffer_put(&c->sndbuf, data, len);
625 // Don't send anything yet if the connection has not fully established yet
627 if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
633 if(!is_reliable(c)) {
634 c->snd.una = c->snd.nxt = c->snd.last;
635 buffer_get(&c->sndbuf, NULL, c->sndbuf.used);
638 if(is_reliable(c) && !timerisset(&c->rtrx_timeout)) {
639 start_retransmit_timer(c);
645 static void swap_ports(struct hdr *hdr) {
646 uint16_t tmp = hdr->src;
651 static void retransmit(struct utcp_connection *c) {
652 if(c->state == CLOSED || c->snd.last == c->snd.una) {
653 debug("Retransmit() called but nothing to retransmit!\n");
654 stop_retransmit_timer(c);
658 struct utcp *utcp = c->utcp;
665 pkt = malloc(sizeof(pkt->hdr) + c->utcp->mtu);
671 pkt->hdr.src = c->src;
672 pkt->hdr.dst = c->dst;
673 pkt->hdr.wnd = c->rcv.wnd;
678 // Send our SYN again
679 pkt->hdr.seq = c->snd.iss;
682 pkt->hdr.aux = 0x0101;
686 pkt->data[3] = c->flags & 0x7;
687 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + 4);
688 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
693 pkt->hdr.seq = c->snd.nxt;
694 pkt->hdr.ack = c->rcv.nxt;
695 pkt->hdr.ctl = SYN | ACK;
696 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr));
697 utcp->send(utcp, pkt, sizeof(pkt->hdr));
705 // Send unacked data again.
706 pkt->hdr.seq = c->snd.una;
707 pkt->hdr.ack = c->rcv.nxt;
709 uint32_t len = seqdiff(c->snd.last, c->snd.una);
711 if(len > utcp->mtu) {
715 if(fin_wanted(c, c->snd.una + len)) {
720 c->snd.nxt = c->snd.una + len;
721 c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
722 buffer_copy(&c->sndbuf, pkt->data, 0, len);
723 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + len);
724 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
731 // We shouldn't need to retransmit anything in this state.
735 stop_retransmit_timer(c);
739 start_retransmit_timer(c);
742 if(utcp->rto > MAX_RTO) {
746 c->rtt_start.tv_sec = 0; // invalidate RTT timer
752 /* Update receive buffer and SACK entries after consuming data.
756 * |.....0000..1111111111.....22222......3333|
759 * 0..3 represent the SACK entries. The ^ indicates up to which point we want
760 * to remove data from the receive buffer. The idea is to substract "len"
761 * from the offset of all the SACK entries, and then remove/cut down entries
762 * that are shifted to before the start of the receive buffer.
764 * There are three cases:
765 * - the SACK entry is after ^, in that case just change the offset.
766 * - the SACK entry starts before and ends after ^, so we have to
767 * change both its offset and size.
768 * - the SACK entry is completely before ^, in that case delete it.
770 static void sack_consume(struct utcp_connection *c, size_t len) {
771 debug("sack_consume %lu\n", (unsigned long)len);
773 if(len > c->rcvbuf.used) {
774 debug("All SACK entries consumed");
779 buffer_get(&c->rcvbuf, NULL, len);
781 for(int i = 0; i < NSACKS && c->sacks[i].len;) {
782 if(len < c->sacks[i].offset) {
783 c->sacks[i].offset -= len;
785 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
786 c->sacks[i].len -= len - c->sacks[i].offset;
787 c->sacks[i].offset = 0;
791 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
792 c->sacks[NSACKS - 1].len = 0;
800 for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
801 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
805 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
806 debug("out of order packet, offset %u\n", offset);
807 // Packet loss or reordering occured. Store the data in the buffer.
808 ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
810 if(rxd < 0 || (size_t)rxd < len) {
814 // Make note of where we put it.
815 for(int i = 0; i < NSACKS; i++) {
816 if(!c->sacks[i].len) { // nothing to merge, add new entry
817 debug("New SACK entry %d\n", i);
818 c->sacks[i].offset = offset;
819 c->sacks[i].len = rxd;
821 } else if(offset < c->sacks[i].offset) {
822 if(offset + rxd < c->sacks[i].offset) { // insert before
823 if(!c->sacks[NSACKS - 1].len) { // only if room left
824 debug("Insert SACK entry at %d\n", i);
825 memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
826 c->sacks[i].offset = offset;
827 c->sacks[i].len = rxd;
829 debug("SACK entries full, dropping packet\n");
834 debug("Merge with start of SACK entry at %d\n", i);
835 c->sacks[i].offset = offset;
838 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
839 if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
840 debug("Merge with end of SACK entry at %d\n", i);
841 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
842 // TODO: handle potential merge with next entry
849 for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
850 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
854 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
855 // Check if we can process out-of-order data now.
856 if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
857 debug("incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
858 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
859 len = max(len, c->sacks[0].offset + c->sacks[0].len);
860 data = c->rcvbuf.data;
864 ssize_t rxd = c->recv(c, data, len);
866 if(rxd < 0 || (size_t)rxd != len) {
867 // TODO: handle the application not accepting all data.
873 sack_consume(c, len);
880 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
881 if(!is_reliable(c)) {
882 c->recv(c, data, len);
883 c->rcv.nxt = seq + len;
887 uint32_t offset = seqdiff(seq, c->rcv.nxt);
889 if(offset + len > c->rcvbuf.maxsize) {
894 handle_out_of_order(c, offset, data, len);
896 handle_in_order(c, data, len);
901 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
902 const uint8_t *ptr = data;
918 print_packet(utcp, "recv", data, len);
920 // Drop packets smaller than the header
924 if(len < sizeof(hdr)) {
929 // Make a copy from the potentially unaligned data to a struct hdr
931 memcpy(&hdr, ptr, sizeof(hdr));
935 // Drop packets with an unknown CTL flag
937 if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
942 // Check for auxiliary headers
944 const uint8_t *init = NULL;
946 uint16_t aux = hdr.aux;
949 size_t auxlen = 4 * (aux >> 8) & 0xf;
950 uint8_t auxtype = aux & 0xff;
959 if(!(hdr.ctl & SYN) || auxlen != 4) {
984 memcpy(&aux, ptr, 2);
989 // Try to match the packet to an existing connection
991 struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
993 // Is it for a new connection?
996 // Ignore RST packets
1002 // Is it a SYN packet and are we LISTENing?
1004 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1005 // If we don't want to accept it, send a RST back
1006 if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
1011 // Try to allocate memory, otherwise send a RST back
1012 c = allocate_connection(utcp, hdr.dst, hdr.src);
1019 // Parse auxilliary information
1026 c->flags = init[3] & 0x7;
1028 c->flags = UTCP_TCP;
1031 // Return SYN+ACK, go to SYN_RECEIVED state
1032 c->snd.wnd = hdr.wnd;
1033 c->rcv.irs = hdr.seq;
1034 c->rcv.nxt = c->rcv.irs + 1;
1035 set_state(c, SYN_RECEIVED);
1042 pkt.hdr.src = c->src;
1043 pkt.hdr.dst = c->dst;
1044 pkt.hdr.ack = c->rcv.irs + 1;
1045 pkt.hdr.seq = c->snd.iss;
1046 pkt.hdr.wnd = c->rcv.wnd;
1047 pkt.hdr.ctl = SYN | ACK;
1050 pkt.hdr.aux = 0x0101;
1054 pkt.data[3] = c->flags & 0x7;
1055 print_packet(c->utcp, "send", &pkt, sizeof(hdr) + 4);
1056 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1059 print_packet(c->utcp, "send", &pkt, sizeof(hdr));
1060 utcp->send(utcp, &pkt, sizeof(hdr));
1063 // No, we don't want your packets, send a RST back
1071 debug("%p state %s\n", c->utcp, strstate[c->state]);
1073 // In case this is for a CLOSED connection, ignore the packet.
1074 // TODO: make it so incoming packets can never match a CLOSED connection.
1076 if(c->state == CLOSED) {
1077 debug("Got packet for closed connection\n");
1081 // It is for an existing connection.
1083 uint32_t prevrcvnxt = c->rcv.nxt;
1085 // 1. Drop invalid packets.
1087 // 1a. Drop packets that should not happen in our current state.
1108 // 1b. Drop packets with a sequence number not in our receive window.
1112 if(c->state == SYN_SENT) {
1114 } else if(len == 0) {
1115 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1117 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1119 // cut already accepted front overlapping
1120 if(rcv_offset < 0) {
1121 acceptable = len > (size_t) - rcv_offset;
1126 hdr.seq -= rcv_offset;
1129 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1134 debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1136 // Ignore unacceptable RST packets.
1141 // Otherwise, continue processing.
1145 c->snd.wnd = hdr.wnd; // TODO: move below
1147 // 1c. Drop packets with an invalid ACK.
1148 // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1149 // (= snd.una + c->sndbuf.used).
1151 if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1152 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1154 // Ignore unacceptable RST packets.
1162 // 2. Handle RST packets
1167 if(!(hdr.ctl & ACK)) {
1171 // The peer has refused our connection.
1172 set_state(c, CLOSED);
1173 errno = ECONNREFUSED;
1176 c->recv(c, NULL, 0);
1186 // We haven't told the application about this connection yet. Silently delete.
1198 // The peer has aborted our connection.
1199 set_state(c, CLOSED);
1203 c->recv(c, NULL, 0);
1215 // As far as the application is concerned, the connection has already been closed.
1216 // If it has called utcp_close() already, we can immediately free this connection.
1222 // Otherwise, immediately move to the CLOSED state.
1223 set_state(c, CLOSED);
1236 if(!(hdr.ctl & ACK)) {
1241 // 3. Advance snd.una
1243 advanced = seqdiff(hdr.ack, c->snd.una);
1244 prevrcvnxt = c->rcv.nxt;
1248 if(c->rtt_start.tv_sec) {
1249 if(c->rtt_seq == hdr.ack) {
1250 struct timeval now, diff;
1251 gettimeofday(&now, NULL);
1252 timersub(&now, &c->rtt_start, &diff);
1253 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
1254 c->rtt_start.tv_sec = 0;
1255 } else if(c->rtt_seq < hdr.ack) {
1256 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1257 c->rtt_start.tv_sec = 0;
1261 int32_t data_acked = advanced;
1269 // TODO: handle FIN as well.
1274 assert(data_acked >= 0);
1276 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1277 assert(data_acked <= bufused);
1280 buffer_get(&c->sndbuf, NULL, data_acked);
1283 // Also advance snd.nxt if possible
1284 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1285 c->snd.nxt = hdr.ack;
1288 c->snd.una = hdr.ack;
1291 c->snd.cwnd += utcp->mtu;
1293 if(c->snd.cwnd > c->sndbuf.maxsize) {
1294 c->snd.cwnd = c->sndbuf.maxsize;
1297 // Check if we have sent a FIN that is now ACKed.
1300 if(c->snd.una == c->snd.last) {
1301 set_state(c, FIN_WAIT_2);
1307 if(c->snd.una == c->snd.last) {
1308 gettimeofday(&c->conn_timeout, NULL);
1309 c->conn_timeout.tv_sec += 60;
1310 set_state(c, TIME_WAIT);
1319 if(!len && is_reliable(c)) {
1322 if(c->dupack == 3) {
1323 debug("Triplicate ACK\n");
1324 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1325 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1326 //Reset the congestion window so we wait for ACKs.
1327 c->snd.nxt = c->snd.una;
1328 c->snd.cwnd = utcp->mtu;
1329 start_retransmit_timer(c);
1337 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
1339 if(c->snd.una == c->snd.last) {
1340 stop_retransmit_timer(c);
1341 } else if(is_reliable(c)) {
1342 start_retransmit_timer(c);
1347 // 5. Process SYN stuff
1353 // This is a SYNACK. It should always have ACKed the SYN.
1358 c->rcv.irs = hdr.seq;
1359 c->rcv.nxt = hdr.seq;
1363 set_state(c, FIN_WAIT_1);
1365 set_state(c, ESTABLISHED);
1368 // TODO: notify application of this somehow.
1379 // Ehm, no. We should never receive a second SYN.
1389 // SYN counts as one sequence number
1393 // 6. Process new data
1395 if(c->state == SYN_RECEIVED) {
1396 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1401 // Are we still LISTENing?
1403 utcp->accept(c, c->src);
1406 if(c->state != ESTABLISHED) {
1407 set_state(c, CLOSED);
1417 // This should never happen.
1432 // Ehm no, We should never receive more data after a FIN.
1442 handle_incoming_data(c, hdr.seq, ptr, len);
1445 // 7. Process FIN stuff
1447 if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
1451 // This should never happen.
1458 set_state(c, CLOSE_WAIT);
1462 set_state(c, CLOSING);
1466 gettimeofday(&c->conn_timeout, NULL);
1467 c->conn_timeout.tv_sec += 60;
1468 set_state(c, TIME_WAIT);
1475 // Ehm, no. We should never receive a second FIN.
1485 // FIN counts as one sequence number
1489 // Inform the application that the peer closed the connection.
1492 c->recv(c, NULL, 0);
1496 // Now we send something back if:
1497 // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
1498 // -> sendatleastone = true
1499 // - or we got an ack, so we should maybe send a bit more data
1500 // -> sendatleastone = false
1502 ack(c, len || prevrcvnxt != c->rcv.nxt);
1514 hdr.ack = hdr.seq + len;
1516 hdr.ctl = RST | ACK;
1519 print_packet(utcp, "send", &hdr, sizeof(hdr));
1520 utcp->send(utcp, &hdr, sizeof(hdr));
1525 int utcp_shutdown(struct utcp_connection *c, int dir) {
1526 debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1534 debug("Error: shutdown() called on closed connection %p\n", c);
1539 if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1544 // TCP does not have a provision for stopping incoming packets.
1545 // The best we can do is to just ignore them.
1546 if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1550 // The rest of the code deals with shutting down writes.
1551 if(dir == UTCP_SHUT_RD) {
1555 // Only process shutting down writes once.
1573 set_state(c, FIN_WAIT_1);
1581 set_state(c, CLOSING);
1594 if(!timerisset(&c->rtrx_timeout)) {
1595 start_retransmit_timer(c);
1601 static bool reset_connection(struct utcp_connection *c) {
1608 debug("Error: abort() called on closed connection %p\n", c);
1625 set_state(c, CLOSED);
1633 set_state(c, CLOSED);
1643 hdr.seq = c->snd.nxt;
1648 print_packet(c->utcp, "send", &hdr, sizeof(hdr));
1649 c->utcp->send(c->utcp, &hdr, sizeof(hdr));
1653 // Closes all the opened connections
1654 void utcp_abort_all_connections(struct utcp *utcp) {
1660 for(int i = 0; i < utcp->nconnections; i++) {
1661 struct utcp_connection *c = utcp->connections[i];
1663 if(c->reapable || c->state == CLOSED) {
1667 utcp_recv_t old_recv = c->recv;
1669 reset_connection(c);
1673 old_recv(c, NULL, 0);
1680 int utcp_close(struct utcp_connection *c) {
1681 if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
1691 int utcp_abort(struct utcp_connection *c) {
1692 if(!reset_connection(c)) {
1701 * One call to this function will loop through all connections,
1702 * checking if something needs to be resent or not.
1703 * The return value is the time to the next timeout in milliseconds,
1704 * or maybe a negative value if the timeout is infinite.
1706 struct timeval utcp_timeout(struct utcp *utcp) {
1708 gettimeofday(&now, NULL);
1709 struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1711 for(int i = 0; i < utcp->nconnections; i++) {
1712 struct utcp_connection *c = utcp->connections[i];
1718 // delete connections that have been utcp_close()d.
1719 if(c->state == CLOSED) {
1721 debug("Reaping %p\n", c);
1729 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1734 c->recv(c, NULL, 0);
1740 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1741 debug("retransmit()\n");
1746 if((c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1747 uint32_t len = buffer_free(&c->sndbuf);
1752 } else if(c->state == CLOSED) {
1757 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <)) {
1758 next = c->conn_timeout;
1761 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <)) {
1762 next = c->rtrx_timeout;
1766 struct timeval diff;
1768 timersub(&next, &now, &diff);
1773 bool utcp_is_active(struct utcp *utcp) {
1778 for(int i = 0; i < utcp->nconnections; i++)
1779 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
1786 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1792 struct utcp *utcp = calloc(1, sizeof(*utcp));
1798 utcp->accept = accept;
1799 utcp->pre_accept = pre_accept;
1802 utcp->mtu = DEFAULT_MTU;
1803 utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1804 utcp->rto = START_RTO; // usec
1809 void utcp_exit(struct utcp *utcp) {
1814 for(int i = 0; i < utcp->nconnections; i++) {
1815 struct utcp_connection *c = utcp->connections[i];
1819 c->recv(c, NULL, 0);
1822 buffer_exit(&c->rcvbuf);
1823 buffer_exit(&c->sndbuf);
1827 free(utcp->connections);
1831 uint16_t utcp_get_mtu(struct utcp *utcp) {
1832 return utcp ? utcp->mtu : 0;
1835 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1836 // TODO: handle overhead of the header
1842 void utcp_reset_timers(struct utcp *utcp) {
1847 struct timeval now, then;
1849 gettimeofday(&now, NULL);
1853 then.tv_sec += utcp->timeout;
1855 for(int i = 0; i < utcp->nconnections; i++) {
1856 utcp->connections[i]->rtrx_timeout = now;
1857 utcp->connections[i]->conn_timeout = then;
1858 utcp->connections[i]->rtt_start.tv_sec = 0;
1861 if(utcp->rto > START_RTO) {
1862 utcp->rto = START_RTO;
1866 int utcp_get_user_timeout(struct utcp *u) {
1867 return u ? u->timeout : 0;
1870 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1872 u->timeout = timeout;
1876 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1877 return c ? c->sndbuf.maxsize : 0;
1880 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1890 return buffer_free(&c->sndbuf);
1897 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1902 c->sndbuf.maxsize = size;
1904 if(c->sndbuf.maxsize != size) {
1905 c->sndbuf.maxsize = -1;
1909 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1910 return c ? c->rcvbuf.maxsize : 0;
1913 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1914 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1915 return buffer_free(&c->rcvbuf);
1921 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1926 c->rcvbuf.maxsize = size;
1928 if(c->rcvbuf.maxsize != size) {
1929 c->rcvbuf.maxsize = -1;
1933 size_t utcp_get_sendq(struct utcp_connection *c) {
1934 return c->sndbuf.used;
1937 size_t utcp_get_recvq(struct utcp_connection *c) {
1938 return c->rcvbuf.used;
1941 bool utcp_get_nodelay(struct utcp_connection *c) {
1942 return c ? c->nodelay : false;
1945 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1947 c->nodelay = nodelay;
1951 bool utcp_get_keepalive(struct utcp_connection *c) {
1952 return c ? c->keepalive : false;
1955 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1957 c->keepalive = keepalive;
1961 size_t utcp_get_outq(struct utcp_connection *c) {
1962 return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
1965 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
1971 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
1977 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
1979 utcp->accept = accept;
1980 utcp->pre_accept = pre_accept;