2 utcp.c -- Userspace TCP
3 Copyright (C) 2014 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"
38 static void debug(const char *format, ...) {
41 vfprintf(stderr, format, ap);
45 static void print_packet(struct utcp *utcp, const char *dir, const void *pkt, size_t len) {
47 if(len < sizeof hdr) {
48 debug("%p %s: short packet (%zu bytes)\n", utcp, dir, len);
52 memcpy(&hdr, pkt, sizeof hdr);
53 fprintf (stderr, "%p %s: len=%zu, src=%u dst=%u seq=%u ack=%u wnd=%u ctl=", utcp, dir, len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd);
63 if(len > sizeof hdr) {
65 for(int i = sizeof hdr; i < len; i++) {
66 const char *data = pkt;
67 debug("%c", data[i] >= 32 ? data[i] : '.');
75 #define print_packet(...)
78 static void set_state(struct utcp_connection *c, enum state state) {
80 if(state == ESTABLISHED)
81 timerclear(&c->conn_timeout);
82 debug("%p new state: %s\n", c->utcp, strstate[state]);
85 static inline void list_connections(struct utcp *utcp) {
86 debug("%p has %d connections:\n", utcp, utcp->nconnections);
87 for(int i = 0; i < utcp->nconnections; i++)
88 debug(" %u -> %u state %s\n", utcp->connections[i]->src, utcp->connections[i]->dst, strstate[utcp->connections[i]->state]);
91 static int32_t seqdiff(uint32_t a, uint32_t b) {
95 // Connections are stored in a sorted list.
96 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
98 static int compare(const void *va, const void *vb) {
99 const struct utcp_connection *a = *(struct utcp_connection **)va;
100 const struct utcp_connection *b = *(struct utcp_connection **)vb;
102 assert(a->src && b->src);
104 int c = (int)a->src - (int)b->src;
107 c = (int)a->dst - (int)b->dst;
111 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
112 if(!utcp->nconnections)
114 struct utcp_connection key = {
118 struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
119 return match ? *match : NULL;
122 static void free_connection(struct utcp_connection *c) {
123 struct utcp *utcp = c->utcp;
124 struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
128 int i = cp - utcp->connections;
129 memmove(cp + i, cp + i + 1, (utcp->nconnections - i - 1) * sizeof *cp);
130 utcp->nconnections--;
136 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
137 // Check whether this combination of src and dst is free
140 if(find_connection(utcp, src, dst)) {
144 } else { // If src == 0, generate a random port number with the high bit set
145 if(utcp->nconnections >= 32767) {
149 src = rand() | 0x8000;
150 while(find_connection(utcp, src, dst))
154 // Allocate memory for the new connection
156 if(utcp->nconnections >= utcp->nallocated) {
157 if(!utcp->nallocated)
158 utcp->nallocated = 4;
160 utcp->nallocated *= 2;
161 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof *utcp->connections);
164 utcp->connections = new_array;
167 struct utcp_connection *c = calloc(1, sizeof *c);
171 c->sndbufsize = DEFAULT_SNDBUFSIZE;
172 c->maxsndbufsize = DEFAULT_MAXSNDBUFSIZE;
173 c->sndbuf = malloc(c->sndbufsize);
179 // Fill in the details
184 c->snd.una = c->snd.iss;
185 c->snd.nxt = c->snd.iss + 1;
186 c->rcv.wnd = utcp->mtu;
187 c->snd.last = c->snd.nxt;
188 c->snd.cwnd = utcp->mtu;
191 // Add it to the sorted list of connections
193 utcp->connections[utcp->nconnections++] = c;
194 qsort(utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
199 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
200 struct utcp_connection *c = allocate_connection(utcp, 0, dst);
210 hdr.seq = c->snd.iss;
212 hdr.wnd = c->rcv.wnd;
216 set_state(c, SYN_SENT);
218 print_packet(utcp, "send", &hdr, sizeof hdr);
219 utcp->send(utcp, &hdr, sizeof hdr);
221 gettimeofday(&c->conn_timeout, NULL);
222 c->conn_timeout.tv_sec += utcp->timeout;
227 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
228 if(c->reapable || c->state != SYN_RECEIVED) {
229 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
233 debug("%p accepted, %p %p\n", c, recv, priv);
236 set_state(c, ESTABLISHED);
239 static void ack(struct utcp_connection *c, bool sendatleastone) {
240 int32_t left = seqdiff(c->snd.last, c->snd.nxt);
241 int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
242 char *data = c->sndbuf + seqdiff(c->snd.nxt, c->snd.una);
252 if(!left && !sendatleastone)
257 char data[c->utcp->mtu];
260 pkt.hdr.src = c->src;
261 pkt.hdr.dst = c->dst;
262 pkt.hdr.ack = c->rcv.nxt;
263 pkt.hdr.wnd = c->snd.wnd;
268 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
269 pkt.hdr.seq = c->snd.nxt;
271 memcpy(pkt.data, data, seglen);
273 c->snd.nxt += seglen;
277 if(c->state != ESTABLISHED && !left && seglen) {
289 print_packet(c->utcp, "send", &pkt, sizeof pkt.hdr + seglen);
290 c->utcp->send(c->utcp, &pkt, sizeof pkt.hdr + seglen);
294 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
296 debug("Error: send() called on closed connection %p\n", c);
306 debug("Error: send() called on unconnected connection %p\n", c);
317 debug("Error: send() called on closing connection %p\n", c);
322 // Add data to send buffer
332 uint32_t bufused = seqdiff(c->snd.nxt, c->snd.una);
334 /* Check our send buffer.
335 * - If it's big enough, just put the data in there.
336 * - If not, decide whether to enlarge if possible.
337 * - Cap len so it doesn't overflow our buffer.
340 if(len > c->sndbufsize - bufused && c->sndbufsize < c->maxsndbufsize) {
342 if(c->sndbufsize > c->maxsndbufsize / 2)
343 newbufsize = c->maxsndbufsize;
345 newbufsize = c->sndbufsize * 2;
346 if(bufused + len > newbufsize) {
347 if(bufused + len > c->maxsndbufsize)
348 newbufsize = c->maxsndbufsize;
350 newbufsize = bufused + len;
352 char *newbuf = realloc(c->sndbuf, newbufsize);
355 c->sndbufsize = newbufsize;
359 if(len > c->sndbufsize - bufused)
360 len = c->sndbufsize - bufused;
363 errno == EWOULDBLOCK;
367 memcpy(c->sndbuf + bufused, data, len);
374 static void swap_ports(struct hdr *hdr) {
375 uint16_t tmp = hdr->src;
380 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
394 print_packet(utcp, "recv", data, len);
396 // Drop packets smaller than the header
399 if(len < sizeof hdr) {
404 // Make a copy from the potentially unaligned data to a struct hdr
406 memcpy(&hdr, data, sizeof hdr);
410 // Drop packets with an unknown CTL flag
412 if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
417 // Try to match the packet to an existing connection
419 struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
421 // Is it for a new connection?
424 // Ignore RST packets
429 // Is it a SYN packet and are we LISTENing?
431 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
432 // If we don't want to accept it, send a RST back
433 if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
438 // Try to allocate memory, otherwise send a RST back
439 c = allocate_connection(utcp, hdr.dst, hdr.src);
445 // Return SYN+ACK, go to SYN_RECEIVED state
446 c->snd.wnd = hdr.wnd;
447 c->rcv.irs = hdr.seq;
448 c->rcv.nxt = c->rcv.irs + 1;
449 set_state(c, SYN_RECEIVED);
453 hdr.ack = c->rcv.irs + 1;
454 hdr.seq = c->snd.iss;
456 print_packet(c->utcp, "send", &hdr, sizeof hdr);
457 utcp->send(utcp, &hdr, sizeof hdr);
459 // No, we don't want your packets, send a RST back
467 debug("%p state %s\n", c->utcp, strstate[c->state]);
469 // In case this is for a CLOSED connection, ignore the packet.
470 // TODO: make it so incoming packets can never match a CLOSED connection.
472 if(c->state == CLOSED)
475 // It is for an existing connection.
477 // 1. Drop invalid packets.
479 // 1a. Drop packets that should not happen in our current state.
496 // 1b. Drop packets with a sequence number not in our receive window.
500 if(c->state == SYN_SENT)
503 // TODO: handle packets overlapping c->rcv.nxt.
505 // Only use this when accepting out-of-order packets.
508 acceptable = hdr.seq == c->rcv.nxt;
510 acceptable = (seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt + c->rcv.wnd) < 0);
513 // We don't accept data when the receive window is zero.
516 // Both start and end of packet must be within the receive window
517 acceptable = (seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt + c->rcv.wnd) < 0)
518 || (seqdiff(hdr.seq + len + 1, c->rcv.nxt) >= 0 && seqdiff(hdr.seq + len - 1, c->rcv.nxt + c->rcv.wnd) < 0);
520 if(c->state != SYN_SENT)
521 acceptable = hdr.seq == c->rcv.nxt;
525 debug("Packet not acceptable, %u <= %u + %zu < %u\n", c->rcv.nxt, hdr.seq, len, c->rcv.nxt + c->rcv.wnd);
526 // Ignore unacceptable RST packets.
529 // Otherwise, send an ACK back in the hope things improve.
533 c->snd.wnd = hdr.wnd; // TODO: move below
535 // 1c. Drop packets with an invalid ACK.
536 // ackno should not roll back, and it should also not be bigger than snd.nxt.
538 if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.nxt) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
539 debug("Packet ack seqno out of range, %u %u %u\n", hdr.ack, c->snd.una, c->snd.nxt);
540 // Ignore unacceptable RST packets.
546 // 2. Handle RST packets
553 // The peer has refused our connection.
554 set_state(c, CLOSED);
555 errno = ECONNREFUSED;
562 // We haven't told the application about this connection yet. Silently delete.
571 // The peer has aborted our connection.
572 set_state(c, CLOSED);
582 // As far as the application is concerned, the connection has already been closed.
583 // If it has called utcp_close() already, we can immediately free this connection.
588 // Otherwise, immediately move to the CLOSED state.
589 set_state(c, CLOSED);
596 // 3. Advance snd.una
598 uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
599 uint32_t prevrcvnxt = c->rcv.nxt;
602 int32_t data_acked = advanced;
609 // TODO: handle FIN as well.
614 assert(data_acked >= 0);
616 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
617 assert(data_acked <= bufused);
619 // Make room in the send buffer.
620 // TODO: try to avoid memmoving too much. Circular buffer?
621 uint32_t left = bufused - data_acked;
622 if(data_acked && left)
623 memmove(c->sndbuf, c->sndbuf + data_acked, left);
625 c->snd.una = hdr.ack;
628 c->snd.cwnd += utcp->mtu;
629 if(c->snd.cwnd > c->maxsndbufsize)
630 c->snd.cwnd = c->maxsndbufsize;
632 // Check if we have sent a FIN that is now ACKed.
635 if(c->snd.una == c->snd.last)
636 set_state(c, FIN_WAIT_2);
639 if(c->snd.una == c->snd.last) {
640 gettimeofday(&c->conn_timeout, NULL);
641 c->conn_timeout.tv_sec += 60;
642 set_state(c, TIME_WAIT);
652 debug("Triplicate ACK\n");
653 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
662 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
663 if(c->snd.una == c->snd.nxt)
664 timerclear(&c->rtrx_timeout);
667 // 5. Process SYN stuff
672 // This is a SYNACK. It should always have ACKed the SYN.
675 c->rcv.irs = hdr.seq;
676 c->rcv.nxt = hdr.seq;
677 set_state(c, ESTABLISHED);
678 // TODO: notify application of this somehow.
688 // Ehm, no. We should never receive a second SYN.
694 // SYN counts as one sequence number
698 // 6. Process new data
700 if(c->state == SYN_RECEIVED) {
701 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
705 // Are we still LISTENing?
707 utcp->accept(c, c->src);
709 if(c->state != ESTABLISHED) {
710 set_state(c, CLOSED);
720 // This should never happen.
730 // Ehm no, We should never receive more data after a FIN.
739 rxd = c->recv(c, data, len);
741 // TODO: once we have a receive buffer, handle the application not accepting all data.
742 fprintf(stderr, "c->recv(%p, %p, %zu) returned %zd\n", c, data, len, rxd);
748 rxd = len; // Bad application, bad!
756 // 7. Process FIN stuff
762 // This should never happen.
765 set_state(c, CLOSE_WAIT);
768 set_state(c, CLOSING);
771 gettimeofday(&c->conn_timeout, NULL);
772 c->conn_timeout.tv_sec += 60;
773 set_state(c, TIME_WAIT);
779 // Ehm, no. We should never receive a second FIN.
785 // FIN counts as one sequence number
789 // Inform the application that the peer closed the connection.
796 // Now we send something back if:
797 // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
798 // -> sendatleastone = true
799 // - or we got an ack, so we should maybe send a bit more data
800 // -> sendatleastone = false
803 ack(c, prevrcvnxt != c->rcv.nxt);
813 hdr.ack = hdr.seq + len;
817 print_packet(utcp, "send", &hdr, sizeof hdr);
818 utcp->send(utcp, &hdr, sizeof hdr);
823 int utcp_shutdown(struct utcp_connection *c, int dir) {
824 debug("%p shutdown %d\n", c ? c->utcp : NULL, dir);
831 debug("Error: shutdown() called on closed connection %p\n", c);
843 set_state(c, CLOSED);
848 set_state(c, FIN_WAIT_1);
854 set_state(c, CLOSING);
869 int utcp_close(struct utcp_connection *c) {
870 if(utcp_shutdown(c, SHUT_RDWR))
876 int utcp_abort(struct utcp_connection *c) {
883 debug("Error: abort() called on closed connection %p\n", c);
898 set_state(c, CLOSED);
906 set_state(c, CLOSED);
916 hdr.seq = c->snd.nxt;
921 print_packet(c->utcp, "send", &hdr, sizeof hdr);
922 c->utcp->send(c->utcp, &hdr, sizeof hdr);
926 static void retransmit(struct utcp_connection *c) {
927 if(c->state == CLOSED || c->snd.nxt == c->snd.una)
930 struct utcp *utcp = c->utcp;
934 char data[c->utcp->mtu];
937 pkt.hdr.src = c->src;
938 pkt.hdr.dst = c->dst;
942 // TODO: this should not happen
946 pkt.hdr.seq = c->snd.iss;
948 pkt.hdr.wnd = c->rcv.wnd;
950 print_packet(c->utcp, "rtrx", &pkt, sizeof pkt.hdr);
951 utcp->send(utcp, &pkt, sizeof pkt.hdr);
955 pkt.hdr.seq = c->snd.nxt;
956 pkt.hdr.ack = c->rcv.nxt;
957 pkt.hdr.ctl = SYN | ACK;
958 print_packet(c->utcp, "rtrx", &pkt, sizeof pkt.hdr);
959 utcp->send(utcp, &pkt, sizeof pkt.hdr);
964 pkt.hdr.seq = c->snd.una;
965 pkt.hdr.ack = c->rcv.nxt;
967 uint32_t len = seqdiff(c->snd.nxt, c->snd.una);
968 if(c->state == FIN_WAIT_1)
973 if(c->state == FIN_WAIT_1)
976 memcpy(pkt.data, c->sndbuf, len);
977 print_packet(c->utcp, "rtrx", &pkt, sizeof pkt.hdr + len);
978 utcp->send(utcp, &pkt, sizeof pkt.hdr + len);
988 * One call to this function will loop through all connections,
989 * checking if something needs to be resent or not.
990 * The return value is the time to the next timeout in milliseconds,
991 * or maybe a negative value if the timeout is infinite.
993 int utcp_timeout(struct utcp *utcp) {
995 gettimeofday(&now, NULL);
996 struct timeval next = {now.tv_sec + 3600, now.tv_usec};
998 for(int i = 0; i < utcp->nconnections; i++) {
999 struct utcp_connection *c = utcp->connections[i];
1003 if(c->state == CLOSED) {
1005 debug("Reaping %p\n", c);
1012 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1016 c->recv(c, NULL, 0);
1020 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1024 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
1025 next = c->conn_timeout;
1027 if(c->snd.nxt != c->snd.una) {
1028 c->rtrx_timeout = now;
1029 c->rtrx_timeout.tv_sec++;
1031 timerclear(&c->rtrx_timeout);
1034 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
1035 next = c->rtrx_timeout;
1038 struct timeval diff;
1039 timersub(&next, &now, &diff);
1042 return diff.tv_sec * 1000 + diff.tv_usec / 1000;
1045 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1046 struct utcp *utcp = calloc(1, sizeof *utcp);
1055 utcp->accept = accept;
1056 utcp->pre_accept = pre_accept;
1065 void utcp_exit(struct utcp *utcp) {
1068 for(int i = 0; i < utcp->nconnections; i++) {
1069 if(!utcp->connections[i]->reapable)
1070 debug("Warning, freeing unclosed connection %p\n", utcp->connections[i]);
1071 free(utcp->connections[i]->sndbuf);
1072 free(utcp->connections[i]);
1074 free(utcp->connections);
1078 uint16_t utcp_get_mtu(struct utcp *utcp) {
1082 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1083 // TODO: handle overhead of the header
1087 int utcp_get_user_timeout(struct utcp *u) {
1091 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1092 u->timeout = timeout;
1095 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1096 return c->maxsndbufsize;
1099 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1100 c->maxsndbufsize = size;
1101 if(c->maxsndbufsize != size)
1102 c->maxsndbufsize = -1;
1105 bool utcp_get_nodelay(struct utcp_connection *c) {
1109 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1110 c->nodelay = nodelay;
1113 bool utcp_get_keepalive(struct utcp_connection *c) {
1114 return c->keepalive;
1117 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1118 c->keepalive = keepalive;
1121 size_t utcp_get_outq(struct utcp_connection *c) {
1122 return seqdiff(c->snd.nxt, c->snd.una);