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);
211 hdr.seq = c->snd.iss;
213 hdr.wnd = c->rcv.wnd;
217 set_state(c, SYN_SENT);
219 print_packet(utcp, "send", &hdr, sizeof hdr);
220 utcp->send(utcp, &hdr, sizeof hdr);
222 gettimeofday(&c->conn_timeout, NULL);
223 c->conn_timeout.tv_sec += utcp->timeout;
228 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
229 if(c->reapable || c->state != SYN_RECEIVED) {
230 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
234 debug("%p accepted, %p %p\n", c, recv, priv);
237 set_state(c, ESTABLISHED);
240 static void ack(struct utcp_connection *c, bool sendatleastone) {
241 int32_t left = seqdiff(c->snd.last, c->snd.nxt);
242 int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
243 char *data = c->sndbuf + seqdiff(c->snd.nxt, c->snd.una);
253 if(!left && !sendatleastone)
258 char data[c->utcp->mtu];
261 pkt.hdr.src = c->src;
262 pkt.hdr.dst = c->dst;
263 pkt.hdr.ack = c->rcv.nxt;
264 pkt.hdr.wnd = c->snd.wnd;
269 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
270 pkt.hdr.seq = c->snd.nxt;
272 memcpy(pkt.data, data, seglen);
274 c->snd.nxt += seglen;
278 if(c->state != ESTABLISHED && !left && seglen) {
290 print_packet(c->utcp, "send", &pkt, sizeof pkt.hdr + seglen);
291 c->utcp->send(c->utcp, &pkt, sizeof pkt.hdr + seglen);
295 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
297 debug("Error: send() called on closed connection %p\n", c);
307 debug("Error: send() called on unconnected connection %p\n", c);
318 debug("Error: send() called on closing connection %p\n", c);
323 // Add data to send buffer
333 uint32_t bufused = seqdiff(c->snd.nxt, c->snd.una);
335 /* Check our send buffer.
336 * - If it's big enough, just put the data in there.
337 * - If not, decide whether to enlarge if possible.
338 * - Cap len so it doesn't overflow our buffer.
341 if(len > c->sndbufsize - bufused && c->sndbufsize < c->maxsndbufsize) {
343 if(c->sndbufsize > c->maxsndbufsize / 2)
344 newbufsize = c->maxsndbufsize;
346 newbufsize = c->sndbufsize * 2;
347 if(bufused + len > newbufsize) {
348 if(bufused + len > c->maxsndbufsize)
349 newbufsize = c->maxsndbufsize;
351 newbufsize = bufused + len;
353 char *newbuf = realloc(c->sndbuf, newbufsize);
356 c->sndbufsize = newbufsize;
360 if(len > c->sndbufsize - bufused)
361 len = c->sndbufsize - bufused;
364 errno == EWOULDBLOCK;
368 memcpy(c->sndbuf + bufused, data, len);
375 static void swap_ports(struct hdr *hdr) {
376 uint16_t tmp = hdr->src;
381 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
395 print_packet(utcp, "recv", data, len);
397 // Drop packets smaller than the header
400 if(len < sizeof hdr) {
405 // Make a copy from the potentially unaligned data to a struct hdr
407 memcpy(&hdr, data, sizeof hdr);
411 // Drop packets with an unknown CTL flag
413 if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
418 // Try to match the packet to an existing connection
420 struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
422 // Is it for a new connection?
425 // Ignore RST packets
430 // Is it a SYN packet and are we LISTENing?
432 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
433 // If we don't want to accept it, send a RST back
434 if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
439 // Try to allocate memory, otherwise send a RST back
440 c = allocate_connection(utcp, hdr.dst, hdr.src);
446 // Return SYN+ACK, go to SYN_RECEIVED state
447 c->snd.wnd = hdr.wnd;
448 c->rcv.irs = hdr.seq;
449 c->rcv.nxt = c->rcv.irs + 1;
450 set_state(c, SYN_RECEIVED);
454 hdr.ack = c->rcv.irs + 1;
455 hdr.seq = c->snd.iss;
457 print_packet(c->utcp, "send", &hdr, sizeof hdr);
458 utcp->send(utcp, &hdr, sizeof hdr);
460 // No, we don't want your packets, send a RST back
468 debug("%p state %s\n", c->utcp, strstate[c->state]);
470 // In case this is for a CLOSED connection, ignore the packet.
471 // TODO: make it so incoming packets can never match a CLOSED connection.
473 if(c->state == CLOSED)
476 // It is for an existing connection.
478 // 1. Drop invalid packets.
480 // 1a. Drop packets that should not happen in our current state.
497 // 1b. Drop packets with a sequence number not in our receive window.
501 if(c->state == SYN_SENT)
504 // TODO: handle packets overlapping c->rcv.nxt.
506 // Only use this when accepting out-of-order packets.
509 acceptable = hdr.seq == c->rcv.nxt;
511 acceptable = (seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt + c->rcv.wnd) < 0);
514 // We don't accept data when the receive window is zero.
517 // Both start and end of packet must be within the receive window
518 acceptable = (seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt + c->rcv.wnd) < 0)
519 || (seqdiff(hdr.seq + len + 1, c->rcv.nxt) >= 0 && seqdiff(hdr.seq + len - 1, c->rcv.nxt + c->rcv.wnd) < 0);
521 if(c->state != SYN_SENT)
522 acceptable = hdr.seq == c->rcv.nxt;
526 debug("Packet not acceptable, %u <= %u + %zu < %u\n", c->rcv.nxt, hdr.seq, len, c->rcv.nxt + c->rcv.wnd);
527 // Ignore unacceptable RST packets.
530 // Otherwise, send an ACK back in the hope things improve.
534 c->snd.wnd = hdr.wnd; // TODO: move below
536 // 1c. Drop packets with an invalid ACK.
537 // ackno should not roll back, and it should also not be bigger than snd.nxt.
539 if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.nxt) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
540 debug("Packet ack seqno out of range, %u %u %u\n", hdr.ack, c->snd.una, c->snd.nxt);
541 // Ignore unacceptable RST packets.
547 // 2. Handle RST packets
554 // The peer has refused our connection.
555 set_state(c, CLOSED);
556 errno = ECONNREFUSED;
563 // We haven't told the application about this connection yet. Silently delete.
572 // The peer has aborted our connection.
573 set_state(c, CLOSED);
583 // As far as the application is concerned, the connection has already been closed.
584 // If it has called utcp_close() already, we can immediately free this connection.
589 // Otherwise, immediately move to the CLOSED state.
590 set_state(c, CLOSED);
597 // 3. Advance snd.una
599 uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
600 uint32_t prevrcvnxt = c->rcv.nxt;
603 int32_t data_acked = advanced;
610 // TODO: handle FIN as well.
615 assert(data_acked >= 0);
617 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
618 assert(data_acked <= bufused);
620 // Make room in the send buffer.
621 // TODO: try to avoid memmoving too much. Circular buffer?
622 uint32_t left = bufused - data_acked;
623 if(data_acked && left)
624 memmove(c->sndbuf, c->sndbuf + data_acked, left);
626 c->snd.una = hdr.ack;
629 c->snd.cwnd += utcp->mtu;
630 if(c->snd.cwnd > c->maxsndbufsize)
631 c->snd.cwnd = c->maxsndbufsize;
633 // Check if we have sent a FIN that is now ACKed.
636 if(c->snd.una == c->snd.last)
637 set_state(c, FIN_WAIT_2);
640 if(c->snd.una == c->snd.last) {
641 gettimeofday(&c->conn_timeout, NULL);
642 c->conn_timeout.tv_sec += 60;
643 set_state(c, TIME_WAIT);
653 debug("Triplicate ACK\n");
654 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
663 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
664 if(c->snd.una == c->snd.nxt)
665 timerclear(&c->rtrx_timeout);
668 // 5. Process SYN stuff
673 // This is a SYNACK. It should always have ACKed the SYN.
676 c->rcv.irs = hdr.seq;
677 c->rcv.nxt = hdr.seq;
678 set_state(c, ESTABLISHED);
679 // TODO: notify application of this somehow.
689 // Ehm, no. We should never receive a second SYN.
695 // SYN counts as one sequence number
699 // 6. Process new data
701 if(c->state == SYN_RECEIVED) {
702 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
706 // Are we still LISTENing?
708 utcp->accept(c, c->src);
710 if(c->state != ESTABLISHED) {
711 set_state(c, CLOSED);
721 // This should never happen.
731 // Ehm no, We should never receive more data after a FIN.
740 rxd = c->recv(c, data, len);
742 // TODO: once we have a receive buffer, handle the application not accepting all data.
743 fprintf(stderr, "c->recv(%p, %p, %zu) returned %zd\n", c, data, len, rxd);
749 rxd = len; // Bad application, bad!
757 // 7. Process FIN stuff
763 // This should never happen.
766 set_state(c, CLOSE_WAIT);
769 set_state(c, CLOSING);
772 gettimeofday(&c->conn_timeout, NULL);
773 c->conn_timeout.tv_sec += 60;
774 set_state(c, TIME_WAIT);
780 // Ehm, no. We should never receive a second FIN.
786 // FIN counts as one sequence number
790 // Inform the application that the peer closed the connection.
797 // Now we send something back if:
798 // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
799 // -> sendatleastone = true
800 // - or we got an ack, so we should maybe send a bit more data
801 // -> sendatleastone = false
804 ack(c, prevrcvnxt != c->rcv.nxt);
814 hdr.ack = hdr.seq + len;
818 print_packet(utcp, "send", &hdr, sizeof hdr);
819 utcp->send(utcp, &hdr, sizeof hdr);
824 int utcp_shutdown(struct utcp_connection *c, int dir) {
825 debug("%p shutdown %d\n", c ? c->utcp : NULL, dir);
832 debug("Error: shutdown() called on closed connection %p\n", c);
844 set_state(c, CLOSED);
849 set_state(c, FIN_WAIT_1);
855 set_state(c, CLOSING);
870 int utcp_close(struct utcp_connection *c) {
871 if(utcp_shutdown(c, SHUT_RDWR))
877 int utcp_abort(struct utcp_connection *c) {
884 debug("Error: abort() called on closed connection %p\n", c);
899 set_state(c, CLOSED);
907 set_state(c, CLOSED);
917 hdr.seq = c->snd.nxt;
922 print_packet(c->utcp, "send", &hdr, sizeof hdr);
923 c->utcp->send(c->utcp, &hdr, sizeof hdr);
927 static void retransmit(struct utcp_connection *c) {
928 if(c->state == CLOSED || c->snd.nxt == c->snd.una)
931 struct utcp *utcp = c->utcp;
935 char data[c->utcp->mtu];
938 pkt.hdr.src = c->src;
939 pkt.hdr.dst = c->dst;
943 // TODO: this should not happen
947 pkt.hdr.seq = c->snd.iss;
949 pkt.hdr.wnd = c->rcv.wnd;
951 print_packet(c->utcp, "rtrx", &pkt, sizeof pkt.hdr);
952 utcp->send(utcp, &pkt, sizeof pkt.hdr);
956 pkt.hdr.seq = c->snd.nxt;
957 pkt.hdr.ack = c->rcv.nxt;
958 pkt.hdr.ctl = SYN | ACK;
959 print_packet(c->utcp, "rtrx", &pkt, sizeof pkt.hdr);
960 utcp->send(utcp, &pkt, sizeof pkt.hdr);
965 pkt.hdr.seq = c->snd.una;
966 pkt.hdr.ack = c->rcv.nxt;
968 uint32_t len = seqdiff(c->snd.nxt, c->snd.una);
969 if(c->state == FIN_WAIT_1)
974 if(c->state == FIN_WAIT_1)
977 memcpy(pkt.data, c->sndbuf, len);
978 print_packet(c->utcp, "rtrx", &pkt, sizeof pkt.hdr + len);
979 utcp->send(utcp, &pkt, sizeof pkt.hdr + len);
989 * One call to this function will loop through all connections,
990 * checking if something needs to be resent or not.
991 * The return value is the time to the next timeout in milliseconds,
992 * or maybe a negative value if the timeout is infinite.
994 int utcp_timeout(struct utcp *utcp) {
996 gettimeofday(&now, NULL);
997 struct timeval next = {now.tv_sec + 3600, now.tv_usec};
999 for(int i = 0; i < utcp->nconnections; i++) {
1000 struct utcp_connection *c = utcp->connections[i];
1004 if(c->state == CLOSED) {
1006 debug("Reaping %p\n", c);
1013 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1017 c->recv(c, NULL, 0);
1021 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1025 if(c->poll && c->sndbufsize < c->maxsndbufsize / 2)
1026 c->poll(c, c->maxsndbufsize - c->sndbufsize);
1028 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
1029 next = c->conn_timeout;
1031 if(c->snd.nxt != c->snd.una) {
1032 c->rtrx_timeout = now;
1033 c->rtrx_timeout.tv_sec++;
1035 timerclear(&c->rtrx_timeout);
1038 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
1039 next = c->rtrx_timeout;
1042 struct timeval diff;
1043 timersub(&next, &now, &diff);
1046 return diff.tv_sec * 1000 + diff.tv_usec / 1000;
1049 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1050 struct utcp *utcp = calloc(1, sizeof *utcp);
1059 utcp->accept = accept;
1060 utcp->pre_accept = pre_accept;
1069 void utcp_exit(struct utcp *utcp) {
1072 for(int i = 0; i < utcp->nconnections; i++) {
1073 if(!utcp->connections[i]->reapable)
1074 debug("Warning, freeing unclosed connection %p\n", utcp->connections[i]);
1075 free(utcp->connections[i]->sndbuf);
1076 free(utcp->connections[i]);
1078 free(utcp->connections);
1082 uint16_t utcp_get_mtu(struct utcp *utcp) {
1086 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1087 // TODO: handle overhead of the header
1091 int utcp_get_user_timeout(struct utcp *u) {
1095 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1096 u->timeout = timeout;
1099 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1100 return c->maxsndbufsize;
1103 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1104 return c->maxsndbufsize - c->sndbufsize;
1107 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1108 c->maxsndbufsize = size;
1109 if(c->maxsndbufsize != size)
1110 c->maxsndbufsize = -1;
1113 bool utcp_get_nodelay(struct utcp_connection *c) {
1117 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1118 c->nodelay = nodelay;
1121 bool utcp_get_keepalive(struct utcp_connection *c) {
1122 return c->keepalive;
1125 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1126 c->keepalive = keepalive;
1129 size_t utcp_get_outq(struct utcp_connection *c) {
1130 return seqdiff(c->snd.nxt, c->snd.una);
1133 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
1137 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {