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.
30 #include <sys/socket.h>
32 #include "utcp_priv.h"
37 static void debug(const char *format, ...) {
40 vfprintf(stderr, format, ap);
44 static void print_packet(struct utcp *utcp, const char *dir, const void *pkt, size_t len) {
46 if(len < sizeof hdr) {
47 debug("%p %s: short packet (%zu bytes)\n", utcp, dir, len);
51 memcpy(&hdr, pkt, sizeof hdr);
52 fprintf (stderr, "%p %s: src=%u dst=%u seq=%u ack=%u wnd=%u ctl=", utcp, dir, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd);
62 if(len > sizeof hdr) {
64 for(int i = sizeof hdr; i < len; i++) {
65 const char *data = pkt;
66 debug("%c", data[i] >= 32 ? data[i] : '.');
74 #define print_packet(...)
77 static void set_state(struct utcp_connection *c, enum state state) {
79 if(state == ESTABLISHED)
80 timerclear(&c->conn_timeout);
81 debug("%p new state: %s\n", c->utcp, strstate[state]);
84 static inline void list_connections(struct utcp *utcp) {
85 debug("%p has %d connections:\n", utcp, utcp->nconnections);
86 for(int i = 0; i < utcp->nconnections; i++)
87 debug(" %u -> %u state %s\n", utcp->connections[i]->src, utcp->connections[i]->dst, strstate[utcp->connections[i]->state]);
90 static int32_t seqdiff(uint32_t a, uint32_t b) {
94 // Connections are stored in a sorted list.
95 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
97 static int compare(const void *va, const void *vb) {
98 const struct utcp_connection *a = *(struct utcp_connection **)va;
99 const struct utcp_connection *b = *(struct utcp_connection **)vb;
100 if(!a->src || !b->src)
102 int c = (int)a->src - (int)b->src;
105 c = (int)a->dst - (int)b->dst;
109 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
110 if(!utcp->nconnections)
112 struct utcp_connection key = {
116 struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
117 return match ? *match : NULL;
120 static void free_connection(struct utcp_connection *c) {
121 struct utcp *utcp = c->utcp;
122 struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
126 int i = cp - utcp->connections;
127 memmove(cp + i, cp + i + 1, (utcp->nconnections - i - 1) * sizeof *cp);
128 utcp->nconnections--;
133 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
134 // Check whether this combination of src and dst is free
137 if(find_connection(utcp, src, dst)) {
141 } else { // If src == 0, generate a random port number with the high bit set
142 if(utcp->nconnections >= 32767) {
146 src = rand() | 0x8000;
147 while(find_connection(utcp, src, dst))
151 // Allocate memory for the new connection
153 if(utcp->nconnections >= utcp->nallocated) {
154 if(!utcp->nallocated)
155 utcp->nallocated = 4;
157 utcp->nallocated *= 2;
158 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof *utcp->connections);
161 utcp->connections = new_array;
164 struct utcp_connection *c = calloc(1, sizeof *c);
168 c->sndbufsize = DEFAULT_SNDBUFSIZE;
169 c->maxsndbufsize = DEFAULT_MAXSNDBUFSIZE;
170 c->sndbuf = malloc(c->sndbufsize);
176 // Fill in the details
181 c->snd.una = c->snd.iss;
182 c->snd.nxt = c->snd.iss + 1;
183 c->rcv.wnd = utcp->mtu;
184 c->snd.last = c->snd.nxt;
185 c->snd.cwnd = utcp->mtu;
188 // Add it to the sorted list of connections
190 utcp->connections[utcp->nconnections++] = c;
191 qsort(utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
196 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
197 struct utcp_connection *c = allocate_connection(utcp, 0, dst);
207 hdr.seq = c->snd.iss;
210 hdr.wnd = c->rcv.wnd;
212 set_state(c, SYN_SENT);
214 print_packet(utcp, "send", &hdr, sizeof hdr);
215 utcp->send(utcp, &hdr, sizeof hdr);
217 gettimeofday(&c->conn_timeout, NULL);
218 c->conn_timeout.tv_sec += utcp->timeout;
223 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
224 if(c->reapable || c->state != SYN_RECEIVED) {
225 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
229 debug("%p accepted, %p %p\n", c, recv, priv);
232 set_state(c, ESTABLISHED);
235 static void ack(struct utcp_connection *c, bool sendatleastone) {
236 int32_t left = seqdiff(c->snd.last, c->snd.nxt);
237 int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
238 char *data = c->sndbuf + seqdiff(c->snd.nxt, c->snd.una);
240 fprintf(stderr, "ack, left=%d, cwndleft=%d, sendatleastone=%d\n", left, cwndleft, sendatleastone);
250 if(!left && !sendatleastone)
255 char data[c->utcp->mtu];
258 pkt.hdr.src = c->src;
259 pkt.hdr.dst = c->dst;
260 pkt.hdr.ack = c->rcv.nxt;
261 pkt.hdr.wnd = c->snd.wnd;
265 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
266 pkt.hdr.seq = c->snd.nxt;
268 memcpy(pkt.data, data, seglen);
270 c->snd.nxt += seglen;
274 if(c->state != ESTABLISHED && !left && seglen) {
286 print_packet(c->utcp, "send", &pkt, sizeof pkt.hdr + seglen);
287 c->utcp->send(c->utcp, &pkt, sizeof pkt.hdr + seglen);
291 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
293 debug("Error: send() called on closed connection %p\n", c);
303 debug("Error: send() called on unconnected connection %p\n", c);
314 debug("Error: send() called on closing connection %p\n", c);
319 // Add data to send buffer
329 uint32_t bufused = seqdiff(c->snd.nxt, c->snd.una);
331 /* Check our send buffer.
332 * - If it's big enough, just put the data in there.
333 * - If not, decide whether to enlarge if possible.
334 * - Cap len so it doesn't overflow our buffer.
337 if(len > c->sndbufsize - bufused && c->sndbufsize < c->maxsndbufsize) {
339 if(c->sndbufsize > c->maxsndbufsize / 2)
340 newbufsize = c->maxsndbufsize;
342 newbufsize = c->sndbufsize * 2;
343 if(bufused + len > newbufsize) {
344 if(bufused + len > c->maxsndbufsize)
345 newbufsize = c->maxsndbufsize;
347 newbufsize = bufused + len;
349 char *newbuf = realloc(c->sndbuf, newbufsize);
352 c->sndbufsize = newbufsize;
356 if(len > c->sndbufsize - bufused)
357 len = c->sndbufsize - bufused;
360 errno == EWOULDBLOCK;
364 memcpy(c->sndbuf + bufused, data, len);
371 static void swap_ports(struct hdr *hdr) {
372 uint16_t tmp = hdr->src;
377 int utcp_recv(struct utcp *utcp, const void *data, size_t len) {
391 print_packet(utcp, "recv", data, len);
393 // Drop packets smaller than the header
396 if(len < sizeof hdr) {
401 // Make a copy from the potentially unaligned data to a struct hdr
403 memcpy(&hdr, data, sizeof hdr);
407 // Drop packets with an unknown CTL flag
409 if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
414 // Try to match the packet to an existing connection
416 struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
418 // Is it for a new connection?
421 // Ignore RST packets
426 // Is it a SYN packet and are we LISTENing?
428 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
429 // If we don't want to accept it, send a RST back
430 if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
435 // Try to allocate memory, otherwise send a RST back
436 c = allocate_connection(utcp, hdr.dst, hdr.src);
442 // Return SYN+ACK, go to SYN_RECEIVED state
443 c->snd.wnd = hdr.wnd;
444 c->rcv.irs = hdr.seq;
445 c->rcv.nxt = c->rcv.irs + 1;
446 set_state(c, SYN_RECEIVED);
450 hdr.ack = c->rcv.irs + 1;
451 hdr.seq = c->snd.iss;
453 print_packet(c->utcp, "send", &hdr, sizeof hdr);
454 utcp->send(utcp, &hdr, sizeof hdr);
456 // No, we don't want your packets, send a RST back
464 debug("%p state %s\n", c->utcp, strstate[c->state]);
466 // In case this is for a CLOSED connection, ignore the packet.
467 // TODO: make it so incoming packets can never match a CLOSED connection.
469 if(c->state == CLOSED)
472 // It is for an existing connection.
474 // 1. Drop invalid packets.
476 // 1a. Drop packets that should not happen in our current state.
493 // 1b. Drop packets with a sequence number not in our receive window.
497 if(c->state == SYN_SENT)
500 // TODO: handle packets overlapping c->rcv.nxt.
502 // Only use this when accepting out-of-order packets.
505 acceptable = hdr.seq == c->rcv.nxt;
507 acceptable = (seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt + c->rcv.wnd) < 0);
510 // We don't accept data when the receive window is zero.
513 // Both start and end of packet must be within the receive window
514 acceptable = (seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt + c->rcv.wnd) < 0)
515 || (seqdiff(hdr.seq + len + 1, c->rcv.nxt) >= 0 && seqdiff(hdr.seq + len - 1, c->rcv.nxt + c->rcv.wnd) < 0);
517 if(c->state != SYN_SENT)
518 acceptable = hdr.seq == c->rcv.nxt;
522 debug("Packet not acceptable, %u <= %u + %zu < %u\n", c->rcv.nxt, hdr.seq, len, c->rcv.nxt + c->rcv.wnd);
523 // Ignore unacceptable RST packets.
526 // Otherwise, send an ACK back in the hope things improve.
530 c->snd.wnd = hdr.wnd; // TODO: move below
532 // 1c. Drop packets with an invalid ACK.
533 // ackno should not roll back, and it should also not be bigger than snd.nxt.
535 if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.nxt) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
536 debug("Packet ack seqno out of range, %u %u %u\n", hdr.ack, c->snd.una, c->snd.nxt);
537 // Ignore unacceptable RST packets.
543 // 2. Handle RST packets
550 // The peer has refused our connection.
551 set_state(c, CLOSED);
552 errno = ECONNREFUSED;
559 // We haven't told the application about this connection yet. Silently delete.
568 // The peer has aborted our connection.
569 set_state(c, CLOSED);
579 // As far as the application is concerned, the connection has already been closed.
580 // If it has called utcp_close() already, we can immediately free this connection.
585 // Otherwise, immediately move to the CLOSED state.
586 set_state(c, CLOSED);
593 // 3. Advance snd.una
595 uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
596 c->snd.una = hdr.ack;
599 debug("%p advanced %u\n", utcp, advanced);
600 // Make room in the send buffer.
601 // TODO: try to avoid memmoving too much. Circular buffer?
602 uint32_t left = seqdiff(c->snd.nxt, hdr.ack);
604 memmove(c->sndbuf, c->sndbuf + advanced, left);
606 c->snd.cwnd += utcp->mtu;
607 if(c->snd.cwnd > c->maxsndbufsize)
608 c->snd.cwnd = c->maxsndbufsize;
609 debug("%p increasing cwnd to %u\n", utcp, c->snd.cwnd);
611 // Check if we have sent a FIN that is now ACKed.
614 if(c->snd.una == c->snd.last)
615 set_state(c, FIN_WAIT_2);
618 if(c->snd.una == c->snd.last) {
619 gettimeofday(&c->conn_timeout, NULL);
620 c->conn_timeout.tv_sec += 60;
621 set_state(c, TIME_WAIT);
631 debug("Triplicate ACK\n");
640 timerclear(&c->conn_timeout); // It should be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
641 if(c->snd.una == c->snd.nxt)
642 timerclear(&c->rtrx_timeout);
645 // 5. Process SYN stuff
650 // This is a SYNACK. It should always have ACKed the SYN.
653 c->rcv.irs = hdr.seq;
654 c->rcv.nxt = hdr.seq;
655 set_state(c, ESTABLISHED);
656 // TODO: notify application of this somehow.
666 // Ehm, no. We should never receive a second SYN.
672 // SYN counts as one sequence number
676 // 6. Process new data
678 if(c->state == SYN_RECEIVED) {
679 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
683 // Are we still LISTENing?
685 utcp->accept(c, c->src);
687 if(c->state != ESTABLISHED) {
688 set_state(c, CLOSED);
698 // This should never happen.
708 // Ehm no, We should never receive more data after a FIN.
717 rxd = c->recv(c, data, len);
721 rxd = len; // Bad application, bad!
729 // 7. Process FIN stuff
735 // This should never happen.
738 set_state(c, CLOSE_WAIT);
741 set_state(c, CLOSING);
744 gettimeofday(&c->conn_timeout, NULL);
745 c->conn_timeout.tv_sec += 60;
746 set_state(c, TIME_WAIT);
752 // Ehm, no. We should never receive a second FIN.
758 // FIN counts as one sequence number
761 // Inform the application that the peer closed the connection.
768 if(!len && !advanced)
771 if(!len && !(hdr.ctl & SYN) && !(hdr.ctl & FIN))
785 hdr.ack = hdr.seq + len;
789 print_packet(utcp, "send", &hdr, sizeof hdr);
790 utcp->send(utcp, &hdr, sizeof hdr);
795 int utcp_shutdown(struct utcp_connection *c, int dir) {
796 debug("%p shutdown %d\n", c->utcp, dir);
803 debug("Error: shutdown() called on closed connection %p\n", c);
815 set_state(c, CLOSED);
820 set_state(c, FIN_WAIT_1);
826 set_state(c, CLOSING);
841 int utcp_close(struct utcp_connection *c) {
842 if(utcp_shutdown(c, SHUT_RDWR))
848 int utcp_abort(struct utcp_connection *c) {
855 debug("Error: abort() called on closed connection %p\n", c);
870 set_state(c, CLOSED);
878 set_state(c, CLOSED);
888 hdr.seq = c->snd.nxt;
893 print_packet(c->utcp, "send", &hdr, sizeof hdr);
894 c->utcp->send(c->utcp, &hdr, sizeof hdr);
898 static void retransmit(struct utcp_connection *c) {
899 if(c->state == CLOSED || c->snd.nxt == c->snd.una)
902 struct utcp *utcp = c->utcp;
906 char data[c->utcp->mtu];
909 pkt.hdr.src = c->src;
910 pkt.hdr.dst = c->dst;
914 // TODO: this should not happen
918 pkt.hdr.seq = c->snd.iss;
920 pkt.hdr.wnd = c->rcv.wnd;
922 print_packet(c->utcp, "rtrx", &pkt, sizeof pkt.hdr);
923 utcp->send(utcp, &pkt, sizeof pkt.hdr);
927 pkt.hdr.seq = c->snd.nxt;
928 pkt.hdr.ack = c->rcv.nxt;
929 pkt.hdr.ctl = SYN | ACK;
930 print_packet(c->utcp, "rtrx", &pkt, sizeof pkt.hdr);
931 utcp->send(utcp, &pkt, sizeof pkt.hdr);
936 pkt.hdr.seq = c->snd.una;
937 pkt.hdr.ack = c->rcv.nxt;
939 uint32_t len = seqdiff(c->snd.nxt, c->snd.una);
940 if(c->state == FIN_WAIT_1)
945 if(c->state == FIN_WAIT_1)
948 memcpy(pkt.data, c->sndbuf, len);
949 print_packet(c->utcp, "rtrx", &pkt, sizeof pkt.hdr + len);
950 utcp->send(utcp, &pkt, sizeof pkt.hdr + len);
960 * One call to this function will loop through all connections,
961 * checking if something needs to be resent or not.
962 * The return value is the time to the next timeout in milliseconds,
963 * or maybe a negative value if the timeout is infinite.
965 int utcp_timeout(struct utcp *utcp) {
967 gettimeofday(&now, NULL);
968 struct timeval next = {now.tv_sec + 3600, now.tv_usec};
970 for(int i = 0; i < utcp->nconnections; i++) {
971 struct utcp_connection *c = utcp->connections[i];
975 if(c->state == CLOSED) {
977 debug("Reaping %p\n", c);
984 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
992 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
996 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
997 next = c->conn_timeout;
999 if(c->snd.nxt != c->snd.una) {
1000 c->rtrx_timeout = now;
1001 c->rtrx_timeout.tv_sec++;
1003 timerclear(&c->rtrx_timeout);
1006 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
1007 next = c->rtrx_timeout;
1010 struct timeval diff;
1011 timersub(&next, &now, &diff);
1014 return diff.tv_sec * 1000 + diff.tv_usec / 1000;
1017 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1018 struct utcp *utcp = calloc(1, sizeof *utcp);
1027 utcp->accept = accept;
1028 utcp->pre_accept = pre_accept;
1037 void utcp_exit(struct utcp *utcp) {
1040 for(int i = 0; i < utcp->nconnections; i++)
1041 free_connection(utcp->connections[i]);
1045 uint16_t utcp_get_mtu(struct utcp *utcp) {
1049 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1050 // TODO: handle overhead of the header
1054 int utcp_get_user_timeout(struct utcp *u) {
1058 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1059 u->timeout = timeout;
1062 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1063 return c->maxsndbufsize;
1066 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1067 c->maxsndbufsize = size;
1068 if(c->maxsndbufsize != size)
1069 c->maxsndbufsize = -1;
1072 bool utcp_get_nodelay(struct utcp_connection *c) {
1076 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1077 c->nodelay = nodelay;
1080 bool utcp_get_keepalive(struct utcp_connection *c) {
1081 return c->keepalive;
1084 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1085 c->keepalive = keepalive;
1088 size_t utcp_get_outq(struct utcp_connection *c) {
1089 return seqdiff(c->snd.nxt, c->snd.una);