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>
34 #include "utcp_priv.h"
49 #define timersub(a, b, r)\
51 (r)->tv_sec = (a)->tv_sec - (b)->tv_sec;\
52 (r)->tv_usec = (a)->tv_usec - (b)->tv_usec;\
54 (r)->tv_sec--, (r)->tv_usec += USEC_PER_SEC;\
58 static inline size_t min(size_t a, size_t b) {
62 static inline size_t max(size_t a, size_t b) {
69 #ifndef UTCP_DEBUG_DATALEN
70 #define UTCP_DEBUG_DATALEN 20
73 static void debug(struct utcp_connection *c, const char *format, ...) {
78 clock_gettime(CLOCK_REALTIME, &tv);
79 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);
82 len += vsnprintf(buf + len, sizeof(buf) - len, format, ap);
85 if(len > 0 && (size_t)len < sizeof(buf)) {
86 fwrite(buf, len, 1, stderr);
90 static void print_packet(struct utcp_connection *c, const char *dir, const void *pkt, size_t len) {
93 if(len < sizeof(hdr)) {
94 debug(c, "%s: short packet (%lu bytes)\n", dir, (unsigned long)len);
98 memcpy(&hdr, pkt, sizeof(hdr));
102 if(len > sizeof(hdr)) {
103 datalen = min(len - sizeof(hdr), UTCP_DEBUG_DATALEN);
109 const uint8_t *data = (uint8_t *)pkt + sizeof(hdr);
110 char str[datalen * 2 + 1];
113 for(uint32_t i = 0; i < datalen; i++) {
114 *p++ = "0123456789ABCDEF"[data[i] >> 4];
115 *p++ = "0123456789ABCDEF"[data[i] & 15];
120 debug(c, "%s: len %lu src %u dst %u seq %u ack %u wnd %u aux %x ctl %s%s%s%s data %s\n",
121 dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux,
122 hdr.ctl & SYN ? "SYN" : "",
123 hdr.ctl & RST ? "RST" : "",
124 hdr.ctl & FIN ? "FIN" : "",
125 hdr.ctl & ACK ? "ACK" : "",
130 static void debug_cwnd(struct utcp_connection *c) {
131 debug(c, "snd.cwnd %u snd.ssthresh %u\n", c->snd.cwnd, ~c->snd.ssthresh ? c->snd.ssthresh : 0);
134 #define debug(...) do {} while(0)
135 #define print_packet(...) do {} while(0)
136 #define debug_cwnd(...) do {} while(0)
139 static void set_state(struct utcp_connection *c, enum state state) {
142 if(state == ESTABLISHED) {
143 timerclear(&c->conn_timeout);
146 debug(c, "state %s\n", strstate[state]);
149 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
150 if(seq != c->snd.last) {
165 static bool is_reliable(struct utcp_connection *c) {
166 return c->flags & UTCP_RELIABLE;
169 static int32_t seqdiff(uint32_t a, uint32_t b) {
174 // TODO: convert to ringbuffers to avoid memmove() operations.
176 // Store data into the buffer
177 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
178 debug(NULL, "buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
180 size_t required = offset + len;
182 if(required > buf->maxsize) {
183 if(offset >= buf->maxsize) {
187 len = buf->maxsize - offset;
188 required = buf->maxsize;
191 if(required > buf->size) {
192 size_t newsize = buf->size;
199 } while(newsize < required);
202 if(newsize > buf->maxsize) {
203 newsize = buf->maxsize;
206 char *newdata = realloc(buf->data, newsize);
216 memcpy(buf->data + offset, data, len);
218 if(required > buf->used) {
219 buf->used = required;
225 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
226 return buffer_put_at(buf, buf->used, data, len);
229 // Get data from the buffer. data can be NULL.
230 static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
231 if(len > buf->used) {
236 memcpy(data, buf->data, len);
239 if(len < buf->used) {
240 memmove(buf->data, buf->data + len, buf->used - len);
247 // Copy data from the buffer without removing it.
248 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
249 if(offset >= buf->used) {
253 if(offset + len > buf->used) {
254 len = buf->used - offset;
257 memcpy(data, buf->data + offset, len);
261 static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
262 memset(buf, 0, sizeof(*buf));
265 buf->data = malloc(len);
273 buf->maxsize = maxlen;
277 static void buffer_exit(struct buffer *buf) {
279 memset(buf, 0, sizeof(*buf));
282 static uint32_t buffer_free(const struct buffer *buf) {
283 return buf->maxsize - buf->used;
286 // Connections are stored in a sorted list.
287 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
289 static int compare(const void *va, const void *vb) {
292 const struct utcp_connection *a = *(struct utcp_connection **)va;
293 const struct utcp_connection *b = *(struct utcp_connection **)vb;
296 assert(a->src && b->src);
298 int c = (int)a->src - (int)b->src;
304 c = (int)a->dst - (int)b->dst;
308 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
309 if(!utcp->nconnections) {
313 struct utcp_connection key = {
317 struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
318 return match ? *match : NULL;
321 static void free_connection(struct utcp_connection *c) {
322 struct utcp *utcp = c->utcp;
323 struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
327 int i = cp - utcp->connections;
328 memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
329 utcp->nconnections--;
331 buffer_exit(&c->rcvbuf);
332 buffer_exit(&c->sndbuf);
336 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
337 // Check whether this combination of src and dst is free
340 if(find_connection(utcp, src, dst)) {
344 } else { // If src == 0, generate a random port number with the high bit set
345 if(utcp->nconnections >= 32767) {
350 src = rand() | 0x8000;
352 while(find_connection(utcp, src, dst)) {
357 // Allocate memory for the new connection
359 if(utcp->nconnections >= utcp->nallocated) {
360 if(!utcp->nallocated) {
361 utcp->nallocated = 4;
363 utcp->nallocated *= 2;
366 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
372 utcp->connections = new_array;
375 struct utcp_connection *c = calloc(1, sizeof(*c));
381 if(!buffer_init(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
386 if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
387 buffer_exit(&c->sndbuf);
392 // Fill in the details
401 c->snd.una = c->snd.iss;
402 c->snd.nxt = c->snd.iss + 1;
403 c->snd.last = c->snd.nxt;
404 c->snd.cwnd = (utcp->mss > 2190 ? 2 : utcp->mss > 1095 ? 3 : 4) * utcp->mss;
405 c->snd.ssthresh = ~0;
409 // Add it to the sorted list of connections
411 utcp->connections[utcp->nconnections++] = c;
412 qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
417 static inline uint32_t absdiff(uint32_t a, uint32_t b) {
425 // Update RTT variables. See RFC 6298.
426 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
428 debug(c, "invalid rtt\n");
432 struct utcp *utcp = c->utcp;
436 utcp->rttvar = rtt / 2;
438 utcp->rttvar = (utcp->rttvar * 3 + absdiff(utcp->srtt, rtt)) / 4;
439 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
442 utcp->rto = utcp->srtt + max(4 * utcp->rttvar, CLOCK_GRANULARITY);
444 if(utcp->rto > MAX_RTO) {
448 debug(c, "rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
451 static void start_retransmit_timer(struct utcp_connection *c) {
452 gettimeofday(&c->rtrx_timeout, NULL);
453 c->rtrx_timeout.tv_usec += c->utcp->rto;
455 while(c->rtrx_timeout.tv_usec >= 1000000) {
456 c->rtrx_timeout.tv_usec -= 1000000;
457 c->rtrx_timeout.tv_sec++;
460 debug(c, "rtrx_timeout %ld.%06lu\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_usec);
463 static void stop_retransmit_timer(struct utcp_connection *c) {
464 timerclear(&c->rtrx_timeout);
465 debug(c, "rtrx_timeout cleared\n");
468 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
469 struct utcp_connection *c = allocate_connection(utcp, 0, dst);
475 assert((flags & ~0x1f) == 0);
486 pkt.hdr.src = c->src;
487 pkt.hdr.dst = c->dst;
488 pkt.hdr.seq = c->snd.iss;
490 pkt.hdr.wnd = c->rcvbuf.maxsize;
492 pkt.hdr.aux = 0x0101;
496 pkt.init[3] = flags & 0x7;
498 set_state(c, SYN_SENT);
500 print_packet(c, "send", &pkt, sizeof(pkt));
501 utcp->send(utcp, &pkt, sizeof(pkt));
503 gettimeofday(&c->conn_timeout, NULL);
504 c->conn_timeout.tv_sec += utcp->timeout;
506 start_retransmit_timer(c);
511 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
512 return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
515 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
516 if(c->reapable || c->state != SYN_RECEIVED) {
517 debug(c, "accept() called on invalid connection in state %s\n", c, strstate[c->state]);
521 debug(c, "accepted %p %p\n", c, recv, priv);
524 set_state(c, ESTABLISHED);
527 static void ack(struct utcp_connection *c, bool sendatleastone) {
528 int32_t left = seqdiff(c->snd.last, c->snd.nxt);
529 int32_t cwndleft = min(c->snd.cwnd, c->snd.wnd) - seqdiff(c->snd.nxt, c->snd.una);
535 } else if(cwndleft < left) {
538 if(!sendatleastone || cwndleft > c->utcp->mss) {
539 left -= left % c->utcp->mss;
543 debug(c, "cwndleft %d left %d\n", cwndleft, left);
545 if(!left && !sendatleastone) {
552 } *pkt = c->utcp->pkt;
554 pkt->hdr.src = c->src;
555 pkt->hdr.dst = c->dst;
556 pkt->hdr.ack = c->rcv.nxt;
557 pkt->hdr.wnd = c->rcvbuf.maxsize;
562 uint32_t seglen = left > c->utcp->mss ? c->utcp->mss : left;
563 pkt->hdr.seq = c->snd.nxt;
565 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
567 c->snd.nxt += seglen;
570 if(seglen && fin_wanted(c, c->snd.nxt)) {
575 if(!c->rtt_start.tv_sec) {
576 // Start RTT measurement
577 gettimeofday(&c->rtt_start, NULL);
578 c->rtt_seq = pkt->hdr.seq + seglen;
579 debug(c, "starting RTT measurement, expecting ack %u\n", c->rtt_seq);
582 print_packet(c, "send", pkt, sizeof(pkt->hdr) + seglen);
583 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
587 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
589 debug(c, "send() called on closed connection\n");
597 debug(c, "send() called on unconnected connection\n");
612 debug(c, "send() called on closed connection\n");
617 // Exit early if we have nothing to send.
628 // Check if we need to be able to buffer all data
630 if(c->flags & UTCP_NO_PARTIAL) {
631 if(len > buffer_free(&c->sndbuf)) {
632 if(len > c->sndbuf.maxsize) {
642 // Add data to send buffer.
644 if(is_reliable(c) || (c->state != SYN_SENT && c->state != SYN_RECEIVED)) {
645 len = buffer_put(&c->sndbuf, data, len);
661 // Don't send anything yet if the connection has not fully established yet
663 if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
669 if(!is_reliable(c)) {
670 c->snd.una = c->snd.nxt = c->snd.last;
671 buffer_get(&c->sndbuf, NULL, c->sndbuf.used);
674 if(is_reliable(c) && !timerisset(&c->rtrx_timeout)) {
675 start_retransmit_timer(c);
678 if(is_reliable(c) && !timerisset(&c->conn_timeout)) {
679 gettimeofday(&c->conn_timeout, NULL);
680 c->conn_timeout.tv_sec += c->utcp->timeout;
686 static void swap_ports(struct hdr *hdr) {
687 uint16_t tmp = hdr->src;
692 static void fast_retransmit(struct utcp_connection *c) {
693 if(c->state == CLOSED || c->snd.last == c->snd.una) {
694 debug(c, "fast_retransmit() called but nothing to retransmit!\n");
698 struct utcp *utcp = c->utcp;
705 pkt = malloc(c->utcp->mtu);
711 pkt->hdr.src = c->src;
712 pkt->hdr.dst = c->dst;
713 pkt->hdr.wnd = c->rcvbuf.maxsize;
722 // Send unacked data again.
723 pkt->hdr.seq = c->snd.una;
724 pkt->hdr.ack = c->rcv.nxt;
726 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
728 if(fin_wanted(c, c->snd.una + len)) {
733 buffer_copy(&c->sndbuf, pkt->data, 0, len);
734 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
735 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
745 static void retransmit(struct utcp_connection *c) {
746 if(c->state == CLOSED || c->snd.last == c->snd.una) {
747 debug(c, "retransmit() called but nothing to retransmit!\n");
748 stop_retransmit_timer(c);
752 struct utcp *utcp = c->utcp;
757 } *pkt = c->utcp->pkt;
759 pkt->hdr.src = c->src;
760 pkt->hdr.dst = c->dst;
761 pkt->hdr.wnd = c->rcvbuf.maxsize;
766 // Send our SYN again
767 pkt->hdr.seq = c->snd.iss;
770 pkt->hdr.aux = 0x0101;
774 pkt->data[3] = c->flags & 0x7;
775 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + 4);
776 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
781 pkt->hdr.seq = c->snd.nxt;
782 pkt->hdr.ack = c->rcv.nxt;
783 pkt->hdr.ctl = SYN | ACK;
784 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr));
785 utcp->send(utcp, pkt, sizeof(pkt->hdr));
793 // Send unacked data again.
794 pkt->hdr.seq = c->snd.una;
795 pkt->hdr.ack = c->rcv.nxt;
797 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
799 if(fin_wanted(c, c->snd.una + len)) {
804 // RFC 5681 slow start after timeout
805 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
806 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
807 c->snd.cwnd = utcp->mss;
810 buffer_copy(&c->sndbuf, pkt->data, 0, len);
811 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
812 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
814 c->snd.nxt = c->snd.una + len;
821 // We shouldn't need to retransmit anything in this state.
825 stop_retransmit_timer(c);
829 start_retransmit_timer(c);
832 if(utcp->rto > MAX_RTO) {
836 c->rtt_start.tv_sec = 0; // invalidate RTT timer
837 c->dupack = 0; // cancel any ongoing fast recovery
843 /* Update receive buffer and SACK entries after consuming data.
847 * |.....0000..1111111111.....22222......3333|
850 * 0..3 represent the SACK entries. The ^ indicates up to which point we want
851 * to remove data from the receive buffer. The idea is to substract "len"
852 * from the offset of all the SACK entries, and then remove/cut down entries
853 * that are shifted to before the start of the receive buffer.
855 * There are three cases:
856 * - the SACK entry is after ^, in that case just change the offset.
857 * - the SACK entry starts before and ends after ^, so we have to
858 * change both its offset and size.
859 * - the SACK entry is completely before ^, in that case delete it.
861 static void sack_consume(struct utcp_connection *c, size_t len) {
862 debug(c, "sack_consume %lu\n", (unsigned long)len);
864 if(len > c->rcvbuf.used) {
865 debug(c, "all SACK entries consumed\n");
870 buffer_get(&c->rcvbuf, NULL, len);
872 for(int i = 0; i < NSACKS && c->sacks[i].len;) {
873 if(len < c->sacks[i].offset) {
874 c->sacks[i].offset -= len;
876 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
877 c->sacks[i].len -= len - c->sacks[i].offset;
878 c->sacks[i].offset = 0;
882 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
883 c->sacks[NSACKS - 1].len = 0;
891 for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
892 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
896 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
897 debug(c, "out of order packet, offset %u\n", offset);
898 // Packet loss or reordering occured. Store the data in the buffer.
899 ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
901 if(rxd < 0 || (size_t)rxd < len) {
905 // Make note of where we put it.
906 for(int i = 0; i < NSACKS; i++) {
907 if(!c->sacks[i].len) { // nothing to merge, add new entry
908 debug(c, "new SACK entry %d\n", i);
909 c->sacks[i].offset = offset;
910 c->sacks[i].len = rxd;
912 } else if(offset < c->sacks[i].offset) {
913 if(offset + rxd < c->sacks[i].offset) { // insert before
914 if(!c->sacks[NSACKS - 1].len) { // only if room left
915 debug(c, "insert SACK entry at %d\n", i);
916 memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
917 c->sacks[i].offset = offset;
918 c->sacks[i].len = rxd;
920 debug(c, "SACK entries full, dropping packet\n");
925 debug(c, "merge with start of SACK entry at %d\n", i);
926 c->sacks[i].offset = offset;
929 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
930 if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
931 debug(c, "merge with end of SACK entry at %d\n", i);
932 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
933 // TODO: handle potential merge with next entry
940 for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
941 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
945 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
946 // Check if we can process out-of-order data now.
947 if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
948 debug(c, "incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
949 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
950 len = max(len, c->sacks[0].offset + c->sacks[0].len);
951 data = c->rcvbuf.data;
955 ssize_t rxd = c->recv(c, data, len);
957 if(rxd < 0 || (size_t)rxd != len) {
958 // TODO: handle the application not accepting all data.
964 sack_consume(c, len);
971 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
972 if(!is_reliable(c)) {
973 c->recv(c, data, len);
974 c->rcv.nxt = seq + len;
978 uint32_t offset = seqdiff(seq, c->rcv.nxt);
980 if(offset + len > c->rcvbuf.maxsize) {
985 handle_out_of_order(c, offset, data, len);
987 handle_in_order(c, data, len);
992 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
993 const uint8_t *ptr = data;
1009 // Drop packets smaller than the header
1013 if(len < sizeof(hdr)) {
1014 print_packet(NULL, "recv", data, len);
1019 // Make a copy from the potentially unaligned data to a struct hdr
1021 memcpy(&hdr, ptr, sizeof(hdr));
1023 // Try to match the packet to an existing connection
1025 struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1026 print_packet(c, "recv", data, len);
1028 // Process the header
1033 // Drop packets with an unknown CTL flag
1035 if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
1036 print_packet(NULL, "recv", data, len);
1041 // Check for auxiliary headers
1043 const uint8_t *init = NULL;
1045 uint16_t aux = hdr.aux;
1048 size_t auxlen = 4 * (aux >> 8) & 0xf;
1049 uint8_t auxtype = aux & 0xff;
1058 if(!(hdr.ctl & SYN) || auxlen != 4) {
1074 if(!(aux & 0x800)) {
1083 memcpy(&aux, ptr, 2);
1088 bool has_data = len || (hdr.ctl & (SYN | FIN));
1090 // Is it for a new connection?
1093 // Ignore RST packets
1099 // Is it a SYN packet and are we LISTENing?
1101 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1102 // If we don't want to accept it, send a RST back
1103 if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
1108 // Try to allocate memory, otherwise send a RST back
1109 c = allocate_connection(utcp, hdr.dst, hdr.src);
1116 // Parse auxilliary information
1123 c->flags = init[3] & 0x7;
1125 c->flags = UTCP_TCP;
1129 // Return SYN+ACK, go to SYN_RECEIVED state
1130 c->snd.wnd = hdr.wnd;
1131 c->rcv.irs = hdr.seq;
1132 c->rcv.nxt = c->rcv.irs + 1;
1133 set_state(c, SYN_RECEIVED);
1140 pkt.hdr.src = c->src;
1141 pkt.hdr.dst = c->dst;
1142 pkt.hdr.ack = c->rcv.irs + 1;
1143 pkt.hdr.seq = c->snd.iss;
1144 pkt.hdr.wnd = c->rcvbuf.maxsize;
1145 pkt.hdr.ctl = SYN | ACK;
1148 pkt.hdr.aux = 0x0101;
1152 pkt.data[3] = c->flags & 0x7;
1153 print_packet(c, "send", &pkt, sizeof(hdr) + 4);
1154 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1157 print_packet(c, "send", &pkt, sizeof(hdr));
1158 utcp->send(utcp, &pkt, sizeof(hdr));
1161 // No, we don't want your packets, send a RST back
1169 debug(c, "state %s\n", strstate[c->state]);
1171 // In case this is for a CLOSED connection, ignore the packet.
1172 // TODO: make it so incoming packets can never match a CLOSED connection.
1174 if(c->state == CLOSED) {
1175 debug(c, "got packet for closed connection\n");
1179 // It is for an existing connection.
1181 // 1. Drop invalid packets.
1183 // 1a. Drop packets that should not happen in our current state.
1204 // 1b. Discard data that is not in our receive window.
1206 if(is_reliable(c)) {
1209 if(c->state == SYN_SENT) {
1211 } else if(len == 0) {
1212 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1214 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1216 // cut already accepted front overlapping
1217 if(rcv_offset < 0) {
1218 acceptable = len > (size_t) - rcv_offset;
1223 hdr.seq -= rcv_offset;
1226 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1231 debug(c, "packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1233 // Ignore unacceptable RST packets.
1238 // Otherwise, continue processing.
1243 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1246 debug(c, "packet out of order, offset %u bytes", rcv_offset);
1249 if(rcv_offset >= 0) {
1250 c->rcv.nxt = hdr.seq + len;
1256 c->snd.wnd = hdr.wnd; // TODO: move below
1258 // 1c. Drop packets with an invalid ACK.
1259 // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1260 // (= snd.una + c->sndbuf.used).
1262 if(!is_reliable(c)) {
1263 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1264 hdr.ack = c->snd.una;
1268 if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1269 debug(c, "packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1271 // Ignore unacceptable RST packets.
1279 // 2. Handle RST packets
1284 if(!(hdr.ctl & ACK)) {
1288 // The peer has refused our connection.
1289 set_state(c, CLOSED);
1290 errno = ECONNREFUSED;
1293 c->recv(c, NULL, 0);
1296 if(c->poll && !c->reapable) {
1307 // We haven't told the application about this connection yet. Silently delete.
1319 // The peer has aborted our connection.
1320 set_state(c, CLOSED);
1324 c->recv(c, NULL, 0);
1327 if(c->poll && !c->reapable) {
1340 // As far as the application is concerned, the connection has already been closed.
1341 // If it has called utcp_close() already, we can immediately free this connection.
1347 // Otherwise, immediately move to the CLOSED state.
1348 set_state(c, CLOSED);
1361 if(!(hdr.ctl & ACK)) {
1366 // 3. Advance snd.una
1368 advanced = seqdiff(hdr.ack, c->snd.una);
1372 if(c->rtt_start.tv_sec) {
1373 if(c->rtt_seq == hdr.ack) {
1374 struct timeval now, diff;
1375 gettimeofday(&now, NULL);
1376 timersub(&now, &c->rtt_start, &diff);
1377 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
1378 c->rtt_start.tv_sec = 0;
1379 } else if(c->rtt_seq < hdr.ack) {
1380 debug(c, "cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1381 c->rtt_start.tv_sec = 0;
1385 int32_t data_acked = advanced;
1393 // TODO: handle FIN as well.
1398 assert(data_acked >= 0);
1401 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1402 assert(data_acked <= bufused);
1406 buffer_get(&c->sndbuf, NULL, data_acked);
1409 // Also advance snd.nxt if possible
1410 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1411 c->snd.nxt = hdr.ack;
1414 c->snd.una = hdr.ack;
1417 if(c->dupack >= 3) {
1418 debug(c, "fast recovery ended\n");
1419 c->snd.cwnd = c->snd.ssthresh;
1425 // Increase the congestion window according to RFC 5681
1426 if(c->snd.cwnd < c->snd.ssthresh) {
1427 c->snd.cwnd += min(advanced, utcp->mss); // eq. 2
1429 c->snd.cwnd += max(1, (utcp->mss * utcp->mss) / c->snd.cwnd); // eq. 3
1432 if(c->snd.cwnd > c->sndbuf.maxsize) {
1433 c->snd.cwnd = c->sndbuf.maxsize;
1438 // Check if we have sent a FIN that is now ACKed.
1441 if(c->snd.una == c->snd.last) {
1442 set_state(c, FIN_WAIT_2);
1448 if(c->snd.una == c->snd.last) {
1449 gettimeofday(&c->conn_timeout, NULL);
1450 c->conn_timeout.tv_sec += utcp->timeout;
1451 set_state(c, TIME_WAIT);
1460 if(!len && is_reliable(c) && c->snd.una != c->snd.last) {
1462 debug(c, "duplicate ACK %d\n", c->dupack);
1464 if(c->dupack == 3) {
1465 // RFC 5681 fast recovery
1466 debug(c, "fast recovery started\n", c->dupack);
1467 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
1468 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
1469 c->snd.cwnd = min(c->snd.ssthresh + 3 * utcp->mss, c->sndbuf.maxsize);
1471 if(c->snd.cwnd > c->sndbuf.maxsize) {
1472 c->snd.cwnd = c->sndbuf.maxsize;
1478 } else if(c->dupack > 3) {
1479 c->snd.cwnd += utcp->mss;
1481 if(c->snd.cwnd > c->sndbuf.maxsize) {
1482 c->snd.cwnd = c->sndbuf.maxsize;
1488 // We got an ACK which indicates the other side did get one of our packets.
1489 // Reset the retransmission timer to avoid going to slow start,
1490 // but don't touch the connection timeout.
1491 start_retransmit_timer(c);
1498 if(c->snd.una == c->snd.last) {
1499 stop_retransmit_timer(c);
1500 timerclear(&c->conn_timeout);
1501 } else if(is_reliable(c)) {
1502 start_retransmit_timer(c);
1503 gettimeofday(&c->conn_timeout, NULL);
1504 c->conn_timeout.tv_sec += utcp->timeout;
1509 // 5. Process SYN stuff
1515 // This is a SYNACK. It should always have ACKed the SYN.
1520 c->rcv.irs = hdr.seq;
1521 c->rcv.nxt = hdr.seq;
1525 set_state(c, FIN_WAIT_1);
1527 set_state(c, ESTABLISHED);
1530 // TODO: notify application of this somehow.
1534 // This is a retransmit of a SYN, send back the SYNACK.
1544 // Ehm, no. We should never receive a second SYN.
1554 // SYN counts as one sequence number
1558 // 6. Process new data
1560 if(c->state == SYN_RECEIVED) {
1561 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1566 // Are we still LISTENing?
1568 utcp->accept(c, c->src);
1571 if(c->state != ESTABLISHED) {
1572 set_state(c, CLOSED);
1582 // This should never happen.
1597 // Ehm no, We should never receive more data after a FIN.
1607 handle_incoming_data(c, hdr.seq, ptr, len);
1610 // 7. Process FIN stuff
1612 if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1616 // This should never happen.
1623 set_state(c, CLOSE_WAIT);
1627 set_state(c, CLOSING);
1631 gettimeofday(&c->conn_timeout, NULL);
1632 c->conn_timeout.tv_sec += utcp->timeout;
1633 set_state(c, TIME_WAIT);
1640 // Ehm, no. We should never receive a second FIN.
1650 // FIN counts as one sequence number
1654 // Inform the application that the peer closed its end of the connection.
1657 c->recv(c, NULL, 0);
1661 // Now we send something back if:
1662 // - we received data, so we have to send back an ACK
1663 // -> sendatleastone = true
1664 // - or we got an ack, so we should maybe send a bit more data
1665 // -> sendatleastone = false
1667 if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1682 hdr.ack = hdr.seq + len;
1684 hdr.ctl = RST | ACK;
1687 print_packet(c, "send", &hdr, sizeof(hdr));
1688 utcp->send(utcp, &hdr, sizeof(hdr));
1693 int utcp_shutdown(struct utcp_connection *c, int dir) {
1694 debug(c, "shutdown %d at %u\n", dir, c ? c->snd.last : 0);
1702 debug(c, "shutdown() called on closed connection\n");
1707 if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1712 // TCP does not have a provision for stopping incoming packets.
1713 // The best we can do is to just ignore them.
1714 if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1718 // The rest of the code deals with shutting down writes.
1719 if(dir == UTCP_SHUT_RD) {
1723 // Only process shutting down writes once.
1741 set_state(c, FIN_WAIT_1);
1749 set_state(c, CLOSING);
1762 if(!timerisset(&c->rtrx_timeout)) {
1763 start_retransmit_timer(c);
1769 static bool reset_connection(struct utcp_connection *c) {
1776 debug(c, "abort() called on closed connection\n");
1793 set_state(c, CLOSED);
1801 set_state(c, CLOSED);
1811 hdr.seq = c->snd.nxt;
1816 print_packet(c, "send", &hdr, sizeof(hdr));
1817 c->utcp->send(c->utcp, &hdr, sizeof(hdr));
1821 // Closes all the opened connections
1822 void utcp_abort_all_connections(struct utcp *utcp) {
1828 for(int i = 0; i < utcp->nconnections; i++) {
1829 struct utcp_connection *c = utcp->connections[i];
1831 if(c->reapable || c->state == CLOSED) {
1835 utcp_recv_t old_recv = c->recv;
1836 utcp_poll_t old_poll = c->poll;
1838 reset_connection(c);
1842 old_recv(c, NULL, 0);
1845 if(old_poll && !c->reapable) {
1854 int utcp_close(struct utcp_connection *c) {
1855 if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
1865 int utcp_abort(struct utcp_connection *c) {
1866 if(!reset_connection(c)) {
1875 * One call to this function will loop through all connections,
1876 * checking if something needs to be resent or not.
1877 * The return value is the time to the next timeout in milliseconds,
1878 * or maybe a negative value if the timeout is infinite.
1880 struct timeval utcp_timeout(struct utcp *utcp) {
1882 gettimeofday(&now, NULL);
1883 struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1885 for(int i = 0; i < utcp->nconnections; i++) {
1886 struct utcp_connection *c = utcp->connections[i];
1892 // delete connections that have been utcp_close()d.
1893 if(c->state == CLOSED) {
1895 debug(c, "reaping\n");
1903 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1908 c->recv(c, NULL, 0);
1911 if(c->poll && !c->reapable) {
1918 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1919 debug(c, "retransmitting after timeout\n");
1924 if((c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1925 uint32_t len = buffer_free(&c->sndbuf);
1930 } else if(c->state == CLOSED) {
1935 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <)) {
1936 next = c->conn_timeout;
1939 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <)) {
1940 next = c->rtrx_timeout;
1944 struct timeval diff;
1946 timersub(&next, &now, &diff);
1951 bool utcp_is_active(struct utcp *utcp) {
1956 for(int i = 0; i < utcp->nconnections; i++)
1957 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
1964 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1970 struct utcp *utcp = calloc(1, sizeof(*utcp));
1976 utcp->accept = accept;
1977 utcp->pre_accept = pre_accept;
1980 utcp_set_mtu(utcp, DEFAULT_MTU);
1981 utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1982 utcp->rto = START_RTO; // usec
1987 void utcp_exit(struct utcp *utcp) {
1992 for(int i = 0; i < utcp->nconnections; i++) {
1993 struct utcp_connection *c = utcp->connections[i];
1997 c->recv(c, NULL, 0);
2000 if(c->poll && !c->reapable) {
2005 buffer_exit(&c->rcvbuf);
2006 buffer_exit(&c->sndbuf);
2010 free(utcp->connections);
2014 uint16_t utcp_get_mtu(struct utcp *utcp) {
2015 return utcp ? utcp->mtu : 0;
2018 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
2023 if (mtu <= sizeof(struct hdr)) {
2027 if (mtu > utcp->mtu) {
2028 char *new = realloc(utcp->pkt, mtu);
2035 utcp->mss = mtu - sizeof(struct hdr);
2038 void utcp_reset_timers(struct utcp *utcp) {
2043 struct timeval now, then;
2045 gettimeofday(&now, NULL);
2049 then.tv_sec += utcp->timeout;
2051 for(int i = 0; i < utcp->nconnections; i++) {
2052 struct utcp_connection *c = utcp->connections[i];
2058 if(timerisset(&c->rtrx_timeout)) {
2059 c->rtrx_timeout = now;
2062 if(timerisset(&c->conn_timeout)) {
2063 c->conn_timeout = then;
2066 c->rtt_start.tv_sec = 0;
2069 if(utcp->rto > START_RTO) {
2070 utcp->rto = START_RTO;
2074 int utcp_get_user_timeout(struct utcp *u) {
2075 return u ? u->timeout : 0;
2078 void utcp_set_user_timeout(struct utcp *u, int timeout) {
2080 u->timeout = timeout;
2084 size_t utcp_get_sndbuf(struct utcp_connection *c) {
2085 return c ? c->sndbuf.maxsize : 0;
2088 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
2098 return buffer_free(&c->sndbuf);
2105 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
2110 c->sndbuf.maxsize = size;
2112 if(c->sndbuf.maxsize != size) {
2113 c->sndbuf.maxsize = -1;
2117 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
2118 return c ? c->rcvbuf.maxsize : 0;
2121 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
2122 if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
2123 return buffer_free(&c->rcvbuf);
2129 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
2134 c->rcvbuf.maxsize = size;
2136 if(c->rcvbuf.maxsize != size) {
2137 c->rcvbuf.maxsize = -1;
2141 size_t utcp_get_sendq(struct utcp_connection *c) {
2142 return c->sndbuf.used;
2145 size_t utcp_get_recvq(struct utcp_connection *c) {
2146 return c->rcvbuf.used;
2149 bool utcp_get_nodelay(struct utcp_connection *c) {
2150 return c ? c->nodelay : false;
2153 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2155 c->nodelay = nodelay;
2159 bool utcp_get_keepalive(struct utcp_connection *c) {
2160 return c ? c->keepalive : false;
2163 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2165 c->keepalive = keepalive;
2169 size_t utcp_get_outq(struct utcp_connection *c) {
2170 return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2173 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2179 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2185 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
2187 utcp->accept = accept;
2188 utcp->pre_accept = pre_accept;
2192 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2193 if(!c || c->reapable) {
2197 if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2202 // If we expect data, start the connection timer.
2203 if(!timerisset(&c->conn_timeout)) {
2204 gettimeofday(&c->conn_timeout, NULL);
2205 c->conn_timeout.tv_sec += c->utcp->timeout;
2208 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2209 if(c->snd.una == c->snd.last) {
2210 timerclear(&c->conn_timeout);
2215 void utcp_offline(struct utcp *utcp, bool offline) {
2217 gettimeofday(&now, NULL);
2219 for(int i = 0; i < utcp->nconnections; i++) {
2220 struct utcp_connection *c = utcp->connections[i];
2226 utcp_expect_data(c, offline);
2229 if(timerisset(&c->rtrx_timeout)) {
2230 c->rtrx_timeout = now;
2233 utcp->connections[i]->rtt_start.tv_sec = 0;
2237 if(!offline && utcp->rto > START_RTO) {
2238 utcp->rto = START_RTO;