2 net_packet.c -- Handles in- and outgoing VPN packets
3 Copyright (C) 1998-2005 Ivo Timmermans,
4 2000-2014 Guus Sliepen <guus@tinc-vpn.org>
5 2010 Timothy Redaelli <timothy@redaelli.eu>
6 2010 Brandon Black <blblack@gmail.com>
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2 of the License, or
11 (at your option) any later version.
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License along
19 with this program; if not, write to the Free Software Foundation, Inc.,
20 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
35 #include "connection.h"
52 #define MAX(a, b) ((a) > (b) ? (a) : (b))
55 /* The minimum size of a probe is 14 bytes, but since we normally use CBC mode
56 encryption, we can add a few extra random bytes without increasing the
57 resulting packet size. */
58 #define MIN_PROBE_SIZE 18
62 static char lzo_wrkmem[LZO1X_999_MEM_COMPRESS > LZO1X_1_MEM_COMPRESS ? LZO1X_999_MEM_COMPRESS : LZO1X_1_MEM_COMPRESS];
65 static void send_udppacket(node_t *, vpn_packet_t *);
67 unsigned replaywin = 32;
68 bool localdiscovery = true;
69 bool udp_discovery = true;
70 int udp_discovery_keepalive_interval = 10;
71 int udp_discovery_interval = 2;
72 int udp_discovery_timeout = 30;
74 #define MAX_SEQNO 1073741824
76 static void try_fix_mtu(node_t *n) {
80 if(n->mtuprobes == 20 || n->minmtu >= n->maxmtu) {
81 if(n->minmtu > n->maxmtu)
82 n->minmtu = n->maxmtu;
84 n->maxmtu = n->minmtu;
86 logger(DEBUG_TRAFFIC, LOG_INFO, "Fixing MTU of %s (%s) to %d after %d probes", n->name, n->hostname, n->mtu, n->mtuprobes);
91 static void udp_probe_timeout_handler(void *data) {
93 if(!n->status.udp_confirmed)
96 logger(DEBUG_TRAFFIC, LOG_INFO, "Too much time has elapsed since last UDP ping response from %s (%s), stopping UDP communication", n->name, n->hostname);
97 n->status.udp_confirmed = false;
104 static void send_udp_probe_reply(node_t *n, vpn_packet_t *packet, length_t len) {
105 if(!n->status.sptps && !n->status.validkey) {
106 logger(DEBUG_TRAFFIC, LOG_INFO, "Trying to send UDP probe reply to %s (%s) but we don't have his key yet", n->name, n->hostname);
110 /* Type 2 probe replies were introduced in protocol 17.3 */
111 if ((n->options >> 24) >= 3) {
113 uint16_t len16 = htons(len);
114 memcpy(DATA(packet) + 1, &len16, 2);
115 packet->len = MIN_PROBE_SIZE;
116 logger(DEBUG_TRAFFIC, LOG_INFO, "Sending type 2 probe reply length %u to %s (%s)", len, n->name, n->hostname);
119 /* Legacy protocol: n won't understand type 2 probe replies. */
121 logger(DEBUG_TRAFFIC, LOG_INFO, "Sending type 1 probe reply length %u to %s (%s)", len, n->name, n->hostname);
124 /* Temporarily set udp_confirmed, so that the reply is sent
125 back exactly the way it came in. */
127 bool udp_confirmed = n->status.udp_confirmed;
128 n->status.udp_confirmed = true;
129 send_udppacket(n, packet);
130 n->status.udp_confirmed = udp_confirmed;
133 static void udp_probe_h(node_t *n, vpn_packet_t *packet, length_t len) {
134 if(!DATA(packet)[0]) {
135 logger(DEBUG_TRAFFIC, LOG_INFO, "Got UDP probe request %d from %s (%s)", packet->len, n->name, n->hostname);
136 return send_udp_probe_reply(n, packet, len);
139 if (DATA(packet)[0] == 2) {
140 // It's a type 2 probe reply, use the length field inside the packet
142 memcpy(&len16, DATA(packet) + 1, 2);
146 logger(DEBUG_TRAFFIC, LOG_INFO, "Got type %d UDP probe reply %d from %s (%s)", DATA(packet)[0], len, n->name, n->hostname);
148 /* It's a valid reply: now we know bidirectional communication
149 is possible using the address and socket that the reply
151 n->status.udp_confirmed = true;
153 // Reset the UDP ping timer.
154 n->udp_ping_sent = now;
157 timeout_del(&n->udp_ping_timeout);
158 timeout_add(&n->udp_ping_timeout, &udp_probe_timeout_handler, n, &(struct timeval){udp_discovery_timeout, 0});
161 if(len > n->maxmtu) {
162 logger(DEBUG_TRAFFIC, LOG_INFO, "Increase in PMTU to %s (%s) detected, restarting PMTU discovery", n->name, n->hostname);
165 /* Set mtuprobes to 1 so that try_mtu() doesn't reset maxmtu */
168 } else if(n->mtuprobes < 0 && len == n->maxmtu) {
169 /* We got a maxmtu sized packet, confirming the PMTU is still valid. */
171 n->mtu_ping_sent = now;
174 /* If applicable, raise the minimum supported MTU */
176 if(n->minmtu < len) {
182 static length_t compress_packet(uint8_t *dest, const uint8_t *source, length_t len, int level) {
184 memcpy(dest, source, len);
186 } else if(level == 10) {
188 lzo_uint lzolen = MAXSIZE;
189 lzo1x_1_compress(source, len, dest, &lzolen, lzo_wrkmem);
194 } else if(level < 10) {
196 unsigned long destlen = MAXSIZE;
197 if(compress2(dest, &destlen, source, len, level) == Z_OK)
204 lzo_uint lzolen = MAXSIZE;
205 lzo1x_999_compress(source, len, dest, &lzolen, lzo_wrkmem);
215 static length_t uncompress_packet(uint8_t *dest, const uint8_t *source, length_t len, int level) {
217 memcpy(dest, source, len);
219 } else if(level > 9) {
221 lzo_uint lzolen = MAXSIZE;
222 if(lzo1x_decompress_safe(source, len, dest, &lzolen, NULL) == LZO_E_OK)
230 unsigned long destlen = MAXSIZE;
231 if(uncompress(dest, &destlen, source, len) == Z_OK)
243 static void receive_packet(node_t *n, vpn_packet_t *packet) {
244 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Received packet of %d bytes from %s (%s)",
245 packet->len, n->name, n->hostname);
248 n->in_bytes += packet->len;
253 static bool try_mac(node_t *n, const vpn_packet_t *inpkt) {
255 return sptps_verify_datagram(&n->sptps, DATA(inpkt), inpkt->len);
257 #ifdef DISABLE_LEGACY
260 if(!n->status.validkey_in || !digest_active(n->indigest) || inpkt->len < sizeof(seqno_t) + digest_length(n->indigest))
263 return digest_verify(n->indigest, inpkt->data, inpkt->len - digest_length(n->indigest), inpkt->data + inpkt->len - digest_length(n->indigest));
267 static bool receive_udppacket(node_t *n, vpn_packet_t *inpkt) {
268 vpn_packet_t pkt1, pkt2;
269 vpn_packet_t *pkt[] = { &pkt1, &pkt2, &pkt1, &pkt2 };
272 pkt1.offset = DEFAULT_PACKET_OFFSET;
273 pkt2.offset = DEFAULT_PACKET_OFFSET;
275 if(n->status.sptps) {
276 if(!n->sptps.state) {
277 if(!n->status.waitingforkey) {
278 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got packet from %s (%s) but we haven't exchanged keys yet", n->name, n->hostname);
281 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got packet from %s (%s) but he hasn't got our key yet", n->name, n->hostname);
285 inpkt->offset += 2 * sizeof(node_id_t);
286 n->status.udppacket = true;
287 bool result = sptps_receive_data(&n->sptps, DATA(inpkt), inpkt->len - 2 * sizeof(node_id_t));
288 n->status.udppacket = false;
291 logger(DEBUG_TRAFFIC, LOG_ERR, "Got bad packet from %s (%s)", n->name, n->hostname);
297 #ifdef DISABLE_LEGACY
300 if(!n->status.validkey_in) {
301 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got packet from %s (%s) but he hasn't got our key yet", n->name, n->hostname);
305 /* Check packet length */
307 if(inpkt->len < sizeof(seqno_t) + digest_length(n->indigest)) {
308 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got too short packet from %s (%s)",
309 n->name, n->hostname);
313 /* It's a legacy UDP packet, the data starts after the seqno */
315 inpkt->offset += sizeof(seqno_t);
317 /* Check the message authentication code */
319 if(digest_active(n->indigest)) {
320 inpkt->len -= digest_length(n->indigest);
321 if(!digest_verify(n->indigest, SEQNO(inpkt), inpkt->len, SEQNO(inpkt) + inpkt->len)) {
322 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got unauthenticated packet from %s (%s)", n->name, n->hostname);
326 /* Decrypt the packet */
328 if(cipher_active(n->incipher)) {
329 vpn_packet_t *outpkt = pkt[nextpkt++];
332 if(!cipher_decrypt(n->incipher, SEQNO(inpkt), inpkt->len, SEQNO(outpkt), &outlen, true)) {
333 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Error decrypting packet from %s (%s)", n->name, n->hostname);
337 outpkt->len = outlen;
341 /* Check the sequence number */
344 memcpy(&seqno, SEQNO(inpkt), sizeof seqno);
345 seqno = ntohl(seqno);
346 inpkt->len -= sizeof seqno;
349 if(seqno != n->received_seqno + 1) {
350 if(seqno >= n->received_seqno + replaywin * 8) {
351 if(n->farfuture++ < replaywin >> 2) {
352 logger(DEBUG_ALWAYS, LOG_WARNING, "Packet from %s (%s) is %d seqs in the future, dropped (%u)",
353 n->name, n->hostname, seqno - n->received_seqno - 1, n->farfuture);
356 logger(DEBUG_ALWAYS, LOG_WARNING, "Lost %d packets from %s (%s)",
357 seqno - n->received_seqno - 1, n->name, n->hostname);
358 memset(n->late, 0, replaywin);
359 } else if (seqno <= n->received_seqno) {
360 if((n->received_seqno >= replaywin * 8 && seqno <= n->received_seqno - replaywin * 8) || !(n->late[(seqno / 8) % replaywin] & (1 << seqno % 8))) {
361 logger(DEBUG_ALWAYS, LOG_WARNING, "Got late or replayed packet from %s (%s), seqno %d, last received %d",
362 n->name, n->hostname, seqno, n->received_seqno);
366 for(int i = n->received_seqno + 1; i < seqno; i++)
367 n->late[(i / 8) % replaywin] |= 1 << i % 8;
372 n->late[(seqno / 8) % replaywin] &= ~(1 << seqno % 8);
375 if(seqno > n->received_seqno)
376 n->received_seqno = seqno;
380 if(n->received_seqno > MAX_SEQNO)
383 /* Decompress the packet */
385 length_t origlen = inpkt->len;
387 if(n->incompression) {
388 vpn_packet_t *outpkt = pkt[nextpkt++];
390 if((outpkt->len = uncompress_packet(DATA(outpkt), DATA(inpkt), inpkt->len, n->incompression)) < 0) {
391 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while uncompressing packet from %s (%s)",
392 n->name, n->hostname);
398 origlen -= MTU/64 + 20;
401 if(inpkt->len > n->maxrecentlen)
402 n->maxrecentlen = inpkt->len;
406 if(!DATA(inpkt)[12] && !DATA(inpkt)[13])
407 udp_probe_h(n, inpkt, origlen);
409 receive_packet(n, inpkt);
414 void receive_tcppacket(connection_t *c, const char *buffer, int len) {
416 outpkt.offset = DEFAULT_PACKET_OFFSET;
418 if(len > sizeof outpkt.data - outpkt.offset)
422 if(c->options & OPTION_TCPONLY)
425 outpkt.priority = -1;
426 memcpy(DATA(&outpkt), buffer, len);
428 receive_packet(c->node, &outpkt);
431 bool receive_tcppacket_sptps(connection_t *c, const char *data, int len) {
432 if (len < sizeof(node_id_t) + sizeof(node_id_t)) {
433 logger(DEBUG_ALWAYS, LOG_ERR, "Got too short TCP SPTPS packet from %s (%s)", c->name, c->hostname);
437 node_t *to = lookup_node_id((node_id_t *)data);
438 data += sizeof(node_id_t); len -= sizeof(node_id_t);
440 logger(DEBUG_PROTOCOL, LOG_ERR, "Got TCP SPTPS packet from %s (%s) with unknown destination ID", c->name, c->hostname);
444 node_t *from = lookup_node_id((node_id_t *)data);
445 data += sizeof(node_id_t); len -= sizeof(node_id_t);
447 logger(DEBUG_PROTOCOL, LOG_ERR, "Got TCP SPTPS packet from %s (%s) with unknown source ID", c->name, c->hostname);
451 /* Help the sender reach us over UDP.
452 Note that we only do this if we're the destination or the static relay;
453 otherwise every hop would initiate its own UDP info message, resulting in elevated chatter. */
454 if(to->via == myself)
455 send_udp_info(myself, from);
457 /* If we're not the final recipient, relay the packet. */
460 send_sptps_data(to, from, 0, data, len);
465 /* The packet is for us */
467 if(!from->status.validkey) {
468 logger(DEBUG_PROTOCOL, LOG_ERR, "Got SPTPS packet from %s (%s) but we don't have a valid key yet", from->name, from->hostname);
471 sptps_receive_data(&from->sptps, data, len);
472 send_mtu_info(myself, from, MTU);
476 static void send_sptps_packet(node_t *n, vpn_packet_t *origpkt) {
477 if(!n->status.validkey && !n->connection)
483 if(!(DATA(origpkt)[12] | DATA(origpkt)[13])) {
484 sptps_send_record(&n->sptps, PKT_PROBE, (char *)DATA(origpkt), origpkt->len);
488 if(routing_mode == RMODE_ROUTER)
493 if(origpkt->len < offset)
498 if(n->outcompression) {
500 int len = compress_packet(DATA(&outpkt) + offset, DATA(origpkt) + offset, origpkt->len - offset, n->outcompression);
502 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while compressing packet to %s (%s)", n->name, n->hostname);
503 } else if(len < origpkt->len - offset) {
504 outpkt.len = len + offset;
506 type |= PKT_COMPRESSED;
510 /* If we have a direct metaconnection to n, and we can't use UDP, then
511 don't bother with SPTPS and just use a "plaintext" PACKET message.
512 We don't really care about end-to-end security since we're not
513 sending the message through any intermediate nodes. */
514 if(n->connection && origpkt->len > n->minmtu)
515 send_tcppacket(n->connection, origpkt);
517 sptps_send_record(&n->sptps, type, DATA(origpkt) + offset, origpkt->len - offset);
521 static void adapt_socket(const sockaddr_t *sa, int *sock) {
522 /* Make sure we have a suitable socket for the chosen address */
523 if(listen_socket[*sock].sa.sa.sa_family != sa->sa.sa_family) {
524 for(int i = 0; i < listen_sockets; i++) {
525 if(listen_socket[i].sa.sa.sa_family == sa->sa.sa_family) {
533 static void choose_udp_address(const node_t *n, const sockaddr_t **sa, int *sock) {
538 /* If the UDP address is confirmed, use it. */
539 if(n->status.udp_confirmed)
542 /* Send every third packet to n->address; that could be set
543 to the node's reflexive UDP address discovered during key
552 /* Otherwise, address are found in edges to this node.
553 So we pick a random edge and a random socket. */
556 int j = rand() % n->edge_tree->count;
557 edge_t *candidate = NULL;
559 for splay_each(edge_t, e, n->edge_tree) {
561 candidate = e->reverse;
567 *sa = &candidate->address;
568 *sock = rand() % listen_sockets;
571 adapt_socket(*sa, sock);
574 static void choose_local_address(const node_t *n, const sockaddr_t **sa, int *sock) {
577 /* Pick one of the edges from this node at random, then use its local address. */
580 int j = rand() % n->edge_tree->count;
581 edge_t *candidate = NULL;
583 for splay_each(edge_t, e, n->edge_tree) {
590 if (candidate && candidate->local_address.sa.sa_family) {
591 *sa = &candidate->local_address;
592 *sock = rand() % listen_sockets;
593 adapt_socket(*sa, sock);
597 static void send_udppacket(node_t *n, vpn_packet_t *origpkt) {
598 vpn_packet_t pkt1, pkt2;
599 vpn_packet_t *pkt[] = { &pkt1, &pkt2, &pkt1, &pkt2 };
600 vpn_packet_t *inpkt = origpkt;
602 vpn_packet_t *outpkt;
603 int origlen = origpkt->len;
605 #if defined(SOL_IP) && defined(IP_TOS)
606 static int priority = 0;
607 int origpriority = origpkt->priority;
610 pkt1.offset = DEFAULT_PACKET_OFFSET;
611 pkt2.offset = DEFAULT_PACKET_OFFSET;
613 if(!n->status.reachable) {
614 logger(DEBUG_TRAFFIC, LOG_INFO, "Trying to send UDP packet to unreachable node %s (%s)", n->name, n->hostname);
619 return send_sptps_packet(n, origpkt);
621 #ifdef DISABLE_LEGACY
624 /* Make sure we have a valid key */
626 if(!n->status.validkey) {
627 logger(DEBUG_TRAFFIC, LOG_INFO,
628 "No valid key known yet for %s (%s), forwarding via TCP",
629 n->name, n->hostname);
630 send_tcppacket(n->nexthop->connection, origpkt);
634 if(n->options & OPTION_PMTU_DISCOVERY && inpkt->len > n->minmtu && (DATA(inpkt)[12] | DATA(inpkt)[13])) {
635 logger(DEBUG_TRAFFIC, LOG_INFO,
636 "Packet for %s (%s) larger than minimum MTU, forwarding via %s",
637 n->name, n->hostname, n != n->nexthop ? n->nexthop->name : "TCP");
640 send_packet(n->nexthop, origpkt);
642 send_tcppacket(n->nexthop->connection, origpkt);
647 /* Compress the packet */
649 if(n->outcompression) {
650 outpkt = pkt[nextpkt++];
652 if((outpkt->len = compress_packet(DATA(outpkt), DATA(inpkt), inpkt->len, n->outcompression)) < 0) {
653 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while compressing packet to %s (%s)",
654 n->name, n->hostname);
661 /* Add sequence number */
663 seqno_t seqno = htonl(++(n->sent_seqno));
664 memcpy(SEQNO(inpkt), &seqno, sizeof seqno);
665 inpkt->len += sizeof seqno;
667 /* Encrypt the packet */
669 if(cipher_active(n->outcipher)) {
670 outpkt = pkt[nextpkt++];
673 if(!cipher_encrypt(n->outcipher, SEQNO(inpkt), inpkt->len, SEQNO(outpkt), &outlen, true)) {
674 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while encrypting packet to %s (%s)", n->name, n->hostname);
678 outpkt->len = outlen;
682 /* Add the message authentication code */
684 if(digest_active(n->outdigest)) {
685 if(!digest_create(n->outdigest, SEQNO(inpkt), inpkt->len, SEQNO(inpkt) + inpkt->len)) {
686 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while encrypting packet to %s (%s)", n->name, n->hostname);
690 inpkt->len += digest_length(n->outdigest);
693 /* Send the packet */
695 const sockaddr_t *sa = NULL;
698 if(n->status.send_locally)
699 choose_local_address(n, &sa, &sock);
701 choose_udp_address(n, &sa, &sock);
703 #if defined(SOL_IP) && defined(IP_TOS)
704 if(priorityinheritance && origpriority != priority
705 && listen_socket[n->sock].sa.sa.sa_family == AF_INET) {
706 priority = origpriority;
707 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Setting outgoing packet priority to %d", priority);
708 if(setsockopt(listen_socket[n->sock].udp.fd, SOL_IP, IP_TOS, &priority, sizeof(priority))) /* SO_PRIORITY doesn't seem to work */
709 logger(DEBUG_ALWAYS, LOG_ERR, "System call `%s' failed: %s", "setsockopt", sockstrerror(sockerrno));
713 if(sendto(listen_socket[sock].udp.fd, SEQNO(inpkt), inpkt->len, 0, &sa->sa, SALEN(sa->sa)) < 0 && !sockwouldblock(sockerrno)) {
714 if(sockmsgsize(sockerrno)) {
715 if(n->maxmtu >= origlen)
716 n->maxmtu = origlen - 1;
717 if(n->mtu >= origlen)
718 n->mtu = origlen - 1;
721 logger(DEBUG_TRAFFIC, LOG_WARNING, "Error sending packet to %s (%s): %s", n->name, n->hostname, sockstrerror(sockerrno));
725 origpkt->len = origlen;
729 bool send_sptps_data(node_t *to, node_t *from, int type, const void *data, size_t len) {
730 node_t *relay = (to->via != myself && (type == PKT_PROBE || (len - SPTPS_DATAGRAM_OVERHEAD) <= to->via->minmtu)) ? to->via : to->nexthop;
731 bool direct = from == myself && to == relay;
732 bool relay_supported = (relay->options >> 24) >= 4;
733 bool tcponly = (myself->options | relay->options) & OPTION_TCPONLY;
735 /* Send it via TCP if it is a handshake packet, TCPOnly is in use, this is a relay packet that the other node cannot understand, or this packet is larger than the MTU. */
737 if(type == SPTPS_HANDSHAKE || tcponly || (!direct && !relay_supported) || (type != PKT_PROBE && (len - SPTPS_DATAGRAM_OVERHEAD) > relay->minmtu)) {
738 if((from != myself || to->status.validkey) && (to->nexthop->connection->options >> 24) >= 7) {
739 char buf[len + sizeof to->id + sizeof from->id]; char* buf_ptr = buf;
740 memcpy(buf_ptr, &to->id, sizeof to->id); buf_ptr += sizeof to->id;
741 memcpy(buf_ptr, &from->id, sizeof from->id); buf_ptr += sizeof from->id;
742 memcpy(buf_ptr, data, len); buf_ptr += len;
743 logger(DEBUG_TRAFFIC, LOG_INFO, "Sending packet from %s (%s) to %s (%s) via %s (%s) (TCP)", from->name, from->hostname, to->name, to->hostname, to->nexthop->name, to->nexthop->hostname);
744 return send_sptps_tcppacket(to->nexthop->connection, buf, sizeof buf);
747 char buf[len * 4 / 3 + 5];
748 b64encode(data, buf, len);
749 /* If no valid key is known yet, send the packets using ANS_KEY requests,
750 to ensure we get to learn the reflexive UDP address. */
751 if(from == myself && !to->status.validkey) {
752 to->incompression = myself->incompression;
753 return send_request(to->nexthop->connection, "%d %s %s %s -1 -1 -1 %d", ANS_KEY, from->name, to->name, buf, to->incompression);
755 return send_request(to->nexthop->connection, "%d %s %s %d %s", REQ_KEY, from->name, to->name, SPTPS_PACKET, buf);
760 if(relay_supported) overhead += sizeof to->id + sizeof from->id;
761 char buf[len + overhead]; char* buf_ptr = buf;
762 if(relay_supported) {
764 /* Inform the recipient that this packet was sent directly. */
765 node_id_t nullid = {};
766 memcpy(buf_ptr, &nullid, sizeof nullid); buf_ptr += sizeof nullid;
768 memcpy(buf_ptr, &to->id, sizeof to->id); buf_ptr += sizeof to->id;
770 memcpy(buf_ptr, &from->id, sizeof from->id); buf_ptr += sizeof from->id;
773 /* TODO: if this copy turns out to be a performance concern, change sptps_send_record() to add some "pre-padding" to the buffer and use that instead */
774 memcpy(buf_ptr, data, len); buf_ptr += len;
776 const sockaddr_t *sa = NULL;
778 if(relay->status.send_locally)
779 choose_local_address(relay, &sa, &sock);
781 choose_udp_address(relay, &sa, &sock);
782 logger(DEBUG_TRAFFIC, LOG_INFO, "Sending packet from %s (%s) to %s (%s) via %s (%s) (UDP)", from->name, from->hostname, to->name, to->hostname, relay->name, relay->hostname);
783 if(sendto(listen_socket[sock].udp.fd, buf, buf_ptr - buf, 0, &sa->sa, SALEN(sa->sa)) < 0 && !sockwouldblock(sockerrno)) {
784 if(sockmsgsize(sockerrno)) {
785 // Compensate for SPTPS overhead
786 len -= SPTPS_DATAGRAM_OVERHEAD;
787 if(relay->maxmtu >= len)
788 relay->maxmtu = len - 1;
789 if(relay->mtu >= len)
790 relay->mtu = len - 1;
793 logger(DEBUG_TRAFFIC, LOG_WARNING, "Error sending UDP SPTPS packet to %s (%s): %s", relay->name, relay->hostname, sockstrerror(sockerrno));
801 bool receive_sptps_record(void *handle, uint8_t type, const void *data, uint16_t len) {
802 node_t *from = handle;
804 if(type == SPTPS_HANDSHAKE) {
805 if(!from->status.validkey) {
806 from->status.validkey = true;
807 from->status.waitingforkey = false;
808 logger(DEBUG_META, LOG_INFO, "SPTPS key exchange with %s (%s) succesful", from->name, from->hostname);
814 logger(DEBUG_ALWAYS, LOG_ERR, "Packet from %s (%s) larger than maximum supported size (%d > %d)", from->name, from->hostname, len, MTU);
819 inpkt.offset = DEFAULT_PACKET_OFFSET;
821 if(type == PKT_PROBE) {
822 if(!from->status.udppacket) {
823 logger(DEBUG_ALWAYS, LOG_ERR, "Got SPTPS PROBE packet from %s (%s) via TCP", from->name, from->hostname);
827 memcpy(DATA(&inpkt), data, len);
828 if(inpkt.len > from->maxrecentlen)
829 from->maxrecentlen = inpkt.len;
830 udp_probe_h(from, &inpkt, len);
834 if(type & ~(PKT_COMPRESSED | PKT_MAC)) {
835 logger(DEBUG_ALWAYS, LOG_ERR, "Unexpected SPTPS record type %d len %d from %s (%s)", type, len, from->name, from->hostname);
839 /* Check if we have the headers we need */
840 if(routing_mode != RMODE_ROUTER && !(type & PKT_MAC)) {
841 logger(DEBUG_TRAFFIC, LOG_ERR, "Received packet from %s (%s) without MAC header (maybe Mode is not set correctly)", from->name, from->hostname);
843 } else if(routing_mode == RMODE_ROUTER && (type & PKT_MAC)) {
844 logger(DEBUG_TRAFFIC, LOG_WARNING, "Received packet from %s (%s) with MAC header (maybe Mode is not set correctly)", from->name, from->hostname);
847 int offset = (type & PKT_MAC) ? 0 : 14;
848 if(type & PKT_COMPRESSED) {
849 length_t ulen = uncompress_packet(DATA(&inpkt) + offset, (const uint8_t *)data, len, from->incompression);
853 inpkt.len = ulen + offset;
855 if(inpkt.len > MAXSIZE)
858 memcpy(DATA(&inpkt) + offset, data, len);
859 inpkt.len = len + offset;
862 /* Generate the Ethernet packet type if necessary */
864 switch(DATA(&inpkt)[14] >> 4) {
866 DATA(&inpkt)[12] = 0x08;
867 DATA(&inpkt)[13] = 0x00;
870 DATA(&inpkt)[12] = 0x86;
871 DATA(&inpkt)[13] = 0xDD;
874 logger(DEBUG_TRAFFIC, LOG_ERR,
875 "Unknown IP version %d while reading packet from %s (%s)",
876 DATA(&inpkt)[14] >> 4, from->name, from->hostname);
881 if(from->status.udppacket && inpkt.len > from->maxrecentlen)
882 from->maxrecentlen = inpkt.len;
884 receive_packet(from, &inpkt);
888 // This function tries to get SPTPS keys, if they aren't already known.
889 // This function makes no guarantees - it is up to the caller to check the node's state to figure out if the keys are available.
890 static void try_sptps(node_t *n) {
891 if(n->status.validkey)
894 logger(DEBUG_TRAFFIC, LOG_INFO, "No valid key known yet for %s (%s)", n->name, n->hostname);
896 if(!n->status.waitingforkey)
898 else if(n->last_req_key + 10 < now.tv_sec) {
899 logger(DEBUG_ALWAYS, LOG_DEBUG, "No key from %s after 10 seconds, restarting SPTPS", n->name);
900 sptps_stop(&n->sptps);
901 n->status.waitingforkey = false;
908 static void send_udp_probe_packet(node_t *n, int len) {
910 packet.offset = DEFAULT_PACKET_OFFSET;
911 memset(DATA(&packet), 0, 14);
912 randomize(DATA(&packet) + 14, len - 14);
916 logger(DEBUG_TRAFFIC, LOG_INFO, "Sending UDP probe length %d to %s (%s)", len, n->name, n->hostname);
918 send_udppacket(n, &packet);
921 // This function tries to establish a UDP tunnel to a node so that packets can be sent.
922 // If a tunnel is already established, it makes sure it stays up.
923 // This function makes no guarantees - it is up to the caller to check the node's state to figure out if UDP is usable.
924 static void try_udp(node_t* n) {
928 /* Send gratuitous probe replies to 1.1 nodes. */
930 if((n->options >> 24) >= 3 && n->status.udp_confirmed) {
931 struct timeval ping_tx_elapsed;
932 timersub(&now, &n->udp_reply_sent, &ping_tx_elapsed);
934 if(ping_tx_elapsed.tv_sec >= udp_discovery_keepalive_interval - 1) {
935 n->udp_reply_sent = now;
936 if(n->maxrecentlen) {
938 pkt.len = n->maxrecentlen;
939 pkt.offset = DEFAULT_PACKET_OFFSET;
940 memset(DATA(&pkt), 0, 14);
941 randomize(DATA(&pkt) + 14, MIN_PROBE_SIZE - 14);
942 send_udp_probe_reply(n, &pkt, pkt.len);
950 struct timeval ping_tx_elapsed;
951 timersub(&now, &n->udp_ping_sent, &ping_tx_elapsed);
953 int interval = n->status.udp_confirmed ? udp_discovery_keepalive_interval : udp_discovery_interval;
955 if(ping_tx_elapsed.tv_sec >= interval) {
956 send_udp_probe_packet(n, MIN_PROBE_SIZE);
957 n->udp_ping_sent = now;
959 if(localdiscovery && !n->status.udp_confirmed && n->prevedge) {
960 n->status.send_locally = true;
961 send_udp_probe_packet(n, MIN_PROBE_SIZE);
962 n->status.send_locally = false;
967 static length_t choose_initial_maxmtu(node_t *n) {
972 const sockaddr_t *sa = NULL;
974 choose_udp_address(n, &sa, &sockindex);
978 sock = socket(sa->sa.sa_family, SOCK_DGRAM, IPPROTO_UDP);
980 logger(DEBUG_TRAFFIC, LOG_ERR, "Creating MTU assessment socket for %s (%s) failed: %s", n->name, n->hostname, sockstrerror(sockerrno));
984 if(connect(sock, &sa->sa, SALEN(sa->sa))) {
985 logger(DEBUG_TRAFFIC, LOG_ERR, "Connecting MTU assessment socket for %s (%s) failed: %s", n->name, n->hostname, sockstrerror(sockerrno));
991 socklen_t ip_mtu_len = sizeof ip_mtu;
992 if(getsockopt(sock, IPPROTO_IP, IP_MTU, &ip_mtu, &ip_mtu_len)) {
993 logger(DEBUG_TRAFFIC, LOG_ERR, "getsockopt(IP_MTU) on %s (%s) failed: %s", n->name, n->hostname, sockstrerror(sockerrno));
1000 /* getsockopt(IP_MTU) returns the MTU of the physical interface.
1001 We need to remove various overheads to get to the tinc MTU. */
1002 length_t mtu = ip_mtu;
1003 mtu -= (sa->sa.sa_family == AF_INET6) ? sizeof(struct ip6_hdr) : sizeof(struct ip);
1005 if(n->status.sptps) {
1006 mtu -= SPTPS_DATAGRAM_OVERHEAD;
1007 if((n->options >> 24) >= 4)
1008 mtu -= sizeof(node_id_t) + sizeof(node_id_t);
1009 #ifndef DISABLE_LEGACY
1011 mtu -= digest_length(n->outdigest);
1013 /* Now it's tricky. We use CBC mode, so the length of the
1014 encrypted payload must be a multiple of the blocksize. The
1015 sequence number is also part of the encrypted payload, so we
1016 must account for it after correcting for the blocksize.
1017 Furthermore, the padding in the last block must be at least
1020 length_t blocksize = cipher_blocksize(n->outcipher);
1033 logger(DEBUG_TRAFFIC, LOG_ERR, "getsockopt(IP_MTU) on %s (%s) returned absurdly small value: %d", n->name, n->hostname, ip_mtu);
1039 logger(DEBUG_TRAFFIC, LOG_INFO, "Using system-provided maximum tinc MTU for %s (%s): %hd", n->name, n->hostname, mtu);
1049 /* This function tries to determines the MTU of a node.
1050 By calling this function repeatedly, n->minmtu will be progressively
1051 increased, and at some point, n->mtu will be fixed to n->minmtu. If the MTU
1052 is already fixed, this function checks if it can be increased.
1055 static void try_mtu(node_t *n) {
1056 if(!(n->options & OPTION_PMTU_DISCOVERY))
1059 if(udp_discovery && !n->status.udp_confirmed) {
1060 n->maxrecentlen = 0;
1067 /* mtuprobes == 0..19: initial discovery, send bursts with 1 second interval, mtuprobes++
1068 mtuprobes == 20: fix MTU, and go to -1
1069 mtuprobes == -1: send one maxmtu and one maxmtu+1 probe every pinginterval
1070 mtuprobes ==-2..-3: send one maxmtu probe every second
1071 mtuprobes == -4: maxmtu no longer valid, reset minmtu and maxmtu and go to 0 */
1073 struct timeval elapsed;
1074 timersub(&now, &n->mtu_ping_sent, &elapsed);
1075 if(n->mtuprobes >= 0) {
1076 if(n->mtuprobes != 0 && elapsed.tv_sec == 0 && elapsed.tv_usec < 333333)
1079 if(n->mtuprobes < -1) {
1080 if(elapsed.tv_sec < 1)
1083 if(elapsed.tv_sec < pinginterval)
1088 n->mtu_ping_sent = now;
1092 if(n->mtuprobes < -3) {
1093 /* We lost three MTU probes, restart discovery */
1094 logger(DEBUG_TRAFFIC, LOG_INFO, "Decrease in PMTU to %s (%s) detected, restarting PMTU discovery", n->name, n->hostname);
1099 if(n->mtuprobes < 0) {
1100 /* After the initial discovery, we only send one maxmtu and one
1101 maxmtu+1 probe to detect PMTU increases. */
1102 send_udp_probe_packet(n, n->maxmtu);
1103 if(n->mtuprobes == -1 && n->maxmtu + 1 < MTU)
1104 send_udp_probe_packet(n, n->maxmtu + 1);
1107 /* Before initial discovery begins, set maxmtu to the most likely value.
1108 If it's underestimated, we will correct it after initial discovery. */
1109 if(n->mtuprobes == 0)
1110 n->maxmtu = choose_initial_maxmtu(n);
1113 /* Decreasing the number of probes per cycle might make the algorithm react faster to lost packets,
1114 but it will typically increase convergence time in the no-loss case. */
1115 const length_t probes_per_cycle = 8;
1117 /* This magic value was determined using math simulations.
1118 It will result in a 1329-byte first probe, followed (if there was a reply) by a 1407-byte probe.
1119 Since 1407 is just below the range of tinc MTUs over typical networks,
1120 this fine-tuning allows tinc to cover a lot of ground very quickly.
1121 This fine-tuning is only valid for maxmtu = MTU; if maxmtu is smaller,
1122 then it's better to use a multiplier of 1. Indeed, this leads to an interesting scenario
1123 if choose_initial_maxmtu() returns the actual MTU value - it will get confirmed with one single probe. */
1124 const float multiplier = (n->maxmtu == MTU) ? 0.97 : 1;
1126 const float cycle_position = probes_per_cycle - (n->mtuprobes % probes_per_cycle) - 1;
1127 const length_t minmtu = MAX(n->minmtu, 512);
1128 const float interval = n->maxmtu - minmtu;
1130 /* The core of the discovery algorithm is this exponential.
1131 It produces very large probes early in the cycle, and then it very quickly decreases the probe size.
1132 This reflects the fact that in the most difficult cases, we don't get any feedback for probes that
1133 are too large, and therefore we need to concentrate on small offsets so that we can quickly converge
1134 on the precise MTU as we are approaching it.
1135 The last probe of the cycle is always 1 byte in size - this is to make sure we'll get at least one
1136 reply per cycle so that we can make progress. */
1137 const length_t offset = powf(interval, multiplier * cycle_position / (probes_per_cycle - 1));
1139 length_t maxmtu = n->maxmtu;
1140 send_udp_probe_packet(n, minmtu + offset);
1141 /* If maxmtu changed, it means the probe was rejected by the system because it was too large.
1142 In that case, we recalculate with the new maxmtu and try again. */
1143 if(n->mtuprobes < 0 || maxmtu == n->maxmtu)
1147 if(n->mtuprobes >= 0)
1152 /* These functions try to establish a tunnel to a node (or its relay) so that
1153 packets can be sent (e.g. exchange keys).
1154 If a tunnel is already established, it tries to improve it (e.g. by trying
1155 to establish a UDP tunnel instead of TCP). This function makes no
1156 guarantees - it is up to the caller to check the node's state to figure out
1157 if TCP and/or UDP is usable. By calling this function repeatedly, the
1158 tunnel is gradually improved until we hit the wall imposed by the underlying
1159 network environment. It is recommended to call this function every time a
1160 packet is sent (or intended to be sent) to a node, so that the tunnel keeps
1161 improving as packets flow, and then gracefully downgrades itself as it goes
1165 static void try_tx_sptps(node_t *n, bool mtu) {
1166 /* If n is a TCP-only neighbor, we'll only use "cleartext" PACKET
1167 messages anyway, so there's no need for SPTPS at all. */
1169 if(n->connection && ((myself->options | n->options) & OPTION_TCPONLY))
1172 /* Otherwise, try to do SPTPS authentication with n if necessary. */
1176 /* Do we need to statically relay packets? */
1178 node_t *via = (n->via == myself) ? n->nexthop : n->via;
1180 /* If the static relay doesn't support SPTPS, everything goes via TCP anyway. */
1182 if((via->options >> 24) < 4)
1185 /* If we do have a static relay, try everything with that one instead. */
1188 return try_tx_sptps(via, mtu);
1190 /* Otherwise, try to establish UDP connectivity. */
1196 /* If we don't have UDP connectivity (yet), we need to use a dynamic relay (nexthop)
1197 while we try to establish direct connectivity. */
1199 if(!n->status.udp_confirmed && n != n->nexthop && (n->nexthop->options >> 24) >= 4)
1200 try_tx_sptps(n->nexthop, mtu);
1203 static void try_tx_legacy(node_t *n, bool mtu) {
1204 /* Does he have our key? If not, send one. */
1206 if(!n->status.validkey_in)
1209 /* Check if we already have a key, or request one. */
1211 if(!n->status.validkey) {
1212 if(n->last_req_key + 10 <= now.tv_sec) {
1214 n->last_req_key = now.tv_sec;
1224 void try_tx(node_t *n, bool mtu) {
1226 try_tx_sptps(n, mtu);
1228 try_tx_legacy(n, mtu);
1231 void send_packet(node_t *n, vpn_packet_t *packet) {
1232 // If it's for myself, write it to the tun/tap device.
1236 memcpy(DATA(packet), mymac.x, ETH_ALEN);
1238 n->out_bytes += packet->len;
1239 devops.write(packet);
1243 logger(DEBUG_TRAFFIC, LOG_ERR, "Sending packet of %d bytes to %s (%s)", packet->len, n->name, n->hostname);
1245 // If the node is not reachable, drop it.
1247 if(!n->status.reachable) {
1248 logger(DEBUG_TRAFFIC, LOG_INFO, "Node %s (%s) is not reachable", n->name, n->hostname);
1252 // Keep track of packet statistics.
1255 n->out_bytes += packet->len;
1257 // Check if it should be sent as an SPTPS packet.
1259 if(n->status.sptps) {
1260 send_sptps_packet(n, packet);
1261 try_tx_sptps(n, true);
1265 // Determine which node to actually send it to.
1267 node_t *via = (packet->priority == -1 || n->via == myself) ? n->nexthop : n->via;
1270 logger(DEBUG_TRAFFIC, LOG_INFO, "Sending packet to %s via %s (%s)", n->name, via->name, n->via->hostname);
1272 // Try to send via UDP, unless TCP is forced.
1274 if(packet->priority == -1 || ((myself->options | via->options) & OPTION_TCPONLY)) {
1275 if(!send_tcppacket(via->connection, packet))
1276 terminate_connection(via->connection, true);
1280 send_udppacket(via, packet);
1281 try_tx_legacy(via, true);
1284 void broadcast_packet(const node_t *from, vpn_packet_t *packet) {
1285 // Always give ourself a copy of the packet.
1287 send_packet(myself, packet);
1289 // In TunnelServer mode, do not forward broadcast packets.
1290 // The MST might not be valid and create loops.
1291 if(tunnelserver || broadcast_mode == BMODE_NONE)
1294 logger(DEBUG_TRAFFIC, LOG_INFO, "Broadcasting packet of %d bytes from %s (%s)",
1295 packet->len, from->name, from->hostname);
1297 switch(broadcast_mode) {
1298 // In MST mode, broadcast packets travel via the Minimum Spanning Tree.
1299 // This guarantees all nodes receive the broadcast packet, and
1300 // usually distributes the sending of broadcast packets over all nodes.
1302 for list_each(connection_t, c, connection_list)
1303 if(c->edge && c->status.mst && c != from->nexthop->connection)
1304 send_packet(c->node, packet);
1307 // In direct mode, we send copies to each node we know of.
1308 // However, this only reaches nodes that can be reached in a single hop.
1309 // We don't have enough information to forward broadcast packets in this case.
1314 for splay_each(node_t, n, node_tree)
1315 if(n->status.reachable && n != myself && ((n->via == myself && n->nexthop == n) || n->via == n))
1316 send_packet(n, packet);
1324 /* We got a packet from some IP address, but we don't know who sent it. Try to
1325 verify the message authentication code against all active session keys.
1326 Since this is actually an expensive operation, we only do a full check once
1327 a minute, the rest of the time we only check against nodes for which we know
1328 an IP address that matches the one from the packet. */
1330 static node_t *try_harder(const sockaddr_t *from, const vpn_packet_t *pkt) {
1331 node_t *match = NULL;
1333 static time_t last_hard_try = 0;
1335 for splay_each(node_t, n, node_tree) {
1336 if(!n->status.reachable || n == myself)
1339 if((n->status.sptps && !n->sptps.instate) || !n->status.validkey_in)
1344 for splay_each(edge_t, e, n->edge_tree) {
1347 if(!sockaddrcmp_noport(from, &e->reverse->address)) {
1354 if(last_hard_try == now.tv_sec)
1359 if(!try_mac(n, pkt))
1367 last_hard_try = now.tv_sec;
1372 void handle_incoming_vpn_data(void *data, int flags) {
1373 listen_socket_t *ls = data;
1376 node_id_t nullid = {};
1377 sockaddr_t addr = {};
1378 socklen_t addrlen = sizeof addr;
1380 bool direct = false;
1383 int len = recvfrom(ls->udp.fd, DATA(&pkt), MAXSIZE, 0, &addr.sa, &addrlen);
1385 if(len <= 0 || len > MAXSIZE) {
1386 if(!sockwouldblock(sockerrno))
1387 logger(DEBUG_ALWAYS, LOG_ERR, "Receiving packet failed: %s", sockstrerror(sockerrno));
1393 sockaddrunmap(&addr); /* Some braindead IPv6 implementations do stupid things. */
1395 // Try to figure out who sent this packet.
1397 node_t *n = lookup_node_udp(&addr);
1399 if(n && !n->status.udp_confirmed)
1400 n = NULL; // Don't believe it if we don't have confirmation yet.
1403 // It might be from a 1.1 node, which might have a source ID in the packet.
1404 pkt.offset = 2 * sizeof(node_id_t);
1405 from = lookup_node_id(SRCID(&pkt));
1406 if(from && !memcmp(DSTID(&pkt), &nullid, sizeof nullid) && from->status.sptps) {
1407 if(sptps_verify_datagram(&from->sptps, DATA(&pkt), pkt.len - 2 * sizeof(node_id_t)))
1416 n = try_harder(&addr, &pkt);
1421 if(debug_level >= DEBUG_PROTOCOL) {
1422 hostname = sockaddr2hostname(&addr);
1423 logger(DEBUG_PROTOCOL, LOG_WARNING, "Received UDP packet from unknown source %s", hostname);
1429 if(n->status.sptps) {
1430 pkt.offset = 2 * sizeof(node_id_t);
1432 if(!memcmp(DSTID(&pkt), &nullid, sizeof nullid)) {
1437 from = lookup_node_id(SRCID(&pkt));
1438 to = lookup_node_id(DSTID(&pkt));
1441 logger(DEBUG_PROTOCOL, LOG_WARNING, "Received UDP packet from %s (%s) with unknown source and/or destination ID", n->name, n->hostname);
1445 /* The packet is supposed to come from the originator or its static relay
1446 (i.e. with no dynamic relays in between).
1447 If it did not, "help" the static relay by sending it UDP info.
1448 Note that we only do this if we're the destination or the static relay;
1449 otherwise every hop would initiate its own UDP info message, resulting in elevated chatter. */
1451 if(n != from->via && to->via == myself)
1452 send_udp_info(myself, from);
1454 /* If we're not the final recipient, relay the packet. */
1457 send_sptps_data(to, from, 0, DATA(&pkt), pkt.len - 2 * sizeof(node_id_t));
1458 try_tx_sptps(to, true);
1467 if(!receive_udppacket(from, &pkt))
1470 n->sock = ls - listen_socket;
1471 if(direct && sockaddrcmp(&addr, &n->address))
1472 update_node_udp(n, &addr);
1474 /* If the packet went through a relay, help the sender find the appropriate MTU
1475 through the relay path. */
1478 send_mtu_info(myself, n, MTU);
1481 void handle_device_data(void *data, int flags) {
1482 vpn_packet_t packet;
1483 packet.offset = DEFAULT_PACKET_OFFSET;
1484 packet.priority = 0;
1486 if(devops.read(&packet)) {
1487 myself->in_packets++;
1488 myself->in_bytes += packet.len;
1489 route(myself, &packet);