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 n->status.udppacket = true;
286 bool result = sptps_receive_data(&n->sptps, DATA(inpkt), inpkt->len);
287 n->status.udppacket = false;
290 /* Uh-oh. It might be that the tunnel is stuck in some corrupted state,
291 so let's restart SPTPS in case that helps. But don't do that too often
292 to prevent storms, and because that would make life a little too easy
293 for external attackers trying to DoS us. */
294 if(n->last_req_key < now.tv_sec - 10) {
295 logger(DEBUG_PROTOCOL, LOG_ERR, "Failed to decode raw TCP packet from %s (%s), restarting SPTPS", n->name, n->hostname);
303 #ifdef DISABLE_LEGACY
306 if(!n->status.validkey_in) {
307 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got packet from %s (%s) but he hasn't got our key yet", n->name, n->hostname);
311 /* Check packet length */
313 if(inpkt->len < sizeof(seqno_t) + digest_length(n->indigest)) {
314 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got too short packet from %s (%s)",
315 n->name, n->hostname);
319 /* It's a legacy UDP packet, the data starts after the seqno */
321 inpkt->offset += sizeof(seqno_t);
323 /* Check the message authentication code */
325 if(digest_active(n->indigest)) {
326 inpkt->len -= digest_length(n->indigest);
327 if(!digest_verify(n->indigest, SEQNO(inpkt), inpkt->len, SEQNO(inpkt) + inpkt->len)) {
328 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got unauthenticated packet from %s (%s)", n->name, n->hostname);
332 /* Decrypt the packet */
334 if(cipher_active(n->incipher)) {
335 vpn_packet_t *outpkt = pkt[nextpkt++];
338 if(!cipher_decrypt(n->incipher, SEQNO(inpkt), inpkt->len, SEQNO(outpkt), &outlen, true)) {
339 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Error decrypting packet from %s (%s)", n->name, n->hostname);
343 outpkt->len = outlen;
347 /* Check the sequence number */
350 memcpy(&seqno, SEQNO(inpkt), sizeof seqno);
351 seqno = ntohl(seqno);
352 inpkt->len -= sizeof seqno;
355 if(seqno != n->received_seqno + 1) {
356 if(seqno >= n->received_seqno + replaywin * 8) {
357 if(n->farfuture++ < replaywin >> 2) {
358 logger(DEBUG_ALWAYS, LOG_WARNING, "Packet from %s (%s) is %d seqs in the future, dropped (%u)",
359 n->name, n->hostname, seqno - n->received_seqno - 1, n->farfuture);
362 logger(DEBUG_ALWAYS, LOG_WARNING, "Lost %d packets from %s (%s)",
363 seqno - n->received_seqno - 1, n->name, n->hostname);
364 memset(n->late, 0, replaywin);
365 } else if (seqno <= n->received_seqno) {
366 if((n->received_seqno >= replaywin * 8 && seqno <= n->received_seqno - replaywin * 8) || !(n->late[(seqno / 8) % replaywin] & (1 << seqno % 8))) {
367 logger(DEBUG_ALWAYS, LOG_WARNING, "Got late or replayed packet from %s (%s), seqno %d, last received %d",
368 n->name, n->hostname, seqno, n->received_seqno);
372 for(int i = n->received_seqno + 1; i < seqno; i++)
373 n->late[(i / 8) % replaywin] |= 1 << i % 8;
378 n->late[(seqno / 8) % replaywin] &= ~(1 << seqno % 8);
381 if(seqno > n->received_seqno)
382 n->received_seqno = seqno;
386 if(n->received_seqno > MAX_SEQNO)
389 /* Decompress the packet */
391 length_t origlen = inpkt->len;
393 if(n->incompression) {
394 vpn_packet_t *outpkt = pkt[nextpkt++];
396 if((outpkt->len = uncompress_packet(DATA(outpkt), DATA(inpkt), inpkt->len, n->incompression)) < 0) {
397 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while uncompressing packet from %s (%s)",
398 n->name, n->hostname);
404 origlen -= MTU/64 + 20;
407 if(inpkt->len > n->maxrecentlen)
408 n->maxrecentlen = inpkt->len;
412 if(!DATA(inpkt)[12] && !DATA(inpkt)[13])
413 udp_probe_h(n, inpkt, origlen);
415 receive_packet(n, inpkt);
420 void receive_tcppacket(connection_t *c, const char *buffer, int len) {
422 outpkt.offset = DEFAULT_PACKET_OFFSET;
424 if(len > sizeof outpkt.data - outpkt.offset)
428 if(c->options & OPTION_TCPONLY)
431 outpkt.priority = -1;
432 memcpy(DATA(&outpkt), buffer, len);
434 receive_packet(c->node, &outpkt);
437 bool receive_tcppacket_sptps(connection_t *c, const char *data, int len) {
438 if (len < sizeof(node_id_t) + sizeof(node_id_t)) {
439 logger(DEBUG_ALWAYS, LOG_ERR, "Got too short TCP SPTPS packet from %s (%s)", c->name, c->hostname);
443 node_t *to = lookup_node_id((node_id_t *)data);
444 data += sizeof(node_id_t); len -= sizeof(node_id_t);
446 logger(DEBUG_PROTOCOL, LOG_ERR, "Got TCP SPTPS packet from %s (%s) with unknown destination ID", c->name, c->hostname);
450 node_t *from = lookup_node_id((node_id_t *)data);
451 data += sizeof(node_id_t); len -= sizeof(node_id_t);
453 logger(DEBUG_PROTOCOL, LOG_ERR, "Got TCP SPTPS packet from %s (%s) with unknown source ID", c->name, c->hostname);
457 /* Help the sender reach us over UDP.
458 Note that we only do this if we're the destination or the static relay;
459 otherwise every hop would initiate its own UDP info message, resulting in elevated chatter. */
460 if(to->via == myself)
461 send_udp_info(myself, from);
463 /* If we're not the final recipient, relay the packet. */
466 send_sptps_data(to, from, 0, data, len);
471 /* The packet is for us */
473 if(!sptps_receive_data(&from->sptps, data, len)) {
474 /* Uh-oh. It might be that the tunnel is stuck in some corrupted state,
475 so let's restart SPTPS in case that helps. But don't do that too often
476 to prevent storms. */
477 if(from->last_req_key < now.tv_sec - 10) {
478 logger(DEBUG_PROTOCOL, LOG_ERR, "Failed to decode raw TCP packet from %s (%s), restarting SPTPS", from->name, from->hostname);
484 send_mtu_info(myself, from, MTU);
488 static void send_sptps_packet(node_t *n, vpn_packet_t *origpkt) {
489 if(!n->status.validkey && !n->connection)
495 if(!(DATA(origpkt)[12] | DATA(origpkt)[13])) {
496 sptps_send_record(&n->sptps, PKT_PROBE, (char *)DATA(origpkt), origpkt->len);
500 if(routing_mode == RMODE_ROUTER)
505 if(origpkt->len < offset)
510 if(n->outcompression) {
512 int len = compress_packet(DATA(&outpkt) + offset, DATA(origpkt) + offset, origpkt->len - offset, n->outcompression);
514 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while compressing packet to %s (%s)", n->name, n->hostname);
515 } else if(len < origpkt->len - offset) {
516 outpkt.len = len + offset;
518 type |= PKT_COMPRESSED;
522 /* If we have a direct metaconnection to n, and we can't use UDP, then
523 don't bother with SPTPS and just use a "plaintext" PACKET message.
524 We don't really care about end-to-end security since we're not
525 sending the message through any intermediate nodes. */
526 if(n->connection && origpkt->len > n->minmtu)
527 send_tcppacket(n->connection, origpkt);
529 sptps_send_record(&n->sptps, type, DATA(origpkt) + offset, origpkt->len - offset);
533 static void adapt_socket(const sockaddr_t *sa, int *sock) {
534 /* Make sure we have a suitable socket for the chosen address */
535 if(listen_socket[*sock].sa.sa.sa_family != sa->sa.sa_family) {
536 for(int i = 0; i < listen_sockets; i++) {
537 if(listen_socket[i].sa.sa.sa_family == sa->sa.sa_family) {
545 static void choose_udp_address(const node_t *n, const sockaddr_t **sa, int *sock) {
550 /* If the UDP address is confirmed, use it. */
551 if(n->status.udp_confirmed)
554 /* Send every third packet to n->address; that could be set
555 to the node's reflexive UDP address discovered during key
564 /* Otherwise, address are found in edges to this node.
565 So we pick a random edge and a random socket. */
568 int j = rand() % n->edge_tree->count;
569 edge_t *candidate = NULL;
571 for splay_each(edge_t, e, n->edge_tree) {
573 candidate = e->reverse;
579 *sa = &candidate->address;
580 *sock = rand() % listen_sockets;
583 adapt_socket(*sa, sock);
586 static void choose_local_address(const node_t *n, const sockaddr_t **sa, int *sock) {
589 /* Pick one of the edges from this node at random, then use its local address. */
592 int j = rand() % n->edge_tree->count;
593 edge_t *candidate = NULL;
595 for splay_each(edge_t, e, n->edge_tree) {
602 if (candidate && candidate->local_address.sa.sa_family) {
603 *sa = &candidate->local_address;
604 *sock = rand() % listen_sockets;
605 adapt_socket(*sa, sock);
609 static void send_udppacket(node_t *n, vpn_packet_t *origpkt) {
610 vpn_packet_t pkt1, pkt2;
611 vpn_packet_t *pkt[] = { &pkt1, &pkt2, &pkt1, &pkt2 };
612 vpn_packet_t *inpkt = origpkt;
614 vpn_packet_t *outpkt;
615 int origlen = origpkt->len;
617 #if defined(SOL_IP) && defined(IP_TOS)
618 static int priority = 0;
619 int origpriority = origpkt->priority;
622 pkt1.offset = DEFAULT_PACKET_OFFSET;
623 pkt2.offset = DEFAULT_PACKET_OFFSET;
625 if(!n->status.reachable) {
626 logger(DEBUG_TRAFFIC, LOG_INFO, "Trying to send UDP packet to unreachable node %s (%s)", n->name, n->hostname);
631 return send_sptps_packet(n, origpkt);
633 #ifdef DISABLE_LEGACY
636 /* Make sure we have a valid key */
638 if(!n->status.validkey) {
639 logger(DEBUG_TRAFFIC, LOG_INFO,
640 "No valid key known yet for %s (%s), forwarding via TCP",
641 n->name, n->hostname);
642 send_tcppacket(n->nexthop->connection, origpkt);
646 if(n->options & OPTION_PMTU_DISCOVERY && inpkt->len > n->minmtu && (DATA(inpkt)[12] | DATA(inpkt)[13])) {
647 logger(DEBUG_TRAFFIC, LOG_INFO,
648 "Packet for %s (%s) larger than minimum MTU, forwarding via %s",
649 n->name, n->hostname, n != n->nexthop ? n->nexthop->name : "TCP");
652 send_packet(n->nexthop, origpkt);
654 send_tcppacket(n->nexthop->connection, origpkt);
659 /* Compress the packet */
661 if(n->outcompression) {
662 outpkt = pkt[nextpkt++];
664 if((outpkt->len = compress_packet(DATA(outpkt), DATA(inpkt), inpkt->len, n->outcompression)) < 0) {
665 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while compressing packet to %s (%s)",
666 n->name, n->hostname);
673 /* Add sequence number */
675 seqno_t seqno = htonl(++(n->sent_seqno));
676 memcpy(SEQNO(inpkt), &seqno, sizeof seqno);
677 inpkt->len += sizeof seqno;
679 /* Encrypt the packet */
681 if(cipher_active(n->outcipher)) {
682 outpkt = pkt[nextpkt++];
685 if(!cipher_encrypt(n->outcipher, SEQNO(inpkt), inpkt->len, SEQNO(outpkt), &outlen, true)) {
686 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while encrypting packet to %s (%s)", n->name, n->hostname);
690 outpkt->len = outlen;
694 /* Add the message authentication code */
696 if(digest_active(n->outdigest)) {
697 if(!digest_create(n->outdigest, SEQNO(inpkt), inpkt->len, SEQNO(inpkt) + inpkt->len)) {
698 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while encrypting packet to %s (%s)", n->name, n->hostname);
702 inpkt->len += digest_length(n->outdigest);
705 /* Send the packet */
707 const sockaddr_t *sa = NULL;
710 if(n->status.send_locally)
711 choose_local_address(n, &sa, &sock);
713 choose_udp_address(n, &sa, &sock);
715 #if defined(SOL_IP) && defined(IP_TOS)
716 if(priorityinheritance && origpriority != priority
717 && listen_socket[n->sock].sa.sa.sa_family == AF_INET) {
718 priority = origpriority;
719 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Setting outgoing packet priority to %d", priority);
720 if(setsockopt(listen_socket[n->sock].udp.fd, SOL_IP, IP_TOS, &priority, sizeof(priority))) /* SO_PRIORITY doesn't seem to work */
721 logger(DEBUG_ALWAYS, LOG_ERR, "System call `%s' failed: %s", "setsockopt", sockstrerror(sockerrno));
725 if(sendto(listen_socket[sock].udp.fd, SEQNO(inpkt), inpkt->len, 0, &sa->sa, SALEN(sa->sa)) < 0 && !sockwouldblock(sockerrno)) {
726 if(sockmsgsize(sockerrno)) {
727 if(n->maxmtu >= origlen)
728 n->maxmtu = origlen - 1;
729 if(n->mtu >= origlen)
730 n->mtu = origlen - 1;
733 logger(DEBUG_TRAFFIC, LOG_WARNING, "Error sending packet to %s (%s): %s", n->name, n->hostname, sockstrerror(sockerrno));
737 origpkt->len = origlen;
741 bool send_sptps_data(node_t *to, node_t *from, int type, const void *data, size_t len) {
742 node_t *relay = (to->via != myself && (type == PKT_PROBE || (len - SPTPS_DATAGRAM_OVERHEAD) <= to->via->minmtu)) ? to->via : to->nexthop;
743 bool direct = from == myself && to == relay;
744 bool relay_supported = (relay->options >> 24) >= 4;
745 bool tcponly = (myself->options | relay->options) & OPTION_TCPONLY;
747 /* 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. */
749 if(type == SPTPS_HANDSHAKE || tcponly || (!direct && !relay_supported) || (type != PKT_PROBE && (len - SPTPS_DATAGRAM_OVERHEAD) > relay->minmtu)) {
750 if(type != SPTPS_HANDSHAKE && (to->nexthop->connection->options >> 24) >= 7) {
751 char buf[len + sizeof to->id + sizeof from->id]; char* buf_ptr = buf;
752 memcpy(buf_ptr, &to->id, sizeof to->id); buf_ptr += sizeof to->id;
753 memcpy(buf_ptr, &from->id, sizeof from->id); buf_ptr += sizeof from->id;
754 memcpy(buf_ptr, data, len); buf_ptr += len;
755 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);
756 return send_sptps_tcppacket(to->nexthop->connection, buf, sizeof buf);
759 char buf[len * 4 / 3 + 5];
760 b64encode(data, buf, len);
761 /* If this is a handshake packet, use ANS_KEY instead of REQ_KEY, for two reasons:
762 - We don't want intermediate nodes to switch to UDP to relay these packets;
763 - ANS_KEY allows us to learn the reflexive UDP address. */
764 if(type == SPTPS_HANDSHAKE) {
765 to->incompression = myself->incompression;
766 return send_request(to->nexthop->connection, "%d %s %s %s -1 -1 -1 %d", ANS_KEY, from->name, to->name, buf, to->incompression);
768 return send_request(to->nexthop->connection, "%d %s %s %d %s", REQ_KEY, from->name, to->name, SPTPS_PACKET, buf);
773 if(relay_supported) overhead += sizeof to->id + sizeof from->id;
774 char buf[len + overhead]; char* buf_ptr = buf;
775 if(relay_supported) {
777 /* Inform the recipient that this packet was sent directly. */
778 node_id_t nullid = {};
779 memcpy(buf_ptr, &nullid, sizeof nullid); buf_ptr += sizeof nullid;
781 memcpy(buf_ptr, &to->id, sizeof to->id); buf_ptr += sizeof to->id;
783 memcpy(buf_ptr, &from->id, sizeof from->id); buf_ptr += sizeof from->id;
786 /* 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 */
787 memcpy(buf_ptr, data, len); buf_ptr += len;
789 const sockaddr_t *sa = NULL;
791 if(relay->status.send_locally)
792 choose_local_address(relay, &sa, &sock);
794 choose_udp_address(relay, &sa, &sock);
795 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);
796 if(sendto(listen_socket[sock].udp.fd, buf, buf_ptr - buf, 0, &sa->sa, SALEN(sa->sa)) < 0 && !sockwouldblock(sockerrno)) {
797 if(sockmsgsize(sockerrno)) {
798 // Compensate for SPTPS overhead
799 len -= SPTPS_DATAGRAM_OVERHEAD;
800 if(relay->maxmtu >= len)
801 relay->maxmtu = len - 1;
802 if(relay->mtu >= len)
803 relay->mtu = len - 1;
806 logger(DEBUG_TRAFFIC, LOG_WARNING, "Error sending UDP SPTPS packet to %s (%s): %s", relay->name, relay->hostname, sockstrerror(sockerrno));
814 bool receive_sptps_record(void *handle, uint8_t type, const void *data, uint16_t len) {
815 node_t *from = handle;
817 if(type == SPTPS_HANDSHAKE) {
818 if(!from->status.validkey) {
819 from->status.validkey = true;
820 from->status.waitingforkey = false;
821 logger(DEBUG_META, LOG_INFO, "SPTPS key exchange with %s (%s) succesful", from->name, from->hostname);
827 logger(DEBUG_ALWAYS, LOG_ERR, "Packet from %s (%s) larger than maximum supported size (%d > %d)", from->name, from->hostname, len, MTU);
832 inpkt.offset = DEFAULT_PACKET_OFFSET;
834 if(type == PKT_PROBE) {
835 if(!from->status.udppacket) {
836 logger(DEBUG_ALWAYS, LOG_ERR, "Got SPTPS PROBE packet from %s (%s) via TCP", from->name, from->hostname);
840 memcpy(DATA(&inpkt), data, len);
841 if(inpkt.len > from->maxrecentlen)
842 from->maxrecentlen = inpkt.len;
843 udp_probe_h(from, &inpkt, len);
847 if(type & ~(PKT_COMPRESSED | PKT_MAC)) {
848 logger(DEBUG_ALWAYS, LOG_ERR, "Unexpected SPTPS record type %d len %d from %s (%s)", type, len, from->name, from->hostname);
852 /* Check if we have the headers we need */
853 if(routing_mode != RMODE_ROUTER && !(type & PKT_MAC)) {
854 logger(DEBUG_TRAFFIC, LOG_ERR, "Received packet from %s (%s) without MAC header (maybe Mode is not set correctly)", from->name, from->hostname);
856 } else if(routing_mode == RMODE_ROUTER && (type & PKT_MAC)) {
857 logger(DEBUG_TRAFFIC, LOG_WARNING, "Received packet from %s (%s) with MAC header (maybe Mode is not set correctly)", from->name, from->hostname);
860 int offset = (type & PKT_MAC) ? 0 : 14;
861 if(type & PKT_COMPRESSED) {
862 length_t ulen = uncompress_packet(DATA(&inpkt) + offset, (const uint8_t *)data, len, from->incompression);
866 inpkt.len = ulen + offset;
868 if(inpkt.len > MAXSIZE)
871 memcpy(DATA(&inpkt) + offset, data, len);
872 inpkt.len = len + offset;
875 /* Generate the Ethernet packet type if necessary */
877 switch(DATA(&inpkt)[14] >> 4) {
879 DATA(&inpkt)[12] = 0x08;
880 DATA(&inpkt)[13] = 0x00;
883 DATA(&inpkt)[12] = 0x86;
884 DATA(&inpkt)[13] = 0xDD;
887 logger(DEBUG_TRAFFIC, LOG_ERR,
888 "Unknown IP version %d while reading packet from %s (%s)",
889 DATA(&inpkt)[14] >> 4, from->name, from->hostname);
894 if(from->status.udppacket && inpkt.len > from->maxrecentlen)
895 from->maxrecentlen = inpkt.len;
897 receive_packet(from, &inpkt);
901 // This function tries to get SPTPS keys, if they aren't already known.
902 // 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.
903 static void try_sptps(node_t *n) {
904 if(n->status.validkey)
907 logger(DEBUG_TRAFFIC, LOG_INFO, "No valid key known yet for %s (%s)", n->name, n->hostname);
909 if(!n->status.waitingforkey)
911 else if(n->last_req_key + 10 < now.tv_sec) {
912 logger(DEBUG_ALWAYS, LOG_DEBUG, "No key from %s after 10 seconds, restarting SPTPS", n->name);
913 sptps_stop(&n->sptps);
914 n->status.waitingforkey = false;
921 static void send_udp_probe_packet(node_t *n, int len) {
923 packet.offset = DEFAULT_PACKET_OFFSET;
924 memset(DATA(&packet), 0, 14);
925 randomize(DATA(&packet) + 14, len - 14);
929 logger(DEBUG_TRAFFIC, LOG_INFO, "Sending UDP probe length %d to %s (%s)", len, n->name, n->hostname);
931 send_udppacket(n, &packet);
934 // This function tries to establish a UDP tunnel to a node so that packets can be sent.
935 // If a tunnel is already established, it makes sure it stays up.
936 // This function makes no guarantees - it is up to the caller to check the node's state to figure out if UDP is usable.
937 static void try_udp(node_t* n) {
941 /* Send gratuitous probe replies to 1.1 nodes. */
943 if((n->options >> 24) >= 3 && n->status.udp_confirmed) {
944 struct timeval ping_tx_elapsed;
945 timersub(&now, &n->udp_reply_sent, &ping_tx_elapsed);
947 if(ping_tx_elapsed.tv_sec >= udp_discovery_keepalive_interval - 1) {
948 n->udp_reply_sent = now;
949 if(n->maxrecentlen) {
951 pkt.len = n->maxrecentlen;
952 pkt.offset = DEFAULT_PACKET_OFFSET;
953 memset(DATA(&pkt), 0, 14);
954 randomize(DATA(&pkt) + 14, MIN_PROBE_SIZE - 14);
955 send_udp_probe_reply(n, &pkt, pkt.len);
963 struct timeval ping_tx_elapsed;
964 timersub(&now, &n->udp_ping_sent, &ping_tx_elapsed);
966 int interval = n->status.udp_confirmed ? udp_discovery_keepalive_interval : udp_discovery_interval;
968 if(ping_tx_elapsed.tv_sec >= interval) {
969 send_udp_probe_packet(n, MIN_PROBE_SIZE);
970 n->udp_ping_sent = now;
972 if(localdiscovery && !n->status.udp_confirmed && n->prevedge) {
973 n->status.send_locally = true;
974 send_udp_probe_packet(n, MIN_PROBE_SIZE);
975 n->status.send_locally = false;
980 static length_t choose_initial_maxmtu(node_t *n) {
985 const sockaddr_t *sa = NULL;
987 choose_udp_address(n, &sa, &sockindex);
991 sock = socket(sa->sa.sa_family, SOCK_DGRAM, IPPROTO_UDP);
993 logger(DEBUG_TRAFFIC, LOG_ERR, "Creating MTU assessment socket for %s (%s) failed: %s", n->name, n->hostname, sockstrerror(sockerrno));
997 if(connect(sock, &sa->sa, SALEN(sa->sa))) {
998 logger(DEBUG_TRAFFIC, LOG_ERR, "Connecting MTU assessment socket for %s (%s) failed: %s", n->name, n->hostname, sockstrerror(sockerrno));
1004 socklen_t ip_mtu_len = sizeof ip_mtu;
1005 if(getsockopt(sock, IPPROTO_IP, IP_MTU, &ip_mtu, &ip_mtu_len)) {
1006 logger(DEBUG_TRAFFIC, LOG_ERR, "getsockopt(IP_MTU) on %s (%s) failed: %s", n->name, n->hostname, sockstrerror(sockerrno));
1013 /* getsockopt(IP_MTU) returns the MTU of the physical interface.
1014 We need to remove various overheads to get to the tinc MTU. */
1015 length_t mtu = ip_mtu;
1016 mtu -= (sa->sa.sa_family == AF_INET6) ? sizeof(struct ip6_hdr) : sizeof(struct ip);
1018 if(n->status.sptps) {
1019 mtu -= SPTPS_DATAGRAM_OVERHEAD;
1020 if((n->options >> 24) >= 4)
1021 mtu -= sizeof(node_id_t) + sizeof(node_id_t);
1022 #ifndef DISABLE_LEGACY
1024 mtu -= digest_length(n->outdigest);
1026 /* Now it's tricky. We use CBC mode, so the length of the
1027 encrypted payload must be a multiple of the blocksize. The
1028 sequence number is also part of the encrypted payload, so we
1029 must account for it after correcting for the blocksize.
1030 Furthermore, the padding in the last block must be at least
1033 length_t blocksize = cipher_blocksize(n->outcipher);
1046 logger(DEBUG_TRAFFIC, LOG_ERR, "getsockopt(IP_MTU) on %s (%s) returned absurdly small value: %d", n->name, n->hostname, ip_mtu);
1052 logger(DEBUG_TRAFFIC, LOG_INFO, "Using system-provided maximum tinc MTU for %s (%s): %hd", n->name, n->hostname, mtu);
1062 /* This function tries to determines the MTU of a node.
1063 By calling this function repeatedly, n->minmtu will be progressively
1064 increased, and at some point, n->mtu will be fixed to n->minmtu. If the MTU
1065 is already fixed, this function checks if it can be increased.
1068 static void try_mtu(node_t *n) {
1069 if(!(n->options & OPTION_PMTU_DISCOVERY))
1072 if(udp_discovery && !n->status.udp_confirmed) {
1073 n->maxrecentlen = 0;
1080 /* mtuprobes == 0..19: initial discovery, send bursts with 1 second interval, mtuprobes++
1081 mtuprobes == 20: fix MTU, and go to -1
1082 mtuprobes == -1: send one maxmtu and one maxmtu+1 probe every pinginterval
1083 mtuprobes ==-2..-3: send one maxmtu probe every second
1084 mtuprobes == -4: maxmtu no longer valid, reset minmtu and maxmtu and go to 0 */
1086 struct timeval elapsed;
1087 timersub(&now, &n->mtu_ping_sent, &elapsed);
1088 if(n->mtuprobes >= 0) {
1089 if(n->mtuprobes != 0 && elapsed.tv_sec == 0 && elapsed.tv_usec < 333333)
1092 if(n->mtuprobes < -1) {
1093 if(elapsed.tv_sec < 1)
1096 if(elapsed.tv_sec < pinginterval)
1101 n->mtu_ping_sent = now;
1105 if(n->mtuprobes < -3) {
1106 /* We lost three MTU probes, restart discovery */
1107 logger(DEBUG_TRAFFIC, LOG_INFO, "Decrease in PMTU to %s (%s) detected, restarting PMTU discovery", n->name, n->hostname);
1112 if(n->mtuprobes < 0) {
1113 /* After the initial discovery, we only send one maxmtu and one
1114 maxmtu+1 probe to detect PMTU increases. */
1115 send_udp_probe_packet(n, n->maxmtu);
1116 if(n->mtuprobes == -1 && n->maxmtu + 1 < MTU)
1117 send_udp_probe_packet(n, n->maxmtu + 1);
1120 /* Before initial discovery begins, set maxmtu to the most likely value.
1121 If it's underestimated, we will correct it after initial discovery. */
1122 if(n->mtuprobes == 0)
1123 n->maxmtu = choose_initial_maxmtu(n);
1126 /* Decreasing the number of probes per cycle might make the algorithm react faster to lost packets,
1127 but it will typically increase convergence time in the no-loss case. */
1128 const length_t probes_per_cycle = 8;
1130 /* This magic value was determined using math simulations.
1131 It will result in a 1329-byte first probe, followed (if there was a reply) by a 1407-byte probe.
1132 Since 1407 is just below the range of tinc MTUs over typical networks,
1133 this fine-tuning allows tinc to cover a lot of ground very quickly.
1134 This fine-tuning is only valid for maxmtu = MTU; if maxmtu is smaller,
1135 then it's better to use a multiplier of 1. Indeed, this leads to an interesting scenario
1136 if choose_initial_maxmtu() returns the actual MTU value - it will get confirmed with one single probe. */
1137 const float multiplier = (n->maxmtu == MTU) ? 0.97 : 1;
1139 const float cycle_position = probes_per_cycle - (n->mtuprobes % probes_per_cycle) - 1;
1140 const length_t minmtu = MAX(n->minmtu, 512);
1141 const float interval = n->maxmtu - minmtu;
1143 /* The core of the discovery algorithm is this exponential.
1144 It produces very large probes early in the cycle, and then it very quickly decreases the probe size.
1145 This reflects the fact that in the most difficult cases, we don't get any feedback for probes that
1146 are too large, and therefore we need to concentrate on small offsets so that we can quickly converge
1147 on the precise MTU as we are approaching it.
1148 The last probe of the cycle is always 1 byte in size - this is to make sure we'll get at least one
1149 reply per cycle so that we can make progress. */
1150 const length_t offset = powf(interval, multiplier * cycle_position / (probes_per_cycle - 1));
1152 length_t maxmtu = n->maxmtu;
1153 send_udp_probe_packet(n, minmtu + offset);
1154 /* If maxmtu changed, it means the probe was rejected by the system because it was too large.
1155 In that case, we recalculate with the new maxmtu and try again. */
1156 if(n->mtuprobes < 0 || maxmtu == n->maxmtu)
1160 if(n->mtuprobes >= 0)
1165 /* These functions try to establish a tunnel to a node (or its relay) so that
1166 packets can be sent (e.g. exchange keys).
1167 If a tunnel is already established, it tries to improve it (e.g. by trying
1168 to establish a UDP tunnel instead of TCP). This function makes no
1169 guarantees - it is up to the caller to check the node's state to figure out
1170 if TCP and/or UDP is usable. By calling this function repeatedly, the
1171 tunnel is gradually improved until we hit the wall imposed by the underlying
1172 network environment. It is recommended to call this function every time a
1173 packet is sent (or intended to be sent) to a node, so that the tunnel keeps
1174 improving as packets flow, and then gracefully downgrades itself as it goes
1178 static void try_tx_sptps(node_t *n, bool mtu) {
1179 /* If n is a TCP-only neighbor, we'll only use "cleartext" PACKET
1180 messages anyway, so there's no need for SPTPS at all. */
1182 if(n->connection && ((myself->options | n->options) & OPTION_TCPONLY))
1185 /* Otherwise, try to do SPTPS authentication with n if necessary. */
1189 /* Do we need to statically relay packets? */
1191 node_t *via = (n->via == myself) ? n->nexthop : n->via;
1193 /* If we do have a static relay, try everything with that one instead, if it supports relaying. */
1196 if((via->options >> 24) < 4)
1198 return try_tx_sptps(via, mtu);
1201 /* Otherwise, try to establish UDP connectivity. */
1207 /* If we don't have UDP connectivity (yet), we need to use a dynamic relay (nexthop)
1208 while we try to establish direct connectivity. */
1210 if(!n->status.udp_confirmed && n != n->nexthop && (n->nexthop->options >> 24) >= 4)
1211 try_tx_sptps(n->nexthop, mtu);
1214 static void try_tx_legacy(node_t *n, bool mtu) {
1215 /* Does he have our key? If not, send one. */
1217 if(!n->status.validkey_in)
1220 /* Check if we already have a key, or request one. */
1222 if(!n->status.validkey) {
1223 if(n->last_req_key + 10 <= now.tv_sec) {
1225 n->last_req_key = now.tv_sec;
1235 void try_tx(node_t *n, bool mtu) {
1237 try_tx_sptps(n, mtu);
1239 try_tx_legacy(n, mtu);
1242 void send_packet(node_t *n, vpn_packet_t *packet) {
1243 // If it's for myself, write it to the tun/tap device.
1247 memcpy(DATA(packet), mymac.x, ETH_ALEN);
1249 n->out_bytes += packet->len;
1250 devops.write(packet);
1254 logger(DEBUG_TRAFFIC, LOG_ERR, "Sending packet of %d bytes to %s (%s)", packet->len, n->name, n->hostname);
1256 // If the node is not reachable, drop it.
1258 if(!n->status.reachable) {
1259 logger(DEBUG_TRAFFIC, LOG_INFO, "Node %s (%s) is not reachable", n->name, n->hostname);
1263 // Keep track of packet statistics.
1266 n->out_bytes += packet->len;
1268 // Check if it should be sent as an SPTPS packet.
1270 if(n->status.sptps) {
1271 send_sptps_packet(n, packet);
1272 try_tx_sptps(n, true);
1276 // Determine which node to actually send it to.
1278 node_t *via = (packet->priority == -1 || n->via == myself) ? n->nexthop : n->via;
1281 logger(DEBUG_TRAFFIC, LOG_INFO, "Sending packet to %s via %s (%s)", n->name, via->name, n->via->hostname);
1283 // Try to send via UDP, unless TCP is forced.
1285 if(packet->priority == -1 || ((myself->options | via->options) & OPTION_TCPONLY)) {
1286 if(!send_tcppacket(via->connection, packet))
1287 terminate_connection(via->connection, true);
1291 send_udppacket(via, packet);
1292 try_tx_legacy(via, true);
1295 void broadcast_packet(const node_t *from, vpn_packet_t *packet) {
1296 // Always give ourself a copy of the packet.
1298 send_packet(myself, packet);
1300 // In TunnelServer mode, do not forward broadcast packets.
1301 // The MST might not be valid and create loops.
1302 if(tunnelserver || broadcast_mode == BMODE_NONE)
1305 logger(DEBUG_TRAFFIC, LOG_INFO, "Broadcasting packet of %d bytes from %s (%s)",
1306 packet->len, from->name, from->hostname);
1308 switch(broadcast_mode) {
1309 // In MST mode, broadcast packets travel via the Minimum Spanning Tree.
1310 // This guarantees all nodes receive the broadcast packet, and
1311 // usually distributes the sending of broadcast packets over all nodes.
1313 for list_each(connection_t, c, connection_list)
1314 if(c->edge && c->status.mst && c != from->nexthop->connection)
1315 send_packet(c->node, packet);
1318 // In direct mode, we send copies to each node we know of.
1319 // However, this only reaches nodes that can be reached in a single hop.
1320 // We don't have enough information to forward broadcast packets in this case.
1325 for splay_each(node_t, n, node_tree)
1326 if(n->status.reachable && n != myself && ((n->via == myself && n->nexthop == n) || n->via == n))
1327 send_packet(n, packet);
1335 /* We got a packet from some IP address, but we don't know who sent it. Try to
1336 verify the message authentication code against all active session keys.
1337 Since this is actually an expensive operation, we only do a full check once
1338 a minute, the rest of the time we only check against nodes for which we know
1339 an IP address that matches the one from the packet. */
1341 static node_t *try_harder(const sockaddr_t *from, const vpn_packet_t *pkt) {
1342 node_t *match = NULL;
1344 static time_t last_hard_try = 0;
1346 for splay_each(node_t, n, node_tree) {
1347 if(!n->status.reachable || n == myself)
1350 if(!n->status.validkey_in && !(n->status.sptps && n->sptps.instate))
1355 for splay_each(edge_t, e, n->edge_tree) {
1358 if(!sockaddrcmp_noport(from, &e->reverse->address)) {
1365 if(last_hard_try == now.tv_sec)
1370 if(!try_mac(n, pkt))
1378 last_hard_try = now.tv_sec;
1383 void handle_incoming_vpn_data(void *data, int flags) {
1384 listen_socket_t *ls = data;
1387 node_id_t nullid = {};
1388 sockaddr_t addr = {};
1389 socklen_t addrlen = sizeof addr;
1391 bool direct = false;
1394 int len = recvfrom(ls->udp.fd, DATA(&pkt), MAXSIZE, 0, &addr.sa, &addrlen);
1396 if(len <= 0 || len > MAXSIZE) {
1397 if(!sockwouldblock(sockerrno))
1398 logger(DEBUG_ALWAYS, LOG_ERR, "Receiving packet failed: %s", sockstrerror(sockerrno));
1404 sockaddrunmap(&addr); /* Some braindead IPv6 implementations do stupid things. */
1406 // Try to figure out who sent this packet.
1408 node_t *n = lookup_node_udp(&addr);
1410 if(n && !n->status.udp_confirmed)
1411 n = NULL; // Don't believe it if we don't have confirmation yet.
1414 // It might be from a 1.1 node, which might have a source ID in the packet.
1415 pkt.offset = 2 * sizeof(node_id_t);
1416 from = lookup_node_id(SRCID(&pkt));
1417 if(from && !memcmp(DSTID(&pkt), &nullid, sizeof nullid) && from->status.sptps) {
1418 if(sptps_verify_datagram(&from->sptps, DATA(&pkt), pkt.len - 2 * sizeof(node_id_t)))
1427 n = try_harder(&addr, &pkt);
1432 if(debug_level >= DEBUG_PROTOCOL) {
1433 hostname = sockaddr2hostname(&addr);
1434 logger(DEBUG_PROTOCOL, LOG_WARNING, "Received UDP packet from unknown source %s", hostname);
1442 if(n->status.sptps) {
1443 bool relay_enabled = (n->options >> 24) >= 4;
1444 if (relay_enabled) {
1445 pkt.offset = 2 * sizeof(node_id_t);
1446 pkt.len -= pkt.offset;
1449 if(!memcmp(DSTID(&pkt), &nullid, sizeof nullid) || !relay_enabled) {
1454 from = lookup_node_id(SRCID(&pkt));
1455 to = lookup_node_id(DSTID(&pkt));
1458 logger(DEBUG_PROTOCOL, LOG_WARNING, "Received UDP packet from %s (%s) with unknown source and/or destination ID", n->name, n->hostname);
1462 /* The packet is supposed to come from the originator or its static relay
1463 (i.e. with no dynamic relays in between).
1464 If it did not, "help" the static relay by sending it UDP info.
1465 Note that we only do this if we're the destination or the static relay;
1466 otherwise every hop would initiate its own UDP info message, resulting in elevated chatter. */
1468 if(n != from->via && to->via == myself)
1469 send_udp_info(myself, from);
1471 /* If we're not the final recipient, relay the packet. */
1474 send_sptps_data(to, from, 0, DATA(&pkt), pkt.len);
1475 try_tx_sptps(to, true);
1483 if(!receive_udppacket(from, &pkt))
1486 n->sock = ls - listen_socket;
1487 if(direct && sockaddrcmp(&addr, &n->address))
1488 update_node_udp(n, &addr);
1490 /* If the packet went through a relay, help the sender find the appropriate MTU
1491 through the relay path. */
1494 send_mtu_info(myself, n, MTU);
1497 void handle_device_data(void *data, int flags) {
1498 vpn_packet_t packet;
1499 packet.offset = DEFAULT_PACKET_OFFSET;
1500 packet.priority = 0;
1502 if(devops.read(&packet)) {
1503 myself->in_packets++;
1504 myself->in_bytes += packet.len;
1505 route(myself, &packet);