2 net_packet.c -- Handles in- and outgoing VPN packets
3 Copyright (C) 1998-2005 Ivo Timmermans,
4 2000-2022 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.
40 #include "address_cache.h"
43 #include "connection.h"
44 #include "compression.h"
58 /* The minimum size of a probe is 14 bytes, but since we normally use CBC mode
59 encryption, we can add a few extra random bytes without increasing the
60 resulting packet size. */
61 #define MIN_PROBE_SIZE 18
65 static char lzo_wrkmem[LZO1X_999_MEM_COMPRESS > LZO1X_1_MEM_COMPRESS ? LZO1X_999_MEM_COMPRESS : LZO1X_1_MEM_COMPRESS];
68 #ifdef HAVE_LZ4_BUILTIN
69 static LZ4_stream_t lz4_stream;
71 static void *lz4_state = NULL;
72 #endif /* HAVE_LZ4_BUILTIN */
74 static void send_udppacket(node_t *, vpn_packet_t *);
76 unsigned replaywin = 32;
77 bool localdiscovery = true;
78 bool udp_discovery = true;
79 int udp_discovery_keepalive_interval = 10;
80 int udp_discovery_interval = 2;
81 int udp_discovery_timeout = 30;
83 #define MAX_SEQNO 1073741824
85 static void try_fix_mtu(node_t *n) {
86 if(n->mtuprobes < 0) {
90 if(n->mtuprobes == 20 || n->minmtu >= n->maxmtu) {
91 if(n->minmtu > n->maxmtu) {
92 n->minmtu = n->maxmtu;
94 n->maxmtu = n->minmtu;
98 logger(DEBUG_TRAFFIC, LOG_INFO, "Fixing MTU of %s (%s) to %d after %d probes", n->name, n->hostname, n->mtu, n->mtuprobes);
103 static void reduce_mtu(node_t *n, int mtu) {
108 if(n->maxmtu > mtu) {
119 static void udp_probe_timeout_handler(void *data) {
122 if(!n->status.udp_confirmed) {
126 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);
127 n->status.udp_confirmed = false;
128 n->udp_ping_rtt = -1;
135 static void send_udp_probe_reply(node_t *n, vpn_packet_t *packet, length_t len) {
136 if(!n->status.sptps && !n->status.validkey) {
137 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);
141 /* Type 2 probe replies were introduced in protocol 17.3 */
142 if((n->options >> 24) >= 3) {
144 uint16_t len16 = htons(len);
145 memcpy(DATA(packet) + 1, &len16, 2);
146 packet->len = MIN_PROBE_SIZE;
147 logger(DEBUG_TRAFFIC, LOG_INFO, "Sending type 2 probe reply length %u to %s (%s)", len, n->name, n->hostname);
150 /* Legacy protocol: n won't understand type 2 probe replies. */
152 logger(DEBUG_TRAFFIC, LOG_INFO, "Sending type 1 probe reply length %u to %s (%s)", len, n->name, n->hostname);
155 /* Temporarily set udp_confirmed, so that the reply is sent
156 back exactly the way it came in. */
158 bool udp_confirmed = n->status.udp_confirmed;
159 n->status.udp_confirmed = true;
160 send_udppacket(n, packet);
161 n->status.udp_confirmed = udp_confirmed;
164 static void udp_probe_h(node_t *n, vpn_packet_t *packet, length_t len) {
165 if(!DATA(packet)[0]) {
166 logger(DEBUG_TRAFFIC, LOG_INFO, "Got UDP probe request %d from %s (%s)", packet->len, n->name, n->hostname);
167 send_udp_probe_reply(n, packet, len);
171 if(DATA(packet)[0] == 2) {
172 // It's a type 2 probe reply, use the length field inside the packet
174 memcpy(&len16, DATA(packet) + 1, 2);
178 if(n->status.ping_sent) { // a probe in flight
179 gettimeofday(&now, NULL);
181 timersub(&now, &n->udp_ping_sent, &rtt);
182 n->udp_ping_rtt = (int)(rtt.tv_sec * 1000000 + rtt.tv_usec);
183 n->status.ping_sent = false;
184 logger(DEBUG_TRAFFIC, LOG_INFO, "Got type %d UDP probe reply %d from %s (%s) rtt=%d.%03d", DATA(packet)[0], len, n->name, n->hostname, n->udp_ping_rtt / 1000, n->udp_ping_rtt % 1000);
186 logger(DEBUG_TRAFFIC, LOG_INFO, "Got type %d UDP probe reply %d from %s (%s)", DATA(packet)[0], len, n->name, n->hostname);
189 /* It's a valid reply: now we know bidirectional communication
190 is possible using the address and socket that the reply
192 if(!n->status.udp_confirmed) {
193 n->status.udp_confirmed = true;
195 if(!n->address_cache) {
196 n->address_cache = open_address_cache(n);
199 reset_address_cache(n->address_cache, &n->address);
202 // Reset the UDP ping timer.
205 timeout_del(&n->udp_ping_timeout);
206 timeout_add(&n->udp_ping_timeout, &udp_probe_timeout_handler, n, &(struct timeval) {
207 udp_discovery_timeout, 0
211 if(len > n->maxmtu) {
212 logger(DEBUG_TRAFFIC, LOG_INFO, "Increase in PMTU to %s (%s) detected, restarting PMTU discovery", n->name, n->hostname);
215 /* Set mtuprobes to 1 so that try_mtu() doesn't reset maxmtu */
218 } else if(n->mtuprobes < 0 && len == n->maxmtu) {
219 /* We got a maxmtu sized packet, confirming the PMTU is still valid. */
221 n->mtu_ping_sent = now;
224 /* If applicable, raise the minimum supported MTU */
226 if(n->minmtu < len) {
233 static length_t compress_packet_lz4(uint8_t *dest, const uint8_t *source, length_t len) {
234 #ifdef HAVE_LZ4_BUILTIN
235 return LZ4_compress_fast_extState(&lz4_stream, (const char *) source, (char *) dest, len, MAXSIZE, 0);
238 /* @FIXME: Put this in a better place, and free() it too. */
239 if(lz4_state == NULL) {
240 lz4_state = malloc(LZ4_sizeofState());
243 if(lz4_state == NULL) {
244 logger(DEBUG_ALWAYS, LOG_ERR, "Failed to allocate lz4_state, error: %i", errno);
248 return LZ4_compress_fast_extState(lz4_state, (const char *) source, (char *) dest, len, MAXSIZE, 0);
249 #endif /* HAVE_LZ4_BUILTIN */
251 #endif /* HAVE_LZ4 */
254 static length_t compress_packet_lzo(uint8_t *dest, const uint8_t *source, length_t len, compression_level_t level) {
255 assert(level == COMPRESS_LZO_LO || level == COMPRESS_LZO_HI);
257 lzo_uint lzolen = MAXSIZE;
260 if(level == COMPRESS_LZO_HI) {
261 result = lzo1x_999_compress(source, len, dest, &lzolen, lzo_wrkmem);
262 } else { // level == COMPRESS_LZO_LO
263 result = lzo1x_1_compress(source, len, dest, &lzolen, lzo_wrkmem);
266 if(result == LZO_E_OK) {
274 static length_t compress_packet(uint8_t *dest, const uint8_t *source, length_t len, compression_level_t level) {
279 return compress_packet_lz4(dest, source, len);
284 case COMPRESS_LZO_HI:
285 case COMPRESS_LZO_LO:
286 return compress_packet_lzo(dest, source, len, level);
290 case COMPRESS_ZLIB_9:
291 case COMPRESS_ZLIB_8:
292 case COMPRESS_ZLIB_7:
293 case COMPRESS_ZLIB_6:
294 case COMPRESS_ZLIB_5:
295 case COMPRESS_ZLIB_4:
296 case COMPRESS_ZLIB_3:
297 case COMPRESS_ZLIB_2:
298 case COMPRESS_ZLIB_1: {
299 unsigned long dest_len = MAXSIZE;
301 if(compress2(dest, &dest_len, source, len, level) == Z_OK) {
311 memcpy(dest, source, len);
319 static length_t uncompress_packet(uint8_t *dest, const uint8_t *source, length_t len, compression_level_t level) {
324 return LZ4_decompress_safe((char *)source, (char *) dest, len, MAXSIZE);
329 case COMPRESS_LZO_HI:
330 case COMPRESS_LZO_LO: {
331 lzo_uint dst_len = MAXSIZE;
333 if(lzo1x_decompress_safe(source, len, dest, &dst_len, NULL) == LZO_E_OK) {
343 case COMPRESS_ZLIB_9:
344 case COMPRESS_ZLIB_8:
345 case COMPRESS_ZLIB_7:
346 case COMPRESS_ZLIB_6:
347 case COMPRESS_ZLIB_5:
348 case COMPRESS_ZLIB_4:
349 case COMPRESS_ZLIB_3:
350 case COMPRESS_ZLIB_2:
351 case COMPRESS_ZLIB_1: {
352 unsigned long destlen = MAXSIZE;
353 static z_stream stream;
356 inflateReset(&stream);
358 inflateInit(&stream);
361 stream.next_in = source;
362 stream.avail_in = len;
363 stream.next_out = dest;
364 stream.avail_out = destlen;
365 stream.total_out = 0;
367 if(inflate(&stream, Z_FINISH) == Z_STREAM_END) {
368 return stream.total_out;
377 memcpy(dest, source, len);
387 static void receive_packet(node_t *n, vpn_packet_t *packet) {
388 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Received packet of %d bytes from %s (%s)",
389 packet->len, n->name, n->hostname);
392 n->in_bytes += packet->len;
397 static bool try_mac(node_t *n, const vpn_packet_t *inpkt) {
398 if(n->status.sptps) {
399 return sptps_verify_datagram(&n->sptps, DATA(inpkt), inpkt->len);
402 #ifdef DISABLE_LEGACY
406 if(!n->status.validkey_in || !digest_active(n->indigest) || (size_t)inpkt->len < sizeof(seqno_t) + digest_length(n->indigest)) {
410 return digest_verify(n->indigest, inpkt->data, inpkt->len - digest_length(n->indigest), inpkt->data + inpkt->len - digest_length(n->indigest));
414 static bool receive_udppacket(node_t *n, vpn_packet_t *inpkt) {
415 if(n->status.sptps) {
416 if(!n->sptps.state) {
417 if(!n->status.waitingforkey) {
418 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got packet from %s (%s) but we haven't exchanged keys yet", n->name, n->hostname);
421 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got packet from %s (%s) but he hasn't got our key yet", n->name, n->hostname);
427 n->status.udppacket = true;
428 bool result = sptps_receive_data(&n->sptps, DATA(inpkt), inpkt->len);
429 n->status.udppacket = false;
432 /* Uh-oh. It might be that the tunnel is stuck in some corrupted state,
433 so let's restart SPTPS in case that helps. But don't do that too often
434 to prevent storms, and because that would make life a little too easy
435 for external attackers trying to DoS us. */
436 if(n->last_req_key < now.tv_sec - 10) {
437 logger(DEBUG_PROTOCOL, LOG_ERR, "Failed to decode raw TCP packet from %s (%s), restarting SPTPS", n->name, n->hostname);
447 #ifdef DISABLE_LEGACY
450 vpn_packet_t pkt1, pkt2;
451 vpn_packet_t *pkt[] = { &pkt1, &pkt2, &pkt1, &pkt2 };
454 pkt1.offset = DEFAULT_PACKET_OFFSET;
455 pkt2.offset = DEFAULT_PACKET_OFFSET;
457 if(!n->status.validkey_in) {
458 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got packet from %s (%s) but he hasn't got our key yet", n->name, n->hostname);
462 /* Check packet length */
464 if((size_t)inpkt->len < sizeof(seqno_t) + digest_length(n->indigest)) {
465 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got too short packet from %s (%s)",
466 n->name, n->hostname);
470 /* It's a legacy UDP packet, the data starts after the seqno */
472 inpkt->offset += sizeof(seqno_t);
474 /* Check the message authentication code */
476 if(digest_active(n->indigest)) {
477 inpkt->len -= digest_length(n->indigest);
479 if(!digest_verify(n->indigest, SEQNO(inpkt), inpkt->len, SEQNO(inpkt) + inpkt->len)) {
480 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got unauthenticated packet from %s (%s)", n->name, n->hostname);
485 /* Decrypt the packet */
487 if(cipher_active(n->incipher)) {
488 vpn_packet_t *outpkt = pkt[nextpkt++];
491 if(!cipher_decrypt(n->incipher, SEQNO(inpkt), inpkt->len, SEQNO(outpkt), &outlen, true)) {
492 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Error decrypting packet from %s (%s)", n->name, n->hostname);
496 outpkt->len = outlen;
500 /* Check the sequence number */
503 memcpy(&seqno, SEQNO(inpkt), sizeof(seqno));
504 seqno = ntohl(seqno);
505 inpkt->len -= sizeof(seqno);
508 if(seqno != n->received_seqno + 1) {
509 if(seqno >= n->received_seqno + replaywin * 8) {
510 if(n->farfuture++ < replaywin >> 2) {
511 logger(DEBUG_TRAFFIC, LOG_WARNING, "Packet from %s (%s) is %d seqs in the future, dropped (%u)",
512 n->name, n->hostname, seqno - n->received_seqno - 1, n->farfuture);
516 logger(DEBUG_TRAFFIC, LOG_WARNING, "Lost %d packets from %s (%s)",
517 seqno - n->received_seqno - 1, n->name, n->hostname);
518 memset(n->late, 0, replaywin);
519 } else if(seqno <= n->received_seqno) {
520 if((n->received_seqno >= replaywin * 8 && seqno <= n->received_seqno - replaywin * 8) || !(n->late[(seqno / 8) % replaywin] & (1 << seqno % 8))) {
521 logger(DEBUG_TRAFFIC, LOG_WARNING, "Got late or replayed packet from %s (%s), seqno %d, last received %d",
522 n->name, n->hostname, seqno, n->received_seqno);
526 for(seqno_t i = n->received_seqno + 1; i < seqno; i++) {
527 n->late[(i / 8) % replaywin] |= 1 << i % 8;
533 n->late[(seqno / 8) % replaywin] &= ~(1 << seqno % 8);
536 if(seqno > n->received_seqno) {
537 n->received_seqno = seqno;
542 if(n->received_seqno > MAX_SEQNO) {
546 /* Decompress the packet */
548 length_t origlen = inpkt->len;
550 if(n->incompression != COMPRESS_NONE) {
551 vpn_packet_t *outpkt = pkt[nextpkt++];
553 if(!(outpkt->len = uncompress_packet(DATA(outpkt), DATA(inpkt), inpkt->len, n->incompression))) {
554 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while uncompressing packet from %s (%s)",
555 n->name, n->hostname);
561 if(origlen > MTU / 64 + 20) {
562 origlen -= MTU / 64 + 20;
568 if(inpkt->len > n->maxrecentlen) {
569 n->maxrecentlen = inpkt->len;
574 if(!DATA(inpkt)[12] && !DATA(inpkt)[13]) {
575 udp_probe_h(n, inpkt, origlen);
577 receive_packet(n, inpkt);
584 void receive_tcppacket(connection_t *c, const char *buffer, size_t len) {
586 outpkt.offset = DEFAULT_PACKET_OFFSET;
588 if(len > sizeof(outpkt.data) - outpkt.offset) {
594 if(c->options & OPTION_TCPONLY) {
597 outpkt.priority = -1;
600 memcpy(DATA(&outpkt), buffer, len);
602 receive_packet(c->node, &outpkt);
605 bool receive_tcppacket_sptps(connection_t *c, const char *data, size_t len) {
606 if(len < sizeof(node_id_t) + sizeof(node_id_t)) {
607 logger(DEBUG_PROTOCOL, LOG_ERR, "Got too short TCP SPTPS packet from %s (%s)", c->name, c->hostname);
611 node_t *to = lookup_node_id((node_id_t *)data);
612 data += sizeof(node_id_t);
613 len -= sizeof(node_id_t);
616 logger(DEBUG_PROTOCOL, LOG_ERR, "Got TCP SPTPS packet from %s (%s) with unknown destination ID", c->name, c->hostname);
620 node_t *from = lookup_node_id((node_id_t *)data);
621 data += sizeof(node_id_t);
622 len -= sizeof(node_id_t);
625 logger(DEBUG_PROTOCOL, LOG_ERR, "Got TCP SPTPS packet from %s (%s) with unknown source ID", c->name, c->hostname);
629 if(!to->status.reachable) {
630 /* This can happen in the form of a race condition
631 if the node just became unreachable. */
632 logger(DEBUG_TRAFFIC, LOG_WARNING, "Cannot relay TCP packet from %s (%s) because the destination, %s (%s), is unreachable", from->name, from->hostname, to->name, to->hostname);
636 /* Help the sender reach us over UDP.
637 Note that we only do this if we're the destination or the static relay;
638 otherwise every hop would initiate its own UDP info message, resulting in elevated chatter. */
639 if(to->via == myself) {
640 send_udp_info(myself, from);
643 /* If we're not the final recipient, relay the packet. */
646 if(to->status.validkey) {
647 send_sptps_data(to, from, 0, data, len);
654 /* The packet is for us */
656 if(!sptps_receive_data(&from->sptps, data, len)) {
657 /* Uh-oh. It might be that the tunnel is stuck in some corrupted state,
658 so let's restart SPTPS in case that helps. But don't do that too often
659 to prevent storms. */
660 if(from->last_req_key < now.tv_sec - 10) {
661 logger(DEBUG_PROTOCOL, LOG_ERR, "Failed to decode raw TCP packet from %s (%s), restarting SPTPS", from->name, from->hostname);
668 send_mtu_info(myself, from, MTU);
672 static void send_sptps_packet(node_t *n, vpn_packet_t *origpkt) {
673 if(!n->status.validkey && !n->connection) {
680 if((!(DATA(origpkt)[12] | DATA(origpkt)[13])) && (n->sptps.outstate)) {
681 sptps_send_record(&n->sptps, PKT_PROBE, DATA(origpkt), origpkt->len);
685 if(routing_mode == RMODE_ROUTER) {
691 if(origpkt->len < offset) {
697 if(n->outcompression != COMPRESS_NONE) {
699 length_t len = compress_packet(DATA(&outpkt) + offset, DATA(origpkt) + offset, origpkt->len - offset, n->outcompression);
702 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while compressing packet to %s (%s)", n->name, n->hostname);
703 } else if(len < origpkt->len - offset) {
704 outpkt.len = len + offset;
706 type |= PKT_COMPRESSED;
710 /* If we have a direct metaconnection to n, and we can't use UDP, then
711 don't bother with SPTPS and just use a "plaintext" PACKET message.
712 We don't really care about end-to-end security since we're not
713 sending the message through any intermediate nodes. */
714 if(n->connection && origpkt->len > n->minmtu) {
715 send_tcppacket(n->connection, origpkt);
717 sptps_send_record(&n->sptps, type, DATA(origpkt) + offset, origpkt->len - offset);
721 static void adapt_socket(const sockaddr_t *sa, size_t *sock) {
722 /* Make sure we have a suitable socket for the chosen address */
723 if(listen_socket[*sock].sa.sa.sa_family != sa->sa.sa_family) {
724 for(int i = 0; i < listen_sockets; i++) {
725 if(listen_socket[i].sa.sa.sa_family == sa->sa.sa_family) {
733 static void choose_udp_address(const node_t *n, const sockaddr_t **sa, size_t *sock) {
738 /* If the UDP address is confirmed, use it. */
739 if(n->status.udp_confirmed) {
743 /* Send every third packet to n->address; that could be set
744 to the node's reflexive UDP address discovered during key
754 /* Otherwise, address are found in edges to this node.
755 So we pick a random edge and a random socket. */
758 unsigned int j = prng(n->edge_tree.count);
759 edge_t *candidate = NULL;
761 for splay_each(edge_t, e, &n->edge_tree) {
763 candidate = e->reverse;
769 *sa = &candidate->address;
770 *sock = prng(listen_sockets);
773 adapt_socket(*sa, sock);
776 static void choose_local_address(const node_t *n, const sockaddr_t **sa, size_t *sock) {
779 /* Pick one of the edges from this node at random, then use its local address. */
782 unsigned int j = prng(n->edge_tree.count);
783 edge_t *candidate = NULL;
785 for splay_each(edge_t, e, &n->edge_tree) {
792 if(candidate && candidate->local_address.sa.sa_family) {
793 *sa = &candidate->local_address;
794 *sock = prng(listen_sockets);
795 adapt_socket(*sa, sock);
799 static void send_udppacket(node_t *n, vpn_packet_t *origpkt) {
800 if(!n->status.reachable) {
801 logger(DEBUG_TRAFFIC, LOG_INFO, "Trying to send UDP packet to unreachable node %s (%s)", n->name, n->hostname);
805 if(n->status.sptps) {
806 send_sptps_packet(n, origpkt);
810 #ifdef DISABLE_LEGACY
813 vpn_packet_t pkt1, pkt2;
814 vpn_packet_t *pkt[] = { &pkt1, &pkt2, &pkt1, &pkt2 };
815 vpn_packet_t *inpkt = origpkt;
817 vpn_packet_t *outpkt;
818 int origlen = origpkt->len;
820 int origpriority = origpkt->priority;
822 pkt1.offset = DEFAULT_PACKET_OFFSET;
823 pkt2.offset = DEFAULT_PACKET_OFFSET;
825 /* Make sure we have a valid key */
827 if(!n->status.validkey) {
828 logger(DEBUG_TRAFFIC, LOG_INFO,
829 "No valid key known yet for %s (%s), forwarding via TCP",
830 n->name, n->hostname);
831 send_tcppacket(n->nexthop->connection, origpkt);
835 if(n->options & OPTION_PMTU_DISCOVERY && inpkt->len > n->minmtu && (DATA(inpkt)[12] | DATA(inpkt)[13])) {
836 logger(DEBUG_TRAFFIC, LOG_INFO,
837 "Packet for %s (%s) larger than minimum MTU, forwarding via %s",
838 n->name, n->hostname, n != n->nexthop ? n->nexthop->name : "TCP");
840 if(n != n->nexthop) {
841 send_packet(n->nexthop, origpkt);
843 send_tcppacket(n->nexthop->connection, origpkt);
849 /* Compress the packet */
851 if(n->outcompression != COMPRESS_NONE) {
852 outpkt = pkt[nextpkt++];
854 if(!(outpkt->len = compress_packet(DATA(outpkt), DATA(inpkt), inpkt->len, n->outcompression))) {
855 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while compressing packet to %s (%s)",
856 n->name, n->hostname);
863 /* Add sequence number */
865 seqno_t seqno = htonl(++(n->sent_seqno));
866 memcpy(SEQNO(inpkt), &seqno, sizeof(seqno));
867 inpkt->len += sizeof(seqno);
869 /* Encrypt the packet */
871 if(cipher_active(n->outcipher)) {
872 outpkt = pkt[nextpkt++];
875 if(!cipher_encrypt(n->outcipher, SEQNO(inpkt), inpkt->len, SEQNO(outpkt), &outlen, true)) {
876 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while encrypting packet to %s (%s)", n->name, n->hostname);
880 outpkt->len = outlen;
884 /* Add the message authentication code */
886 if(digest_active(n->outdigest)) {
887 if(!digest_create(n->outdigest, SEQNO(inpkt), inpkt->len, SEQNO(inpkt) + inpkt->len)) {
888 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while encrypting packet to %s (%s)", n->name, n->hostname);
892 inpkt->len += digest_length(n->outdigest);
895 /* Send the packet */
897 const sockaddr_t *sa = NULL;
900 if(n->status.send_locally) {
901 choose_local_address(n, &sa, &sock);
905 choose_udp_address(n, &sa, &sock);
908 if(priorityinheritance && origpriority != listen_socket[sock].priority) {
909 listen_socket[sock].priority = origpriority;
911 switch(sa->sa.sa_family) {
915 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Setting IPv4 outgoing packet priority to %d", origpriority);
917 if(setsockopt(listen_socket[sock].udp.fd, IPPROTO_IP, IP_TOS, (void *)&origpriority, sizeof(origpriority))) { /* SO_PRIORITY doesn't seem to work */
918 logger(DEBUG_ALWAYS, LOG_ERR, "System call `%s' failed: %s", "setsockopt", sockstrerror(sockerrno));
923 #if defined(IPV6_TCLASS)
926 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Setting IPv6 outgoing packet priority to %d", origpriority);
928 if(setsockopt(listen_socket[sock].udp.fd, IPPROTO_IPV6, IPV6_TCLASS, (void *)&origpriority, sizeof(origpriority))) { /* SO_PRIORITY doesn't seem to work */
929 logger(DEBUG_ALWAYS, LOG_ERR, "System call `%s' failed: %s", "setsockopt", sockstrerror(sockerrno));
940 if(sendto(listen_socket[sock].udp.fd, (void *)SEQNO(inpkt), inpkt->len, 0, &sa->sa, SALEN(sa->sa)) < 0 && !sockwouldblock(sockerrno)) {
941 if(sockmsgsize(sockerrno)) {
942 reduce_mtu(n, origlen - 1);
944 logger(DEBUG_TRAFFIC, LOG_WARNING, "Error sending packet to %s (%s): %s", n->name, n->hostname, sockstrerror(sockerrno));
949 origpkt->len = origlen;
953 bool send_sptps_data(node_t *to, node_t *from, int type, const void *data, size_t len) {
954 size_t origlen = len - SPTPS_DATAGRAM_OVERHEAD;
955 node_t *relay = (to->via != myself && (type == PKT_PROBE || origlen <= to->via->minmtu)) ? to->via : to->nexthop;
956 bool direct = from == myself && to == relay;
957 bool relay_supported = (relay->options >> 24) >= 4;
958 bool tcponly = (myself->options | relay->options) & OPTION_TCPONLY;
960 /* 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. */
962 if(type == SPTPS_HANDSHAKE || tcponly || (!direct && !relay_supported) || (type != PKT_PROBE && origlen > relay->minmtu)) {
963 if(type != SPTPS_HANDSHAKE && (to->nexthop->connection->options >> 24) >= 7) {
964 uint8_t buf[len + sizeof(to->id) + sizeof(from->id)];
965 uint8_t *buf_ptr = buf;
966 memcpy(buf_ptr, &to->id, sizeof(to->id));
967 buf_ptr += sizeof(to->id);
968 memcpy(buf_ptr, &from->id, sizeof(from->id));
969 buf_ptr += sizeof(from->id);
970 memcpy(buf_ptr, data, len);
971 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);
972 return send_sptps_tcppacket(to->nexthop->connection, buf, sizeof(buf));
975 char buf[B64_SIZE(len)];
976 b64encode_tinc(data, buf, len);
978 /* If this is a handshake packet, use ANS_KEY instead of REQ_KEY, for two reasons:
979 - We don't want intermediate nodes to switch to UDP to relay these packets;
980 - ANS_KEY allows us to learn the reflexive UDP address. */
981 if(type == SPTPS_HANDSHAKE) {
982 to->incompression = myself->incompression;
983 return send_request(to->nexthop->connection, "%d %s %s %s -1 -1 -1 %d", ANS_KEY, from->name, to->name, buf, to->incompression);
985 return send_request(to->nexthop->connection, "%d %s %s %d %s", REQ_KEY, from->name, to->name, SPTPS_PACKET, buf);
991 if(relay_supported) {
992 overhead += sizeof(to->id) + sizeof(from->id);
995 char buf[len + overhead];
998 if(relay_supported) {
1000 /* Inform the recipient that this packet was sent directly. */
1001 node_id_t nullid = {0};
1002 memcpy(buf_ptr, &nullid, sizeof(nullid));
1003 buf_ptr += sizeof(nullid);
1005 memcpy(buf_ptr, &to->id, sizeof(to->id));
1006 buf_ptr += sizeof(to->id);
1009 memcpy(buf_ptr, &from->id, sizeof(from->id));
1010 buf_ptr += sizeof(from->id);
1014 /* 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 */
1015 memcpy(buf_ptr, data, len);
1018 const sockaddr_t *sa = NULL;
1021 if(relay->status.send_locally) {
1022 choose_local_address(relay, &sa, &sock);
1026 choose_udp_address(relay, &sa, &sock);
1029 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);
1031 if(sendto(listen_socket[sock].udp.fd, buf, buf_ptr - buf, 0, &sa->sa, SALEN(sa->sa)) < 0 && !sockwouldblock(sockerrno)) {
1032 if(sockmsgsize(sockerrno)) {
1033 reduce_mtu(relay, (int)origlen - 1);
1035 logger(DEBUG_TRAFFIC, LOG_WARNING, "Error sending UDP SPTPS packet to %s (%s): %s", relay->name, relay->hostname, sockstrerror(sockerrno));
1043 bool receive_sptps_record(void *handle, uint8_t type, const void *data, uint16_t len) {
1044 node_t *from = handle;
1046 if(type == SPTPS_HANDSHAKE) {
1047 if(!from->status.validkey) {
1048 from->status.validkey = true;
1049 from->status.waitingforkey = false;
1050 logger(DEBUG_META, LOG_INFO, "SPTPS key exchange with %s (%s) successful", from->name, from->hostname);
1057 logger(DEBUG_ALWAYS, LOG_ERR, "Packet from %s (%s) larger than maximum supported size (%d > %d)", from->name, from->hostname, len, MTU);
1062 inpkt.offset = DEFAULT_PACKET_OFFSET;
1065 if(type == PKT_PROBE) {
1066 if(!from->status.udppacket) {
1067 logger(DEBUG_ALWAYS, LOG_ERR, "Got SPTPS PROBE packet from %s (%s) via TCP", from->name, from->hostname);
1072 memcpy(DATA(&inpkt), data, len);
1074 if(inpkt.len > from->maxrecentlen) {
1075 from->maxrecentlen = inpkt.len;
1078 udp_probe_h(from, &inpkt, len);
1082 if(type & ~(PKT_COMPRESSED | PKT_MAC)) {
1083 logger(DEBUG_ALWAYS, LOG_ERR, "Unexpected SPTPS record type %d len %d from %s (%s)", type, len, from->name, from->hostname);
1087 /* Check if we have the headers we need */
1088 if(routing_mode != RMODE_ROUTER && !(type & PKT_MAC)) {
1089 logger(DEBUG_TRAFFIC, LOG_ERR, "Received packet from %s (%s) without MAC header (maybe Mode is not set correctly)", from->name, from->hostname);
1091 } else if(routing_mode == RMODE_ROUTER && (type & PKT_MAC)) {
1092 logger(DEBUG_TRAFFIC, LOG_WARNING, "Received packet from %s (%s) with MAC header (maybe Mode is not set correctly)", from->name, from->hostname);
1095 int offset = (type & PKT_MAC) ? 0 : 14;
1097 if(type & PKT_COMPRESSED) {
1098 length_t ulen = uncompress_packet(DATA(&inpkt) + offset, (const uint8_t *)data, len, from->incompression);
1103 inpkt.len = ulen + offset;
1106 if(inpkt.len > MAXSIZE) {
1110 memcpy(DATA(&inpkt) + offset, data, len);
1111 inpkt.len = len + offset;
1114 /* Generate the Ethernet packet type if necessary */
1116 switch(DATA(&inpkt)[14] >> 4) {
1118 DATA(&inpkt)[12] = 0x08;
1119 DATA(&inpkt)[13] = 0x00;
1123 DATA(&inpkt)[12] = 0x86;
1124 DATA(&inpkt)[13] = 0xDD;
1128 logger(DEBUG_TRAFFIC, LOG_ERR,
1129 "Unknown IP version %d while reading packet from %s (%s)",
1130 DATA(&inpkt)[14] >> 4, from->name, from->hostname);
1135 if(from->status.udppacket && inpkt.len > from->maxrecentlen) {
1136 from->maxrecentlen = inpkt.len;
1139 receive_packet(from, &inpkt);
1143 // This function tries to get SPTPS keys, if they aren't already known.
1144 // 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.
1145 static void try_sptps(node_t *n) {
1146 if(n->status.validkey) {
1150 logger(DEBUG_TRAFFIC, LOG_INFO, "No valid key known yet for %s (%s)", n->name, n->hostname);
1152 if(!n->status.waitingforkey) {
1154 } else if(n->last_req_key + 10 < now.tv_sec) {
1155 logger(DEBUG_ALWAYS, LOG_DEBUG, "No key from %s after 10 seconds, restarting SPTPS", n->name);
1156 sptps_stop(&n->sptps);
1157 n->status.waitingforkey = false;
1164 static void send_udp_probe_packet(node_t *n, size_t len) {
1165 vpn_packet_t packet;
1167 if(len > sizeof(packet.data)) {
1168 logger(DEBUG_TRAFFIC, LOG_INFO, "Truncating probe length %lu to %s (%s)", (unsigned long)len, n->name, n->hostname);
1169 len = sizeof(packet.data);
1172 packet.offset = DEFAULT_PACKET_OFFSET;
1173 memset(DATA(&packet), 0, 14);
1174 randomize(DATA(&packet) + 14, len - 14);
1176 packet.priority = 0;
1178 logger(DEBUG_TRAFFIC, LOG_INFO, "Sending UDP probe length %lu to %s (%s)", (unsigned long)len, n->name, n->hostname);
1180 send_udppacket(n, &packet);
1183 // This function tries to establish a UDP tunnel to a node so that packets can be sent.
1184 // If a tunnel is already established, it makes sure it stays up.
1185 // This function makes no guarantees - it is up to the caller to check the node's state to figure out if UDP is usable.
1186 static void try_udp(node_t *n) {
1187 if(!udp_discovery) {
1191 /* Send gratuitous probe replies to 1.1 nodes. */
1193 if((n->options >> 24) >= 3 && n->status.udp_confirmed) {
1194 struct timeval ping_tx_elapsed;
1195 timersub(&now, &n->udp_reply_sent, &ping_tx_elapsed);
1197 if(ping_tx_elapsed.tv_sec >= udp_discovery_keepalive_interval - 1) {
1198 n->udp_reply_sent = now;
1200 if(n->maxrecentlen) {
1202 pkt.len = n->maxrecentlen;
1203 pkt.offset = DEFAULT_PACKET_OFFSET;
1204 memset(DATA(&pkt), 0, 14);
1205 randomize(DATA(&pkt) + 14, MIN_PROBE_SIZE - 14);
1206 send_udp_probe_reply(n, &pkt, pkt.len);
1207 n->maxrecentlen = 0;
1214 struct timeval ping_tx_elapsed;
1215 timersub(&now, &n->udp_ping_sent, &ping_tx_elapsed);
1217 int interval = n->status.udp_confirmed
1218 ? udp_discovery_keepalive_interval
1219 : udp_discovery_interval;
1221 if(ping_tx_elapsed.tv_sec >= interval) {
1222 gettimeofday(&now, NULL);
1223 n->udp_ping_sent = now; // a probe in flight
1224 n->status.ping_sent = true;
1225 send_udp_probe_packet(n, MIN_PROBE_SIZE);
1227 if(localdiscovery && !n->status.udp_confirmed && n->prevedge) {
1228 n->status.send_locally = true;
1229 send_udp_probe_packet(n, MIN_PROBE_SIZE);
1230 n->status.send_locally = false;
1235 static length_t choose_initial_maxmtu(node_t *n) {
1240 const sockaddr_t *sa = NULL;
1242 choose_udp_address(n, &sa, &sockindex);
1248 sock = socket(sa->sa.sa_family, SOCK_DGRAM, IPPROTO_UDP);
1251 logger(DEBUG_TRAFFIC, LOG_ERR, "Creating MTU assessment socket for %s (%s) failed: %s", n->name, n->hostname, sockstrerror(sockerrno));
1255 if(connect(sock, &sa->sa, SALEN(sa->sa))) {
1256 logger(DEBUG_TRAFFIC, LOG_ERR, "Connecting MTU assessment socket for %s (%s) failed: %s", n->name, n->hostname, sockstrerror(sockerrno));
1262 socklen_t ip_mtu_len = sizeof(ip_mtu);
1264 if(getsockopt(sock, IPPROTO_IP, IP_MTU, (void *)&ip_mtu, &ip_mtu_len)) {
1265 logger(DEBUG_TRAFFIC, LOG_ERR, "getsockopt(IP_MTU) on %s (%s) failed: %s", n->name, n->hostname, sockstrerror(sockerrno));
1272 if(ip_mtu < MINMTU) {
1273 logger(DEBUG_TRAFFIC, LOG_ERR, "getsockopt(IP_MTU) on %s (%s) returned absurdly small value: %d", n->name, n->hostname, ip_mtu);
1277 /* getsockopt(IP_MTU) returns the MTU of the physical interface.
1278 We need to remove various overheads to get to the tinc MTU. */
1279 length_t mtu = ip_mtu;
1280 mtu -= (sa->sa.sa_family == AF_INET6) ? sizeof(struct ip6_hdr) : sizeof(struct ip);
1283 if(n->status.sptps) {
1284 mtu -= SPTPS_DATAGRAM_OVERHEAD;
1286 if((n->options >> 24) >= 4) {
1287 mtu -= sizeof(node_id_t) + sizeof(node_id_t);
1290 #ifndef DISABLE_LEGACY
1292 mtu -= digest_length(n->outdigest);
1294 /* Now it's tricky. We use CBC mode, so the length of the
1295 encrypted payload must be a multiple of the blocksize. The
1296 sequence number is also part of the encrypted payload, so we
1297 must account for it after correcting for the blocksize.
1298 Furthermore, the padding in the last block must be at least
1301 length_t blocksize = cipher_blocksize(n->outcipher);
1317 logger(DEBUG_TRAFFIC, LOG_INFO, "Using system-provided maximum tinc MTU for %s (%s): %hd", n->name, n->hostname, mtu);
1326 /* This function tries to determines the MTU of a node.
1327 By calling this function repeatedly, n->minmtu will be progressively
1328 increased, and at some point, n->mtu will be fixed to n->minmtu. If the MTU
1329 is already fixed, this function checks if it can be increased.
1332 static void try_mtu(node_t *n) {
1333 if(!(n->options & OPTION_PMTU_DISCOVERY)) {
1337 if(udp_discovery && !n->status.udp_confirmed) {
1338 n->maxrecentlen = 0;
1345 /* mtuprobes == 0..19: initial discovery, send bursts with 1 second interval, mtuprobes++
1346 mtuprobes == 20: fix MTU, and go to -1
1347 mtuprobes == -1: send one maxmtu and one maxmtu+1 probe every pinginterval
1348 mtuprobes ==-2..-3: send one maxmtu probe every second
1349 mtuprobes == -4: maxmtu no longer valid, reset minmtu and maxmtu and go to 0 */
1351 struct timeval elapsed;
1352 timersub(&now, &n->mtu_ping_sent, &elapsed);
1354 if(n->mtuprobes >= 0) {
1355 if(n->mtuprobes != 0 && elapsed.tv_sec == 0 && elapsed.tv_usec < 333333) {
1359 if(n->mtuprobes < -1) {
1360 if(elapsed.tv_sec < 1) {
1364 if(elapsed.tv_sec < pinginterval) {
1370 n->mtu_ping_sent = now;
1374 if(n->mtuprobes < -3) {
1375 /* We lost three MTU probes, restart discovery */
1376 logger(DEBUG_TRAFFIC, LOG_INFO, "Decrease in PMTU to %s (%s) detected, restarting PMTU discovery", n->name, n->hostname);
1381 if(n->mtuprobes < 0) {
1382 /* After the initial discovery, we only send one maxmtu and one
1383 maxmtu+1 probe to detect PMTU increases. */
1384 send_udp_probe_packet(n, n->maxmtu);
1386 if(n->mtuprobes == -1 && n->maxmtu + 1 < MTU) {
1387 send_udp_probe_packet(n, n->maxmtu + 1);
1392 /* Before initial discovery begins, set maxmtu to the most likely value.
1393 If it's underestimated, we will correct it after initial discovery. */
1394 if(n->mtuprobes == 0) {
1395 n->maxmtu = choose_initial_maxmtu(n);
1399 /* Decreasing the number of probes per cycle might make the algorithm react faster to lost packets,
1400 but it will typically increase convergence time in the no-loss case. */
1401 const length_t probes_per_cycle = 8;
1403 /* This magic value was determined using math simulations.
1404 It will result in a 1329-byte first probe, followed (if there was a reply) by a 1407-byte probe.
1405 Since 1407 is just below the range of tinc MTUs over typical networks,
1406 this fine-tuning allows tinc to cover a lot of ground very quickly.
1407 This fine-tuning is only valid for maxmtu = MTU; if maxmtu is smaller,
1408 then it's better to use a multiplier of 1. Indeed, this leads to an interesting scenario
1409 if choose_initial_maxmtu() returns the actual MTU value - it will get confirmed with one single probe. */
1410 const float multiplier = (n->maxmtu == MTU) ? 0.97f : 1.0f;
1412 const float cycle_position = (float) probes_per_cycle - (float)(n->mtuprobes % probes_per_cycle) - 1.0f;
1413 const length_t minmtu = MAX(n->minmtu, MINMTU);
1414 const float interval = (float)(n->maxmtu - minmtu);
1416 length_t offset = 0;
1418 /* powf can be underflowed if n->maxmtu is less than 512 due to the minmtu MAX bound */
1420 /* The core of the discovery algorithm is this exponential.
1421 It produces very large probes early in the cycle, and then it very quickly decreases the probe size.
1422 This reflects the fact that in the most difficult cases, we don't get any feedback for probes that
1423 are too large, and therefore we need to concentrate on small offsets so that we can quickly converge
1424 on the precise MTU as we are approaching it.
1425 The last probe of the cycle is always 1 byte in size - this is to make sure we'll get at least one
1426 reply per cycle so that we can make progress. */
1427 offset = lrintf(powf(interval, multiplier * cycle_position / (float)(probes_per_cycle - 1)));
1430 length_t maxmtu = n->maxmtu;
1431 send_udp_probe_packet(n, minmtu + offset);
1433 /* If maxmtu changed, it means the probe was rejected by the system because it was too large.
1434 In that case, we recalculate with the new maxmtu and try again. */
1435 if(n->mtuprobes < 0 || maxmtu == n->maxmtu) {
1440 if(n->mtuprobes >= 0) {
1446 /* These functions try to establish a tunnel to a node (or its relay) so that
1447 packets can be sent (e.g. exchange keys).
1448 If a tunnel is already established, it tries to improve it (e.g. by trying
1449 to establish a UDP tunnel instead of TCP). This function makes no
1450 guarantees - it is up to the caller to check the node's state to figure out
1451 if TCP and/or UDP is usable. By calling this function repeatedly, the
1452 tunnel is gradually improved until we hit the wall imposed by the underlying
1453 network environment. It is recommended to call this function every time a
1454 packet is sent (or intended to be sent) to a node, so that the tunnel keeps
1455 improving as packets flow, and then gracefully downgrades itself as it goes
1459 static void try_tx_sptps(node_t *n, bool mtu) {
1460 /* If n is a TCP-only neighbor, we'll only use "cleartext" PACKET
1461 messages anyway, so there's no need for SPTPS at all. */
1463 if(n->connection && ((myself->options | n->options) & OPTION_TCPONLY)) {
1467 /* Otherwise, try to do SPTPS authentication with n if necessary. */
1471 /* Do we need to statically relay packets? */
1473 node_t *via = (n->via == myself) ? n->nexthop : n->via;
1475 /* If we do have a static relay, try everything with that one instead, if it supports relaying. */
1478 if((via->options >> 24) < 4) {
1486 /* Otherwise, try to establish UDP connectivity. */
1494 /* If we don't have UDP connectivity (yet), we need to use a dynamic relay (nexthop)
1495 while we try to establish direct connectivity. */
1497 if(!n->status.udp_confirmed && n != n->nexthop && (n->nexthop->options >> 24) >= 4) {
1498 try_tx(n->nexthop, mtu);
1502 static void try_tx_legacy(node_t *n, bool mtu) {
1503 /* Does he have our key? If not, send one. */
1505 if(!n->status.validkey_in) {
1509 /* Check if we already have a key, or request one. */
1511 if(!n->status.validkey) {
1512 if(n->last_req_key + 10 <= now.tv_sec) {
1514 n->last_req_key = now.tv_sec;
1527 void try_tx(node_t *n, bool mtu) {
1528 if(!n->status.reachable) {
1532 if(n->status.sptps) {
1533 try_tx_sptps(n, mtu);
1535 try_tx_legacy(n, mtu);
1539 void send_packet(node_t *n, vpn_packet_t *packet) {
1540 // If it's for myself, write it to the tun/tap device.
1544 memcpy(DATA(packet), mymac.x, ETH_ALEN);
1545 // Use an arbitrary fake source address.
1546 memcpy(DATA(packet) + ETH_ALEN, DATA(packet), ETH_ALEN);
1547 DATA(packet)[ETH_ALEN * 2 - 1] ^= 0xFF;
1551 n->out_bytes += packet->len;
1552 devops.write(packet);
1556 logger(DEBUG_TRAFFIC, LOG_ERR, "Sending packet of %d bytes to %s (%s)", packet->len, n->name, n->hostname);
1558 // If the node is not reachable, drop it.
1560 if(!n->status.reachable) {
1561 logger(DEBUG_TRAFFIC, LOG_INFO, "Node %s (%s) is not reachable", n->name, n->hostname);
1565 // Keep track of packet statistics.
1568 n->out_bytes += packet->len;
1570 // Check if it should be sent as an SPTPS packet.
1572 if(n->status.sptps) {
1573 send_sptps_packet(n, packet);
1578 // Determine which node to actually send it to.
1580 node_t *via = (packet->priority == -1 || n->via == myself) ? n->nexthop : n->via;
1583 logger(DEBUG_TRAFFIC, LOG_INFO, "Sending packet to %s via %s (%s)", n->name, via->name, n->via->hostname);
1586 // Try to send via UDP, unless TCP is forced.
1588 if(packet->priority == -1 || ((myself->options | via->options) & OPTION_TCPONLY)) {
1589 if(!send_tcppacket(via->connection, packet)) {
1590 terminate_connection(via->connection, true);
1596 send_udppacket(via, packet);
1600 void broadcast_packet(const node_t *from, vpn_packet_t *packet) {
1601 // Always give ourself a copy of the packet.
1602 if(from != myself) {
1603 send_packet(myself, packet);
1606 // In TunnelServer mode, do not forward broadcast packets.
1607 // The MST might not be valid and create loops.
1608 if(tunnelserver || broadcast_mode == BMODE_NONE) {
1612 logger(DEBUG_TRAFFIC, LOG_INFO, "Broadcasting packet of %d bytes from %s (%s)",
1613 packet->len, from->name, from->hostname);
1615 switch(broadcast_mode) {
1616 // In MST mode, broadcast packets travel via the Minimum Spanning Tree.
1617 // This guarantees all nodes receive the broadcast packet, and
1618 // usually distributes the sending of broadcast packets over all nodes.
1620 for list_each(connection_t, c, &connection_list)
1621 if(c->edge && c->status.mst && c != from->nexthop->connection) {
1622 send_packet(c->node, packet);
1627 // In direct mode, we send copies to each node we know of.
1628 // However, this only reaches nodes that can be reached in a single hop.
1629 // We don't have enough information to forward broadcast packets in this case.
1631 if(from != myself) {
1635 for splay_each(node_t, n, &node_tree)
1636 if(n->status.reachable && n != myself && ((n->via == myself && n->nexthop == n) || n->via == n)) {
1637 send_packet(n, packet);
1647 /* We got a packet from some IP address, but we don't know who sent it. Try to
1648 verify the message authentication code against all active session keys.
1649 Since this is actually an expensive operation, we only do a full check once
1650 a minute, the rest of the time we only check against nodes for which we know
1651 an IP address that matches the one from the packet. */
1653 static node_t *try_harder(const sockaddr_t *from, const vpn_packet_t *pkt) {
1654 node_t *match = NULL;
1656 static time_t last_hard_try = 0;
1658 for splay_each(node_t, n, &node_tree) {
1659 if(!n->status.reachable || n == myself) {
1663 if(!n->status.validkey_in && !(n->status.sptps && n->sptps.instate)) {
1669 for splay_each(edge_t, e, &n->edge_tree) {
1674 if(!sockaddrcmp_noport(from, &e->reverse->address)) {
1681 if(last_hard_try == now.tv_sec) {
1688 if(!try_mac(n, pkt)) {
1697 last_hard_try = now.tv_sec;
1703 static void handle_incoming_vpn_packet(listen_socket_t *ls, vpn_packet_t *pkt, sockaddr_t *addr) {
1705 node_id_t nullid = {0};
1707 bool direct = false;
1709 sockaddrunmap(addr); /* Some braindead IPv6 implementations do stupid things. */
1711 // Try to figure out who sent this packet.
1713 node_t *n = lookup_node_udp(addr);
1715 if(n && !n->status.udp_confirmed) {
1716 n = NULL; // Don't believe it if we don't have confirmation yet.
1720 // It might be from a 1.1 node, which might have a source ID in the packet.
1721 pkt->offset = 2 * sizeof(node_id_t);
1722 from = lookup_node_id(SRCID(pkt));
1724 if(from && from->status.sptps && !memcmp(DSTID(pkt), &nullid, sizeof(nullid))) {
1725 if(sptps_verify_datagram(&from->sptps, DATA(pkt), pkt->len - 2 * sizeof(node_id_t))) {
1735 n = try_harder(addr, pkt);
1741 if(debug_level >= DEBUG_PROTOCOL) {
1742 hostname = sockaddr2hostname(addr);
1743 logger(DEBUG_PROTOCOL, LOG_WARNING, "Received UDP packet from unknown source %s", hostname);
1752 if(n->status.sptps) {
1753 bool relay_enabled = (n->options >> 24) >= 4;
1756 pkt->offset = 2 * sizeof(node_id_t);
1757 pkt->len -= pkt->offset;
1760 if(!relay_enabled || !memcmp(DSTID(pkt), &nullid, sizeof(nullid))) {
1765 from = lookup_node_id(SRCID(pkt));
1766 to = lookup_node_id(DSTID(pkt));
1770 logger(DEBUG_PROTOCOL, LOG_WARNING, "Received UDP packet from %s (%s) with unknown source and/or destination ID", n->name, n->hostname);
1774 if(!to->status.reachable) {
1775 /* This can happen in the form of a race condition
1776 if the node just became unreachable. */
1777 logger(DEBUG_TRAFFIC, LOG_WARNING, "Cannot relay packet from %s (%s) because the destination, %s (%s), is unreachable", from->name, from->hostname, to->name, to->hostname);
1781 /* The packet is supposed to come from the originator or its static relay
1782 (i.e. with no dynamic relays in between).
1783 If it did not, "help" the static relay by sending it UDP info.
1784 Note that we only do this if we're the destination or the static relay;
1785 otherwise every hop would initiate its own UDP info message, resulting in elevated chatter. */
1787 if(n != from->via && to->via == myself) {
1788 send_udp_info(myself, from);
1791 /* If we're not the final recipient, relay the packet. */
1794 send_sptps_data(to, from, 0, DATA(pkt), pkt->len);
1803 if(!receive_udppacket(from, pkt)) {
1807 n->sock = ls - listen_socket;
1809 if(direct && sockaddrcmp(addr, &n->address)) {
1810 update_node_udp(n, addr);
1813 /* If the packet went through a relay, help the sender find the appropriate MTU
1814 through the relay path. */
1817 send_mtu_info(myself, n, MTU);
1821 void handle_incoming_vpn_data(void *data, int flags) {
1824 listen_socket_t *ls = data;
1826 #ifdef HAVE_RECVMMSG
1828 static ssize_t num = MAX_MSG;
1829 static vpn_packet_t pkt[MAX_MSG];
1830 static sockaddr_t addr[MAX_MSG];
1831 static struct mmsghdr msg[MAX_MSG];
1832 static struct iovec iov[MAX_MSG];
1834 for(int i = 0; i < num; i++) {
1837 iov[i] = (struct iovec) {
1838 .iov_base = DATA(&pkt[i]),
1842 msg[i].msg_hdr = (struct msghdr) {
1843 .msg_name = &addr[i].sa,
1844 .msg_namelen = sizeof(addr)[i],
1850 num = recvmmsg(ls->udp.fd, msg, MAX_MSG, MSG_DONTWAIT, NULL);
1853 if(!sockwouldblock(sockerrno)) {
1854 logger(DEBUG_ALWAYS, LOG_ERR, "Receiving packet failed: %s", sockstrerror(sockerrno));
1860 for(int i = 0; i < num; i++) {
1861 pkt[i].len = msg[i].msg_len;
1863 if(pkt[i].len <= 0 || pkt[i].len > MAXSIZE) {
1867 handle_incoming_vpn_packet(ls, &pkt[i], &addr[i]);
1872 sockaddr_t addr = {0};
1873 socklen_t addrlen = sizeof(addr);
1876 ssize_t len = recvfrom(ls->udp.fd, (void *)DATA(&pkt), MAXSIZE, 0, &addr.sa, &addrlen);
1878 if(len <= 0 || (size_t)len > MAXSIZE) {
1879 if(!sockwouldblock(sockerrno)) {
1880 logger(DEBUG_ALWAYS, LOG_ERR, "Receiving packet failed: %s", sockstrerror(sockerrno));
1888 handle_incoming_vpn_packet(ls, &pkt, &addr);
1892 void handle_device_data(void *data, int flags) {
1895 vpn_packet_t packet;
1896 packet.offset = DEFAULT_PACKET_OFFSET;
1897 packet.priority = 0;
1898 static int errors = 0;
1900 if(devops.read(&packet)) {
1902 myself->in_packets++;
1903 myself->in_bytes += packet.len;
1904 route(myself, &packet);
1906 usleep(errors * 50000);
1910 logger(DEBUG_ALWAYS, LOG_ERR, "Too many errors from %s, exiting!", device);