2 net_packet.c -- Handles in- and outgoing VPN packets
3 Copyright (C) 1998-2005 Ivo Timmermans,
4 2000-2021 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"
59 #define MAX(a, b) ((a) > (b) ? (a) : (b))
62 /* The minimum size of a probe is 14 bytes, but since we normally use CBC mode
63 encryption, we can add a few extra random bytes without increasing the
64 resulting packet size. */
65 #define MIN_PROBE_SIZE 18
69 static char lzo_wrkmem[LZO1X_999_MEM_COMPRESS > LZO1X_1_MEM_COMPRESS ? LZO1X_999_MEM_COMPRESS : LZO1X_1_MEM_COMPRESS];
72 #ifdef HAVE_LZ4_BUILTIN
73 static LZ4_stream_t lz4_stream;
75 static void *lz4_state = NULL;
76 #endif /* HAVE_LZ4_BUILTIN */
78 static void send_udppacket(node_t *, vpn_packet_t *);
80 unsigned replaywin = 32;
81 bool localdiscovery = true;
82 bool udp_discovery = true;
83 int udp_discovery_keepalive_interval = 10;
84 int udp_discovery_interval = 2;
85 int udp_discovery_timeout = 30;
87 #define MAX_SEQNO 1073741824
89 static void try_fix_mtu(node_t *n) {
90 if(n->mtuprobes < 0) {
94 if(n->mtuprobes == 20 || n->minmtu >= n->maxmtu) {
95 if(n->minmtu > n->maxmtu) {
96 n->minmtu = n->maxmtu;
98 n->maxmtu = n->minmtu;
102 logger(DEBUG_TRAFFIC, LOG_INFO, "Fixing MTU of %s (%s) to %d after %d probes", n->name, n->hostname, n->mtu, n->mtuprobes);
107 static void udp_probe_timeout_handler(void *data) {
110 if(!n->status.udp_confirmed) {
114 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);
115 n->status.udp_confirmed = false;
116 n->udp_ping_rtt = -1;
123 static void send_udp_probe_reply(node_t *n, vpn_packet_t *packet, length_t len) {
124 if(!n->status.sptps && !n->status.validkey) {
125 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);
129 /* Type 2 probe replies were introduced in protocol 17.3 */
130 if((n->options >> 24) >= 3) {
132 uint16_t len16 = htons(len);
133 memcpy(DATA(packet) + 1, &len16, 2);
134 packet->len = MIN_PROBE_SIZE;
135 logger(DEBUG_TRAFFIC, LOG_INFO, "Sending type 2 probe reply length %u to %s (%s)", len, n->name, n->hostname);
138 /* Legacy protocol: n won't understand type 2 probe replies. */
140 logger(DEBUG_TRAFFIC, LOG_INFO, "Sending type 1 probe reply length %u to %s (%s)", len, n->name, n->hostname);
143 /* Temporarily set udp_confirmed, so that the reply is sent
144 back exactly the way it came in. */
146 bool udp_confirmed = n->status.udp_confirmed;
147 n->status.udp_confirmed = true;
148 send_udppacket(n, packet);
149 n->status.udp_confirmed = udp_confirmed;
152 static void udp_probe_h(node_t *n, vpn_packet_t *packet, length_t len) {
153 if(!DATA(packet)[0]) {
154 logger(DEBUG_TRAFFIC, LOG_INFO, "Got UDP probe request %d from %s (%s)", packet->len, n->name, n->hostname);
155 send_udp_probe_reply(n, packet, len);
159 if(DATA(packet)[0] == 2) {
160 // It's a type 2 probe reply, use the length field inside the packet
162 memcpy(&len16, DATA(packet) + 1, 2);
166 if(n->status.ping_sent) { // a probe in flight
167 gettimeofday(&now, NULL);
169 timersub(&now, &n->udp_ping_sent, &rtt);
170 n->udp_ping_rtt = (int)(rtt.tv_sec * 1000000 + rtt.tv_usec);
171 n->status.ping_sent = false;
172 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);
174 logger(DEBUG_TRAFFIC, LOG_INFO, "Got type %d UDP probe reply %d from %s (%s)", DATA(packet)[0], len, n->name, n->hostname);
177 /* It's a valid reply: now we know bidirectional communication
178 is possible using the address and socket that the reply
180 if(!n->status.udp_confirmed) {
181 n->status.udp_confirmed = true;
183 if(!n->address_cache) {
184 n->address_cache = open_address_cache(n);
187 reset_address_cache(n->address_cache, &n->address);
190 // Reset the UDP ping timer.
193 timeout_del(&n->udp_ping_timeout);
194 timeout_add(&n->udp_ping_timeout, &udp_probe_timeout_handler, n, &(struct timeval) {
195 udp_discovery_timeout, 0
199 if(len > n->maxmtu) {
200 logger(DEBUG_TRAFFIC, LOG_INFO, "Increase in PMTU to %s (%s) detected, restarting PMTU discovery", n->name, n->hostname);
203 /* Set mtuprobes to 1 so that try_mtu() doesn't reset maxmtu */
206 } else if(n->mtuprobes < 0 && len == n->maxmtu) {
207 /* We got a maxmtu sized packet, confirming the PMTU is still valid. */
209 n->mtu_ping_sent = now;
212 /* If applicable, raise the minimum supported MTU */
214 if(n->minmtu < len) {
221 static length_t compress_packet_lz4(uint8_t *dest, const uint8_t *source, length_t len) {
222 #ifdef HAVE_LZ4_BUILTIN
223 return LZ4_compress_fast_extState(&lz4_stream, (const char *) source, (char *) dest, len, MAXSIZE, 0);
226 /* @FIXME: Put this in a better place, and free() it too. */
227 if(lz4_state == NULL) {
228 lz4_state = malloc(LZ4_sizeofState());
231 if(lz4_state == NULL) {
232 logger(DEBUG_ALWAYS, LOG_ERR, "Failed to allocate lz4_state, error: %i", errno);
236 return LZ4_compress_fast_extState(lz4_state, (const char *) source, (char *) dest, len, MAXSIZE, 0);
237 #endif /* HAVE_LZ4_BUILTIN */
239 #endif /* HAVE_LZ4 */
242 static length_t compress_packet_lzo(uint8_t *dest, const uint8_t *source, length_t len, compression_level_t level) {
243 assert(level == COMPRESS_LZO_LO || level == COMPRESS_LZO_HI);
245 lzo_uint lzolen = MAXSIZE;
248 if(level == COMPRESS_LZO_HI) {
249 result = lzo1x_999_compress(source, len, dest, &lzolen, lzo_wrkmem);
250 } else { // level == COMPRESS_LZO_LO
251 result = lzo1x_1_compress(source, len, dest, &lzolen, lzo_wrkmem);
254 if(result == LZO_E_OK) {
262 static length_t compress_packet(uint8_t *dest, const uint8_t *source, length_t len, compression_level_t level) {
267 return compress_packet_lz4(dest, source, len);
272 case COMPRESS_LZO_HI:
273 case COMPRESS_LZO_LO:
274 return compress_packet_lzo(dest, source, len, level);
278 case COMPRESS_ZLIB_9:
279 case COMPRESS_ZLIB_8:
280 case COMPRESS_ZLIB_7:
281 case COMPRESS_ZLIB_6:
282 case COMPRESS_ZLIB_5:
283 case COMPRESS_ZLIB_4:
284 case COMPRESS_ZLIB_3:
285 case COMPRESS_ZLIB_2:
286 case COMPRESS_ZLIB_1: {
287 unsigned long dest_len = MAXSIZE;
289 if(compress2(dest, &dest_len, source, len, level) == Z_OK) {
299 memcpy(dest, source, len);
307 static length_t uncompress_packet(uint8_t *dest, const uint8_t *source, length_t len, compression_level_t level) {
312 return LZ4_decompress_safe((char *)source, (char *) dest, len, MAXSIZE);
317 case COMPRESS_LZO_HI:
318 case COMPRESS_LZO_LO: {
319 lzo_uint dst_len = MAXSIZE;
321 if(lzo1x_decompress_safe(source, len, dest, &dst_len, NULL) == LZO_E_OK) {
331 case COMPRESS_ZLIB_9:
332 case COMPRESS_ZLIB_8:
333 case COMPRESS_ZLIB_7:
334 case COMPRESS_ZLIB_6:
335 case COMPRESS_ZLIB_5:
336 case COMPRESS_ZLIB_4:
337 case COMPRESS_ZLIB_3:
338 case COMPRESS_ZLIB_2:
339 case COMPRESS_ZLIB_1: {
340 unsigned long destlen = MAXSIZE;
341 static z_stream stream;
344 inflateReset(&stream);
346 inflateInit(&stream);
349 stream.next_in = source;
350 stream.avail_in = len;
351 stream.next_out = dest;
352 stream.avail_out = destlen;
353 stream.total_out = 0;
355 if(inflate(&stream, Z_FINISH) == Z_STREAM_END) {
356 return stream.total_out;
365 memcpy(dest, source, len);
375 static void receive_packet(node_t *n, vpn_packet_t *packet) {
376 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Received packet of %d bytes from %s (%s)",
377 packet->len, n->name, n->hostname);
380 n->in_bytes += packet->len;
385 static bool try_mac(node_t *n, const vpn_packet_t *inpkt) {
386 if(n->status.sptps) {
387 return sptps_verify_datagram(&n->sptps, DATA(inpkt), inpkt->len);
390 #ifdef DISABLE_LEGACY
394 if(!n->status.validkey_in || !digest_active(n->indigest) || (size_t)inpkt->len < sizeof(seqno_t) + digest_length(n->indigest)) {
398 return digest_verify(n->indigest, inpkt->data, inpkt->len - digest_length(n->indigest), inpkt->data + inpkt->len - digest_length(n->indigest));
402 static bool receive_udppacket(node_t *n, vpn_packet_t *inpkt) {
403 if(n->status.sptps) {
404 if(!n->sptps.state) {
405 if(!n->status.waitingforkey) {
406 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got packet from %s (%s) but we haven't exchanged keys yet", n->name, n->hostname);
409 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got packet from %s (%s) but he hasn't got our key yet", n->name, n->hostname);
415 n->status.udppacket = true;
416 bool result = sptps_receive_data(&n->sptps, DATA(inpkt), inpkt->len);
417 n->status.udppacket = false;
420 /* Uh-oh. It might be that the tunnel is stuck in some corrupted state,
421 so let's restart SPTPS in case that helps. But don't do that too often
422 to prevent storms, and because that would make life a little too easy
423 for external attackers trying to DoS us. */
424 if(n->last_req_key < now.tv_sec - 10) {
425 logger(DEBUG_PROTOCOL, LOG_ERR, "Failed to decode raw TCP packet from %s (%s), restarting SPTPS", n->name, n->hostname);
435 #ifdef DISABLE_LEGACY
438 vpn_packet_t pkt1, pkt2;
439 vpn_packet_t *pkt[] = { &pkt1, &pkt2, &pkt1, &pkt2 };
442 pkt1.offset = DEFAULT_PACKET_OFFSET;
443 pkt2.offset = DEFAULT_PACKET_OFFSET;
445 if(!n->status.validkey_in) {
446 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got packet from %s (%s) but he hasn't got our key yet", n->name, n->hostname);
450 /* Check packet length */
452 if((size_t)inpkt->len < sizeof(seqno_t) + digest_length(n->indigest)) {
453 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got too short packet from %s (%s)",
454 n->name, n->hostname);
458 /* It's a legacy UDP packet, the data starts after the seqno */
460 inpkt->offset += sizeof(seqno_t);
462 /* Check the message authentication code */
464 if(digest_active(n->indigest)) {
465 inpkt->len -= digest_length(n->indigest);
467 if(!digest_verify(n->indigest, SEQNO(inpkt), inpkt->len, SEQNO(inpkt) + inpkt->len)) {
468 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got unauthenticated packet from %s (%s)", n->name, n->hostname);
473 /* Decrypt the packet */
475 if(cipher_active(n->incipher)) {
476 vpn_packet_t *outpkt = pkt[nextpkt++];
479 if(!cipher_decrypt(n->incipher, SEQNO(inpkt), inpkt->len, SEQNO(outpkt), &outlen, true)) {
480 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Error decrypting packet from %s (%s)", n->name, n->hostname);
484 outpkt->len = outlen;
488 /* Check the sequence number */
491 memcpy(&seqno, SEQNO(inpkt), sizeof(seqno));
492 seqno = ntohl(seqno);
493 inpkt->len -= sizeof(seqno);
496 if(seqno != n->received_seqno + 1) {
497 if(seqno >= n->received_seqno + replaywin * 8) {
498 if(n->farfuture++ < replaywin >> 2) {
499 logger(DEBUG_TRAFFIC, LOG_WARNING, "Packet from %s (%s) is %d seqs in the future, dropped (%u)",
500 n->name, n->hostname, seqno - n->received_seqno - 1, n->farfuture);
504 logger(DEBUG_TRAFFIC, LOG_WARNING, "Lost %d packets from %s (%s)",
505 seqno - n->received_seqno - 1, n->name, n->hostname);
506 memset(n->late, 0, replaywin);
507 } else if(seqno <= n->received_seqno) {
508 if((n->received_seqno >= replaywin * 8 && seqno <= n->received_seqno - replaywin * 8) || !(n->late[(seqno / 8) % replaywin] & (1 << seqno % 8))) {
509 logger(DEBUG_TRAFFIC, LOG_WARNING, "Got late or replayed packet from %s (%s), seqno %d, last received %d",
510 n->name, n->hostname, seqno, n->received_seqno);
514 for(seqno_t i = n->received_seqno + 1; i < seqno; i++) {
515 n->late[(i / 8) % replaywin] |= 1 << i % 8;
521 n->late[(seqno / 8) % replaywin] &= ~(1 << seqno % 8);
524 if(seqno > n->received_seqno) {
525 n->received_seqno = seqno;
530 if(n->received_seqno > MAX_SEQNO) {
534 /* Decompress the packet */
536 length_t origlen = inpkt->len;
538 if(n->incompression != COMPRESS_NONE) {
539 vpn_packet_t *outpkt = pkt[nextpkt++];
541 if(!(outpkt->len = uncompress_packet(DATA(outpkt), DATA(inpkt), inpkt->len, n->incompression))) {
542 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while uncompressing packet from %s (%s)",
543 n->name, n->hostname);
549 if(origlen > MTU / 64 + 20) {
550 origlen -= MTU / 64 + 20;
556 if(inpkt->len > n->maxrecentlen) {
557 n->maxrecentlen = inpkt->len;
562 if(!DATA(inpkt)[12] && !DATA(inpkt)[13]) {
563 udp_probe_h(n, inpkt, origlen);
565 receive_packet(n, inpkt);
572 void receive_tcppacket(connection_t *c, const char *buffer, size_t len) {
574 outpkt.offset = DEFAULT_PACKET_OFFSET;
576 if(len > sizeof(outpkt.data) - outpkt.offset) {
582 if(c->options & OPTION_TCPONLY) {
585 outpkt.priority = -1;
588 memcpy(DATA(&outpkt), buffer, len);
590 receive_packet(c->node, &outpkt);
593 bool receive_tcppacket_sptps(connection_t *c, const char *data, size_t len) {
594 if(len < sizeof(node_id_t) + sizeof(node_id_t)) {
595 logger(DEBUG_PROTOCOL, LOG_ERR, "Got too short TCP SPTPS packet from %s (%s)", c->name, c->hostname);
599 node_t *to = lookup_node_id((node_id_t *)data);
600 data += sizeof(node_id_t);
601 len -= sizeof(node_id_t);
604 logger(DEBUG_PROTOCOL, LOG_ERR, "Got TCP SPTPS packet from %s (%s) with unknown destination ID", c->name, c->hostname);
608 node_t *from = lookup_node_id((node_id_t *)data);
609 data += sizeof(node_id_t);
610 len -= sizeof(node_id_t);
613 logger(DEBUG_PROTOCOL, LOG_ERR, "Got TCP SPTPS packet from %s (%s) with unknown source ID", c->name, c->hostname);
617 if(!to->status.reachable) {
618 /* This can happen in the form of a race condition
619 if the node just became unreachable. */
620 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);
624 /* Help the sender reach us over UDP.
625 Note that we only do this if we're the destination or the static relay;
626 otherwise every hop would initiate its own UDP info message, resulting in elevated chatter. */
627 if(to->via == myself) {
628 send_udp_info(myself, from);
631 /* If we're not the final recipient, relay the packet. */
634 if(to->status.validkey) {
635 send_sptps_data(to, from, 0, data, len);
642 /* The packet is for us */
644 if(!sptps_receive_data(&from->sptps, data, len)) {
645 /* Uh-oh. It might be that the tunnel is stuck in some corrupted state,
646 so let's restart SPTPS in case that helps. But don't do that too often
647 to prevent storms. */
648 if(from->last_req_key < now.tv_sec - 10) {
649 logger(DEBUG_PROTOCOL, LOG_ERR, "Failed to decode raw TCP packet from %s (%s), restarting SPTPS", from->name, from->hostname);
656 send_mtu_info(myself, from, MTU);
660 static void send_sptps_packet(node_t *n, vpn_packet_t *origpkt) {
661 if(!n->status.validkey && !n->connection) {
668 if((!(DATA(origpkt)[12] | DATA(origpkt)[13])) && (n->sptps.outstate)) {
669 sptps_send_record(&n->sptps, PKT_PROBE, DATA(origpkt), origpkt->len);
673 if(routing_mode == RMODE_ROUTER) {
679 if(origpkt->len < offset) {
685 if(n->outcompression != COMPRESS_NONE) {
687 length_t len = compress_packet(DATA(&outpkt) + offset, DATA(origpkt) + offset, origpkt->len - offset, n->outcompression);
690 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while compressing packet to %s (%s)", n->name, n->hostname);
691 } else if(len < origpkt->len - offset) {
692 outpkt.len = len + offset;
694 type |= PKT_COMPRESSED;
698 /* If we have a direct metaconnection to n, and we can't use UDP, then
699 don't bother with SPTPS and just use a "plaintext" PACKET message.
700 We don't really care about end-to-end security since we're not
701 sending the message through any intermediate nodes. */
702 if(n->connection && origpkt->len > n->minmtu) {
703 send_tcppacket(n->connection, origpkt);
705 sptps_send_record(&n->sptps, type, DATA(origpkt) + offset, origpkt->len - offset);
709 static void adapt_socket(const sockaddr_t *sa, size_t *sock) {
710 /* Make sure we have a suitable socket for the chosen address */
711 if(listen_socket[*sock].sa.sa.sa_family != sa->sa.sa_family) {
712 for(int i = 0; i < listen_sockets; i++) {
713 if(listen_socket[i].sa.sa.sa_family == sa->sa.sa_family) {
721 static void choose_udp_address(const node_t *n, const sockaddr_t **sa, size_t *sock) {
726 /* If the UDP address is confirmed, use it. */
727 if(n->status.udp_confirmed) {
731 /* Send every third packet to n->address; that could be set
732 to the node's reflexive UDP address discovered during key
742 /* Otherwise, address are found in edges to this node.
743 So we pick a random edge and a random socket. */
746 unsigned int j = rand() % n->edge_tree.count;
747 edge_t *candidate = NULL;
749 for splay_each(edge_t, e, &n->edge_tree) {
751 candidate = e->reverse;
757 *sa = &candidate->address;
758 *sock = rand() % listen_sockets;
761 adapt_socket(*sa, sock);
764 static void choose_local_address(const node_t *n, const sockaddr_t **sa, size_t *sock) {
767 /* Pick one of the edges from this node at random, then use its local address. */
770 unsigned int j = rand() % n->edge_tree.count;
771 edge_t *candidate = NULL;
773 for splay_each(edge_t, e, &n->edge_tree) {
780 if(candidate && candidate->local_address.sa.sa_family) {
781 *sa = &candidate->local_address;
782 *sock = rand() % listen_sockets;
783 adapt_socket(*sa, sock);
787 static void send_udppacket(node_t *n, vpn_packet_t *origpkt) {
788 if(!n->status.reachable) {
789 logger(DEBUG_TRAFFIC, LOG_INFO, "Trying to send UDP packet to unreachable node %s (%s)", n->name, n->hostname);
793 if(n->status.sptps) {
794 send_sptps_packet(n, origpkt);
798 #ifdef DISABLE_LEGACY
801 vpn_packet_t pkt1, pkt2;
802 vpn_packet_t *pkt[] = { &pkt1, &pkt2, &pkt1, &pkt2 };
803 vpn_packet_t *inpkt = origpkt;
805 vpn_packet_t *outpkt;
806 int origlen = origpkt->len;
808 int origpriority = origpkt->priority;
810 pkt1.offset = DEFAULT_PACKET_OFFSET;
811 pkt2.offset = DEFAULT_PACKET_OFFSET;
813 /* Make sure we have a valid key */
815 if(!n->status.validkey) {
816 logger(DEBUG_TRAFFIC, LOG_INFO,
817 "No valid key known yet for %s (%s), forwarding via TCP",
818 n->name, n->hostname);
819 send_tcppacket(n->nexthop->connection, origpkt);
823 if(n->options & OPTION_PMTU_DISCOVERY && inpkt->len > n->minmtu && (DATA(inpkt)[12] | DATA(inpkt)[13])) {
824 logger(DEBUG_TRAFFIC, LOG_INFO,
825 "Packet for %s (%s) larger than minimum MTU, forwarding via %s",
826 n->name, n->hostname, n != n->nexthop ? n->nexthop->name : "TCP");
828 if(n != n->nexthop) {
829 send_packet(n->nexthop, origpkt);
831 send_tcppacket(n->nexthop->connection, origpkt);
837 /* Compress the packet */
839 if(n->outcompression != COMPRESS_NONE) {
840 outpkt = pkt[nextpkt++];
842 if(!(outpkt->len = compress_packet(DATA(outpkt), DATA(inpkt), inpkt->len, n->outcompression))) {
843 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while compressing packet to %s (%s)",
844 n->name, n->hostname);
851 /* Add sequence number */
853 seqno_t seqno = htonl(++(n->sent_seqno));
854 memcpy(SEQNO(inpkt), &seqno, sizeof(seqno));
855 inpkt->len += sizeof(seqno);
857 /* Encrypt the packet */
859 if(cipher_active(n->outcipher)) {
860 outpkt = pkt[nextpkt++];
863 if(!cipher_encrypt(n->outcipher, SEQNO(inpkt), inpkt->len, SEQNO(outpkt), &outlen, true)) {
864 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while encrypting packet to %s (%s)", n->name, n->hostname);
868 outpkt->len = outlen;
872 /* Add the message authentication code */
874 if(digest_active(n->outdigest)) {
875 if(!digest_create(n->outdigest, SEQNO(inpkt), inpkt->len, SEQNO(inpkt) + inpkt->len)) {
876 logger(DEBUG_TRAFFIC, LOG_ERR, "Error while encrypting packet to %s (%s)", n->name, n->hostname);
880 inpkt->len += digest_length(n->outdigest);
883 /* Send the packet */
885 const sockaddr_t *sa = NULL;
888 if(n->status.send_locally) {
889 choose_local_address(n, &sa, &sock);
893 choose_udp_address(n, &sa, &sock);
896 if(priorityinheritance && origpriority != listen_socket[sock].priority) {
897 listen_socket[sock].priority = origpriority;
899 switch(sa->sa.sa_family) {
903 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Setting IPv4 outgoing packet priority to %d", origpriority);
905 if(setsockopt(listen_socket[sock].udp.fd, IPPROTO_IP, IP_TOS, (void *)&origpriority, sizeof(origpriority))) { /* SO_PRIORITY doesn't seem to work */
906 logger(DEBUG_ALWAYS, LOG_ERR, "System call `%s' failed: %s", "setsockopt", sockstrerror(sockerrno));
911 #if defined(IPV6_TCLASS)
914 logger(DEBUG_TRAFFIC, LOG_DEBUG, "Setting IPv6 outgoing packet priority to %d", origpriority);
916 if(setsockopt(listen_socket[sock].udp.fd, IPPROTO_IPV6, IPV6_TCLASS, (void *)&origpriority, sizeof(origpriority))) { /* SO_PRIORITY doesn't seem to work */
917 logger(DEBUG_ALWAYS, LOG_ERR, "System call `%s' failed: %s", "setsockopt", sockstrerror(sockerrno));
928 if(sendto(listen_socket[sock].udp.fd, (void *)SEQNO(inpkt), inpkt->len, 0, &sa->sa, SALEN(sa->sa)) < 0 && !sockwouldblock(sockerrno)) {
929 if(sockmsgsize(sockerrno)) {
930 if(n->maxmtu >= origlen) {
931 n->maxmtu = origlen - 1;
934 if(n->mtu >= origlen) {
935 n->mtu = origlen - 1;
940 logger(DEBUG_TRAFFIC, LOG_WARNING, "Error sending packet to %s (%s): %s", n->name, n->hostname, sockstrerror(sockerrno));
945 origpkt->len = origlen;
949 bool send_sptps_data(node_t *to, node_t *from, int type, const void *data, size_t len) {
950 node_t *relay = (to->via != myself && (type == PKT_PROBE || (len - SPTPS_DATAGRAM_OVERHEAD) <= to->via->minmtu)) ? to->via : to->nexthop;
951 bool direct = from == myself && to == relay;
952 bool relay_supported = (relay->options >> 24) >= 4;
953 bool tcponly = (myself->options | relay->options) & OPTION_TCPONLY;
955 /* 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. */
957 if(type == SPTPS_HANDSHAKE || tcponly || (!direct && !relay_supported) || (type != PKT_PROBE && (len - SPTPS_DATAGRAM_OVERHEAD) > relay->minmtu)) {
958 if(type != SPTPS_HANDSHAKE && (to->nexthop->connection->options >> 24) >= 7) {
959 uint8_t buf[len + sizeof(to->id) + sizeof(from->id)];
960 uint8_t *buf_ptr = buf;
961 memcpy(buf_ptr, &to->id, sizeof(to->id));
962 buf_ptr += sizeof(to->id);
963 memcpy(buf_ptr, &from->id, sizeof(from->id));
964 buf_ptr += sizeof(from->id);
965 memcpy(buf_ptr, data, len);
966 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);
967 return send_sptps_tcppacket(to->nexthop->connection, buf, sizeof(buf));
970 char buf[len * 4 / 3 + 5];
971 b64encode(data, buf, len);
973 /* If this is a handshake packet, use ANS_KEY instead of REQ_KEY, for two reasons:
974 - We don't want intermediate nodes to switch to UDP to relay these packets;
975 - ANS_KEY allows us to learn the reflexive UDP address. */
976 if(type == SPTPS_HANDSHAKE) {
977 to->incompression = myself->incompression;
978 return send_request(to->nexthop->connection, "%d %s %s %s -1 -1 -1 %d", ANS_KEY, from->name, to->name, buf, to->incompression);
980 return send_request(to->nexthop->connection, "%d %s %s %d %s", REQ_KEY, from->name, to->name, SPTPS_PACKET, buf);
986 if(relay_supported) {
987 overhead += sizeof(to->id) + sizeof(from->id);
990 uint8_t buf[len + overhead];
991 uint8_t *buf_ptr = buf;
993 if(relay_supported) {
995 /* Inform the recipient that this packet was sent directly. */
996 node_id_t nullid = {0};
997 memcpy(buf_ptr, &nullid, sizeof(nullid));
998 buf_ptr += sizeof(nullid);
1000 memcpy(buf_ptr, &to->id, sizeof(to->id));
1001 buf_ptr += sizeof(to->id);
1004 memcpy(buf_ptr, &from->id, sizeof(from->id));
1005 buf_ptr += sizeof(from->id);
1009 /* 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 */
1010 memcpy(buf_ptr, data, len);
1013 const sockaddr_t *sa = NULL;
1016 if(relay->status.send_locally) {
1017 choose_local_address(relay, &sa, &sock);
1021 choose_udp_address(relay, &sa, &sock);
1024 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);
1026 if(sendto(listen_socket[sock].udp.fd, buf, buf_ptr - buf, 0, &sa->sa, SALEN(sa->sa)) < 0 && !sockwouldblock(sockerrno)) {
1027 if(sockmsgsize(sockerrno)) {
1028 // Compensate for SPTPS overhead
1029 len -= SPTPS_DATAGRAM_OVERHEAD;
1031 if(relay->maxmtu >= len) {
1032 relay->maxmtu = len - 1;
1035 if(relay->mtu >= len) {
1036 relay->mtu = len - 1;
1041 logger(DEBUG_TRAFFIC, LOG_WARNING, "Error sending UDP SPTPS packet to %s (%s): %s", relay->name, relay->hostname, sockstrerror(sockerrno));
1049 bool receive_sptps_record(void *handle, uint8_t type, const void *data, uint16_t len) {
1050 node_t *from = handle;
1052 if(type == SPTPS_HANDSHAKE) {
1053 if(!from->status.validkey) {
1054 from->status.validkey = true;
1055 from->status.waitingforkey = false;
1056 logger(DEBUG_META, LOG_INFO, "SPTPS key exchange with %s (%s) successful", from->name, from->hostname);
1063 logger(DEBUG_ALWAYS, LOG_ERR, "Packet from %s (%s) larger than maximum supported size (%d > %d)", from->name, from->hostname, len, MTU);
1068 inpkt.offset = DEFAULT_PACKET_OFFSET;
1071 if(type == PKT_PROBE) {
1072 if(!from->status.udppacket) {
1073 logger(DEBUG_ALWAYS, LOG_ERR, "Got SPTPS PROBE packet from %s (%s) via TCP", from->name, from->hostname);
1078 memcpy(DATA(&inpkt), data, len);
1080 if(inpkt.len > from->maxrecentlen) {
1081 from->maxrecentlen = inpkt.len;
1084 udp_probe_h(from, &inpkt, len);
1088 if(type & ~(PKT_COMPRESSED | PKT_MAC)) {
1089 logger(DEBUG_ALWAYS, LOG_ERR, "Unexpected SPTPS record type %d len %d from %s (%s)", type, len, from->name, from->hostname);
1093 /* Check if we have the headers we need */
1094 if(routing_mode != RMODE_ROUTER && !(type & PKT_MAC)) {
1095 logger(DEBUG_TRAFFIC, LOG_ERR, "Received packet from %s (%s) without MAC header (maybe Mode is not set correctly)", from->name, from->hostname);
1097 } else if(routing_mode == RMODE_ROUTER && (type & PKT_MAC)) {
1098 logger(DEBUG_TRAFFIC, LOG_WARNING, "Received packet from %s (%s) with MAC header (maybe Mode is not set correctly)", from->name, from->hostname);
1101 int offset = (type & PKT_MAC) ? 0 : 14;
1103 if(type & PKT_COMPRESSED) {
1104 length_t ulen = uncompress_packet(DATA(&inpkt) + offset, (const uint8_t *)data, len, from->incompression);
1109 inpkt.len = ulen + offset;
1112 if(inpkt.len > MAXSIZE) {
1116 memcpy(DATA(&inpkt) + offset, data, len);
1117 inpkt.len = len + offset;
1120 /* Generate the Ethernet packet type if necessary */
1122 switch(DATA(&inpkt)[14] >> 4) {
1124 DATA(&inpkt)[12] = 0x08;
1125 DATA(&inpkt)[13] = 0x00;
1129 DATA(&inpkt)[12] = 0x86;
1130 DATA(&inpkt)[13] = 0xDD;
1134 logger(DEBUG_TRAFFIC, LOG_ERR,
1135 "Unknown IP version %d while reading packet from %s (%s)",
1136 DATA(&inpkt)[14] >> 4, from->name, from->hostname);
1141 if(from->status.udppacket && inpkt.len > from->maxrecentlen) {
1142 from->maxrecentlen = inpkt.len;
1145 receive_packet(from, &inpkt);
1149 // This function tries to get SPTPS keys, if they aren't already known.
1150 // 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.
1151 static void try_sptps(node_t *n) {
1152 if(n->status.validkey) {
1156 logger(DEBUG_TRAFFIC, LOG_INFO, "No valid key known yet for %s (%s)", n->name, n->hostname);
1158 if(!n->status.waitingforkey) {
1160 } else if(n->last_req_key + 10 < now.tv_sec) {
1161 logger(DEBUG_ALWAYS, LOG_DEBUG, "No key from %s after 10 seconds, restarting SPTPS", n->name);
1162 sptps_stop(&n->sptps);
1163 n->status.waitingforkey = false;
1170 static void send_udp_probe_packet(node_t *n, size_t len) {
1171 vpn_packet_t packet;
1173 if(len > sizeof(packet.data)) {
1174 logger(DEBUG_TRAFFIC, LOG_INFO, "Truncating probe length %zu to %s (%s)", len, n->name, n->hostname);
1175 len = sizeof(packet.data);
1178 packet.offset = DEFAULT_PACKET_OFFSET;
1179 memset(DATA(&packet), 0, 14);
1180 randomize(DATA(&packet) + 14, len - 14);
1182 packet.priority = 0;
1184 logger(DEBUG_TRAFFIC, LOG_INFO, "Sending UDP probe length %zu to %s (%s)", len, n->name, n->hostname);
1186 send_udppacket(n, &packet);
1189 // This function tries to establish a UDP tunnel to a node so that packets can be sent.
1190 // If a tunnel is already established, it makes sure it stays up.
1191 // This function makes no guarantees - it is up to the caller to check the node's state to figure out if UDP is usable.
1192 static void try_udp(node_t *n) {
1193 if(!udp_discovery) {
1197 /* Send gratuitous probe replies to 1.1 nodes. */
1199 if((n->options >> 24) >= 3 && n->status.udp_confirmed) {
1200 struct timeval ping_tx_elapsed;
1201 timersub(&now, &n->udp_reply_sent, &ping_tx_elapsed);
1203 if(ping_tx_elapsed.tv_sec >= udp_discovery_keepalive_interval - 1) {
1204 n->udp_reply_sent = now;
1206 if(n->maxrecentlen) {
1208 pkt.len = n->maxrecentlen;
1209 pkt.offset = DEFAULT_PACKET_OFFSET;
1210 memset(DATA(&pkt), 0, 14);
1211 randomize(DATA(&pkt) + 14, MIN_PROBE_SIZE - 14);
1212 send_udp_probe_reply(n, &pkt, pkt.len);
1213 n->maxrecentlen = 0;
1220 struct timeval ping_tx_elapsed;
1221 timersub(&now, &n->udp_ping_sent, &ping_tx_elapsed);
1223 int interval = n->status.udp_confirmed
1224 ? udp_discovery_keepalive_interval
1225 : udp_discovery_interval;
1227 if(ping_tx_elapsed.tv_sec >= interval) {
1228 gettimeofday(&now, NULL);
1229 n->udp_ping_sent = now; // a probe in flight
1230 n->status.ping_sent = true;
1231 send_udp_probe_packet(n, MIN_PROBE_SIZE);
1233 if(localdiscovery && !n->status.udp_confirmed && n->prevedge) {
1234 n->status.send_locally = true;
1235 send_udp_probe_packet(n, MIN_PROBE_SIZE);
1236 n->status.send_locally = false;
1241 static length_t choose_initial_maxmtu(node_t *n) {
1246 const sockaddr_t *sa = NULL;
1248 choose_udp_address(n, &sa, &sockindex);
1254 sock = socket(sa->sa.sa_family, SOCK_DGRAM, IPPROTO_UDP);
1257 logger(DEBUG_TRAFFIC, LOG_ERR, "Creating MTU assessment socket for %s (%s) failed: %s", n->name, n->hostname, sockstrerror(sockerrno));
1261 if(connect(sock, &sa->sa, SALEN(sa->sa))) {
1262 logger(DEBUG_TRAFFIC, LOG_ERR, "Connecting MTU assessment socket for %s (%s) failed: %s", n->name, n->hostname, sockstrerror(sockerrno));
1268 socklen_t ip_mtu_len = sizeof(ip_mtu);
1270 if(getsockopt(sock, IPPROTO_IP, IP_MTU, &ip_mtu, &ip_mtu_len)) {
1271 logger(DEBUG_TRAFFIC, LOG_ERR, "getsockopt(IP_MTU) on %s (%s) failed: %s", n->name, n->hostname, sockstrerror(sockerrno));
1278 /* getsockopt(IP_MTU) returns the MTU of the physical interface.
1279 We need to remove various overheads to get to the tinc MTU. */
1280 length_t mtu = ip_mtu;
1281 mtu -= (sa->sa.sa_family == AF_INET6) ? sizeof(struct ip6_hdr) : sizeof(struct ip);
1284 if(n->status.sptps) {
1285 mtu -= SPTPS_DATAGRAM_OVERHEAD;
1287 if((n->options >> 24) >= 4) {
1288 mtu -= sizeof(node_id_t) + sizeof(node_id_t);
1291 #ifndef DISABLE_LEGACY
1293 mtu -= digest_length(n->outdigest);
1295 /* Now it's tricky. We use CBC mode, so the length of the
1296 encrypted payload must be a multiple of the blocksize. The
1297 sequence number is also part of the encrypted payload, so we
1298 must account for it after correcting for the blocksize.
1299 Furthermore, the padding in the last block must be at least
1302 length_t blocksize = cipher_blocksize(n->outcipher);
1315 logger(DEBUG_TRAFFIC, LOG_ERR, "getsockopt(IP_MTU) on %s (%s) returned absurdly small value: %d", n->name, n->hostname, ip_mtu);
1323 logger(DEBUG_TRAFFIC, LOG_INFO, "Using system-provided maximum tinc MTU for %s (%s): %hd", n->name, n->hostname, mtu);
1332 /* This function tries to determines the MTU of a node.
1333 By calling this function repeatedly, n->minmtu will be progressively
1334 increased, and at some point, n->mtu will be fixed to n->minmtu. If the MTU
1335 is already fixed, this function checks if it can be increased.
1338 static void try_mtu(node_t *n) {
1339 if(!(n->options & OPTION_PMTU_DISCOVERY)) {
1343 if(udp_discovery && !n->status.udp_confirmed) {
1344 n->maxrecentlen = 0;
1351 /* mtuprobes == 0..19: initial discovery, send bursts with 1 second interval, mtuprobes++
1352 mtuprobes == 20: fix MTU, and go to -1
1353 mtuprobes == -1: send one maxmtu and one maxmtu+1 probe every pinginterval
1354 mtuprobes ==-2..-3: send one maxmtu probe every second
1355 mtuprobes == -4: maxmtu no longer valid, reset minmtu and maxmtu and go to 0 */
1357 struct timeval elapsed;
1358 timersub(&now, &n->mtu_ping_sent, &elapsed);
1360 if(n->mtuprobes >= 0) {
1361 if(n->mtuprobes != 0 && elapsed.tv_sec == 0 && elapsed.tv_usec < 333333) {
1365 if(n->mtuprobes < -1) {
1366 if(elapsed.tv_sec < 1) {
1370 if(elapsed.tv_sec < pinginterval) {
1376 n->mtu_ping_sent = now;
1380 if(n->mtuprobes < -3) {
1381 /* We lost three MTU probes, restart discovery */
1382 logger(DEBUG_TRAFFIC, LOG_INFO, "Decrease in PMTU to %s (%s) detected, restarting PMTU discovery", n->name, n->hostname);
1387 if(n->mtuprobes < 0) {
1388 /* After the initial discovery, we only send one maxmtu and one
1389 maxmtu+1 probe to detect PMTU increases. */
1390 send_udp_probe_packet(n, n->maxmtu);
1392 if(n->mtuprobes == -1 && n->maxmtu + 1 < MTU) {
1393 send_udp_probe_packet(n, n->maxmtu + 1);
1398 /* Before initial discovery begins, set maxmtu to the most likely value.
1399 If it's underestimated, we will correct it after initial discovery. */
1400 if(n->mtuprobes == 0) {
1401 n->maxmtu = choose_initial_maxmtu(n);
1405 /* Decreasing the number of probes per cycle might make the algorithm react faster to lost packets,
1406 but it will typically increase convergence time in the no-loss case. */
1407 const length_t probes_per_cycle = 8;
1409 /* This magic value was determined using math simulations.
1410 It will result in a 1329-byte first probe, followed (if there was a reply) by a 1407-byte probe.
1411 Since 1407 is just below the range of tinc MTUs over typical networks,
1412 this fine-tuning allows tinc to cover a lot of ground very quickly.
1413 This fine-tuning is only valid for maxmtu = MTU; if maxmtu is smaller,
1414 then it's better to use a multiplier of 1. Indeed, this leads to an interesting scenario
1415 if choose_initial_maxmtu() returns the actual MTU value - it will get confirmed with one single probe. */
1416 const float multiplier = (n->maxmtu == MTU) ? 0.97f : 1.0f;
1418 const float cycle_position = (float) probes_per_cycle - (float)(n->mtuprobes % probes_per_cycle) - 1.0f;
1419 const length_t minmtu = MAX(n->minmtu, 512);
1420 const float interval = (float)(n->maxmtu - minmtu);
1422 length_t offset = 0;
1424 /* powf can be underflowed if n->maxmtu is less than 512 due to the minmtu MAX bound */
1426 /* The core of the discovery algorithm is this exponential.
1427 It produces very large probes early in the cycle, and then it very quickly decreases the probe size.
1428 This reflects the fact that in the most difficult cases, we don't get any feedback for probes that
1429 are too large, and therefore we need to concentrate on small offsets so that we can quickly converge
1430 on the precise MTU as we are approaching it.
1431 The last probe of the cycle is always 1 byte in size - this is to make sure we'll get at least one
1432 reply per cycle so that we can make progress. */
1433 offset = (length_t) powf(interval, multiplier * cycle_position / ((float) probes_per_cycle - 1.0f));
1436 length_t maxmtu = n->maxmtu;
1437 send_udp_probe_packet(n, minmtu + offset);
1439 /* If maxmtu changed, it means the probe was rejected by the system because it was too large.
1440 In that case, we recalculate with the new maxmtu and try again. */
1441 if(n->mtuprobes < 0 || maxmtu == n->maxmtu) {
1446 if(n->mtuprobes >= 0) {
1452 /* These functions try to establish a tunnel to a node (or its relay) so that
1453 packets can be sent (e.g. exchange keys).
1454 If a tunnel is already established, it tries to improve it (e.g. by trying
1455 to establish a UDP tunnel instead of TCP). This function makes no
1456 guarantees - it is up to the caller to check the node's state to figure out
1457 if TCP and/or UDP is usable. By calling this function repeatedly, the
1458 tunnel is gradually improved until we hit the wall imposed by the underlying
1459 network environment. It is recommended to call this function every time a
1460 packet is sent (or intended to be sent) to a node, so that the tunnel keeps
1461 improving as packets flow, and then gracefully downgrades itself as it goes
1465 static void try_tx_sptps(node_t *n, bool mtu) {
1466 /* If n is a TCP-only neighbor, we'll only use "cleartext" PACKET
1467 messages anyway, so there's no need for SPTPS at all. */
1469 if(n->connection && ((myself->options | n->options) & OPTION_TCPONLY)) {
1473 /* Otherwise, try to do SPTPS authentication with n if necessary. */
1477 /* Do we need to statically relay packets? */
1479 node_t *via = (n->via == myself) ? n->nexthop : n->via;
1481 /* If we do have a static relay, try everything with that one instead, if it supports relaying. */
1484 if((via->options >> 24) < 4) {
1492 /* Otherwise, try to establish UDP connectivity. */
1500 /* If we don't have UDP connectivity (yet), we need to use a dynamic relay (nexthop)
1501 while we try to establish direct connectivity. */
1503 if(!n->status.udp_confirmed && n != n->nexthop && (n->nexthop->options >> 24) >= 4) {
1504 try_tx(n->nexthop, mtu);
1508 static void try_tx_legacy(node_t *n, bool mtu) {
1509 /* Does he have our key? If not, send one. */
1511 if(!n->status.validkey_in) {
1515 /* Check if we already have a key, or request one. */
1517 if(!n->status.validkey) {
1518 if(n->last_req_key + 10 <= now.tv_sec) {
1520 n->last_req_key = now.tv_sec;
1533 void try_tx(node_t *n, bool mtu) {
1534 if(!n->status.reachable) {
1538 if(n->status.sptps) {
1539 try_tx_sptps(n, mtu);
1541 try_tx_legacy(n, mtu);
1545 void send_packet(node_t *n, vpn_packet_t *packet) {
1546 // If it's for myself, write it to the tun/tap device.
1550 memcpy(DATA(packet), mymac.x, ETH_ALEN);
1551 // Use an arbitrary fake source address.
1552 memcpy(DATA(packet) + ETH_ALEN, DATA(packet), ETH_ALEN);
1553 DATA(packet)[ETH_ALEN * 2 - 1] ^= 0xFF;
1557 n->out_bytes += packet->len;
1558 devops.write(packet);
1562 logger(DEBUG_TRAFFIC, LOG_ERR, "Sending packet of %d bytes to %s (%s)", packet->len, n->name, n->hostname);
1564 // If the node is not reachable, drop it.
1566 if(!n->status.reachable) {
1567 logger(DEBUG_TRAFFIC, LOG_INFO, "Node %s (%s) is not reachable", n->name, n->hostname);
1571 // Keep track of packet statistics.
1574 n->out_bytes += packet->len;
1576 // Check if it should be sent as an SPTPS packet.
1578 if(n->status.sptps) {
1579 send_sptps_packet(n, packet);
1584 // Determine which node to actually send it to.
1586 node_t *via = (packet->priority == -1 || n->via == myself) ? n->nexthop : n->via;
1589 logger(DEBUG_TRAFFIC, LOG_INFO, "Sending packet to %s via %s (%s)", n->name, via->name, n->via->hostname);
1592 // Try to send via UDP, unless TCP is forced.
1594 if(packet->priority == -1 || ((myself->options | via->options) & OPTION_TCPONLY)) {
1595 if(!send_tcppacket(via->connection, packet)) {
1596 terminate_connection(via->connection, true);
1602 send_udppacket(via, packet);
1606 void broadcast_packet(const node_t *from, vpn_packet_t *packet) {
1607 // Always give ourself a copy of the packet.
1608 if(from != myself) {
1609 send_packet(myself, packet);
1612 // In TunnelServer mode, do not forward broadcast packets.
1613 // The MST might not be valid and create loops.
1614 if(tunnelserver || broadcast_mode == BMODE_NONE) {
1618 logger(DEBUG_TRAFFIC, LOG_INFO, "Broadcasting packet of %d bytes from %s (%s)",
1619 packet->len, from->name, from->hostname);
1621 switch(broadcast_mode) {
1622 // In MST mode, broadcast packets travel via the Minimum Spanning Tree.
1623 // This guarantees all nodes receive the broadcast packet, and
1624 // usually distributes the sending of broadcast packets over all nodes.
1626 for list_each(connection_t, c, &connection_list)
1627 if(c->edge && c->status.mst && c != from->nexthop->connection) {
1628 send_packet(c->node, packet);
1633 // In direct mode, we send copies to each node we know of.
1634 // However, this only reaches nodes that can be reached in a single hop.
1635 // We don't have enough information to forward broadcast packets in this case.
1637 if(from != myself) {
1641 for splay_each(node_t, n, &node_tree)
1642 if(n->status.reachable && n != myself && ((n->via == myself && n->nexthop == n) || n->via == n)) {
1643 send_packet(n, packet);
1653 /* We got a packet from some IP address, but we don't know who sent it. Try to
1654 verify the message authentication code against all active session keys.
1655 Since this is actually an expensive operation, we only do a full check once
1656 a minute, the rest of the time we only check against nodes for which we know
1657 an IP address that matches the one from the packet. */
1659 static node_t *try_harder(const sockaddr_t *from, const vpn_packet_t *pkt) {
1660 node_t *match = NULL;
1662 static time_t last_hard_try = 0;
1664 for splay_each(node_t, n, &node_tree) {
1665 if(!n->status.reachable || n == myself) {
1669 if(!n->status.validkey_in && !(n->status.sptps && n->sptps.instate)) {
1675 for splay_each(edge_t, e, &n->edge_tree) {
1680 if(!sockaddrcmp_noport(from, &e->reverse->address)) {
1687 if(last_hard_try == now.tv_sec) {
1694 if(!try_mac(n, pkt)) {
1703 last_hard_try = now.tv_sec;
1709 static void handle_incoming_vpn_packet(listen_socket_t *ls, vpn_packet_t *pkt, sockaddr_t *addr) {
1711 node_id_t nullid = {0};
1713 bool direct = false;
1715 sockaddrunmap(addr); /* Some braindead IPv6 implementations do stupid things. */
1717 // Try to figure out who sent this packet.
1719 node_t *n = lookup_node_udp(addr);
1721 if(n && !n->status.udp_confirmed) {
1722 n = NULL; // Don't believe it if we don't have confirmation yet.
1726 // It might be from a 1.1 node, which might have a source ID in the packet.
1727 pkt->offset = 2 * sizeof(node_id_t);
1728 from = lookup_node_id(SRCID(pkt));
1730 if(from && from->status.sptps && !memcmp(DSTID(pkt), &nullid, sizeof(nullid))) {
1731 if(sptps_verify_datagram(&from->sptps, DATA(pkt), pkt->len - 2 * sizeof(node_id_t))) {
1741 n = try_harder(addr, pkt);
1747 if(debug_level >= DEBUG_PROTOCOL) {
1748 hostname = sockaddr2hostname(addr);
1749 logger(DEBUG_PROTOCOL, LOG_WARNING, "Received UDP packet from unknown source %s", hostname);
1758 if(n->status.sptps) {
1759 bool relay_enabled = (n->options >> 24) >= 4;
1762 pkt->offset = 2 * sizeof(node_id_t);
1763 pkt->len -= pkt->offset;
1766 if(!relay_enabled || !memcmp(DSTID(pkt), &nullid, sizeof(nullid))) {
1771 from = lookup_node_id(SRCID(pkt));
1772 to = lookup_node_id(DSTID(pkt));
1776 logger(DEBUG_PROTOCOL, LOG_WARNING, "Received UDP packet from %s (%s) with unknown source and/or destination ID", n->name, n->hostname);
1780 if(!to->status.reachable) {
1781 /* This can happen in the form of a race condition
1782 if the node just became unreachable. */
1783 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);
1787 /* The packet is supposed to come from the originator or its static relay
1788 (i.e. with no dynamic relays in between).
1789 If it did not, "help" the static relay by sending it UDP info.
1790 Note that we only do this if we're the destination or the static relay;
1791 otherwise every hop would initiate its own UDP info message, resulting in elevated chatter. */
1793 if(n != from->via && to->via == myself) {
1794 send_udp_info(myself, from);
1797 /* If we're not the final recipient, relay the packet. */
1800 send_sptps_data(to, from, 0, DATA(pkt), pkt->len);
1809 if(!receive_udppacket(from, pkt)) {
1813 n->sock = ls - listen_socket;
1815 if(direct && sockaddrcmp(addr, &n->address)) {
1816 update_node_udp(n, addr);
1819 /* If the packet went through a relay, help the sender find the appropriate MTU
1820 through the relay path. */
1823 send_mtu_info(myself, n, MTU);
1827 void handle_incoming_vpn_data(void *data, int flags) {
1830 listen_socket_t *ls = data;
1832 #ifdef HAVE_RECVMMSG
1834 static ssize_t num = MAX_MSG;
1835 static vpn_packet_t pkt[MAX_MSG];
1836 static sockaddr_t addr[MAX_MSG];
1837 static struct mmsghdr msg[MAX_MSG];
1838 static struct iovec iov[MAX_MSG];
1840 for(int i = 0; i < num; i++) {
1843 iov[i] = (struct iovec) {
1844 .iov_base = DATA(&pkt[i]),
1848 msg[i].msg_hdr = (struct msghdr) {
1849 .msg_name = &addr[i].sa,
1850 .msg_namelen = sizeof(addr)[i],
1856 num = recvmmsg(ls->udp.fd, msg, MAX_MSG, MSG_DONTWAIT, NULL);
1859 if(!sockwouldblock(sockerrno)) {
1860 logger(DEBUG_ALWAYS, LOG_ERR, "Receiving packet failed: %s", sockstrerror(sockerrno));
1866 for(int i = 0; i < num; i++) {
1867 pkt[i].len = msg[i].msg_len;
1869 if(pkt[i].len <= 0 || pkt[i].len > MAXSIZE) {
1873 handle_incoming_vpn_packet(ls, &pkt[i], &addr[i]);
1878 sockaddr_t addr = {0};
1879 socklen_t addrlen = sizeof(addr);
1882 ssize_t len = recvfrom(ls->udp.fd, (void *)DATA(&pkt), MAXSIZE, 0, &addr.sa, &addrlen);
1884 if(len <= 0 || (size_t)len > MAXSIZE) {
1885 if(!sockwouldblock(sockerrno)) {
1886 logger(DEBUG_ALWAYS, LOG_ERR, "Receiving packet failed: %s", sockstrerror(sockerrno));
1894 handle_incoming_vpn_packet(ls, &pkt, &addr);
1898 void handle_device_data(void *data, int flags) {
1901 vpn_packet_t packet;
1902 packet.offset = DEFAULT_PACKET_OFFSET;
1903 packet.priority = 0;
1904 static int errors = 0;
1906 if(devops.read(&packet)) {
1908 myself->in_packets++;
1909 myself->in_bytes += packet.len;
1910 route(myself, &packet);
1912 usleep(errors * 50000);
1916 logger(DEBUG_ALWAYS, LOG_ERR, "Too many errors from %s, exiting!", device);