+// This function tries to get SPTPS keys, if they aren't already known.
+// 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.
+static void try_sptps(node_t *n) {
+ if(n->status.validkey)
+ return;
+
+ logger(DEBUG_TRAFFIC, LOG_INFO, "No valid key known yet for %s (%s)", n->name, n->hostname);
+
+ if(!n->status.waitingforkey)
+ send_req_key(n);
+ else if(n->last_req_key + 10 < now.tv_sec) {
+ logger(DEBUG_ALWAYS, LOG_DEBUG, "No key from %s after 10 seconds, restarting SPTPS", n->name);
+ sptps_stop(&n->sptps);
+ n->status.waitingforkey = false;
+ send_req_key(n);
+ }
+
+ return;
+}
+
+static void send_udp_probe_packet(node_t *n, int len) {
+ vpn_packet_t packet;
+ packet.offset = DEFAULT_PACKET_OFFSET;
+ memset(DATA(&packet), 0, 14);
+ randomize(DATA(&packet) + 14, len - 14);
+ packet.len = len;
+ packet.priority = 0;
+
+ logger(DEBUG_TRAFFIC, LOG_INFO, "Sending UDP probe length %d to %s (%s)", len, n->name, n->hostname);
+
+ send_udppacket(n, &packet);
+}
+
+// This function tries to establish a UDP tunnel to a node so that packets can be sent.
+// If a tunnel is already established, it makes sure it stays up.
+// This function makes no guarantees - it is up to the caller to check the node's state to figure out if UDP is usable.
+static void try_udp(node_t* n) {
+ if(!udp_discovery)
+ return;
+
+ struct timeval ping_tx_elapsed;
+ timersub(&now, &n->udp_ping_sent, &ping_tx_elapsed);
+
+ int interval = n->status.udp_confirmed ? udp_discovery_keepalive_interval : udp_discovery_interval;
+
+ if(ping_tx_elapsed.tv_sec >= interval) {
+ send_udp_probe_packet(n, MAX(n->minmtu, 16));
+ n->udp_ping_sent = now;
+
+ if(localdiscovery && !n->status.udp_confirmed && n->prevedge) {
+ n->status.send_locally = true;
+ send_udp_probe_packet(n, 16);
+ n->status.send_locally = false;
+ }
+ }
+}
+
+static length_t choose_initial_maxmtu(node_t *n) {
+#ifdef IP_MTU
+
+ int sock = -1;
+
+ const sockaddr_t *sa = NULL;
+ int sockindex;
+ choose_udp_address(n, &sa, &sockindex);
+ if(!sa)
+ return MTU;
+
+ sock = socket(sa->sa.sa_family, SOCK_DGRAM, IPPROTO_UDP);
+ if(sock < 0) {
+ logger(DEBUG_TRAFFIC, LOG_ERR, "Creating MTU assessment socket for %s (%s) failed: %s", n->name, n->hostname, sockstrerror(sockerrno));
+ return MTU;
+ }
+
+ if(connect(sock, &sa->sa, SALEN(sa->sa))) {
+ logger(DEBUG_TRAFFIC, LOG_ERR, "Connecting MTU assessment socket for %s (%s) failed: %s", n->name, n->hostname, sockstrerror(sockerrno));
+ close(sock);
+ return MTU;
+ }
+
+ int ip_mtu;
+ socklen_t ip_mtu_len = sizeof ip_mtu;
+ if(getsockopt(sock, IPPROTO_IP, IP_MTU, &ip_mtu, &ip_mtu_len)) {
+ logger(DEBUG_TRAFFIC, LOG_ERR, "getsockopt(IP_MTU) on %s (%s) failed: %s", n->name, n->hostname, sockstrerror(sockerrno));
+ close(sock);
+ return MTU;
+ }
+
+ close(sock);
+
+ /* getsockopt(IP_MTU) returns the MTU of the physical interface.
+ We need to remove various overheads to get to the tinc MTU. */
+ length_t mtu = ip_mtu;
+ mtu -= (sa->sa.sa_family == AF_INET6) ? sizeof(struct ip6_hdr) : sizeof(struct ip);
+ mtu -= 8; /* UDP */
+ if(n->status.sptps) {
+ mtu -= SPTPS_DATAGRAM_OVERHEAD;
+ if((n->options >> 24) >= 4)
+ mtu -= sizeof(node_id_t) + sizeof(node_id_t);
+ }
+
+ if (mtu < 512) {
+ logger(DEBUG_TRAFFIC, LOG_ERR, "getsockopt(IP_MTU) on %s (%s) returned absurdly small value: %d", n->name, n->hostname, ip_mtu);
+ return MTU;
+ }
+ if (mtu > MTU)
+ return MTU;
+
+ logger(DEBUG_TRAFFIC, LOG_INFO, "Using system-provided maximum tinc MTU for %s (%s): %hd", n->name, n->hostname, mtu);
+ return mtu;
+
+#else
+
+ return MTU;
+
+#endif
+}
+
+// This function tries to determines the MTU of a node.
+// By calling this function repeatedly, n->minmtu will be progressively increased, and at some point, n->mtu will be fixed to n->minmtu.
+// If the MTU is already fixed, this function checks if it can be increased.
+static void try_mtu(node_t *n) {
+ if(!(n->options & OPTION_PMTU_DISCOVERY))
+ return;
+
+ if(udp_discovery && !n->status.udp_confirmed) {
+ n->mtuprobes = 0;
+ n->minmtu = 0;
+ n->maxmtu = MTU;
+ return;
+ }
+
+ /* mtuprobes == 0..19: initial discovery, send bursts with 1 second interval, mtuprobes++
+ mtuprobes == 20: fix MTU, and go to -1
+ mtuprobes == -1: send one >maxmtu probe every pingtimeout */
+
+ struct timeval elapsed;
+ timersub(&now, &n->probe_sent_time, &elapsed);
+ if(n->mtuprobes >= 0) {
+ if(n->mtuprobes != 0 && elapsed.tv_sec == 0 && elapsed.tv_usec < 333333)
+ return;
+ } else {
+ if(elapsed.tv_sec < pingtimeout)
+ return;
+ }
+
+ try_fix_mtu(n);
+
+ if(n->mtuprobes < 0) {
+ /* After the initial discovery, we only send one >maxmtu probe
+ to detect PMTU increases. */
+ if(n->maxmtu + 1 < MTU)
+ send_udp_probe_packet(n, n->maxmtu + 1);
+ } else {
+ /* Before initial discovery begins, set maxmtu to the most likely value.
+ If it's underestimated, we will correct it after initial discovery. */
+ if(n->mtuprobes == 0)
+ n->maxmtu = choose_initial_maxmtu(n);
+
+ for (;;) {
+ /* Decreasing the number of probes per cycle might make the algorithm react faster to lost packets,
+ but it will typically increase convergence time in the no-loss case. */
+ const length_t probes_per_cycle = 8;
+
+ /* This magic value was determined using math simulations.
+ It will result in a 1329-byte first probe, followed (if there was a reply) by a 1407-byte probe.
+ Since 1407 is just below the range of tinc MTUs over typical networks,
+ this fine-tuning allows tinc to cover a lot of ground very quickly.
+ This fine-tuning is only valid for maxmtu = MTU; if maxmtu is smaller,
+ then it's better to use a multiplier of 1. Indeed, this leads to an interesting scenario
+ if choose_initial_maxmtu() returns the actual MTU value - it will get confirmed with one single probe. */
+ const float multiplier = (n->maxmtu == MTU) ? 0.97 : 1;
+
+ const float cycle_position = probes_per_cycle - (n->mtuprobes % probes_per_cycle) - 1;
+ const length_t minmtu = MAX(n->minmtu, 512);
+ const float interval = n->maxmtu - minmtu;
+
+ /* The core of the discovery algorithm is this exponential.
+ It produces very large probes early in the cycle, and then it very quickly decreases the probe size.
+ This reflects the fact that in the most difficult cases, we don't get any feedback for probes that
+ are too large, and therefore we need to concentrate on small offsets so that we can quickly converge
+ on the precise MTU as we are approaching it.
+ The last probe of the cycle is always 1 byte in size - this is to make sure we'll get at least one
+ reply per cycle so that we can make progress. */
+ const length_t offset = powf(interval, multiplier * cycle_position / (probes_per_cycle - 1));
+
+ length_t maxmtu = n->maxmtu;
+ send_udp_probe_packet(n, minmtu + offset);
+ /* If maxmtu changed, it means the probe was rejected by the system because it was too large.
+ In that case, we recalculate with the new maxmtu and try again. */
+ if(n->mtuprobes < 0 || maxmtu == n->maxmtu)
+ break;
+ }
+
+ if(n->mtuprobes >= 0)
+ n->mtuprobes++;
+ }
+
+ n->probe_counter = 0;
+ n->probe_sent_time = now;
+ n->probe_time = now;
+
+ /* Calculate the packet loss of incoming traffic by comparing the rate of
+ packets received to the rate with which the sequence number has increased.
+ TODO: this is unrelated to PMTU discovery - it should be moved elsewhere.
+ */
+
+ if(n->received > n->prev_received)
+ n->packetloss = 1.0 - (n->received - n->prev_received) / (float)(n->received_seqno - n->prev_received_seqno);
+ else
+ n->packetloss = n->received_seqno <= n->prev_received_seqno;
+
+ n->prev_received_seqno = n->received_seqno;
+ n->prev_received = n->received;
+}
+
+// This function tries to establish a tunnel to a node (or its relay) so that packets can be sent (e.g. get SPTPS keys).
+// If a tunnel is already established, it tries to improve it (e.g. by trying to establish a UDP tunnel instead of TCP).
+// This function makes no guarantees - it is up to the caller to check the node's state to figure out if TCP and/or UDP is usable.
+// By calling this function repeatedly, the tunnel is gradually improved until we hit the wall imposed by the underlying network environment.
+// It is recommended to call this function every time a packet is sent (or intended to be sent) to a node,
+// so that the tunnel keeps improving as packets flow, and then gracefully downgrades itself as it goes idle.
+static void try_tx(node_t *n) {
+ /* If n is a TCP-only neighbor, we'll only use "cleartext" PACKET
+ messages anyway, so there's no need for SPTPS at all. Otherwise, get the keys. */
+ if(n->status.sptps && !(n->connection && ((myself->options | n->options) & OPTION_TCPONLY))) {
+ try_sptps(n);
+ if (!n->status.validkey)
+ return;
+ }
+
+ node_t *via = (n->via == myself) ? n->nexthop : n->via;
+
+ if((myself->options | via->options) & OPTION_TCPONLY)
+ return;
+
+ if(!n->status.sptps && !via->status.validkey && via->last_req_key + 10 <= now.tv_sec) {
+ send_req_key(via);
+ via->last_req_key = now.tv_sec;
+ } else if(via == n || !n->status.sptps || (via->options >> 24) >= 4) {
+ try_udp(via);
+ try_mtu(via);
+ }
+
+ /* If we don't know how to reach "via" yet, then try to reach it through a relay. */
+ if(n->status.sptps && !via->status.udp_confirmed && via->nexthop != via && (via->nexthop->options >> 24) >= 4)
+ try_tx(via->nexthop);
+}
+