#define MAX_SEQNO 1073741824
-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 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;
+static void try_fix_mtu(node_t *n) {
+ if(n->mtuprobes < 0)
return;
- }
-
- /* mtuprobes == 0..29: initial discovery, send bursts with 1 second interval, mtuprobes++
- mtuprobes == 30: fix MTU, and go to 31
- mtuprobes == 31: send one >maxmtu probe every pingtimeout */
-
- struct timeval now;
- gettimeofday(&now, NULL);
- struct timeval elapsed;
- timersub(&now, &n->probe_sent_time, &elapsed);
- if(n->mtuprobes < 31) {
- if(n->mtuprobes != 0 && elapsed.tv_sec < 1)
- return;
- } else {
- if(elapsed.tv_sec < pingtimeout)
- return;
- }
- if(n->mtuprobes == 30 || (n->mtuprobes < 30 && n->minmtu >= n->maxmtu)) {
+ if(n->mtuprobes == 20 || n->minmtu >= n->maxmtu) {
if(n->minmtu > n->maxmtu)
n->minmtu = n->maxmtu;
else
n->maxmtu = n->minmtu;
n->mtu = n->minmtu;
logger(DEBUG_TRAFFIC, LOG_INFO, "Fixing MTU of %s (%s) to %d after %d probes", n->name, n->hostname, n->mtu, n->mtuprobes);
- n->mtuprobes = 31;
- }
-
- int timeout;
- if(n->mtuprobes == 31) {
- /* After the initial discovery, we only send one >maxmtu probe
- to detect PMTU increases. */
- if(n->maxmtu + 8 < MTU)
- send_udp_probe_packet(n, n->maxmtu + 8);
- } else {
- /* Probes are sent in batches of three, with random sizes between the
- lower and upper boundaries for the MTU thus far discovered. */
- for (int i = 0; i < 3; i++) {
- int len = n->maxmtu;
- if(n->minmtu < n->maxmtu)
- len = n->minmtu + 1 + rand() % (n->maxmtu - n->minmtu);
-
- send_udp_probe_packet(n, MAX(len, 64));
- }
- n->mtuprobes++;
+ n->mtuprobes = -1;
}
-
- 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;
}
static void udp_probe_timeout_handler(void *data) {
timeout_add(&n->udp_ping_timeout, &udp_probe_timeout_handler, n, &(struct timeval){udp_discovery_timeout, 0});
}
- if(probelen >= n->maxmtu + 8) {
+ if(probelen >= n->maxmtu + 1) {
logger(DEBUG_TRAFFIC, LOG_INFO, "Increase in PMTU to %s (%s) detected, restarting PMTU discovery", n->name, n->hostname);
n->maxmtu = MTU;
- n->mtuprobes = 10;
+ n->mtuprobes = 0;
return;
}
if(probelen > n->maxmtu)
probelen = n->maxmtu;
- if(n->minmtu < probelen)
+ if(n->minmtu < probelen) {
n->minmtu = probelen;
+ try_fix_mtu(n);
+ }
- /* Calculate RTT and bandwidth.
+ /* Calculate RTT.
The RTT is the time between the MTU probe burst was sent and the first
- reply is received. The bandwidth is measured using the time between the
- arrival of the first and third probe reply (or type 2 probe requests).
+ reply is received.
*/
struct timeval now, diff;
if(n->probe_counter == 1) {
n->rtt = diff.tv_sec + diff.tv_usec * 1e-6;
n->probe_time = probe_timestamp;
- } else if(n->probe_counter == 3) {
- /* TODO: this will never fire after initial MTU discovery. */
- struct timeval probe_timestamp_diff;
- timersub(&probe_timestamp, &n->probe_time, &probe_timestamp_diff);
- n->bandwidth = 2.0 * probelen / (probe_timestamp_diff.tv_sec + probe_timestamp_diff.tv_usec * 1e-6);
- logger(DEBUG_TRAFFIC, LOG_DEBUG, "%s (%s) RTT %.2f ms, burst bandwidth %.3f Mbit/s, rx packet loss %.2f %%", n->name, n->hostname, n->rtt * 1e3, n->bandwidth * 8e-6, n->packetloss * 1e2);
+ logger(DEBUG_TRAFFIC, LOG_DEBUG, "%s (%s) RTT %.2f ms, rx packet loss %.2f %%", n->name, n->hostname, n->rtt * 1e3, n->packetloss * 1e2);
}
}
}
n->maxmtu = origlen - 1;
if(n->mtu >= origlen)
n->mtu = origlen - 1;
+ try_fix_mtu(n);
} else
logger(DEBUG_TRAFFIC, LOG_WARNING, "Error sending packet to %s (%s): %s", n->name, n->hostname, sockstrerror(sockerrno));
}
relay->maxmtu = len - 1;
if(relay->mtu >= len)
relay->mtu = len - 1;
+ try_fix_mtu(relay);
} else {
logger(DEBUG_TRAFFIC, LOG_WARNING, "Error sending UDP SPTPS packet to %s (%s): %s", relay->name, relay->hostname, sockstrerror(sockerrno));
return false;
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.
}
}
+// 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 now;
+ gettimeofday(&now, NULL);
+ 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);
+
+ int timeout;
+ 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 {
+ /* 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. */
+ const float multiplier = 0.97;
+
+ 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));
+
+ send_udp_probe_packet(n, minmtu + offset);
+ 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.
projects / tinc / blobdiff