Fix warnings when compiling for Windows.
[tinc] / src / net_packet.c
index 84d4915..5a85642 100644 (file)
@@ -1,7 +1,7 @@
 /*
     net_packet.c -- Handles in- and outgoing VPN packets
     Copyright (C) 1998-2005 Ivo Timmermans,
-                  2000-2014 Guus Sliepen <guus@tinc-vpn.org>
+                  2000-2018 Guus Sliepen <guus@tinc-vpn.org>
                   2010      Timothy Redaelli <timothy@redaelli.eu>
                   2010      Brandon Black <blblack@gmail.com>
 
@@ -23,6 +23,7 @@
 #include "system.h"
 
 #ifdef HAVE_ZLIB
+#define ZLIB_CONST
 #include <zlib.h>
 #endif
 
@@ -30,6 +31,7 @@
 #include LZO1X_H
 #endif
 
+#include "address_cache.h"
 #include "cipher.h"
 #include "conf.h"
 #include "connection.h"
 #define MAX(a, b) ((a) > (b) ? (a) : (b))
 #endif
 
+/* The minimum size of a probe is 14 bytes, but since we normally use CBC mode
+   encryption, we can add a few extra random bytes without increasing the
+   resulting packet size. */
+#define MIN_PROBE_SIZE 18
+
 int keylifetime = 0;
 #ifdef HAVE_LZO
 static char lzo_wrkmem[LZO1X_999_MEM_COMPRESS > LZO1X_1_MEM_COMPRESS ? LZO1X_999_MEM_COMPRESS : LZO1X_1_MEM_COMPRESS];
@@ -59,23 +66,27 @@ static char lzo_wrkmem[LZO1X_999_MEM_COMPRESS > LZO1X_1_MEM_COMPRESS ? LZO1X_999
 
 static void send_udppacket(node_t *, vpn_packet_t *);
 
-unsigned replaywin = 16;
+unsigned replaywin = 32;
 bool localdiscovery = true;
 bool udp_discovery = true;
-int udp_discovery_interval = 9;
+int udp_discovery_keepalive_interval = 10;
+int udp_discovery_interval = 2;
 int udp_discovery_timeout = 30;
 
 #define MAX_SEQNO 1073741824
 
 static void try_fix_mtu(node_t *n) {
-       if(n->mtuprobes < 0)
+       if(n->mtuprobes < 0) {
                return;
+       }
 
        if(n->mtuprobes == 20 || n->minmtu >= n->maxmtu) {
-               if(n->minmtu > n->maxmtu)
+               if(n->minmtu > n->maxmtu) {
                        n->minmtu = n->maxmtu;
-               else
+               } 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 = -1;
@@ -84,106 +95,114 @@ static void try_fix_mtu(node_t *n) {
 
 static void udp_probe_timeout_handler(void *data) {
        node_t *n = data;
-       if(!n->status.udp_confirmed)
+
+       if(!n->status.udp_confirmed) {
                return;
+       }
 
        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);
        n->status.udp_confirmed = false;
+       n->udp_ping_rtt = -1;
+       n->maxrecentlen = 0;
        n->mtuprobes = 0;
        n->minmtu = 0;
        n->maxmtu = MTU;
 }
 
+static void send_udp_probe_reply(node_t *n, vpn_packet_t *packet, length_t len) {
+       if(!n->status.sptps && !n->status.validkey) {
+               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);
+               return;
+       }
+
+       /* Type 2 probe replies were introduced in protocol 17.3 */
+       if((n->options >> 24) >= 3) {
+               DATA(packet)[0] = 2;
+               uint16_t len16 = htons(len);
+               memcpy(DATA(packet) + 1, &len16, 2);
+               packet->len = MIN_PROBE_SIZE;
+               logger(DEBUG_TRAFFIC, LOG_INFO, "Sending type 2 probe reply length %u to %s (%s)", len, n->name, n->hostname);
+
+       } else {
+               /* Legacy protocol: n won't understand type 2 probe replies. */
+               DATA(packet)[0] = 1;
+               logger(DEBUG_TRAFFIC, LOG_INFO, "Sending type 1 probe reply length %u to %s (%s)", len, n->name, n->hostname);
+       }
+
+       /* Temporarily set udp_confirmed, so that the reply is sent
+          back exactly the way it came in. */
+
+       bool udp_confirmed = n->status.udp_confirmed;
+       n->status.udp_confirmed = true;
+       send_udppacket(n, packet);
+       n->status.udp_confirmed = udp_confirmed;
+}
+
 static void udp_probe_h(node_t *n, vpn_packet_t *packet, length_t len) {
        if(!DATA(packet)[0]) {
                logger(DEBUG_TRAFFIC, LOG_INFO, "Got UDP probe request %d from %s (%s)", packet->len, n->name, n->hostname);
+               send_udp_probe_reply(n, packet, len);
+               return;
+       }
 
-               /* It's a probe request, send back a reply */
-
-               /* Type 2 probe replies were introduced in protocol 17.3 */
-               if ((n->options >> 24) >= 3) {
-                       uint8_t *data = DATA(packet);
-                       *data++ = 2;
-                       uint16_t len16 = htons(len); memcpy(data, &len16, 2); data += 2;
-                       struct timeval now;
-                       gettimeofday(&now, NULL);
-                       uint32_t sec = htonl(now.tv_sec); memcpy(data, &sec, 4); data += 4;
-                       uint32_t usec = htonl(now.tv_usec); memcpy(data, &usec, 4); data += 4;
-                       packet->len -= 10;
-               } else {
-                       /* Legacy protocol: n won't understand type 2 probe replies. */
-                       DATA(packet)[0] = 1;
-               }
-
-               /* Temporarily set udp_confirmed, so that the reply is sent
-                  back exactly the way it came in. */
+       if(DATA(packet)[0] == 2) {
+               // It's a type 2 probe reply, use the length field inside the packet
+               uint16_t len16;
+               memcpy(&len16, DATA(packet) + 1, 2);
+               len = ntohs(len16);
+       }
 
-               bool udp_confirmed = n->status.udp_confirmed;
-               n->status.udp_confirmed = true;
-               send_udppacket(n, packet);
-               n->status.udp_confirmed = udp_confirmed;
+       if(n->udp_ping_sent.tv_sec != 0) {  // a probe in flight
+               gettimeofday(&now, NULL);
+               struct timeval rtt;
+               timersub(&now, &n->udp_ping_sent, &rtt);
+               n->udp_ping_rtt = rtt.tv_sec * 1000000 + rtt.tv_usec;
+               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);
        } else {
-               length_t probelen = len;
-               if (DATA(packet)[0] == 2) {
-                       if (len < 3)
-                               logger(DEBUG_TRAFFIC, LOG_WARNING, "Received invalid (too short) UDP probe reply from %s (%s)", n->name, n->hostname);
-                       else {
-                               uint16_t probelen16; memcpy(&probelen16, DATA(packet) + 1, 2); probelen = ntohs(probelen16);
-                       }
-               }
-               logger(DEBUG_TRAFFIC, LOG_INFO, "Got type %d UDP probe reply %d from %s (%s)", DATA(packet)[0], probelen, n->name, n->hostname);
+               logger(DEBUG_TRAFFIC, LOG_INFO, "Got type %d UDP probe reply %d from %s (%s)", DATA(packet)[0], len, n->name, n->hostname);
+       }
 
-               /* It's a valid reply: now we know bidirectional communication
-                  is possible using the address and socket that the reply
-                  packet used. */
+       /* It's a valid reply: now we know bidirectional communication
+          is possible using the address and socket that the reply
+          packet used. */
+       if(!n->status.udp_confirmed) {
                n->status.udp_confirmed = true;
 
-               if(udp_discovery) {
-                       timeout_del(&n->udp_ping_timeout);
-                       timeout_add(&n->udp_ping_timeout, &udp_probe_timeout_handler, n, &(struct timeval){udp_discovery_timeout, 0});
+               if(!n->address_cache) {
+                       n->address_cache = open_address_cache(n);
                }
 
-               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;
-                       /* Set mtuprobes to 1 so that try_mtu() doesn't reset maxmtu */
-                       n->mtuprobes = 1;
-                       return;
-               }
+               reset_address_cache(n->address_cache, &n->address);
+       }
 
-               /* If applicable, raise the minimum supported MTU */
+       // Reset the UDP ping timer. (no probe in flight)
+       n->udp_ping_sent.tv_sec = 0;
 
-               if(probelen > n->maxmtu)
-                       probelen = n->maxmtu;
-               if(n->minmtu < probelen) {
-                       n->minmtu = probelen;
-                       try_fix_mtu(n);
-               }
-
-               /* Calculate RTT.
-                  The RTT is the time between the MTU probe burst was sent and the first
-                  reply is received.
-                */
+       if(udp_discovery) {
+               timeout_del(&n->udp_ping_timeout);
+               timeout_add(&n->udp_ping_timeout, &udp_probe_timeout_handler, n, &(struct timeval) {
+                       udp_discovery_timeout, 0
+               });
+       }
 
-               struct timeval now, diff;
-               gettimeofday(&now, NULL);
-               timersub(&now, &n->probe_time, &diff);
+       if(len > n->maxmtu) {
+               logger(DEBUG_TRAFFIC, LOG_INFO, "Increase in PMTU to %s (%s) detected, restarting PMTU discovery", n->name, n->hostname);
+               n->minmtu = len;
+               n->maxmtu = MTU;
+               /* Set mtuprobes to 1 so that try_mtu() doesn't reset maxmtu */
+               n->mtuprobes = 1;
+               return;
+       } else if(n->mtuprobes < 0 && len == n->maxmtu) {
+               /* We got a maxmtu sized packet, confirming the PMTU is still valid. */
+               n->mtuprobes = -1;
+               n->mtu_ping_sent = now;
+       }
 
-               struct timeval probe_timestamp = now;
-               if (DATA(packet)[0] == 2 && packet->len >= 11) {
-                       uint32_t sec; memcpy(&sec, DATA(packet) + 3, 4);
-                       uint32_t usec; memcpy(&usec, DATA(packet) + 7, 4);
-                       probe_timestamp.tv_sec = ntohl(sec);
-                       probe_timestamp.tv_usec = ntohl(usec);
-               }
-               
-               n->probe_counter++;
+       /* If applicable, raise the minimum supported MTU */
 
-               if(n->probe_counter == 1) {
-                       n->rtt = diff.tv_sec + diff.tv_usec * 1e-6;
-                       n->probe_time = probe_timestamp;
-                       logger(DEBUG_TRAFFIC, LOG_DEBUG, "%s (%s) RTT %.2f ms, rx packet loss %.2f %%", n->name, n->hostname, n->rtt * 1e3, n->packetloss * 1e2);
-               }
+       if(n->minmtu < len) {
+               n->minmtu = len;
+               try_fix_mtu(n);
        }
 }
 
@@ -197,27 +216,28 @@ static length_t compress_packet(uint8_t *dest, const uint8_t *source, length_t l
                lzo1x_1_compress(source, len, dest, &lzolen, lzo_wrkmem);
                return lzolen;
 #else
-               return -1;
+               return 0;
 #endif
        } else if(level < 10) {
 #ifdef HAVE_ZLIB
                unsigned long destlen = MAXSIZE;
-               if(compress2(dest, &destlen, source, len, level) == Z_OK)
+
+               if(compress2(dest, &destlen, source, len, level) == Z_OK) {
                        return destlen;
-               else
+               else
 #endif
-                       return -1;
+                       return 0;
        } else {
 #ifdef HAVE_LZO
                lzo_uint lzolen = MAXSIZE;
                lzo1x_999_compress(source, len, dest, &lzolen, lzo_wrkmem);
                return lzolen;
 #else
-               return -1;
+               return 0;
 #endif
        }
 
-       return -1;
+       return 0;
 }
 
 static length_t uncompress_packet(uint8_t *dest, const uint8_t *source, length_t len, int level) {
@@ -227,30 +247,48 @@ static length_t uncompress_packet(uint8_t *dest, const uint8_t *source, length_t
        } else if(level > 9) {
 #ifdef HAVE_LZO
                lzo_uint lzolen = MAXSIZE;
-               if(lzo1x_decompress_safe(source, len, dest, &lzolen, NULL) == LZO_E_OK)
+
+               if(lzo1x_decompress_safe(source, len, dest, &lzolen, NULL) == LZO_E_OK) {
                        return lzolen;
-               else
+               else
 #endif
-                       return -1;
+                       return 0;
        }
+
 #ifdef HAVE_ZLIB
        else {
                unsigned long destlen = MAXSIZE;
-               if(uncompress(dest, &destlen, source, len) == Z_OK)
-                       return destlen;
-               else
-                       return -1;
+               static z_stream stream;
+
+               if(stream.next_in) {
+                       inflateReset(&stream);
+               } else {
+                       inflateInit(&stream);
+               }
+
+               stream.next_in = source;
+               stream.avail_in = len;
+               stream.next_out = dest;
+               stream.avail_out = destlen;
+               stream.total_out = 0;
+
+               if(inflate(&stream, Z_FINISH) == Z_STREAM_END) {
+                       return stream.total_out;
+               } else {
+                       return 0;
+               }
        }
+
 #endif
 
-       return -1;
+       return 0;
 }
 
 /* VPN packet I/O */
 
 static void receive_packet(node_t *n, vpn_packet_t *packet) {
        logger(DEBUG_TRAFFIC, LOG_DEBUG, "Received packet of %d bytes from %s (%s)",
-                          packet->len, n->name, n->hostname);
+              packet->len, n->name, n->hostname);
 
        n->in_packets++;
        n->in_bytes += packet->len;
@@ -259,27 +297,23 @@ static void receive_packet(node_t *n, vpn_packet_t *packet) {
 }
 
 static bool try_mac(node_t *n, const vpn_packet_t *inpkt) {
-       if(n->status.sptps)
+       if(n->status.sptps) {
                return sptps_verify_datagram(&n->sptps, DATA(inpkt), inpkt->len);
+       }
 
 #ifdef DISABLE_LEGACY
        return false;
 #else
-       if(!digest_active(n->indigest) || inpkt->len < sizeof(seqno_t) + digest_length(n->indigest))
+
+       if(!n->status.validkey_in || !digest_active(n->indigest) || (size_t)inpkt->len < sizeof(seqno_t) + digest_length(n->indigest)) {
                return false;
+       }
 
-       return digest_verify(n->indigest, SEQNO(inpkt), inpkt->len - digest_length(n->indigest), DATA(inpkt) + inpkt->len - digest_length(n->indigest));
+       return digest_verify(n->indigest, inpkt->data, inpkt->len - digest_length(n->indigest), inpkt->data + inpkt->len - digest_length(n->indigest));
 #endif
 }
 
 static bool receive_udppacket(node_t *n, vpn_packet_t *inpkt) {
-       vpn_packet_t pkt1, pkt2;
-       vpn_packet_t *pkt[] = { &pkt1, &pkt2, &pkt1, &pkt2 };
-       int nextpkt = 0;
-       size_t outlen;
-       pkt1.offset = DEFAULT_PACKET_OFFSET;
-       pkt2.offset = DEFAULT_PACKET_OFFSET;
-
        if(n->status.sptps) {
                if(!n->sptps.state) {
                        if(!n->status.waitingforkey) {
@@ -288,29 +322,50 @@ static bool receive_udppacket(node_t *n, vpn_packet_t *inpkt) {
                        } else {
                                logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got packet from %s (%s) but he hasn't got our key yet", n->name, n->hostname);
                        }
+
                        return false;
                }
-               inpkt->offset += 2 * sizeof(node_id_t);
-               if(!sptps_receive_data(&n->sptps, DATA(inpkt), inpkt->len - 2 * sizeof(node_id_t))) {
-                       logger(DEBUG_TRAFFIC, LOG_ERR, "Got bad packet from %s (%s)", n->name, n->hostname);
+
+               n->status.udppacket = true;
+               bool result = sptps_receive_data(&n->sptps, DATA(inpkt), inpkt->len);
+               n->status.udppacket = false;
+
+               if(!result) {
+                       /* Uh-oh. It might be that the tunnel is stuck in some corrupted state,
+                          so let's restart SPTPS in case that helps. But don't do that too often
+                          to prevent storms, and because that would make life a little too easy
+                          for external attackers trying to DoS us. */
+                       if(n->last_req_key < now.tv_sec - 10) {
+                               logger(DEBUG_PROTOCOL, LOG_ERR, "Failed to decode raw TCP packet from %s (%s), restarting SPTPS", n->name, n->hostname);
+                               send_req_key(n);
+                       }
+
                        return false;
                }
+
                return true;
        }
 
 #ifdef DISABLE_LEGACY
        return false;
 #else
-       if(!n->status.validkey) {
+       vpn_packet_t pkt1, pkt2;
+       vpn_packet_t *pkt[] = { &pkt1, &pkt2, &pkt1, &pkt2 };
+       int nextpkt = 0;
+       size_t outlen;
+       pkt1.offset = DEFAULT_PACKET_OFFSET;
+       pkt2.offset = DEFAULT_PACKET_OFFSET;
+
+       if(!n->status.validkey_in) {
                logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got packet from %s (%s) but he hasn't got our key yet", n->name, n->hostname);
                return false;
        }
 
        /* Check packet length */
 
-       if(inpkt->len < sizeof(seqno_t) + digest_length(n->indigest)) {
+       if((size_t)inpkt->len < sizeof(seqno_t) + digest_length(n->indigest)) {
                logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got too short packet from %s (%s)",
-                                       n->name, n->hostname);
+                      n->name, n->hostname);
                return false;
        }
 
@@ -322,11 +377,13 @@ static bool receive_udppacket(node_t *n, vpn_packet_t *inpkt) {
 
        if(digest_active(n->indigest)) {
                inpkt->len -= digest_length(n->indigest);
+
                if(!digest_verify(n->indigest, SEQNO(inpkt), inpkt->len, SEQNO(inpkt) + inpkt->len)) {
                        logger(DEBUG_TRAFFIC, LOG_DEBUG, "Got unauthenticated packet from %s (%s)", n->name, n->hostname);
                        return false;
                }
        }
+
        /* Decrypt the packet */
 
        if(cipher_active(n->incipher)) {
@@ -345,30 +402,32 @@ static bool receive_udppacket(node_t *n, vpn_packet_t *inpkt) {
        /* Check the sequence number */
 
        seqno_t seqno;
-       memcpy(&seqno, SEQNO(inpkt), sizeof seqno);
+       memcpy(&seqno, SEQNO(inpkt), sizeof(seqno));
        seqno = ntohl(seqno);
-       inpkt->len -= sizeof seqno;
+       inpkt->len -= sizeof(seqno);
 
        if(replaywin) {
                if(seqno != n->received_seqno + 1) {
                        if(seqno >= n->received_seqno + replaywin * 8) {
                                if(n->farfuture++ < replaywin >> 2) {
-                                       logger(DEBUG_ALWAYS, LOG_WARNING, "Packet from %s (%s) is %d seqs in the future, dropped (%u)",
-                                               n->name, n->hostname, seqno - n->received_seqno - 1, n->farfuture);
+                                       logger(DEBUG_TRAFFIC, LOG_WARNING, "Packet from %s (%s) is %d seqs in the future, dropped (%u)",
+                                              n->name, n->hostname, seqno - n->received_seqno - 1, n->farfuture);
                                        return false;
                                }
-                               logger(DEBUG_ALWAYS, LOG_WARNING, "Lost %d packets from %s (%s)",
-                                               seqno - n->received_seqno - 1, n->name, n->hostname);
+
+                               logger(DEBUG_TRAFFIC, LOG_WARNING, "Lost %d packets from %s (%s)",
+                                      seqno - n->received_seqno - 1, n->name, n->hostname);
                                memset(n->late, 0, replaywin);
-                       } else if (seqno <= n->received_seqno) {
+                       } else if(seqno <= n->received_seqno) {
                                if((n->received_seqno >= replaywin * 8 && seqno <= n->received_seqno - replaywin * 8) || !(n->late[(seqno / 8) % replaywin] & (1 << seqno % 8))) {
-                                       logger(DEBUG_ALWAYS, LOG_WARNING, "Got late or replayed packet from %s (%s), seqno %d, last received %d",
-                                               n->name, n->hostname, seqno, n->received_seqno);
+                                       logger(DEBUG_TRAFFIC, LOG_WARNING, "Got late or replayed packet from %s (%s), seqno %d, last received %d",
+                                              n->name, n->hostname, seqno, n->received_seqno);
                                        return false;
                                }
                        } else {
-                               for(int i = n->received_seqno + 1; i < seqno; i++)
+                               for(seqno_t i = n->received_seqno + 1; i < seqno; i++) {
                                        n->late[(i / 8) % replaywin] |= 1 << i % 8;
+                               }
                        }
                }
 
@@ -376,13 +435,15 @@ static bool receive_udppacket(node_t *n, vpn_packet_t *inpkt) {
                n->late[(seqno / 8) % replaywin] &= ~(1 << seqno % 8);
        }
 
-       if(seqno > n->received_seqno)
+       if(seqno > n->received_seqno) {
                n->received_seqno = seqno;
+       }
 
        n->received++;
 
-       if(n->received_seqno > MAX_SEQNO)
+       if(n->received_seqno > MAX_SEQNO) {
                regenerate_key();
+       }
 
        /* Decompress the packet */
 
@@ -391,70 +452,152 @@ static bool receive_udppacket(node_t *n, vpn_packet_t *inpkt) {
        if(n->incompression) {
                vpn_packet_t *outpkt = pkt[nextpkt++];
 
-               if((outpkt->len = uncompress_packet(DATA(outpkt), DATA(inpkt), inpkt->len, n->incompression)) < 0) {
+               if(!(outpkt->len = uncompress_packet(DATA(outpkt), DATA(inpkt), inpkt->len, n->incompression))) {
                        logger(DEBUG_TRAFFIC, LOG_ERR, "Error while uncompressing packet from %s (%s)",
-                                                n->name, n->hostname);
+                              n->name, n->hostname);
                        return false;
                }
 
                inpkt = outpkt;
 
-               origlen -= MTU/64 + 20;
+               if(origlen > MTU / 64 + 20) {
+                       origlen -= MTU / 64 + 20;
+               } else {
+                       origlen = 0;
+               }
+       }
+
+       if(inpkt->len > n->maxrecentlen) {
+               n->maxrecentlen = inpkt->len;
        }
 
        inpkt->priority = 0;
 
-       if(!DATA(inpkt)[12] && !DATA(inpkt)[13])
+       if(!DATA(inpkt)[12] && !DATA(inpkt)[13]) {
                udp_probe_h(n, inpkt, origlen);
-       else
+       } else {
                receive_packet(n, inpkt);
+       }
+
        return true;
 #endif
 }
 
-void receive_tcppacket(connection_t *c, const char *buffer, int len) {
+void receive_tcppacket(connection_t *c, const char *buffer, size_t len) {
        vpn_packet_t outpkt;
        outpkt.offset = DEFAULT_PACKET_OFFSET;
 
-       if(len > sizeof outpkt.data - outpkt.offset)
+       if(len > sizeof(outpkt.data) - outpkt.offset) {
                return;
+       }
 
        outpkt.len = len;
-       if(c->options & OPTION_TCPONLY)
+
+       if(c->options & OPTION_TCPONLY) {
                outpkt.priority = 0;
-       else
+       } else {
                outpkt.priority = -1;
+       }
+
        memcpy(DATA(&outpkt), buffer, len);
 
        receive_packet(c->node, &outpkt);
 }
 
+bool receive_tcppacket_sptps(connection_t *c, const char *data, size_t len) {
+       if(len < sizeof(node_id_t) + sizeof(node_id_t)) {
+               logger(DEBUG_PROTOCOL, LOG_ERR, "Got too short TCP SPTPS packet from %s (%s)", c->name, c->hostname);
+               return false;
+       }
+
+       node_t *to = lookup_node_id((node_id_t *)data);
+       data += sizeof(node_id_t);
+       len -= sizeof(node_id_t);
+
+       if(!to) {
+               logger(DEBUG_PROTOCOL, LOG_ERR, "Got TCP SPTPS packet from %s (%s) with unknown destination ID", c->name, c->hostname);
+               return true;
+       }
+
+       node_t *from = lookup_node_id((node_id_t *)data);
+       data += sizeof(node_id_t);
+       len -= sizeof(node_id_t);
+
+       if(!from) {
+               logger(DEBUG_PROTOCOL, LOG_ERR, "Got TCP SPTPS packet from %s (%s) with unknown source ID", c->name, c->hostname);
+               return true;
+       }
+
+       if(!to->status.reachable) {
+               /* This can happen in the form of a race condition
+                  if the node just became unreachable. */
+               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);
+               return true;
+       }
+
+       /* Help the sender reach us over UDP.
+          Note that we only do this if we're the destination or the static relay;
+          otherwise every hop would initiate its own UDP info message, resulting in elevated chatter. */
+       if(to->via == myself) {
+               send_udp_info(myself, from);
+       }
+
+       /* If we're not the final recipient, relay the packet. */
+
+       if(to != myself) {
+               send_sptps_data(to, from, 0, data, len);
+               try_tx(to, true);
+               return true;
+       }
+
+       /* The packet is for us */
+
+       if(!sptps_receive_data(&from->sptps, data, len)) {
+               /* Uh-oh. It might be that the tunnel is stuck in some corrupted state,
+                  so let's restart SPTPS in case that helps. But don't do that too often
+                  to prevent storms. */
+               if(from->last_req_key < now.tv_sec - 10) {
+                       logger(DEBUG_PROTOCOL, LOG_ERR, "Failed to decode raw TCP packet from %s (%s), restarting SPTPS", from->name, from->hostname);
+                       send_req_key(from);
+               }
+
+               return true;
+       }
+
+       send_mtu_info(myself, from, MTU);
+       return true;
+}
+
 static void send_sptps_packet(node_t *n, vpn_packet_t *origpkt) {
-       if(!n->status.validkey && !n->connection)
+       if(!n->status.validkey && !n->connection) {
                return;
+       }
 
        uint8_t type = 0;
        int offset = 0;
 
-       if(!(DATA(origpkt)[12] | DATA(origpkt)[13])) {
+       if((!(DATA(origpkt)[12] | DATA(origpkt)[13])) && (n->sptps.outstate))  {
                sptps_send_record(&n->sptps, PKT_PROBE, (char *)DATA(origpkt), origpkt->len);
                return;
        }
 
-       if(routing_mode == RMODE_ROUTER)
+       if(routing_mode == RMODE_ROUTER) {
                offset = 14;
-       else
+       } else {
                type = PKT_MAC;
+       }
 
-       if(origpkt->len < offset)
+       if(origpkt->len < offset) {
                return;
+       }
 
        vpn_packet_t outpkt;
 
        if(n->outcompression) {
                outpkt.offset = 0;
-               int len = compress_packet(DATA(&outpkt) + offset, DATA(origpkt) + offset, origpkt->len - offset, n->outcompression);
-               if(len < 0) {
+               length_t len = compress_packet(DATA(&outpkt) + offset, DATA(origpkt) + offset, origpkt->len - offset, n->outcompression);
+
+               if(!len) {
                        logger(DEBUG_TRAFFIC, LOG_ERR, "Error while compressing packet to %s (%s)", n->name, n->hostname);
                } else if(len < origpkt->len - offset) {
                        outpkt.len = len + offset;
@@ -467,10 +610,12 @@ static void send_sptps_packet(node_t *n, vpn_packet_t *origpkt) {
           don't bother with SPTPS and just use a "plaintext" PACKET message.
           We don't really care about end-to-end security since we're not
           sending the message through any intermediate nodes. */
-       if(n->connection && origpkt->len > n->minmtu)
+       if(n->connection && origpkt->len > n->minmtu) {
                send_tcppacket(n->connection, origpkt);
-       else
+       } else {
                sptps_send_record(&n->sptps, type, DATA(origpkt) + offset, origpkt->len - offset);
+       }
+
        return;
 }
 
@@ -492,14 +637,16 @@ static void choose_udp_address(const node_t *n, const sockaddr_t **sa, int *sock
        *sock = n->sock;
 
        /* If the UDP address is confirmed, use it. */
-       if(n->status.udp_confirmed)
+       if(n->status.udp_confirmed) {
                return;
+       }
 
        /* Send every third packet to n->address; that could be set
           to the node's reflexive UDP address discovered during key
           exchange. */
 
        static int x = 0;
+
        if(++x >= 3) {
                x = 0;
                return;
@@ -543,7 +690,7 @@ static void choose_local_address(const node_t *n, const sockaddr_t **sa, int *so
                }
        }
 
-       if (candidate && candidate->local_address.sa.sa_family) {
+       if(candidate && candidate->local_address.sa.sa_family) {
                *sa = &candidate->local_address;
                *sock = rand() % listen_sockets;
                adapt_socket(*sa, sock);
@@ -551,6 +698,19 @@ static void choose_local_address(const node_t *n, const sockaddr_t **sa, int *so
 }
 
 static void send_udppacket(node_t *n, vpn_packet_t *origpkt) {
+       if(!n->status.reachable) {
+               logger(DEBUG_TRAFFIC, LOG_INFO, "Trying to send UDP packet to unreachable node %s (%s)", n->name, n->hostname);
+               return;
+       }
+
+       if(n->status.sptps) {
+               send_sptps_packet(n, origpkt);
+               return;
+       }
+
+#ifdef DISABLE_LEGACY
+       return;
+#else
        vpn_packet_t pkt1, pkt2;
        vpn_packet_t *pkt[] = { &pkt1, &pkt2, &pkt1, &pkt2 };
        vpn_packet_t *inpkt = origpkt;
@@ -558,44 +718,31 @@ static void send_udppacket(node_t *n, vpn_packet_t *origpkt) {
        vpn_packet_t *outpkt;
        int origlen = origpkt->len;
        size_t outlen;
-#if defined(SOL_IP) && defined(IP_TOS)
-       static int priority = 0;
        int origpriority = origpkt->priority;
-#endif
 
        pkt1.offset = DEFAULT_PACKET_OFFSET;
        pkt2.offset = DEFAULT_PACKET_OFFSET;
 
-       if(!n->status.reachable) {
-               logger(DEBUG_TRAFFIC, LOG_INFO, "Trying to send UDP packet to unreachable node %s (%s)", n->name, n->hostname);
-               return;
-       }
-
-       if(n->status.sptps)
-               return send_sptps_packet(n, origpkt);
-
-#ifdef DISABLE_LEGACY
-       return;
-#else
        /* Make sure we have a valid key */
 
        if(!n->status.validkey) {
                logger(DEBUG_TRAFFIC, LOG_INFO,
-                                  "No valid key known yet for %s (%s), forwarding via TCP",
-                                  n->name, n->hostname);
+                      "No valid key known yet for %s (%s), forwarding via TCP",
+                      n->name, n->hostname);
                send_tcppacket(n->nexthop->connection, origpkt);
                return;
        }
 
        if(n->options & OPTION_PMTU_DISCOVERY && inpkt->len > n->minmtu && (DATA(inpkt)[12] | DATA(inpkt)[13])) {
                logger(DEBUG_TRAFFIC, LOG_INFO,
-                               "Packet for %s (%s) larger than minimum MTU, forwarding via %s",
-                               n->name, n->hostname, n != n->nexthop ? n->nexthop->name : "TCP");
+                      "Packet for %s (%s) larger than minimum MTU, forwarding via %s",
+                      n->name, n->hostname, n != n->nexthop ? n->nexthop->name : "TCP");
 
-               if(n != n->nexthop)
+               if(n != n->nexthop) {
                        send_packet(n->nexthop, origpkt);
-               else
+               } else {
                        send_tcppacket(n->nexthop->connection, origpkt);
+               }
 
                return;
        }
@@ -605,9 +752,9 @@ static void send_udppacket(node_t *n, vpn_packet_t *origpkt) {
        if(n->outcompression) {
                outpkt = pkt[nextpkt++];
 
-               if((outpkt->len = compress_packet(DATA(outpkt), DATA(inpkt), inpkt->len, n->outcompression)) < 0) {
+               if(!(outpkt->len = compress_packet(DATA(outpkt), DATA(inpkt), inpkt->len, n->outcompression))) {
                        logger(DEBUG_TRAFFIC, LOG_ERR, "Error while compressing packet to %s (%s)",
-                                  n->name, n->hostname);
+                              n->name, n->hostname);
                        return;
                }
 
@@ -617,8 +764,8 @@ static void send_udppacket(node_t *n, vpn_packet_t *origpkt) {
        /* Add sequence number */
 
        seqno_t seqno = htonl(++(n->sent_seqno));
-       memcpy(SEQNO(inpkt), &seqno, sizeof seqno);
-       inpkt->len += sizeof seqno;
+       memcpy(SEQNO(inpkt), &seqno, sizeof(seqno));
+       inpkt->len += sizeof(seqno);
 
        /* Encrypt the packet */
 
@@ -651,30 +798,60 @@ static void send_udppacket(node_t *n, vpn_packet_t *origpkt) {
        const sockaddr_t *sa = NULL;
        int sock;
 
-       if(n->status.send_locally)
+       if(n->status.send_locally) {
                choose_local_address(n, &sa, &sock);
-       if(!sa)
-               choose_udp_address(n, &sa, &sock);
+       }
 
-#if defined(SOL_IP) && defined(IP_TOS)
-       if(priorityinheritance && origpriority != priority
-          && listen_socket[n->sock].sa.sa.sa_family == AF_INET) {
-               priority = origpriority;
-               logger(DEBUG_TRAFFIC, LOG_DEBUG, "Setting outgoing packet priority to %d", priority);
-               if(setsockopt(listen_socket[n->sock].udp.fd, SOL_IP, IP_TOS, &priority, sizeof(priority))) /* SO_PRIORITY doesn't seem to work */
-                       logger(DEBUG_ALWAYS, LOG_ERR, "System call `%s' failed: %s", "setsockopt", sockstrerror(sockerrno));
+       if(!sa) {
+               choose_udp_address(n, &sa, &sock);
        }
+
+       if(priorityinheritance && origpriority != listen_socket[sock].priority) {
+               listen_socket[sock].priority = origpriority;
+
+               switch(sa->sa.sa_family) {
+#if defined(IP_TOS)
+
+               case AF_INET:
+                       logger(DEBUG_TRAFFIC, LOG_DEBUG, "Setting IPv4 outgoing packet priority to %d", origpriority);
+
+                       if(setsockopt(listen_socket[sock].udp.fd, IPPROTO_IP, IP_TOS, (void *)&origpriority, sizeof(origpriority))) { /* SO_PRIORITY doesn't seem to work */
+                               logger(DEBUG_ALWAYS, LOG_ERR, "System call `%s' failed: %s", "setsockopt", sockstrerror(sockerrno));
+                       }
+
+                       break;
+#endif
+#if defined(IPV6_TCLASS)
+
+               case AF_INET6:
+                       logger(DEBUG_TRAFFIC, LOG_DEBUG, "Setting IPv6 outgoing packet priority to %d", origpriority);
+
+                       if(setsockopt(listen_socket[sock].udp.fd, IPPROTO_IPV6, IPV6_TCLASS, (void *)&origpriority, sizeof(origpriority))) { /* SO_PRIORITY doesn't seem to work */
+                               logger(DEBUG_ALWAYS, LOG_ERR, "System call `%s' failed: %s", "setsockopt", sockstrerror(sockerrno));
+                       }
+
+                       break;
 #endif
 
-       if(sendto(listen_socket[sock].udp.fd, SEQNO(inpkt), inpkt->len, 0, &sa->sa, SALEN(sa->sa)) < 0 && !sockwouldblock(sockerrno)) {
+               default:
+                       break;
+               }
+       }
+
+       if(sendto(listen_socket[sock].udp.fd, (void *)SEQNO(inpkt), inpkt->len, 0, &sa->sa, SALEN(sa->sa)) < 0 && !sockwouldblock(sockerrno)) {
                if(sockmsgsize(sockerrno)) {
-                       if(n->maxmtu >= origlen)
+                       if(n->maxmtu >= origlen) {
                                n->maxmtu = origlen - 1;
-                       if(n->mtu >= origlen)
+                       }
+
+                       if(n->mtu >= origlen) {
                                n->mtu = origlen - 1;
+                       }
+
                        try_fix_mtu(n);
-               } else
+               } else {
                        logger(DEBUG_TRAFFIC, LOG_WARNING, "Error sending packet to %s (%s): %s", n->name, n->hostname, sockstrerror(sockerrno));
+               }
        }
 
 end:
@@ -682,61 +859,96 @@ end:
 #endif
 }
 
-static bool send_sptps_data_priv(node_t *to, node_t *from, int type, const void *data, size_t len) {
+bool send_sptps_data(node_t *to, node_t *from, int type, const void *data, size_t len) {
        node_t *relay = (to->via != myself && (type == PKT_PROBE || (len - SPTPS_DATAGRAM_OVERHEAD) <= to->via->minmtu)) ? to->via : to->nexthop;
        bool direct = from == myself && to == relay;
        bool relay_supported = (relay->options >> 24) >= 4;
        bool tcponly = (myself->options | relay->options) & OPTION_TCPONLY;
 
-       /* 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.
-          TODO: When relaying, the original sender does not know the end-to-end PMTU (it only knows the PMTU of the first hop).
-                This can lead to scenarios where large packets are sent over UDP to relay, but then relay has no choice but fall back to TCP. */
+       /* 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. */
 
        if(type == SPTPS_HANDSHAKE || tcponly || (!direct && !relay_supported) || (type != PKT_PROBE && (len - SPTPS_DATAGRAM_OVERHEAD) > relay->minmtu)) {
+               if(type != SPTPS_HANDSHAKE && (to->nexthop->connection->options >> 24) >= 7) {
+                       char buf[len + sizeof(to->id) + sizeof(from->id)];
+                       char *buf_ptr = buf;
+                       memcpy(buf_ptr, &to->id, sizeof(to->id));
+                       buf_ptr += sizeof(to->id);
+                       memcpy(buf_ptr, &from->id, sizeof(from->id));
+                       buf_ptr += sizeof(from->id);
+                       memcpy(buf_ptr, data, len);
+                       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);
+                       return send_sptps_tcppacket(to->nexthop->connection, buf, sizeof(buf));
+               }
+
                char buf[len * 4 / 3 + 5];
                b64encode(data, buf, len);
-               /* If no valid key is known yet, send the packets using ANS_KEY requests,
-                  to ensure we get to learn the reflexive UDP address. */
-               if(from == myself && !to->status.validkey) {
+
+               /* If this is a handshake packet, use ANS_KEY instead of REQ_KEY, for two reasons:
+                   - We don't want intermediate nodes to switch to UDP to relay these packets;
+                   - ANS_KEY allows us to learn the reflexive UDP address. */
+               if(type == SPTPS_HANDSHAKE) {
                        to->incompression = myself->incompression;
                        return send_request(to->nexthop->connection, "%d %s %s %s -1 -1 -1 %d", ANS_KEY, from->name, to->name, buf, to->incompression);
                } else {
-                       return send_request(to->nexthop->connection, "%d %s %s %d %s", REQ_KEY, from->name, to->name, REQ_SPTPS, buf);
+                       return send_request(to->nexthop->connection, "%d %s %s %d %s", REQ_KEY, from->name, to->name, SPTPS_PACKET, buf);
                }
        }
 
        size_t overhead = 0;
-       if(relay_supported) overhead += sizeof to->id + sizeof from->id;
-       char buf[len + overhead]; char* buf_ptr = buf;
+
+       if(relay_supported) {
+               overhead += sizeof(to->id) + sizeof(from->id);
+       }
+
+       char buf[len + overhead];
+       char *buf_ptr = buf;
+
        if(relay_supported) {
                if(direct) {
                        /* Inform the recipient that this packet was sent directly. */
-                       node_id_t nullid = {};
-                       memcpy(buf_ptr, &nullid, sizeof nullid); buf_ptr += sizeof nullid;
+                       node_id_t nullid = {0};
+                       memcpy(buf_ptr, &nullid, sizeof(nullid));
+                       buf_ptr += sizeof(nullid);
                } else {
-                       memcpy(buf_ptr, &to->id, sizeof to->id); buf_ptr += sizeof to->id;
+                       memcpy(buf_ptr, &to->id, sizeof(to->id));
+                       buf_ptr += sizeof(to->id);
                }
-               memcpy(buf_ptr, &from->id, sizeof from->id); buf_ptr += sizeof from->id;
+
+               memcpy(buf_ptr, &from->id, sizeof(from->id));
+               buf_ptr += sizeof(from->id);
 
        }
+
        /* 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 */
-       memcpy(buf_ptr, data, len); buf_ptr += len;
+       memcpy(buf_ptr, data, len);
+       buf_ptr += len;
 
        const sockaddr_t *sa = NULL;
        int sock;
-       if(relay->status.send_locally)
+
+       if(relay->status.send_locally) {
                choose_local_address(relay, &sa, &sock);
-       if(!sa)
+       }
+
+       if(!sa) {
                choose_udp_address(relay, &sa, &sock);
-       logger(DEBUG_TRAFFIC, LOG_INFO, "Sending packet from %s (%s) to %s (%s) via %s (%s)", from->name, from->hostname, to->name, to->hostname, relay->name, relay->hostname);
+       }
+
+       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);
+
        if(sendto(listen_socket[sock].udp.fd, buf, buf_ptr - buf, 0, &sa->sa, SALEN(sa->sa)) < 0 && !sockwouldblock(sockerrno)) {
                if(sockmsgsize(sockerrno)) {
                        // Compensate for SPTPS overhead
                        len -= SPTPS_DATAGRAM_OVERHEAD;
-                       if(relay->maxmtu >= len)
+
+                       if(relay->maxmtu >= len) {
                                relay->maxmtu = len - 1;
-                       if(relay->mtu >= len)
+                       }
+
+                       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));
@@ -747,10 +959,6 @@ static bool send_sptps_data_priv(node_t *to, node_t *from, int type, const void
        return true;
 }
 
-bool send_sptps_data(void *handle, uint8_t type, const void *data, size_t len) {
-       return send_sptps_data_priv(handle, myself, type, data, len);
-}
-
 bool receive_sptps_record(void *handle, uint8_t type, const void *data, uint16_t len) {
        node_t *from = handle;
 
@@ -758,8 +966,9 @@ bool receive_sptps_record(void *handle, uint8_t type, const void *data, uint16_t
                if(!from->status.validkey) {
                        from->status.validkey = true;
                        from->status.waitingforkey = false;
-                       logger(DEBUG_META, LOG_INFO, "SPTPS key exchange with %s (%s) succesful", from->name, from->hostname);
+                       logger(DEBUG_META, LOG_INFO, "SPTPS key exchange with %s (%s) successful", from->name, from->hostname);
                }
+
                return true;
        }
 
@@ -770,10 +979,21 @@ bool receive_sptps_record(void *handle, uint8_t type, const void *data, uint16_t
 
        vpn_packet_t inpkt;
        inpkt.offset = DEFAULT_PACKET_OFFSET;
+       inpkt.priority = 0;
 
        if(type == PKT_PROBE) {
+               if(!from->status.udppacket) {
+                       logger(DEBUG_ALWAYS, LOG_ERR, "Got SPTPS PROBE packet from %s (%s) via TCP", from->name, from->hostname);
+                       return false;
+               }
+
                inpkt.len = len;
                memcpy(DATA(&inpkt), data, len);
+
+               if(inpkt.len > from->maxrecentlen) {
+                       from->maxrecentlen = inpkt.len;
+               }
+
                udp_probe_h(from, &inpkt, len);
                return true;
        }
@@ -792,15 +1012,19 @@ bool receive_sptps_record(void *handle, uint8_t type, const void *data, uint16_t
        }
 
        int offset = (type & PKT_MAC) ? 0 : 14;
+
        if(type & PKT_COMPRESSED) {
                length_t ulen = uncompress_packet(DATA(&inpkt) + offset, (const uint8_t *)data, len, from->incompression);
-               if(ulen < 0) {
+
+               if(!ulen) {
                        return false;
                } else {
                        inpkt.len = ulen + offset;
                }
-               if(inpkt.len > MAXSIZE)
+
+               if(inpkt.len > MAXSIZE) {
                        abort();
+               }
        } else {
                memcpy(DATA(&inpkt) + offset, data, len);
                inpkt.len = len + offset;
@@ -809,22 +1033,28 @@ bool receive_sptps_record(void *handle, uint8_t type, const void *data, uint16_t
        /* Generate the Ethernet packet type if necessary */
        if(offset) {
                switch(DATA(&inpkt)[14] >> 4) {
-                       case 4:
-                               DATA(&inpkt)[12] = 0x08;
-                               DATA(&inpkt)[13] = 0x00;
-                               break;
-                       case 6:
-                               DATA(&inpkt)[12] = 0x86;
-                               DATA(&inpkt)[13] = 0xDD;
-                               break;
-                       default:
-                               logger(DEBUG_TRAFFIC, LOG_ERR,
-                                                  "Unknown IP version %d while reading packet from %s (%s)",
-                                                  DATA(&inpkt)[14] >> 4, from->name, from->hostname);
-                               return false;
+               case 4:
+                       DATA(&inpkt)[12] = 0x08;
+                       DATA(&inpkt)[13] = 0x00;
+                       break;
+
+               case 6:
+                       DATA(&inpkt)[12] = 0x86;
+                       DATA(&inpkt)[13] = 0xDD;
+                       break;
+
+               default:
+                       logger(DEBUG_TRAFFIC, LOG_ERR,
+                              "Unknown IP version %d while reading packet from %s (%s)",
+                              DATA(&inpkt)[14] >> 4, from->name, from->hostname);
+                       return false;
                }
        }
 
+       if(from->status.udppacket && inpkt.len > from->maxrecentlen) {
+               from->maxrecentlen = inpkt.len;
+       }
+
        receive_packet(from, &inpkt);
        return true;
 }
@@ -832,14 +1062,15 @@ bool receive_sptps_record(void *handle, uint8_t type, const void *data, uint16_t
 // 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)
+       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)
+       if(!n->status.waitingforkey) {
                send_req_key(n);
-       else if(n->last_req_key + 10 < now.tv_sec) {
+       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;
@@ -865,22 +1096,47 @@ static void send_udp_probe_packet(node_t *n, int len) {
 // 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_tn) {
-       if(!udp_discovery)
+static void try_udp(node_t *n) {
+       if(!udp_discovery) {
                return;
+       }
+
+       /* Send gratuitous probe replies to 1.1 nodes. */
+
+       if((n->options >> 24) >= 3 && n->status.udp_confirmed) {
+               struct timeval ping_tx_elapsed;
+               timersub(&now, &n->udp_reply_sent, &ping_tx_elapsed);
+
+               if(ping_tx_elapsed.tv_sec >= udp_discovery_keepalive_interval - 1) {
+                       n->udp_reply_sent = now;
+
+                       if(n->maxrecentlen) {
+                               vpn_packet_t pkt;
+                               pkt.len = n->maxrecentlen;
+                               pkt.offset = DEFAULT_PACKET_OFFSET;
+                               memset(DATA(&pkt), 0, 14);
+                               randomize(DATA(&pkt) + 14, MIN_PROBE_SIZE - 14);
+                               send_udp_probe_reply(n, &pkt, pkt.len);
+                               n->maxrecentlen = 0;
+                       }
+               }
+       }
+
+       /* Probe request */
 
-       struct timeval now;
-       gettimeofday(&now, NULL);
        struct timeval ping_tx_elapsed;
        timersub(&now, &n->udp_ping_sent, &ping_tx_elapsed);
 
-       if(ping_tx_elapsed.tv_sec >= udp_discovery_interval) {
-               send_udp_probe_packet(n, MAX(n->minmtu, 16));
-               n->udp_ping_sent = now;
+       int interval = n->status.udp_confirmed ? udp_discovery_keepalive_interval : udp_discovery_interval;
+
+       if(ping_tx_elapsed.tv_sec >= interval) {
+               gettimeofday(&now, NULL);
+               n->udp_ping_sent = now; // a probe in flight
+               send_udp_probe_packet(n, MIN_PROBE_SIZE);
 
                if(localdiscovery && !n->status.udp_confirmed && n->prevedge) {
                        n->status.send_locally = true;
-                       send_udp_probe_packet(n, 16);
+                       send_udp_probe_packet(n, MIN_PROBE_SIZE);
                        n->status.send_locally = false;
                }
        }
@@ -894,10 +1150,13 @@ static length_t choose_initial_maxmtu(node_t *n) {
        const sockaddr_t *sa = NULL;
        int sockindex;
        choose_udp_address(n, &sa, &sockindex);
-       if(!sa)
+
+       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;
@@ -910,7 +1169,8 @@ static length_t choose_initial_maxmtu(node_t *n) {
        }
 
        int ip_mtu;
-       socklen_t ip_mtu_len = sizeof 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);
@@ -924,37 +1184,68 @@ static length_t choose_initial_maxmtu(node_t *n) {
        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)
+
+               if((n->options >> 24) >= 4) {
                        mtu -= sizeof(node_id_t) + sizeof(node_id_t);
+               }
+
+#ifndef DISABLE_LEGACY
+       } else {
+               mtu -= digest_length(n->outdigest);
+
+               /* Now it's tricky. We use CBC mode, so the length of the
+                  encrypted payload must be a multiple of the blocksize. The
+                  sequence number is also part of the encrypted payload, so we
+                  must account for it after correcting for the blocksize.
+                  Furthermore, the padding in the last block must be at least
+                  1 byte. */
+
+               length_t blocksize = cipher_blocksize(n->outcipher);
+
+               if(blocksize > 1) {
+                       mtu /= blocksize;
+                       mtu *= blocksize;
+                       mtu--;
+               }
+
+               mtu -= 4; // seqno
+#endif
        }
 
-       if (mtu < 512) {
+       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)
+
+       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
-
+       (void)n;
        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.
+/* 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))
+       if(!(n->options & OPTION_PMTU_DISCOVERY)) {
                return;
+       }
 
        if(udp_discovery && !n->status.udp_confirmed) {
+               n->maxrecentlen = 0;
                n->mtuprobes = 0;
                n->minmtu = 0;
                n->maxmtu = MTU;
@@ -963,310 +1254,450 @@ static void try_mtu(node_t *n) {
 
        /* 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 */
+          mtuprobes ==    -1: send one maxmtu and one maxmtu+1 probe every pinginterval
+          mtuprobes ==-2..-3: send one maxmtu probe every second
+          mtuprobes ==    -4: maxmtu no longer valid, reset minmtu and maxmtu and go to 0 */
 
-       struct timeval now;
-       gettimeofday(&now, NULL);
        struct timeval elapsed;
-       timersub(&now, &n->probe_sent_time, &elapsed);
+       timersub(&now, &n->mtu_ping_sent, &elapsed);
+
        if(n->mtuprobes >= 0) {
-               if(n->mtuprobes != 0 && elapsed.tv_sec == 0 && elapsed.tv_usec < 333333)
+               if(n->mtuprobes != 0 && elapsed.tv_sec == 0 && elapsed.tv_usec < 333333) {
                        return;
+               }
        } else {
-               if(elapsed.tv_sec < pingtimeout)
-                       return;
+               if(n->mtuprobes < -1) {
+                       if(elapsed.tv_sec < 1) {
+                               return;
+                       }
+               } else {
+                       if(elapsed.tv_sec < pinginterval) {
+                               return;
+                       }
+               }
        }
 
+       n->mtu_ping_sent = now;
+
        try_fix_mtu(n);
 
-       int timeout;
+       if(n->mtuprobes < -3) {
+               /* We lost three MTU probes, restart discovery */
+               logger(DEBUG_TRAFFIC, LOG_INFO, "Decrease in PMTU to %s (%s) detected, restarting PMTU discovery", n->name, n->hostname);
+               n->mtuprobes = 0;
+               n->minmtu = 0;
+       }
+
        if(n->mtuprobes < 0) {
-               /* After the initial discovery, we only send one >maxmtu probe
-                  to detect PMTU increases. */
-               if(n->maxmtu + 1 < MTU)
+               /* After the initial discovery, we only send one maxmtu and one
+                  maxmtu+1 probe to detect PMTU increases. */
+               send_udp_probe_packet(n, n->maxmtu);
+
+               if(n->mtuprobes == -1 && n->maxmtu + 1 < MTU) {
                        send_udp_probe_packet(n, n->maxmtu + 1);
+               }
+
+               n->mtuprobes--;
        } 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)
+               if(n->mtuprobes == 0) {
                        n->maxmtu = choose_initial_maxmtu(n);
+               }
 
-               /* 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));
-
-               send_udp_probe_packet(n, minmtu + offset);
-               if(n->mtuprobes >= 0)
+               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;
+/* These functions try to establish a tunnel to a node (or its relay) so that
+   packets can be sent (e.g. exchange 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.
+*/
 
-       /* 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.
-        */
+static void try_tx_sptps(node_t *n, bool mtu) {
+       /* If n is a TCP-only neighbor, we'll only use "cleartext" PACKET
+          messages anyway, so there's no need for SPTPS at all. */
 
-       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;
+       if(n->connection && ((myself->options | n->options) & OPTION_TCPONLY)) {
+               return;
+       }
 
-       n->prev_received_seqno = n->received_seqno;
-       n->prev_received = n->received;
-}
+       /* Otherwise, try to do SPTPS authentication with n if necessary. */
 
-// 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)
+       try_sptps(n);
+
+       /* Do we need to statically relay packets? */
+
+       node_t *via = (n->via == myself) ? n->nexthop : n->via;
+
+       /* If we do have a static relay, try everything with that one instead, if it supports relaying. */
+
+       if(via != n) {
+               if((via->options >> 24) < 4) {
                        return;
+               }
+
+               try_tx(via, mtu);
+               return;
        }
 
-       node_t *via = (n->via == myself) ? n->nexthop : n->via;
-       
-       if((myself->options | via->options) & OPTION_TCPONLY)
+       /* Otherwise, try to establish UDP connectivity. */
+
+       try_udp(n);
+
+       if(mtu) {
+               try_mtu(n);
+       }
+
+       /* If we don't have UDP connectivity (yet), we need to use a dynamic relay (nexthop)
+          while we try to establish direct connectivity. */
+
+       if(!n->status.udp_confirmed && n != n->nexthop && (n->nexthop->options >> 24) >= 4) {
+               try_tx(n->nexthop, mtu);
+       }
+}
+
+static void try_tx_legacy(node_t *n, bool mtu) {
+       /* Does he have our key? If not, send one. */
+
+       if(!n->status.validkey_in) {
+               send_ans_key(n);
+       }
+
+       /* Check if we already have a key, or request one. */
+
+       if(!n->status.validkey) {
+               if(n->last_req_key + 10 <= now.tv_sec) {
+                       send_req_key(n);
+                       n->last_req_key = now.tv_sec;
+               }
+
                return;
+       }
+
+       try_udp(n);
+
+       if(mtu) {
+               try_mtu(n);
+       }
+}
 
-       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);
+void try_tx(node_t *n, bool mtu) {
+       if(!n->status.reachable) {
+               return;
        }
 
-       /* 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);
+       if(n->status.sptps) {
+               try_tx_sptps(n, mtu);
+       } else {
+               try_tx_legacy(n, mtu);
+       }
 }
 
-/*
-  send a packet to the given vpn ip.
-*/
 void send_packet(node_t *n, vpn_packet_t *packet) {
-       node_t *via;
+       // If it's for myself, write it to the tun/tap device.
 
        if(n == myself) {
-               if(overwrite_mac)
-                        memcpy(DATA(packet), mymac.x, ETH_ALEN);
+               if(overwrite_mac) {
+                       memcpy(DATA(packet), mymac.x, ETH_ALEN);
+                       // Use an arbitrary fake source address.
+                       memcpy(DATA(packet) + ETH_ALEN, DATA(packet), ETH_ALEN);
+                       DATA(packet)[ETH_ALEN * 2 - 1] ^= 0xFF;
+               }
+
                n->out_packets++;
                n->out_bytes += packet->len;
                devops.write(packet);
                return;
        }
 
-       logger(DEBUG_TRAFFIC, LOG_ERR, "Sending packet of %d bytes to %s (%s)",
-                          packet->len, n->name, n->hostname);
+       logger(DEBUG_TRAFFIC, LOG_ERR, "Sending packet of %d bytes to %s (%s)", packet->len, n->name, n->hostname);
+
+       // If the node is not reachable, drop it.
 
        if(!n->status.reachable) {
-               logger(DEBUG_TRAFFIC, LOG_INFO, "Node %s (%s) is not reachable",
-                                  n->name, n->hostname);
+               logger(DEBUG_TRAFFIC, LOG_INFO, "Node %s (%s) is not reachable", n->name, n->hostname);
                return;
        }
 
+       // Keep track of packet statistics.
+
        n->out_packets++;
        n->out_bytes += packet->len;
 
+       // Check if it should be sent as an SPTPS packet.
+
        if(n->status.sptps) {
                send_sptps_packet(n, packet);
-               goto end;
+               try_tx(n, true);
+               return;
        }
 
-       via = (packet->priority == -1 || n->via == myself) ? n->nexthop : n->via;
+       // Determine which node to actually send it to.
 
-       if(via != n)
-               logger(DEBUG_TRAFFIC, LOG_INFO, "Sending packet to %s via %s (%s)",
-                          n->name, via->name, n->via->hostname);
+       node_t *via = (packet->priority == -1 || n->via == myself) ? n->nexthop : n->via;
+
+       if(via != n) {
+               logger(DEBUG_TRAFFIC, LOG_INFO, "Sending packet to %s via %s (%s)", n->name, via->name, n->via->hostname);
+       }
+
+       // Try to send via UDP, unless TCP is forced.
 
        if(packet->priority == -1 || ((myself->options | via->options) & OPTION_TCPONLY)) {
-               if(!send_tcppacket(via->connection, packet))
+               if(!send_tcppacket(via->connection, packet)) {
                        terminate_connection(via->connection, true);
-       } else
-               send_udppacket(via, packet);
+               }
 
-end:
-       /* Try to improve the tunnel.
-          Note that we do this *after* we send the packet because sending actual packets take priority
-          with regard to the send buffer space and latency. */
-       try_tx(n);
-}
+               return;
+       }
 
-/* Broadcast a packet using the minimum spanning tree */
+       send_udppacket(via, packet);
+       try_tx(via, true);
+}
 
 void broadcast_packet(const node_t *from, vpn_packet_t *packet) {
        // Always give ourself a copy of the packet.
-       if(from != myself)
+       if(from != myself) {
                send_packet(myself, packet);
+       }
 
        // In TunnelServer mode, do not forward broadcast packets.
        // The MST might not be valid and create loops.
-       if(tunnelserver || broadcast_mode == BMODE_NONE)
+       if(tunnelserver || broadcast_mode == BMODE_NONE) {
                return;
+       }
 
        logger(DEBUG_TRAFFIC, LOG_INFO, "Broadcasting packet of %d bytes from %s (%s)",
-                          packet->len, from->name, from->hostname);
+              packet->len, from->name, from->hostname);
 
        switch(broadcast_mode) {
-               // In MST mode, broadcast packets travel via the Minimum Spanning Tree.
-               // This guarantees all nodes receive the broadcast packet, and
-               // usually distributes the sending of broadcast packets over all nodes.
-               case BMODE_MST:
-                       for list_each(connection_t, c, connection_list)
-                               if(c->edge && c->status.mst && c != from->nexthop->connection)
-                                       send_packet(c->node, packet);
-                       break;
+       // In MST mode, broadcast packets travel via the Minimum Spanning Tree.
+       // This guarantees all nodes receive the broadcast packet, and
+       // usually distributes the sending of broadcast packets over all nodes.
+       case BMODE_MST:
+               for list_each(connection_t, c, connection_list)
+                       if(c->edge && c->status.mst && c != from->nexthop->connection) {
+                               send_packet(c->node, packet);
+                       }
 
-               // In direct mode, we send copies to each node we know of.
-               // However, this only reaches nodes that can be reached in a single hop.
-               // We don't have enough information to forward broadcast packets in this case.
-               case BMODE_DIRECT:
-                       if(from != myself)
-                               break;
+               break;
 
-                       for splay_each(node_t, n, node_tree)
-                               if(n->status.reachable && n != myself && ((n->via == myself && n->nexthop == n) || n->via == n))
-                                       send_packet(n, packet);
+       // In direct mode, we send copies to each node we know of.
+       // However, this only reaches nodes that can be reached in a single hop.
+       // We don't have enough information to forward broadcast packets in this case.
+       case BMODE_DIRECT:
+               if(from != myself) {
                        break;
+               }
 
-               default:
-                       break;
+               for splay_each(node_t, n, node_tree)
+                       if(n->status.reachable && n != myself && ((n->via == myself && n->nexthop == n) || n->via == n)) {
+                               send_packet(n, packet);
+                       }
+
+               break;
+
+       default:
+               break;
        }
 }
 
+/* We got a packet from some IP address, but we don't know who sent it.  Try to
+   verify the message authentication code against all active session keys.
+   Since this is actually an expensive operation, we only do a full check once
+   a minute, the rest of the time we only check against nodes for which we know
+   an IP address that matches the one from the packet.  */
+
 static node_t *try_harder(const sockaddr_t *from, const vpn_packet_t *pkt) {
-       node_t *n = NULL;
+       node_t *match = NULL;
        bool hard = false;
        static time_t last_hard_try = 0;
 
-       for splay_each(edge_t, e, edge_weight_tree) {
-               if(!e->to->status.reachable || e->to == myself)
+       for splay_each(node_t, n, node_tree) {
+               if(!n->status.reachable || n == myself) {
                        continue;
+               }
+
+               if(!n->status.validkey_in && !(n->status.sptps && n->sptps.instate)) {
+                       continue;
+               }
+
+               bool soft = false;
 
-               if(sockaddrcmp_noport(from, &e->address)) {
-                       if(last_hard_try == now.tv_sec)
+               for splay_each(edge_t, e, n->edge_tree) {
+                       if(!e->reverse) {
                                continue;
+                       }
+
+                       if(!sockaddrcmp_noport(from, &e->reverse->address)) {
+                               soft = true;
+                               break;
+                       }
+               }
+
+               if(!soft) {
+                       if(last_hard_try == now.tv_sec) {
+                               continue;
+                       }
+
                        hard = true;
                }
 
-               if(!try_mac(e->to, pkt))
+               if(!try_mac(n, pkt)) {
                        continue;
+               }
 
-               n = e->to;
+               match = n;
                break;
        }
 
-       if(hard)
+       if(hard) {
                last_hard_try = now.tv_sec;
+       }
 
-       last_hard_try = now.tv_sec;
-       return n;
+       return match;
 }
 
-void handle_incoming_vpn_data(void *data, int flags) {
-       listen_socket_t *ls = data;
-       vpn_packet_t pkt;
+static void handle_incoming_vpn_packet(listen_socket_t *ls, vpn_packet_t *pkt, sockaddr_t *addr) {
        char *hostname;
-       node_id_t nullid = {};
-       sockaddr_t addr = {};
-       socklen_t addrlen = sizeof addr;
+       node_id_t nullid = {0};
        node_t *from, *to;
        bool direct = false;
 
-       pkt.offset = 0;
-       int len = recvfrom(ls->udp.fd, DATA(&pkt), MAXSIZE, 0, &addr.sa, &addrlen);
-
-       if(len <= 0 || len > MAXSIZE) {
-               if(!sockwouldblock(sockerrno))
-                       logger(DEBUG_ALWAYS, LOG_ERR, "Receiving packet failed: %s", sockstrerror(sockerrno));
-               return;
-       }
-
-       pkt.len = len;
-
-       sockaddrunmap(&addr); /* Some braindead IPv6 implementations do stupid things. */
+       sockaddrunmap(addr); /* Some braindead IPv6 implementations do stupid things. */
 
        // Try to figure out who sent this packet.
 
-       node_t *n = lookup_node_udp(&addr);
+       node_t *n = lookup_node_udp(addr);
+
+       if(n && !n->status.udp_confirmed) {
+               n = NULL;        // Don't believe it if we don't have confirmation yet.
+       }
 
        if(!n) {
                // It might be from a 1.1 node, which might have a source ID in the packet.
-               pkt.offset = 2 * sizeof(node_id_t);
-               from = lookup_node_id(SRCID(&pkt));
-               if(from && !memcmp(DSTID(&pkt), &nullid, sizeof nullid) && from->status.sptps) {
-                       if(sptps_verify_datagram(&from->sptps, DATA(&pkt), pkt.len - 2 * sizeof(node_id_t)))
+               pkt->offset = 2 * sizeof(node_id_t);
+               from = lookup_node_id(SRCID(pkt));
+
+               if(from && !memcmp(DSTID(pkt), &nullid, sizeof(nullid)) && from->status.sptps) {
+                       if(sptps_verify_datagram(&from->sptps, DATA(pkt), pkt->len - 2 * sizeof(node_id_t))) {
                                n = from;
-                       else
+                       } else {
                                goto skip_harder;
+                       }
                }
        }
 
        if(!n) {
-               pkt.offset = 0;
-               n = try_harder(&addr, &pkt);
+               pkt->offset = 0;
+               n = try_harder(addr, pkt);
        }
 
 skip_harder:
+
        if(!n) {
                if(debug_level >= DEBUG_PROTOCOL) {
-                       hostname = sockaddr2hostname(&addr);
+                       hostname = sockaddr2hostname(addr);
                        logger(DEBUG_PROTOCOL, LOG_WARNING, "Received UDP packet from unknown source %s", hostname);
                        free(hostname);
                }
+
                return;
        }
 
+       pkt->offset = 0;
+
        if(n->status.sptps) {
-               pkt.offset = 2 * sizeof(node_id_t);
+               bool relay_enabled = (n->options >> 24) >= 4;
+
+               if(relay_enabled) {
+                       pkt->offset = 2 * sizeof(node_id_t);
+                       pkt->len -= pkt->offset;
+               }
 
-               if(!memcmp(DSTID(&pkt), &nullid, sizeof nullid)) {
+               if(!memcmp(DSTID(pkt), &nullid, sizeof(nullid)) || !relay_enabled) {
                        direct = true;
                        from = n;
                        to = myself;
                } else {
-                       from = lookup_node_id(SRCID(&pkt));
-                       to = lookup_node_id(DSTID(&pkt));
+                       from = lookup_node_id(SRCID(pkt));
+                       to = lookup_node_id(DSTID(pkt));
                }
+
                if(!from || !to) {
                        logger(DEBUG_PROTOCOL, LOG_WARNING, "Received UDP packet from %s (%s) with unknown source and/or destination ID", n->name, n->hostname);
                        return;
                }
 
+               if(!to->status.reachable) {
+                       /* This can happen in the form of a race condition
+                          if the node just became unreachable. */
+                       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);
+                       return;
+               }
+
+               /* The packet is supposed to come from the originator or its static relay
+                  (i.e. with no dynamic relays in between).
+                  If it did not, "help" the static relay by sending it UDP info.
+                  Note that we only do this if we're the destination or the static relay;
+                  otherwise every hop would initiate its own UDP info message, resulting in elevated chatter. */
+
+               if(n != from->via && to->via == myself) {
+                       send_udp_info(myself, from);
+               }
+
+               /* If we're not the final recipient, relay the packet. */
+
                if(to != myself) {
-                       send_sptps_data_priv(to, n, 0, DATA(&pkt), pkt.len - 2 * sizeof(node_id_t));
+                       send_sptps_data(to, from, 0, DATA(pkt), pkt->len);
+                       try_tx(to, true);
                        return;
                }
        } else {
@@ -1274,23 +1705,115 @@ skip_harder:
                from = n;
        }
 
-       pkt.offset = 0;
-       if(!receive_udppacket(from, &pkt))
+       if(!receive_udppacket(from, pkt)) {
                return;
+       }
 
        n->sock = ls - listen_socket;
-       if(direct && sockaddrcmp(&addr, &n->address))
-               update_node_udp(n, &addr);
+
+       if(direct && sockaddrcmp(addr, &n->address)) {
+               update_node_udp(n, addr);
+       }
+
+       /* If the packet went through a relay, help the sender find the appropriate MTU
+          through the relay path. */
+
+       if(!direct) {
+               send_mtu_info(myself, n, MTU);
+       }
+}
+
+void handle_incoming_vpn_data(void *data, int flags) {
+       (void)data;
+       (void)flags;
+       listen_socket_t *ls = data;
+
+#ifdef HAVE_RECVMMSG
+#define MAX_MSG 64
+       static int num = MAX_MSG;
+       static vpn_packet_t pkt[MAX_MSG];
+       static sockaddr_t addr[MAX_MSG];
+       static struct mmsghdr msg[MAX_MSG];
+       static struct iovec iov[MAX_MSG];
+
+       for(int i = 0; i < num; i++) {
+               pkt[i].offset = 0;
+
+               iov[i] = (struct iovec) {
+                       .iov_base = DATA(&pkt[i]),
+                       .iov_len = MAXSIZE,
+               };
+
+               msg[i].msg_hdr = (struct msghdr) {
+                       .msg_name = &addr[i].sa,
+                       .msg_namelen = sizeof(addr)[i],
+                       .msg_iov = &iov[i],
+                       .msg_iovlen = 1,
+               };
+       }
+
+       num = recvmmsg(ls->udp.fd, msg, MAX_MSG, MSG_DONTWAIT, NULL);
+
+       if(num < 0) {
+               if(!sockwouldblock(sockerrno)) {
+                       logger(DEBUG_ALWAYS, LOG_ERR, "Receiving packet failed: %s", sockstrerror(sockerrno));
+               }
+
+               return;
+       }
+
+       for(int i = 0; i < num; i++) {
+               pkt[i].len = msg[i].msg_len;
+
+               if(pkt[i].len <= 0 || pkt[i].len > MAXSIZE) {
+                       continue;
+               }
+
+               handle_incoming_vpn_packet(ls, &pkt[i], &addr[i]);
+       }
+
+#else
+       vpn_packet_t pkt;
+       sockaddr_t addr = {0};
+       socklen_t addrlen = sizeof(addr);
+
+       pkt.offset = 0;
+       int len = recvfrom(ls->udp.fd, (void *)DATA(&pkt), MAXSIZE, 0, &addr.sa, &addrlen);
+
+       if(len <= 0 || (size_t)len > MAXSIZE) {
+               if(!sockwouldblock(sockerrno)) {
+                       logger(DEBUG_ALWAYS, LOG_ERR, "Receiving packet failed: %s", sockstrerror(sockerrno));
+               }
+
+               return;
+       }
+
+       pkt.len = len;
+
+       handle_incoming_vpn_packet(ls, &pkt, &addr);
+#endif
 }
 
 void handle_device_data(void *data, int flags) {
+       (void)data;
+       (void)flags;
        vpn_packet_t packet;
        packet.offset = DEFAULT_PACKET_OFFSET;
        packet.priority = 0;
+       static int errors = 0;
 
        if(devops.read(&packet)) {
+               errors = 0;
                myself->in_packets++;
                myself->in_bytes += packet.len;
                route(myself, &packet);
+       } else {
+               usleep(errors * 50000);
+               errors++;
+
+               if(errors > 10) {
+                       logger(DEBUG_ALWAYS, LOG_ERR, "Too many errors from %s, exiting!", device);
+                       event_exit();
+               }
        }
 }