6d8a64ec30b1f445366d0848ebb2f1009ac245fe
[l2tpns.git] / cluster.c
1 // L2TPNS Clustering Stuff
2
3 char const *cvs_id_cluster = "$Id: cluster.c,v 1.45 2005/07/31 10:04:09 bodea Exp $";
4
5 #include <stdio.h>
6 #include <stdlib.h>
7 #include <stdarg.h>
8 #include <unistd.h>
9 #include <inttypes.h>
10 #include <sys/file.h>
11 #include <sys/stat.h>
12 #include <sys/socket.h>
13 #include <netinet/in.h>
14 #include <arpa/inet.h>
15 #include <sys/ioctl.h>
16 #include <net/if.h>
17 #include <string.h>
18 #include <malloc.h>
19 #include <errno.h>
20 #include <libcli.h>
21
22 #include "l2tpns.h"
23 #include "cluster.h"
24 #include "util.h"
25 #include "tbf.h"
26
27 #ifdef BGP
28 #include "bgp.h"
29 #endif
30 /*
31 * All cluster packets have the same format.
32 *
33 * One or more instances of
34 * a 32 bit 'type' id.
35 * a 32 bit 'extra' data dependant on the 'type'.
36 * zero or more bytes of structure data, dependant on the type.
37 *
38 */
39
40 // Module variables.
41 extern int cluster_sockfd; // The filedescriptor for the cluster communications port.
42
43 in_addr_t my_address = 0; // The network address of my ethernet port.
44 static int walk_session_number = 0; // The next session to send when doing the slow table walk.
45 static int walk_tunnel_number = 0; // The next tunnel to send when doing the slow table walk.
46 int forked = 0; // Sanity check: CLI must not diddle with heartbeat table
47
48 #define MAX_HEART_SIZE (8192) // Maximum size of heartbeat packet. Must be less than max IP packet size :)
49 #define MAX_CHANGES (MAX_HEART_SIZE/(sizeof(sessiont) + sizeof(int) ) - 2) // Assumes a session is the biggest type!
50
51 static struct {
52 int type;
53 int id;
54 } cluster_changes[MAX_CHANGES]; // Queue of changed structures that need to go out when next heartbeat.
55
56 static struct {
57 int seq;
58 int size;
59 uint8_t data[MAX_HEART_SIZE];
60 } past_hearts[HB_HISTORY_SIZE]; // Ring buffer of heartbeats that we've recently sent out. Needed so
61 // we can re-transmit if needed.
62
63 static struct {
64 in_addr_t peer;
65 uint32_t basetime;
66 clockt timestamp;
67 int uptodate;
68 } peers[CLUSTER_MAX_SIZE]; // List of all the peers we've heard from.
69 static int num_peers; // Number of peers in list.
70
71 static int rle_decompress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize);
72 static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize);
73
74 //
75 // Create a listening socket
76 //
77 // This joins the cluster multi-cast group.
78 //
79 int cluster_init()
80 {
81 struct sockaddr_in addr;
82 struct sockaddr_in interface_addr;
83 struct ip_mreq mreq;
84 struct ifreq ifr;
85 int opt;
86
87 config->cluster_undefined_sessions = MAXSESSION-1;
88 config->cluster_undefined_tunnels = MAXTUNNEL-1;
89
90 if (!config->cluster_address)
91 return 0;
92 if (!*config->cluster_interface)
93 return 0;
94
95 cluster_sockfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
96
97 memset(&addr, 0, sizeof(addr));
98 addr.sin_family = AF_INET;
99 addr.sin_port = htons(CLUSTERPORT);
100 addr.sin_addr.s_addr = INADDR_ANY;
101 setsockopt(cluster_sockfd, SOL_SOCKET, SO_REUSEADDR, &addr, sizeof(addr));
102
103 opt = fcntl(cluster_sockfd, F_GETFL, 0);
104 fcntl(cluster_sockfd, F_SETFL, opt | O_NONBLOCK);
105
106 if (bind(cluster_sockfd, (void *) &addr, sizeof(addr)) < 0)
107 {
108 LOG(0, 0, 0, "Failed to bind cluster socket: %s\n", strerror(errno));
109 return -1;
110 }
111
112 strcpy(ifr.ifr_name, config->cluster_interface);
113 if (ioctl(cluster_sockfd, SIOCGIFADDR, &ifr) < 0)
114 {
115 LOG(0, 0, 0, "Failed to get interface address for (%s): %s\n", config->cluster_interface, strerror(errno));
116 return -1;
117 }
118
119 memcpy(&interface_addr, &ifr.ifr_addr, sizeof(interface_addr));
120 my_address = interface_addr.sin_addr.s_addr;
121
122 // Join multicast group.
123 mreq.imr_multiaddr.s_addr = config->cluster_address;
124 mreq.imr_interface = interface_addr.sin_addr;
125
126
127 opt = 0; // Turn off multicast loopback.
128 setsockopt(cluster_sockfd, IPPROTO_IP, IP_MULTICAST_LOOP, &opt, sizeof(opt));
129
130 if (setsockopt(cluster_sockfd, IPPROTO_IP, IP_ADD_MEMBERSHIP, &mreq, sizeof(mreq)) < 0)
131 {
132 LOG(0, 0, 0, "Failed to setsockopt (join mcast group): %s\n", strerror(errno));
133 return -1;
134 }
135
136 if (setsockopt(cluster_sockfd, IPPROTO_IP, IP_MULTICAST_IF, &interface_addr, sizeof(interface_addr)) < 0)
137 {
138 LOG(0, 0, 0, "Failed to setsockopt (set mcast interface): %s\n", strerror(errno));
139 return -1;
140 }
141
142 config->cluster_last_hb = TIME;
143 config->cluster_seq_number = -1;
144
145 return cluster_sockfd;
146 }
147
148
149 //
150 // Send a chunk of data to the entire cluster (usually via the multicast
151 // address ).
152 //
153
154 static int cluster_send_data(void *data, int datalen)
155 {
156 struct sockaddr_in addr = {0};
157
158 if (!cluster_sockfd) return -1;
159 if (!config->cluster_address) return 0;
160
161 addr.sin_addr.s_addr = config->cluster_address;
162 addr.sin_port = htons(CLUSTERPORT);
163 addr.sin_family = AF_INET;
164
165 LOG(5, 0, 0, "Cluster send data: %d bytes\n", datalen);
166
167 if (sendto(cluster_sockfd, data, datalen, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
168 {
169 LOG(0, 0, 0, "sendto: %s\n", strerror(errno));
170 return -1;
171 }
172
173 return 0;
174 }
175
176 //
177 // Add a chunk of data to a heartbeat packet.
178 // Maintains the format. Assumes that the caller
179 // has passed in a big enough buffer!
180 //
181 static void add_type(uint8_t **p, int type, int more, uint8_t *data, int size)
182 {
183 *((uint32_t *) (*p)) = type;
184 *p += sizeof(uint32_t);
185
186 *((uint32_t *)(*p)) = more;
187 *p += sizeof(uint32_t);
188
189 if (data && size > 0) {
190 memcpy(*p, data, size);
191 *p += size;
192 }
193 }
194
195 // advertise our presence via BGP or gratuitous ARP
196 static void advertise_routes(void)
197 {
198 #ifdef BGP
199 if (bgp_configured)
200 bgp_enable_routing(1);
201 else
202 #endif /* BGP */
203 if (config->send_garp)
204 send_garp(config->bind_address); // Start taking traffic.
205 }
206
207 // withdraw our routes (BGP only)
208 static void withdraw_routes(void)
209 {
210 #ifdef BGP
211 if (bgp_configured)
212 bgp_enable_routing(0);
213 #endif /* BGP */
214 }
215
216 static void cluster_uptodate(void)
217 {
218 if (config->cluster_iam_uptodate)
219 return;
220
221 if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels)
222 return;
223
224 config->cluster_iam_uptodate = 1;
225
226 LOG(0, 0, 0, "Now uptodate with master.\n");
227 advertise_routes();
228 }
229
230 //
231 // Send a unicast UDP packet to a peer with 'data' as the
232 // contents.
233 //
234 static int peer_send_data(in_addr_t peer, uint8_t *data, int size)
235 {
236 struct sockaddr_in addr = {0};
237
238 if (!cluster_sockfd) return -1;
239 if (!config->cluster_address) return 0;
240
241 if (!peer) // Odd??
242 return -1;
243
244 addr.sin_addr.s_addr = peer;
245 addr.sin_port = htons(CLUSTERPORT);
246 addr.sin_family = AF_INET;
247
248 LOG_HEX(5, "Peer send", data, size);
249
250 if (sendto(cluster_sockfd, data, size, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
251 {
252 LOG(0, 0, 0, "sendto: %s\n", strerror(errno));
253 return -1;
254 }
255
256 return 0;
257 }
258
259 //
260 // Send a structured message to a peer with a single element of type 'type'.
261 //
262 static int peer_send_message(in_addr_t peer, int type, int more, uint8_t *data, int size)
263 {
264 uint8_t buf[65536]; // Vast overkill.
265 uint8_t *p = buf;
266
267 LOG(4, 0, 0, "Sending message to peer (type %d, more %d, size %d)\n", type, more, size);
268 add_type(&p, type, more, data, size);
269
270 return peer_send_data(peer, buf, (p-buf) );
271 }
272
273 // send a packet to the master
274 static int _forward_packet(uint8_t *data, int size, in_addr_t addr, int port, int type)
275 {
276 uint8_t buf[65536]; // Vast overkill.
277 uint8_t *p = buf;
278
279 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
280 return -1;
281
282 LOG(4, 0, 0, "Forwarding packet from %s to master (size %d)\n", fmtaddr(addr, 0), size);
283
284 STAT(c_forwarded);
285 add_type(&p, type, addr, (uint8_t *) &port, sizeof(port)); // ick. should be uint16_t
286 memcpy(p, data, size);
287 p += size;
288
289 return peer_send_data(config->cluster_master_address, buf, (p - buf));
290 }
291
292 //
293 // Forward a state changing packet to the master.
294 //
295 // The master just processes the payload as if it had
296 // received it off the tun device.
297 //
298 int master_forward_packet(uint8_t *data, int size, in_addr_t addr, int port)
299 {
300 return _forward_packet(data, size, addr, port, C_FORWARD);
301 }
302
303 // Forward a DAE RADIUS packet to the master.
304 int master_forward_dae_packet(uint8_t *data, int size, in_addr_t addr, int port)
305 {
306 return _forward_packet(data, size, addr, port, C_FORWARD_DAE);
307 }
308
309 //
310 // Forward a throttled packet to the master for handling.
311 //
312 // The master just drops the packet into the appropriate
313 // token bucket queue, and lets normal processing take care
314 // of it.
315 //
316 int master_throttle_packet(int tbfid, uint8_t *data, int size)
317 {
318 uint8_t buf[65536]; // Vast overkill.
319 uint8_t *p = buf;
320
321 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
322 return -1;
323
324 LOG(4, 0, 0, "Throttling packet master (size %d, tbfid %d)\n", size, tbfid);
325
326 add_type(&p, C_THROTTLE, tbfid, data, size);
327
328 return peer_send_data(config->cluster_master_address, buf, (p-buf) );
329
330 }
331
332 //
333 // Forward a walled garden packet to the master for handling.
334 //
335 // The master just writes the packet straight to the tun
336 // device (where is will normally loop through the
337 // firewall rules, and come back in on the tun device)
338 //
339 // (Note that this must be called with the tun header
340 // as the start of the data).
341 int master_garden_packet(sessionidt s, uint8_t *data, int size)
342 {
343 uint8_t buf[65536]; // Vast overkill.
344 uint8_t *p = buf;
345
346 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
347 return -1;
348
349 LOG(4, 0, 0, "Walled garden packet to master (size %d)\n", size);
350
351 add_type(&p, C_GARDEN, s, data, size);
352
353 return peer_send_data(config->cluster_master_address, buf, (p-buf));
354
355 }
356
357 //
358 // Send a chunk of data as a heartbeat..
359 // We save it in the history buffer as we do so.
360 //
361 static void send_heartbeat(int seq, uint8_t *data, int size)
362 {
363 int i;
364
365 if (size > sizeof(past_hearts[0].data))
366 {
367 LOG(0, 0, 0, "Tried to heartbeat something larger than the maximum packet!\n");
368 kill(0, SIGTERM);
369 exit(1);
370 }
371 i = seq % HB_HISTORY_SIZE;
372 past_hearts[i].seq = seq;
373 past_hearts[i].size = size;
374 memcpy(&past_hearts[i].data, data, size); // Save it.
375 cluster_send_data(data, size);
376 }
377
378 //
379 // Send an 'i am alive' message to every machine in the cluster.
380 //
381 void cluster_send_ping(time_t basetime)
382 {
383 uint8_t buff[100 + sizeof(pingt)];
384 uint8_t *p = buff;
385 pingt x;
386
387 if (config->cluster_iam_master && basetime) // We're heartbeating so no need to ping.
388 return;
389
390 LOG(5, 0, 0, "Sending cluster ping...\n");
391
392 x.ver = 1;
393 x.addr = config->bind_address;
394 x.undef = config->cluster_undefined_sessions + config->cluster_undefined_tunnels;
395 x.basetime = basetime;
396
397 add_type(&p, C_PING, basetime, (uint8_t *) &x, sizeof(x));
398 cluster_send_data(buff, (p-buff) );
399 }
400
401 //
402 // Walk the session counters looking for non-zero ones to send
403 // to the master. We send up to 600 of them at one time.
404 // We examine a maximum of 3000 sessions.
405 // (50k max session should mean that we normally
406 // examine the entire session table every 25 seconds).
407
408 #define MAX_B_RECS (600)
409 void master_update_counts(void)
410 {
411 int i, c;
412 bytest b[MAX_B_RECS+1];
413
414 if (config->cluster_iam_master) // Only happens on the slaves.
415 return;
416
417 if (!config->cluster_master_address) // If we don't have a master, skip it for a while.
418 return;
419
420 // C_BYTES format changed in 2.1.0 (cluster version 5)
421 // during upgrade from previous versions, hang onto our counters
422 // for a bit until the new master comes up
423 if (config->cluster_last_hb_ver < 5)
424 return;
425
426 i = MAX_B_RECS * 5; // Examine max 3000 sessions;
427 if (config->cluster_highest_sessionid > i)
428 i = config->cluster_highest_sessionid;
429
430 for ( c = 0; i > 0 ; --i) {
431 // Next session to look at.
432 walk_session_number++;
433 if ( walk_session_number > config->cluster_highest_sessionid)
434 walk_session_number = 1;
435
436 if (!sess_local[walk_session_number].cin && !sess_local[walk_session_number].cout)
437 continue; // Unchanged. Skip it.
438
439 b[c].sid = walk_session_number;
440 b[c].pin = sess_local[walk_session_number].pin;
441 b[c].pout = sess_local[walk_session_number].pout;
442 b[c].cin = sess_local[walk_session_number].cin;
443 b[c].cout = sess_local[walk_session_number].cout;
444
445 // Reset counters.
446 sess_local[walk_session_number].pin = sess_local[walk_session_number].pout = 0;
447 sess_local[walk_session_number].cin = sess_local[walk_session_number].cout = 0;
448
449 if (++c > MAX_B_RECS) // Send a max of 600 elements in a packet.
450 break;
451 }
452
453 if (!c) // Didn't find any that changes. Get out of here!
454 return;
455
456
457 // Forward the data to the master.
458 LOG(4, 0, 0, "Sending byte counters to master (%d elements)\n", c);
459 peer_send_message(config->cluster_master_address, C_BYTES, c, (uint8_t *) &b, sizeof(b[0]) * c);
460 return;
461 }
462
463 //
464 // On the master, check how our slaves are going. If
465 // one of them's not up-to-date we'll heartbeat faster.
466 // If we don't have any of them, then we need to turn
467 // on our own packet handling!
468 //
469 void cluster_check_slaves(void)
470 {
471 int i;
472 static int have_peers = 0;
473 int had_peers = have_peers;
474 clockt t = TIME;
475
476 if (!config->cluster_iam_master)
477 return; // Only runs on the master...
478
479 config->cluster_iam_uptodate = 1; // cleared in loop below
480
481 for (i = have_peers = 0; i < num_peers; i++)
482 {
483 if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
484 continue; // Stale peer! Skip them.
485
486 if (!peers[i].basetime)
487 continue; // Shutdown peer! Skip them.
488
489 if (peers[i].uptodate)
490 have_peers++;
491 else
492 config->cluster_iam_uptodate = 0; // Start fast heartbeats
493 }
494
495 // in a cluster, withdraw/add routes when we get a peer/lose peers
496 if (have_peers != had_peers)
497 {
498 if (had_peers < config->cluster_master_min_adv &&
499 have_peers >= config->cluster_master_min_adv)
500 withdraw_routes();
501
502 else if (had_peers >= config->cluster_master_min_adv &&
503 have_peers < config->cluster_master_min_adv)
504 advertise_routes();
505 }
506 }
507
508 //
509 // Check that we have a master. If it's been too
510 // long since we heard from a master then hold an election.
511 //
512 void cluster_check_master(void)
513 {
514 int i, count, tcount, high_unique_id = 0;
515 int last_free = 0;
516 clockt t = TIME;
517 static int probed = 0;
518 int have_peers;
519
520 if (config->cluster_iam_master)
521 return; // Only runs on the slaves...
522
523 // If the master is late (missed 2 hearbeats by a second and a
524 // hair) it may be that the switch has dropped us from the
525 // multicast group, try unicasting probes to the master
526 // which will hopefully respond with a unicast heartbeat that
527 // will allow us to limp along until the querier next runs.
528 if (config->cluster_master_address
529 && TIME > (config->cluster_last_hb + 2 * config->cluster_hb_interval + 11))
530 {
531 if (!probed || (TIME > (probed + 2 * config->cluster_hb_interval)))
532 {
533 probed = TIME;
534 LOG(1, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
535 0.1 * (TIME - (config->cluster_last_hb + config->cluster_hb_interval)));
536
537 peer_send_message(config->cluster_master_address,
538 C_LASTSEEN, config->cluster_seq_number, NULL, 0);
539 }
540 } else { // We got a recent heartbeat; reset the probe flag.
541 probed = 0;
542 }
543
544 if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout))
545 return; // Everything's ok!
546
547 config->cluster_last_hb = TIME + 1; // Just the one election thanks.
548 config->cluster_master_address = 0;
549
550 LOG(0, 0, 0, "Master timed out! Holding election...\n");
551
552 // In the process of shutting down, can't be master
553 if (main_quit)
554 return;
555
556 for (i = have_peers = 0; i < num_peers; i++)
557 {
558 if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
559 continue; // Stale peer! Skip them.
560
561 if (!peers[i].basetime)
562 continue; // Shutdown peer! Skip them.
563
564 if (peers[i].basetime < basetime) {
565 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
566 return; // They'll win the election. Get out of here.
567 }
568
569 if (peers[i].basetime == basetime &&
570 peers[i].peer > my_address) {
571 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
572 return; // They'll win the election. Wait for them to come up.
573 }
574
575 if (peers[i].uptodate)
576 have_peers++;
577 }
578
579 // Wow. it's been ages since I last heard a heartbeat
580 // and I'm better than an of my peers so it's time
581 // to become a master!!!
582
583 config->cluster_iam_master = 1;
584
585 LOG(0, 0, 0, "I am declaring myself the master!\n");
586
587 if (have_peers < config->cluster_master_min_adv)
588 advertise_routes();
589 else
590 withdraw_routes();
591
592 if (config->cluster_seq_number == -1)
593 config->cluster_seq_number = 0;
594
595 //
596 // Go through and mark all the tunnels as defined.
597 // Count the highest used tunnel number as well.
598 //
599 config->cluster_highest_tunnelid = 0;
600 for (i = 0, tcount = 0; i < MAXTUNNEL; ++i) {
601 if (tunnel[i].state == TUNNELUNDEF)
602 tunnel[i].state = TUNNELFREE;
603
604 if (tunnel[i].state != TUNNELFREE && i > config->cluster_highest_tunnelid)
605 config->cluster_highest_tunnelid = i;
606 }
607
608 //
609 // Go through and mark all the sessions as being defined.
610 // reset the idle timeouts.
611 // add temporary byte counters to permanent ones.
612 // Re-string the free list.
613 // Find the ID of the highest session.
614 last_free = 0;
615 high_unique_id = 0;
616 config->cluster_highest_sessionid = 0;
617 for (i = 0, count = 0; i < MAXSESSION; ++i) {
618 if (session[i].tunnel == T_UNDEF) {
619 session[i].tunnel = T_FREE;
620 ++count;
621 }
622
623 if (!session[i].opened) { // Unused session. Add to free list.
624 memset(&session[i], 0, sizeof(session[i]));
625 session[i].tunnel = T_FREE;
626 session[last_free].next = i;
627 session[i].next = 0;
628 last_free = i;
629 continue;
630 }
631
632 // Reset idle timeouts..
633 session[i].last_packet = time_now;
634
635 // Reset die relative to our uptime rather than the old master's
636 if (session[i].die) session[i].die = TIME;
637
638 // Accumulate un-sent byte/packet counters.
639 increment_counter(&session[i].cin, &session[i].cin_wrap, sess_local[i].cin);
640 increment_counter(&session[i].cout, &session[i].cout_wrap, sess_local[i].cout);
641 session[i].cin_delta += sess_local[i].cin;
642 session[i].cout_delta += sess_local[i].cout;
643
644 session[i].pin += sess_local[i].pin;
645 session[i].pout += sess_local[i].pout;
646
647 sess_local[i].cin = sess_local[i].cout = 0;
648 sess_local[i].pin = sess_local[i].pout = 0;
649
650 sess_local[i].radius = 0; // Reset authentication as the radius blocks aren't up to date.
651
652 if (session[i].unique_id >= high_unique_id) // This is different to the index into the session table!!!
653 high_unique_id = session[i].unique_id+1;
654
655 session[i].tbf_in = session[i].tbf_out = 0; // Remove stale pointers from old master.
656 throttle_session(i, session[i].throttle_in, session[i].throttle_out);
657
658 config->cluster_highest_sessionid = i;
659 }
660
661 session[last_free].next = 0; // End of chain.
662 last_id = high_unique_id; // Keep track of the highest used session ID.
663
664 become_master();
665
666 rebuild_address_pool();
667
668 // If we're not the very first master, this is a big issue!
669 if(count>0)
670 LOG(0, 0, 0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count);
671
672 config->cluster_undefined_sessions = 0;
673 config->cluster_undefined_tunnels = 0;
674 config->cluster_iam_uptodate = 1; // assume all peers are up-to-date
675
676 // FIXME. We need to fix up the tunnel control message
677 // queue here! There's a number of other variables we
678 // should also update.
679 }
680
681
682 //
683 // Check that our session table is validly matching what the
684 // master has in mind.
685 //
686 // In particular, if we have too many sessions marked 'undefined'
687 // we fix it up here, and we ensure that the 'first free session'
688 // pointer is valid.
689 //
690 static void cluster_check_sessions(int highsession, int freesession_ptr, int hightunnel)
691 {
692 int i;
693
694 sessionfree = freesession_ptr; // Keep the freesession ptr valid.
695
696 if (config->cluster_iam_uptodate)
697 return;
698
699 if (highsession > config->cluster_undefined_sessions && hightunnel > config->cluster_undefined_tunnels)
700 return;
701
702 // Clear out defined sessions, counting the number of
703 // undefs remaining.
704 config->cluster_undefined_sessions = 0;
705 for (i = 1 ; i < MAXSESSION; ++i) {
706 if (i > highsession) {
707 if (session[i].tunnel == T_UNDEF) session[i].tunnel = T_FREE; // Defined.
708 continue;
709 }
710
711 if (session[i].tunnel == T_UNDEF)
712 ++config->cluster_undefined_sessions;
713 }
714
715 // Clear out defined tunnels, counting the number of
716 // undefs remaining.
717 config->cluster_undefined_tunnels = 0;
718 for (i = 1 ; i < MAXTUNNEL; ++i) {
719 if (i > hightunnel) {
720 if (tunnel[i].state == TUNNELUNDEF) tunnel[i].state = TUNNELFREE; // Defined.
721 continue;
722 }
723
724 if (tunnel[i].state == TUNNELUNDEF)
725 ++config->cluster_undefined_tunnels;
726 }
727
728
729 if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels) {
730 LOG(2, 0, 0, "Cleared undefined sessions/tunnels. %d sess (high %d), %d tunn (high %d)\n",
731 config->cluster_undefined_sessions, highsession, config->cluster_undefined_tunnels, hightunnel);
732 return;
733 }
734
735 // Are we up to date?
736
737 if (!config->cluster_iam_uptodate)
738 cluster_uptodate();
739 }
740
741 static int hb_add_type(uint8_t **p, int type, int id)
742 {
743 switch (type) {
744 case C_CSESSION: { // Compressed C_SESSION.
745 uint8_t c[sizeof(sessiont) * 2]; // Bigger than worst case.
746 uint8_t *d = (uint8_t *) &session[id];
747 uint8_t *orig = d;
748 int size;
749
750 size = rle_compress( &d, sizeof(sessiont), c, sizeof(c) );
751
752 // Did we compress the full structure, and is the size actually
753 // reduced??
754 if ( (d - orig) == sizeof(sessiont) && size < sizeof(sessiont) ) {
755 add_type(p, C_CSESSION, id, c, size);
756 break;
757 }
758 // Failed to compress : Fall through.
759 }
760 case C_SESSION:
761 add_type(p, C_SESSION, id, (uint8_t *) &session[id], sizeof(sessiont));
762 break;
763
764 case C_CTUNNEL: { // Compressed C_TUNNEL
765 uint8_t c[sizeof(tunnelt) * 2]; // Bigger than worst case.
766 uint8_t *d = (uint8_t *) &tunnel[id];
767 uint8_t *orig = d;
768 int size;
769
770 size = rle_compress( &d, sizeof(tunnelt), c, sizeof(c) );
771
772 // Did we compress the full structure, and is the size actually
773 // reduced??
774 if ( (d - orig) == sizeof(tunnelt) && size < sizeof(tunnelt) ) {
775 add_type(p, C_CTUNNEL, id, c, size);
776 break;
777 }
778 // Failed to compress : Fall through.
779 }
780 case C_TUNNEL:
781 add_type(p, C_TUNNEL, id, (uint8_t *) &tunnel[id], sizeof(tunnelt));
782 break;
783 default:
784 LOG(0, 0, 0, "Found an invalid type in heart queue! (%d)\n", type);
785 kill(0, SIGTERM);
786 exit(1);
787 }
788 return 0;
789 }
790
791 //
792 // Send a heartbeat, incidently sending out any queued changes..
793 //
794 void cluster_heartbeat()
795 {
796 int i, count = 0, tcount = 0;
797 uint8_t buff[MAX_HEART_SIZE + sizeof(heartt) + sizeof(int) ];
798 heartt h;
799 uint8_t *p = buff;
800
801 if (!config->cluster_iam_master) // Only the master does this.
802 return;
803
804 config->cluster_table_version += config->cluster_num_changes;
805
806 // Fill out the heartbeat header.
807 memset(&h, 0, sizeof(h));
808
809 h.version = HB_VERSION;
810 h.seq = config->cluster_seq_number;
811 h.basetime = basetime;
812 h.clusterid = config->bind_address; // Will this do??
813 h.basetime = basetime;
814 h.highsession = config->cluster_highest_sessionid;
815 h.freesession = sessionfree;
816 h.hightunnel = config->cluster_highest_tunnelid;
817 h.size_sess = sizeof(sessiont); // Just in case.
818 h.size_tunn = sizeof(tunnelt);
819 h.interval = config->cluster_hb_interval;
820 h.timeout = config->cluster_hb_timeout;
821 h.table_version = config->cluster_table_version;
822
823 add_type(&p, C_HEARTBEAT, HB_VERSION, (uint8_t *) &h, sizeof(h));
824
825 for (i = 0; i < config->cluster_num_changes; ++i) {
826 hb_add_type(&p, cluster_changes[i].type, cluster_changes[i].id);
827 }
828
829 if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
830 LOG(0, 0, 0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", (int) (p - buff));
831 kill(0, SIGTERM);
832 exit(1);
833 }
834
835 //
836 // Fill out the packet with sessions from the session table...
837 // (not forgetting to leave space so we can get some tunnels in too )
838 while ( (p + sizeof(uint32_t) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) {
839
840 if (!walk_session_number) // session #0 isn't valid.
841 ++walk_session_number;
842
843 if (count >= config->cluster_highest_sessionid) // If we're a small cluster, don't go wild.
844 break;
845
846 hb_add_type(&p, C_CSESSION, walk_session_number);
847 walk_session_number = (1+walk_session_number)%(config->cluster_highest_sessionid+1); // +1 avoids divide by zero.
848
849 ++count; // Count the number of extra sessions we're sending.
850 }
851
852 //
853 // Fill out the packet with tunnels from the tunnel table...
854 // This effectively means we walk the tunnel table more quickly
855 // than the session table. This is good because stuffing up a
856 // tunnel is a much bigger deal than stuffing up a session.
857 //
858 while ( (p + sizeof(uint32_t) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {
859
860 if (!walk_tunnel_number) // tunnel #0 isn't valid.
861 ++walk_tunnel_number;
862
863 if (tcount >= config->cluster_highest_tunnelid)
864 break;
865
866 hb_add_type(&p, C_CTUNNEL, walk_tunnel_number);
867 walk_tunnel_number = (1+walk_tunnel_number)%(config->cluster_highest_tunnelid+1); // +1 avoids divide by zero.
868
869 ++tcount;
870 }
871
872 //
873 // Did we do something wrong?
874 if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
875 LOG(0, 0, 0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", (int) (p - buff));
876 kill(0, SIGTERM);
877 exit(1);
878 }
879
880 LOG(3, 0, 0, "Sending v%d heartbeat #%d, change #%" PRIu64 " with %d changes "
881 "(%d x-sess, %d x-tunnels, %d highsess, %d hightun, size %d)\n",
882 HB_VERSION, h.seq, h.table_version, config->cluster_num_changes,
883 count, tcount, config->cluster_highest_sessionid,
884 config->cluster_highest_tunnelid, (int) (p - buff));
885
886 config->cluster_num_changes = 0;
887
888 send_heartbeat(h.seq, buff, (p-buff) ); // Send out the heartbeat to the cluster, keeping a copy of it.
889
890 config->cluster_seq_number = (config->cluster_seq_number+1)%HB_MAX_SEQ; // Next seq number to use.
891 }
892
893 //
894 // A structure of type 'type' has changed; Add it to the queue to send.
895 //
896 static int type_changed(int type, int id)
897 {
898 int i;
899
900 for (i = 0 ; i < config->cluster_num_changes ; ++i)
901 if ( cluster_changes[i].id == id &&
902 cluster_changes[i].type == type)
903 return 0; // Already marked for change.
904
905 cluster_changes[i].type = type;
906 cluster_changes[i].id = id;
907 ++config->cluster_num_changes;
908
909 if (config->cluster_num_changes > MAX_CHANGES)
910 cluster_heartbeat(); // flush now
911
912 return 1;
913 }
914
915
916 // A particular session has been changed!
917 int cluster_send_session(int sid)
918 {
919 if (!config->cluster_iam_master) {
920 LOG(0, sid, 0, "I'm not a master, but I just tried to change a session!\n");
921 return -1;
922 }
923
924 if (forked) {
925 LOG(0, sid, 0, "cluster_send_session called from child process!\n");
926 return -1;
927 }
928
929 return type_changed(C_CSESSION, sid);
930 }
931
932 // A particular tunnel has been changed!
933 int cluster_send_tunnel(int tid)
934 {
935 if (!config->cluster_iam_master) {
936 LOG(0, 0, tid, "I'm not a master, but I just tried to change a tunnel!\n");
937 return -1;
938 }
939
940 return type_changed(C_CTUNNEL, tid);
941 }
942
943
944 //
945 // We're a master, and a slave has just told us that it's
946 // missed a packet. We'll resend it every packet since
947 // the last one it's seen.
948 //
949 static int cluster_catchup_slave(int seq, in_addr_t slave)
950 {
951 int s;
952 int diff;
953
954 LOG(1, 0, 0, "Slave %s sent LASTSEEN with seq %d\n", fmtaddr(slave, 0), seq);
955 if (!config->cluster_iam_master) {
956 LOG(1, 0, 0, "Got LASTSEEN but I'm not a master! Redirecting it to %s.\n",
957 fmtaddr(config->cluster_master_address, 0));
958
959 peer_send_message(slave, C_MASTER, config->cluster_master_address, NULL, 0);
960 return 0;
961 }
962
963 diff = config->cluster_seq_number - seq; // How many packet do we need to send?
964 if (diff < 0)
965 diff += HB_MAX_SEQ;
966
967 if (diff >= HB_HISTORY_SIZE) { // Ouch. We don't have the packet to send it!
968 LOG(0, 0, 0, "A slave asked for message %d when our seq number is %d. Killing it.\n",
969 seq, config->cluster_seq_number);
970 return peer_send_message(slave, C_KILL, seq, NULL, 0);// Kill the slave. Nothing else to do.
971 }
972
973 LOG(1, 0, 0, "Sending %d catchup packets to slave %s\n", diff, fmtaddr(slave, 0) );
974
975 // Now resend every packet that it missed, in order.
976 while (seq != config->cluster_seq_number) {
977 s = seq % HB_HISTORY_SIZE;
978 if (seq != past_hearts[s].seq) {
979 LOG(0, 0, 0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n",
980 fmtaddr(slave, 0), seq, past_hearts[s].seq, s, config->cluster_seq_number);
981 return -1; // What to do here!?
982 }
983 peer_send_data(slave, past_hearts[s].data, past_hearts[s].size);
984 seq = (seq+1)%HB_MAX_SEQ; // Increment to next seq number.
985 }
986 return 0; // All good!
987 }
988
989 //
990 // We've heard from another peer! Add it to the list
991 // that we select from at election time.
992 //
993 static int cluster_add_peer(in_addr_t peer, time_t basetime, pingt *pp, int size)
994 {
995 int i;
996 in_addr_t clusterid;
997 pingt p;
998
999 // Allow for backward compatability.
1000 // Just the ping packet into a new structure to allow
1001 // for the possibility that we might have received
1002 // more or fewer elements than we were expecting.
1003 if (size > sizeof(p))
1004 size = sizeof(p);
1005
1006 memset( (void *) &p, 0, sizeof(p) );
1007 memcpy( (void *) &p, (void *) pp, size);
1008
1009 clusterid = p.addr;
1010 if (clusterid != config->bind_address)
1011 {
1012 // Is this for us?
1013 LOG(4, 0, 0, "Skipping ping from %s (different cluster)\n", fmtaddr(peer, 0));
1014 return 0;
1015 }
1016
1017 for (i = 0; i < num_peers ; ++i)
1018 {
1019 if (peers[i].peer != peer)
1020 continue;
1021
1022 // This peer already exists. Just update the timestamp.
1023 peers[i].basetime = basetime;
1024 peers[i].timestamp = TIME;
1025 peers[i].uptodate = !p.undef;
1026 break;
1027 }
1028
1029 // Is this the master shutting down??
1030 if (peer == config->cluster_master_address) {
1031 LOG(3, 0, 0, "Master %s %s\n", fmtaddr(config->cluster_master_address, 0),
1032 basetime ? "has restarted!" : "shutting down...");
1033
1034 config->cluster_master_address = 0;
1035 config->cluster_last_hb = 0; // Force an election.
1036 cluster_check_master();
1037 }
1038
1039 if (i >= num_peers)
1040 {
1041 LOG(4, 0, 0, "Adding %s as a peer\n", fmtaddr(peer, 0));
1042
1043 // Not found. Is there a stale slot to re-use?
1044 for (i = 0; i < num_peers ; ++i)
1045 {
1046 if (!peers[i].basetime) // Shutdown
1047 break;
1048
1049 if ((peers[i].timestamp + config->cluster_hb_timeout * 10) < TIME) // Stale.
1050 break;
1051 }
1052
1053 if (i >= CLUSTER_MAX_SIZE)
1054 {
1055 // Too many peers!!
1056 LOG(0, 0, 0, "Tried to add %s as a peer, but I already have %d of them!\n", fmtaddr(peer, 0), i);
1057 return -1;
1058 }
1059
1060 peers[i].peer = peer;
1061 peers[i].basetime = basetime;
1062 peers[i].timestamp = TIME;
1063 peers[i].uptodate = !p.undef;
1064 if (i == num_peers)
1065 ++num_peers;
1066
1067 LOG(1, 0, 0, "Added %s as a new peer. Now %d peers\n", fmtaddr(peer, 0), num_peers);
1068 }
1069
1070 return 1;
1071 }
1072
1073 // A slave responds with C_MASTER when it gets a message which should have gone to a master.
1074 static int cluster_set_master(in_addr_t peer, in_addr_t master)
1075 {
1076 if (config->cluster_iam_master) // Sanity...
1077 return 0;
1078
1079 LOG(3, 0, 0, "Peer %s set the master to %s...\n", fmtaddr(peer, 0),
1080 fmtaddr(master, 1));
1081
1082 config->cluster_master_address = master;
1083 if (master)
1084 {
1085 // catchup with new master
1086 peer_send_message(master, C_LASTSEEN, config->cluster_seq_number, NULL, 0);
1087
1088 // delay next election
1089 config->cluster_last_hb = TIME;
1090 }
1091
1092 // run election (or reset "probed" if master was set)
1093 cluster_check_master();
1094 return 0;
1095 }
1096
1097 /* Handle the slave updating the byte counters for the master. */
1098 //
1099 // Note that we don't mark the session as dirty; We rely on
1100 // the slow table walk to propogate this back out to the slaves.
1101 //
1102 static int cluster_handle_bytes(uint8_t *data, int size)
1103 {
1104 bytest *b;
1105
1106 b = (bytest *) data;
1107
1108 LOG(3, 0, 0, "Got byte counter update (size %d)\n", size);
1109
1110 /* Loop around, adding the byte
1111 counts to each of the sessions. */
1112
1113 while (size >= sizeof(*b) ) {
1114 if (b->sid > MAXSESSION) {
1115 LOG(0, 0, 0, "Got C_BYTES with session #%d!\n", b->sid);
1116 return -1; /* Abort processing */
1117 }
1118
1119 session[b->sid].pin += b->pin;
1120 session[b->sid].pout += b->pout;
1121
1122 increment_counter(&session[b->sid].cin, &session[b->sid].cin_wrap, b->cin);
1123 increment_counter(&session[b->sid].cout, &session[b->sid].cout_wrap, b->cout);
1124
1125 session[b->sid].cin_delta += b->cin;
1126 session[b->sid].cout_delta += b->cout;
1127
1128 if (b->cin)
1129 session[b->sid].last_packet = time_now; // Reset idle timer!
1130
1131 size -= sizeof(*b);
1132 ++b;
1133 }
1134
1135 if (size != 0)
1136 LOG(0, 0, 0, "Got C_BYTES with %d bytes of trailing junk!\n", size);
1137
1138 return size;
1139 }
1140
1141 //
1142 // Handle receiving a session structure in a heartbeat packet.
1143 //
1144 static int cluster_recv_session(int more, uint8_t *p)
1145 {
1146 if (more >= MAXSESSION) {
1147 LOG(0, 0, 0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
1148 return -1;
1149 }
1150
1151 if (session[more].tunnel == T_UNDEF) {
1152 if (config->cluster_iam_uptodate) { // Sanity.
1153 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined session!\n");
1154 } else {
1155 --config->cluster_undefined_sessions;
1156 }
1157 }
1158
1159 load_session(more, (sessiont *) p); // Copy session into session table..
1160
1161 LOG(5, more, 0, "Received session update (%d undef)\n", config->cluster_undefined_sessions);
1162
1163 if (!config->cluster_iam_uptodate)
1164 cluster_uptodate(); // Check to see if we're up to date.
1165
1166 return 0;
1167 }
1168
1169 static int cluster_recv_tunnel(int more, uint8_t *p)
1170 {
1171 if (more >= MAXTUNNEL) {
1172 LOG(0, 0, 0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
1173 return -1;
1174 }
1175
1176 if (tunnel[more].state == TUNNELUNDEF) {
1177 if (config->cluster_iam_uptodate) { // Sanity.
1178 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined tunnel!\n");
1179 } else {
1180 --config->cluster_undefined_tunnels;
1181 }
1182 }
1183
1184 memcpy(&tunnel[more], p, sizeof(tunnel[more]) );
1185
1186 //
1187 // Clear tunnel control messages. These are dynamically allocated.
1188 // If we get unlucky, this may cause the tunnel to drop!
1189 //
1190 tunnel[more].controls = tunnel[more].controle = NULL;
1191 tunnel[more].controlc = 0;
1192
1193 LOG(5, 0, more, "Received tunnel update\n");
1194
1195 if (!config->cluster_iam_uptodate)
1196 cluster_uptodate(); // Check to see if we're up to date.
1197
1198 return 0;
1199 }
1200
1201
1202 // pre v5 heartbeat session structure
1203 struct oldsession {
1204 sessionidt next;
1205 sessionidt far;
1206 tunnelidt tunnel;
1207 in_addr_t ip;
1208 int ip_pool_index;
1209 unsigned long unique_id;
1210 uint16_t nr;
1211 uint16_t ns;
1212 uint32_t magic;
1213 uint32_t cin, cout;
1214 uint32_t pin, pout;
1215 uint32_t total_cin;
1216 uint32_t total_cout;
1217 uint32_t id;
1218 uint16_t throttle_in;
1219 uint16_t throttle_out;
1220 clockt opened;
1221 clockt die;
1222 time_t last_packet;
1223 in_addr_t dns1, dns2;
1224 routet route[MAXROUTE];
1225 uint16_t radius;
1226 uint16_t mru;
1227 uint16_t tbf_in;
1228 uint16_t tbf_out;
1229 uint8_t l2tp_flags;
1230 uint8_t reserved_old_snoop;
1231 uint8_t walled_garden;
1232 uint8_t flags1;
1233 char random_vector[MAXTEL];
1234 int random_vector_length;
1235 char user[129];
1236 char called[MAXTEL];
1237 char calling[MAXTEL];
1238 uint32_t tx_connect_speed;
1239 uint32_t rx_connect_speed;
1240 uint32_t flags;
1241 #define SF_IPCP_ACKED 1 // Has this session seen an IPCP Ack?
1242 #define SF_LCP_ACKED 2 // LCP negotiated
1243 #define SF_CCP_ACKED 4 // CCP negotiated
1244 in_addr_t snoop_ip;
1245 uint16_t snoop_port;
1246 uint16_t sid;
1247 uint8_t filter_in;
1248 uint8_t filter_out;
1249 char reserved[18];
1250 };
1251
1252 static uint8_t *convert_session(struct oldsession *old)
1253 {
1254 static sessiont new;
1255 int i;
1256
1257 memset(&new, 0, sizeof(new));
1258
1259 new.next = old->next;
1260 new.far = old->far;
1261 new.tunnel = old->tunnel;
1262 new.l2tp_flags = old->l2tp_flags;
1263 new.ip = old->ip;
1264 new.ip_pool_index = old->ip_pool_index;
1265 new.unique_id = old->unique_id;
1266 new.nr = old->nr;
1267 new.ns = old->ns;
1268 new.magic = old->magic;
1269 new.pin = old->pin;
1270 new.pout = old->pout;
1271 new.cin = old->total_cin;
1272 new.cout = old->total_cout;
1273 new.cin_delta = old->cin;
1274 new.cout_delta = old->cout;
1275 new.throttle_in = old->throttle_in;
1276 new.throttle_out = old->throttle_out;
1277 new.filter_in = old->filter_in;
1278 new.filter_out = old->filter_out;
1279 new.mru = old->mru;
1280 new.opened = old->opened;
1281 new.die = old->die;
1282 new.last_packet = old->last_packet;
1283 new.dns1 = old->dns1;
1284 new.dns2 = old->dns2;
1285 new.tbf_in = old->tbf_in;
1286 new.tbf_out = old->tbf_out;
1287 new.random_vector_length = old->random_vector_length;
1288 new.tx_connect_speed = old->tx_connect_speed;
1289 new.rx_connect_speed = old->rx_connect_speed;
1290 new.snoop_ip = old->snoop_ip;
1291 new.snoop_port = old->snoop_port;
1292 new.walled_garden = old->walled_garden;
1293
1294 memcpy(new.random_vector, old->random_vector, sizeof(new.random_vector));
1295 memcpy(new.user, old->user, sizeof(new.user));
1296 memcpy(new.called, old->called, sizeof(new.called));
1297 memcpy(new.calling, old->calling, sizeof(new.calling));
1298
1299 for (i = 0; i < MAXROUTE; i++)
1300 memcpy(&new.route[i], &old->route[i], sizeof(new.route[i]));
1301
1302 if (new.opened)
1303 {
1304 new.ppp.phase = Establish;
1305 if (old->flags & (SF_IPCP_ACKED|SF_LCP_ACKED))
1306 {
1307 new.ppp.phase = Network;
1308 new.ppp.lcp = Opened;
1309 new.ppp.ipcp = (old->flags & SF_IPCP_ACKED) ? Opened : Starting;
1310 new.ppp.ccp = (old->flags & SF_CCP_ACKED) ? Opened : Stopped;
1311 }
1312
1313 // no PPPv6 in old session
1314 new.ppp.ipv6cp = Stopped;
1315 }
1316
1317 return (uint8_t *) &new;
1318 }
1319
1320 //
1321 // Process a heartbeat..
1322 //
1323 // v3: added interval, timeout
1324 // v4: added table_version
1325 // v5: added ipv6, re-ordered session structure
1326 static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t *p, in_addr_t addr)
1327 {
1328 heartt *h;
1329 int s = size - (p-data);
1330 int i, type;
1331 int hb_ver = more;
1332
1333 #if HB_VERSION != 5
1334 # error "need to update cluster_process_heartbeat()"
1335 #endif
1336
1337 // we handle versions 3 through 5
1338 if (hb_ver < 3 || hb_ver > HB_VERSION) {
1339 LOG(0, 0, 0, "Received a heartbeat version that I don't support (%d)!\n", hb_ver);
1340 return -1; // Ignore it??
1341 }
1342
1343 // Ok. It's a heartbeat packet from a cluster master!
1344 if (s < sizeof(*h))
1345 goto shortpacket;
1346
1347 h = (heartt *) p;
1348 p += sizeof(*h);
1349 s -= sizeof(*h);
1350
1351 if (h->clusterid != config->bind_address)
1352 return -1; // It's not part of our cluster.
1353
1354 if (config->cluster_iam_master) { // Sanity...
1355 // Note that this MUST match the election process above!
1356
1357 LOG(0, 0, 0, "I just got a heartbeat from master %s, but _I_ am the master!\n", fmtaddr(addr, 0));
1358 if (!h->basetime) {
1359 LOG(0, 0, 0, "Heartbeat with zero basetime! Ignoring\n");
1360 return -1; // Skip it.
1361 }
1362
1363 if (hb_ver >= 4) {
1364 if (h->table_version > config->cluster_table_version) {
1365 LOG(0, 0, 0, "They've seen more state changes (%" PRIu64 " vs my %" PRIu64 ") so I'm gone!\n",
1366 h->table_version, config->cluster_table_version);
1367
1368 kill(0, SIGTERM);
1369 exit(1);
1370 }
1371 if (h->table_version < config->cluster_table_version)
1372 return -1;
1373 }
1374
1375 if (basetime > h->basetime) {
1376 LOG(0, 0, 0, "They're an older master than me so I'm gone!\n");
1377 kill(0, SIGTERM);
1378 exit(1);
1379 }
1380
1381 if (basetime < h->basetime)
1382 return -1;
1383
1384 if (my_address < addr) { // Tie breaker.
1385 LOG(0, 0, 0, "They're a higher IP address than me, so I'm gone!\n");
1386 kill(0, SIGTERM);
1387 exit(1);
1388 }
1389
1390 //
1391 // Send it a unicast heartbeat to see give it a chance to die.
1392 // NOTE: It's actually safe to do seq-number - 1 without checking
1393 // for wrap around.
1394 //
1395 cluster_catchup_slave(config->cluster_seq_number - 1, addr);
1396
1397 return -1; // Skip it.
1398 }
1399
1400 //
1401 // Try and guard against a stray master appearing.
1402 //
1403 // Ignore heartbeats received from another master before the
1404 // timeout (less a smidgen) for the old master has elapsed.
1405 //
1406 // Note that after a clean failover, the cluster_master_address
1407 // is cleared, so this doesn't run.
1408 //
1409 if (config->cluster_master_address && addr != config->cluster_master_address) {
1410 LOG(0, 0, 0, "Ignoring stray heartbeat from %s, current master %s has not yet timed out (last heartbeat %.1f seconds ago).\n",
1411 fmtaddr(addr, 0), fmtaddr(config->cluster_master_address, 1),
1412 0.1 * (TIME - config->cluster_last_hb));
1413 return -1; // ignore
1414 }
1415
1416 if (config->cluster_seq_number == -1) // Don't have one. Just align to the master...
1417 config->cluster_seq_number = h->seq;
1418
1419 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1420 config->cluster_last_hb_ver = hb_ver; // remember what cluster version the master is using
1421
1422 if (config->cluster_seq_number != h->seq) { // Out of sequence heartbeat!
1423 static int lastseen_seq = 0;
1424 static time_t lastseen_time = 0;
1425
1426 // limit to once per second for a particular seq#
1427 int ask = (config->cluster_seq_number != lastseen_seq || time_now != lastseen_time);
1428
1429 LOG(1, 0, 0, "HB: Got seq# %d but was expecting %d. %s.\n",
1430 h->seq, config->cluster_seq_number,
1431 ask ? "Asking for resend" : "Ignoring");
1432
1433 if (ask)
1434 {
1435 lastseen_seq = config->cluster_seq_number;
1436 lastseen_time = time_now;
1437 peer_send_message(addr, C_LASTSEEN, config->cluster_seq_number, NULL, 0);
1438 }
1439
1440 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1441
1442 // Just drop the packet. The master will resend it as part of the catchup.
1443
1444 return 0;
1445 }
1446 // Save the packet in our buffer.
1447 // This is needed in case we become the master.
1448 config->cluster_seq_number = (h->seq+1)%HB_MAX_SEQ;
1449 i = h->seq % HB_HISTORY_SIZE;
1450 past_hearts[i].seq = h->seq;
1451 past_hearts[i].size = size;
1452 memcpy(&past_hearts[i].data, data, size); // Save it.
1453
1454
1455 // Check that we don't have too many undefined sessions, and
1456 // that the free session pointer is correct.
1457 cluster_check_sessions(h->highsession, h->freesession, h->hightunnel);
1458
1459 if (h->interval != config->cluster_hb_interval)
1460 {
1461 LOG(2, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
1462 h->interval, config->cluster_hb_interval);
1463
1464 config->cluster_hb_interval = h->interval;
1465 }
1466
1467 if (h->timeout != config->cluster_hb_timeout)
1468 {
1469 LOG(2, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
1470 h->timeout, config->cluster_hb_timeout);
1471
1472 config->cluster_hb_timeout = h->timeout;
1473 }
1474
1475 // Ok. process the packet...
1476 while ( s > 0) {
1477
1478 type = *((uint32_t *) p);
1479 p += sizeof(uint32_t);
1480 s -= sizeof(uint32_t);
1481
1482 more = *((uint32_t *) p);
1483 p += sizeof(uint32_t);
1484 s -= sizeof(uint32_t);
1485
1486 switch (type) {
1487 case C_CSESSION: { // Compressed session structure.
1488 uint8_t c[ sizeof(sessiont) + 2];
1489 int size;
1490 uint8_t *orig_p = p;
1491
1492 size = rle_decompress((uint8_t **) &p, s, c, sizeof(c) );
1493 s -= (p - orig_p);
1494
1495 // session struct changed with v5
1496 if (hb_ver < 5)
1497 {
1498 if (size != sizeof(struct oldsession)) {
1499 LOG(0, 0, 0, "DANGER: Received a v%d CSESSION that didn't decompress correctly!\n", hb_ver);
1500 // Now what? Should exit! No-longer up to date!
1501 break;
1502 }
1503 cluster_recv_session(more, convert_session((struct oldsession *) c));
1504 break;
1505 }
1506
1507 if (size != sizeof(sessiont) ) { // Ouch! Very very bad!
1508 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1509 // Now what? Should exit! No-longer up to date!
1510 break;
1511 }
1512
1513 cluster_recv_session(more, c);
1514 break;
1515 }
1516 case C_SESSION:
1517 if (hb_ver < 5)
1518 {
1519 if (s < sizeof(struct oldsession))
1520 goto shortpacket;
1521
1522 cluster_recv_session(more, convert_session((struct oldsession *) p));
1523
1524 p += sizeof(struct oldsession);
1525 s -= sizeof(struct oldsession);
1526 break;
1527 }
1528
1529 if ( s < sizeof(session[more]))
1530 goto shortpacket;
1531
1532 cluster_recv_session(more, p);
1533
1534 p += sizeof(session[more]);
1535 s -= sizeof(session[more]);
1536 break;
1537
1538 case C_CTUNNEL: { // Compressed tunnel structure.
1539 uint8_t c[ sizeof(tunnelt) + 2];
1540 int size;
1541 uint8_t *orig_p = p;
1542
1543 size = rle_decompress((uint8_t **) &p, s, c, sizeof(c));
1544 s -= (p - orig_p);
1545
1546 if (size != sizeof(tunnelt) ) { // Ouch! Very very bad!
1547 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1548 // Now what? Should exit! No-longer up to date!
1549 break;
1550 }
1551
1552 cluster_recv_tunnel(more, c);
1553 break;
1554
1555 }
1556 case C_TUNNEL:
1557 if ( s < sizeof(tunnel[more]))
1558 goto shortpacket;
1559
1560 cluster_recv_tunnel(more, p);
1561
1562 p += sizeof(tunnel[more]);
1563 s -= sizeof(tunnel[more]);
1564 break;
1565 default:
1566 LOG(0, 0, 0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type);
1567 return -1; // can't process any more of the packet!!
1568 }
1569 }
1570
1571 if (config->cluster_master_address != addr)
1572 {
1573 LOG(0, 0, 0, "My master just changed from %s to %s!\n",
1574 fmtaddr(config->cluster_master_address, 0), fmtaddr(addr, 1));
1575
1576 config->cluster_master_address = addr;
1577 }
1578
1579 config->cluster_last_hb = TIME; // Successfully received a heartbeat!
1580 config->cluster_table_version = h->table_version;
1581 return 0;
1582
1583 shortpacket:
1584 LOG(0, 0, 0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
1585 return -1;
1586 }
1587
1588 //
1589 // We got a packet on the cluster port!
1590 // Handle pings, lastseens, and heartbeats!
1591 //
1592 int processcluster(uint8_t *data, int size, in_addr_t addr)
1593 {
1594 int type, more;
1595 uint8_t *p = data;
1596 int s = size;
1597
1598 if (addr == my_address)
1599 return -1; // Ignore it. Something looped back the multicast!
1600
1601 LOG(5, 0, 0, "Process cluster: %d bytes from %s\n", size, fmtaddr(addr, 0));
1602
1603 if (s <= 0) // Any data there??
1604 return -1;
1605
1606 if (s < 8)
1607 goto shortpacket;
1608
1609 type = *((uint32_t *) p);
1610 p += sizeof(uint32_t);
1611 s -= sizeof(uint32_t);
1612
1613 more = *((uint32_t *) p);
1614 p += sizeof(uint32_t);
1615 s -= sizeof(uint32_t);
1616
1617 switch (type)
1618 {
1619 case C_PING: // Update the peers table.
1620 return cluster_add_peer(addr, more, (pingt *) p, s);
1621
1622 case C_MASTER: // Our master is wrong
1623 return cluster_set_master(addr, more);
1624
1625 case C_LASTSEEN: // Catch up a slave (slave missed a packet).
1626 return cluster_catchup_slave(more, addr);
1627
1628 case C_FORWARD: // Forwarded control packet. pass off to processudp.
1629 case C_FORWARD_DAE: // Forwarded DAE packet. pass off to processdae.
1630 if (!config->cluster_iam_master)
1631 {
1632 LOG(0, 0, 0, "I'm not the master, but I got a C_FORWARD_%s from %s?\n",
1633 type == C_FORWARD_DAE ? "_DAE" : "", fmtaddr(addr, 0));
1634
1635 return -1;
1636 }
1637 else
1638 {
1639 struct sockaddr_in a;
1640 a.sin_addr.s_addr = more;
1641
1642 a.sin_port = *(int *) p;
1643 s -= sizeof(int);
1644 p += sizeof(int);
1645
1646 LOG(4, 0, 0, "Got a forwarded %spacket... (%s:%d)\n",
1647 type == C_FORWARD_DAE ? "DAE " : "", fmtaddr(more, 0), a.sin_port);
1648
1649 STAT(recv_forward);
1650 if (type == C_FORWARD_DAE)
1651 processdae(p, s, &a, sizeof(a));
1652 else
1653 processudp(p, s, &a);
1654
1655 return 0;
1656 }
1657
1658 case C_THROTTLE: { // Receive a forwarded packet from a slave.
1659 if (!config->cluster_iam_master) {
1660 LOG(0, 0, 0, "I'm not the master, but I got a C_THROTTLE from %s?\n", fmtaddr(addr, 0));
1661 return -1;
1662 }
1663
1664 tbf_queue_packet(more, p, s); // The TBF id tells wether it goes in or out.
1665 return 0;
1666 }
1667 case C_GARDEN:
1668 // Receive a walled garden packet from a slave.
1669 if (!config->cluster_iam_master) {
1670 LOG(0, 0, 0, "I'm not the master, but I got a C_GARDEN from %s?\n", fmtaddr(addr, 0));
1671 return -1;
1672 }
1673
1674 tun_write(p, s);
1675 return 0;
1676
1677 case C_BYTES:
1678 if (!config->cluster_iam_master) {
1679 LOG(0, 0, 0, "I'm not the master, but I got a C_BYTES from %s?\n", fmtaddr(addr, 0));
1680 return -1;
1681 }
1682
1683 return cluster_handle_bytes(p, s);
1684
1685 case C_KILL: // The master asked us to die!? (usually because we're too out of date).
1686 if (config->cluster_iam_master) {
1687 LOG(0, 0, 0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", fmtaddr(addr, 0), more);
1688 return -1;
1689 }
1690 if (more != config->cluster_seq_number) {
1691 LOG(0, 0, 0, "The master asked us to die but the seq number didn't match!?\n");
1692 return -1;
1693 }
1694
1695 if (addr != config->cluster_master_address) {
1696 LOG(0, 0, 0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%s)\n",
1697 fmtaddr(addr, 0), fmtaddr(config->cluster_master_address, 1));
1698 // We can only warn about it. The master might really have switched!
1699 }
1700
1701 LOG(0, 0, 0, "Received a valid C_KILL: I'm going to die now.\n");
1702 kill(0, SIGTERM);
1703 exit(0); // Lets be paranoid;
1704 return -1; // Just signalling the compiler.
1705
1706 case C_HEARTBEAT:
1707 LOG(4, 0, 0, "Got a heartbeat from %s\n", fmtaddr(addr, 0));
1708 return cluster_process_heartbeat(data, size, more, p, addr);
1709
1710 default:
1711 LOG(0, 0, 0, "Strange type packet received on cluster socket (%d)\n", type);
1712 return -1;
1713 }
1714 return 0;
1715
1716 shortpacket:
1717 LOG(0, 0, 0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n");
1718 return -1;
1719 }
1720
1721 //====================================================================================================
1722
1723 int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
1724 {
1725 int i;
1726
1727 if (CLI_HELP_REQUESTED)
1728 return CLI_HELP_NO_ARGS;
1729
1730 cli_print(cli, "Cluster status : %s", config->cluster_iam_master ? "Master" : "Slave" );
1731 cli_print(cli, "My address : %s", fmtaddr(my_address, 0));
1732 cli_print(cli, "VIP address : %s", fmtaddr(config->bind_address, 0));
1733 cli_print(cli, "Multicast address: %s", fmtaddr(config->cluster_address, 0));
1734 cli_print(cli, "Multicast i'face : %s", config->cluster_interface);
1735
1736 if (!config->cluster_iam_master) {
1737 cli_print(cli, "My master : %s (last heartbeat %.1f seconds old)",
1738 config->cluster_master_address
1739 ? fmtaddr(config->cluster_master_address, 0)
1740 : "Not defined",
1741 0.1 * (TIME - config->cluster_last_hb));
1742 cli_print(cli, "Uptodate : %s", config->cluster_iam_uptodate ? "Yes" : "No");
1743 cli_print(cli, "Table version # : %" PRIu64, config->cluster_table_version);
1744 cli_print(cli, "Next sequence number expected: %d", config->cluster_seq_number);
1745 cli_print(cli, "%d sessions undefined of %d", config->cluster_undefined_sessions, config->cluster_highest_sessionid);
1746 cli_print(cli, "%d tunnels undefined of %d", config->cluster_undefined_tunnels, config->cluster_highest_tunnelid);
1747 } else {
1748 cli_print(cli, "Table version # : %" PRIu64, config->cluster_table_version);
1749 cli_print(cli, "Next heartbeat # : %d", config->cluster_seq_number);
1750 cli_print(cli, "Highest session : %d", config->cluster_highest_sessionid);
1751 cli_print(cli, "Highest tunnel : %d", config->cluster_highest_tunnelid);
1752 cli_print(cli, "%d changes queued for sending", config->cluster_num_changes);
1753 }
1754 cli_print(cli, "%d peers.", num_peers);
1755
1756 if (num_peers)
1757 cli_print(cli, "%20s %10s %8s", "Address", "Basetime", "Age");
1758 for (i = 0; i < num_peers; ++i) {
1759 cli_print(cli, "%20s %10u %8d", fmtaddr(peers[i].peer, 0),
1760 peers[i].basetime, TIME - peers[i].timestamp);
1761 }
1762 return CLI_OK;
1763 }
1764
1765 //
1766 // Simple run-length-encoding compression.
1767 // Format is
1768 // 1 byte < 128 = count of non-zero bytes following. // Not legal to be zero.
1769 // n non-zero bytes;
1770 // or
1771 // 1 byte > 128 = (count - 128) run of zero bytes. //
1772 // repeat.
1773 // count == 0 indicates end of compressed stream.
1774 //
1775 // Compress from 'src' into 'dst'. return number of bytes
1776 // used from 'dst'.
1777 // Updates *src_p to indicate 1 past last bytes used.
1778 //
1779 // We could get an extra byte in the zero runs by storing (count-1)
1780 // but I'm playing it safe.
1781 //
1782 // Worst case is a 50% expansion in space required (trying to
1783 // compress { 0x00, 0x01 } * N )
1784 static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
1785 {
1786 int count;
1787 int orig_dsize = dsize;
1788 uint8_t *x, *src;
1789 src = *src_p;
1790
1791 while (ssize > 0 && dsize > 2) {
1792 count = 0;
1793 x = dst++; --dsize; // Reserve space for count byte..
1794
1795 if (*src) { // Copy a run of non-zero bytes.
1796 while (*src && count < 127 && ssize > 0 && dsize > 1) { // Count number of non-zero bytes.
1797 *dst++ = *src++;
1798 --dsize; --ssize;
1799 ++count;
1800 }
1801 *x = count; // Store number of non-zero bytes. Guarenteed to be non-zero!
1802
1803 } else { // Compress a run of zero bytes.
1804 while (*src == 0 && count < 127 && ssize > 0) {
1805 ++src;
1806 --ssize;
1807 ++count;
1808 }
1809 *x = count | 0x80 ;
1810 }
1811 }
1812
1813 *dst++ = 0x0; // Add Stop byte.
1814 --dsize;
1815
1816 *src_p = src;
1817 return (orig_dsize - dsize);
1818 }
1819
1820 //
1821 // Decompress the buffer into **p.
1822 // 'psize' is the size of the decompression buffer available.
1823 //
1824 // Returns the number of bytes decompressed.
1825 //
1826 // Decompresses from '*src_p' into 'dst'.
1827 // Return the number of dst bytes used.
1828 // Updates the 'src_p' pointer to point to the
1829 // first un-used byte.
1830 static int rle_decompress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
1831 {
1832 int count;
1833 int orig_dsize = dsize;
1834 uint8_t *src = *src_p;
1835
1836 while (ssize >0 && dsize > 0) { // While there's more to decompress, and there's room in the decompress buffer...
1837 count = *src++; --ssize; // get the count byte from the source.
1838 if (count == 0x0) // End marker reached? If so, finish.
1839 break;
1840
1841 if (count & 0x80) { // Decompress a run of zeros
1842 for (count &= 0x7f ; count > 0 && dsize > 0; --count) {
1843 *dst++ = 0x0;
1844 --dsize;
1845 }
1846 } else { // Copy run of non-zero bytes.
1847 for ( ; count > 0 && ssize && dsize; --count) { // Copy non-zero bytes across.
1848 *dst++ = *src++;
1849 --ssize; --dsize;
1850 }
1851 }
1852 }
1853 *src_p = src;
1854 return (orig_dsize - dsize);
1855 }