a9e820976ad1bab4fe6876c698614b6185f2b343
[l2tpns.git] / cluster.c
1 // L2TPNS Clustering Stuff
2
3 char const *cvs_id_cluster = "$Id: cluster.c,v 1.39 2005-05-26 12:17:30 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 int cluster_sockfd = 0; // 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 char 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(char **p, int type, int more, char *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, char *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, char *data, int size)
263 {
264 char buf[65536]; // Vast overkill.
265 char *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 //
274 // Forward a state changing packet to the master.
275 //
276 // The master just processes the payload as if it had
277 // received it off the tun device.
278 //
279 int master_forward_packet(char *data, int size, in_addr_t addr, int port)
280 {
281 char buf[65536]; // Vast overkill.
282 char *p = buf;
283
284 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
285 return -1;
286
287 LOG(4, 0, 0, "Forwarding packet from %s to master (size %d)\n", fmtaddr(addr, 0), size);
288
289 STAT(c_forwarded);
290 add_type(&p, C_FORWARD, addr, (char *) &port, sizeof(port));
291 memcpy(p, data, size);
292 p += size;
293
294 return peer_send_data(config->cluster_master_address, buf, (p - buf));
295 }
296
297 //
298 // Forward a throttled packet to the master for handling.
299 //
300 // The master just drops the packet into the appropriate
301 // token bucket queue, and lets normal processing take care
302 // of it.
303 //
304 int master_throttle_packet(int tbfid, char *data, int size)
305 {
306 char buf[65536]; // Vast overkill.
307 char *p = buf;
308
309 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
310 return -1;
311
312 LOG(4, 0, 0, "Throttling packet master (size %d, tbfid %d)\n", size, tbfid);
313
314 add_type(&p, C_THROTTLE, tbfid, data, size);
315
316 return peer_send_data(config->cluster_master_address, buf, (p-buf) );
317
318 }
319
320 //
321 // Forward a walled garden packet to the master for handling.
322 //
323 // The master just writes the packet straight to the tun
324 // device (where is will normally loop through the
325 // firewall rules, and come back in on the tun device)
326 //
327 // (Note that this must be called with the tun header
328 // as the start of the data).
329 int master_garden_packet(sessionidt s, char *data, int size)
330 {
331 char buf[65536]; // Vast overkill.
332 char *p = buf;
333
334 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
335 return -1;
336
337 LOG(4, 0, 0, "Walled garden packet to master (size %d)\n", size);
338
339 add_type(&p, C_GARDEN, s, data, size);
340
341 return peer_send_data(config->cluster_master_address, buf, (p-buf));
342
343 }
344
345 //
346 // Send a chunk of data as a heartbeat..
347 // We save it in the history buffer as we do so.
348 //
349 static void send_heartbeat(int seq, char *data, int size)
350 {
351 int i;
352
353 if (size > sizeof(past_hearts[0].data))
354 {
355 LOG(0, 0, 0, "Tried to heartbeat something larger than the maximum packet!\n");
356 kill(0, SIGTERM);
357 exit(1);
358 }
359 i = seq % HB_HISTORY_SIZE;
360 past_hearts[i].seq = seq;
361 past_hearts[i].size = size;
362 memcpy(&past_hearts[i].data, data, size); // Save it.
363 cluster_send_data(data, size);
364 }
365
366 //
367 // Send an 'i am alive' message to every machine in the cluster.
368 //
369 void cluster_send_ping(time_t basetime)
370 {
371 char buff[100 + sizeof(pingt)];
372 char *p = buff;
373 pingt x;
374
375 if (config->cluster_iam_master && basetime) // We're heartbeating so no need to ping.
376 return;
377
378 LOG(5, 0, 0, "Sending cluster ping...\n");
379
380 x.ver = 1;
381 x.addr = config->bind_address;
382 x.undef = config->cluster_undefined_sessions + config->cluster_undefined_tunnels;
383 x.basetime = basetime;
384
385 add_type(&p, C_PING, basetime, (char *) &x, sizeof(x));
386 cluster_send_data(buff, (p-buff) );
387 }
388
389 //
390 // Walk the session counters looking for non-zero ones to send
391 // to the master. We send up to 600 of them at one time.
392 // We examine a maximum of 3000 sessions.
393 // (50k max session should mean that we normally
394 // examine the entire session table every 25 seconds).
395
396 #define MAX_B_RECS (600)
397 void master_update_counts(void)
398 {
399 int i, c;
400 bytest b[MAX_B_RECS+1];
401
402 if (config->cluster_iam_master) // Only happens on the slaves.
403 return;
404
405 if (!config->cluster_master_address) // If we don't have a master, skip it for a while.
406 return;
407
408 i = MAX_B_RECS * 5; // Examine max 2000 sessions;
409 if (config->cluster_highest_sessionid > i)
410 i = config->cluster_highest_sessionid;
411
412 for ( c = 0; i > 0 ; --i) {
413 // Next session to look at.
414 walk_session_number++;
415 if ( walk_session_number > config->cluster_highest_sessionid)
416 walk_session_number = 1;
417
418 if (!sess_local[walk_session_number].cin && !sess_local[walk_session_number].cout)
419 continue; // Unused. Skip it.
420
421 b[c].sid = walk_session_number;
422 b[c].in = sess_local[walk_session_number].cin;
423 b[c].out = sess_local[walk_session_number].cout;
424
425 if (++c > MAX_B_RECS) // Send a max of 400 elements in a packet.
426 break;
427
428 // Reset counters.
429 sess_local[walk_session_number].cin = sess_local[walk_session_number].cout = 0;
430 }
431
432 if (!c) // Didn't find any that changes. Get out of here!
433 return;
434
435
436 // Forward the data to the master.
437 LOG(4, 0, 0, "Sending byte counters to master (%d elements)\n", c);
438 peer_send_message(config->cluster_master_address, C_BYTES, c, (char *) &b, sizeof(b[0]) * c);
439 return;
440 }
441
442 //
443 // On the master, check how our slaves are going. If
444 // one of them's not up-to-date we'll heartbeat faster.
445 // If we don't have any of them, then we need to turn
446 // on our own packet handling!
447 //
448 void cluster_check_slaves(void)
449 {
450 int i;
451 static int have_peers = 0;
452 int had_peers = have_peers;
453 clockt t = TIME;
454
455 if (!config->cluster_iam_master)
456 return; // Only runs on the master...
457
458 config->cluster_iam_uptodate = 1; // cleared in loop below
459
460 for (i = have_peers = 0; i < num_peers; i++)
461 {
462 if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
463 continue; // Stale peer! Skip them.
464
465 if (!peers[i].basetime)
466 continue; // Shutdown peer! Skip them.
467
468 if (peers[i].uptodate)
469 have_peers++;
470 else
471 config->cluster_iam_uptodate = 0; // Start fast heartbeats
472 }
473
474 // in a cluster, withdraw/add routes when we get a peer/lose peers
475 if (have_peers != had_peers)
476 {
477 if (had_peers < config->cluster_master_min_adv &&
478 have_peers >= config->cluster_master_min_adv)
479 withdraw_routes();
480
481 else if (had_peers >= config->cluster_master_min_adv &&
482 have_peers < config->cluster_master_min_adv)
483 advertise_routes();
484 }
485 }
486
487 //
488 // Check that we have a master. If it's been too
489 // long since we heard from a master then hold an election.
490 //
491 void cluster_check_master(void)
492 {
493 int i, count, tcount, high_unique_id = 0;
494 int last_free = 0;
495 clockt t = TIME;
496 static int probed = 0;
497 int have_peers;
498
499 if (config->cluster_iam_master)
500 return; // Only runs on the slaves...
501
502 // If the master is late (missed 2 hearbeats by a second and a
503 // hair) it may be that the switch has dropped us from the
504 // multicast group, try unicasting probes to the master
505 // which will hopefully respond with a unicast heartbeat that
506 // will allow us to limp along until the querier next runs.
507 if (config->cluster_master_address
508 && TIME > (config->cluster_last_hb + 2 * config->cluster_hb_interval + 11))
509 {
510 if (!probed || (TIME > (probed + 2 * config->cluster_hb_interval)))
511 {
512 probed = TIME;
513 LOG(1, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
514 0.1 * (TIME - (config->cluster_last_hb + config->cluster_hb_interval)));
515
516 peer_send_message(config->cluster_master_address,
517 C_LASTSEEN, config->cluster_seq_number, NULL, 0);
518 }
519 } else { // We got a recent heartbeat; reset the probe flag.
520 probed = 0;
521 }
522
523 if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout))
524 return; // Everything's ok!
525
526 config->cluster_last_hb = TIME + 1; // Just the one election thanks.
527
528 LOG(0, 0, 0, "Master timed out! Holding election...\n");
529
530 for (i = have_peers = 0; i < num_peers; i++)
531 {
532 if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
533 continue; // Stale peer! Skip them.
534
535 if (!peers[i].basetime)
536 continue; // Shutdown peer! Skip them.
537
538 if (peers[i].basetime < basetime) {
539 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
540 return; // They'll win the election. Get out of here.
541 }
542
543 if (peers[i].basetime == basetime &&
544 peers[i].peer > my_address) {
545 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
546 return; // They'll win the election. Wait for them to come up.
547 }
548
549 if (peers[i].uptodate)
550 have_peers++;
551 }
552
553 // Wow. it's been ages since I last heard a heartbeat
554 // and I'm better than an of my peers so it's time
555 // to become a master!!!
556
557 config->cluster_iam_master = 1;
558 config->cluster_master_address = 0;
559
560 LOG(0, 0, 0, "I am declaring myself the master!\n");
561
562 if (have_peers < config->cluster_master_min_adv)
563 advertise_routes();
564 else
565 withdraw_routes();
566
567 if (config->cluster_seq_number == -1)
568 config->cluster_seq_number = 0;
569
570 //
571 // Go through and mark all the tunnels as defined.
572 // Count the highest used tunnel number as well.
573 //
574 config->cluster_highest_tunnelid = 0;
575 for (i = 0, tcount = 0; i < MAXTUNNEL; ++i) {
576 if (tunnel[i].state == TUNNELUNDEF)
577 tunnel[i].state = TUNNELFREE;
578
579 if (tunnel[i].state != TUNNELFREE && i > config->cluster_highest_tunnelid)
580 config->cluster_highest_tunnelid = i;
581 }
582
583 //
584 // Go through and mark all the sessions as being defined.
585 // reset the idle timeouts.
586 // add temporary byte counters to permanent ones.
587 // Re-string the free list.
588 // Find the ID of the highest session.
589 last_free = 0;
590 high_unique_id = 0;
591 config->cluster_highest_sessionid = 0;
592 for (i = 0, count = 0; i < MAXSESSION; ++i) {
593 if (session[i].tunnel == T_UNDEF) {
594 session[i].tunnel = T_FREE;
595 ++count;
596 }
597
598 if (!session[i].opened) { // Unused session. Add to free list.
599 memset(&session[i], 0, sizeof(session[i]));
600 session[i].tunnel = T_FREE;
601 session[last_free].next = i;
602 session[i].next = 0;
603 last_free = i;
604 continue;
605 }
606
607 // Reset idle timeouts..
608 session[i].last_packet = time_now;
609
610 // Reset die relative to our uptime rather than the old master's
611 if (session[i].die) session[i].die = TIME;
612
613 // Accumulate un-sent byte counters.
614 session[i].cin += sess_local[i].cin;
615 session[i].cout += sess_local[i].cout;
616 session[i].total_cin += sess_local[i].cin;
617 session[i].total_cout += sess_local[i].cout;
618
619 sess_local[i].cin = sess_local[i].cout = 0;
620
621 sess_local[i].radius = 0; // Reset authentication as the radius blocks aren't up to date.
622
623 if (session[i].unique_id >= high_unique_id) // This is different to the index into the session table!!!
624 high_unique_id = session[i].unique_id+1;
625
626 session[i].tbf_in = session[i].tbf_out = 0; // Remove stale pointers from old master.
627 throttle_session(i, session[i].throttle_in, session[i].throttle_out);
628
629 config->cluster_highest_sessionid = i;
630 }
631
632 session[last_free].next = 0; // End of chain.
633 last_id = high_unique_id; // Keep track of the highest used session ID.
634
635 become_master();
636
637 rebuild_address_pool();
638
639 // If we're not the very first master, this is a big issue!
640 if(count>0)
641 LOG(0, 0, 0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count);
642
643 config->cluster_undefined_sessions = 0;
644 config->cluster_undefined_tunnels = 0;
645 config->cluster_iam_uptodate = 1; // assume all peers are up-to-date
646
647 // FIXME. We need to fix up the tunnel control message
648 // queue here! There's a number of other variables we
649 // should also update.
650 }
651
652
653 //
654 // Check that our session table is validly matching what the
655 // master has in mind.
656 //
657 // In particular, if we have too many sessions marked 'undefined'
658 // we fix it up here, and we ensure that the 'first free session'
659 // pointer is valid.
660 //
661 static void cluster_check_sessions(int highsession, int freesession_ptr, int hightunnel)
662 {
663 int i;
664
665 sessionfree = freesession_ptr; // Keep the freesession ptr valid.
666
667 if (config->cluster_iam_uptodate)
668 return;
669
670 if (highsession > config->cluster_undefined_sessions && hightunnel > config->cluster_undefined_tunnels)
671 return;
672
673 // Clear out defined sessions, counting the number of
674 // undefs remaining.
675 config->cluster_undefined_sessions = 0;
676 for (i = 1 ; i < MAXSESSION; ++i) {
677 if (i > highsession) {
678 if (session[i].tunnel == T_UNDEF) session[i].tunnel = T_FREE; // Defined.
679 continue;
680 }
681
682 if (session[i].tunnel == T_UNDEF)
683 ++config->cluster_undefined_sessions;
684 }
685
686 // Clear out defined tunnels, counting the number of
687 // undefs remaining.
688 config->cluster_undefined_tunnels = 0;
689 for (i = 1 ; i < MAXTUNNEL; ++i) {
690 if (i > hightunnel) {
691 if (tunnel[i].state == TUNNELUNDEF) tunnel[i].state = TUNNELFREE; // Defined.
692 continue;
693 }
694
695 if (tunnel[i].state == TUNNELUNDEF)
696 ++config->cluster_undefined_tunnels;
697 }
698
699
700 if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels) {
701 LOG(2, 0, 0, "Cleared undefined sessions/tunnels. %d sess (high %d), %d tunn (high %d)\n",
702 config->cluster_undefined_sessions, highsession, config->cluster_undefined_tunnels, hightunnel);
703 return;
704 }
705
706 // Are we up to date?
707
708 if (!config->cluster_iam_uptodate)
709 cluster_uptodate();
710 }
711
712 static int hb_add_type(char **p, int type, int id)
713 {
714 switch (type) {
715 case C_CSESSION: { // Compressed C_SESSION.
716 uint8_t c[sizeof(sessiont) * 2]; // Bigger than worst case.
717 uint8_t *d = (uint8_t *) &session[id];
718 uint8_t *orig = d;
719 int size;
720
721 size = rle_compress( &d, sizeof(sessiont), c, sizeof(c) );
722
723 // Did we compress the full structure, and is the size actually
724 // reduced??
725 if ( (d - orig) == sizeof(sessiont) && size < sizeof(sessiont) ) {
726 add_type(p, C_CSESSION, id, (char *) c, size);
727 break;
728 }
729 // Failed to compress : Fall through.
730 }
731 case C_SESSION: add_type(p, C_SESSION, id,
732 (char *) &session[id], sizeof(sessiont));
733 break;
734
735 case C_CTUNNEL: { // Compressed C_TUNNEL
736 uint8_t c[sizeof(tunnelt) * 2]; // Bigger than worst case.
737 uint8_t *d = (uint8_t *) &tunnel[id];
738 uint8_t *orig = d;
739 int size;
740
741 size = rle_compress( &d, sizeof(tunnelt), c, sizeof(c) );
742
743 // Did we compress the full structure, and is the size actually
744 // reduced??
745 if ( (d - orig) == sizeof(tunnelt) && size < sizeof(tunnelt) ) {
746 add_type(p, C_CTUNNEL, id, c, size);
747 break;
748 }
749 // Failed to compress : Fall through.
750 }
751 case C_TUNNEL: add_type(p, C_TUNNEL, id,
752 (char *) &tunnel[id], sizeof(tunnelt));
753 break;
754 default:
755 LOG(0, 0, 0, "Found an invalid type in heart queue! (%d)\n", type);
756 kill(0, SIGTERM);
757 exit(1);
758 }
759 return 0;
760 }
761
762 //
763 // Send a heartbeat, incidently sending out any queued changes..
764 //
765 void cluster_heartbeat()
766 {
767 int i, count = 0, tcount = 0;
768 char buff[MAX_HEART_SIZE + sizeof(heartt) + sizeof(int) ];
769 heartt h;
770 char *p = buff;
771
772 if (!config->cluster_iam_master) // Only the master does this.
773 return;
774
775 config->cluster_table_version += config->cluster_num_changes;
776
777 // Fill out the heartbeat header.
778 memset(&h, 0, sizeof(h));
779
780 h.version = HB_VERSION;
781 h.seq = config->cluster_seq_number;
782 h.basetime = basetime;
783 h.clusterid = config->bind_address; // Will this do??
784 h.basetime = basetime;
785 h.highsession = config->cluster_highest_sessionid;
786 h.freesession = sessionfree;
787 h.hightunnel = config->cluster_highest_tunnelid;
788 h.size_sess = sizeof(sessiont); // Just in case.
789 h.size_tunn = sizeof(tunnelt);
790 h.interval = config->cluster_hb_interval;
791 h.timeout = config->cluster_hb_timeout;
792 h.table_version = config->cluster_table_version;
793
794 add_type(&p, C_HEARTBEAT, HB_VERSION, (char *) &h, sizeof(h));
795
796 for (i = 0; i < config->cluster_num_changes; ++i) {
797 hb_add_type(&p, cluster_changes[i].type, cluster_changes[i].id);
798 }
799
800 if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
801 LOG(0, 0, 0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", (int) (p - buff));
802 kill(0, SIGTERM);
803 exit(1);
804 }
805
806 //
807 // Fill out the packet with sessions from the session table...
808 // (not forgetting to leave space so we can get some tunnels in too )
809 while ( (p + sizeof(uint32_t) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) {
810
811 if (!walk_session_number) // session #0 isn't valid.
812 ++walk_session_number;
813
814 if (count >= config->cluster_highest_sessionid) // If we're a small cluster, don't go wild.
815 break;
816
817 hb_add_type(&p, C_CSESSION, walk_session_number);
818 walk_session_number = (1+walk_session_number)%(config->cluster_highest_sessionid+1); // +1 avoids divide by zero.
819
820 ++count; // Count the number of extra sessions we're sending.
821 }
822
823 //
824 // Fill out the packet with tunnels from the tunnel table...
825 // This effectively means we walk the tunnel table more quickly
826 // than the session table. This is good because stuffing up a
827 // tunnel is a much bigger deal than stuffing up a session.
828 //
829 while ( (p + sizeof(uint32_t) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {
830
831 if (!walk_tunnel_number) // tunnel #0 isn't valid.
832 ++walk_tunnel_number;
833
834 if (tcount >= config->cluster_highest_tunnelid)
835 break;
836
837 hb_add_type(&p, C_CTUNNEL, walk_tunnel_number);
838 walk_tunnel_number = (1+walk_tunnel_number)%(config->cluster_highest_tunnelid+1); // +1 avoids divide by zero.
839
840 ++tcount;
841 }
842
843 //
844 // Did we do something wrong?
845 if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
846 LOG(0, 0, 0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", (int) (p - buff));
847 kill(0, SIGTERM);
848 exit(1);
849 }
850
851 LOG(3, 0, 0, "Sending v%d heartbeat #%d, change #%" PRIu64 " with %d changes "
852 "(%d x-sess, %d x-tunnels, %d highsess, %d hightun, size %d)\n",
853 HB_VERSION, h.seq, h.table_version, config->cluster_num_changes,
854 count, tcount, config->cluster_highest_sessionid,
855 config->cluster_highest_tunnelid, (int) (p - buff));
856
857 config->cluster_num_changes = 0;
858
859 send_heartbeat(h.seq, buff, (p-buff) ); // Send out the heartbeat to the cluster, keeping a copy of it.
860
861 config->cluster_seq_number = (config->cluster_seq_number+1)%HB_MAX_SEQ; // Next seq number to use.
862 }
863
864 //
865 // A structure of type 'type' has changed; Add it to the queue to send.
866 //
867 static int type_changed(int type, int id)
868 {
869 int i;
870
871 for (i = 0 ; i < config->cluster_num_changes ; ++i)
872 if ( cluster_changes[i].id == id &&
873 cluster_changes[i].type == type)
874 return 0; // Already marked for change.
875
876 cluster_changes[i].type = type;
877 cluster_changes[i].id = id;
878 ++config->cluster_num_changes;
879
880 if (config->cluster_num_changes > MAX_CHANGES)
881 cluster_heartbeat(); // flush now
882
883 return 1;
884 }
885
886
887 // A particular session has been changed!
888 int cluster_send_session(int sid)
889 {
890 if (!config->cluster_iam_master) {
891 LOG(0, sid, 0, "I'm not a master, but I just tried to change a session!\n");
892 return -1;
893 }
894
895 if (forked) {
896 LOG(0, sid, 0, "cluster_send_session called from child process!\n");
897 return -1;
898 }
899
900 return type_changed(C_CSESSION, sid);
901 }
902
903 // A particular tunnel has been changed!
904 int cluster_send_tunnel(int tid)
905 {
906 if (!config->cluster_iam_master) {
907 LOG(0, 0, tid, "I'm not a master, but I just tried to change a tunnel!\n");
908 return -1;
909 }
910
911 return type_changed(C_CTUNNEL, tid);
912 }
913
914
915 //
916 // We're a master, and a slave has just told us that it's
917 // missed a packet. We'll resend it every packet since
918 // the last one it's seen.
919 //
920 static int cluster_catchup_slave(int seq, in_addr_t slave)
921 {
922 int s;
923 int diff;
924
925 LOG(1, 0, 0, "Slave %s sent LASTSEEN with seq %d\n", fmtaddr(slave, 0), seq);
926 if (!config->cluster_iam_master) {
927 LOG(1, 0, 0, "Got LASTSEEN but I'm not a master! Redirecting it to %s.\n",
928 fmtaddr(config->cluster_master_address, 0));
929
930 peer_send_message(slave, C_MASTER, config->cluster_master_address, NULL, 0);
931 return 0;
932 }
933
934 diff = config->cluster_seq_number - seq; // How many packet do we need to send?
935 if (diff < 0)
936 diff += HB_MAX_SEQ;
937
938 if (diff >= HB_HISTORY_SIZE) { // Ouch. We don't have the packet to send it!
939 LOG(0, 0, 0, "A slave asked for message %d when our seq number is %d. Killing it.\n",
940 seq, config->cluster_seq_number);
941 return peer_send_message(slave, C_KILL, seq, NULL, 0);// Kill the slave. Nothing else to do.
942 }
943
944 LOG(1, 0, 0, "Sending %d catchup packets to slave %s\n", diff, fmtaddr(slave, 0) );
945
946 // Now resend every packet that it missed, in order.
947 while (seq != config->cluster_seq_number) {
948 s = seq % HB_HISTORY_SIZE;
949 if (seq != past_hearts[s].seq) {
950 LOG(0, 0, 0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n",
951 fmtaddr(slave, 0), seq, past_hearts[s].seq, s, config->cluster_seq_number);
952 return -1; // What to do here!?
953 }
954 peer_send_data(slave, past_hearts[s].data, past_hearts[s].size);
955 seq = (seq+1)%HB_MAX_SEQ; // Increment to next seq number.
956 }
957 return 0; // All good!
958 }
959
960 //
961 // We've heard from another peer! Add it to the list
962 // that we select from at election time.
963 //
964 static int cluster_add_peer(in_addr_t peer, time_t basetime, pingt *pp, int size)
965 {
966 int i;
967 in_addr_t clusterid;
968 pingt p;
969
970 // Allow for backward compatability.
971 // Just the ping packet into a new structure to allow
972 // for the possibility that we might have received
973 // more or fewer elements than we were expecting.
974 if (size > sizeof(p))
975 size = sizeof(p);
976
977 memset( (void *) &p, 0, sizeof(p) );
978 memcpy( (void *) &p, (void *) pp, size);
979
980 clusterid = p.addr;
981 if (clusterid != config->bind_address)
982 {
983 // Is this for us?
984 LOG(4, 0, 0, "Skipping ping from %s (different cluster)\n", fmtaddr(peer, 0));
985 return 0;
986 }
987
988 for (i = 0; i < num_peers ; ++i)
989 {
990 if (peers[i].peer != peer)
991 continue;
992
993 // This peer already exists. Just update the timestamp.
994 peers[i].basetime = basetime;
995 peers[i].timestamp = TIME;
996 peers[i].uptodate = !p.undef;
997 break;
998 }
999
1000 // Is this the master shutting down??
1001 if (peer == config->cluster_master_address) {
1002 LOG(3, 0, 0, "Master %s %s\n", fmtaddr(config->cluster_master_address, 0),
1003 basetime ? "has restarted!" : "shutting down...");
1004
1005 config->cluster_master_address = 0;
1006 config->cluster_last_hb = 0; // Force an election.
1007 cluster_check_master();
1008 return 0;
1009 }
1010
1011 if (i >= num_peers)
1012 {
1013 LOG(4, 0, 0, "Adding %s as a peer\n", fmtaddr(peer, 0));
1014
1015 // Not found. Is there a stale slot to re-use?
1016 for (i = 0; i < num_peers ; ++i)
1017 {
1018 if (!peers[i].basetime) // Shutdown
1019 break;
1020
1021 if ((peers[i].timestamp + config->cluster_hb_timeout * 10) < TIME) // Stale.
1022 break;
1023 }
1024
1025 if (i >= CLUSTER_MAX_SIZE)
1026 {
1027 // Too many peers!!
1028 LOG(0, 0, 0, "Tried to add %s as a peer, but I already have %d of them!\n", fmtaddr(peer, 0), i);
1029 return -1;
1030 }
1031
1032 peers[i].peer = peer;
1033 peers[i].basetime = basetime;
1034 peers[i].timestamp = TIME;
1035 peers[i].uptodate = !p.undef;
1036 if (i == num_peers)
1037 ++num_peers;
1038
1039 LOG(1, 0, 0, "Added %s as a new peer. Now %d peers\n", fmtaddr(peer, 0), num_peers);
1040 }
1041
1042 return 1;
1043 }
1044
1045 // A slave responds with C_MASTER when it gets a message which should have gone to a master.
1046 static int cluster_set_master(in_addr_t peer, in_addr_t master)
1047 {
1048 if (config->cluster_iam_master) // Sanity...
1049 return 0;
1050
1051 LOG(3, 0, 0, "Peer %s set the master to %s...\n", fmtaddr(peer, 0),
1052 fmtaddr(master, 1));
1053
1054 config->cluster_master_address = master;
1055 cluster_check_master();
1056 return 0;
1057 }
1058
1059 /* Handle the slave updating the byte counters for the master. */
1060 //
1061 // Note that we don't mark the session as dirty; We rely on
1062 // the slow table walk to propogate this back out to the slaves.
1063 //
1064 static int cluster_handle_bytes(char *data, int size)
1065 {
1066 bytest *b;
1067
1068 b = (bytest *) data;
1069
1070 LOG(3, 0, 0, "Got byte counter update (size %d)\n", size);
1071
1072 /* Loop around, adding the byte
1073 counts to each of the sessions. */
1074
1075 while (size >= sizeof(*b) ) {
1076 if (b->sid > MAXSESSION) {
1077 LOG(0, 0, 0, "Got C_BYTES with session #%d!\n", b->sid);
1078 return -1; /* Abort processing */
1079 }
1080
1081 session[b->sid].total_cin += b->in;
1082 session[b->sid].total_cout += b->out;
1083
1084 session[b->sid].cin += b->in;
1085 session[b->sid].cout += b->out;
1086
1087 if (b->in)
1088 session[b->sid].last_packet = time_now; // Reset idle timer!
1089
1090 size -= sizeof(*b);
1091 ++b;
1092 }
1093
1094 if (size != 0)
1095 LOG(0, 0, 0, "Got C_BYTES with %d bytes of trailing junk!\n", size);
1096
1097 return size;
1098 }
1099
1100 //
1101 // Handle receiving a session structure in a heartbeat packet.
1102 //
1103 static int cluster_recv_session(int more, uint8_t *p)
1104 {
1105 if (more >= MAXSESSION) {
1106 LOG(0, 0, 0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
1107 return -1;
1108 }
1109
1110 if (session[more].tunnel == T_UNDEF) {
1111 if (config->cluster_iam_uptodate) { // Sanity.
1112 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined session!\n");
1113 } else {
1114 --config->cluster_undefined_sessions;
1115 }
1116 }
1117
1118 load_session(more, (sessiont *) p); // Copy session into session table..
1119
1120 LOG(5, more, 0, "Received session update (%d undef)\n", config->cluster_undefined_sessions);
1121
1122 if (!config->cluster_iam_uptodate)
1123 cluster_uptodate(); // Check to see if we're up to date.
1124
1125 return 0;
1126 }
1127
1128 static int cluster_recv_tunnel(int more, uint8_t *p)
1129 {
1130 if (more >= MAXTUNNEL) {
1131 LOG(0, 0, 0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
1132 return -1;
1133 }
1134
1135 if (tunnel[more].state == TUNNELUNDEF) {
1136 if (config->cluster_iam_uptodate) { // Sanity.
1137 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined tunnel!\n");
1138 } else {
1139 --config->cluster_undefined_tunnels;
1140 }
1141 }
1142
1143 memcpy(&tunnel[more], p, sizeof(tunnel[more]) );
1144
1145 //
1146 // Clear tunnel control messages. These are dynamically allocated.
1147 // If we get unlucky, this may cause the tunnel to drop!
1148 //
1149 tunnel[more].controls = tunnel[more].controle = NULL;
1150 tunnel[more].controlc = 0;
1151
1152 LOG(5, 0, more, "Received tunnel update\n");
1153
1154 if (!config->cluster_iam_uptodate)
1155 cluster_uptodate(); // Check to see if we're up to date.
1156
1157 return 0;
1158 }
1159
1160
1161 // pre v5 heartbeat session structure
1162 struct oldsession {
1163 sessionidt next;
1164 sessionidt far;
1165 tunnelidt tunnel;
1166 in_addr_t ip;
1167 int ip_pool_index;
1168 unsigned long unique_id;
1169 uint16_t nr;
1170 uint16_t ns;
1171 uint32_t magic;
1172 uint32_t cin, cout;
1173 uint32_t pin, pout;
1174 uint32_t total_cin;
1175 uint32_t total_cout;
1176 uint32_t id;
1177 uint16_t throttle_in;
1178 uint16_t throttle_out;
1179 clockt opened;
1180 clockt die;
1181 time_t last_packet;
1182 in_addr_t dns1, dns2;
1183 routet route[MAXROUTE];
1184 uint16_t radius;
1185 uint16_t mru;
1186 uint16_t tbf_in;
1187 uint16_t tbf_out;
1188 uint8_t l2tp_flags;
1189 uint8_t reserved_old_snoop;
1190 uint8_t walled_garden;
1191 uint8_t flags1;
1192 char random_vector[MAXTEL];
1193 int random_vector_length;
1194 char user[129];
1195 char called[MAXTEL];
1196 char calling[MAXTEL];
1197 uint32_t tx_connect_speed;
1198 uint32_t rx_connect_speed;
1199 uint32_t flags;
1200 in_addr_t snoop_ip;
1201 uint16_t snoop_port;
1202 uint16_t sid;
1203 uint8_t filter_in;
1204 uint8_t filter_out;
1205 char reserved[18];
1206 };
1207
1208 static uint8_t *convert_session(struct oldsession *old)
1209 {
1210 static sessiont new;
1211 int i;
1212
1213 memset(&new, 0, sizeof(new));
1214
1215 new.next = old->next;
1216 new.far = old->far;
1217 new.tunnel = old->tunnel;
1218 new.l2tp_flags = old->l2tp_flags;
1219 new.flags = old->flags;
1220 new.ip = old->ip;
1221 new.ip_pool_index = old->ip_pool_index;
1222 new.unique_id = old->unique_id;
1223 new.nr = old->nr;
1224 new.ns = old->ns;
1225 new.magic = old->magic;
1226 new.cin = old->cin;
1227 new.cout = old->cout;
1228 new.pin = old->pin;
1229 new.pout = old->pout;
1230 new.total_cin = old->total_cin;
1231 new.total_cout = old->total_cout;
1232 new.throttle_in = old->throttle_in;
1233 new.throttle_out = old->throttle_out;
1234 new.filter_in = old->filter_in;
1235 new.filter_out = old->filter_out;
1236 new.mru = old->mru;
1237 new.opened = old->opened;
1238 new.die = old->die;
1239 new.last_packet = old->last_packet;
1240 new.dns1 = old->dns1;
1241 new.dns2 = old->dns2;
1242 new.tbf_in = old->tbf_in;
1243 new.tbf_out = old->tbf_out;
1244 new.random_vector_length = old->random_vector_length;
1245 new.tx_connect_speed = old->tx_connect_speed;
1246 new.rx_connect_speed = old->rx_connect_speed;
1247 new.snoop_ip = old->snoop_ip;
1248 new.snoop_port = old->snoop_port;
1249 new.walled_garden = old->walled_garden;
1250
1251 memcpy(new.random_vector, old->random_vector, sizeof(new.random_vector));
1252 memcpy(new.user, old->user, sizeof(new.user));
1253 memcpy(new.called, old->called, sizeof(new.called));
1254 memcpy(new.calling, old->calling, sizeof(new.calling));
1255
1256 for (i = 0; i < MAXROUTE; i++)
1257 memcpy(&new.route[i], &old->route[i], sizeof(new.route[i]));
1258
1259 return (uint8_t *) &new;
1260 }
1261
1262 //
1263 // Process a heartbeat..
1264 //
1265 // v3: added interval, timeout
1266 // v4: added table_version
1267 // v5: added ipv6, re-ordered session structure
1268 static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t *p, in_addr_t addr)
1269 {
1270 heartt *h;
1271 int s = size - (p-data);
1272 int i, type;
1273 int hb_ver = more;
1274
1275 #if HB_VERSION != 5
1276 # error "need to update cluster_process_heartbeat()"
1277 #endif
1278
1279 // we handle versions 3 through 5
1280 if (hb_ver < 3 || hb_ver > HB_VERSION) {
1281 LOG(0, 0, 0, "Received a heartbeat version that I don't support (%d)!\n", hb_ver);
1282 return -1; // Ignore it??
1283 }
1284
1285 // Ok. It's a heartbeat packet from a cluster master!
1286 if (s < sizeof(*h))
1287 goto shortpacket;
1288
1289 h = (heartt *) p;
1290 p += sizeof(*h);
1291 s -= sizeof(*h);
1292
1293 if (h->clusterid != config->bind_address)
1294 return -1; // It's not part of our cluster.
1295
1296 if (config->cluster_iam_master) { // Sanity...
1297 // Note that this MUST match the election process above!
1298
1299 LOG(0, 0, 0, "I just got a heartbeat from master %s, but _I_ am the master!\n", fmtaddr(addr, 0));
1300 if (!h->basetime) {
1301 LOG(0, 0, 0, "Heartbeat with zero basetime! Ignoring\n");
1302 return -1; // Skip it.
1303 }
1304
1305 if (hb_ver >= 4) {
1306 if (h->table_version > config->cluster_table_version) {
1307 LOG(0, 0, 0, "They've seen more state changes (%" PRIu64 " vs my %" PRIu64 ") so I'm gone!\n",
1308 h->table_version, config->cluster_table_version);
1309
1310 kill(0, SIGTERM);
1311 exit(1);
1312 }
1313 if (h->table_version < config->cluster_table_version)
1314 return -1;
1315 }
1316
1317 if (basetime > h->basetime) {
1318 LOG(0, 0, 0, "They're an older master than me so I'm gone!\n");
1319 kill(0, SIGTERM);
1320 exit(1);
1321 }
1322
1323 if (basetime < h->basetime)
1324 return -1;
1325
1326 if (my_address < addr) { // Tie breaker.
1327 LOG(0, 0, 0, "They're a higher IP address than me, so I'm gone!\n");
1328 kill(0, SIGTERM);
1329 exit(1);
1330 }
1331
1332 //
1333 // Send it a unicast heartbeat to see give it a chance to die.
1334 // NOTE: It's actually safe to do seq-number - 1 without checking
1335 // for wrap around.
1336 //
1337 cluster_catchup_slave(config->cluster_seq_number - 1, addr);
1338
1339 return -1; // Skip it.
1340 }
1341
1342 //
1343 // Try and guard against a stray master appearing.
1344 //
1345 // Ignore heartbeats received from another master before the
1346 // timeout (less a smidgen) for the old master has elapsed.
1347 //
1348 // Note that after a clean failover, the cluster_master_address
1349 // is cleared, so this doesn't run.
1350 //
1351 if (config->cluster_master_address && addr != config->cluster_master_address
1352 && (config->cluster_last_hb + config->cluster_hb_timeout - 11) > TIME) {
1353 LOG(0, 0, 0, "Ignoring stray heartbeat from %s, current master %s has not yet timed out (last heartbeat %.1f seconds ago).\n",
1354 fmtaddr(addr, 0), fmtaddr(config->cluster_master_address, 1),
1355 0.1 * (TIME - config->cluster_last_hb));
1356 return -1; // ignore
1357 }
1358
1359 if (config->cluster_seq_number == -1) // Don't have one. Just align to the master...
1360 config->cluster_seq_number = h->seq;
1361
1362 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1363
1364 if (config->cluster_seq_number != h->seq) { // Out of sequence heartbeat!
1365 static int lastseen_seq = 0;
1366 static time_t lastseen_time = 0;
1367
1368 // limit to once per second for a particular seq#
1369 int ask = (config->cluster_seq_number != lastseen_seq || time_now != lastseen_time);
1370
1371 LOG(1, 0, 0, "HB: Got seq# %d but was expecting %d. %s.\n",
1372 h->seq, config->cluster_seq_number,
1373 ask ? "Asking for resend" : "Ignoring");
1374
1375 if (ask)
1376 {
1377 lastseen_seq = config->cluster_seq_number;
1378 lastseen_time = time_now;
1379 peer_send_message(addr, C_LASTSEEN, config->cluster_seq_number, NULL, 0);
1380 }
1381
1382 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1383
1384 // Just drop the packet. The master will resend it as part of the catchup.
1385
1386 return 0;
1387 }
1388 // Save the packet in our buffer.
1389 // This is needed in case we become the master.
1390 config->cluster_seq_number = (h->seq+1)%HB_MAX_SEQ;
1391 i = h->seq % HB_HISTORY_SIZE;
1392 past_hearts[i].seq = h->seq;
1393 past_hearts[i].size = size;
1394 memcpy(&past_hearts[i].data, data, size); // Save it.
1395
1396
1397 // Check that we don't have too many undefined sessions, and
1398 // that the free session pointer is correct.
1399 cluster_check_sessions(h->highsession, h->freesession, h->hightunnel);
1400
1401 if (h->interval != config->cluster_hb_interval)
1402 {
1403 LOG(2, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
1404 h->interval, config->cluster_hb_interval);
1405
1406 config->cluster_hb_interval = h->interval;
1407 }
1408
1409 if (h->timeout != config->cluster_hb_timeout)
1410 {
1411 LOG(2, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
1412 h->timeout, config->cluster_hb_timeout);
1413
1414 config->cluster_hb_timeout = h->timeout;
1415 }
1416
1417 // Ok. process the packet...
1418 while ( s > 0) {
1419
1420 type = *((uint32_t *) p);
1421 p += sizeof(uint32_t);
1422 s -= sizeof(uint32_t);
1423
1424 more = *((uint32_t *) p);
1425 p += sizeof(uint32_t);
1426 s -= sizeof(uint32_t);
1427
1428 switch (type) {
1429 case C_CSESSION: { // Compressed session structure.
1430 uint8_t c[ sizeof(sessiont) + 2];
1431 int size;
1432 uint8_t *orig_p = p;
1433
1434 size = rle_decompress((uint8_t **) &p, s, c, sizeof(c) );
1435 s -= (p - orig_p);
1436
1437 // session struct changed with v5
1438 if (hb_ver < 5)
1439 {
1440 if (size != sizeof(struct oldsession)) {
1441 LOG(0, 0, 0, "DANGER: Received a v%d CSESSION that didn't decompress correctly!\n", hb_ver);
1442 // Now what? Should exit! No-longer up to date!
1443 break;
1444 }
1445 cluster_recv_session(more, convert_session((struct oldsession *) c));
1446 break;
1447 }
1448
1449 if (size != sizeof(sessiont) ) { // Ouch! Very very bad!
1450 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1451 // Now what? Should exit! No-longer up to date!
1452 break;
1453 }
1454
1455 cluster_recv_session(more, c);
1456 break;
1457 }
1458 case C_SESSION:
1459 if (hb_ver < 5)
1460 {
1461 if (s < sizeof(struct oldsession))
1462 goto shortpacket;
1463
1464 cluster_recv_session(more, convert_session((struct oldsession *) p));
1465
1466 p += sizeof(struct oldsession);
1467 s -= sizeof(struct oldsession);
1468 break;
1469 }
1470
1471 if ( s < sizeof(session[more]))
1472 goto shortpacket;
1473
1474 cluster_recv_session(more, p);
1475
1476 p += sizeof(session[more]);
1477 s -= sizeof(session[more]);
1478 break;
1479
1480 case C_CTUNNEL: { // Compressed tunnel structure.
1481 uint8_t c[ sizeof(tunnelt) + 2];
1482 int size;
1483 uint8_t *orig_p = p;
1484
1485 size = rle_decompress((uint8_t **) &p, s, c, sizeof(c));
1486 s -= (p - orig_p);
1487
1488 if (size != sizeof(tunnelt) ) { // Ouch! Very very bad!
1489 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1490 // Now what? Should exit! No-longer up to date!
1491 break;
1492 }
1493
1494 cluster_recv_tunnel(more, c);
1495 break;
1496
1497 }
1498 case C_TUNNEL:
1499 if ( s < sizeof(tunnel[more]))
1500 goto shortpacket;
1501
1502 cluster_recv_tunnel(more, p);
1503
1504 p += sizeof(tunnel[more]);
1505 s -= sizeof(tunnel[more]);
1506 break;
1507 default:
1508 LOG(0, 0, 0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type);
1509 return -1; // can't process any more of the packet!!
1510 }
1511 }
1512
1513 if (config->cluster_master_address != addr)
1514 {
1515 LOG(0, 0, 0, "My master just changed from %s to %s!\n",
1516 fmtaddr(config->cluster_master_address, 0), fmtaddr(addr, 1));
1517
1518 config->cluster_master_address = addr;
1519 }
1520
1521 config->cluster_last_hb = TIME; // Successfully received a heartbeat!
1522 config->cluster_table_version = h->table_version;
1523 return 0;
1524
1525 shortpacket:
1526 LOG(0, 0, 0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
1527 return -1;
1528 }
1529
1530 //
1531 // We got a packet on the cluster port!
1532 // Handle pings, lastseens, and heartbeats!
1533 //
1534 int processcluster(char *data, int size, in_addr_t addr)
1535 {
1536 int type, more;
1537 char *p = data;
1538 int s = size;
1539
1540 if (addr == my_address)
1541 return -1; // Ignore it. Something looped back the multicast!
1542
1543 LOG(5, 0, 0, "Process cluster: %d bytes from %s\n", size, fmtaddr(addr, 0));
1544
1545 if (s <= 0) // Any data there??
1546 return -1;
1547
1548 if (s < 8)
1549 goto shortpacket;
1550
1551 type = *((uint32_t *) p);
1552 p += sizeof(uint32_t);
1553 s -= sizeof(uint32_t);
1554
1555 more = *((uint32_t *) p);
1556 p += sizeof(uint32_t);
1557 s -= sizeof(uint32_t);
1558
1559 switch (type) {
1560 case C_PING: // Update the peers table.
1561 return cluster_add_peer(addr, more, (pingt *) p, s);
1562
1563 case C_MASTER: // Our master is wrong
1564 return cluster_set_master(addr, more);
1565
1566 case C_LASTSEEN: // Catch up a slave (slave missed a packet).
1567 return cluster_catchup_slave(more, addr);
1568
1569 case C_FORWARD: { // Forwarded control packet. pass off to processudp.
1570 struct sockaddr_in a;
1571 a.sin_addr.s_addr = more;
1572
1573 a.sin_port = *(int *) p;
1574 s -= sizeof(int);
1575 p += sizeof(int);
1576
1577 if (!config->cluster_iam_master) { // huh?
1578 LOG(0, 0, 0, "I'm not the master, but I got a C_FORWARD from %s?\n", fmtaddr(addr, 0));
1579 return -1;
1580 }
1581
1582 LOG(4, 0, 0, "Got a forwarded packet... (%s:%d)\n", fmtaddr(more, 0), a.sin_port);
1583 STAT(recv_forward);
1584 processudp(p, s, &a);
1585 return 0;
1586 }
1587 case C_THROTTLE: { // Receive a forwarded packet from a slave.
1588 if (!config->cluster_iam_master) {
1589 LOG(0, 0, 0, "I'm not the master, but I got a C_THROTTLE from %s?\n", fmtaddr(addr, 0));
1590 return -1;
1591 }
1592
1593 tbf_queue_packet(more, p, s); // The TBF id tells wether it goes in or out.
1594 return 0;
1595 }
1596 case C_GARDEN:
1597 // Receive a walled garden packet from a slave.
1598 if (!config->cluster_iam_master) {
1599 LOG(0, 0, 0, "I'm not the master, but I got a C_GARDEN from %s?\n", fmtaddr(addr, 0));
1600 return -1;
1601 }
1602
1603 tun_write(p, s);
1604 return 0;
1605
1606 case C_BYTES:
1607 if (!config->cluster_iam_master) {
1608 LOG(0, 0, 0, "I'm not the master, but I got a C_BYTES from %s?\n", fmtaddr(addr, 0));
1609 return -1;
1610 }
1611
1612 return cluster_handle_bytes(p, s);
1613
1614 case C_KILL: // The master asked us to die!? (usually because we're too out of date).
1615 if (config->cluster_iam_master) {
1616 LOG(0, 0, 0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", fmtaddr(addr, 0), more);
1617 return -1;
1618 }
1619 if (more != config->cluster_seq_number) {
1620 LOG(0, 0, 0, "The master asked us to die but the seq number didn't match!?\n");
1621 return -1;
1622 }
1623
1624 if (addr != config->cluster_master_address) {
1625 LOG(0, 0, 0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%s)\n",
1626 fmtaddr(addr, 0), fmtaddr(config->cluster_master_address, 1));
1627 // We can only warn about it. The master might really have switched!
1628 }
1629
1630 LOG(0, 0, 0, "Received a valid C_KILL: I'm going to die now.\n");
1631 kill(0, SIGTERM);
1632 exit(0); // Lets be paranoid;
1633 return -1; // Just signalling the compiler.
1634
1635 case C_HEARTBEAT:
1636 LOG(4, 0, 0, "Got a heartbeat from %s\n", fmtaddr(addr, 0));
1637 return cluster_process_heartbeat(data, size, more, p, addr);
1638
1639 default:
1640 LOG(0, 0, 0, "Strange type packet received on cluster socket (%d)\n", type);
1641 return -1;
1642 }
1643 return 0;
1644
1645 shortpacket:
1646 LOG(0, 0, 0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n");
1647 return -1;
1648 }
1649
1650 //====================================================================================================
1651
1652 int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
1653 {
1654 int i;
1655
1656 if (CLI_HELP_REQUESTED)
1657 return CLI_HELP_NO_ARGS;
1658
1659 cli_print(cli, "Cluster status : %s", config->cluster_iam_master ? "Master" : "Slave" );
1660 cli_print(cli, "My address : %s", fmtaddr(my_address, 0));
1661 cli_print(cli, "VIP address : %s", fmtaddr(config->bind_address, 0));
1662 cli_print(cli, "Multicast address: %s", fmtaddr(config->cluster_address, 0));
1663 cli_print(cli, "Multicast i'face : %s", config->cluster_interface);
1664
1665 if (!config->cluster_iam_master) {
1666 cli_print(cli, "My master : %s (last heartbeat %.1f seconds old)",
1667 config->cluster_master_address
1668 ? fmtaddr(config->cluster_master_address, 0)
1669 : "Not defined",
1670 0.1 * (TIME - config->cluster_last_hb));
1671 cli_print(cli, "Uptodate : %s", config->cluster_iam_uptodate ? "Yes" : "No");
1672 cli_print(cli, "Table version # : %" PRIu64, config->cluster_table_version);
1673 cli_print(cli, "Next sequence number expected: %d", config->cluster_seq_number);
1674 cli_print(cli, "%d sessions undefined of %d", config->cluster_undefined_sessions, config->cluster_highest_sessionid);
1675 cli_print(cli, "%d tunnels undefined of %d", config->cluster_undefined_tunnels, config->cluster_highest_tunnelid);
1676 } else {
1677 cli_print(cli, "Table version # : %" PRIu64, config->cluster_table_version);
1678 cli_print(cli, "Next heartbeat # : %d", config->cluster_seq_number);
1679 cli_print(cli, "Highest session : %d", config->cluster_highest_sessionid);
1680 cli_print(cli, "Highest tunnel : %d", config->cluster_highest_tunnelid);
1681 cli_print(cli, "%d changes queued for sending", config->cluster_num_changes);
1682 }
1683 cli_print(cli, "%d peers.", num_peers);
1684
1685 if (num_peers)
1686 cli_print(cli, "%20s %10s %8s", "Address", "Basetime", "Age");
1687 for (i = 0; i < num_peers; ++i) {
1688 cli_print(cli, "%20s %10u %8d", fmtaddr(peers[i].peer, 0),
1689 peers[i].basetime, TIME - peers[i].timestamp);
1690 }
1691 return CLI_OK;
1692 }
1693
1694 //
1695 // Simple run-length-encoding compression.
1696 // Format is
1697 // 1 byte < 128 = count of non-zero bytes following. // Not legal to be zero.
1698 // n non-zero bytes;
1699 // or
1700 // 1 byte > 128 = (count - 128) run of zero bytes. //
1701 // repeat.
1702 // count == 0 indicates end of compressed stream.
1703 //
1704 // Compress from 'src' into 'dst'. return number of bytes
1705 // used from 'dst'.
1706 // Updates *src_p to indicate 1 past last bytes used.
1707 //
1708 // We could get an extra byte in the zero runs by storing (count-1)
1709 // but I'm playing it safe.
1710 //
1711 // Worst case is a 50% expansion in space required (trying to
1712 // compress { 0x00, 0x01 } * N )
1713 static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
1714 {
1715 int count;
1716 int orig_dsize = dsize;
1717 uint8_t *x, *src;
1718 src = *src_p;
1719
1720 while (ssize > 0 && dsize > 2) {
1721 count = 0;
1722 x = dst++; --dsize; // Reserve space for count byte..
1723
1724 if (*src) { // Copy a run of non-zero bytes.
1725 while (*src && count < 127 && ssize > 0 && dsize > 1) { // Count number of non-zero bytes.
1726 *dst++ = *src++;
1727 --dsize; --ssize;
1728 ++count;
1729 }
1730 *x = count; // Store number of non-zero bytes. Guarenteed to be non-zero!
1731
1732 } else { // Compress a run of zero bytes.
1733 while (*src == 0 && count < 127 && ssize > 0) {
1734 ++src;
1735 --ssize;
1736 ++count;
1737 }
1738 *x = count | 0x80 ;
1739 }
1740 }
1741
1742 *dst++ = 0x0; // Add Stop byte.
1743 --dsize;
1744
1745 *src_p = src;
1746 return (orig_dsize - dsize);
1747 }
1748
1749 //
1750 // Decompress the buffer into **p.
1751 // 'psize' is the size of the decompression buffer available.
1752 //
1753 // Returns the number of bytes decompressed.
1754 //
1755 // Decompresses from '*src_p' into 'dst'.
1756 // Return the number of dst bytes used.
1757 // Updates the 'src_p' pointer to point to the
1758 // first un-used byte.
1759 static int rle_decompress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
1760 {
1761 int count;
1762 int orig_dsize = dsize;
1763 char *src = *src_p;
1764
1765 while (ssize >0 && dsize > 0) { // While there's more to decompress, and there's room in the decompress buffer...
1766 count = *src++; --ssize; // get the count byte from the source.
1767 if (count == 0x0) // End marker reached? If so, finish.
1768 break;
1769
1770 if (count & 0x80) { // Decompress a run of zeros
1771 for (count &= 0x7f ; count > 0 && dsize > 0; --count) {
1772 *dst++ = 0x0;
1773 --dsize;
1774 }
1775 } else { // Copy run of non-zero bytes.
1776 for ( ; count > 0 && ssize && dsize; --count) { // Copy non-zero bytes across.
1777 *dst++ = *src++;
1778 --ssize; --dsize;
1779 }
1780 }
1781 }
1782 *src_p = src;
1783 return (orig_dsize - dsize);
1784 }