backout regular_cleanup changes
[l2tpns.git] / cluster.c
1 // L2TPNS Clustering Stuff
2
3 char const *cvs_id_cluster = "$Id: cluster.c,v 1.26.2.11 2005/05/30 02:55:40 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
47 #define MAX_HEART_SIZE (8192) // Maximum size of heartbeat packet. Must be less than max IP packet size :)
48 #define MAX_CHANGES (MAX_HEART_SIZE/(sizeof(sessiont) + sizeof(int) ) - 2) // Assumes a session is the biggest type!
49
50 static struct {
51 int type;
52 int id;
53 } cluster_changes[MAX_CHANGES]; // Queue of changed structures that need to go out when next heartbeat.
54
55 static struct {
56 int seq;
57 int size;
58 char data[MAX_HEART_SIZE];
59 } past_hearts[HB_HISTORY_SIZE]; // Ring buffer of heartbeats that we've recently sent out. Needed so
60 // we can re-transmit if needed.
61
62 static struct {
63 in_addr_t peer;
64 uint32_t basetime;
65 clockt timestamp;
66 int uptodate;
67 } peers[CLUSTER_MAX_SIZE]; // List of all the peers we've heard from.
68 static int num_peers; // Number of peers in list.
69
70 static int rle_decompress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize);
71 static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize);
72
73 //
74 // Create a listening socket
75 //
76 // This joins the cluster multi-cast group.
77 //
78 int cluster_init()
79 {
80 struct sockaddr_in addr;
81 struct sockaddr_in interface_addr;
82 struct ip_mreq mreq;
83 struct ifreq ifr;
84 int opt;
85
86 config->cluster_undefined_sessions = MAXSESSION-1;
87 config->cluster_undefined_tunnels = MAXTUNNEL-1;
88
89 if (!config->cluster_address)
90 return 0;
91 if (!*config->cluster_interface)
92 return 0;
93
94 cluster_sockfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
95
96 memset(&addr, 0, sizeof(addr));
97 addr.sin_family = AF_INET;
98 addr.sin_port = htons(CLUSTERPORT);
99 addr.sin_addr.s_addr = INADDR_ANY;
100 setsockopt(cluster_sockfd, SOL_SOCKET, SO_REUSEADDR, &addr, sizeof(addr));
101
102 opt = fcntl(cluster_sockfd, F_GETFL, 0);
103 fcntl(cluster_sockfd, F_SETFL, opt | O_NONBLOCK);
104
105 if (bind(cluster_sockfd, (void *) &addr, sizeof(addr)) < 0)
106 {
107 LOG(0, 0, 0, "Failed to bind cluster socket: %s\n", strerror(errno));
108 return -1;
109 }
110
111 strcpy(ifr.ifr_name, config->cluster_interface);
112 if (ioctl(cluster_sockfd, SIOCGIFADDR, &ifr) < 0)
113 {
114 LOG(0, 0, 0, "Failed to get interface address for (%s): %s\n", config->cluster_interface, strerror(errno));
115 return -1;
116 }
117
118 memcpy(&interface_addr, &ifr.ifr_addr, sizeof(interface_addr));
119 my_address = interface_addr.sin_addr.s_addr;
120
121 // Join multicast group.
122 mreq.imr_multiaddr.s_addr = config->cluster_address;
123 mreq.imr_interface = interface_addr.sin_addr;
124
125
126 opt = 0; // Turn off multicast loopback.
127 setsockopt(cluster_sockfd, IPPROTO_IP, IP_MULTICAST_LOOP, &opt, sizeof(opt));
128
129 if (setsockopt(cluster_sockfd, IPPROTO_IP, IP_ADD_MEMBERSHIP, &mreq, sizeof(mreq)) < 0)
130 {
131 LOG(0, 0, 0, "Failed to setsockopt (join mcast group): %s\n", strerror(errno));
132 return -1;
133 }
134
135 if (setsockopt(cluster_sockfd, IPPROTO_IP, IP_MULTICAST_IF, &interface_addr, sizeof(interface_addr)) < 0)
136 {
137 LOG(0, 0, 0, "Failed to setsockopt (set mcast interface): %s\n", strerror(errno));
138 return -1;
139 }
140
141 config->cluster_last_hb = TIME;
142 config->cluster_seq_number = -1;
143
144 return cluster_sockfd;
145 }
146
147
148 //
149 // Send a chunk of data to the entire cluster (usually via the multicast
150 // address ).
151 //
152
153 static int cluster_send_data(void *data, int datalen)
154 {
155 struct sockaddr_in addr = {0};
156
157 if (!cluster_sockfd) return -1;
158 if (!config->cluster_address) return 0;
159
160 addr.sin_addr.s_addr = config->cluster_address;
161 addr.sin_port = htons(CLUSTERPORT);
162 addr.sin_family = AF_INET;
163
164 LOG(5, 0, 0, "Cluster send data: %d bytes\n", datalen);
165
166 if (sendto(cluster_sockfd, data, datalen, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
167 {
168 LOG(0, 0, 0, "sendto: %s\n", strerror(errno));
169 return -1;
170 }
171
172 return 0;
173 }
174
175 //
176 // Add a chunk of data to a heartbeat packet.
177 // Maintains the format. Assumes that the caller
178 // has passed in a big enough buffer!
179 //
180 static void add_type(char **p, int type, int more, char *data, int size)
181 {
182 *((uint32_t *) (*p)) = type;
183 *p += sizeof(uint32_t);
184
185 *((uint32_t *)(*p)) = more;
186 *p += sizeof(uint32_t);
187
188 if (data && size > 0) {
189 memcpy(*p, data, size);
190 *p += size;
191 }
192 }
193
194 // advertise our presence via BGP or gratuitous ARP
195 static void advertise_routes(void)
196 {
197 #ifdef BGP
198 if (bgp_configured)
199 bgp_enable_routing(1);
200 else
201 #endif /* BGP */
202 if (config->send_garp)
203 send_garp(config->bind_address); // Start taking traffic.
204 }
205
206 // withdraw our routes (BGP only)
207 static void withdraw_routes(void)
208 {
209 #ifdef BGP
210 if (bgp_configured)
211 bgp_enable_routing(0);
212 #endif /* BGP */
213 }
214
215 static void cluster_uptodate(void)
216 {
217 if (config->cluster_iam_uptodate)
218 return;
219
220 if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels)
221 return;
222
223 config->cluster_iam_uptodate = 1;
224
225 LOG(0, 0, 0, "Now uptodate with master.\n");
226 advertise_routes();
227 }
228
229 //
230 // Send a unicast UDP packet to a peer with 'data' as the
231 // contents.
232 //
233 static int peer_send_data(in_addr_t peer, char *data, int size)
234 {
235 struct sockaddr_in addr = {0};
236
237 if (!cluster_sockfd) return -1;
238 if (!config->cluster_address) return 0;
239
240 if (!peer) // Odd??
241 return -1;
242
243 addr.sin_addr.s_addr = peer;
244 addr.sin_port = htons(CLUSTERPORT);
245 addr.sin_family = AF_INET;
246
247 LOG_HEX(5, "Peer send", data, size);
248
249 if (sendto(cluster_sockfd, data, size, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
250 {
251 LOG(0, 0, 0, "sendto: %s\n", strerror(errno));
252 return -1;
253 }
254
255 return 0;
256 }
257
258 //
259 // Send a structured message to a peer with a single element of type 'type'.
260 //
261 static int peer_send_message(in_addr_t peer, int type, int more, char *data, int size)
262 {
263 char buf[65536]; // Vast overkill.
264 char *p = buf;
265
266 LOG(4, 0, 0, "Sending message to peer (type %d, more %d, size %d)\n", type, more, size);
267 add_type(&p, type, more, data, size);
268
269 return peer_send_data(peer, buf, (p-buf) );
270 }
271
272 //
273 // Forward a state changing packet to the master.
274 //
275 // The master just processes the payload as if it had
276 // received it off the tun device.
277 //
278 int master_forward_packet(char *data, int size, in_addr_t addr, int port)
279 {
280 char buf[65536]; // Vast overkill.
281 char *p = buf;
282
283 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
284 return -1;
285
286 LOG(4, 0, 0, "Forwarding packet from %s to master (size %d)\n", fmtaddr(addr, 0), size);
287
288 STAT(c_forwarded);
289 add_type(&p, C_FORWARD, addr, (char *) &port, sizeof(port));
290 memcpy(p, data, size);
291 p += size;
292
293 return peer_send_data(config->cluster_master_address, buf, (p - buf));
294 }
295
296 //
297 // Forward a throttled packet to the master for handling.
298 //
299 // The master just drops the packet into the appropriate
300 // token bucket queue, and lets normal processing take care
301 // of it.
302 //
303 int master_throttle_packet(int tbfid, char *data, int size)
304 {
305 char buf[65536]; // Vast overkill.
306 char *p = buf;
307
308 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
309 return -1;
310
311 LOG(4, 0, 0, "Throttling packet master (size %d, tbfid %d)\n", size, tbfid);
312
313 add_type(&p, C_THROTTLE, tbfid, data, size);
314
315 return peer_send_data(config->cluster_master_address, buf, (p-buf) );
316
317 }
318
319 //
320 // Forward a walled garden packet to the master for handling.
321 //
322 // The master just writes the packet straight to the tun
323 // device (where is will normally loop through the
324 // firewall rules, and come back in on the tun device)
325 //
326 // (Note that this must be called with the tun header
327 // as the start of the data).
328 int master_garden_packet(sessionidt s, char *data, int size)
329 {
330 char buf[65536]; // Vast overkill.
331 char *p = buf;
332
333 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
334 return -1;
335
336 LOG(4, 0, 0, "Walled garden packet to master (size %d)\n", size);
337
338 add_type(&p, C_GARDEN, s, data, size);
339
340 return peer_send_data(config->cluster_master_address, buf, (p-buf));
341
342 }
343
344 //
345 // Send a chunk of data as a heartbeat..
346 // We save it in the history buffer as we do so.
347 //
348 static void send_heartbeat(int seq, char *data, int size)
349 {
350 int i;
351
352 if (size > sizeof(past_hearts[0].data))
353 {
354 LOG(0, 0, 0, "Tried to heartbeat something larger than the maximum packet!\n");
355 kill(0, SIGTERM);
356 exit(1);
357 }
358 i = seq % HB_HISTORY_SIZE;
359 past_hearts[i].seq = seq;
360 past_hearts[i].size = size;
361 memcpy(&past_hearts[i].data, data, size); // Save it.
362 cluster_send_data(data, size);
363 }
364
365 //
366 // Send an 'i am alive' message to every machine in the cluster.
367 //
368 void cluster_send_ping(time_t basetime)
369 {
370 char buff[100 + sizeof(pingt)];
371 char *p = buff;
372 pingt x;
373
374 if (config->cluster_iam_master && basetime) // We're heartbeating so no need to ping.
375 return;
376
377 LOG(5, 0, 0, "Sending cluster ping...\n");
378
379 x.ver = 1;
380 x.addr = config->bind_address;
381 x.undef = config->cluster_undefined_sessions + config->cluster_undefined_tunnels;
382 x.basetime = basetime;
383
384 add_type(&p, C_PING, basetime, (char *) &x, sizeof(x));
385 cluster_send_data(buff, (p-buff) );
386 }
387
388 //
389 // Walk the session counters looking for non-zero ones to send
390 // to the master. We send up to 600 of them at one time.
391 // We examine a maximum of 3000 sessions.
392 // (50k max session should mean that we normally
393 // examine the entire session table every 25 seconds).
394
395 #define MAX_B_RECS (600)
396 void master_update_counts(void)
397 {
398 int i, c;
399 bytest b[MAX_B_RECS+1];
400
401 if (config->cluster_iam_master) // Only happens on the slaves.
402 return;
403
404 if (!config->cluster_master_address) // If we don't have a master, skip it for a while.
405 return;
406
407 i = MAX_B_RECS * 5; // Examine max 2000 sessions;
408 if (config->cluster_highest_sessionid > i)
409 i = config->cluster_highest_sessionid;
410
411 for ( c = 0; i > 0 ; --i) {
412 // Next session to look at.
413 walk_session_number++;
414 if ( walk_session_number > config->cluster_highest_sessionid)
415 walk_session_number = 1;
416
417 if (!sess_local[walk_session_number].cin && !sess_local[walk_session_number].cout)
418 continue; // Unused. Skip it.
419
420 b[c].sid = walk_session_number;
421 b[c].in = sess_local[walk_session_number].cin;
422 b[c].out = sess_local[walk_session_number].cout;
423
424 if (++c > MAX_B_RECS) // Send a max of 400 elements in a packet.
425 break;
426
427 // Reset counters.
428 sess_local[walk_session_number].cin = sess_local[walk_session_number].cout = 0;
429 }
430
431 if (!c) // Didn't find any that changes. Get out of here!
432 return;
433
434
435 // Forward the data to the master.
436 LOG(4, 0, 0, "Sending byte counters to master (%d elements)\n", c);
437 peer_send_message(config->cluster_master_address, C_BYTES, c, (char *) &b, sizeof(b[0]) * c);
438 return;
439 }
440
441 //
442 // On the master, check how our slaves are going. If
443 // one of them's not up-to-date we'll heartbeat faster.
444 // If we don't have any of them, then we need to turn
445 // on our own packet handling!
446 //
447 void cluster_check_slaves(void)
448 {
449 int i;
450 static int have_peers = 0;
451 int had_peers = have_peers;
452 clockt t = TIME;
453
454 if (!config->cluster_iam_master)
455 return; // Only runs on the master...
456
457 config->cluster_iam_uptodate = 1; // cleared in loop below
458
459 for (i = have_peers = 0; i < num_peers; i++)
460 {
461 if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
462 continue; // Stale peer! Skip them.
463
464 if (!peers[i].basetime)
465 continue; // Shutdown peer! Skip them.
466
467 if (peers[i].uptodate)
468 have_peers++;
469 else
470 config->cluster_iam_uptodate = 0; // Start fast heartbeats
471 }
472
473 // in a cluster, withdraw/add routes when we get a peer/lose peers
474 if (have_peers != had_peers)
475 {
476 if (had_peers < config->cluster_master_min_adv &&
477 have_peers >= config->cluster_master_min_adv)
478 withdraw_routes();
479
480 else if (had_peers >= config->cluster_master_min_adv &&
481 have_peers < config->cluster_master_min_adv)
482 advertise_routes();
483 }
484 }
485
486 //
487 // Check that we have a master. If it's been too
488 // long since we heard from a master then hold an election.
489 //
490 void cluster_check_master(void)
491 {
492 int i, count, tcount, high_unique_id = 0;
493 int last_free = 0;
494 clockt t = TIME;
495 static int probed = 0;
496 int have_peers;
497
498 if (config->cluster_iam_master)
499 return; // Only runs on the slaves...
500
501 // If the master is late (missed 2 hearbeats by a second and a
502 // hair) it may be that the switch has dropped us from the
503 // multicast group, try unicasting probes to the master
504 // which will hopefully respond with a unicast heartbeat that
505 // will allow us to limp along until the querier next runs.
506 if (config->cluster_master_address
507 && TIME > (config->cluster_last_hb + 2 * config->cluster_hb_interval + 11))
508 {
509 if (!probed || (TIME > (probed + 2 * config->cluster_hb_interval)))
510 {
511 probed = TIME;
512 LOG(1, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
513 0.1 * (TIME - (config->cluster_last_hb + config->cluster_hb_interval)));
514
515 peer_send_message(config->cluster_master_address,
516 C_LASTSEEN, config->cluster_seq_number, NULL, 0);
517 }
518 } else { // We got a recent heartbeat; reset the probe flag.
519 probed = 0;
520 }
521
522 if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout))
523 return; // Everything's ok!
524
525 config->cluster_last_hb = TIME + 1; // Just the one election thanks.
526
527 LOG(0, 0, 0, "Master timed out! Holding election...\n");
528
529 // In the process of shutting down, can't be master
530 if (main_quit)
531 return;
532
533 for (i = have_peers = 0; i < num_peers; i++)
534 {
535 if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
536 continue; // Stale peer! Skip them.
537
538 if (!peers[i].basetime)
539 continue; // Shutdown peer! Skip them.
540
541 if (peers[i].basetime < basetime) {
542 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
543 return; // They'll win the election. Get out of here.
544 }
545
546 if (peers[i].basetime == basetime &&
547 peers[i].peer > my_address) {
548 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
549 return; // They'll win the election. Wait for them to come up.
550 }
551
552 if (peers[i].uptodate)
553 have_peers++;
554 }
555
556 // Wow. it's been ages since I last heard a heartbeat
557 // and I'm better than an of my peers so it's time
558 // to become a master!!!
559
560 config->cluster_iam_master = 1;
561 config->cluster_master_address = 0;
562
563 LOG(0, 0, 0, "I am declaring myself the master!\n");
564
565 if (have_peers < config->cluster_master_min_adv)
566 advertise_routes();
567 else
568 withdraw_routes();
569
570 if (config->cluster_seq_number == -1)
571 config->cluster_seq_number = 0;
572
573 //
574 // Go through and mark all the tunnels as defined.
575 // Count the highest used tunnel number as well.
576 //
577 config->cluster_highest_tunnelid = 0;
578 for (i = 0, tcount = 0; i < MAXTUNNEL; ++i) {
579 if (tunnel[i].state == TUNNELUNDEF)
580 tunnel[i].state = TUNNELFREE;
581
582 if (tunnel[i].state != TUNNELFREE && i > config->cluster_highest_tunnelid)
583 config->cluster_highest_tunnelid = i;
584 }
585
586 //
587 // Go through and mark all the sessions as being defined.
588 // reset the idle timeouts.
589 // add temporary byte counters to permanent ones.
590 // Re-string the free list.
591 // Find the ID of the highest session.
592 last_free = 0;
593 high_unique_id = 0;
594 config->cluster_highest_sessionid = 0;
595 for (i = 0, count = 0; i < MAXSESSION; ++i) {
596 if (session[i].tunnel == T_UNDEF) {
597 session[i].tunnel = T_FREE;
598 ++count;
599 }
600
601 if (!session[i].opened) { // Unused session. Add to free list.
602 memset(&session[i], 0, sizeof(session[i]));
603 session[i].tunnel = T_FREE;
604 session[last_free].next = i;
605 session[i].next = 0;
606 last_free = i;
607 continue;
608 }
609
610 // Reset idle timeouts..
611 session[i].last_packet = time_now;
612
613 // Reset die relative to our uptime rather than the old master's
614 if (session[i].die) session[i].die = TIME;
615
616 // Accumulate un-sent byte counters.
617 session[i].cin += sess_local[i].cin;
618 session[i].cout += sess_local[i].cout;
619 session[i].total_cin += sess_local[i].cin;
620 session[i].total_cout += sess_local[i].cout;
621
622 sess_local[i].cin = sess_local[i].cout = 0;
623
624 session[i].radius = 0; // Reset authentication as the radius blocks aren't up to date.
625
626 if (session[i].unique_id >= high_unique_id) // This is different to the index into the session table!!!
627 high_unique_id = session[i].unique_id+1;
628
629 session[i].tbf_in = session[i].tbf_out = 0; // Remove stale pointers from old master.
630 throttle_session(i, session[i].throttle_in, session[i].throttle_out);
631
632 config->cluster_highest_sessionid = i;
633 }
634
635 session[last_free].next = 0; // End of chain.
636 last_id = high_unique_id; // Keep track of the highest used session ID.
637
638 become_master();
639
640 rebuild_address_pool();
641
642 // If we're not the very first master, this is a big issue!
643 if(count>0)
644 LOG(0, 0, 0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count);
645
646 config->cluster_undefined_sessions = 0;
647 config->cluster_undefined_tunnels = 0;
648 config->cluster_iam_uptodate = 1; // assume all peers are up-to-date
649
650 // FIXME. We need to fix up the tunnel control message
651 // queue here! There's a number of other variables we
652 // should also update.
653 }
654
655
656 //
657 // Check that our session table is validly matching what the
658 // master has in mind.
659 //
660 // In particular, if we have too many sessions marked 'undefined'
661 // we fix it up here, and we ensure that the 'first free session'
662 // pointer is valid.
663 //
664 static void cluster_check_sessions(int highsession, int freesession_ptr, int hightunnel)
665 {
666 int i;
667
668 sessionfree = freesession_ptr; // Keep the freesession ptr valid.
669
670 if (config->cluster_iam_uptodate)
671 return;
672
673 if (highsession > config->cluster_undefined_sessions && hightunnel > config->cluster_undefined_tunnels)
674 return;
675
676 // Clear out defined sessions, counting the number of
677 // undefs remaining.
678 config->cluster_undefined_sessions = 0;
679 for (i = 1 ; i < MAXSESSION; ++i) {
680 if (i > highsession) {
681 if (session[i].tunnel == T_UNDEF) session[i].tunnel = T_FREE; // Defined.
682 continue;
683 }
684
685 if (session[i].tunnel == T_UNDEF)
686 ++config->cluster_undefined_sessions;
687 }
688
689 // Clear out defined tunnels, counting the number of
690 // undefs remaining.
691 config->cluster_undefined_tunnels = 0;
692 for (i = 1 ; i < MAXTUNNEL; ++i) {
693 if (i > hightunnel) {
694 if (tunnel[i].state == TUNNELUNDEF) tunnel[i].state = TUNNELFREE; // Defined.
695 continue;
696 }
697
698 if (tunnel[i].state == TUNNELUNDEF)
699 ++config->cluster_undefined_tunnels;
700 }
701
702
703 if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels) {
704 LOG(2, 0, 0, "Cleared undefined sessions/tunnels. %d sess (high %d), %d tunn (high %d)\n",
705 config->cluster_undefined_sessions, highsession, config->cluster_undefined_tunnels, hightunnel);
706 return;
707 }
708
709 // Are we up to date?
710
711 if (!config->cluster_iam_uptodate)
712 cluster_uptodate();
713 }
714
715 static int hb_add_type(char **p, int type, int id)
716 {
717 switch (type) {
718 case C_CSESSION: { // Compressed C_SESSION.
719 uint8_t c[sizeof(sessiont) * 2]; // Bigger than worst case.
720 uint8_t *d = (uint8_t *) &session[id];
721 uint8_t *orig = d;
722 int size;
723
724 size = rle_compress( &d, sizeof(sessiont), c, sizeof(c) );
725
726 // Did we compress the full structure, and is the size actually
727 // reduced??
728 if ( (d - orig) == sizeof(sessiont) && size < sizeof(sessiont) ) {
729 add_type(p, C_CSESSION, id, (char *) c, size);
730 break;
731 }
732 // Failed to compress : Fall through.
733 }
734 case C_SESSION: add_type(p, C_SESSION, id,
735 (char *) &session[id], sizeof(sessiont));
736 break;
737
738 case C_CTUNNEL: { // Compressed C_TUNNEL
739 uint8_t c[sizeof(tunnelt) * 2]; // Bigger than worst case.
740 uint8_t *d = (uint8_t *) &tunnel[id];
741 uint8_t *orig = d;
742 int size;
743
744 size = rle_compress( &d, sizeof(tunnelt), c, sizeof(c) );
745
746 // Did we compress the full structure, and is the size actually
747 // reduced??
748 if ( (d - orig) == sizeof(tunnelt) && size < sizeof(tunnelt) ) {
749 add_type(p, C_CTUNNEL, id, c, size);
750 break;
751 }
752 // Failed to compress : Fall through.
753 }
754 case C_TUNNEL: add_type(p, C_TUNNEL, id,
755 (char *) &tunnel[id], sizeof(tunnelt));
756 break;
757 default:
758 LOG(0, 0, 0, "Found an invalid type in heart queue! (%d)\n", type);
759 kill(0, SIGTERM);
760 exit(1);
761 }
762 return 0;
763 }
764
765 //
766 // Send a heartbeat, incidently sending out any queued changes..
767 //
768 void cluster_heartbeat()
769 {
770 int i, count = 0, tcount = 0;
771 char buff[MAX_HEART_SIZE + sizeof(heartt) + sizeof(int) ];
772 heartt h;
773 char *p = buff;
774
775 if (!config->cluster_iam_master) // Only the master does this.
776 return;
777
778 config->cluster_table_version += config->cluster_num_changes;
779
780 // Fill out the heartbeat header.
781 memset(&h, 0, sizeof(h));
782
783 h.version = HB_VERSION;
784 h.seq = config->cluster_seq_number;
785 h.basetime = basetime;
786 h.clusterid = config->bind_address; // Will this do??
787 h.basetime = basetime;
788 h.highsession = config->cluster_highest_sessionid;
789 h.freesession = sessionfree;
790 h.hightunnel = config->cluster_highest_tunnelid;
791 h.size_sess = sizeof(sessiont); // Just in case.
792 h.size_tunn = sizeof(tunnelt);
793 h.interval = config->cluster_hb_interval;
794 h.timeout = config->cluster_hb_timeout;
795 h.table_version = config->cluster_table_version;
796
797 add_type(&p, C_HEARTBEAT, HB_VERSION, (char *) &h, sizeof(h));
798
799 for (i = 0; i < config->cluster_num_changes; ++i) {
800 hb_add_type(&p, cluster_changes[i].type, cluster_changes[i].id);
801 }
802
803 if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
804 LOG(0, 0, 0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", (int) (p - buff));
805 kill(0, SIGTERM);
806 exit(1);
807 }
808
809 //
810 // Fill out the packet with sessions from the session table...
811 // (not forgetting to leave space so we can get some tunnels in too )
812 while ( (p + sizeof(uint32_t) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) {
813
814 if (!walk_session_number) // session #0 isn't valid.
815 ++walk_session_number;
816
817 if (count >= config->cluster_highest_sessionid) // If we're a small cluster, don't go wild.
818 break;
819
820 hb_add_type(&p, C_CSESSION, walk_session_number);
821 walk_session_number = (1+walk_session_number)%(config->cluster_highest_sessionid+1); // +1 avoids divide by zero.
822
823 ++count; // Count the number of extra sessions we're sending.
824 }
825
826 //
827 // Fill out the packet with tunnels from the tunnel table...
828 // This effectively means we walk the tunnel table more quickly
829 // than the session table. This is good because stuffing up a
830 // tunnel is a much bigger deal than stuffing up a session.
831 //
832 while ( (p + sizeof(uint32_t) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {
833
834 if (!walk_tunnel_number) // tunnel #0 isn't valid.
835 ++walk_tunnel_number;
836
837 if (tcount >= config->cluster_highest_tunnelid)
838 break;
839
840 hb_add_type(&p, C_CTUNNEL, walk_tunnel_number);
841 walk_tunnel_number = (1+walk_tunnel_number)%(config->cluster_highest_tunnelid+1); // +1 avoids divide by zero.
842
843 ++tcount;
844 }
845
846 //
847 // Did we do something wrong?
848 if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
849 LOG(0, 0, 0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", (int) (p - buff));
850 kill(0, SIGTERM);
851 exit(1);
852 }
853
854 LOG(3, 0, 0, "Sending v%d heartbeat #%d, change #%" PRIu64 " with %d changes "
855 "(%d x-sess, %d x-tunnels, %d highsess, %d hightun, size %d)\n",
856 HB_VERSION, h.seq, h.table_version, config->cluster_num_changes,
857 count, tcount, config->cluster_highest_sessionid,
858 config->cluster_highest_tunnelid, (int) (p - buff));
859
860 config->cluster_num_changes = 0;
861
862 send_heartbeat(h.seq, buff, (p-buff) ); // Send out the heartbeat to the cluster, keeping a copy of it.
863
864 config->cluster_seq_number = (config->cluster_seq_number+1)%HB_MAX_SEQ; // Next seq number to use.
865 }
866
867 //
868 // A structure of type 'type' has changed; Add it to the queue to send.
869 //
870 static int type_changed(int type, int id)
871 {
872 int i;
873
874 for (i = 0 ; i < config->cluster_num_changes ; ++i)
875 if ( cluster_changes[i].id == id &&
876 cluster_changes[i].type == type)
877 return 0; // Already marked for change.
878
879 cluster_changes[i].type = type;
880 cluster_changes[i].id = id;
881 ++config->cluster_num_changes;
882
883 if (config->cluster_num_changes > MAX_CHANGES)
884 cluster_heartbeat(); // flush now
885
886 return 1;
887 }
888
889
890 // A particular session has been changed!
891 int cluster_send_session(int sid)
892 {
893 if (!config->cluster_iam_master) {
894 LOG(0, sid, 0, "I'm not a master, but I just tried to change a session!\n");
895 return -1;
896 }
897
898 return type_changed(C_CSESSION, sid);
899 }
900
901 // A particular tunnel has been changed!
902 int cluster_send_tunnel(int tid)
903 {
904 if (!config->cluster_iam_master) {
905 LOG(0, 0, tid, "I'm not a master, but I just tried to change a tunnel!\n");
906 return -1;
907 }
908
909 return type_changed(C_CTUNNEL, tid);
910 }
911
912
913 //
914 // We're a master, and a slave has just told us that it's
915 // missed a packet. We'll resend it every packet since
916 // the last one it's seen.
917 //
918 static int cluster_catchup_slave(int seq, in_addr_t slave)
919 {
920 int s;
921 int diff;
922
923 LOG(1, 0, 0, "Slave %s sent LASTSEEN with seq %d\n", fmtaddr(slave, 0), seq);
924 if (!config->cluster_iam_master) {
925 LOG(1, 0, 0, "Got LASTSEEN but I'm not a master! Redirecting it to %s.\n",
926 fmtaddr(config->cluster_master_address, 0));
927
928 peer_send_message(slave, C_MASTER, config->cluster_master_address, NULL, 0);
929 return 0;
930 }
931
932 diff = config->cluster_seq_number - seq; // How many packet do we need to send?
933 if (diff < 0)
934 diff += HB_MAX_SEQ;
935
936 if (diff >= HB_HISTORY_SIZE) { // Ouch. We don't have the packet to send it!
937 LOG(0, 0, 0, "A slave asked for message %d when our seq number is %d. Killing it.\n",
938 seq, config->cluster_seq_number);
939 return peer_send_message(slave, C_KILL, seq, NULL, 0);// Kill the slave. Nothing else to do.
940 }
941
942 LOG(1, 0, 0, "Sending %d catchup packets to slave %s\n", diff, fmtaddr(slave, 0) );
943
944 // Now resend every packet that it missed, in order.
945 while (seq != config->cluster_seq_number) {
946 s = seq % HB_HISTORY_SIZE;
947 if (seq != past_hearts[s].seq) {
948 LOG(0, 0, 0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n",
949 fmtaddr(slave, 0), seq, past_hearts[s].seq, s, config->cluster_seq_number);
950 return -1; // What to do here!?
951 }
952 peer_send_data(slave, past_hearts[s].data, past_hearts[s].size);
953 seq = (seq+1)%HB_MAX_SEQ; // Increment to next seq number.
954 }
955 return 0; // All good!
956 }
957
958 //
959 // We've heard from another peer! Add it to the list
960 // that we select from at election time.
961 //
962 static int cluster_add_peer(in_addr_t peer, time_t basetime, pingt *pp, int size)
963 {
964 int i;
965 in_addr_t clusterid;
966 pingt p;
967
968 // Allow for backward compatability.
969 // Just the ping packet into a new structure to allow
970 // for the possibility that we might have received
971 // more or fewer elements than we were expecting.
972 if (size > sizeof(p))
973 size = sizeof(p);
974
975 memset( (void *) &p, 0, sizeof(p) );
976 memcpy( (void *) &p, (void *) pp, size);
977
978 clusterid = p.addr;
979 if (clusterid != config->bind_address)
980 {
981 // Is this for us?
982 LOG(4, 0, 0, "Skipping ping from %s (different cluster)\n", fmtaddr(peer, 0));
983 return 0;
984 }
985
986 for (i = 0; i < num_peers ; ++i)
987 {
988 if (peers[i].peer != peer)
989 continue;
990
991 // This peer already exists. Just update the timestamp.
992 peers[i].basetime = basetime;
993 peers[i].timestamp = TIME;
994 peers[i].uptodate = !p.undef;
995 break;
996 }
997
998 // Is this the master shutting down??
999 if (peer == config->cluster_master_address) {
1000 LOG(3, 0, 0, "Master %s %s\n", fmtaddr(config->cluster_master_address, 0),
1001 basetime ? "has restarted!" : "shutting down...");
1002
1003 config->cluster_master_address = 0;
1004 config->cluster_last_hb = 0; // Force an election.
1005 cluster_check_master();
1006 return 0;
1007 }
1008
1009 if (i >= num_peers)
1010 {
1011 LOG(4, 0, 0, "Adding %s as a peer\n", fmtaddr(peer, 0));
1012
1013 // Not found. Is there a stale slot to re-use?
1014 for (i = 0; i < num_peers ; ++i)
1015 {
1016 if (!peers[i].basetime) // Shutdown
1017 break;
1018
1019 if ((peers[i].timestamp + config->cluster_hb_timeout * 10) < TIME) // Stale.
1020 break;
1021 }
1022
1023 if (i >= CLUSTER_MAX_SIZE)
1024 {
1025 // Too many peers!!
1026 LOG(0, 0, 0, "Tried to add %s as a peer, but I already have %d of them!\n", fmtaddr(peer, 0), i);
1027 return -1;
1028 }
1029
1030 peers[i].peer = peer;
1031 peers[i].basetime = basetime;
1032 peers[i].timestamp = TIME;
1033 peers[i].uptodate = !p.undef;
1034 if (i == num_peers)
1035 ++num_peers;
1036
1037 LOG(1, 0, 0, "Added %s as a new peer. Now %d peers\n", fmtaddr(peer, 0), num_peers);
1038 }
1039
1040 return 1;
1041 }
1042
1043 // A slave responds with C_MASTER when it gets a message which should have gone to a master.
1044 static int cluster_set_master(in_addr_t peer, in_addr_t master)
1045 {
1046 if (config->cluster_iam_master) // Sanity...
1047 return 0;
1048
1049 LOG(3, 0, 0, "Peer %s set the master to %s...\n", fmtaddr(peer, 0),
1050 fmtaddr(master, 1));
1051
1052 config->cluster_master_address = master;
1053 cluster_check_master();
1054 return 0;
1055 }
1056
1057 /* Handle the slave updating the byte counters for the master. */
1058 //
1059 // Note that we don't mark the session as dirty; We rely on
1060 // the slow table walk to propogate this back out to the slaves.
1061 //
1062 static int cluster_handle_bytes(char *data, int size)
1063 {
1064 bytest *b;
1065
1066 b = (bytest *) data;
1067
1068 LOG(3, 0, 0, "Got byte counter update (size %d)\n", size);
1069
1070 /* Loop around, adding the byte
1071 counts to each of the sessions. */
1072
1073 while (size >= sizeof(*b) ) {
1074 if (b->sid > MAXSESSION) {
1075 LOG(0, 0, 0, "Got C_BYTES with session #%d!\n", b->sid);
1076 return -1; /* Abort processing */
1077 }
1078
1079 session[b->sid].total_cin += b->in;
1080 session[b->sid].total_cout += b->out;
1081
1082 session[b->sid].cin += b->in;
1083 session[b->sid].cout += b->out;
1084
1085 if (b->in)
1086 session[b->sid].last_packet = time_now; // Reset idle timer!
1087
1088 size -= sizeof(*b);
1089 ++b;
1090 }
1091
1092 if (size != 0)
1093 LOG(0, 0, 0, "Got C_BYTES with %d bytes of trailing junk!\n", size);
1094
1095 return size;
1096 }
1097
1098 //
1099 // Handle receiving a session structure in a heartbeat packet.
1100 //
1101 static int cluster_recv_session(int more, uint8_t *p)
1102 {
1103 if (more >= MAXSESSION) {
1104 LOG(0, 0, 0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
1105 return -1;
1106 }
1107
1108 if (session[more].tunnel == T_UNDEF) {
1109 if (config->cluster_iam_uptodate) { // Sanity.
1110 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined session!\n");
1111 } else {
1112 --config->cluster_undefined_sessions;
1113 }
1114 }
1115
1116 load_session(more, (sessiont *) p); // Copy session into session table..
1117
1118 LOG(5, more, 0, "Received session update (%d undef)\n", config->cluster_undefined_sessions);
1119
1120 if (!config->cluster_iam_uptodate)
1121 cluster_uptodate(); // Check to see if we're up to date.
1122
1123 return 0;
1124 }
1125
1126 static int cluster_recv_tunnel(int more, uint8_t *p)
1127 {
1128 if (more >= MAXTUNNEL) {
1129 LOG(0, 0, 0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
1130 return -1;
1131 }
1132
1133 if (tunnel[more].state == TUNNELUNDEF) {
1134 if (config->cluster_iam_uptodate) { // Sanity.
1135 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined tunnel!\n");
1136 } else {
1137 --config->cluster_undefined_tunnels;
1138 }
1139 }
1140
1141 memcpy(&tunnel[more], p, sizeof(tunnel[more]) );
1142
1143 //
1144 // Clear tunnel control messages. These are dynamically allocated.
1145 // If we get unlucky, this may cause the tunnel to drop!
1146 //
1147 tunnel[more].controls = tunnel[more].controle = NULL;
1148 tunnel[more].controlc = 0;
1149
1150 LOG(5, 0, more, "Received tunnel update\n");
1151
1152 if (!config->cluster_iam_uptodate)
1153 cluster_uptodate(); // Check to see if we're up to date.
1154
1155 return 0;
1156 }
1157
1158
1159 //
1160 // Process a heartbeat..
1161 //
1162 // v3: added interval, timeout
1163 // v4: added table_version
1164 static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t *p, in_addr_t addr)
1165 {
1166 heartt *h;
1167 int s = size - (p-data);
1168 int i, type;
1169
1170 #if HB_VERSION != 4
1171 # error "need to update cluster_process_heartbeat()"
1172 #endif
1173
1174 // we handle versions 3 through 4
1175 if (more < 3 || more > HB_VERSION) {
1176 LOG(0, 0, 0, "Received a heartbeat version that I don't support (%d)!\n", more);
1177 return -1; // Ignore it??
1178 }
1179
1180 // Ok. It's a heartbeat packet from a cluster master!
1181 if (s < sizeof(*h))
1182 goto shortpacket;
1183
1184 h = (heartt *) p;
1185 p += sizeof(*h);
1186 s -= sizeof(*h);
1187
1188 if (h->clusterid != config->bind_address)
1189 return -1; // It's not part of our cluster.
1190
1191 if (config->cluster_iam_master) { // Sanity...
1192 // Note that this MUST match the election process above!
1193
1194 LOG(0, 0, 0, "I just got a heartbeat from master %s, but _I_ am the master!\n", fmtaddr(addr, 0));
1195 if (!h->basetime) {
1196 LOG(0, 0, 0, "Heartbeat with zero basetime! Ignoring\n");
1197 return -1; // Skip it.
1198 }
1199
1200 if (more >= 4) {
1201 if (h->table_version > config->cluster_table_version) {
1202 LOG(0, 0, 0, "They've seen more state changes (%" PRIu64 " vs my %" PRIu64 ") so I'm gone!\n",
1203 h->table_version, config->cluster_table_version);
1204
1205 kill(0, SIGTERM);
1206 exit(1);
1207 }
1208 if (h->table_version < config->cluster_table_version)
1209 return -1;
1210 }
1211
1212 if (basetime > h->basetime) {
1213 LOG(0, 0, 0, "They're an older master than me so I'm gone!\n");
1214 kill(0, SIGTERM);
1215 exit(1);
1216 }
1217
1218 if (basetime < h->basetime)
1219 return -1;
1220
1221 if (my_address < addr) { // Tie breaker.
1222 LOG(0, 0, 0, "They're a higher IP address than me, so I'm gone!\n");
1223 kill(0, SIGTERM);
1224 exit(1);
1225 }
1226
1227 //
1228 // Send it a unicast heartbeat to see give it a chance to die.
1229 // NOTE: It's actually safe to do seq-number - 1 without checking
1230 // for wrap around.
1231 //
1232 cluster_catchup_slave(config->cluster_seq_number - 1, addr);
1233
1234 return -1; // Skip it.
1235 }
1236
1237 //
1238 // Try and guard against a stray master appearing.
1239 //
1240 // Ignore heartbeats received from another master before the
1241 // timeout (less a smidgen) for the old master has elapsed.
1242 //
1243 // Note that after a clean failover, the cluster_master_address
1244 // is cleared, so this doesn't run.
1245 //
1246 if (config->cluster_master_address && addr != config->cluster_master_address
1247 && (config->cluster_last_hb + config->cluster_hb_timeout - 11) > TIME) {
1248 LOG(0, 0, 0, "Ignoring stray heartbeat from %s, current master %s has not yet timed out (last heartbeat %.1f seconds ago).\n",
1249 fmtaddr(addr, 0), fmtaddr(config->cluster_master_address, 1),
1250 0.1 * (TIME - config->cluster_last_hb));
1251 return -1; // ignore
1252 }
1253
1254 if (config->cluster_seq_number == -1) // Don't have one. Just align to the master...
1255 config->cluster_seq_number = h->seq;
1256
1257 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1258
1259 if (config->cluster_seq_number != h->seq) { // Out of sequence heartbeat!
1260 static int lastseen_seq = 0;
1261 static time_t lastseen_time = 0;
1262
1263 // limit to once per second for a particular seq#
1264 int ask = (config->cluster_seq_number != lastseen_seq || time_now != lastseen_time);
1265
1266 LOG(1, 0, 0, "HB: Got seq# %d but was expecting %d. %s.\n",
1267 h->seq, config->cluster_seq_number,
1268 ask ? "Asking for resend" : "Ignoring");
1269
1270 if (ask)
1271 {
1272 lastseen_seq = config->cluster_seq_number;
1273 lastseen_time = time_now;
1274 peer_send_message(addr, C_LASTSEEN, config->cluster_seq_number, NULL, 0);
1275 }
1276
1277 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1278
1279 // Just drop the packet. The master will resend it as part of the catchup.
1280
1281 return 0;
1282 }
1283 // Save the packet in our buffer.
1284 // This is needed in case we become the master.
1285 config->cluster_seq_number = (h->seq+1)%HB_MAX_SEQ;
1286 i = h->seq % HB_HISTORY_SIZE;
1287 past_hearts[i].seq = h->seq;
1288 past_hearts[i].size = size;
1289 memcpy(&past_hearts[i].data, data, size); // Save it.
1290
1291
1292 // Check that we don't have too many undefined sessions, and
1293 // that the free session pointer is correct.
1294 cluster_check_sessions(h->highsession, h->freesession, h->hightunnel);
1295
1296 if (h->interval != config->cluster_hb_interval)
1297 {
1298 LOG(2, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
1299 h->interval, config->cluster_hb_interval);
1300
1301 config->cluster_hb_interval = h->interval;
1302 }
1303
1304 if (h->timeout != config->cluster_hb_timeout)
1305 {
1306 LOG(2, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
1307 h->timeout, config->cluster_hb_timeout);
1308
1309 config->cluster_hb_timeout = h->timeout;
1310 }
1311
1312 // Ok. process the packet...
1313 while ( s > 0) {
1314
1315 type = *((uint32_t *) p);
1316 p += sizeof(uint32_t);
1317 s -= sizeof(uint32_t);
1318
1319 more = *((uint32_t *) p);
1320 p += sizeof(uint32_t);
1321 s -= sizeof(uint32_t);
1322
1323 switch (type) {
1324 case C_CSESSION: { // Compressed session structure.
1325 uint8_t c[ sizeof(sessiont) + 2];
1326 int size;
1327 uint8_t *orig_p = p;
1328
1329 size = rle_decompress((uint8_t **) &p, s, c, sizeof(c) );
1330 s -= (p - orig_p);
1331
1332 if (size != sizeof(sessiont) ) { // Ouch! Very very bad!
1333 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1334 // Now what? Should exit! No-longer up to date!
1335 break;
1336 }
1337
1338 cluster_recv_session(more, c);
1339 break;
1340 }
1341 case C_SESSION:
1342 if ( s < sizeof(session[more]))
1343 goto shortpacket;
1344
1345 cluster_recv_session(more, p);
1346
1347 p += sizeof(session[more]);
1348 s -= sizeof(session[more]);
1349 break;
1350
1351 case C_CTUNNEL: { // Compressed tunnel structure.
1352 uint8_t c[ sizeof(tunnelt) + 2];
1353 int size;
1354 uint8_t *orig_p = p;
1355
1356 size = rle_decompress((uint8_t **) &p, s, c, sizeof(c));
1357 s -= (p - orig_p);
1358
1359 if (size != sizeof(tunnelt) ) { // Ouch! Very very bad!
1360 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1361 // Now what? Should exit! No-longer up to date!
1362 break;
1363 }
1364
1365 cluster_recv_tunnel(more, c);
1366 break;
1367
1368 }
1369 case C_TUNNEL:
1370 if ( s < sizeof(tunnel[more]))
1371 goto shortpacket;
1372
1373 cluster_recv_tunnel(more, p);
1374
1375 p += sizeof(tunnel[more]);
1376 s -= sizeof(tunnel[more]);
1377 break;
1378 default:
1379 LOG(0, 0, 0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type);
1380 return -1; // can't process any more of the packet!!
1381 }
1382 }
1383
1384 if (config->cluster_master_address != addr)
1385 {
1386 LOG(0, 0, 0, "My master just changed from %s to %s!\n",
1387 fmtaddr(config->cluster_master_address, 0), fmtaddr(addr, 1));
1388
1389 config->cluster_master_address = addr;
1390 }
1391
1392 config->cluster_last_hb = TIME; // Successfully received a heartbeat!
1393 config->cluster_table_version = h->table_version;
1394 return 0;
1395
1396 shortpacket:
1397 LOG(0, 0, 0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
1398 return -1;
1399 }
1400
1401 //
1402 // We got a packet on the cluster port!
1403 // Handle pings, lastseens, and heartbeats!
1404 //
1405 int processcluster(char *data, int size, in_addr_t addr)
1406 {
1407 int type, more;
1408 char *p = data;
1409 int s = size;
1410
1411 if (addr == my_address)
1412 return -1; // Ignore it. Something looped back the multicast!
1413
1414 LOG(5, 0, 0, "Process cluster: %d bytes from %s\n", size, fmtaddr(addr, 0));
1415
1416 if (s <= 0) // Any data there??
1417 return -1;
1418
1419 if (s < 8)
1420 goto shortpacket;
1421
1422 type = *((uint32_t *) p);
1423 p += sizeof(uint32_t);
1424 s -= sizeof(uint32_t);
1425
1426 more = *((uint32_t *) p);
1427 p += sizeof(uint32_t);
1428 s -= sizeof(uint32_t);
1429
1430 switch (type) {
1431 case C_PING: // Update the peers table.
1432 return cluster_add_peer(addr, more, (pingt *) p, s);
1433
1434 case C_MASTER: // Our master is wrong
1435 return cluster_set_master(addr, more);
1436
1437 case C_LASTSEEN: // Catch up a slave (slave missed a packet).
1438 return cluster_catchup_slave(more, addr);
1439
1440 case C_FORWARD: { // Forwarded control packet. pass off to processudp.
1441 struct sockaddr_in a;
1442 a.sin_addr.s_addr = more;
1443
1444 a.sin_port = *(int *) p;
1445 s -= sizeof(int);
1446 p += sizeof(int);
1447
1448 if (!config->cluster_iam_master) { // huh?
1449 LOG(0, 0, 0, "I'm not the master, but I got a C_FORWARD from %s?\n", fmtaddr(addr, 0));
1450 return -1;
1451 }
1452
1453 LOG(4, 0, 0, "Got a forwarded packet... (%s:%d)\n", fmtaddr(more, 0), a.sin_port);
1454 STAT(recv_forward);
1455 processudp(p, s, &a);
1456 return 0;
1457 }
1458 case C_THROTTLE: { // Receive a forwarded packet from a slave.
1459 if (!config->cluster_iam_master) {
1460 LOG(0, 0, 0, "I'm not the master, but I got a C_THROTTLE from %s?\n", fmtaddr(addr, 0));
1461 return -1;
1462 }
1463
1464 tbf_queue_packet(more, p, s); // The TBF id tells wether it goes in or out.
1465 return 0;
1466 }
1467 case C_GARDEN:
1468 // Receive a walled garden packet from a slave.
1469 if (!config->cluster_iam_master) {
1470 LOG(0, 0, 0, "I'm not the master, but I got a C_GARDEN from %s?\n", fmtaddr(addr, 0));
1471 return -1;
1472 }
1473
1474 tun_write(p, s);
1475 return 0;
1476
1477 case C_BYTES:
1478 if (!config->cluster_iam_master) {
1479 LOG(0, 0, 0, "I'm not the master, but I got a C_BYTES from %s?\n", fmtaddr(addr, 0));
1480 return -1;
1481 }
1482
1483 return cluster_handle_bytes(p, s);
1484
1485 case C_KILL: // The master asked us to die!? (usually because we're too out of date).
1486 if (config->cluster_iam_master) {
1487 LOG(0, 0, 0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", fmtaddr(addr, 0), more);
1488 return -1;
1489 }
1490 if (more != config->cluster_seq_number) {
1491 LOG(0, 0, 0, "The master asked us to die but the seq number didn't match!?\n");
1492 return -1;
1493 }
1494
1495 if (addr != config->cluster_master_address) {
1496 LOG(0, 0, 0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%s)\n",
1497 fmtaddr(addr, 0), fmtaddr(config->cluster_master_address, 1));
1498 // We can only warn about it. The master might really have switched!
1499 }
1500
1501 LOG(0, 0, 0, "Received a valid C_KILL: I'm going to die now.\n");
1502 kill(0, SIGTERM);
1503 exit(0); // Lets be paranoid;
1504 return -1; // Just signalling the compiler.
1505
1506 case C_HEARTBEAT:
1507 LOG(4, 0, 0, "Got a heartbeat from %s\n", fmtaddr(addr, 0));
1508 return cluster_process_heartbeat(data, size, more, p, addr);
1509
1510 default:
1511 LOG(0, 0, 0, "Strange type packet received on cluster socket (%d)\n", type);
1512 return -1;
1513 }
1514 return 0;
1515
1516 shortpacket:
1517 LOG(0, 0, 0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n");
1518 return -1;
1519 }
1520
1521 //====================================================================================================
1522
1523 int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
1524 {
1525 int i;
1526
1527 if (CLI_HELP_REQUESTED)
1528 return CLI_HELP_NO_ARGS;
1529
1530 cli_print(cli, "Cluster status : %s", config->cluster_iam_master ? "Master" : "Slave" );
1531 cli_print(cli, "My address : %s", fmtaddr(my_address, 0));
1532 cli_print(cli, "VIP address : %s", fmtaddr(config->bind_address, 0));
1533 cli_print(cli, "Multicast address: %s", fmtaddr(config->cluster_address, 0));
1534 cli_print(cli, "Multicast i'face : %s", config->cluster_interface);
1535
1536 if (!config->cluster_iam_master) {
1537 cli_print(cli, "My master : %s (last heartbeat %.1f seconds old)",
1538 config->cluster_master_address
1539 ? fmtaddr(config->cluster_master_address, 0)
1540 : "Not defined",
1541 0.1 * (TIME - config->cluster_last_hb));
1542 cli_print(cli, "Uptodate : %s", config->cluster_iam_uptodate ? "Yes" : "No");
1543 cli_print(cli, "Table version # : %" PRIu64, config->cluster_table_version);
1544 cli_print(cli, "Next sequence number expected: %d", config->cluster_seq_number);
1545 cli_print(cli, "%d sessions undefined of %d", config->cluster_undefined_sessions, config->cluster_highest_sessionid);
1546 cli_print(cli, "%d tunnels undefined of %d", config->cluster_undefined_tunnels, config->cluster_highest_tunnelid);
1547 } else {
1548 cli_print(cli, "Table version # : %" PRIu64, config->cluster_table_version);
1549 cli_print(cli, "Next heartbeat # : %d", config->cluster_seq_number);
1550 cli_print(cli, "Highest session : %d", config->cluster_highest_sessionid);
1551 cli_print(cli, "Highest tunnel : %d", config->cluster_highest_tunnelid);
1552 cli_print(cli, "%d changes queued for sending", config->cluster_num_changes);
1553 }
1554 cli_print(cli, "%d peers.", num_peers);
1555
1556 if (num_peers)
1557 cli_print(cli, "%20s %10s %8s", "Address", "Basetime", "Age");
1558 for (i = 0; i < num_peers; ++i) {
1559 cli_print(cli, "%20s %10u %8d", fmtaddr(peers[i].peer, 0),
1560 peers[i].basetime, TIME - peers[i].timestamp);
1561 }
1562 return CLI_OK;
1563 }
1564
1565 //
1566 // Simple run-length-encoding compression.
1567 // Format is
1568 // 1 byte < 128 = count of non-zero bytes following. // Not legal to be zero.
1569 // n non-zero bytes;
1570 // or
1571 // 1 byte > 128 = (count - 128) run of zero bytes. //
1572 // repeat.
1573 // count == 0 indicates end of compressed stream.
1574 //
1575 // Compress from 'src' into 'dst'. return number of bytes
1576 // used from 'dst'.
1577 // Updates *src_p to indicate 1 past last bytes used.
1578 //
1579 // We could get an extra byte in the zero runs by storing (count-1)
1580 // but I'm playing it safe.
1581 //
1582 // Worst case is a 50% expansion in space required (trying to
1583 // compress { 0x00, 0x01 } * N )
1584 static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
1585 {
1586 int count;
1587 int orig_dsize = dsize;
1588 uint8_t *x, *src;
1589 src = *src_p;
1590
1591 while (ssize > 0 && dsize > 2) {
1592 count = 0;
1593 x = dst++; --dsize; // Reserve space for count byte..
1594
1595 if (*src) { // Copy a run of non-zero bytes.
1596 while (*src && count < 127 && ssize > 0 && dsize > 1) { // Count number of non-zero bytes.
1597 *dst++ = *src++;
1598 --dsize; --ssize;
1599 ++count;
1600 }
1601 *x = count; // Store number of non-zero bytes. Guarenteed to be non-zero!
1602
1603 } else { // Compress a run of zero bytes.
1604 while (*src == 0 && count < 127 && ssize > 0) {
1605 ++src;
1606 --ssize;
1607 ++count;
1608 }
1609 *x = count | 0x80 ;
1610 }
1611 }
1612
1613 *dst++ = 0x0; // Add Stop byte.
1614 --dsize;
1615
1616 *src_p = src;
1617 return (orig_dsize - dsize);
1618 }
1619
1620 //
1621 // Decompress the buffer into **p.
1622 // 'psize' is the size of the decompression buffer available.
1623 //
1624 // Returns the number of bytes decompressed.
1625 //
1626 // Decompresses from '*src_p' into 'dst'.
1627 // Return the number of dst bytes used.
1628 // Updates the 'src_p' pointer to point to the
1629 // first un-used byte.
1630 static int rle_decompress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
1631 {
1632 int count;
1633 int orig_dsize = dsize;
1634 char *src = *src_p;
1635
1636 while (ssize >0 && dsize > 0) { // While there's more to decompress, and there's room in the decompress buffer...
1637 count = *src++; --ssize; // get the count byte from the source.
1638 if (count == 0x0) // End marker reached? If so, finish.
1639 break;
1640
1641 if (count & 0x80) { // Decompress a run of zeros
1642 for (count &= 0x7f ; count > 0 && dsize > 0; --count) {
1643 *dst++ = 0x0;
1644 --dsize;
1645 }
1646 } else { // Copy run of non-zero bytes.
1647 for ( ; count > 0 && ssize && dsize; --count) { // Copy non-zero bytes across.
1648 *dst++ = *src++;
1649 --ssize; --dsize;
1650 }
1651 }
1652 }
1653 *src_p = src;
1654 return (orig_dsize - dsize);
1655 }