Any ping from what we think is the master should force an election.
[l2tpns.git] / cluster.c
1 // L2TPNS Clustering Stuff
2
3 char const *cvs_id_cluster = "$Id: cluster.c,v 1.26.2.7 2005/05/22 04:15:32 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(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 static void cluster_uptodate(void)
207 {
208 if (config->cluster_iam_uptodate)
209 return;
210
211 if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels)
212 return;
213
214 config->cluster_iam_uptodate = 1;
215
216 LOG(0, 0, 0, "Now uptodate with master.\n");
217 advertise();
218 }
219
220 //
221 // Send a unicast UDP packet to a peer with 'data' as the
222 // contents.
223 //
224 static int peer_send_data(in_addr_t peer, char *data, int size)
225 {
226 struct sockaddr_in addr = {0};
227
228 if (!cluster_sockfd) return -1;
229 if (!config->cluster_address) return 0;
230
231 if (!peer) // Odd??
232 return -1;
233
234 addr.sin_addr.s_addr = peer;
235 addr.sin_port = htons(CLUSTERPORT);
236 addr.sin_family = AF_INET;
237
238 LOG_HEX(5, "Peer send", data, size);
239
240 if (sendto(cluster_sockfd, data, size, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
241 {
242 LOG(0, 0, 0, "sendto: %s\n", strerror(errno));
243 return -1;
244 }
245
246 return 0;
247 }
248
249 //
250 // Send a structured message to a peer with a single element of type 'type'.
251 //
252 static int peer_send_message(in_addr_t peer, int type, int more, char *data, int size)
253 {
254 char buf[65536]; // Vast overkill.
255 char *p = buf;
256
257 LOG(4, 0, 0, "Sending message to peer (type %d, more %d, size %d)\n", type, more, size);
258 add_type(&p, type, more, data, size);
259
260 return peer_send_data(peer, buf, (p-buf) );
261 }
262
263 //
264 // Forward a state changing packet to the master.
265 //
266 // The master just processes the payload as if it had
267 // received it off the tun device.
268 //
269 int master_forward_packet(char *data, int size, in_addr_t addr, int port)
270 {
271 char buf[65536]; // Vast overkill.
272 char *p = buf;
273
274 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
275 return -1;
276
277 LOG(4, 0, 0, "Forwarding packet from %s to master (size %d)\n", fmtaddr(addr, 0), size);
278
279 STAT(c_forwarded);
280 add_type(&p, C_FORWARD, addr, (char *) &port, sizeof(port));
281 memcpy(p, data, size);
282 p += size;
283
284 return peer_send_data(config->cluster_master_address, buf, (p - buf));
285 }
286
287 //
288 // Forward a throttled packet to the master for handling.
289 //
290 // The master just drops the packet into the appropriate
291 // token bucket queue, and lets normal processing take care
292 // of it.
293 //
294 int master_throttle_packet(int tbfid, char *data, int size)
295 {
296 char buf[65536]; // Vast overkill.
297 char *p = buf;
298
299 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
300 return -1;
301
302 LOG(4, 0, 0, "Throttling packet master (size %d, tbfid %d)\n", size, tbfid);
303
304 add_type(&p, C_THROTTLE, tbfid, data, size);
305
306 return peer_send_data(config->cluster_master_address, buf, (p-buf) );
307
308 }
309
310 //
311 // Forward a walled garden packet to the master for handling.
312 //
313 // The master just writes the packet straight to the tun
314 // device (where is will normally loop through the
315 // firewall rules, and come back in on the tun device)
316 //
317 // (Note that this must be called with the tun header
318 // as the start of the data).
319 int master_garden_packet(sessionidt s, char *data, int size)
320 {
321 char buf[65536]; // Vast overkill.
322 char *p = buf;
323
324 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
325 return -1;
326
327 LOG(4, 0, 0, "Walled garden packet to master (size %d)\n", size);
328
329 add_type(&p, C_GARDEN, s, data, size);
330
331 return peer_send_data(config->cluster_master_address, buf, (p-buf));
332
333 }
334
335 //
336 // Send a chunk of data as a heartbeat..
337 // We save it in the history buffer as we do so.
338 //
339 static void send_heartbeat(int seq, char *data, int size)
340 {
341 int i;
342
343 if (size > sizeof(past_hearts[0].data))
344 {
345 LOG(0, 0, 0, "Tried to heartbeat something larger than the maximum packet!\n");
346 kill(0, SIGTERM);
347 exit(1);
348 }
349 i = seq % HB_HISTORY_SIZE;
350 past_hearts[i].seq = seq;
351 past_hearts[i].size = size;
352 memcpy(&past_hearts[i].data, data, size); // Save it.
353 cluster_send_data(data, size);
354 }
355
356 //
357 // Send an 'i am alive' message to every machine in the cluster, or to a single peer
358 //
359 static void send_ping(time_t basetime, in_addr_t peer)
360 {
361 char buff[100 + sizeof(pingt)];
362 char *p = buff;
363 pingt x;
364
365 if (config->cluster_iam_master && basetime) // We're heartbeating so no need to ping.
366 return;
367
368 x.ver = 1;
369 x.addr = config->bind_address;
370 x.undef = config->cluster_undefined_sessions + config->cluster_undefined_tunnels;
371 x.basetime = basetime;
372
373 add_type(&p, C_PING, basetime, (char *) &x, sizeof(x));
374
375 if (peer)
376 peer_send_data(peer, buff, (p-buff));
377 else
378 cluster_send_data(buff, (p-buff) );
379 }
380
381 void cluster_send_ping(time_t basetime)
382 {
383 LOG(5, 0, 0, "Sending cluster ping...\n");
384 send_ping(0, basetime);
385 }
386
387 void peer_send_ping(in_addr_t peer, time_t basetime)
388 {
389 LOG(5, 0, 0, "Sending unicast ping to %s...\n", fmtaddr(peer, 0));
390 send_ping(peer, basetime);
391 }
392
393 //
394 // Walk the session counters looking for non-zero ones to send
395 // to the master. We send up to 600 of them at one time.
396 // We examine a maximum of 3000 sessions.
397 // (50k max session should mean that we normally
398 // examine the entire session table every 25 seconds).
399
400 #define MAX_B_RECS (600)
401 void master_update_counts(void)
402 {
403 int i, c;
404 bytest b[MAX_B_RECS+1];
405
406 if (config->cluster_iam_master) // Only happens on the slaves.
407 return;
408
409 if (!config->cluster_master_address) // If we don't have a master, skip it for a while.
410 return;
411
412 i = MAX_B_RECS * 5; // Examine max 2000 sessions;
413 if (config->cluster_highest_sessionid > i)
414 i = config->cluster_highest_sessionid;
415
416 for ( c = 0; i > 0 ; --i) {
417 // Next session to look at.
418 walk_session_number++;
419 if ( walk_session_number > config->cluster_highest_sessionid)
420 walk_session_number = 1;
421
422 if (!sess_local[walk_session_number].cin && !sess_local[walk_session_number].cout)
423 continue; // Unused. Skip it.
424
425 b[c].sid = walk_session_number;
426 b[c].in = sess_local[walk_session_number].cin;
427 b[c].out = sess_local[walk_session_number].cout;
428
429 if (++c > MAX_B_RECS) // Send a max of 400 elements in a packet.
430 break;
431
432 // Reset counters.
433 sess_local[walk_session_number].cin = sess_local[walk_session_number].cout = 0;
434 }
435
436 if (!c) // Didn't find any that changes. Get out of here!
437 return;
438
439
440 // Forward the data to the master.
441 LOG(4, 0, 0, "Sending byte counters to master (%d elements)\n", c);
442 peer_send_message(config->cluster_master_address, C_BYTES, c, (char *) &b, sizeof(b[0]) * c);
443 return;
444 }
445
446 //
447 // On the master, check how our slaves are going. If
448 // one of them's not up-to-date we'll heartbeat faster.
449 // If we don't have any of them, then we need to turn
450 // on our own packet handling!
451 //
452 void cluster_check_slaves(void)
453 {
454 int i;
455 static int have_peers = 0;
456 int had_peers = have_peers;
457 clockt t = TIME;
458
459 if (!config->cluster_iam_master)
460 return; // Only runs on the master...
461
462 config->cluster_iam_uptodate = 1; // cleared in loop below
463
464 for (i = have_peers = 0; i < num_peers; i++)
465 {
466 if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
467 continue; // Stale peer! Skip them.
468
469 if (!peers[i].basetime)
470 continue; // Shutdown peer! Skip them.
471
472 if (peers[i].uptodate)
473 have_peers = 1;
474
475 if (!peers[i].uptodate)
476 config->cluster_iam_uptodate = 0; // Start fast heartbeats
477 }
478
479 #ifdef BGP
480 // in a cluster, withdraw/add routes when we get a peer/lose all peers
481 if (bgp_configured && have_peers != had_peers)
482 bgp_enable_routing(!have_peers);
483 #endif /* BGP */
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
497 if (config->cluster_iam_master)
498 return; // Only runs on the slaves...
499
500 // If the master is late (missed 2 hearbeats by a second and a
501 // hair) it may be that the switch has dropped us from the
502 // multicast group, try unicasting probes to the master
503 // which will hopefully respond with a unicast heartbeat that
504 // will allow us to limp along until the querier next runs.
505 if (config->cluster_master_address
506 && TIME > (config->cluster_last_hb + 2 * config->cluster_hb_interval + 11))
507 {
508 if (!probed || (TIME > (probed + 2 * config->cluster_hb_interval)))
509 {
510 probed = TIME;
511 LOG(1, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
512 0.1 * (TIME - (config->cluster_last_hb + config->cluster_hb_interval)));
513
514 peer_send_message(config->cluster_master_address,
515 C_LASTSEEN, config->cluster_seq_number, NULL, 0);
516 }
517 } else { // We got a recent heartbeat; reset the probe flag.
518 probed = 0;
519 }
520
521 if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout))
522 return; // Everything's ok!
523
524 config->cluster_last_hb = TIME + 1; // Just the one election thanks.
525
526 LOG(0, 0, 0, "Master timed out! Holding election...\n");
527
528 for (i = 0; i < num_peers; i++)
529 {
530 if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
531 continue; // Stale peer! Skip them.
532
533 if (!peers[i].basetime)
534 continue; // Shutdown peer! Skip them.
535
536 if (peers[i].basetime < basetime) {
537 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
538 return; // They'll win the election. Get out of here.
539 }
540
541 if (peers[i].basetime == basetime &&
542 peers[i].peer > my_address) {
543 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
544 return; // They'll win the election. Wait for them to come up.
545 }
546 }
547
548 // Wow. it's been ages since I last heard a heartbeat
549 // and I'm better than an of my peers so it's time
550 // to become a master!!!
551
552 config->cluster_iam_master = 1;
553 config->cluster_master_address = 0;
554
555 LOG(0, 0, 0, "I am declaring myself the master!\n");
556
557 if (config->cluster_seq_number == -1)
558 config->cluster_seq_number = 0;
559
560 //
561 // Go through and mark all the tunnels as defined.
562 // Count the highest used tunnel number as well.
563 //
564 config->cluster_highest_tunnelid = 0;
565 for (i = 0, tcount = 0; i < MAXTUNNEL; ++i) {
566 if (tunnel[i].state == TUNNELUNDEF)
567 tunnel[i].state = TUNNELFREE;
568
569 if (tunnel[i].state != TUNNELFREE && i > config->cluster_highest_tunnelid)
570 config->cluster_highest_tunnelid = i;
571 }
572
573 //
574 // Go through and mark all the sessions as being defined.
575 // reset the idle timeouts.
576 // add temporary byte counters to permanent ones.
577 // Re-string the free list.
578 // Find the ID of the highest session.
579 last_free = 0;
580 high_unique_id = 0;
581 config->cluster_highest_sessionid = 0;
582 for (i = 0, count = 0; i < MAXSESSION; ++i) {
583 if (session[i].tunnel == T_UNDEF) {
584 session[i].tunnel = T_FREE;
585 ++count;
586 }
587
588 if (!session[i].opened) { // Unused session. Add to free list.
589 memset(&session[i], 0, sizeof(session[i]));
590 session[i].tunnel = T_FREE;
591 session[last_free].next = i;
592 session[i].next = 0;
593 last_free = i;
594 continue;
595 }
596
597 // Reset idle timeouts..
598 session[i].last_packet = time_now;
599
600 // Reset die relative to our uptime rather than the old master's
601 if (session[i].die) session[i].die = TIME;
602
603 // Accumulate un-sent byte counters.
604 session[i].cin += sess_local[i].cin;
605 session[i].cout += sess_local[i].cout;
606 session[i].total_cin += sess_local[i].cin;
607 session[i].total_cout += sess_local[i].cout;
608
609 sess_local[i].cin = sess_local[i].cout = 0;
610
611 session[i].radius = 0; // Reset authentication as the radius blocks aren't up to date.
612
613 if (session[i].unique_id >= high_unique_id) // This is different to the index into the session table!!!
614 high_unique_id = session[i].unique_id+1;
615
616 session[i].tbf_in = session[i].tbf_out = 0; // Remove stale pointers from old master.
617 throttle_session(i, session[i].throttle_in, session[i].throttle_out);
618
619 config->cluster_highest_sessionid = i;
620 }
621
622 session[last_free].next = 0; // End of chain.
623 last_id = high_unique_id; // Keep track of the highest used session ID.
624
625 become_master();
626
627 rebuild_address_pool();
628
629 // If we're not the very first master, this is a big issue!
630 if(count>0)
631 LOG(0, 0, 0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count);
632
633 config->cluster_undefined_sessions = 0;
634 config->cluster_undefined_tunnels = 0;
635 config->cluster_iam_uptodate = 1; // assume all peers are up-to-date
636
637 if (!num_peers) // lone master
638 advertise();
639 #ifdef BGP
640 else if (bgp_configured)
641 bgp_enable_routing(0);
642 #endif /* BGP */
643
644
645 // FIXME. We need to fix up the tunnel control message
646 // queue here! There's a number of other variables we
647 // should also update.
648 }
649
650
651 //
652 // Check that our session table is validly matching what the
653 // master has in mind.
654 //
655 // In particular, if we have too many sessions marked 'undefined'
656 // we fix it up here, and we ensure that the 'first free session'
657 // pointer is valid.
658 //
659 static void cluster_check_sessions(int highsession, int freesession_ptr, int hightunnel)
660 {
661 int i;
662
663 sessionfree = freesession_ptr; // Keep the freesession ptr valid.
664
665 if (config->cluster_iam_uptodate)
666 return;
667
668 if (highsession > config->cluster_undefined_sessions && hightunnel > config->cluster_undefined_tunnels)
669 return;
670
671 // Clear out defined sessions, counting the number of
672 // undefs remaining.
673 config->cluster_undefined_sessions = 0;
674 for (i = 1 ; i < MAXSESSION; ++i) {
675 if (i > highsession) {
676 if (session[i].tunnel == T_UNDEF) session[i].tunnel = T_FREE; // Defined.
677 continue;
678 }
679
680 if (session[i].tunnel == T_UNDEF)
681 ++config->cluster_undefined_sessions;
682 }
683
684 // Clear out defined tunnels, counting the number of
685 // undefs remaining.
686 config->cluster_undefined_tunnels = 0;
687 for (i = 1 ; i < MAXTUNNEL; ++i) {
688 if (i > hightunnel) {
689 if (tunnel[i].state == TUNNELUNDEF) tunnel[i].state = TUNNELFREE; // Defined.
690 continue;
691 }
692
693 if (tunnel[i].state == TUNNELUNDEF)
694 ++config->cluster_undefined_tunnels;
695 }
696
697
698 if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels) {
699 LOG(2, 0, 0, "Cleared undefined sessions/tunnels. %d sess (high %d), %d tunn (high %d)\n",
700 config->cluster_undefined_sessions, highsession, config->cluster_undefined_tunnels, hightunnel);
701 return;
702 }
703
704 // Are we up to date?
705
706 if (!config->cluster_iam_uptodate)
707 cluster_uptodate();
708 }
709
710 static int hb_add_type(char **p, int type, int id)
711 {
712 switch (type) {
713 case C_CSESSION: { // Compressed C_SESSION.
714 uint8_t c[sizeof(sessiont) * 2]; // Bigger than worst case.
715 uint8_t *d = (uint8_t *) &session[id];
716 uint8_t *orig = d;
717 int size;
718
719 size = rle_compress( &d, sizeof(sessiont), c, sizeof(c) );
720
721 // Did we compress the full structure, and is the size actually
722 // reduced??
723 if ( (d - orig) == sizeof(sessiont) && size < sizeof(sessiont) ) {
724 add_type(p, C_CSESSION, id, (char *) c, size);
725 break;
726 }
727 // Failed to compress : Fall through.
728 }
729 case C_SESSION: add_type(p, C_SESSION, id,
730 (char *) &session[id], sizeof(sessiont));
731 break;
732
733 case C_CTUNNEL: { // Compressed C_TUNNEL
734 uint8_t c[sizeof(tunnelt) * 2]; // Bigger than worst case.
735 uint8_t *d = (uint8_t *) &tunnel[id];
736 uint8_t *orig = d;
737 int size;
738
739 size = rle_compress( &d, sizeof(tunnelt), c, sizeof(c) );
740
741 // Did we compress the full structure, and is the size actually
742 // reduced??
743 if ( (d - orig) == sizeof(tunnelt) && size < sizeof(tunnelt) ) {
744 add_type(p, C_CTUNNEL, id, c, size);
745 break;
746 }
747 // Failed to compress : Fall through.
748 }
749 case C_TUNNEL: add_type(p, C_TUNNEL, id,
750 (char *) &tunnel[id], sizeof(tunnelt));
751 break;
752 default:
753 LOG(0, 0, 0, "Found an invalid type in heart queue! (%d)\n", type);
754 kill(0, SIGTERM);
755 exit(1);
756 }
757 return 0;
758 }
759
760 //
761 // Send a heartbeat, incidently sending out any queued changes..
762 //
763 void cluster_heartbeat()
764 {
765 int i, count = 0, tcount = 0;
766 char buff[MAX_HEART_SIZE + sizeof(heartt) + sizeof(int) ];
767 heartt h;
768 char *p = buff;
769
770 if (!config->cluster_iam_master) // Only the master does this.
771 return;
772
773 config->cluster_table_version += config->cluster_num_changes;
774
775 // Fill out the heartbeat header.
776 memset(&h, 0, sizeof(h));
777
778 h.version = HB_VERSION;
779 h.seq = config->cluster_seq_number;
780 h.basetime = basetime;
781 h.clusterid = config->bind_address; // Will this do??
782 h.basetime = basetime;
783 h.highsession = config->cluster_highest_sessionid;
784 h.freesession = sessionfree;
785 h.hightunnel = config->cluster_highest_tunnelid;
786 h.size_sess = sizeof(sessiont); // Just in case.
787 h.size_tunn = sizeof(tunnelt);
788 h.interval = config->cluster_hb_interval;
789 h.timeout = config->cluster_hb_timeout;
790 h.table_version = config->cluster_table_version;
791
792 add_type(&p, C_HEARTBEAT, HB_VERSION, (char *) &h, sizeof(h));
793
794 for (i = 0; i < config->cluster_num_changes; ++i) {
795 hb_add_type(&p, cluster_changes[i].type, cluster_changes[i].id);
796 }
797
798 if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
799 LOG(0, 0, 0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", (int) (p - buff));
800 kill(0, SIGTERM);
801 exit(1);
802 }
803
804 //
805 // Fill out the packet with sessions from the session table...
806 // (not forgetting to leave space so we can get some tunnels in too )
807 while ( (p + sizeof(uint32_t) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) {
808
809 if (!walk_session_number) // session #0 isn't valid.
810 ++walk_session_number;
811
812 if (count >= config->cluster_highest_sessionid) // If we're a small cluster, don't go wild.
813 break;
814
815 hb_add_type(&p, C_CSESSION, walk_session_number);
816 walk_session_number = (1+walk_session_number)%(config->cluster_highest_sessionid+1); // +1 avoids divide by zero.
817
818 ++count; // Count the number of extra sessions we're sending.
819 }
820
821 //
822 // Fill out the packet with tunnels from the tunnel table...
823 // This effectively means we walk the tunnel table more quickly
824 // than the session table. This is good because stuffing up a
825 // tunnel is a much bigger deal than stuffing up a session.
826 //
827 while ( (p + sizeof(uint32_t) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {
828
829 if (!walk_tunnel_number) // tunnel #0 isn't valid.
830 ++walk_tunnel_number;
831
832 if (tcount >= config->cluster_highest_tunnelid)
833 break;
834
835 hb_add_type(&p, C_CTUNNEL, walk_tunnel_number);
836 walk_tunnel_number = (1+walk_tunnel_number)%(config->cluster_highest_tunnelid+1); // +1 avoids divide by zero.
837
838 ++tcount;
839 }
840
841 //
842 // Did we do something wrong?
843 if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
844 LOG(0, 0, 0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", (int) (p - buff));
845 kill(0, SIGTERM);
846 exit(1);
847 }
848
849 LOG(3, 0, 0, "Sending v%d heartbeat #%d, change #%" PRIu64 " with %d changes "
850 "(%d x-sess, %d x-tunnels, %d highsess, %d hightun, size %d)\n",
851 HB_VERSION, h.seq, h.table_version, config->cluster_num_changes,
852 count, tcount, config->cluster_highest_sessionid,
853 config->cluster_highest_tunnelid, (int) (p - buff));
854
855 config->cluster_num_changes = 0;
856
857 send_heartbeat(h.seq, buff, (p-buff) ); // Send out the heartbeat to the cluster, keeping a copy of it.
858
859 config->cluster_seq_number = (config->cluster_seq_number+1)%HB_MAX_SEQ; // Next seq number to use.
860 }
861
862 //
863 // A structure of type 'type' has changed; Add it to the queue to send.
864 //
865 static int type_changed(int type, int id)
866 {
867 int i;
868
869 for (i = 0 ; i < config->cluster_num_changes ; ++i)
870 if ( cluster_changes[i].id == id &&
871 cluster_changes[i].type == type)
872 return 0; // Already marked for change.
873
874 cluster_changes[i].type = type;
875 cluster_changes[i].id = id;
876 ++config->cluster_num_changes;
877
878 if (config->cluster_num_changes > MAX_CHANGES)
879 cluster_heartbeat(); // flush now
880
881 return 1;
882 }
883
884
885 // A particular session has been changed!
886 int cluster_send_session(int sid)
887 {
888 if (!config->cluster_iam_master) {
889 LOG(0, sid, 0, "I'm not a master, but I just tried to change a session!\n");
890 return -1;
891 }
892
893 return type_changed(C_CSESSION, sid);
894 }
895
896 // A particular tunnel has been changed!
897 int cluster_send_tunnel(int tid)
898 {
899 if (!config->cluster_iam_master) {
900 LOG(0, 0, tid, "I'm not a master, but I just tried to change a tunnel!\n");
901 return -1;
902 }
903
904 return type_changed(C_CTUNNEL, tid);
905 }
906
907
908 //
909 // We're a master, and a slave has just told us that it's
910 // missed a packet. We'll resend it every packet since
911 // the last one it's seen.
912 //
913 static int cluster_catchup_slave(int seq, in_addr_t slave)
914 {
915 int s;
916 int diff;
917
918 LOG(1, 0, 0, "Slave %s sent LASTSEEN with seq %d\n", fmtaddr(slave, 0), seq);
919 if (!config->cluster_iam_master) {
920 LOG(1, 0, 0, "Got LASTSEEN but I'm not a master! Sending a PING.\n");
921 // Send a ping to the slave so they know we're no longer a master
922 peer_send_ping(slave, basetime);
923 return 0;
924 }
925
926 diff = config->cluster_seq_number - seq; // How many packet do we need to send?
927 if (diff < 0)
928 diff += HB_MAX_SEQ;
929
930 if (diff >= HB_HISTORY_SIZE) { // Ouch. We don't have the packet to send it!
931 LOG(0, 0, 0, "A slave asked for message %d when our seq number is %d. Killing it.\n",
932 seq, config->cluster_seq_number);
933 return peer_send_message(slave, C_KILL, seq, NULL, 0);// Kill the slave. Nothing else to do.
934 }
935
936 LOG(1, 0, 0, "Sending %d catchup packets to slave %s\n", diff, fmtaddr(slave, 0) );
937
938 // Now resend every packet that it missed, in order.
939 while (seq != config->cluster_seq_number) {
940 s = seq % HB_HISTORY_SIZE;
941 if (seq != past_hearts[s].seq) {
942 LOG(0, 0, 0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n",
943 fmtaddr(slave, 0), seq, past_hearts[s].seq, s, config->cluster_seq_number);
944 return -1; // What to do here!?
945 }
946 peer_send_data(slave, past_hearts[s].data, past_hearts[s].size);
947 seq = (seq+1)%HB_MAX_SEQ; // Increment to next seq number.
948 }
949 return 0; // All good!
950 }
951
952 //
953 // We've heard from another peer! Add it to the list
954 // that we select from at election time.
955 //
956 static int cluster_add_peer(in_addr_t peer, time_t basetime, pingt *pp, int size)
957 {
958 int i;
959 in_addr_t clusterid;
960 pingt p;
961
962 // Allow for backward compatability.
963 // Just the ping packet into a new structure to allow
964 // for the possibility that we might have received
965 // more or fewer elements than we were expecting.
966 if (size > sizeof(p))
967 size = sizeof(p);
968
969 memset( (void *) &p, 0, sizeof(p) );
970 memcpy( (void *) &p, (void *) pp, size);
971
972 clusterid = p.addr;
973 if (clusterid != config->bind_address)
974 {
975 // Is this for us?
976 LOG(4, 0, 0, "Skipping ping from %s (different cluster)\n", fmtaddr(peer, 0));
977 return 0;
978 }
979
980 for (i = 0; i < num_peers ; ++i)
981 {
982 if (peers[i].peer != peer)
983 continue;
984
985 // This peer already exists. Just update the timestamp.
986 peers[i].basetime = basetime;
987 peers[i].timestamp = TIME;
988 peers[i].uptodate = !p.undef;
989 break;
990 }
991
992 // Is this the master shutting down??
993 if (peer == config->cluster_master_address) {
994 LOG(3, 0, 0, "Master %s %s\n", fmtaddr(config->cluster_master_address, 0),
995 basetime ? "has restarted!" : "shutting down...");
996
997 config->cluster_master_address = 0;
998 config->cluster_last_hb = 0; // Force an election.
999 cluster_check_master();
1000 return 0;
1001 }
1002
1003 if (i >= num_peers)
1004 {
1005 LOG(4, 0, 0, "Adding %s as a peer\n", fmtaddr(peer, 0));
1006
1007 // Not found. Is there a stale slot to re-use?
1008 for (i = 0; i < num_peers ; ++i)
1009 {
1010 if (!peers[i].basetime) // Shutdown
1011 break;
1012
1013 if ((peers[i].timestamp + config->cluster_hb_timeout * 10) < TIME) // Stale.
1014 break;
1015 }
1016
1017 if (i >= CLUSTER_MAX_SIZE)
1018 {
1019 // Too many peers!!
1020 LOG(0, 0, 0, "Tried to add %s as a peer, but I already have %d of them!\n", fmtaddr(peer, 0), i);
1021 return -1;
1022 }
1023
1024 peers[i].peer = peer;
1025 peers[i].basetime = basetime;
1026 peers[i].timestamp = TIME;
1027 peers[i].uptodate = !p.undef;
1028 if (i == num_peers)
1029 ++num_peers;
1030
1031 LOG(1, 0, 0, "Added %s as a new peer. Now %d peers\n", fmtaddr(peer, 0), num_peers);
1032 }
1033
1034 return 1;
1035 }
1036
1037 /* Handle the slave updating the byte counters for the master. */
1038 //
1039 // Note that we don't mark the session as dirty; We rely on
1040 // the slow table walk to propogate this back out to the slaves.
1041 //
1042 static int cluster_handle_bytes(char *data, int size)
1043 {
1044 bytest *b;
1045
1046 b = (bytest *) data;
1047
1048 LOG(3, 0, 0, "Got byte counter update (size %d)\n", size);
1049
1050 /* Loop around, adding the byte
1051 counts to each of the sessions. */
1052
1053 while (size >= sizeof(*b) ) {
1054 if (b->sid > MAXSESSION) {
1055 LOG(0, 0, 0, "Got C_BYTES with session #%d!\n", b->sid);
1056 return -1; /* Abort processing */
1057 }
1058
1059 session[b->sid].total_cin += b->in;
1060 session[b->sid].total_cout += b->out;
1061
1062 session[b->sid].cin += b->in;
1063 session[b->sid].cout += b->out;
1064
1065 if (b->in)
1066 session[b->sid].last_packet = time_now; // Reset idle timer!
1067
1068 size -= sizeof(*b);
1069 ++b;
1070 }
1071
1072 if (size != 0)
1073 LOG(0, 0, 0, "Got C_BYTES with %d bytes of trailing junk!\n", size);
1074
1075 return size;
1076 }
1077
1078 //
1079 // Handle receiving a session structure in a heartbeat packet.
1080 //
1081 static int cluster_recv_session(int more, uint8_t *p)
1082 {
1083 if (more >= MAXSESSION) {
1084 LOG(0, 0, 0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
1085 return -1;
1086 }
1087
1088 if (session[more].tunnel == T_UNDEF) {
1089 if (config->cluster_iam_uptodate) { // Sanity.
1090 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined session!\n");
1091 } else {
1092 --config->cluster_undefined_sessions;
1093 }
1094 }
1095
1096 load_session(more, (sessiont *) p); // Copy session into session table..
1097
1098 LOG(5, more, 0, "Received session update (%d undef)\n", config->cluster_undefined_sessions);
1099
1100 if (!config->cluster_iam_uptodate)
1101 cluster_uptodate(); // Check to see if we're up to date.
1102
1103 return 0;
1104 }
1105
1106 static int cluster_recv_tunnel(int more, uint8_t *p)
1107 {
1108 if (more >= MAXTUNNEL) {
1109 LOG(0, 0, 0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
1110 return -1;
1111 }
1112
1113 if (tunnel[more].state == TUNNELUNDEF) {
1114 if (config->cluster_iam_uptodate) { // Sanity.
1115 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined tunnel!\n");
1116 } else {
1117 --config->cluster_undefined_tunnels;
1118 }
1119 }
1120
1121 memcpy(&tunnel[more], p, sizeof(tunnel[more]) );
1122
1123 //
1124 // Clear tunnel control messages. These are dynamically allocated.
1125 // If we get unlucky, this may cause the tunnel to drop!
1126 //
1127 tunnel[more].controls = tunnel[more].controle = NULL;
1128 tunnel[more].controlc = 0;
1129
1130 LOG(5, 0, more, "Received tunnel update\n");
1131
1132 if (!config->cluster_iam_uptodate)
1133 cluster_uptodate(); // Check to see if we're up to date.
1134
1135 return 0;
1136 }
1137
1138
1139 //
1140 // Process a heartbeat..
1141 //
1142 // v3: added interval, timeout
1143 // v4: added table_version
1144 static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t *p, in_addr_t addr)
1145 {
1146 heartt *h;
1147 int s = size - (p-data);
1148 int i, type;
1149
1150 #if HB_VERSION != 4
1151 # error "need to update cluster_process_heartbeat()"
1152 #endif
1153
1154 // we handle versions 3 through 4
1155 if (more < 3 || more > HB_VERSION) {
1156 LOG(0, 0, 0, "Received a heartbeat version that I don't support (%d)!\n", more);
1157 return -1; // Ignore it??
1158 }
1159
1160 // Ok. It's a heartbeat packet from a cluster master!
1161 if (s < sizeof(*h))
1162 goto shortpacket;
1163
1164 h = (heartt *) p;
1165 p += sizeof(*h);
1166 s -= sizeof(*h);
1167
1168 if (h->clusterid != config->bind_address)
1169 return -1; // It's not part of our cluster.
1170
1171 if (config->cluster_iam_master) { // Sanity...
1172 // Note that this MUST match the election process above!
1173
1174 LOG(0, 0, 0, "I just got a heartbeat from master %s, but _I_ am the master!\n", fmtaddr(addr, 0));
1175 if (!h->basetime) {
1176 LOG(0, 0, 0, "Heartbeat with zero basetime! Ignoring\n");
1177 return -1; // Skip it.
1178 }
1179
1180 if (more >= 4) {
1181 if (h->table_version > config->cluster_table_version) {
1182 LOG(0, 0, 0, "They've seen more state changes (%" PRIu64 " vs my %" PRIu64 ") so I'm gone!\n",
1183 h->table_version, config->cluster_table_version);
1184
1185 kill(0, SIGTERM);
1186 exit(1);
1187 }
1188 if (h->table_version < config->cluster_table_version)
1189 return -1;
1190 }
1191
1192 if (basetime > h->basetime) {
1193 LOG(0, 0, 0, "They're an older master than me so I'm gone!\n");
1194 kill(0, SIGTERM);
1195 exit(1);
1196 }
1197
1198 if (basetime < h->basetime)
1199 return -1;
1200
1201 if (my_address < addr) { // Tie breaker.
1202 LOG(0, 0, 0, "They're a higher IP address than me, so I'm gone!\n");
1203 kill(0, SIGTERM);
1204 exit(1);
1205 }
1206
1207 //
1208 // Send it a unicast heartbeat to see give it a chance to die.
1209 // NOTE: It's actually safe to do seq-number - 1 without checking
1210 // for wrap around.
1211 //
1212 cluster_catchup_slave(config->cluster_seq_number - 1, addr);
1213
1214 return -1; // Skip it.
1215 }
1216
1217 //
1218 // Try and guard against a stray master appearing.
1219 //
1220 // Ignore heartbeats received from another master before the
1221 // timeout (less a smidgen) for the old master has elapsed.
1222 //
1223 // Note that after a clean failover, the cluster_master_address
1224 // is cleared, so this doesn't run.
1225 //
1226 if (config->cluster_master_address && addr != config->cluster_master_address
1227 && (config->cluster_last_hb + config->cluster_hb_timeout - 11) > TIME) {
1228 LOG(0, 0, 0, "Ignoring stray heartbeat from %s, current master %s has not yet timed out (last heartbeat %.1f seconds ago).\n",
1229 fmtaddr(addr, 0), fmtaddr(config->cluster_master_address, 1),
1230 0.1 * (TIME - config->cluster_last_hb));
1231 return -1; // ignore
1232 }
1233
1234 if (config->cluster_seq_number == -1) // Don't have one. Just align to the master...
1235 config->cluster_seq_number = h->seq;
1236
1237 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1238
1239 if (config->cluster_seq_number != h->seq) { // Out of sequence heartbeat!
1240 static int lastseen_seq = 0;
1241 static time_t lastseen_time = 0;
1242
1243 // limit to once per second for a particular seq#
1244 int ask = (config->cluster_seq_number != lastseen_seq || time_now != lastseen_time);
1245
1246 LOG(1, 0, 0, "HB: Got seq# %d but was expecting %d. %s.\n",
1247 h->seq, config->cluster_seq_number,
1248 ask ? "Asking for resend" : "Ignoring");
1249
1250 if (ask)
1251 {
1252 lastseen_seq = config->cluster_seq_number;
1253 lastseen_time = time_now;
1254 peer_send_message(addr, C_LASTSEEN, config->cluster_seq_number, NULL, 0);
1255 }
1256
1257 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1258
1259 // Just drop the packet. The master will resend it as part of the catchup.
1260
1261 return 0;
1262 }
1263 // Save the packet in our buffer.
1264 // This is needed in case we become the master.
1265 config->cluster_seq_number = (h->seq+1)%HB_MAX_SEQ;
1266 i = h->seq % HB_HISTORY_SIZE;
1267 past_hearts[i].seq = h->seq;
1268 past_hearts[i].size = size;
1269 memcpy(&past_hearts[i].data, data, size); // Save it.
1270
1271
1272 // Check that we don't have too many undefined sessions, and
1273 // that the free session pointer is correct.
1274 cluster_check_sessions(h->highsession, h->freesession, h->hightunnel);
1275
1276 if (h->interval != config->cluster_hb_interval)
1277 {
1278 LOG(2, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
1279 h->interval, config->cluster_hb_interval);
1280
1281 config->cluster_hb_interval = h->interval;
1282 }
1283
1284 if (h->timeout != config->cluster_hb_timeout)
1285 {
1286 LOG(2, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
1287 h->timeout, config->cluster_hb_timeout);
1288
1289 config->cluster_hb_timeout = h->timeout;
1290 }
1291
1292 // Ok. process the packet...
1293 while ( s > 0) {
1294
1295 type = *((uint32_t *) p);
1296 p += sizeof(uint32_t);
1297 s -= sizeof(uint32_t);
1298
1299 more = *((uint32_t *) p);
1300 p += sizeof(uint32_t);
1301 s -= sizeof(uint32_t);
1302
1303 switch (type) {
1304 case C_CSESSION: { // Compressed session structure.
1305 uint8_t c[ sizeof(sessiont) + 2];
1306 int size;
1307 uint8_t *orig_p = p;
1308
1309 size = rle_decompress((uint8_t **) &p, s, c, sizeof(c) );
1310 s -= (p - orig_p);
1311
1312 if (size != sizeof(sessiont) ) { // Ouch! Very very bad!
1313 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1314 // Now what? Should exit! No-longer up to date!
1315 break;
1316 }
1317
1318 cluster_recv_session(more, c);
1319 break;
1320 }
1321 case C_SESSION:
1322 if ( s < sizeof(session[more]))
1323 goto shortpacket;
1324
1325 cluster_recv_session(more, p);
1326
1327 p += sizeof(session[more]);
1328 s -= sizeof(session[more]);
1329 break;
1330
1331 case C_CTUNNEL: { // Compressed tunnel structure.
1332 uint8_t c[ sizeof(tunnelt) + 2];
1333 int size;
1334 uint8_t *orig_p = p;
1335
1336 size = rle_decompress((uint8_t **) &p, s, c, sizeof(c));
1337 s -= (p - orig_p);
1338
1339 if (size != sizeof(tunnelt) ) { // Ouch! Very very bad!
1340 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1341 // Now what? Should exit! No-longer up to date!
1342 break;
1343 }
1344
1345 cluster_recv_tunnel(more, c);
1346 break;
1347
1348 }
1349 case C_TUNNEL:
1350 if ( s < sizeof(tunnel[more]))
1351 goto shortpacket;
1352
1353 cluster_recv_tunnel(more, p);
1354
1355 p += sizeof(tunnel[more]);
1356 s -= sizeof(tunnel[more]);
1357 break;
1358 default:
1359 LOG(0, 0, 0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type);
1360 return -1; // can't process any more of the packet!!
1361 }
1362 }
1363
1364 if (config->cluster_master_address != addr)
1365 {
1366 LOG(0, 0, 0, "My master just changed from %s to %s!\n",
1367 fmtaddr(config->cluster_master_address, 0), fmtaddr(addr, 1));
1368
1369 config->cluster_master_address = addr;
1370 }
1371
1372 config->cluster_last_hb = TIME; // Successfully received a heartbeat!
1373 config->cluster_table_version = h->table_version;
1374 return 0;
1375
1376 shortpacket:
1377 LOG(0, 0, 0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
1378 return -1;
1379 }
1380
1381 //
1382 // We got a packet on the cluster port!
1383 // Handle pings, lastseens, and heartbeats!
1384 //
1385 int processcluster(char *data, int size, in_addr_t addr)
1386 {
1387 int type, more;
1388 char *p = data;
1389 int s = size;
1390
1391 if (addr == my_address)
1392 return -1; // Ignore it. Something looped back the multicast!
1393
1394 LOG(5, 0, 0, "Process cluster: %d bytes from %s\n", size, fmtaddr(addr, 0));
1395
1396 if (s <= 0) // Any data there??
1397 return -1;
1398
1399 if (s < 8)
1400 goto shortpacket;
1401
1402 type = *((uint32_t *) p);
1403 p += sizeof(uint32_t);
1404 s -= sizeof(uint32_t);
1405
1406 more = *((uint32_t *) p);
1407 p += sizeof(uint32_t);
1408 s -= sizeof(uint32_t);
1409
1410 switch (type) {
1411 case C_PING: // Update the peers table.
1412 return cluster_add_peer(addr, more, (pingt *) p, s);
1413
1414 case C_LASTSEEN: // Catch up a slave (slave missed a packet).
1415 return cluster_catchup_slave(more, addr);
1416
1417 case C_FORWARD: { // Forwarded control packet. pass off to processudp.
1418 struct sockaddr_in a;
1419 a.sin_addr.s_addr = more;
1420
1421 a.sin_port = *(int *) p;
1422 s -= sizeof(int);
1423 p += sizeof(int);
1424
1425 if (!config->cluster_iam_master) { // huh?
1426 LOG(0, 0, 0, "I'm not the master, but I got a C_FORWARD from %s?\n", fmtaddr(addr, 0));
1427 return -1;
1428 }
1429
1430 LOG(4, 0, 0, "Got a forwarded packet... (%s:%d)\n", fmtaddr(more, 0), a.sin_port);
1431 STAT(recv_forward);
1432 processudp(p, s, &a);
1433 return 0;
1434 }
1435 case C_THROTTLE: { // Receive a forwarded packet from a slave.
1436 if (!config->cluster_iam_master) {
1437 LOG(0, 0, 0, "I'm not the master, but I got a C_THROTTLE from %s?\n", fmtaddr(addr, 0));
1438 return -1;
1439 }
1440
1441 tbf_queue_packet(more, p, s); // The TBF id tells wether it goes in or out.
1442 return 0;
1443 }
1444 case C_GARDEN:
1445 // Receive a walled garden packet from a slave.
1446 if (!config->cluster_iam_master) {
1447 LOG(0, 0, 0, "I'm not the master, but I got a C_GARDEN from %s?\n", fmtaddr(addr, 0));
1448 return -1;
1449 }
1450
1451 tun_write(p, s);
1452 return 0;
1453
1454 case C_BYTES:
1455 return cluster_handle_bytes(p, s);
1456
1457 case C_KILL: // The master asked us to die!? (usually because we're too out of date).
1458 if (config->cluster_iam_master) {
1459 LOG(0, 0, 0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", fmtaddr(addr, 0), more);
1460 return -1;
1461 }
1462 if (more != config->cluster_seq_number) {
1463 LOG(0, 0, 0, "The master asked us to die but the seq number didn't match!?\n");
1464 return -1;
1465 }
1466
1467 if (addr != config->cluster_master_address) {
1468 LOG(0, 0, 0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%s)\n",
1469 fmtaddr(addr, 0), fmtaddr(config->cluster_master_address, 1));
1470 // We can only warn about it. The master might really have switched!
1471 }
1472
1473 LOG(0, 0, 0, "Received a valid C_KILL: I'm going to die now.\n");
1474 kill(0, SIGTERM);
1475 exit(0); // Lets be paranoid;
1476 return -1; // Just signalling the compiler.
1477
1478 case C_HEARTBEAT:
1479 LOG(4, 0, 0, "Got a heartbeat from %s\n", fmtaddr(addr, 0));
1480 return cluster_process_heartbeat(data, size, more, p, addr);
1481
1482 default:
1483 LOG(0, 0, 0, "Strange type packet received on cluster socket (%d)\n", type);
1484 return -1;
1485 }
1486 return 0;
1487
1488 shortpacket:
1489 LOG(0, 0, 0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n");
1490 return -1;
1491 }
1492
1493 //====================================================================================================
1494
1495 int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
1496 {
1497 int i;
1498
1499 if (CLI_HELP_REQUESTED)
1500 return CLI_HELP_NO_ARGS;
1501
1502 cli_print(cli, "Cluster status : %s", config->cluster_iam_master ? "Master" : "Slave" );
1503 cli_print(cli, "My address : %s", fmtaddr(my_address, 0));
1504 cli_print(cli, "VIP address : %s", fmtaddr(config->bind_address, 0));
1505 cli_print(cli, "Multicast address: %s", fmtaddr(config->cluster_address, 0));
1506 cli_print(cli, "Multicast i'face : %s", config->cluster_interface);
1507
1508 if (!config->cluster_iam_master) {
1509 cli_print(cli, "My master : %s (last heartbeat %.1f seconds old)",
1510 config->cluster_master_address
1511 ? fmtaddr(config->cluster_master_address, 0)
1512 : "Not defined",
1513 0.1 * (TIME - config->cluster_last_hb));
1514 cli_print(cli, "Uptodate : %s", config->cluster_iam_uptodate ? "Yes" : "No");
1515 cli_print(cli, "Table version # : %" PRIu64, config->cluster_table_version);
1516 cli_print(cli, "Next sequence number expected: %d", config->cluster_seq_number);
1517 cli_print(cli, "%d sessions undefined of %d", config->cluster_undefined_sessions, config->cluster_highest_sessionid);
1518 cli_print(cli, "%d tunnels undefined of %d", config->cluster_undefined_tunnels, config->cluster_highest_tunnelid);
1519 } else {
1520 cli_print(cli, "Table version # : %" PRIu64, config->cluster_table_version);
1521 cli_print(cli, "Next heartbeat # : %d", config->cluster_seq_number);
1522 cli_print(cli, "Highest session : %d", config->cluster_highest_sessionid);
1523 cli_print(cli, "Highest tunnel : %d", config->cluster_highest_tunnelid);
1524 cli_print(cli, "%d changes queued for sending", config->cluster_num_changes);
1525 }
1526 cli_print(cli, "%d peers.", num_peers);
1527
1528 if (num_peers)
1529 cli_print(cli, "%20s %10s %8s", "Address", "Basetime", "Age");
1530 for (i = 0; i < num_peers; ++i) {
1531 cli_print(cli, "%20s %10u %8d", fmtaddr(peers[i].peer, 0),
1532 peers[i].basetime, TIME - peers[i].timestamp);
1533 }
1534 return CLI_OK;
1535 }
1536
1537 //
1538 // Simple run-length-encoding compression.
1539 // Format is
1540 // 1 byte < 128 = count of non-zero bytes following. // Not legal to be zero.
1541 // n non-zero bytes;
1542 // or
1543 // 1 byte > 128 = (count - 128) run of zero bytes. //
1544 // repeat.
1545 // count == 0 indicates end of compressed stream.
1546 //
1547 // Compress from 'src' into 'dst'. return number of bytes
1548 // used from 'dst'.
1549 // Updates *src_p to indicate 1 past last bytes used.
1550 //
1551 // We could get an extra byte in the zero runs by storing (count-1)
1552 // but I'm playing it safe.
1553 //
1554 // Worst case is a 50% expansion in space required (trying to
1555 // compress { 0x00, 0x01 } * N )
1556 static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
1557 {
1558 int count;
1559 int orig_dsize = dsize;
1560 uint8_t *x, *src;
1561 src = *src_p;
1562
1563 while (ssize > 0 && dsize > 2) {
1564 count = 0;
1565 x = dst++; --dsize; // Reserve space for count byte..
1566
1567 if (*src) { // Copy a run of non-zero bytes.
1568 while (*src && count < 127 && ssize > 0 && dsize > 1) { // Count number of non-zero bytes.
1569 *dst++ = *src++;
1570 --dsize; --ssize;
1571 ++count;
1572 }
1573 *x = count; // Store number of non-zero bytes. Guarenteed to be non-zero!
1574
1575 } else { // Compress a run of zero bytes.
1576 while (*src == 0 && count < 127 && ssize > 0) {
1577 ++src;
1578 --ssize;
1579 ++count;
1580 }
1581 *x = count | 0x80 ;
1582 }
1583 }
1584
1585 *dst++ = 0x0; // Add Stop byte.
1586 --dsize;
1587
1588 *src_p = src;
1589 return (orig_dsize - dsize);
1590 }
1591
1592 //
1593 // Decompress the buffer into **p.
1594 // 'psize' is the size of the decompression buffer available.
1595 //
1596 // Returns the number of bytes decompressed.
1597 //
1598 // Decompresses from '*src_p' into 'dst'.
1599 // Return the number of dst bytes used.
1600 // Updates the 'src_p' pointer to point to the
1601 // first un-used byte.
1602 static int rle_decompress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
1603 {
1604 int count;
1605 int orig_dsize = dsize;
1606 char *src = *src_p;
1607
1608 while (ssize >0 && dsize > 0) { // While there's more to decompress, and there's room in the decompress buffer...
1609 count = *src++; --ssize; // get the count byte from the source.
1610 if (count == 0x0) // End marker reached? If so, finish.
1611 break;
1612
1613 if (count & 0x80) { // Decompress a run of zeros
1614 for (count &= 0x7f ; count > 0 && dsize > 0; --count) {
1615 *dst++ = 0x0;
1616 --dsize;
1617 }
1618 } else { // Copy run of non-zero bytes.
1619 for ( ; count > 0 && ssize && dsize; --count) { // Copy non-zero bytes across.
1620 *dst++ = *src++;
1621 --ssize; --dsize;
1622 }
1623 }
1624 }
1625 *src_p = src;
1626 return (orig_dsize - dsize);
1627 }