2a5e6e3a4b9c1267dd43e64cae1910b3feeb37c7
[l2tpns.git] / cluster.c
1 // L2TPNS Clustering Stuff
2
3 char const *cvs_id_cluster = "$Id: cluster.c,v 1.22 2004-12-13 05:09:55 bodea Exp $";
4
5 #include <stdio.h>
6 #include <sys/file.h>
7 #include <sys/stat.h>
8 #include <sys/socket.h>
9 #include <netinet/in.h>
10 #include <arpa/inet.h>
11 #include <sys/ioctl.h>
12 #include <net/if.h>
13 #include <string.h>
14 #include <malloc.h>
15 #include <errno.h>
16 #include <stdlib.h>
17 #include <stdarg.h>
18 #include <unistd.h>
19 #include <stdio.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 ipt 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 u32 peer;
64 time_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(u8 ** src_p, int ssize, u8 *dst, int dsize);
71 static int rle_compress(u8 ** src_p, int ssize, u8 *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 * ( (u32*)(*p) ) = type;
183 *p += sizeof(u32);
184
185 * ( (u32*)(*p) ) = more;
186 *p += sizeof(u32);
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(u32 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(u32 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, u32 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 //
289 // Forward a throttled packet to the master for handling.
290 //
291 // The master just drops the packet into the appropriate
292 // token bucket queue, and lets normal processing take care
293 // of it.
294 //
295 int master_throttle_packet(int tbfid, char *data, int size)
296 {
297 char buf[65536]; // Vast overkill.
298 char *p = buf;
299
300 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
301 return -1;
302
303 LOG(4, 0, 0, "Throttling packet master (size %d, tbfid %d)\n", size, tbfid);
304
305 add_type(&p, C_THROTTLE, tbfid, data, size);
306
307 return peer_send_data(config->cluster_master_address, buf, (p-buf) );
308
309 }
310
311 //
312 // Forward a walled garden packet to the master for handling.
313 //
314 // The master just writes the packet straight to the tun
315 // device (where is will normally loop through the
316 // firewall rules, and come back in on the tun device)
317 //
318 // (Note that this must be called with the tun header
319 // as the start of the data).
320 int master_garden_packet(sessionidt s, char *data, int size)
321 {
322 char buf[65536]; // Vast overkill.
323 char *p = buf;
324
325 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
326 return -1;
327
328 LOG(4, 0, 0, "Walled garden packet to master (size %d)\n", size);
329
330 add_type(&p, C_GARDEN, s, data, size);
331
332 return peer_send_data(config->cluster_master_address, buf, (p-buf));
333
334 }
335
336 //
337 // Send a chunk of data as a heartbeat..
338 // We save it in the history buffer as we do so.
339 //
340 static void send_heartbeat(int seq, char * data, int size)
341 {
342 int i;
343
344 if (size > sizeof(past_hearts[0].data))
345 {
346 LOG(0, 0, 0, "Tried to heartbeat something larger than the maximum packet!\n");
347 kill(0, SIGTERM);
348 exit(1);
349 }
350 i = seq % HB_HISTORY_SIZE;
351 past_hearts[i].seq = seq;
352 past_hearts[i].size = size;
353 memcpy(&past_hearts[i].data, data, size); // Save it.
354 cluster_send_data(data, size);
355 }
356
357 //
358 // Send an 'i am alive' message to every machine in the cluster.
359 //
360 void cluster_send_ping(time_t basetime)
361 {
362 char buff[100 + sizeof(pingt)];
363 char *p = buff;
364 pingt x;
365
366 if (config->cluster_iam_master && basetime) // We're heartbeating so no need to ping.
367 return;
368
369 LOG(5, 0, 0, "Sending cluster ping...\n");
370
371 x.ver = 1;
372 x.addr = config->bind_address;
373 x.undef = config->cluster_undefined_sessions + config->cluster_undefined_tunnels;
374 x.basetime = basetime;
375
376 add_type(&p, C_PING, basetime, (char *) &x, sizeof(x));
377 cluster_send_data(buff, (p-buff) );
378 }
379
380 //
381 // Walk the session counters looking for non-zero ones to send
382 // to the master. We send up to 600 of them at one time.
383 // We examine a maximum of 3000 sessions.
384 // (50k max session should mean that we normally
385 // examine the entire session table every 25 seconds).
386
387 #define MAX_B_RECS (600)
388 void master_update_counts(void)
389 {
390 int i, c;
391 bytest b[MAX_B_RECS+1];
392
393 if (config->cluster_iam_master) // Only happens on the slaves.
394 return;
395
396 if (!config->cluster_master_address) // If we don't have a master, skip it for a while.
397 return;
398
399 i = MAX_B_RECS * 5; // Examine max 2000 sessions;
400 if (config->cluster_highest_sessionid > i)
401 i = config->cluster_highest_sessionid;
402
403 for ( c = 0; i > 0 ; --i) {
404 // Next session to look at.
405 walk_session_number++;
406 if ( walk_session_number > config->cluster_highest_sessionid)
407 walk_session_number = 1;
408
409 if (!sess_count[walk_session_number].cin && !sess_count[walk_session_number].cout)
410 continue; // Unused. Skip it.
411
412 b[c].sid = walk_session_number;
413 b[c].in = sess_count[walk_session_number].cin;
414 b[c].out = sess_count[walk_session_number].cout;
415
416 if (++c > MAX_B_RECS) // Send a max of 400 elements in a packet.
417 break;
418
419 // Reset counters.
420 sess_count[walk_session_number].cin = sess_count[walk_session_number].cout = 0;
421 }
422
423 if (!c) // Didn't find any that changes. Get out of here!
424 return;
425
426
427 // Forward the data to the master.
428 LOG(4, 0, 0, "Sending byte counters to master (%d elements)\n", c);
429 peer_send_message(config->cluster_master_address, C_BYTES, c, (char*) &b, sizeof(b[0]) * c);
430 return;
431 }
432
433 //
434 // On the master, check how our slaves are going. If
435 // one of them's not up-to-date we'll heartbeat faster.
436 // If we don't have any of them, then we need to turn
437 // on our own packet handling!
438 //
439 void cluster_check_slaves(void)
440 {
441 int i;
442 static int have_peers = 0;
443 int had_peers = have_peers;
444 clockt t = TIME;
445
446 if (!config->cluster_iam_master)
447 return; // Only runs on the master...
448
449 config->cluster_iam_uptodate = 1; // cleared in loop below
450
451 for (i = have_peers = 0; i < num_peers; i++)
452 {
453 if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
454 continue; // Stale peer! Skip them.
455
456 if (!peers[i].basetime)
457 continue; // Shutdown peer! Skip them.
458
459 if (peers[i].uptodate)
460 have_peers = 1;
461
462 if (!peers[i].uptodate)
463 config->cluster_iam_uptodate = 0; // Start fast heartbeats
464 }
465
466 #ifdef BGP
467 // in a cluster, withdraw/add routes when we get a peer/lose all peers
468 if (bgp_configured && have_peers != had_peers)
469 bgp_enable_routing(!have_peers);
470 #endif /* BGP */
471 }
472
473 //
474 // Check that we have a master. If it's been too
475 // long since we heard from a master then hold an election.
476 //
477 void cluster_check_master(void)
478 {
479 int i, count, tcount, high_unique_id = 0;
480 int last_free = 0;
481 clockt t = TIME;
482 static int probed = 0;
483
484 if (config->cluster_iam_master)
485 return; // Only runs on the slaves...
486
487 // If the master is late (missed 2 hearbeats by a second and a
488 // hair) it may be that the switch has dropped us from the
489 // multicast group, try unicasting one probe to the master
490 // which will hopefully respond with a unicast heartbeat that
491 // will allow us to limp along until the querier next runs.
492 if (TIME > (config->cluster_last_hb + 2 * config->cluster_hb_interval + 11))
493 {
494 if (!probed && config->cluster_master_address)
495 {
496 probed = 1;
497 LOG(1, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
498 0.1 * (TIME - (config->cluster_last_hb + config->cluster_hb_interval)));
499
500 peer_send_message(config->cluster_master_address,
501 C_LASTSEEN, config->cluster_seq_number, NULL, 0);
502 }
503 } else { // We got a recent heartbeat; reset the probe flag.
504 probed = 0;
505 }
506
507 if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout))
508 return; // Everything's ok!
509
510 config->cluster_last_hb = TIME + 1; // Just the one election thanks.
511
512 LOG(0, 0, 0, "Master timed out! Holding election...\n");
513
514 for (i = 0; i < num_peers; i++)
515 {
516 if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
517 continue; // Stale peer! Skip them.
518
519 if (!peers[i].basetime)
520 continue; // Shutdown peer! Skip them.
521
522 if (peers[i].basetime < basetime) {
523 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
524 return; // They'll win the election. Get out of here.
525 }
526
527 if (peers[i].basetime == basetime &&
528 peers[i].peer > my_address) {
529 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
530 return; // They'll win the election. Wait for them to come up.
531 }
532 }
533
534 // Wow. it's been ages since I last heard a heartbeat
535 // and I'm better than an of my peers so it's time
536 // to become a master!!!
537
538 config->cluster_iam_master = 1;
539 config->cluster_master_address = 0;
540
541 LOG(0, 0, 0, "I am declaring myself the master!\n");
542
543 if (config->cluster_seq_number == -1)
544 config->cluster_seq_number = 0;
545
546 //
547 // Go through and mark all the tunnels as defined.
548 // Count the highest used tunnel number as well.
549 //
550 config->cluster_highest_tunnelid = 0;
551 for (i = 0, tcount = 0; i < MAXTUNNEL; ++i) {
552 if (tunnel[i].state == TUNNELUNDEF)
553 tunnel[i].state = TUNNELFREE;
554
555 if (tunnel[i].state != TUNNELFREE && i > config->cluster_highest_tunnelid)
556 config->cluster_highest_tunnelid = i;
557 }
558
559 //
560 // Go through and mark all the sessions as being defined.
561 // reset the idle timeouts.
562 // add temporary byte counters to permanent ones.
563 // Re-string the free list.
564 // Find the ID of the highest session.
565 last_free = 0;
566 high_unique_id = 0;
567 config->cluster_highest_sessionid = 0;
568 for (i = 0, count = 0; i < MAXSESSION; ++i) {
569 if (session[i].tunnel == T_UNDEF) {
570 session[i].tunnel = T_FREE;
571 ++count;
572 }
573
574 if (session[i].tunnel == T_FREE) { // Unused session. Add to free list.
575 session[last_free].next = i;
576 session[i].next = 0;
577 last_free = i;
578 }
579
580 // Reset all the idle timeouts..
581 session[i].last_packet = time_now;
582
583 // Accumulate un-sent byte counters.
584 session[i].cin += sess_count[i].cin;
585 session[i].cout += sess_count[i].cout;
586 session[i].total_cin += sess_count[i].cin;
587 session[i].total_cout += sess_count[i].cout;
588
589 sess_count[i].cin = sess_count[i].cout = 0;
590
591 session[i].radius = 0; // Reset authentication as the radius blocks aren't up to date.
592
593 if (session[i].unique_id >= high_unique_id) // This is different to the index into the session table!!!
594 high_unique_id = session[i].unique_id+1;
595
596
597 session[i].tbf_in = session[i].tbf_out = 0; // Remove stale pointers from old master.
598 throttle_session(i, session[i].throttle_in, session[i].throttle_out);
599
600 if (session[i].tunnel != T_FREE && i > config->cluster_highest_sessionid)
601 config->cluster_highest_sessionid = i;
602 }
603
604 session[last_free].next = 0; // End of chain.
605 last_id = high_unique_id; // Keep track of the highest used session ID.
606
607 become_master();
608
609 rebuild_address_pool();
610
611 // If we're not the very first master, this is a big issue!
612 if(count>0)
613 LOG(0, 0, 0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count);
614
615 config->cluster_undefined_sessions = 0;
616 config->cluster_undefined_tunnels = 0;
617 config->cluster_iam_uptodate = 1; // assume all peers are up-to-date
618
619 if (!num_peers) // lone master
620 advertise();
621
622 // FIXME. We need to fix up the tunnel control message
623 // queue here! There's a number of other variables we
624 // should also update.
625 }
626
627
628 //
629 // Check that our session table is validly matching what the
630 // master has in mind.
631 //
632 // In particular, if we have too many sessions marked 'undefined'
633 // we fix it up here, and we ensure that the 'first free session'
634 // pointer is valid.
635 //
636 static void cluster_check_sessions(int highsession, int freesession_ptr, int hightunnel)
637 {
638 int i;
639
640 sessionfree = freesession_ptr; // Keep the freesession ptr valid.
641
642 if (config->cluster_iam_uptodate)
643 return;
644
645 if (highsession > config->cluster_undefined_sessions && hightunnel > config->cluster_undefined_tunnels)
646 return;
647
648 // Clear out defined sessions, counting the number of
649 // undefs remaining.
650 config->cluster_undefined_sessions = 0;
651 for (i = 1 ; i < MAXSESSION; ++i) {
652 if (i > highsession) {
653 session[i].tunnel = 0; // Defined.
654 continue;
655 }
656 if (session[i].tunnel != T_UNDEF)
657 continue;
658 ++config->cluster_undefined_sessions;
659 }
660
661 // Clear out defined tunnels, counting the number of
662 // undefs remaining.
663 config->cluster_undefined_tunnels = 0;
664 for (i = 1 ; i < MAXTUNNEL; ++i) {
665 if (i > hightunnel) {
666 tunnel[i].state = TUNNELFREE; // Defined.
667 continue;
668 }
669 if (tunnel[i].state != TUNNELUNDEF)
670 continue;
671 ++config->cluster_undefined_tunnels;
672 }
673
674
675 if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels) {
676 LOG(2, 0, 0, "Cleared undefined sessions/tunnels. %d sess (high %d), %d tunn (high %d)\n",
677 config->cluster_undefined_sessions, highsession, config->cluster_undefined_tunnels, hightunnel);
678 return;
679 }
680
681 // Are we up to date?
682
683 if (!config->cluster_iam_uptodate)
684 cluster_uptodate();
685 }
686
687 static int hb_add_type(char **p, int type, int id)
688 {
689 switch (type) {
690 case C_CSESSION: { // Compressed C_SESSION.
691 u8 c[sizeof(sessiont) * 2]; // Bigger than worst case.
692 u8 *d = (u8 *) &session[id];
693 u8 *orig = d;
694 int size;
695
696 size = rle_compress( &d, sizeof(sessiont), c, sizeof(c) );
697
698 // Did we compress the full structure, and is the size actually
699 // reduced??
700 if ( (d - orig) == sizeof(sessiont) && size < sizeof(sessiont) ) {
701 add_type(p, C_CSESSION, id, (char*) c, size);
702 break;
703 }
704 // Failed to compress : Fall through.
705 }
706 case C_SESSION: add_type(p, C_SESSION, id,
707 (char*) &session[id], sizeof(sessiont));
708 break;
709
710 case C_CTUNNEL: { // Compressed C_TUNNEL
711 u8 c[sizeof(tunnelt) * 2]; // Bigger than worst case.
712 u8 *d = (u8 *) &tunnel[id];
713 u8 *orig = d;
714 int size;
715
716 size = rle_compress( &d, sizeof(tunnelt), c, sizeof(c) );
717
718 // Did we compress the full structure, and is the size actually
719 // reduced??
720 if ( (d - orig) == sizeof(tunnelt) && size < sizeof(tunnelt) ) {
721 add_type(p, C_CTUNNEL, id, c, size);
722 break;
723 }
724 // Failed to compress : Fall through.
725 }
726 case C_TUNNEL: add_type(p, C_TUNNEL, id,
727 (char*) &tunnel[id], sizeof(tunnelt));
728 break;
729 default:
730 LOG(0, 0, 0, "Found an invalid type in heart queue! (%d)\n", type);
731 kill(0, SIGTERM);
732 exit(1);
733 }
734 return 0;
735 }
736
737 //
738 // Send a heartbeat, incidently sending out any queued changes..
739 //
740 void cluster_heartbeat()
741 {
742 int i, count = 0, tcount = 0;
743 char buff[MAX_HEART_SIZE + sizeof(heartt) + sizeof(int) ];
744 heartt h;
745 char *p = buff;
746
747 if (!config->cluster_iam_master) // Only the master does this.
748 return;
749
750 config->cluster_table_version += config->cluster_num_changes;
751
752 // Fill out the heartbeat header.
753 memset(&h, 0, sizeof(h));
754
755 h.version = HB_VERSION;
756 h.seq = config->cluster_seq_number;
757 h.basetime = basetime;
758 h.clusterid = config->bind_address; // Will this do??
759 h.basetime = basetime;
760 h.highsession = config->cluster_highest_sessionid;
761 h.freesession = sessionfree;
762 h.hightunnel = config->cluster_highest_tunnelid;
763 h.size_sess = sizeof(sessiont); // Just in case.
764 h.size_tunn = sizeof(tunnelt);
765 h.interval = config->cluster_hb_interval;
766 h.timeout = config->cluster_hb_timeout;
767 h.table_version = config->cluster_table_version;
768
769 add_type(&p, C_HEARTBEAT, HB_VERSION, (char*) &h, sizeof(h));
770
771 for (i = 0; i < config->cluster_num_changes; ++i) {
772 hb_add_type(&p, cluster_changes[i].type, cluster_changes[i].id);
773 }
774
775 if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
776 LOG(0, 0, 0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", p - buff);
777 kill(0, SIGTERM);
778 exit(1);
779 }
780
781 //
782 // Fill out the packet with sessions from the session table...
783 // (not forgetting to leave space so we can get some tunnels in too )
784 while ( (p + sizeof(u32) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) {
785
786 if (!walk_session_number) // session #0 isn't valid.
787 ++walk_session_number;
788
789 if (count >= config->cluster_highest_sessionid) // If we're a small cluster, don't go wild.
790 break;
791
792 hb_add_type(&p, C_CSESSION, walk_session_number);
793 walk_session_number = (1+walk_session_number)%(config->cluster_highest_sessionid+1); // +1 avoids divide by zero.
794
795 ++count; // Count the number of extra sessions we're sending.
796 }
797
798 //
799 // Fill out the packet with tunnels from the tunnel table...
800 // This effectively means we walk the tunnel table more quickly
801 // than the session table. This is good because stuffing up a
802 // tunnel is a much bigger deal than stuffing up a session.
803 //
804 while ( (p + sizeof(u32) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {
805
806 if (!walk_tunnel_number) // tunnel #0 isn't valid.
807 ++walk_tunnel_number;
808
809 if (tcount >= config->cluster_highest_tunnelid)
810 break;
811
812 hb_add_type(&p, C_CTUNNEL, walk_tunnel_number);
813 walk_tunnel_number = (1+walk_tunnel_number)%(config->cluster_highest_tunnelid+1); // +1 avoids divide by zero.
814
815 ++tcount;
816 }
817
818 //
819 // Did we do something wrong?
820 if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
821 LOG(0, 0, 0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", p - buff);
822 kill(0, SIGTERM);
823 exit(1);
824 }
825
826 LOG(3, 0, 0, "Sending v%d heartbeat #%d, change #%llu with %d changes "
827 "(%d x-sess, %d x-tunnels, %d highsess, %d hightun, size %d)\n",
828 HB_VERSION, h.seq, h.table_version, config->cluster_num_changes,
829 count, tcount, config->cluster_highest_sessionid,
830 config->cluster_highest_tunnelid, (p-buff));
831
832 config->cluster_num_changes = 0;
833
834 send_heartbeat(h.seq, buff, (p-buff) ); // Send out the heartbeat to the cluster, keeping a copy of it.
835
836 config->cluster_seq_number = (config->cluster_seq_number+1)%HB_MAX_SEQ; // Next seq number to use.
837 }
838
839 //
840 // A structure of type 'type' has changed; Add it to the queue to send.
841 //
842 static int type_changed(int type, int id)
843 {
844 int i;
845
846 for (i = 0 ; i < config->cluster_num_changes ; ++i)
847 if ( cluster_changes[i].id == id &&
848 cluster_changes[i].type == type)
849 return 0; // Already marked for change.
850
851 cluster_changes[i].type = type;
852 cluster_changes[i].id = id;
853 ++config->cluster_num_changes;
854
855 if (config->cluster_num_changes > MAX_CHANGES)
856 cluster_heartbeat(); // flush now
857
858 return 1;
859 }
860
861
862 // A particular session has been changed!
863 int cluster_send_session(int sid)
864 {
865 if (!config->cluster_iam_master) {
866 LOG(0, sid, 0, "I'm not a master, but I just tried to change a session!\n");
867 return -1;
868 }
869
870 return type_changed(C_CSESSION, sid);
871 }
872
873 // A particular tunnel has been changed!
874 int cluster_send_tunnel(int tid)
875 {
876 if (!config->cluster_iam_master) {
877 LOG(0, 0, tid, "I'm not a master, but I just tried to change a tunnel!\n");
878 return -1;
879 }
880
881 return type_changed(C_CTUNNEL, tid);
882 }
883
884
885 //
886 // We're a master, and a slave has just told us that it's
887 // missed a packet. We'll resend it every packet since
888 // the last one it's seen.
889 //
890 static int cluster_catchup_slave(int seq, u32 slave)
891 {
892 int s;
893 int diff;
894
895 LOG(1, 0, 0, "Slave %s sent LASTSEEN with seq %d\n", fmtaddr(slave, 0), seq);
896
897 diff = config->cluster_seq_number - seq; // How many packet do we need to send?
898 if (diff < 0)
899 diff += HB_MAX_SEQ;
900
901 if (diff >= HB_HISTORY_SIZE) { // Ouch. We don't have the packet to send it!
902 LOG(0, 0, 0, "A slave asked for message %d when our seq number is %d. Killing it.\n",
903 seq, config->cluster_seq_number);
904 return peer_send_message(slave, C_KILL, seq, NULL, 0);// Kill the slave. Nothing else to do.
905 }
906
907 // Now resend every packet that it missed, in order.
908 while (seq != config->cluster_seq_number) {
909 s = seq%HB_HISTORY_SIZE;
910 if (seq != past_hearts[s].seq) {
911 LOG(0, 0, 0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n",
912 fmtaddr(slave, 0), seq, past_hearts[s].seq, s, config->cluster_seq_number);
913 return -1; // What to do here!?
914 }
915 peer_send_data(slave, past_hearts[s].data, past_hearts[s].size);
916 seq = (seq+1)%HB_MAX_SEQ; // Increment to next seq number.
917 }
918 return 0; // All good!
919 }
920
921 //
922 // We've heard from another peer! Add it to the list
923 // that we select from at election time.
924 //
925 static int cluster_add_peer(u32 peer, time_t basetime, pingt *pp, int size)
926 {
927 int i;
928 u32 clusterid;
929 pingt p;
930
931 // Allow for backward compatability.
932 // Just the ping packet into a new structure to allow
933 // for the possibility that we might have received
934 // more or fewer elements than we were expecting.
935 if (size > sizeof(p))
936 size = sizeof(p);
937
938 memset( (void*) &p, 0, sizeof(p) );
939 memcpy( (void*) &p, (void*) pp, size);
940
941 clusterid = p.addr;
942 if (clusterid != config->bind_address)
943 {
944 // Is this for us?
945 LOG(4, 0, 0, "Skipping ping from %s (different cluster)\n", fmtaddr(peer, 0));
946 return 0;
947 }
948
949 for (i = 0; i < num_peers ; ++i)
950 {
951 if (peers[i].peer != peer)
952 continue;
953
954 // This peer already exists. Just update the timestamp.
955 peers[i].basetime = basetime;
956 peers[i].timestamp = TIME;
957 peers[i].uptodate = !p.undef;
958 break;
959 }
960
961 // Is this the master shutting down??
962 if (peer == config->cluster_master_address && !basetime) {
963 LOG(3, 0, 0, "Master %s shutting down...\n", fmtaddr(config->cluster_master_address, 0));
964 config->cluster_master_address = 0;
965 config->cluster_last_hb = 0; // Force an election.
966 cluster_check_master();
967 return 0;
968 }
969
970 if (i >= num_peers)
971 {
972 LOG(4, 0, 0, "Adding %s as a peer\n", fmtaddr(peer, 0));
973
974 // Not found. Is there a stale slot to re-use?
975 for (i = 0; i < num_peers ; ++i)
976 {
977 if (!peers[i].basetime) // Shutdown
978 break;
979
980 if ((peers[i].timestamp + config->cluster_hb_timeout * 10) < TIME) // Stale.
981 break;
982 }
983
984 if (i >= CLUSTER_MAX_SIZE)
985 {
986 // Too many peers!!
987 LOG(0, 0, 0, "Tried to add %s as a peer, but I already have %d of them!\n", fmtaddr(peer, 0), i);
988 return -1;
989 }
990
991 peers[i].peer = peer;
992 peers[i].basetime = basetime;
993 peers[i].timestamp = TIME;
994 peers[i].uptodate = !p.undef;
995 if (i == num_peers)
996 ++num_peers;
997
998 LOG(1, 0, 0, "Added %s as a new peer. Now %d peers\n", fmtaddr(peer, 0), num_peers);
999 }
1000
1001 return 1;
1002 }
1003
1004 /* Handle the slave updating the byte counters for the master. */
1005 //
1006 // Note that we don't mark the session as dirty; We rely on
1007 // the slow table walk to propogate this back out to the slaves.
1008 //
1009 static int cluster_handle_bytes(char * data, int size)
1010 {
1011 bytest * b;
1012
1013 b = (bytest*) data;
1014
1015 LOG(3, 0, 0, "Got byte counter update (size %d)\n", size);
1016
1017 /* Loop around, adding the byte
1018 counts to each of the sessions. */
1019
1020 while (size >= sizeof(*b) ) {
1021 if (b->sid > MAXSESSION) {
1022 LOG(0, 0, 0, "Got C_BYTES with session #%d!\n", b->sid);
1023 return -1; /* Abort processing */
1024 }
1025
1026 session[b->sid].total_cin += b->in;
1027 session[b->sid].total_cout += b->out;
1028
1029 session[b->sid].cin += b->in;
1030 session[b->sid].cout += b->out;
1031 session[b->sid].last_packet = time_now; // Reset idle timer!
1032
1033 size -= sizeof(*b);
1034 ++b;
1035 }
1036
1037 if (size != 0)
1038 LOG(0, 0, 0, "Got C_BYTES with %d bytes of trailing junk!\n", size);
1039
1040 return size;
1041 }
1042
1043 //
1044 // Handle receiving a session structure in a heartbeat packet.
1045 //
1046 static int cluster_recv_session(int more , u8 * p)
1047 {
1048 if (more >= MAXSESSION) {
1049 LOG(0, 0, 0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
1050 return -1;
1051 }
1052
1053 if (session[more].tunnel == T_UNDEF) {
1054 if (config->cluster_iam_uptodate) { // Sanity.
1055 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined session!\n");
1056 } else {
1057 --config->cluster_undefined_sessions;
1058 }
1059 }
1060
1061 load_session(more, (sessiont*) p); // Copy session into session table..
1062
1063 LOG(5, more, 0, "Received session update (%d undef)\n", config->cluster_undefined_sessions);
1064
1065 if (!config->cluster_iam_uptodate)
1066 cluster_uptodate(); // Check to see if we're up to date.
1067
1068 return 0;
1069 }
1070
1071 static int cluster_recv_tunnel(int more, u8 *p)
1072 {
1073 if (more >= MAXTUNNEL) {
1074 LOG(0, 0, 0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
1075 return -1;
1076 }
1077
1078 if (tunnel[more].state == TUNNELUNDEF) {
1079 if (config->cluster_iam_uptodate) { // Sanity.
1080 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined tunnel!\n");
1081 } else {
1082 --config->cluster_undefined_tunnels;
1083 }
1084 }
1085
1086 memcpy(&tunnel[more], p, sizeof(tunnel[more]) );
1087
1088 //
1089 // Clear tunnel control messages. These are dynamically allocated.
1090 // If we get unlucky, this may cause the tunnel to drop!
1091 //
1092 tunnel[more].controls = tunnel[more].controle = NULL;
1093 tunnel[more].controlc = 0;
1094
1095 LOG(5, 0, more, "Received tunnel update\n");
1096
1097 if (!config->cluster_iam_uptodate)
1098 cluster_uptodate(); // Check to see if we're up to date.
1099
1100 return 0;
1101 }
1102
1103
1104 //
1105 // Process a heartbeat..
1106 //
1107 // v3: added interval, timeout
1108 // v4: added table_version
1109 static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32 addr)
1110 {
1111 heartt * h;
1112 int s = size - (p-data);
1113 int i, type;
1114
1115 #if HB_VERSION != 4
1116 # error "need to update cluster_process_heartbeat()"
1117 #endif
1118
1119 // we handle versions 3 through 4
1120 if (more < 3 || more > HB_VERSION) {
1121 LOG(0, 0, 0, "Received a heartbeat version that I don't support (%d)!\n", more);
1122 return -1; // Ignore it??
1123 }
1124
1125 // Ok. It's a heartbeat packet from a cluster master!
1126 if (s < sizeof(*h))
1127 goto shortpacket;
1128
1129 h = (heartt*) p;
1130 p += sizeof(*h);
1131 s -= sizeof(*h);
1132
1133 if (h->clusterid != config->bind_address)
1134 return -1; // It's not part of our cluster.
1135
1136 if (config->cluster_iam_master) { // Sanity...
1137 // Note that this MUST match the election process above!
1138
1139 LOG(0, 0, 0, "I just got a heartbeat from master %s, but _I_ am the master!\n", fmtaddr(addr, 0));
1140 if (!h->basetime) {
1141 LOG(0, 0, 0, "Heartbeat with zero basetime! Ignoring\n");
1142 return -1; // Skip it.
1143 }
1144 if (more >= 4 && h->table_version > config->cluster_table_version) {
1145 LOG(0, 0, 0, "They've seen more state changes (%llu vs my %llu) so I'm gone!\n",
1146 h->table_version, config->cluster_table_version);
1147
1148 kill(0, SIGTERM);
1149 exit(1);
1150 }
1151 if (basetime > h->basetime) {
1152 LOG(0, 0, 0, "They're an older master than me so I'm gone!\n");
1153 kill(0, SIGTERM);
1154 exit(1);
1155 }
1156 if (basetime == h->basetime && my_address < addr) { // Tie breaker.
1157 LOG(0, 0, 0, "They're a higher IP address than me, so I'm gone!\n");
1158 kill(0, SIGTERM);
1159 exit(1);
1160 }
1161 return -1; // Skip it.
1162 }
1163
1164 if (config->cluster_seq_number == -1) // Don't have one. Just align to the master...
1165 config->cluster_seq_number = h->seq;
1166
1167 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1168
1169 if (config->cluster_seq_number != h->seq) { // Out of sequence heartbeat!
1170 LOG(1, 0, 0, "HB: Got seq# %d but was expecting %d. asking for resend.\n", h->seq, config->cluster_seq_number);
1171
1172 peer_send_message(addr, C_LASTSEEN, config->cluster_seq_number, NULL, 0);
1173
1174 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1175
1176 // Just drop the packet. The master will resend it as part of the catchup.
1177
1178 return 0;
1179 }
1180 // Save the packet in our buffer.
1181 // This is needed in case we become the master.
1182 config->cluster_seq_number = (h->seq+1)%HB_MAX_SEQ;
1183 i = h->seq % HB_HISTORY_SIZE;
1184 past_hearts[i].seq = h->seq;
1185 past_hearts[i].size = size;
1186 memcpy(&past_hearts[i].data, data, size); // Save it.
1187
1188
1189 // Check that we don't have too many undefined sessions, and
1190 // that the free session pointer is correct.
1191 cluster_check_sessions(h->highsession, h->freesession, h->hightunnel);
1192
1193 if (h->interval != config->cluster_hb_interval)
1194 {
1195 LOG(2, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
1196 h->interval, config->cluster_hb_interval);
1197
1198 config->cluster_hb_interval = h->interval;
1199 }
1200
1201 if (h->timeout != config->cluster_hb_timeout)
1202 {
1203 LOG(2, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
1204 h->timeout, config->cluster_hb_timeout);
1205
1206 config->cluster_hb_timeout = h->timeout;
1207 }
1208
1209 // Ok. process the packet...
1210 while ( s > 0) {
1211
1212 type = * ((u32*) p);
1213 p += sizeof(u32);
1214 s -= sizeof(u32);
1215
1216 more = * ((u32*) p);
1217 p += sizeof(u32);
1218 s -= sizeof(u32);
1219
1220 switch (type) {
1221 case C_CSESSION: { // Compressed session structure.
1222 u8 c [ sizeof(sessiont) + 2];
1223 int size;
1224 u8 * orig_p = p;
1225
1226 size = rle_decompress((u8 **) &p, s, c, sizeof(c) );
1227 s -= (p - orig_p);
1228
1229 if (size != sizeof(sessiont) ) { // Ouch! Very very bad!
1230 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1231 // Now what? Should exit! No-longer up to date!
1232 break;
1233 }
1234
1235 cluster_recv_session(more, c);
1236 break;
1237 }
1238 case C_SESSION:
1239 if ( s < sizeof(session[more]))
1240 goto shortpacket;
1241
1242 cluster_recv_session(more, p);
1243
1244 p += sizeof(session[more]);
1245 s -= sizeof(session[more]);
1246 break;
1247
1248 case C_CTUNNEL: { // Compressed tunnel structure.
1249 u8 c [ sizeof(tunnelt) + 2];
1250 int size;
1251 u8 * orig_p = p;
1252
1253 size = rle_decompress( (u8 **) &p, s, c, sizeof(c) );
1254 s -= (p - orig_p);
1255
1256 if (size != sizeof(tunnelt) ) { // Ouch! Very very bad!
1257 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1258 // Now what? Should exit! No-longer up to date!
1259 break;
1260 }
1261
1262 cluster_recv_tunnel(more, c);
1263 break;
1264
1265 }
1266 case C_TUNNEL:
1267 if ( s < sizeof(tunnel[more]))
1268 goto shortpacket;
1269
1270 cluster_recv_tunnel(more, p);
1271
1272 p += sizeof(tunnel[more]);
1273 s -= sizeof(tunnel[more]);
1274 break;
1275 default:
1276 LOG(0, 0, 0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type);
1277 return -1; // can't process any more of the packet!!
1278 }
1279 }
1280
1281 if (config->cluster_master_address != addr)
1282 {
1283 LOG(0, 0, 0, "My master just changed from %s to %s!\n",
1284 fmtaddr(config->cluster_master_address, 0), fmtaddr(addr, 1));
1285
1286 config->cluster_master_address = addr;
1287 }
1288
1289 config->cluster_last_hb = TIME; // Successfully received a heartbeat!
1290 config->cluster_table_version = h->table_version;
1291 return 0;
1292
1293 shortpacket:
1294 LOG(0, 0, 0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
1295 return -1;
1296 }
1297
1298 //
1299 // We got a packet on the cluster port!
1300 // Handle pings, lastseens, and heartbeats!
1301 //
1302 int processcluster(char * data, int size, u32 addr)
1303 {
1304 int type, more;
1305 char * p = data;
1306 int s = size;
1307
1308 if (addr == my_address)
1309 return -1; // Ignore it. Something looped back the multicast!
1310
1311 LOG(5, 0, 0, "Process cluster: %d bytes from %s\n", size, fmtaddr(addr, 0));
1312
1313 if (s <= 0) // Any data there??
1314 return -1;
1315
1316 if (s < 8)
1317 goto shortpacket;
1318
1319 type = * ((u32*) p);
1320 p += sizeof(u32);
1321 s -= sizeof(u32);
1322
1323 more = * ((u32*) p);
1324 p += sizeof(u32);
1325 s -= sizeof(u32);
1326
1327 switch (type) {
1328 case C_PING: // Update the peers table.
1329 return cluster_add_peer(addr, more, (pingt*)p, s);
1330
1331 case C_LASTSEEN: // Catch up a slave (slave missed a packet).
1332 return cluster_catchup_slave(more, addr);
1333
1334 case C_FORWARD: { // Forwarded control packet. pass off to processudp.
1335 struct sockaddr_in a;
1336 a.sin_addr.s_addr = more;
1337
1338 a.sin_port = * (int*) p;
1339 s -= sizeof(int);
1340 p += sizeof(int);
1341
1342 if (!config->cluster_iam_master) { // huh?
1343 LOG(0, 0, 0, "I'm not the master, but I got a C_FORWARD from %s?\n", fmtaddr(addr, 0));
1344 return -1;
1345 }
1346
1347 LOG(4, 0, 0, "Got a forwarded packet... (%s:%d)\n", fmtaddr(more, 0), a.sin_port);
1348 STAT(recv_forward);
1349 processudp(p, s, &a);
1350 return 0;
1351 }
1352 case C_THROTTLE: { // Receive a forwarded packet from a slave.
1353 if (!config->cluster_iam_master) {
1354 LOG(0, 0, 0, "I'm not the master, but I got a C_THROTTLE from %s?\n", fmtaddr(addr, 0));
1355 return -1;
1356 }
1357
1358 tbf_queue_packet(more, p, s); // The TBF id tells wether it goes in or out.
1359 return 0;
1360 }
1361 case C_GARDEN:
1362 // Receive a walled garden packet from a slave.
1363 if (!config->cluster_iam_master) {
1364 LOG(0, 0, 0, "I'm not the master, but I got a C_GARDEN from %s?\n", fmtaddr(addr, 0));
1365 return -1;
1366 }
1367
1368 tun_write(p, s);
1369 return 0;
1370
1371 case C_BYTES:
1372 return cluster_handle_bytes(p, s);
1373
1374 case C_KILL: // The master asked us to die!? (usually because we're too out of date).
1375 if (config->cluster_iam_master) {
1376 LOG(0, 0, 0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", fmtaddr(addr, 0), more);
1377 return -1;
1378 }
1379 if (more != config->cluster_seq_number) {
1380 LOG(0, 0, 0, "The master asked us to die but the seq number didn't match!?\n");
1381 return -1;
1382 }
1383
1384 if (addr != config->cluster_master_address) {
1385 LOG(0, 0, 0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%s)\n",
1386 fmtaddr(addr, 0), fmtaddr(config->cluster_master_address, 1));
1387 // We can only warn about it. The master might really have switched!
1388 }
1389
1390 LOG(0, 0, 0, "Received a valid C_KILL: I'm going to die now.\n");
1391 kill(0, SIGTERM);
1392 exit(0); // Lets be paranoid;
1393 return -1; // Just signalling the compiler.
1394
1395 case C_HEARTBEAT:
1396 LOG(4, 0, 0, "Got a heartbeat from %s\n", fmtaddr(addr, 0));
1397 return cluster_process_heartbeat(data, size, more, p, addr);
1398
1399 default:
1400 LOG(0, 0, 0, "Strange type packet received on cluster socket (%d)\n", type);
1401 return -1;
1402 }
1403 return 0;
1404
1405 shortpacket:
1406 LOG(0, 0, 0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n");
1407 return -1;
1408 }
1409
1410 //====================================================================================================
1411
1412 int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
1413 {
1414 int i;
1415
1416 if (CLI_HELP_REQUESTED)
1417 return CLI_HELP_NO_ARGS;
1418
1419 cli_print(cli, "Cluster status : %s", config->cluster_iam_master ? "Master" : "Slave" );
1420 cli_print(cli, "My address : %s", fmtaddr(my_address, 0));
1421 cli_print(cli, "VIP address : %s", fmtaddr(config->bind_address, 0));
1422 cli_print(cli, "Multicast address: %s", fmtaddr(config->cluster_address, 0));
1423 cli_print(cli, "Multicast i'face : %s", config->cluster_interface);
1424
1425 if (!config->cluster_iam_master) {
1426 cli_print(cli, "My master : %s (last heartbeat %.1f seconds old)",
1427 config->cluster_master_address
1428 ? fmtaddr(config->cluster_master_address, 0)
1429 : "Not defined",
1430 0.1 * (TIME - config->cluster_last_hb));
1431 cli_print(cli, "Uptodate : %s", config->cluster_iam_uptodate ? "Yes" : "No");
1432 cli_print(cli, "Table version # : %llu", config->cluster_table_version);
1433 cli_print(cli, "Next sequence number expected: %d", config->cluster_seq_number);
1434 cli_print(cli, "%d sessions undefined of %d", config->cluster_undefined_sessions, config->cluster_highest_sessionid);
1435 cli_print(cli, "%d tunnels undefined of %d", config->cluster_undefined_tunnels, config->cluster_highest_tunnelid);
1436 } else {
1437 cli_print(cli, "Table version # : %llu", config->cluster_table_version);
1438 cli_print(cli, "Next heartbeat # : %d", config->cluster_seq_number);
1439 cli_print(cli, "Highest session : %d", config->cluster_highest_sessionid);
1440 cli_print(cli, "Highest tunnel : %d", config->cluster_highest_tunnelid);
1441 cli_print(cli, "%d changes queued for sending", config->cluster_num_changes);
1442 }
1443 cli_print(cli, "%d peers.", num_peers);
1444
1445 if (num_peers)
1446 cli_print(cli, "%20s %10s %8s", "Address", "Basetime", "Age");
1447 for (i = 0; i < num_peers; ++i) {
1448 cli_print(cli, "%20s %10d %8d", fmtaddr(peers[i].peer, 0),
1449 peers[i].basetime, TIME - peers[i].timestamp);
1450 }
1451 return CLI_OK;
1452 }
1453
1454 //
1455 // Simple run-length-encoding compression.
1456 // Format is
1457 // 1 byte < 128 = count of non-zero bytes following. // Not legal to be zero.
1458 // n non-zero bytes;
1459 // or
1460 // 1 byte > 128 = (count - 128) run of zero bytes. //
1461 // repeat.
1462 // count == 0 indicates end of compressed stream.
1463 //
1464 // Compress from 'src' into 'dst'. return number of bytes
1465 // used from 'dst'.
1466 // Updates *src_p to indicate 1 past last bytes used.
1467 //
1468 // We could get an extra byte in the zero runs by storing (count-1)
1469 // but I'm playing it safe.
1470 //
1471 // Worst case is a 50% expansion in space required (trying to
1472 // compress { 0x00, 0x01 } * N )
1473 static int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize)
1474 {
1475 int count;
1476 int orig_dsize = dsize;
1477 u8 * x,*src;
1478 src = *src_p;
1479
1480 while (ssize > 0 && dsize > 2) {
1481 count = 0;
1482 x = dst++; --dsize; // Reserve space for count byte..
1483
1484 if (*src) { // Copy a run of non-zero bytes.
1485 while (*src && count < 127 && ssize > 0 && dsize > 1) { // Count number of non-zero bytes.
1486 *dst++ = *src++;
1487 --dsize; --ssize;
1488 ++count;
1489 }
1490 *x = count; // Store number of non-zero bytes. Guarenteed to be non-zero!
1491
1492 } else { // Compress a run of zero bytes.
1493 while (*src == 0 && count < 127 && ssize > 0) {
1494 ++src;
1495 --ssize;
1496 ++count;
1497 }
1498 *x = count | 0x80 ;
1499 }
1500 }
1501
1502 *dst++ = 0x0; // Add Stop byte.
1503 --dsize;
1504
1505 *src_p = src;
1506 return (orig_dsize - dsize);
1507 }
1508
1509 //
1510 // Decompress the buffer into **p.
1511 // 'psize' is the size of the decompression buffer available.
1512 //
1513 // Returns the number of bytes decompressed.
1514 //
1515 // Decompresses from '*src_p' into 'dst'.
1516 // Return the number of dst bytes used.
1517 // Updates the 'src_p' pointer to point to the
1518 // first un-used byte.
1519 static int rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize)
1520 {
1521 int count;
1522 int orig_dsize = dsize;
1523 char * src = *src_p;
1524
1525 while (ssize >0 && dsize > 0) { // While there's more to decompress, and there's room in the decompress buffer...
1526 count = *src++; --ssize; // get the count byte from the source.
1527 if (count == 0x0) // End marker reached? If so, finish.
1528 break;
1529
1530 if (count & 0x80) { // Decompress a run of zeros
1531 for (count &= 0x7f ; count > 0 && dsize > 0; --count) {
1532 *dst++ = 0x0;
1533 --dsize;
1534 }
1535 } else { // Copy run of non-zero bytes.
1536 for ( ; count > 0 && ssize && dsize; --count) { // Copy non-zero bytes across.
1537 *dst++ = *src++;
1538 --ssize; --dsize;
1539 }
1540 }
1541 }
1542 *src_p = src;
1543 return (orig_dsize - dsize);
1544 }