Use session[s].opened consistently when checking for in-use sessions (rather than...
[l2tpns.git] / cluster.c
1 // L2TPNS Clustering Stuff
2
3 char const *cvs_id_cluster = "$Id: cluster.c,v 1.26.2.3 2005/02/14 05:59:27 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.
358 //
359 void cluster_send_ping(time_t basetime)
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 LOG(5, 0, 0, "Sending cluster ping...\n");
369
370 x.ver = 1;
371 x.addr = config->bind_address;
372 x.undef = config->cluster_undefined_sessions + config->cluster_undefined_tunnels;
373 x.basetime = basetime;
374
375 add_type(&p, C_PING, basetime, (char *) &x, sizeof(x));
376 cluster_send_data(buff, (p-buff) );
377 }
378
379 //
380 // Walk the session counters looking for non-zero ones to send
381 // to the master. We send up to 600 of them at one time.
382 // We examine a maximum of 3000 sessions.
383 // (50k max session should mean that we normally
384 // examine the entire session table every 25 seconds).
385
386 #define MAX_B_RECS (600)
387 void master_update_counts(void)
388 {
389 int i, c;
390 bytest b[MAX_B_RECS+1];
391
392 if (config->cluster_iam_master) // Only happens on the slaves.
393 return;
394
395 if (!config->cluster_master_address) // If we don't have a master, skip it for a while.
396 return;
397
398 i = MAX_B_RECS * 5; // Examine max 2000 sessions;
399 if (config->cluster_highest_sessionid > i)
400 i = config->cluster_highest_sessionid;
401
402 for ( c = 0; i > 0 ; --i) {
403 // Next session to look at.
404 walk_session_number++;
405 if ( walk_session_number > config->cluster_highest_sessionid)
406 walk_session_number = 1;
407
408 if (!sess_local[walk_session_number].cin && !sess_local[walk_session_number].cout)
409 continue; // Unused. Skip it.
410
411 b[c].sid = walk_session_number;
412 b[c].in = sess_local[walk_session_number].cin;
413 b[c].out = sess_local[walk_session_number].cout;
414
415 if (++c > MAX_B_RECS) // Send a max of 400 elements in a packet.
416 break;
417
418 // Reset counters.
419 sess_local[walk_session_number].cin = sess_local[walk_session_number].cout = 0;
420 }
421
422 if (!c) // Didn't find any that changes. Get out of here!
423 return;
424
425
426 // Forward the data to the master.
427 LOG(4, 0, 0, "Sending byte counters to master (%d elements)\n", c);
428 peer_send_message(config->cluster_master_address, C_BYTES, c, (char *) &b, sizeof(b[0]) * c);
429 return;
430 }
431
432 //
433 // On the master, check how our slaves are going. If
434 // one of them's not up-to-date we'll heartbeat faster.
435 // If we don't have any of them, then we need to turn
436 // on our own packet handling!
437 //
438 void cluster_check_slaves(void)
439 {
440 int i;
441 static int have_peers = 0;
442 int had_peers = have_peers;
443 clockt t = TIME;
444
445 if (!config->cluster_iam_master)
446 return; // Only runs on the master...
447
448 config->cluster_iam_uptodate = 1; // cleared in loop below
449
450 for (i = have_peers = 0; i < num_peers; i++)
451 {
452 if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
453 continue; // Stale peer! Skip them.
454
455 if (!peers[i].basetime)
456 continue; // Shutdown peer! Skip them.
457
458 if (peers[i].uptodate)
459 have_peers = 1;
460
461 if (!peers[i].uptodate)
462 config->cluster_iam_uptodate = 0; // Start fast heartbeats
463 }
464
465 #ifdef BGP
466 // in a cluster, withdraw/add routes when we get a peer/lose all peers
467 if (bgp_configured && have_peers != had_peers)
468 bgp_enable_routing(!have_peers);
469 #endif /* BGP */
470 }
471
472 //
473 // Check that we have a master. If it's been too
474 // long since we heard from a master then hold an election.
475 //
476 void cluster_check_master(void)
477 {
478 int i, count, tcount, high_unique_id = 0;
479 int last_free = 0;
480 clockt t = TIME;
481 static int probed = 0;
482
483 if (config->cluster_iam_master)
484 return; // Only runs on the slaves...
485
486 // If the master is late (missed 2 hearbeats by a second and a
487 // hair) it may be that the switch has dropped us from the
488 // multicast group, try unicasting one probe to the master
489 // which will hopefully respond with a unicast heartbeat that
490 // will allow us to limp along until the querier next runs.
491 if (TIME > (config->cluster_last_hb + 2 * config->cluster_hb_interval + 11))
492 {
493 if (!probed && config->cluster_master_address)
494 {
495 probed = 1;
496 LOG(1, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
497 0.1 * (TIME - (config->cluster_last_hb + config->cluster_hb_interval)));
498
499 peer_send_message(config->cluster_master_address,
500 C_LASTSEEN, config->cluster_seq_number, NULL, 0);
501 }
502 } else { // We got a recent heartbeat; reset the probe flag.
503 probed = 0;
504 }
505
506 if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout))
507 return; // Everything's ok!
508
509 config->cluster_last_hb = TIME + 1; // Just the one election thanks.
510
511 LOG(0, 0, 0, "Master timed out! Holding election...\n");
512
513 for (i = 0; i < num_peers; i++)
514 {
515 if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
516 continue; // Stale peer! Skip them.
517
518 if (!peers[i].basetime)
519 continue; // Shutdown peer! Skip them.
520
521 if (peers[i].basetime < basetime) {
522 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
523 return; // They'll win the election. Get out of here.
524 }
525
526 if (peers[i].basetime == basetime &&
527 peers[i].peer > my_address) {
528 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
529 return; // They'll win the election. Wait for them to come up.
530 }
531 }
532
533 // Wow. it's been ages since I last heard a heartbeat
534 // and I'm better than an of my peers so it's time
535 // to become a master!!!
536
537 config->cluster_iam_master = 1;
538 config->cluster_master_address = 0;
539
540 LOG(0, 0, 0, "I am declaring myself the master!\n");
541
542 if (config->cluster_seq_number == -1)
543 config->cluster_seq_number = 0;
544
545 //
546 // Go through and mark all the tunnels as defined.
547 // Count the highest used tunnel number as well.
548 //
549 config->cluster_highest_tunnelid = 0;
550 for (i = 0, tcount = 0; i < MAXTUNNEL; ++i) {
551 if (tunnel[i].state == TUNNELUNDEF)
552 tunnel[i].state = TUNNELFREE;
553
554 if (tunnel[i].state != TUNNELFREE && i > config->cluster_highest_tunnelid)
555 config->cluster_highest_tunnelid = i;
556 }
557
558 //
559 // Go through and mark all the sessions as being defined.
560 // reset the idle timeouts.
561 // add temporary byte counters to permanent ones.
562 // Re-string the free list.
563 // Find the ID of the highest session.
564 last_free = 0;
565 high_unique_id = 0;
566 config->cluster_highest_sessionid = 0;
567 for (i = 0, count = 0; i < MAXSESSION; ++i) {
568 if (session[i].tunnel == T_UNDEF) {
569 session[i].tunnel = T_FREE;
570 ++count;
571 }
572
573 if (!session[i].opened) { // Unused session. Add to free list.
574 memset(&session[i], 0, sizeof(session[i]));
575 session[i].tunnel = T_FREE;
576 session[last_free].next = i;
577 session[i].next = 0;
578 last_free = i;
579 continue;
580 }
581
582 // Reset all the idle timeouts..
583 session[i].last_packet = time_now;
584
585 // Accumulate un-sent byte counters.
586 session[i].cin += sess_local[i].cin;
587 session[i].cout += sess_local[i].cout;
588 session[i].total_cin += sess_local[i].cin;
589 session[i].total_cout += sess_local[i].cout;
590
591 sess_local[i].cin = sess_local[i].cout = 0;
592
593 session[i].radius = 0; // Reset authentication as the radius blocks aren't up to date.
594
595 if (session[i].unique_id >= high_unique_id) // This is different to the index into the session table!!!
596 high_unique_id = session[i].unique_id+1;
597
598 session[i].tbf_in = session[i].tbf_out = 0; // Remove stale pointers from old master.
599 throttle_session(i, session[i].throttle_in, session[i].throttle_out);
600
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 if (session[i].tunnel == T_UNDEF) session[i].tunnel = T_FREE; // Defined.
654 continue;
655 }
656
657 if (session[i].tunnel == T_UNDEF)
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 if (tunnel[i].state == TUNNELUNDEF) tunnel[i].state = TUNNELFREE; // Defined.
667 continue;
668 }
669
670 if (tunnel[i].state == TUNNELUNDEF)
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 uint8_t c[sizeof(sessiont) * 2]; // Bigger than worst case.
692 uint8_t *d = (uint8_t *) &session[id];
693 uint8_t *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 uint8_t c[sizeof(tunnelt) * 2]; // Bigger than worst case.
712 uint8_t *d = (uint8_t *) &tunnel[id];
713 uint8_t *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", (int) (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(uint32_t) * 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(uint32_t) * 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", (int) (p - buff));
822 kill(0, SIGTERM);
823 exit(1);
824 }
825
826 LOG(3, 0, 0, "Sending v%d heartbeat #%d, change #%" PRIu64 " 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, (int) (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, in_addr_t 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(in_addr_t peer, time_t basetime, pingt *pp, int size)
926 {
927 int i;
928 in_addr_t 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, uint8_t *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, uint8_t *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(uint8_t *data, int size, int more, uint8_t *p, in_addr_t 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
1145 if (more >= 4) {
1146 if (h->table_version > config->cluster_table_version) {
1147 LOG(0, 0, 0, "They've seen more state changes (%" PRIu64 " vs my %" PRIu64 ") so I'm gone!\n",
1148 h->table_version, config->cluster_table_version);
1149
1150 kill(0, SIGTERM);
1151 exit(1);
1152 }
1153 if (h->table_version < config->cluster_table_version)
1154 return -1;
1155 }
1156
1157 if (basetime > h->basetime) {
1158 LOG(0, 0, 0, "They're an older master than me so I'm gone!\n");
1159 kill(0, SIGTERM);
1160 exit(1);
1161 }
1162
1163 if (basetime < h->basetime)
1164 return -1;
1165
1166 if (my_address < addr) { // Tie breaker.
1167 LOG(0, 0, 0, "They're a higher IP address than me, so I'm gone!\n");
1168 kill(0, SIGTERM);
1169 exit(1);
1170 }
1171
1172 return -1; // Skip it.
1173 }
1174
1175 if (config->cluster_seq_number == -1) // Don't have one. Just align to the master...
1176 config->cluster_seq_number = h->seq;
1177
1178 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1179
1180 if (config->cluster_seq_number != h->seq) { // Out of sequence heartbeat!
1181 static int lastseen_seq = 0;
1182 static time_t lastseen_time = 0;
1183
1184 // limit to once per second for a particular seq#
1185 int ask = (config->cluster_seq_number != lastseen_seq || time_now != lastseen_time);
1186
1187 LOG(1, 0, 0, "HB: Got seq# %d but was expecting %d. %s.\n",
1188 h->seq, config->cluster_seq_number,
1189 ask ? "Asking for resend" : "Ignoring");
1190
1191 if (ask)
1192 {
1193 lastseen_seq = config->cluster_seq_number;
1194 lastseen_time = time_now;
1195 peer_send_message(addr, C_LASTSEEN, config->cluster_seq_number, NULL, 0);
1196 }
1197
1198 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1199
1200 // Just drop the packet. The master will resend it as part of the catchup.
1201
1202 return 0;
1203 }
1204 // Save the packet in our buffer.
1205 // This is needed in case we become the master.
1206 config->cluster_seq_number = (h->seq+1)%HB_MAX_SEQ;
1207 i = h->seq % HB_HISTORY_SIZE;
1208 past_hearts[i].seq = h->seq;
1209 past_hearts[i].size = size;
1210 memcpy(&past_hearts[i].data, data, size); // Save it.
1211
1212
1213 // Check that we don't have too many undefined sessions, and
1214 // that the free session pointer is correct.
1215 cluster_check_sessions(h->highsession, h->freesession, h->hightunnel);
1216
1217 if (h->interval != config->cluster_hb_interval)
1218 {
1219 LOG(2, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
1220 h->interval, config->cluster_hb_interval);
1221
1222 config->cluster_hb_interval = h->interval;
1223 }
1224
1225 if (h->timeout != config->cluster_hb_timeout)
1226 {
1227 LOG(2, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
1228 h->timeout, config->cluster_hb_timeout);
1229
1230 config->cluster_hb_timeout = h->timeout;
1231 }
1232
1233 // Ok. process the packet...
1234 while ( s > 0) {
1235
1236 type = *((uint32_t *) p);
1237 p += sizeof(uint32_t);
1238 s -= sizeof(uint32_t);
1239
1240 more = *((uint32_t *) p);
1241 p += sizeof(uint32_t);
1242 s -= sizeof(uint32_t);
1243
1244 switch (type) {
1245 case C_CSESSION: { // Compressed session structure.
1246 uint8_t c[ sizeof(sessiont) + 2];
1247 int size;
1248 uint8_t *orig_p = p;
1249
1250 size = rle_decompress((uint8_t **) &p, s, c, sizeof(c) );
1251 s -= (p - orig_p);
1252
1253 if (size != sizeof(sessiont) ) { // Ouch! Very very bad!
1254 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1255 // Now what? Should exit! No-longer up to date!
1256 break;
1257 }
1258
1259 cluster_recv_session(more, c);
1260 break;
1261 }
1262 case C_SESSION:
1263 if ( s < sizeof(session[more]))
1264 goto shortpacket;
1265
1266 cluster_recv_session(more, p);
1267
1268 p += sizeof(session[more]);
1269 s -= sizeof(session[more]);
1270 break;
1271
1272 case C_CTUNNEL: { // Compressed tunnel structure.
1273 uint8_t c[ sizeof(tunnelt) + 2];
1274 int size;
1275 uint8_t *orig_p = p;
1276
1277 size = rle_decompress((uint8_t **) &p, s, c, sizeof(c));
1278 s -= (p - orig_p);
1279
1280 if (size != sizeof(tunnelt) ) { // Ouch! Very very bad!
1281 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1282 // Now what? Should exit! No-longer up to date!
1283 break;
1284 }
1285
1286 cluster_recv_tunnel(more, c);
1287 break;
1288
1289 }
1290 case C_TUNNEL:
1291 if ( s < sizeof(tunnel[more]))
1292 goto shortpacket;
1293
1294 cluster_recv_tunnel(more, p);
1295
1296 p += sizeof(tunnel[more]);
1297 s -= sizeof(tunnel[more]);
1298 break;
1299 default:
1300 LOG(0, 0, 0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type);
1301 return -1; // can't process any more of the packet!!
1302 }
1303 }
1304
1305 if (config->cluster_master_address != addr)
1306 {
1307 LOG(0, 0, 0, "My master just changed from %s to %s!\n",
1308 fmtaddr(config->cluster_master_address, 0), fmtaddr(addr, 1));
1309
1310 config->cluster_master_address = addr;
1311 }
1312
1313 config->cluster_last_hb = TIME; // Successfully received a heartbeat!
1314 config->cluster_table_version = h->table_version;
1315 return 0;
1316
1317 shortpacket:
1318 LOG(0, 0, 0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
1319 return -1;
1320 }
1321
1322 //
1323 // We got a packet on the cluster port!
1324 // Handle pings, lastseens, and heartbeats!
1325 //
1326 int processcluster(char *data, int size, in_addr_t addr)
1327 {
1328 int type, more;
1329 char *p = data;
1330 int s = size;
1331
1332 if (addr == my_address)
1333 return -1; // Ignore it. Something looped back the multicast!
1334
1335 LOG(5, 0, 0, "Process cluster: %d bytes from %s\n", size, fmtaddr(addr, 0));
1336
1337 if (s <= 0) // Any data there??
1338 return -1;
1339
1340 if (s < 8)
1341 goto shortpacket;
1342
1343 type = *((uint32_t *) p);
1344 p += sizeof(uint32_t);
1345 s -= sizeof(uint32_t);
1346
1347 more = *((uint32_t *) p);
1348 p += sizeof(uint32_t);
1349 s -= sizeof(uint32_t);
1350
1351 switch (type) {
1352 case C_PING: // Update the peers table.
1353 return cluster_add_peer(addr, more, (pingt *) p, s);
1354
1355 case C_LASTSEEN: // Catch up a slave (slave missed a packet).
1356 return cluster_catchup_slave(more, addr);
1357
1358 case C_FORWARD: { // Forwarded control packet. pass off to processudp.
1359 struct sockaddr_in a;
1360 a.sin_addr.s_addr = more;
1361
1362 a.sin_port = *(int *) p;
1363 s -= sizeof(int);
1364 p += sizeof(int);
1365
1366 if (!config->cluster_iam_master) { // huh?
1367 LOG(0, 0, 0, "I'm not the master, but I got a C_FORWARD from %s?\n", fmtaddr(addr, 0));
1368 return -1;
1369 }
1370
1371 LOG(4, 0, 0, "Got a forwarded packet... (%s:%d)\n", fmtaddr(more, 0), a.sin_port);
1372 STAT(recv_forward);
1373 processudp(p, s, &a);
1374 return 0;
1375 }
1376 case C_THROTTLE: { // Receive a forwarded packet from a slave.
1377 if (!config->cluster_iam_master) {
1378 LOG(0, 0, 0, "I'm not the master, but I got a C_THROTTLE from %s?\n", fmtaddr(addr, 0));
1379 return -1;
1380 }
1381
1382 tbf_queue_packet(more, p, s); // The TBF id tells wether it goes in or out.
1383 return 0;
1384 }
1385 case C_GARDEN:
1386 // Receive a walled garden packet from a slave.
1387 if (!config->cluster_iam_master) {
1388 LOG(0, 0, 0, "I'm not the master, but I got a C_GARDEN from %s?\n", fmtaddr(addr, 0));
1389 return -1;
1390 }
1391
1392 tun_write(p, s);
1393 return 0;
1394
1395 case C_BYTES:
1396 return cluster_handle_bytes(p, s);
1397
1398 case C_KILL: // The master asked us to die!? (usually because we're too out of date).
1399 if (config->cluster_iam_master) {
1400 LOG(0, 0, 0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", fmtaddr(addr, 0), more);
1401 return -1;
1402 }
1403 if (more != config->cluster_seq_number) {
1404 LOG(0, 0, 0, "The master asked us to die but the seq number didn't match!?\n");
1405 return -1;
1406 }
1407
1408 if (addr != config->cluster_master_address) {
1409 LOG(0, 0, 0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%s)\n",
1410 fmtaddr(addr, 0), fmtaddr(config->cluster_master_address, 1));
1411 // We can only warn about it. The master might really have switched!
1412 }
1413
1414 LOG(0, 0, 0, "Received a valid C_KILL: I'm going to die now.\n");
1415 kill(0, SIGTERM);
1416 exit(0); // Lets be paranoid;
1417 return -1; // Just signalling the compiler.
1418
1419 case C_HEARTBEAT:
1420 LOG(4, 0, 0, "Got a heartbeat from %s\n", fmtaddr(addr, 0));
1421 return cluster_process_heartbeat(data, size, more, p, addr);
1422
1423 default:
1424 LOG(0, 0, 0, "Strange type packet received on cluster socket (%d)\n", type);
1425 return -1;
1426 }
1427 return 0;
1428
1429 shortpacket:
1430 LOG(0, 0, 0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n");
1431 return -1;
1432 }
1433
1434 //====================================================================================================
1435
1436 int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
1437 {
1438 int i;
1439
1440 if (CLI_HELP_REQUESTED)
1441 return CLI_HELP_NO_ARGS;
1442
1443 cli_print(cli, "Cluster status : %s", config->cluster_iam_master ? "Master" : "Slave" );
1444 cli_print(cli, "My address : %s", fmtaddr(my_address, 0));
1445 cli_print(cli, "VIP address : %s", fmtaddr(config->bind_address, 0));
1446 cli_print(cli, "Multicast address: %s", fmtaddr(config->cluster_address, 0));
1447 cli_print(cli, "Multicast i'face : %s", config->cluster_interface);
1448
1449 if (!config->cluster_iam_master) {
1450 cli_print(cli, "My master : %s (last heartbeat %.1f seconds old)",
1451 config->cluster_master_address
1452 ? fmtaddr(config->cluster_master_address, 0)
1453 : "Not defined",
1454 0.1 * (TIME - config->cluster_last_hb));
1455 cli_print(cli, "Uptodate : %s", config->cluster_iam_uptodate ? "Yes" : "No");
1456 cli_print(cli, "Table version # : %" PRIu64, config->cluster_table_version);
1457 cli_print(cli, "Next sequence number expected: %d", config->cluster_seq_number);
1458 cli_print(cli, "%d sessions undefined of %d", config->cluster_undefined_sessions, config->cluster_highest_sessionid);
1459 cli_print(cli, "%d tunnels undefined of %d", config->cluster_undefined_tunnels, config->cluster_highest_tunnelid);
1460 } else {
1461 cli_print(cli, "Table version # : %" PRIu64, config->cluster_table_version);
1462 cli_print(cli, "Next heartbeat # : %d", config->cluster_seq_number);
1463 cli_print(cli, "Highest session : %d", config->cluster_highest_sessionid);
1464 cli_print(cli, "Highest tunnel : %d", config->cluster_highest_tunnelid);
1465 cli_print(cli, "%d changes queued for sending", config->cluster_num_changes);
1466 }
1467 cli_print(cli, "%d peers.", num_peers);
1468
1469 if (num_peers)
1470 cli_print(cli, "%20s %10s %8s", "Address", "Basetime", "Age");
1471 for (i = 0; i < num_peers; ++i) {
1472 cli_print(cli, "%20s %10u %8d", fmtaddr(peers[i].peer, 0),
1473 peers[i].basetime, TIME - peers[i].timestamp);
1474 }
1475 return CLI_OK;
1476 }
1477
1478 //
1479 // Simple run-length-encoding compression.
1480 // Format is
1481 // 1 byte < 128 = count of non-zero bytes following. // Not legal to be zero.
1482 // n non-zero bytes;
1483 // or
1484 // 1 byte > 128 = (count - 128) run of zero bytes. //
1485 // repeat.
1486 // count == 0 indicates end of compressed stream.
1487 //
1488 // Compress from 'src' into 'dst'. return number of bytes
1489 // used from 'dst'.
1490 // Updates *src_p to indicate 1 past last bytes used.
1491 //
1492 // We could get an extra byte in the zero runs by storing (count-1)
1493 // but I'm playing it safe.
1494 //
1495 // Worst case is a 50% expansion in space required (trying to
1496 // compress { 0x00, 0x01 } * N )
1497 static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
1498 {
1499 int count;
1500 int orig_dsize = dsize;
1501 uint8_t *x, *src;
1502 src = *src_p;
1503
1504 while (ssize > 0 && dsize > 2) {
1505 count = 0;
1506 x = dst++; --dsize; // Reserve space for count byte..
1507
1508 if (*src) { // Copy a run of non-zero bytes.
1509 while (*src && count < 127 && ssize > 0 && dsize > 1) { // Count number of non-zero bytes.
1510 *dst++ = *src++;
1511 --dsize; --ssize;
1512 ++count;
1513 }
1514 *x = count; // Store number of non-zero bytes. Guarenteed to be non-zero!
1515
1516 } else { // Compress a run of zero bytes.
1517 while (*src == 0 && count < 127 && ssize > 0) {
1518 ++src;
1519 --ssize;
1520 ++count;
1521 }
1522 *x = count | 0x80 ;
1523 }
1524 }
1525
1526 *dst++ = 0x0; // Add Stop byte.
1527 --dsize;
1528
1529 *src_p = src;
1530 return (orig_dsize - dsize);
1531 }
1532
1533 //
1534 // Decompress the buffer into **p.
1535 // 'psize' is the size of the decompression buffer available.
1536 //
1537 // Returns the number of bytes decompressed.
1538 //
1539 // Decompresses from '*src_p' into 'dst'.
1540 // Return the number of dst bytes used.
1541 // Updates the 'src_p' pointer to point to the
1542 // first un-used byte.
1543 static int rle_decompress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
1544 {
1545 int count;
1546 int orig_dsize = dsize;
1547 char *src = *src_p;
1548
1549 while (ssize >0 && dsize > 0) { // While there's more to decompress, and there's room in the decompress buffer...
1550 count = *src++; --ssize; // get the count byte from the source.
1551 if (count == 0x0) // End marker reached? If so, finish.
1552 break;
1553
1554 if (count & 0x80) { // Decompress a run of zeros
1555 for (count &= 0x7f ; count > 0 && dsize > 0; --count) {
1556 *dst++ = 0x0;
1557 --dsize;
1558 }
1559 } else { // Copy run of non-zero bytes.
1560 for ( ; count > 0 && ssize && dsize; --count) { // Copy non-zero bytes across.
1561 *dst++ = *src++;
1562 --ssize; --dsize;
1563 }
1564 }
1565 }
1566 *src_p = src;
1567 return (orig_dsize - dsize);
1568 }