add "throttle=N"
[l2tpns.git] / cluster.c
1 // L2TPNS Clustering Stuff
2
3 char const *cvs_id_cluster = "$Id: cluster.c,v 1.26.2.5 2005/05/02 09:03:27 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 idle timeouts..
583 session[i].last_packet = time_now;
584
585 // Reset die relative to our uptime rather than the old master's
586 if (session[i].die) session[i].die = TIME;
587
588 // Accumulate un-sent byte counters.
589 session[i].cin += sess_local[i].cin;
590 session[i].cout += sess_local[i].cout;
591 session[i].total_cin += sess_local[i].cin;
592 session[i].total_cout += sess_local[i].cout;
593
594 sess_local[i].cin = sess_local[i].cout = 0;
595
596 session[i].radius = 0; // Reset authentication as the radius blocks aren't up to date.
597
598 if (session[i].unique_id >= high_unique_id) // This is different to the index into the session table!!!
599 high_unique_id = session[i].unique_id+1;
600
601 session[i].tbf_in = session[i].tbf_out = 0; // Remove stale pointers from old master.
602 throttle_session(i, session[i].throttle_in, session[i].throttle_out);
603
604 config->cluster_highest_sessionid = i;
605 }
606
607 session[last_free].next = 0; // End of chain.
608 last_id = high_unique_id; // Keep track of the highest used session ID.
609
610 become_master();
611
612 rebuild_address_pool();
613
614 // If we're not the very first master, this is a big issue!
615 if(count>0)
616 LOG(0, 0, 0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count);
617
618 config->cluster_undefined_sessions = 0;
619 config->cluster_undefined_tunnels = 0;
620 config->cluster_iam_uptodate = 1; // assume all peers are up-to-date
621
622 if (!num_peers) // lone master
623 advertise();
624
625 // FIXME. We need to fix up the tunnel control message
626 // queue here! There's a number of other variables we
627 // should also update.
628 }
629
630
631 //
632 // Check that our session table is validly matching what the
633 // master has in mind.
634 //
635 // In particular, if we have too many sessions marked 'undefined'
636 // we fix it up here, and we ensure that the 'first free session'
637 // pointer is valid.
638 //
639 static void cluster_check_sessions(int highsession, int freesession_ptr, int hightunnel)
640 {
641 int i;
642
643 sessionfree = freesession_ptr; // Keep the freesession ptr valid.
644
645 if (config->cluster_iam_uptodate)
646 return;
647
648 if (highsession > config->cluster_undefined_sessions && hightunnel > config->cluster_undefined_tunnels)
649 return;
650
651 // Clear out defined sessions, counting the number of
652 // undefs remaining.
653 config->cluster_undefined_sessions = 0;
654 for (i = 1 ; i < MAXSESSION; ++i) {
655 if (i > highsession) {
656 if (session[i].tunnel == T_UNDEF) session[i].tunnel = T_FREE; // Defined.
657 continue;
658 }
659
660 if (session[i].tunnel == T_UNDEF)
661 ++config->cluster_undefined_sessions;
662 }
663
664 // Clear out defined tunnels, counting the number of
665 // undefs remaining.
666 config->cluster_undefined_tunnels = 0;
667 for (i = 1 ; i < MAXTUNNEL; ++i) {
668 if (i > hightunnel) {
669 if (tunnel[i].state == TUNNELUNDEF) tunnel[i].state = TUNNELFREE; // Defined.
670 continue;
671 }
672
673 if (tunnel[i].state == TUNNELUNDEF)
674 ++config->cluster_undefined_tunnels;
675 }
676
677
678 if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels) {
679 LOG(2, 0, 0, "Cleared undefined sessions/tunnels. %d sess (high %d), %d tunn (high %d)\n",
680 config->cluster_undefined_sessions, highsession, config->cluster_undefined_tunnels, hightunnel);
681 return;
682 }
683
684 // Are we up to date?
685
686 if (!config->cluster_iam_uptodate)
687 cluster_uptodate();
688 }
689
690 static int hb_add_type(char **p, int type, int id)
691 {
692 switch (type) {
693 case C_CSESSION: { // Compressed C_SESSION.
694 uint8_t c[sizeof(sessiont) * 2]; // Bigger than worst case.
695 uint8_t *d = (uint8_t *) &session[id];
696 uint8_t *orig = d;
697 int size;
698
699 size = rle_compress( &d, sizeof(sessiont), c, sizeof(c) );
700
701 // Did we compress the full structure, and is the size actually
702 // reduced??
703 if ( (d - orig) == sizeof(sessiont) && size < sizeof(sessiont) ) {
704 add_type(p, C_CSESSION, id, (char *) c, size);
705 break;
706 }
707 // Failed to compress : Fall through.
708 }
709 case C_SESSION: add_type(p, C_SESSION, id,
710 (char *) &session[id], sizeof(sessiont));
711 break;
712
713 case C_CTUNNEL: { // Compressed C_TUNNEL
714 uint8_t c[sizeof(tunnelt) * 2]; // Bigger than worst case.
715 uint8_t *d = (uint8_t *) &tunnel[id];
716 uint8_t *orig = d;
717 int size;
718
719 size = rle_compress( &d, sizeof(tunnelt), c, sizeof(c) );
720
721 // Did we compress the full structure, and is the size actually
722 // reduced??
723 if ( (d - orig) == sizeof(tunnelt) && size < sizeof(tunnelt) ) {
724 add_type(p, C_CTUNNEL, id, c, size);
725 break;
726 }
727 // Failed to compress : Fall through.
728 }
729 case C_TUNNEL: add_type(p, C_TUNNEL, id,
730 (char *) &tunnel[id], sizeof(tunnelt));
731 break;
732 default:
733 LOG(0, 0, 0, "Found an invalid type in heart queue! (%d)\n", type);
734 kill(0, SIGTERM);
735 exit(1);
736 }
737 return 0;
738 }
739
740 //
741 // Send a heartbeat, incidently sending out any queued changes..
742 //
743 void cluster_heartbeat()
744 {
745 int i, count = 0, tcount = 0;
746 char buff[MAX_HEART_SIZE + sizeof(heartt) + sizeof(int) ];
747 heartt h;
748 char *p = buff;
749
750 if (!config->cluster_iam_master) // Only the master does this.
751 return;
752
753 config->cluster_table_version += config->cluster_num_changes;
754
755 // Fill out the heartbeat header.
756 memset(&h, 0, sizeof(h));
757
758 h.version = HB_VERSION;
759 h.seq = config->cluster_seq_number;
760 h.basetime = basetime;
761 h.clusterid = config->bind_address; // Will this do??
762 h.basetime = basetime;
763 h.highsession = config->cluster_highest_sessionid;
764 h.freesession = sessionfree;
765 h.hightunnel = config->cluster_highest_tunnelid;
766 h.size_sess = sizeof(sessiont); // Just in case.
767 h.size_tunn = sizeof(tunnelt);
768 h.interval = config->cluster_hb_interval;
769 h.timeout = config->cluster_hb_timeout;
770 h.table_version = config->cluster_table_version;
771
772 add_type(&p, C_HEARTBEAT, HB_VERSION, (char *) &h, sizeof(h));
773
774 for (i = 0; i < config->cluster_num_changes; ++i) {
775 hb_add_type(&p, cluster_changes[i].type, cluster_changes[i].id);
776 }
777
778 if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
779 LOG(0, 0, 0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", (int) (p - buff));
780 kill(0, SIGTERM);
781 exit(1);
782 }
783
784 //
785 // Fill out the packet with sessions from the session table...
786 // (not forgetting to leave space so we can get some tunnels in too )
787 while ( (p + sizeof(uint32_t) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) {
788
789 if (!walk_session_number) // session #0 isn't valid.
790 ++walk_session_number;
791
792 if (count >= config->cluster_highest_sessionid) // If we're a small cluster, don't go wild.
793 break;
794
795 hb_add_type(&p, C_CSESSION, walk_session_number);
796 walk_session_number = (1+walk_session_number)%(config->cluster_highest_sessionid+1); // +1 avoids divide by zero.
797
798 ++count; // Count the number of extra sessions we're sending.
799 }
800
801 //
802 // Fill out the packet with tunnels from the tunnel table...
803 // This effectively means we walk the tunnel table more quickly
804 // than the session table. This is good because stuffing up a
805 // tunnel is a much bigger deal than stuffing up a session.
806 //
807 while ( (p + sizeof(uint32_t) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {
808
809 if (!walk_tunnel_number) // tunnel #0 isn't valid.
810 ++walk_tunnel_number;
811
812 if (tcount >= config->cluster_highest_tunnelid)
813 break;
814
815 hb_add_type(&p, C_CTUNNEL, walk_tunnel_number);
816 walk_tunnel_number = (1+walk_tunnel_number)%(config->cluster_highest_tunnelid+1); // +1 avoids divide by zero.
817
818 ++tcount;
819 }
820
821 //
822 // Did we do something wrong?
823 if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
824 LOG(0, 0, 0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", (int) (p - buff));
825 kill(0, SIGTERM);
826 exit(1);
827 }
828
829 LOG(3, 0, 0, "Sending v%d heartbeat #%d, change #%" PRIu64 " with %d changes "
830 "(%d x-sess, %d x-tunnels, %d highsess, %d hightun, size %d)\n",
831 HB_VERSION, h.seq, h.table_version, config->cluster_num_changes,
832 count, tcount, config->cluster_highest_sessionid,
833 config->cluster_highest_tunnelid, (int) (p - buff));
834
835 config->cluster_num_changes = 0;
836
837 send_heartbeat(h.seq, buff, (p-buff) ); // Send out the heartbeat to the cluster, keeping a copy of it.
838
839 config->cluster_seq_number = (config->cluster_seq_number+1)%HB_MAX_SEQ; // Next seq number to use.
840 }
841
842 //
843 // A structure of type 'type' has changed; Add it to the queue to send.
844 //
845 static int type_changed(int type, int id)
846 {
847 int i;
848
849 for (i = 0 ; i < config->cluster_num_changes ; ++i)
850 if ( cluster_changes[i].id == id &&
851 cluster_changes[i].type == type)
852 return 0; // Already marked for change.
853
854 cluster_changes[i].type = type;
855 cluster_changes[i].id = id;
856 ++config->cluster_num_changes;
857
858 if (config->cluster_num_changes > MAX_CHANGES)
859 cluster_heartbeat(); // flush now
860
861 return 1;
862 }
863
864
865 // A particular session has been changed!
866 int cluster_send_session(int sid)
867 {
868 if (!config->cluster_iam_master) {
869 LOG(0, sid, 0, "I'm not a master, but I just tried to change a session!\n");
870 return -1;
871 }
872
873 return type_changed(C_CSESSION, sid);
874 }
875
876 // A particular tunnel has been changed!
877 int cluster_send_tunnel(int tid)
878 {
879 if (!config->cluster_iam_master) {
880 LOG(0, 0, tid, "I'm not a master, but I just tried to change a tunnel!\n");
881 return -1;
882 }
883
884 return type_changed(C_CTUNNEL, tid);
885 }
886
887
888 //
889 // We're a master, and a slave has just told us that it's
890 // missed a packet. We'll resend it every packet since
891 // the last one it's seen.
892 //
893 static int cluster_catchup_slave(int seq, in_addr_t slave)
894 {
895 int s;
896 int diff;
897
898 LOG(1, 0, 0, "Slave %s sent LASTSEEN with seq %d\n", fmtaddr(slave, 0), seq);
899
900 diff = config->cluster_seq_number - seq; // How many packet do we need to send?
901 if (diff < 0)
902 diff += HB_MAX_SEQ;
903
904 if (diff >= HB_HISTORY_SIZE) { // Ouch. We don't have the packet to send it!
905 LOG(0, 0, 0, "A slave asked for message %d when our seq number is %d. Killing it.\n",
906 seq, config->cluster_seq_number);
907 return peer_send_message(slave, C_KILL, seq, NULL, 0);// Kill the slave. Nothing else to do.
908 }
909
910 // Now resend every packet that it missed, in order.
911 while (seq != config->cluster_seq_number) {
912 s = seq%HB_HISTORY_SIZE;
913 if (seq != past_hearts[s].seq) {
914 LOG(0, 0, 0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n",
915 fmtaddr(slave, 0), seq, past_hearts[s].seq, s, config->cluster_seq_number);
916 return -1; // What to do here!?
917 }
918 peer_send_data(slave, past_hearts[s].data, past_hearts[s].size);
919 seq = (seq+1)%HB_MAX_SEQ; // Increment to next seq number.
920 }
921 return 0; // All good!
922 }
923
924 //
925 // We've heard from another peer! Add it to the list
926 // that we select from at election time.
927 //
928 static int cluster_add_peer(in_addr_t peer, time_t basetime, pingt *pp, int size)
929 {
930 int i;
931 in_addr_t clusterid;
932 pingt p;
933
934 // Allow for backward compatability.
935 // Just the ping packet into a new structure to allow
936 // for the possibility that we might have received
937 // more or fewer elements than we were expecting.
938 if (size > sizeof(p))
939 size = sizeof(p);
940
941 memset( (void *) &p, 0, sizeof(p) );
942 memcpy( (void *) &p, (void *) pp, size);
943
944 clusterid = p.addr;
945 if (clusterid != config->bind_address)
946 {
947 // Is this for us?
948 LOG(4, 0, 0, "Skipping ping from %s (different cluster)\n", fmtaddr(peer, 0));
949 return 0;
950 }
951
952 for (i = 0; i < num_peers ; ++i)
953 {
954 if (peers[i].peer != peer)
955 continue;
956
957 // This peer already exists. Just update the timestamp.
958 peers[i].basetime = basetime;
959 peers[i].timestamp = TIME;
960 peers[i].uptodate = !p.undef;
961 break;
962 }
963
964 // Is this the master shutting down??
965 if (peer == config->cluster_master_address && !basetime) {
966 LOG(3, 0, 0, "Master %s shutting down...\n", fmtaddr(config->cluster_master_address, 0));
967 config->cluster_master_address = 0;
968 config->cluster_last_hb = 0; // Force an election.
969 cluster_check_master();
970 return 0;
971 }
972
973 if (i >= num_peers)
974 {
975 LOG(4, 0, 0, "Adding %s as a peer\n", fmtaddr(peer, 0));
976
977 // Not found. Is there a stale slot to re-use?
978 for (i = 0; i < num_peers ; ++i)
979 {
980 if (!peers[i].basetime) // Shutdown
981 break;
982
983 if ((peers[i].timestamp + config->cluster_hb_timeout * 10) < TIME) // Stale.
984 break;
985 }
986
987 if (i >= CLUSTER_MAX_SIZE)
988 {
989 // Too many peers!!
990 LOG(0, 0, 0, "Tried to add %s as a peer, but I already have %d of them!\n", fmtaddr(peer, 0), i);
991 return -1;
992 }
993
994 peers[i].peer = peer;
995 peers[i].basetime = basetime;
996 peers[i].timestamp = TIME;
997 peers[i].uptodate = !p.undef;
998 if (i == num_peers)
999 ++num_peers;
1000
1001 LOG(1, 0, 0, "Added %s as a new peer. Now %d peers\n", fmtaddr(peer, 0), num_peers);
1002 }
1003
1004 return 1;
1005 }
1006
1007 /* Handle the slave updating the byte counters for the master. */
1008 //
1009 // Note that we don't mark the session as dirty; We rely on
1010 // the slow table walk to propogate this back out to the slaves.
1011 //
1012 static int cluster_handle_bytes(char *data, int size)
1013 {
1014 bytest *b;
1015
1016 b = (bytest *) data;
1017
1018 LOG(3, 0, 0, "Got byte counter update (size %d)\n", size);
1019
1020 /* Loop around, adding the byte
1021 counts to each of the sessions. */
1022
1023 while (size >= sizeof(*b) ) {
1024 if (b->sid > MAXSESSION) {
1025 LOG(0, 0, 0, "Got C_BYTES with session #%d!\n", b->sid);
1026 return -1; /* Abort processing */
1027 }
1028
1029 session[b->sid].total_cin += b->in;
1030 session[b->sid].total_cout += b->out;
1031
1032 session[b->sid].cin += b->in;
1033 session[b->sid].cout += b->out;
1034
1035 if (b->in)
1036 session[b->sid].last_packet = time_now; // Reset idle timer!
1037
1038 size -= sizeof(*b);
1039 ++b;
1040 }
1041
1042 if (size != 0)
1043 LOG(0, 0, 0, "Got C_BYTES with %d bytes of trailing junk!\n", size);
1044
1045 return size;
1046 }
1047
1048 //
1049 // Handle receiving a session structure in a heartbeat packet.
1050 //
1051 static int cluster_recv_session(int more, uint8_t *p)
1052 {
1053 if (more >= MAXSESSION) {
1054 LOG(0, 0, 0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
1055 return -1;
1056 }
1057
1058 if (session[more].tunnel == T_UNDEF) {
1059 if (config->cluster_iam_uptodate) { // Sanity.
1060 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined session!\n");
1061 } else {
1062 --config->cluster_undefined_sessions;
1063 }
1064 }
1065
1066 load_session(more, (sessiont *) p); // Copy session into session table..
1067
1068 LOG(5, more, 0, "Received session update (%d undef)\n", config->cluster_undefined_sessions);
1069
1070 if (!config->cluster_iam_uptodate)
1071 cluster_uptodate(); // Check to see if we're up to date.
1072
1073 return 0;
1074 }
1075
1076 static int cluster_recv_tunnel(int more, uint8_t *p)
1077 {
1078 if (more >= MAXTUNNEL) {
1079 LOG(0, 0, 0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
1080 return -1;
1081 }
1082
1083 if (tunnel[more].state == TUNNELUNDEF) {
1084 if (config->cluster_iam_uptodate) { // Sanity.
1085 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined tunnel!\n");
1086 } else {
1087 --config->cluster_undefined_tunnels;
1088 }
1089 }
1090
1091 memcpy(&tunnel[more], p, sizeof(tunnel[more]) );
1092
1093 //
1094 // Clear tunnel control messages. These are dynamically allocated.
1095 // If we get unlucky, this may cause the tunnel to drop!
1096 //
1097 tunnel[more].controls = tunnel[more].controle = NULL;
1098 tunnel[more].controlc = 0;
1099
1100 LOG(5, 0, more, "Received tunnel update\n");
1101
1102 if (!config->cluster_iam_uptodate)
1103 cluster_uptodate(); // Check to see if we're up to date.
1104
1105 return 0;
1106 }
1107
1108
1109 //
1110 // Process a heartbeat..
1111 //
1112 // v3: added interval, timeout
1113 // v4: added table_version
1114 static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t *p, in_addr_t addr)
1115 {
1116 heartt *h;
1117 int s = size - (p-data);
1118 int i, type;
1119
1120 #if HB_VERSION != 4
1121 # error "need to update cluster_process_heartbeat()"
1122 #endif
1123
1124 // we handle versions 3 through 4
1125 if (more < 3 || more > HB_VERSION) {
1126 LOG(0, 0, 0, "Received a heartbeat version that I don't support (%d)!\n", more);
1127 return -1; // Ignore it??
1128 }
1129
1130 // Ok. It's a heartbeat packet from a cluster master!
1131 if (s < sizeof(*h))
1132 goto shortpacket;
1133
1134 h = (heartt *) p;
1135 p += sizeof(*h);
1136 s -= sizeof(*h);
1137
1138 if (h->clusterid != config->bind_address)
1139 return -1; // It's not part of our cluster.
1140
1141 if (config->cluster_iam_master) { // Sanity...
1142 // Note that this MUST match the election process above!
1143
1144 LOG(0, 0, 0, "I just got a heartbeat from master %s, but _I_ am the master!\n", fmtaddr(addr, 0));
1145 if (!h->basetime) {
1146 LOG(0, 0, 0, "Heartbeat with zero basetime! Ignoring\n");
1147 return -1; // Skip it.
1148 }
1149
1150 if (more >= 4) {
1151 if (h->table_version > config->cluster_table_version) {
1152 LOG(0, 0, 0, "They've seen more state changes (%" PRIu64 " vs my %" PRIu64 ") so I'm gone!\n",
1153 h->table_version, config->cluster_table_version);
1154
1155 kill(0, SIGTERM);
1156 exit(1);
1157 }
1158 if (h->table_version < config->cluster_table_version)
1159 return -1;
1160 }
1161
1162 if (basetime > h->basetime) {
1163 LOG(0, 0, 0, "They're an older master than me so I'm gone!\n");
1164 kill(0, SIGTERM);
1165 exit(1);
1166 }
1167
1168 if (basetime < h->basetime)
1169 return -1;
1170
1171 if (my_address < addr) { // Tie breaker.
1172 LOG(0, 0, 0, "They're a higher IP address than me, so I'm gone!\n");
1173 kill(0, SIGTERM);
1174 exit(1);
1175 }
1176
1177 return -1; // Skip it.
1178 }
1179
1180 if (config->cluster_seq_number == -1) // Don't have one. Just align to the master...
1181 config->cluster_seq_number = h->seq;
1182
1183 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1184
1185 if (config->cluster_seq_number != h->seq) { // Out of sequence heartbeat!
1186 static int lastseen_seq = 0;
1187 static time_t lastseen_time = 0;
1188
1189 // limit to once per second for a particular seq#
1190 int ask = (config->cluster_seq_number != lastseen_seq || time_now != lastseen_time);
1191
1192 LOG(1, 0, 0, "HB: Got seq# %d but was expecting %d. %s.\n",
1193 h->seq, config->cluster_seq_number,
1194 ask ? "Asking for resend" : "Ignoring");
1195
1196 if (ask)
1197 {
1198 lastseen_seq = config->cluster_seq_number;
1199 lastseen_time = time_now;
1200 peer_send_message(addr, C_LASTSEEN, config->cluster_seq_number, NULL, 0);
1201 }
1202
1203 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1204
1205 // Just drop the packet. The master will resend it as part of the catchup.
1206
1207 return 0;
1208 }
1209 // Save the packet in our buffer.
1210 // This is needed in case we become the master.
1211 config->cluster_seq_number = (h->seq+1)%HB_MAX_SEQ;
1212 i = h->seq % HB_HISTORY_SIZE;
1213 past_hearts[i].seq = h->seq;
1214 past_hearts[i].size = size;
1215 memcpy(&past_hearts[i].data, data, size); // Save it.
1216
1217
1218 // Check that we don't have too many undefined sessions, and
1219 // that the free session pointer is correct.
1220 cluster_check_sessions(h->highsession, h->freesession, h->hightunnel);
1221
1222 if (h->interval != config->cluster_hb_interval)
1223 {
1224 LOG(2, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
1225 h->interval, config->cluster_hb_interval);
1226
1227 config->cluster_hb_interval = h->interval;
1228 }
1229
1230 if (h->timeout != config->cluster_hb_timeout)
1231 {
1232 LOG(2, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
1233 h->timeout, config->cluster_hb_timeout);
1234
1235 config->cluster_hb_timeout = h->timeout;
1236 }
1237
1238 // Ok. process the packet...
1239 while ( s > 0) {
1240
1241 type = *((uint32_t *) p);
1242 p += sizeof(uint32_t);
1243 s -= sizeof(uint32_t);
1244
1245 more = *((uint32_t *) p);
1246 p += sizeof(uint32_t);
1247 s -= sizeof(uint32_t);
1248
1249 switch (type) {
1250 case C_CSESSION: { // Compressed session structure.
1251 uint8_t c[ sizeof(sessiont) + 2];
1252 int size;
1253 uint8_t *orig_p = p;
1254
1255 size = rle_decompress((uint8_t **) &p, s, c, sizeof(c) );
1256 s -= (p - orig_p);
1257
1258 if (size != sizeof(sessiont) ) { // Ouch! Very very bad!
1259 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1260 // Now what? Should exit! No-longer up to date!
1261 break;
1262 }
1263
1264 cluster_recv_session(more, c);
1265 break;
1266 }
1267 case C_SESSION:
1268 if ( s < sizeof(session[more]))
1269 goto shortpacket;
1270
1271 cluster_recv_session(more, p);
1272
1273 p += sizeof(session[more]);
1274 s -= sizeof(session[more]);
1275 break;
1276
1277 case C_CTUNNEL: { // Compressed tunnel structure.
1278 uint8_t c[ sizeof(tunnelt) + 2];
1279 int size;
1280 uint8_t *orig_p = p;
1281
1282 size = rle_decompress((uint8_t **) &p, s, c, sizeof(c));
1283 s -= (p - orig_p);
1284
1285 if (size != sizeof(tunnelt) ) { // Ouch! Very very bad!
1286 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1287 // Now what? Should exit! No-longer up to date!
1288 break;
1289 }
1290
1291 cluster_recv_tunnel(more, c);
1292 break;
1293
1294 }
1295 case C_TUNNEL:
1296 if ( s < sizeof(tunnel[more]))
1297 goto shortpacket;
1298
1299 cluster_recv_tunnel(more, p);
1300
1301 p += sizeof(tunnel[more]);
1302 s -= sizeof(tunnel[more]);
1303 break;
1304 default:
1305 LOG(0, 0, 0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type);
1306 return -1; // can't process any more of the packet!!
1307 }
1308 }
1309
1310 if (config->cluster_master_address != addr)
1311 {
1312 LOG(0, 0, 0, "My master just changed from %s to %s!\n",
1313 fmtaddr(config->cluster_master_address, 0), fmtaddr(addr, 1));
1314
1315 config->cluster_master_address = addr;
1316 }
1317
1318 config->cluster_last_hb = TIME; // Successfully received a heartbeat!
1319 config->cluster_table_version = h->table_version;
1320 return 0;
1321
1322 shortpacket:
1323 LOG(0, 0, 0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
1324 return -1;
1325 }
1326
1327 //
1328 // We got a packet on the cluster port!
1329 // Handle pings, lastseens, and heartbeats!
1330 //
1331 int processcluster(char *data, int size, in_addr_t addr)
1332 {
1333 int type, more;
1334 char *p = data;
1335 int s = size;
1336
1337 if (addr == my_address)
1338 return -1; // Ignore it. Something looped back the multicast!
1339
1340 LOG(5, 0, 0, "Process cluster: %d bytes from %s\n", size, fmtaddr(addr, 0));
1341
1342 if (s <= 0) // Any data there??
1343 return -1;
1344
1345 if (s < 8)
1346 goto shortpacket;
1347
1348 type = *((uint32_t *) p);
1349 p += sizeof(uint32_t);
1350 s -= sizeof(uint32_t);
1351
1352 more = *((uint32_t *) p);
1353 p += sizeof(uint32_t);
1354 s -= sizeof(uint32_t);
1355
1356 switch (type) {
1357 case C_PING: // Update the peers table.
1358 return cluster_add_peer(addr, more, (pingt *) p, s);
1359
1360 case C_LASTSEEN: // Catch up a slave (slave missed a packet).
1361 return cluster_catchup_slave(more, addr);
1362
1363 case C_FORWARD: { // Forwarded control packet. pass off to processudp.
1364 struct sockaddr_in a;
1365 a.sin_addr.s_addr = more;
1366
1367 a.sin_port = *(int *) p;
1368 s -= sizeof(int);
1369 p += sizeof(int);
1370
1371 if (!config->cluster_iam_master) { // huh?
1372 LOG(0, 0, 0, "I'm not the master, but I got a C_FORWARD from %s?\n", fmtaddr(addr, 0));
1373 return -1;
1374 }
1375
1376 LOG(4, 0, 0, "Got a forwarded packet... (%s:%d)\n", fmtaddr(more, 0), a.sin_port);
1377 STAT(recv_forward);
1378 processudp(p, s, &a);
1379 return 0;
1380 }
1381 case C_THROTTLE: { // Receive a forwarded packet from a slave.
1382 if (!config->cluster_iam_master) {
1383 LOG(0, 0, 0, "I'm not the master, but I got a C_THROTTLE from %s?\n", fmtaddr(addr, 0));
1384 return -1;
1385 }
1386
1387 tbf_queue_packet(more, p, s); // The TBF id tells wether it goes in or out.
1388 return 0;
1389 }
1390 case C_GARDEN:
1391 // Receive a walled garden packet from a slave.
1392 if (!config->cluster_iam_master) {
1393 LOG(0, 0, 0, "I'm not the master, but I got a C_GARDEN from %s?\n", fmtaddr(addr, 0));
1394 return -1;
1395 }
1396
1397 tun_write(p, s);
1398 return 0;
1399
1400 case C_BYTES:
1401 return cluster_handle_bytes(p, s);
1402
1403 case C_KILL: // The master asked us to die!? (usually because we're too out of date).
1404 if (config->cluster_iam_master) {
1405 LOG(0, 0, 0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", fmtaddr(addr, 0), more);
1406 return -1;
1407 }
1408 if (more != config->cluster_seq_number) {
1409 LOG(0, 0, 0, "The master asked us to die but the seq number didn't match!?\n");
1410 return -1;
1411 }
1412
1413 if (addr != config->cluster_master_address) {
1414 LOG(0, 0, 0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%s)\n",
1415 fmtaddr(addr, 0), fmtaddr(config->cluster_master_address, 1));
1416 // We can only warn about it. The master might really have switched!
1417 }
1418
1419 LOG(0, 0, 0, "Received a valid C_KILL: I'm going to die now.\n");
1420 kill(0, SIGTERM);
1421 exit(0); // Lets be paranoid;
1422 return -1; // Just signalling the compiler.
1423
1424 case C_HEARTBEAT:
1425 LOG(4, 0, 0, "Got a heartbeat from %s\n", fmtaddr(addr, 0));
1426 return cluster_process_heartbeat(data, size, more, p, addr);
1427
1428 default:
1429 LOG(0, 0, 0, "Strange type packet received on cluster socket (%d)\n", type);
1430 return -1;
1431 }
1432 return 0;
1433
1434 shortpacket:
1435 LOG(0, 0, 0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n");
1436 return -1;
1437 }
1438
1439 //====================================================================================================
1440
1441 int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
1442 {
1443 int i;
1444
1445 if (CLI_HELP_REQUESTED)
1446 return CLI_HELP_NO_ARGS;
1447
1448 cli_print(cli, "Cluster status : %s", config->cluster_iam_master ? "Master" : "Slave" );
1449 cli_print(cli, "My address : %s", fmtaddr(my_address, 0));
1450 cli_print(cli, "VIP address : %s", fmtaddr(config->bind_address, 0));
1451 cli_print(cli, "Multicast address: %s", fmtaddr(config->cluster_address, 0));
1452 cli_print(cli, "Multicast i'face : %s", config->cluster_interface);
1453
1454 if (!config->cluster_iam_master) {
1455 cli_print(cli, "My master : %s (last heartbeat %.1f seconds old)",
1456 config->cluster_master_address
1457 ? fmtaddr(config->cluster_master_address, 0)
1458 : "Not defined",
1459 0.1 * (TIME - config->cluster_last_hb));
1460 cli_print(cli, "Uptodate : %s", config->cluster_iam_uptodate ? "Yes" : "No");
1461 cli_print(cli, "Table version # : %" PRIu64, config->cluster_table_version);
1462 cli_print(cli, "Next sequence number expected: %d", config->cluster_seq_number);
1463 cli_print(cli, "%d sessions undefined of %d", config->cluster_undefined_sessions, config->cluster_highest_sessionid);
1464 cli_print(cli, "%d tunnels undefined of %d", config->cluster_undefined_tunnels, config->cluster_highest_tunnelid);
1465 } else {
1466 cli_print(cli, "Table version # : %" PRIu64, config->cluster_table_version);
1467 cli_print(cli, "Next heartbeat # : %d", config->cluster_seq_number);
1468 cli_print(cli, "Highest session : %d", config->cluster_highest_sessionid);
1469 cli_print(cli, "Highest tunnel : %d", config->cluster_highest_tunnelid);
1470 cli_print(cli, "%d changes queued for sending", config->cluster_num_changes);
1471 }
1472 cli_print(cli, "%d peers.", num_peers);
1473
1474 if (num_peers)
1475 cli_print(cli, "%20s %10s %8s", "Address", "Basetime", "Age");
1476 for (i = 0; i < num_peers; ++i) {
1477 cli_print(cli, "%20s %10u %8d", fmtaddr(peers[i].peer, 0),
1478 peers[i].basetime, TIME - peers[i].timestamp);
1479 }
1480 return CLI_OK;
1481 }
1482
1483 //
1484 // Simple run-length-encoding compression.
1485 // Format is
1486 // 1 byte < 128 = count of non-zero bytes following. // Not legal to be zero.
1487 // n non-zero bytes;
1488 // or
1489 // 1 byte > 128 = (count - 128) run of zero bytes. //
1490 // repeat.
1491 // count == 0 indicates end of compressed stream.
1492 //
1493 // Compress from 'src' into 'dst'. return number of bytes
1494 // used from 'dst'.
1495 // Updates *src_p to indicate 1 past last bytes used.
1496 //
1497 // We could get an extra byte in the zero runs by storing (count-1)
1498 // but I'm playing it safe.
1499 //
1500 // Worst case is a 50% expansion in space required (trying to
1501 // compress { 0x00, 0x01 } * N )
1502 static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
1503 {
1504 int count;
1505 int orig_dsize = dsize;
1506 uint8_t *x, *src;
1507 src = *src_p;
1508
1509 while (ssize > 0 && dsize > 2) {
1510 count = 0;
1511 x = dst++; --dsize; // Reserve space for count byte..
1512
1513 if (*src) { // Copy a run of non-zero bytes.
1514 while (*src && count < 127 && ssize > 0 && dsize > 1) { // Count number of non-zero bytes.
1515 *dst++ = *src++;
1516 --dsize; --ssize;
1517 ++count;
1518 }
1519 *x = count; // Store number of non-zero bytes. Guarenteed to be non-zero!
1520
1521 } else { // Compress a run of zero bytes.
1522 while (*src == 0 && count < 127 && ssize > 0) {
1523 ++src;
1524 --ssize;
1525 ++count;
1526 }
1527 *x = count | 0x80 ;
1528 }
1529 }
1530
1531 *dst++ = 0x0; // Add Stop byte.
1532 --dsize;
1533
1534 *src_p = src;
1535 return (orig_dsize - dsize);
1536 }
1537
1538 //
1539 // Decompress the buffer into **p.
1540 // 'psize' is the size of the decompression buffer available.
1541 //
1542 // Returns the number of bytes decompressed.
1543 //
1544 // Decompresses from '*src_p' into 'dst'.
1545 // Return the number of dst bytes used.
1546 // Updates the 'src_p' pointer to point to the
1547 // first un-used byte.
1548 static int rle_decompress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
1549 {
1550 int count;
1551 int orig_dsize = dsize;
1552 char *src = *src_p;
1553
1554 while (ssize >0 && dsize > 0) { // While there's more to decompress, and there's room in the decompress buffer...
1555 count = *src++; --ssize; // get the count byte from the source.
1556 if (count == 0x0) // End marker reached? If so, finish.
1557 break;
1558
1559 if (count & 0x80) { // Decompress a run of zeros
1560 for (count &= 0x7f ; count > 0 && dsize > 0; --count) {
1561 *dst++ = 0x0;
1562 --dsize;
1563 }
1564 } else { // Copy run of non-zero bytes.
1565 for ( ; count > 0 && ssize && dsize; --count) { // Copy non-zero bytes across.
1566 *dst++ = *src++;
1567 --ssize; --dsize;
1568 }
1569 }
1570 }
1571 *src_p = src;
1572 return (orig_dsize - dsize);
1573 }