a251a77a8faad4d8ba297cca98d172658114e9c9
[l2tpns.git] / cluster.c
1 // L2TPNS Clustering Stuff
2
3 char const *cvs_id_cluster = "$Id: cluster.c,v 1.7 2004-07-07 09:09:53 bodea Exp $";
4
5 #include <stdio.h>
6 #include <sys/file.h>
7 #include <sys/stat.h>
8 #include <sys/socket.h>
9 #include <netinet/in.h>
10 #include <arpa/inet.h>
11 #include <sys/ioctl.h>
12 #include <net/if.h>
13 #include <string.h>
14 #include <malloc.h>
15 #include <errno.h>
16 #include <stdlib.h>
17 #include <stdarg.h>
18 #include <unistd.h>
19 #include <stdio.h>
20 #include <libcli.h>
21
22 #include "l2tpns.h"
23 #include "cluster.h"
24 #include "util.h"
25 #include "tbf.h"
26
27 #ifdef BGP
28 #include "bgp.h"
29 #endif
30 /*
31 * All cluster packets have the same format.
32 *
33 * One or more instances of
34 * a 32 bit 'type' id.
35 * a 32 bit 'extra' data dependant on the 'type'.
36 * zero or more bytes of structure data, dependant on the type.
37 *
38 */
39
40 // Module variables.
41 int cluster_sockfd = 0; // The filedescriptor for the cluster communications port.
42
43 ipt my_address = 0; // The network address of my ethernet port.
44 static int walk_session_number = 0; // The next session to send when doing the slow table walk.
45 static int walk_tunnel_number = 0; // The next tunnel to send when doing the slow table walk.
46
47 #define MAX_HEART_SIZE (8192) // Maximum size of heartbeat packet. Must be less than max IP packet size :)
48 #define MAX_CHANGES (MAX_HEART_SIZE/(sizeof(sessiont) + sizeof(int) ) - 2) // Assumes a session is the biggest type!
49
50 static struct {
51 int type;
52 int id;
53 } cluster_changes[MAX_CHANGES]; // Queue of changed structures that need to go out when next heartbeat.
54
55 static struct {
56 int seq;
57 int size;
58 char data[MAX_HEART_SIZE];
59 } past_hearts[HB_HISTORY_SIZE]; // Ring buffer of heartbeats that we've recently sent out. Needed so
60 // we can re-transmit if needed.
61
62 static struct {
63 u32 peer;
64 time_t basetime;
65 clockt timestamp;
66 int uptodate;
67 } peers[CLUSTER_MAX_SIZE]; // List of all the peers we've heard from.
68 static int num_peers; // Number of peers in list.
69 static int have_peers; // At least one up to date peer
70
71 int rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize);
72 int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize);
73
74 //
75 // Create a listening socket
76 //
77 // This joins the cluster multi-cast group.
78 //
79 int cluster_init()
80 {
81 struct sockaddr_in addr;
82 struct sockaddr_in interface_addr;
83 struct ip_mreq mreq;
84 struct ifreq ifr;
85 int opt = 0;
86
87 config->cluster_undefined_sessions = MAXSESSION-1;
88 config->cluster_undefined_tunnels = MAXTUNNEL-1;
89
90 if (!config->cluster_address)
91 return 0;
92 if (!*config->cluster_interface)
93 return 0;
94
95 cluster_sockfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
96
97 memset(&addr, 0, sizeof(addr));
98 addr.sin_family = AF_INET;
99 addr.sin_port = htons(CLUSTERPORT);
100 addr.sin_addr.s_addr = INADDR_ANY;
101 setsockopt(cluster_sockfd, SOL_SOCKET, SO_REUSEADDR, &addr, sizeof(addr));
102
103 if (bind(cluster_sockfd, (void *) &addr, sizeof(addr)) < 0)
104 {
105 log(0, 0, 0, 0, "Failed to bind cluster socket: %s\n", strerror(errno));
106 return -1;
107 }
108
109 strcpy(ifr.ifr_name, config->cluster_interface);
110 if (ioctl(cluster_sockfd, SIOCGIFADDR, &ifr) < 0) {
111 log(0, 0, 0, 0, "Failed to get interface address for (%s): %s\n", config->cluster_interface, strerror(errno));
112 return -1;
113 }
114
115 memcpy(&interface_addr, &ifr.ifr_addr, sizeof(interface_addr) );
116 my_address = interface_addr.sin_addr.s_addr;
117
118 // Join multicast group.
119 mreq.imr_multiaddr.s_addr = config->cluster_address;
120 mreq.imr_interface = interface_addr.sin_addr;
121
122
123 opt = 0; // Turn off multicast loopback.
124 setsockopt(cluster_sockfd, IPPROTO_IP, IP_MULTICAST_LOOP, &opt, sizeof(opt));
125
126 if (setsockopt(cluster_sockfd, IPPROTO_IP, IP_ADD_MEMBERSHIP, &mreq, sizeof(mreq)) < 0) {
127 log(0, 0, 0, 0, "Failed to setsockopt (join mcast group): %s\n", strerror(errno));
128 return -1;
129 }
130
131 if (setsockopt (cluster_sockfd, IPPROTO_IP, IP_MULTICAST_IF, &interface_addr, sizeof(interface_addr)) < 0) {
132 log(0, 0, 0, 0, "Failed to setsockopt (set mcast interface): %s\n", strerror(errno));
133 return -1;
134 }
135
136 config->cluster_last_hb = TIME;
137 config->cluster_seq_number = -1;
138
139 return cluster_sockfd;
140 }
141
142
143 //
144 // Send a chunk of data to the entire cluster (usually via the multicast
145 // address ).
146 //
147
148 int cluster_send_data(void *data, int datalen)
149 {
150 struct sockaddr_in addr = {0};
151
152 if (!cluster_sockfd) return -1;
153 if (!config->cluster_address) return 0;
154
155 addr.sin_addr.s_addr = config->cluster_address;
156 addr.sin_port = htons(CLUSTERPORT);
157 addr.sin_family = AF_INET;
158
159 log(5,0,0,0, "Cluster send data: %d bytes\n", datalen);
160
161 if (sendto(cluster_sockfd, data, datalen, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
162 {
163 log(0, 0, 0, 0, "sendto: %s\n", strerror(errno));
164 return -1;
165 }
166
167 return 0;
168 }
169
170 //
171 // Add a chunk of data to a heartbeat packet.
172 // Maintains the format. Assumes that the caller
173 // has passed in a big enough buffer!
174 //
175 static void add_type(char ** p, int type, int more, char * data, int size)
176 {
177 * ( (u32*)(*p) ) = type;
178 *p += sizeof(u32);
179
180 * ( (u32*)(*p) ) = more;
181 *p += sizeof(u32);
182
183 if (data && size > 0) {
184 memcpy(*p, data, size);
185 (*p) += size;
186 }
187 }
188
189 void cluster_uptodate(void)
190 {
191 if (config->cluster_iam_uptodate)
192 return;
193
194 if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels)
195 return;
196
197 config->cluster_iam_uptodate = 1;
198
199 log(0,0,0,0, "Now uptodate with master.\n");
200
201 #ifdef BGP
202 if (bgp_configured)
203 bgp_enable_routing(1);
204 else
205 #endif /* BGP */
206 if (config->send_garp)
207 send_garp(config->bind_address); // Start taking traffic.
208 }
209
210 //
211 // Send a unicast UDP packet to a peer with 'data' as the
212 // contents.
213 //
214 int peer_send_data(u32 peer, char * data, int size)
215 {
216 struct sockaddr_in addr = {0};
217
218 if (!cluster_sockfd) return -1;
219 if (!config->cluster_address) return 0;
220
221 if (!peer) // Odd??
222 return -1;
223
224 addr.sin_addr.s_addr = peer;
225 addr.sin_port = htons(CLUSTERPORT);
226 addr.sin_family = AF_INET;
227
228 log_hex(5, "Peer send", data, size);
229
230 if (sendto(cluster_sockfd, data, size, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
231 {
232 log(0, 0, 0, 0, "sendto: %s\n", strerror(errno));
233 return -1;
234 }
235
236 return 0;
237 }
238
239 //
240 // Send a structured message to a peer with a single element of type 'type'.
241 //
242 int peer_send_message(u32 peer, int type, int more, char * data, int size)
243 {
244 char buf[65536]; // Vast overkill.
245 char * p = buf;
246
247 log(4,0,0,0, "Sending message to peer (type %d, more %d, size %d)\n", type, more, size);
248 add_type(&p, type, more, data, size);
249
250 return peer_send_data(peer, buf, (p-buf) );
251 }
252
253 //
254 // Forward a state changing packet to the master.
255 //
256 // The master just processes the payload as if it had
257 // received it off the tap device.
258 //
259 int master_forward_packet(char * data, int size, u32 addr, int port)
260 {
261 char buf[65536]; // Vast overkill.
262 char * p = buf;
263
264 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
265 return -1;
266
267 log(4,0,0,0, "Forwarding packet from %s to master (size %d)\n", inet_toa(addr), size);
268
269 STAT(c_forwarded);
270 add_type(&p, C_FORWARD, addr, (char*) &port, sizeof(port) );
271 memcpy(p, data, size);
272 p += size;
273
274 return peer_send_data(config->cluster_master_address, buf, (p-buf) );
275
276 }
277
278 //
279 // Forward a throttled packet to the master for handling.
280 //
281 // The master just drops the packet into the appropriate
282 // token bucket queue, and lets normal processing take care
283 // of it.
284 //
285 int master_throttle_packet(int tbfid, char * data, int size)
286 {
287 char buf[65536]; // Vast overkill.
288 char * p = buf;
289
290 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
291 return -1;
292
293 log(4,0,0,0, "Throttling packet master (size %d, tbfid %d)\n", size, tbfid);
294
295 add_type(&p, C_THROTTLE, tbfid, data, size);
296
297 return peer_send_data(config->cluster_master_address, buf, (p-buf) );
298
299 }
300
301 //
302 // Forward a walled garden packet to the master for handling.
303 //
304 // The master just writes the packet straight to the tun
305 // device (where is will normally loop through the
306 // firewall rules, and come back in on the tun device)
307 //
308 // (Note that this must be called with the tun header
309 // as the start of the data).
310 int master_garden_packet(sessionidt s, char *data, int size)
311 {
312 char buf[65536]; // Vast overkill.
313 char *p = buf;
314
315 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
316 return -1;
317
318 log(4,0,0,0, "Walled garden packet to master (size %d)\n", size);
319
320 add_type(&p, C_GARDEN, s, data, size);
321
322 return peer_send_data(config->cluster_master_address, buf, (p-buf));
323
324 }
325
326 //
327 // Send a chunk of data as a heartbeat..
328 // We save it in the history buffer as we do so.
329 //
330 static void send_heartbeat(int seq, char * data, int size)
331 {
332 int i;
333 static int last_seq = -1;
334
335 if (last_seq != -1 && (seq != (last_seq+1)%HB_MAX_SEQ) ) {
336 log(0,0,0,0, "FATAL: Sequence number skipped! (%d != %d)\n",
337 seq, last_seq);
338 }
339 last_seq = seq;
340
341 if (size > sizeof(past_hearts[0].data)) {
342 log(0,0,0,0, "Tried to heartbeat something larger than the maximum packet!\n");
343 kill(0, SIGTERM);
344 exit(1);
345 }
346 i = seq % HB_HISTORY_SIZE;
347 past_hearts[i].seq = seq;
348 past_hearts[i].size = size;
349 memcpy(&past_hearts[i].data, data, size); // Save it.
350 cluster_send_data(data, size);
351 }
352
353 //
354 // Send an 'i am alive' message to every machine in the cluster.
355 //
356 void cluster_send_ping(time_t basetime)
357 {
358 char buff[100 + sizeof(pingt)];
359 char *p = buff;
360 pingt x;
361
362 if (config->cluster_iam_master && basetime) // We're heartbeating so no need to ping.
363 return;
364
365 log(5,0,0,0, "Sending cluster ping...\n");
366
367 x.ver = 1;
368 x.addr = config->bind_address;
369 x.undef = config->cluster_undefined_sessions + config->cluster_undefined_tunnels;
370 x.basetime = basetime;
371
372 add_type(&p, C_PING, basetime, (char *) &x, sizeof(x));
373 cluster_send_data(buff, (p-buff) );
374 }
375
376 //
377 // Walk the session counters looking for non-zero ones to send
378 // to the master. We send up to 100 of them at one time.
379 // We examine a maximum of 2000 sessions.
380 // (50k max session should mean that we normally
381 // examine the entire session table every 25 seconds).
382
383 #define MAX_B_RECS (400)
384 void master_update_counts(void)
385 {
386 int i, c;
387 bytest b[MAX_B_RECS+1];
388
389 if (config->cluster_iam_master) // Only happens on the slaves.
390 return;
391
392 if (!config->cluster_master_address) // If we don't have a master, skip it for a while.
393 return;
394
395 i = MAX_B_RECS * 5; // Examine max 2000 sessions;
396 if (config->cluster_highest_sessionid > i)
397 i = config->cluster_highest_sessionid;
398
399 for ( c = 0; i > 0 ; --i) {
400 // Next session to look at.
401 walk_session_number++;
402 if ( walk_session_number > config->cluster_highest_sessionid)
403 walk_session_number = 1;
404
405 if (!sess_count[walk_session_number].cin && !sess_count[walk_session_number].cout)
406 continue; // Unused. Skip it.
407
408 b[c].sid = walk_session_number;
409 b[c].in = sess_count[walk_session_number].cin;
410 b[c].out = sess_count[walk_session_number].cout;
411
412 if (++c > MAX_B_RECS) // Send a max of 400 elements in a packet.
413 break;
414
415 // Reset counters.
416 sess_count[walk_session_number].cin = sess_count[walk_session_number].cout = 0;
417 }
418
419 if (!c) // Didn't find any that changes. Get out of here!
420 return;
421
422
423 // Forward the data to the master.
424 log(4,0,0,0, "Sending byte counters to master (%d elements)\n", c);
425 peer_send_message(config->cluster_master_address, C_BYTES, c, (char*) &b, sizeof(b[0]) * c);
426 return;
427 }
428
429 //
430 // Check that we have a master. If it's been too
431 // long since we heard from a master then hold an election.
432 //
433 void cluster_check_master(void)
434 {
435 int i, count, tcount, high_sid = 0;
436 int last_free = 0;
437 int had_peers = have_peers;
438 clockt t = TIME;
439 static int probed = 0;
440
441 if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout))
442 {
443 // If the master is late (missed 2 hearbeats by a second and a
444 // hair) it may be that the switch has dropped us from the
445 // multicast group, try unicasting one probe to the master
446 // which will hopefully respond with a unicast heartbeat that
447 // will allow us to limp along until the querier next runs.
448 if (config->cluster_master_address
449 && TIME > (config->cluster_last_hb + 2 * config->cluster_hb_interval + 11))
450 {
451 if (!probed)
452 {
453 probed = 1;
454 log(1, 0, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
455 TIME - (config->cluster_last_hb + config->cluster_hb_interval));
456
457 peer_send_message(config->cluster_master_address,
458 C_LASTSEEN, config->cluster_seq_number, NULL, 0);
459 }
460 } else { // We got a recent heartbeat; reset the probe flag.
461 probed = 0;
462 }
463
464 if (!config->cluster_iam_master)
465 return; // Everything's ok. return.
466
467 // Master needs to check peer state
468 }
469
470 config->cluster_last_hb = TIME + 1;
471
472 if (config->cluster_iam_master)
473 config->cluster_iam_uptodate = 1; // cleared in loop below
474 else
475 log(0,0,0,0, "Master timed out! Holding election...\n");
476
477
478 for (i = have_peers = 0; i < num_peers; i++)
479 {
480 if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
481 continue; // Stale peer! Skip them.
482
483 if (!peers[i].basetime)
484 continue; // Shutdown peer! Skip them.
485
486 if (peers[i].uptodate)
487 have_peers = 1;
488
489 if (config->cluster_iam_master)
490 {
491 if (!peers[i].uptodate)
492 config->cluster_iam_uptodate = 0; // Start fast heartbeats
493
494 continue;
495 }
496
497 if (peers[i].basetime < basetime) {
498 log(1,0,0,0, "Expecting %s to become master\n", inet_toa(peers[i].peer) );
499 return; // They'll win the election. Get out of here.
500 }
501
502 if (peers[i].basetime == basetime &&
503 peers[i].peer > my_address) {
504 log(1,0,0,0, "Expecting %s to become master\n", inet_toa(peers[i].peer) );
505 return; // They'll win the election. Wait for them to come up.
506 }
507 }
508
509 if (config->cluster_iam_master) // If we're the master, we've already won
510 {
511 #ifdef BGP
512 // master lost all slaves, need to handle traffic ourself
513 if (bgp_configured && had_peers && !have_peers)
514 bgp_enable_routing(1);
515 #endif /* BGP */
516 return;
517 }
518
519 // Wow. it's been ages since I last heard a heartbeat
520 // and I'm better than an of my peers so it's time
521 // to become a master!!!
522
523 config->cluster_iam_master = 1;
524 config->cluster_master_address = 0;
525
526 log(0,0,0,0, "I am declaring myself the master!\n");
527
528 #ifdef BGP
529 if (bgp_configured && have_peers)
530 bgp_enable_routing(0); /* stop handling traffic */
531 #endif /* BGP */
532
533 if (config->cluster_seq_number == -1)
534 config->cluster_seq_number = 0;
535
536 //
537 // Go through and mark all the tunnels as defined.
538 // Count the highest used tunnel number as well.
539 //
540 config->cluster_highest_tunnelid = 0;
541 for (i = 0, tcount = 0; i < MAXTUNNEL; ++i) {
542 if (tunnel[i].state == TUNNELUNDEF)
543 tunnel[i].state = TUNNELFREE;
544
545 if (tunnel[i].state != TUNNELFREE && i > config->cluster_highest_tunnelid)
546 config->cluster_highest_tunnelid = i;
547 }
548
549 //
550 // Go through and mark all the sessions as being defined.
551 // reset the idle timeouts.
552 // add temporary byte counters to permanent ones.
553 // Re-string the free list.
554 // Find the ID of the highest session.
555 last_free = 0;
556 high_sid = 0;
557 config->cluster_highest_sessionid = 0;
558 for (i = 0, count = 0; i < MAXSESSION; ++i) {
559 if (session[i].tunnel == T_UNDEF) {
560 session[i].tunnel = T_FREE;
561 ++count;
562 }
563
564 if (session[i].tunnel == T_FREE) { // Unused session. Add to free list.
565 session[last_free].next = i;
566 session[i].next = 0;
567 last_free = i;
568 }
569
570 // Reset all the idle timeouts..
571 session[i].last_packet = time_now;
572
573 // Accumulate un-sent byte counters.
574 session[i].cin += sess_count[i].cin;
575 session[i].cout += sess_count[i].cout;
576 session[i].total_cin += sess_count[i].cin;
577 session[i].total_cout += sess_count[i].cout;
578
579 sess_count[i].cin = sess_count[i].cout = 0;
580
581 session[i].radius = 0; // Reset authentication as the radius blocks aren't up to date.
582
583 if (session[i].sid >= high_sid) // This is different to the index into the session table!!!
584 high_sid = session[i].sid+1;
585
586
587 session[i].tbf_in = session[i].tbf_out = 0; // Remove stale pointers from old master.
588 throttle_session(i, session[i].throttle);
589
590 // I'm unsure about this. --mo
591 // It's potentially a good thing, but it could send a
592 // LOT of packets.
593 // if (session[i].throttle)
594 // cluster_send_session(s); // Tell the slaves about the new tbf indexes.
595
596 if (session[i].tunnel != T_FREE && i > config->cluster_highest_sessionid)
597 config->cluster_highest_sessionid = i;
598
599 }
600
601 session[last_free].next = 0; // End of chain.
602 last_sid = high_sid; // Keep track of the highest used session ID.
603
604 become_master();
605
606 rebuild_address_pool();
607
608 // If we're not the very first master, this is a big issue!
609 if(count>0)
610 log(0,0,0,0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count);
611
612 config->cluster_undefined_sessions = 0;
613 config->cluster_undefined_tunnels = 0;
614 config->cluster_iam_uptodate = 1; // assume all peers are up-to-date
615
616 // FIXME. We need to fix up the tunnel control message
617 // queue here! There's a number of other variables we
618 // should also update.
619 }
620
621
622 //
623 // Check that our session table is validly matching what the
624 // master has in mind.
625 //
626 // In particular, if we have too many sessions marked 'undefined'
627 // we fix it up here, and we ensure that the 'first free session'
628 // pointer is valid.
629 //
630 static void cluster_check_sessions(int highsession, int freesession_ptr, int hightunnel)
631 {
632 int i;
633
634 sessionfree = freesession_ptr; // Keep the freesession ptr valid.
635
636 if (config->cluster_iam_uptodate)
637 return;
638
639 if (highsession > config->cluster_undefined_sessions && hightunnel > config->cluster_undefined_tunnels)
640 return;
641
642 // Clear out defined sessions, counting the number of
643 // undefs remaining.
644 config->cluster_undefined_sessions = 0;
645 for (i = 1 ; i < MAXSESSION; ++i) {
646 if (i > highsession) {
647 session[i].tunnel = 0; // Defined.
648 continue;
649 }
650 if (session[i].tunnel != T_UNDEF)
651 continue;
652 ++config->cluster_undefined_sessions;
653 }
654
655 // Clear out defined tunnels, counting the number of
656 // undefs remaining.
657 config->cluster_undefined_tunnels = 0;
658 for (i = 1 ; i < MAXTUNNEL; ++i) {
659 if (i > hightunnel) {
660 tunnel[i].state = TUNNELFREE; // Defined.
661 continue;
662 }
663 if (tunnel[i].state != TUNNELUNDEF)
664 continue;
665 ++config->cluster_undefined_tunnels;
666 }
667
668
669 if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels) {
670 log(2,0,0,0, "Cleared undefined sessions/tunnels. %d sess (high %d), %d tunn (high %d)\n",
671 config->cluster_undefined_sessions, highsession, config->cluster_undefined_tunnels, hightunnel);
672 return;
673 }
674
675 // Are we up to date?
676
677 if (!config->cluster_iam_uptodate)
678 cluster_uptodate();
679 }
680
681 int hb_add_type(char **p, int type, int id)
682 {
683 switch (type) {
684 case C_CSESSION: { // Compressed C_SESSION.
685 u8 c[sizeof(sessiont) * 2]; // Bigger than worst case.
686 u8 *d = (u8 *) &session[id];
687 u8 *orig = d;
688 int size;
689
690 size = rle_compress( &d, sizeof(sessiont), c, sizeof(c) );
691
692 // Did we compress the full structure, and is the size actually
693 // reduced??
694 if ( (d - orig) == sizeof(sessiont) && size < sizeof(sessiont) ) {
695 add_type(p, C_CSESSION, id, (char*) c, size);
696 break;
697 }
698 // Failed to compress : Fall through.
699 }
700 case C_SESSION: add_type(p, C_SESSION, id,
701 (char*) &session[id], sizeof(sessiont));
702 break;
703
704 case C_CTUNNEL: { // Compressed C_TUNNEL
705 u8 c[sizeof(tunnelt) * 2]; // Bigger than worst case.
706 u8 *d = (u8 *) &tunnel[id];
707 u8 *orig = d;
708 int size;
709
710 size = rle_compress( &d, sizeof(tunnelt), c, sizeof(c) );
711
712 // Did we compress the full structure, and is the size actually
713 // reduced??
714 if ( (d - orig) == sizeof(tunnelt) && size < sizeof(tunnelt) ) {
715 add_type(p, C_CTUNNEL, id, c, size);
716 break;
717 }
718 // Failed to compress : Fall through.
719 }
720 case C_TUNNEL: add_type(p, C_TUNNEL, id,
721 (char*) &tunnel[id], sizeof(tunnelt));
722 break;
723 default:
724 log(0,0,0,0, "Found an invalid type in heart queue! (%d)\n", type);
725 kill(0, SIGTERM);
726 exit(1);
727 }
728 return 0;
729 }
730
731 //
732 // Send a heartbeat, incidently sending out any queued changes..
733 //
734 void cluster_heartbeat()
735 {
736 int i, count = 0, tcount = 0;
737 char buff[MAX_HEART_SIZE + sizeof(heartt) + sizeof(int) ];
738 heartt h;
739 char *p = buff;
740
741 if (!config->cluster_iam_master) // Only the master does this.
742 return;
743
744 // Fill out the heartbeat header.
745 memset(&h, 0, sizeof(h));
746
747 h.version = HB_VERSION;
748 h.seq = config->cluster_seq_number;
749 h.basetime = basetime;
750 h.clusterid = config->bind_address; // Will this do??
751 h.basetime = basetime;
752 h.highsession = config->cluster_highest_sessionid;
753 h.freesession = sessionfree;
754 h.hightunnel = config->cluster_highest_tunnelid;
755 h.size_sess = sizeof(sessiont); // Just in case.
756 h.size_tunn = sizeof(tunnelt);
757 h.interval = config->cluster_hb_interval;
758 h.timeout = config->cluster_hb_timeout;
759
760 add_type(&p, C_HEARTBEAT, HB_VERSION, (char*) &h, sizeof(h));
761
762 for (i = 0; i < config->cluster_num_changes; ++i) {
763 hb_add_type(&p, cluster_changes[i].type, cluster_changes[i].id);
764 }
765
766 if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
767 log(0,0,0,0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", p - buff);
768 kill(0, SIGTERM);
769 exit(1);
770 }
771
772 //
773 // Fill out the packet with sessions from the session table...
774 // (not forgetting to leave space so we can get some tunnels in too )
775 while ( (p + sizeof(u32) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) {
776
777 if (!walk_session_number) // session #0 isn't valid.
778 ++walk_session_number;
779
780 if (count >= config->cluster_highest_sessionid) // If we're a small cluster, don't go wild.
781 break;
782
783 hb_add_type(&p, C_CSESSION, walk_session_number);
784 walk_session_number = (1+walk_session_number)%(config->cluster_highest_sessionid+1); // +1 avoids divide by zero.
785
786 ++count; // Count the number of extra sessions we're sending.
787 }
788
789 //
790 // Fill out the packet with tunnels from the tunnel table...
791 // This effectively means we walk the tunnel table more quickly
792 // than the session table. This is good because stuffing up a
793 // tunnel is a much bigger deal than stuffing up a session.
794 //
795 while ( (p + sizeof(u32) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {
796
797 if (!walk_tunnel_number) // tunnel #0 isn't valid.
798 ++walk_tunnel_number;
799
800 if (tcount >= config->cluster_highest_tunnelid)
801 break;
802
803 hb_add_type(&p, C_CTUNNEL, walk_tunnel_number);
804 walk_tunnel_number = (1+walk_tunnel_number)%(config->cluster_highest_tunnelid+1); // +1 avoids divide by zero.
805
806 ++tcount;
807 }
808
809 //
810 // Did we do something wrong?
811 if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
812 log(0,0,0,0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", p - buff);
813 kill(0, SIGTERM);
814 exit(1);
815 }
816
817 log(3,0,0,0, "Sending heartbeat #%d with %d changes (%d x-sess, %d x-tunnels, %d highsess, %d hightun size %d)\n",
818 h.seq, config->cluster_num_changes, count, tcount, config->cluster_highest_sessionid,
819 config->cluster_highest_tunnelid, (p-buff));
820
821 config->cluster_num_changes = 0;
822
823 send_heartbeat(h.seq, buff, (p-buff) ); // Send out the heartbeat to the cluster, keeping a copy of it.
824
825 config->cluster_seq_number = (config->cluster_seq_number+1)%HB_MAX_SEQ; // Next seq number to use.
826 }
827
828 //
829 // A structure of type 'type' has changed; Add it to the queue to send.
830 //
831 int type_changed(int type, int id)
832 {
833 int i;
834
835 for (i = 0 ; i < config->cluster_num_changes ; ++i)
836 if ( cluster_changes[i].id == id &&
837 cluster_changes[i].type == type)
838 return 0; // Already marked for change.
839
840 cluster_changes[i].type = type;
841 cluster_changes[i].id = id;
842 ++config->cluster_num_changes;
843
844 if (config->cluster_num_changes > MAX_CHANGES)
845 cluster_heartbeat(); // flush now
846
847 return 1;
848 }
849
850
851 // A particular session has been changed!
852 int cluster_send_session(int sid)
853 {
854 if (!config->cluster_iam_master) {
855 log(0,0,sid,0, "I'm not a master, but I just tried to change a session!\n");
856 return -1;
857 }
858
859 return type_changed(C_CSESSION, sid);
860 }
861
862 // A particular tunnel has been changed!
863 int cluster_send_tunnel(int tid)
864 {
865 if (!config->cluster_iam_master) {
866 log(0,0,0,tid, "I'm not a master, but I just tried to change a tunnel!\n");
867 return -1;
868 }
869
870 return type_changed(C_CTUNNEL, tid);
871 }
872
873
874 //
875 // We're a master, and a slave has just told us that it's
876 // missed a packet. We'll resend it every packet since
877 // the last one it's seen.
878 //
879 int cluster_catchup_slave(int seq, u32 slave)
880 {
881 int s;
882 int diff;
883
884 log(1,0,0,0, "Slave %s sent LASTSEEN with seq %d\n", inet_toa(slave), seq);
885
886 diff = config->cluster_seq_number - seq; // How many packet do we need to send?
887 if (diff < 0)
888 diff += HB_MAX_SEQ;
889
890 if (diff >= HB_HISTORY_SIZE) { // Ouch. We don't have the packet to send it!
891 log(0,0,0,0, "A slaved asked for message %d when our seq number is %d. Killing it.\n",
892 seq, config->cluster_seq_number);
893 return peer_send_message(slave, C_KILL, seq, NULL, 0);// Kill the slave. Nothing else to do.
894 }
895
896 // Now resend every packet that it missed, in order.
897 while (seq != config->cluster_seq_number) {
898 s = seq%HB_HISTORY_SIZE;
899 if (seq != past_hearts[s].seq) {
900 int i;
901 log(0,0,0,0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n",
902 inet_toa(slave), seq, past_hearts[s].seq, s, config->cluster_seq_number);
903
904 for (i = 0; i < HB_HISTORY_SIZE; ++i) {
905 log(0,0,0,0, "\tentry %3d: seq %d (size %d)\n", i, past_hearts[s].seq, past_hearts[s].size);
906 }
907 return -1; // What to do here!?
908 }
909 peer_send_data(slave, past_hearts[s].data, past_hearts[s].size);
910 seq = (seq+1)%HB_MAX_SEQ; // Increment to next seq number.
911 }
912 return 0; // All good!
913 }
914
915 //
916 // We've heard from another peer! Add it to the list
917 // that we select from at election time.
918 //
919 int cluster_add_peer(u32 peer, time_t basetime, pingt *p)
920 {
921 int i;
922 u32 clusterid;
923
924 clusterid = p->addr;
925 if (clusterid != config->bind_address)
926 {
927 // Is this for us?
928 log(4,0,0,0, "Skipping ping from %s (different cluster)\n", inet_toa(peer));
929 return 0;
930 }
931
932 for (i = 0; i < num_peers ; ++i)
933 {
934 if (peers[i].peer != peer)
935 continue;
936
937 // This peer already exists. Just update the timestamp.
938 peers[i].basetime = basetime;
939 peers[i].timestamp = TIME;
940 peers[i].uptodate = !p->undef;
941 break;
942 }
943
944 // Is this the master shutting down??
945 if (peer == config->cluster_master_address && !basetime) {
946 config->cluster_master_address = 0;
947 config->cluster_last_hb = 0; // Force an election.
948 cluster_check_master();
949 return 0;
950 }
951
952 if (i >= num_peers)
953 {
954 log(4,0,0,0, "Adding %s as a peer\n", inet_toa(peer));
955
956 // Not found. Is there a stale slot to re-use?
957 for (i = 0; i < num_peers ; ++i)
958 {
959 if (!peers[i].basetime) // Shutdown
960 break;
961
962 if ((peers[i].timestamp + config->cluster_hb_timeout * 10) < TIME) // Stale.
963 break;
964 }
965
966 if (i >= CLUSTER_MAX_SIZE)
967 {
968 // Too many peers!!
969 log(0,0,0,0, "Tried to add %s as a peer, but I already have %d of them!\n", inet_toa(peer), i);
970 return -1;
971 }
972
973 peers[i].peer = peer;
974 peers[i].basetime = basetime;
975 peers[i].timestamp = TIME;
976 peers[i].uptodate = !p->undef;
977 if (i == num_peers)
978 ++num_peers;
979
980 log(1,0,0,0, "Added %s as a new peer. Now %d peers\n", inet_toa(peer), num_peers);
981 }
982
983 if (peers[i].uptodate)
984 {
985 #ifdef BGP
986 /* drop routes if we've now got a peer */
987 if (config->cluster_iam_master && bgp_configured && !have_peers)
988 bgp_enable_routing(0);
989 #endif /* BGP */
990 have_peers = 1;
991 }
992 else if (config->cluster_iam_master)
993 {
994 config->cluster_iam_uptodate = 0; // increase heart-rate...
995 }
996
997
998 return 1;
999 }
1000
1001 /* Handle the slave updating the byte counters for the master. */
1002 //
1003 // Note that we don't mark the session as dirty; We rely on
1004 // the slow table walk to propogate this back out to the slaves.
1005 //
1006 int cluster_handle_bytes(char * data, int size)
1007 {
1008 bytest * b;
1009
1010 b = (bytest*) data;
1011
1012 log(3,0,0,0, "Got byte counter update (size %d)\n", size);
1013
1014 /* Loop around, adding the byte
1015 counts to each of the sessions. */
1016
1017 while (size >= sizeof(*b) ) {
1018 if (b->sid > MAXSESSION) {
1019 log(0,0,0,0, "Got C_BYTES with session #%d!\n", b->sid);
1020 return -1; /* Abort processing */
1021 }
1022
1023 session[b->sid].total_cin += b->in;
1024 session[b->sid].total_cout += b->out;
1025
1026 session[b->sid].cin += b->in;
1027 session[b->sid].cout += b->out;
1028 session[b->sid].last_packet = time_now; // Reset idle timer!
1029
1030 size -= sizeof(*b);
1031 ++b;
1032 }
1033
1034 if (size != 0)
1035 log(0,0,0,0, "Got C_BYTES with %d bytes of trailing junk!\n", size);
1036
1037 return size;
1038 }
1039
1040 //
1041 // Handle receiving a session structure in a heartbeat packet.
1042 //
1043 static int cluster_recv_session(int more , u8 * p)
1044 {
1045 if (more >= MAXSESSION) {
1046 log(0,0,0,0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
1047 return -1;
1048 }
1049
1050 if (session[more].tunnel == T_UNDEF) {
1051 if (config->cluster_iam_uptodate) { // Sanity.
1052 log(0,0,0,0, "I thought I was uptodate but I just found an undefined session!\n");
1053 } else {
1054 --config->cluster_undefined_sessions;
1055 }
1056 }
1057
1058 load_session(more, (sessiont*) p); // Copy session into session table..
1059
1060 log(5,0,more,0, "Received session update (%d undef)\n", config->cluster_undefined_sessions);
1061
1062 if (!config->cluster_iam_uptodate)
1063 cluster_uptodate(); // Check to see if we're up to date.
1064
1065 return 0;
1066 }
1067
1068 static int cluster_recv_tunnel(int more, u8 *p)
1069 {
1070 if (more >= MAXTUNNEL) {
1071 log(0,0,0,0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
1072 return -1;
1073 }
1074
1075 if (tunnel[more].state == TUNNELUNDEF) {
1076 if (config->cluster_iam_uptodate) { // Sanity.
1077 log(0,0,0,0, "I thought I was uptodate but I just found an undefined tunnel!\n");
1078 } else {
1079 --config->cluster_undefined_tunnels;
1080 }
1081 }
1082
1083 memcpy(&tunnel[more], p, sizeof(tunnel[more]) );
1084
1085 //
1086 // Clear tunnel control messages. These are dynamically allocated.
1087 // If we get unlucky, this may cause the tunnel to drop!
1088 //
1089 tunnel[more].controls = tunnel[more].controle = NULL;
1090 tunnel[more].controlc = 0;
1091
1092 log(5,0,0,more, "Received tunnel update\n");
1093
1094 if (!config->cluster_iam_uptodate)
1095 cluster_uptodate(); // Check to see if we're up to date.
1096
1097 return 0;
1098 }
1099
1100
1101 //
1102 // Process a heartbeat..
1103 //
1104 static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32 addr)
1105 {
1106 heartt * h;
1107 int s = size - (p-data);
1108 int i, type;
1109
1110 #if HB_VERSION != 3
1111 # error "need to update cluster_process_heartbeat()"
1112 #endif
1113
1114 // we handle version 2+
1115 if (more < 2 || more > HB_VERSION) {
1116 log(0,0,0,0, "Received a heartbeat version that I don't support (%d)!\n", more);
1117 return -1; // Ignore it??
1118 }
1119
1120 // Ok. It's a heartbeat packet from a cluster master!
1121 if (s < sizeof(*h))
1122 goto shortpacket;
1123
1124 h = (heartt*) p;
1125 p += sizeof(*h);
1126 s -= sizeof(*h);
1127
1128 if (h->clusterid != config->bind_address)
1129 return -1; // It's not part of our cluster.
1130
1131 if (config->cluster_iam_master) { // Sanity...
1132 // Note that this MUST match the election process above!
1133
1134 log(0,0,0,0, "I just got a packet claiming to be from a master but _I_ am the master!\n");
1135 if (!h->basetime) {
1136 log(0,0,0,0, "Heartbeat from addr %s with zero basetime!\n", inet_toa(addr) );
1137 return -1; // Skip it.
1138 }
1139 if (basetime > h->basetime) {
1140 log(0,0,0,0, "They're (%s) an older master than me so I'm gone!\n", inet_toa(addr));
1141 kill(0, SIGTERM);
1142 exit(1);
1143 }
1144 if (basetime == h->basetime && my_address < addr) { // Tie breaker.
1145 log(0,0,0,0, "They're a higher IP address than me, so I'm gone!\n");
1146 kill(0, SIGTERM);
1147 exit(1);
1148 }
1149 return -1; // Skip it.
1150 }
1151
1152 if (config->cluster_seq_number == -1) // Don't have one. Just align to the master...
1153 config->cluster_seq_number = h->seq;
1154
1155 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1156
1157 if (config->cluster_seq_number != h->seq) { // Out of sequence heartbeat!
1158 log(1,0,0,0, "HB: Got seq# %d but was expecting %d. asking for resend.\n", h->seq, config->cluster_seq_number);
1159
1160 peer_send_message(addr, C_LASTSEEN, config->cluster_seq_number, NULL, 0);
1161
1162 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1163
1164 // Just drop the packet. The master will resend it as part of the catchup.
1165
1166 return 0;
1167 }
1168 // Save the packet in our buffer.
1169 // This is needed in case we become the master.
1170 config->cluster_seq_number = (h->seq+1)%HB_MAX_SEQ;
1171 i = h->seq % HB_HISTORY_SIZE;
1172 past_hearts[i].seq = h->seq;
1173 past_hearts[i].size = size;
1174 memcpy(&past_hearts[i].data, data, size); // Save it.
1175
1176
1177 // Check that we don't have too many undefined sessions, and
1178 // that the free session pointer is correct.
1179 cluster_check_sessions(h->highsession, h->freesession, h->hightunnel);
1180
1181 if (more > 2) // reserved section of heartt was not initialized prior to v3
1182 {
1183 if (h->interval != config->cluster_hb_interval)
1184 {
1185 log(2, 0, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
1186 h->interval, config->cluster_hb_interval);
1187
1188 config->cluster_hb_interval = h->interval;
1189 }
1190
1191 if (h->timeout != config->cluster_hb_timeout)
1192 {
1193 log(2, 0, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
1194 h->timeout, config->cluster_hb_timeout);
1195
1196 config->cluster_hb_timeout = h->timeout;
1197 }
1198 }
1199
1200 // Ok. process the packet...
1201 while ( s > 0) {
1202
1203 type = * ((u32*) p);
1204 p += sizeof(u32);
1205 s -= sizeof(u32);
1206
1207 more = * ((u32*) p);
1208 p += sizeof(u32);
1209 s -= sizeof(u32);
1210
1211 switch (type) {
1212 case C_CSESSION: { // Compressed session structure.
1213 u8 c [ sizeof(sessiont) + 2];
1214 int size;
1215 u8 * orig_p = p;
1216
1217 size = rle_decompress((u8 **) &p, s, c, sizeof(c) );
1218 s -= (p - orig_p);
1219
1220 if (size != sizeof(sessiont) ) { // Ouch! Very very bad!
1221 log(0,0,0,0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1222 // Now what? Should exit! No-longer up to date!
1223 break;
1224 }
1225
1226 cluster_recv_session(more, c);
1227 break;
1228 }
1229 case C_SESSION:
1230 if ( s < sizeof(session[more]))
1231 goto shortpacket;
1232
1233 cluster_recv_session(more, p);
1234
1235 p += sizeof(session[more]);
1236 s -= sizeof(session[more]);
1237 break;
1238
1239 case C_CTUNNEL: { // Compressed tunnel structure.
1240 u8 c [ sizeof(tunnelt) + 2];
1241 int size;
1242 u8 * orig_p = p;
1243
1244 size = rle_decompress( (u8 **) &p, s, c, sizeof(c) );
1245 s -= (p - orig_p);
1246
1247 if (size != sizeof(tunnelt) ) { // Ouch! Very very bad!
1248 log(0,0,0,0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1249 // Now what? Should exit! No-longer up to date!
1250 break;
1251 }
1252
1253 cluster_recv_tunnel(more, c);
1254 break;
1255
1256 }
1257 case C_TUNNEL:
1258 if ( s < sizeof(tunnel[more]))
1259 goto shortpacket;
1260
1261 cluster_recv_tunnel(more, p);
1262
1263 p += sizeof(tunnel[more]);
1264 s -= sizeof(tunnel[more]);
1265 break;
1266 default:
1267 log(0,0,0,0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type);
1268 return -1; // can't process any more of the packet!!
1269 }
1270 }
1271 if (config->cluster_master_address != addr)
1272 {
1273 char *str;
1274 str = strdup(inet_toa(config->cluster_master_address));
1275 log(0,0,0,0, "My master just changed from %s to %s!\n", str, inet_toa(addr));
1276 if (str) free(str);
1277 }
1278
1279 config->cluster_master_address = addr;
1280 config->cluster_last_hb = TIME; // Successfully received a heartbeat!
1281 return 0;
1282
1283 shortpacket:
1284 log(0,0,0,0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
1285 return -1;
1286 }
1287
1288 //
1289 // We got a packet on the cluster port!
1290 // Handle pings, lastseens, and heartbeats!
1291 //
1292 int processcluster(char * data, int size, u32 addr)
1293 {
1294 int type, more;
1295 char * p = data;
1296 int s = size;
1297
1298 if (addr == my_address)
1299 return -1; // Ignore it. Something looped back the multicast!
1300
1301 log(5,0,0,0, "Process cluster: %d bytes from %s\n", size, inet_toa(addr));
1302
1303 if (s <= 0) // Any data there??
1304 return -1;
1305
1306 if (s < 8)
1307 goto shortpacket;
1308
1309 type = * ((u32*) p);
1310 p += sizeof(u32);
1311 s -= sizeof(u32);
1312
1313 more = * ((u32*) p);
1314 p += sizeof(u32);
1315 s -= sizeof(u32);
1316
1317 switch (type) {
1318 case C_PING: // Update the peers table.
1319 return cluster_add_peer(addr, more, (pingt*)p);
1320
1321 case C_LASTSEEN: // Catch up a slave (slave missed a packet).
1322 return cluster_catchup_slave(more, addr);
1323
1324 case C_FORWARD: { // Forwarded control packet. pass off to processudp.
1325 struct sockaddr_in a;
1326 a.sin_addr.s_addr = more;
1327
1328 a.sin_port = * (int*) p;
1329 s -= sizeof(int);
1330 p += sizeof(int);
1331
1332 if (!config->cluster_iam_master) { // huh?
1333 log(0,0,0,0, "I'm not the master, but I got a C_FORWARD from %s?\n", inet_toa(addr));
1334 return -1;
1335 }
1336
1337 log(4,0,0,0, "Got a forwarded packet... (%s:%d)\n", inet_toa(more), a.sin_port);
1338 STAT(recv_forward);
1339 processudp(p, s, &a);
1340 return 0;
1341 }
1342 case C_THROTTLE: { // Receive a forwarded packet from a slave.
1343 if (!config->cluster_iam_master) {
1344 log(0,0,0,0, "I'm not the master, but I got a C_THROTTLE from %s?\n", inet_toa(addr));
1345 return -1;
1346 }
1347
1348 tbf_queue_packet(more, p, s); // The TBF id tells wether it goes in or out.
1349 return 0;
1350 }
1351 case C_GARDEN:
1352 // Receive a walled garden packet from a slave.
1353 if (!config->cluster_iam_master) {
1354 log(0,0,0,0, "I'm not the master, but I got a C_GARDEN from %s?\n", inet_toa(addr));
1355 return -1;
1356 }
1357
1358 tun_write(p, s);
1359 return 0;
1360
1361 case C_BYTES:
1362 return cluster_handle_bytes(p, s);
1363
1364 case C_KILL: // The master asked us to die!? (usually because we're too out of date).
1365 if (config->cluster_iam_master) {
1366 log(0,0,0,0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", inet_toa(addr), more);
1367 return -1;
1368 }
1369 if (more != config->cluster_seq_number) {
1370 log(0,0,0,0, "The master asked us to die but the seq number didn't match!?\n");
1371 return -1;
1372 }
1373
1374 if (addr != config->cluster_master_address) {
1375 log(0,0,0,0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%x)\n",
1376 inet_toa(addr), config->cluster_master_address);
1377 // We can only warn about it. The master might really have switched!
1378 }
1379
1380 log(0,0,0,0, "Received a valid C_KILL: I'm going to die now.\n");
1381 kill(0, SIGTERM);
1382 exit(0); // Lets be paranoid;
1383 return -1; // Just signalling the compiler.
1384
1385 case C_HEARTBEAT:
1386 log(4,0,0,0, "Got a heartbeat from %s\n", inet_toa(addr));
1387 return cluster_process_heartbeat(data, size, more, p, addr);
1388
1389 default:
1390 log(0,0,0,0, "Strange type packet received on cluster socket (%d)\n", type);
1391 return -1;
1392 }
1393 return 0;
1394
1395 shortpacket:
1396 log(0,0,0,0, "I got an cluster heartbeat packet! This means I'm probably out of sync!!\n");
1397 return -1;
1398 }
1399
1400 //====================================================================================================
1401
1402 int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
1403 {
1404 int i;
1405
1406 if (CLI_HELP_REQUESTED)
1407 return CLI_HELP_NO_ARGS;
1408
1409 cli_print(cli, "Cluster status : %s", config->cluster_iam_master ? "Master" : "Slave" );
1410 cli_print(cli, "My address : %s", inet_toa(my_address));
1411 cli_print(cli, "VIP address : %s", inet_toa(config->bind_address));
1412 cli_print(cli, "Multicast address: %s", inet_toa(config->cluster_address));
1413 cli_print(cli, "Multicast i'face : %s", config->cluster_interface);
1414
1415 if (!config->cluster_iam_master) {
1416 cli_print(cli, "My master : %s (last heartbeat %.1f seconds old)",
1417 config->cluster_master_address ? inet_toa(config->cluster_master_address) : "Not defined",
1418 0.1 * (TIME - config->cluster_last_hb));
1419 cli_print(cli, "Uptodate : %s", config->cluster_iam_uptodate ? "Yes" : "No");
1420 cli_print(cli, "Next sequence number expected: %d", config->cluster_seq_number);
1421 cli_print(cli, "%d sessions undefined of %d", config->cluster_undefined_sessions, config->cluster_highest_sessionid);
1422 cli_print(cli, "%d tunnels undefined of %d", config->cluster_undefined_tunnels, config->cluster_highest_tunnelid);
1423 } else {
1424 cli_print(cli, "Next heartbeat # : %d", config->cluster_seq_number);
1425 cli_print(cli, "Highest session : %d", config->cluster_highest_sessionid);
1426 cli_print(cli, "Highest tunnel : %d", config->cluster_highest_tunnelid);
1427 cli_print(cli, "%d changes queued for sending", config->cluster_num_changes);
1428 }
1429 cli_print(cli, "%d peers.", num_peers);
1430
1431 if (num_peers)
1432 cli_print(cli, "%20s %10s %8s", "Address", "Basetime", "Age");
1433 for (i = 0; i < num_peers; ++i) {
1434 cli_print(cli, "%20s %10d %8d", inet_toa(peers[i].peer),
1435 peers[i].basetime, TIME - peers[i].timestamp);
1436 }
1437 return CLI_OK;
1438 }
1439
1440 //
1441 // Simple run-length-encoding compression.
1442 // Format is
1443 // 1 byte < 128 = count of non-zero bytes following. // Not legal to be zero.
1444 // n non-zero bytes;
1445 // or
1446 // 1 byte > 128 = (count - 128) run of zero bytes. //
1447 // repeat.
1448 // count == 0 indicates end of compressed stream.
1449 //
1450 // Compress from 'src' into 'dst'. return number of bytes
1451 // used from 'dst'.
1452 // Updates *src_p to indicate 1 past last bytes used.
1453 //
1454 // We could get an extra byte in the zero runs by storing (count-1)
1455 // but I'm playing it safe.
1456 //
1457 // Worst case is a 50% expansion in space required (trying to
1458 // compress { 0x00, 0x01 } * N )
1459 int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize)
1460 {
1461 int count;
1462 int orig_dsize = dsize;
1463 u8 * x,*src;
1464 src = *src_p;
1465
1466 while (ssize > 0 && dsize > 2) {
1467 count = 0;
1468 x = dst++; --dsize; // Reserve space for count byte..
1469
1470 if (*src) { // Copy a run of non-zero bytes.
1471 while (*src && count < 127 && ssize > 0 && dsize > 1) { // Count number of non-zero bytes.
1472 *dst++ = *src++;
1473 --dsize; --ssize;
1474 ++count;
1475 }
1476 *x = count; // Store number of non-zero bytes. Guarenteed to be non-zero!
1477
1478 } else { // Compress a run of zero bytes.
1479 while (*src == 0 && count < 127 && ssize > 0) {
1480 ++src;
1481 --ssize;
1482 ++count;
1483 }
1484 *x = count | 0x80 ;
1485 }
1486 }
1487
1488 *dst++ = 0x0; // Add Stop byte.
1489 --dsize;
1490
1491 *src_p = src;
1492 return (orig_dsize - dsize);
1493 }
1494
1495 //
1496 // Decompress the buffer into **p.
1497 // 'psize' is the size of the decompression buffer available.
1498 //
1499 // Returns the number of bytes decompressed.
1500 //
1501 // Decompresses from '*src_p' into 'dst'.
1502 // Return the number of dst bytes used.
1503 // Updates the 'src_p' pointer to point to the
1504 // first un-used byte.
1505 int rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize)
1506 {
1507 int count;
1508 int orig_dsize = dsize;
1509 char * src = *src_p;
1510
1511 while (ssize >0 && dsize > 0) { // While there's more to decompress, and there's room in the decompress buffer...
1512 count = *src++; --ssize; // get the count byte from the source.
1513 if (count == 0x0) // End marker reached? If so, finish.
1514 break;
1515
1516 if (count & 0x80) { // Decompress a run of zeros
1517 for (count &= 0x7f ; count > 0 && dsize > 0; --count) {
1518 *dst++ = 0x0;
1519 --dsize;
1520 }
1521 } else { // Copy run of non-zero bytes.
1522 for ( ; count > 0 && ssize && dsize; --count) { // Copy non-zero bytes across.
1523 *dst++ = *src++;
1524 --ssize; --dsize;
1525 }
1526 }
1527 }
1528 *src_p = src;
1529 return (orig_dsize - dsize);
1530 }