l2tpns-dont-timeshift-unidirectional-traffic.patch
[l2tpns.git] / cluster.c
1 // L2TPNS Clustering Stuff
2
3 char const *cvs_id_cluster = "$Id: cluster.c,v 1.11 2004-09-02 04:18:07 fred_nerk Exp $";
4
5 #include <stdio.h>
6 #include <sys/file.h>
7 #include <sys/stat.h>
8 #include <sys/socket.h>
9 #include <netinet/in.h>
10 #include <arpa/inet.h>
11 #include <sys/ioctl.h>
12 #include <net/if.h>
13 #include <string.h>
14 #include <malloc.h>
15 #include <errno.h>
16 #include <stdlib.h>
17 #include <stdarg.h>
18 #include <unistd.h>
19 #include <stdio.h>
20 #include <libcli.h>
21
22 #include "l2tpns.h"
23 #include "cluster.h"
24 #include "util.h"
25 #include "tbf.h"
26
27 #ifdef BGP
28 #include "bgp.h"
29 #endif
30 /*
31 * All cluster packets have the same format.
32 *
33 * One or more instances of
34 * a 32 bit 'type' id.
35 * a 32 bit 'extra' data dependant on the 'type'.
36 * zero or more bytes of structure data, dependant on the type.
37 *
38 */
39
40 // Module variables.
41 int cluster_sockfd = 0; // The filedescriptor for the cluster communications port.
42
43 ipt my_address = 0; // The network address of my ethernet port.
44 static int walk_session_number = 0; // The next session to send when doing the slow table walk.
45 static int walk_tunnel_number = 0; // The next tunnel to send when doing the slow table walk.
46
47 #define MAX_HEART_SIZE (8192) // Maximum size of heartbeat packet. Must be less than max IP packet size :)
48 #define MAX_CHANGES (MAX_HEART_SIZE/(sizeof(sessiont) + sizeof(int) ) - 2) // Assumes a session is the biggest type!
49
50 static struct {
51 int type;
52 int id;
53 } cluster_changes[MAX_CHANGES]; // Queue of changed structures that need to go out when next heartbeat.
54
55 static struct {
56 int seq;
57 int size;
58 char data[MAX_HEART_SIZE];
59 } past_hearts[HB_HISTORY_SIZE]; // Ring buffer of heartbeats that we've recently sent out. Needed so
60 // we can re-transmit if needed.
61
62 static struct {
63 u32 peer;
64 time_t basetime;
65 clockt timestamp;
66 int uptodate;
67 } peers[CLUSTER_MAX_SIZE]; // List of all the peers we've heard from.
68 static int num_peers; // Number of peers in list.
69
70 int rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize);
71 int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize);
72
73 //
74 // Create a listening socket
75 //
76 // This joins the cluster multi-cast group.
77 //
78 int cluster_init()
79 {
80 struct sockaddr_in addr;
81 struct sockaddr_in interface_addr;
82 struct ip_mreq mreq;
83 struct ifreq ifr;
84 int opt = 0;
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 if (bind(cluster_sockfd, (void *) &addr, sizeof(addr)) < 0)
103 {
104 log(0, 0, 0, 0, "Failed to bind cluster socket: %s\n", strerror(errno));
105 return -1;
106 }
107
108 strcpy(ifr.ifr_name, config->cluster_interface);
109 if (ioctl(cluster_sockfd, SIOCGIFADDR, &ifr) < 0)
110 {
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 {
128 log(0, 0, 0, 0, "Failed to setsockopt (join mcast group): %s\n", strerror(errno));
129 return -1;
130 }
131
132 if (setsockopt (cluster_sockfd, IPPROTO_IP, IP_MULTICAST_IF, &interface_addr, sizeof(interface_addr)) < 0)
133 {
134 log(0, 0, 0, 0, "Failed to setsockopt (set mcast interface): %s\n", strerror(errno));
135 return -1;
136 }
137
138 config->cluster_last_hb = TIME;
139 config->cluster_seq_number = -1;
140
141 return cluster_sockfd;
142 }
143
144
145 //
146 // Send a chunk of data to the entire cluster (usually via the multicast
147 // address ).
148 //
149
150 int cluster_send_data(void *data, int datalen)
151 {
152 struct sockaddr_in addr = {0};
153
154 if (!cluster_sockfd) return -1;
155 if (!config->cluster_address) return 0;
156
157 addr.sin_addr.s_addr = config->cluster_address;
158 addr.sin_port = htons(CLUSTERPORT);
159 addr.sin_family = AF_INET;
160
161 log(5,0,0,0, "Cluster send data: %d bytes\n", datalen);
162
163 if (sendto(cluster_sockfd, data, datalen, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
164 {
165 log(0, 0, 0, 0, "sendto: %s\n", strerror(errno));
166 return -1;
167 }
168
169 return 0;
170 }
171
172 //
173 // Add a chunk of data to a heartbeat packet.
174 // Maintains the format. Assumes that the caller
175 // has passed in a big enough buffer!
176 //
177 static void add_type(char ** p, int type, int more, char * data, int size)
178 {
179 * ( (u32*)(*p) ) = type;
180 *p += sizeof(u32);
181
182 * ( (u32*)(*p) ) = more;
183 *p += sizeof(u32);
184
185 if (data && size > 0) {
186 memcpy(*p, data, size);
187 (*p) += size;
188 }
189 }
190
191 void cluster_uptodate(void)
192 {
193 if (config->cluster_iam_uptodate)
194 return;
195
196 if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels)
197 return;
198
199 config->cluster_iam_uptodate = 1;
200
201 log(0,0,0,0, "Now uptodate with master.\n");
202
203 #ifdef BGP
204 if (bgp_configured)
205 bgp_enable_routing(1);
206 else
207 #endif /* BGP */
208 if (config->send_garp)
209 send_garp(config->bind_address); // Start taking traffic.
210 }
211
212 //
213 // Send a unicast UDP packet to a peer with 'data' as the
214 // contents.
215 //
216 int peer_send_data(u32 peer, char * data, int size)
217 {
218 struct sockaddr_in addr = {0};
219
220 if (!cluster_sockfd) return -1;
221 if (!config->cluster_address) return 0;
222
223 if (!peer) // Odd??
224 return -1;
225
226 addr.sin_addr.s_addr = peer;
227 addr.sin_port = htons(CLUSTERPORT);
228 addr.sin_family = AF_INET;
229
230 log_hex(5, "Peer send", data, size);
231
232 if (sendto(cluster_sockfd, data, size, MSG_NOSIGNAL, (void *) &addr, sizeof(addr)) < 0)
233 {
234 log(0, 0, 0, 0, "sendto: %s\n", strerror(errno));
235 return -1;
236 }
237
238 return 0;
239 }
240
241 //
242 // Send a structured message to a peer with a single element of type 'type'.
243 //
244 int peer_send_message(u32 peer, int type, int more, char * data, int size)
245 {
246 char buf[65536]; // Vast overkill.
247 char * p = buf;
248
249 log(4,0,0,0, "Sending message to peer (type %d, more %d, size %d)\n", type, more, size);
250 add_type(&p, type, more, data, size);
251
252 return peer_send_data(peer, buf, (p-buf) );
253 }
254
255 //
256 // Forward a state changing packet to the master.
257 //
258 // The master just processes the payload as if it had
259 // received it off the tap device.
260 //
261 int master_forward_packet(char * data, int size, u32 addr, int port)
262 {
263 char buf[65536]; // Vast overkill.
264 char * p = buf;
265
266 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
267 return -1;
268
269 log(4,0,0,0, "Forwarding packet from %s to master (size %d)\n", inet_toa(addr), size);
270
271 STAT(c_forwarded);
272 add_type(&p, C_FORWARD, addr, (char*) &port, sizeof(port) );
273 memcpy(p, data, size);
274 p += size;
275
276 return peer_send_data(config->cluster_master_address, buf, (p-buf) );
277
278 }
279
280 //
281 // Forward a throttled packet to the master for handling.
282 //
283 // The master just drops the packet into the appropriate
284 // token bucket queue, and lets normal processing take care
285 // of it.
286 //
287 int master_throttle_packet(int tbfid, char * data, int size)
288 {
289 char buf[65536]; // Vast overkill.
290 char * p = buf;
291
292 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
293 return -1;
294
295 log(4,0,0,0, "Throttling packet master (size %d, tbfid %d)\n", size, tbfid);
296
297 add_type(&p, C_THROTTLE, tbfid, data, size);
298
299 return peer_send_data(config->cluster_master_address, buf, (p-buf) );
300
301 }
302
303 //
304 // Forward a walled garden packet to the master for handling.
305 //
306 // The master just writes the packet straight to the tun
307 // device (where is will normally loop through the
308 // firewall rules, and come back in on the tun device)
309 //
310 // (Note that this must be called with the tun header
311 // as the start of the data).
312 int master_garden_packet(sessionidt s, char *data, int size)
313 {
314 char buf[65536]; // Vast overkill.
315 char *p = buf;
316
317 if (!config->cluster_master_address) // No election has been held yet. Just skip it.
318 return -1;
319
320 log(4,0,0,0, "Walled garden packet to master (size %d)\n", size);
321
322 add_type(&p, C_GARDEN, s, data, size);
323
324 return peer_send_data(config->cluster_master_address, buf, (p-buf));
325
326 }
327
328 //
329 // Send a chunk of data as a heartbeat..
330 // We save it in the history buffer as we do so.
331 //
332 static void send_heartbeat(int seq, char * data, int size)
333 {
334 int i;
335
336 if (size > sizeof(past_hearts[0].data))
337 {
338 log(0,0,0,0, "Tried to heartbeat something larger than the maximum packet!\n");
339 kill(0, SIGTERM);
340 exit(1);
341 }
342 i = seq % HB_HISTORY_SIZE;
343 past_hearts[i].seq = seq;
344 past_hearts[i].size = size;
345 memcpy(&past_hearts[i].data, data, size); // Save it.
346 cluster_send_data(data, size);
347 }
348
349 //
350 // Send an 'i am alive' message to every machine in the cluster.
351 //
352 void cluster_send_ping(time_t basetime)
353 {
354 char buff[100 + sizeof(pingt)];
355 char *p = buff;
356 pingt x;
357
358 if (config->cluster_iam_master && basetime) // We're heartbeating so no need to ping.
359 return;
360
361 log(5,0,0,0, "Sending cluster ping...\n");
362
363 x.ver = 1;
364 x.addr = config->bind_address;
365 x.undef = config->cluster_undefined_sessions + config->cluster_undefined_tunnels;
366 x.basetime = basetime;
367
368 add_type(&p, C_PING, basetime, (char *) &x, sizeof(x));
369 cluster_send_data(buff, (p-buff) );
370 }
371
372 //
373 // Walk the session counters looking for non-zero ones to send
374 // to the master. We send up to 100 of them at one time.
375 // We examine a maximum of 2000 sessions.
376 // (50k max session should mean that we normally
377 // examine the entire session table every 25 seconds).
378
379 #define MAX_B_RECS (400)
380 void master_update_counts(void)
381 {
382 int i, c;
383 bytest b[MAX_B_RECS+1];
384
385 if (config->cluster_iam_master) // Only happens on the slaves.
386 return;
387
388 if (!config->cluster_master_address) // If we don't have a master, skip it for a while.
389 return;
390
391 i = MAX_B_RECS * 5; // Examine max 2000 sessions;
392 if (config->cluster_highest_sessionid > i)
393 i = config->cluster_highest_sessionid;
394
395 for ( c = 0; i > 0 ; --i) {
396 // Next session to look at.
397 walk_session_number++;
398 if ( walk_session_number > config->cluster_highest_sessionid)
399 walk_session_number = 1;
400
401 if (!sess_count[walk_session_number].cin && !sess_count[walk_session_number].cout)
402 continue; // Unused. Skip it.
403
404 b[c].sid = walk_session_number;
405 b[c].in = sess_count[walk_session_number].cin;
406 b[c].out = sess_count[walk_session_number].cout;
407
408 if (++c > MAX_B_RECS) // Send a max of 400 elements in a packet.
409 break;
410
411 // Reset counters.
412 sess_count[walk_session_number].cin = sess_count[walk_session_number].cout = 0;
413 }
414
415 if (!c) // Didn't find any that changes. Get out of here!
416 return;
417
418
419 // Forward the data to the master.
420 log(4,0,0,0, "Sending byte counters to master (%d elements)\n", c);
421 peer_send_message(config->cluster_master_address, C_BYTES, c, (char*) &b, sizeof(b[0]) * c);
422 return;
423 }
424
425 //
426 // On the master, check how our slaves are going. If
427 // one of them's not up-to-date we'll heartbeat faster.
428 // If we don't have any of them, then we need to turn
429 // on our own packet handling!
430 //
431 void cluster_check_slaves(void)
432 {
433 int i;
434 static int have_peers = 0;
435 int had_peers = have_peers;
436 clockt t = TIME;
437
438 if (!config->cluster_iam_master)
439 return; // Only runs on the master...
440
441 config->cluster_iam_uptodate = 1; // cleared in loop below
442
443 for (i = have_peers = 0; i < num_peers; i++)
444 {
445 if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
446 continue; // Stale peer! Skip them.
447
448 if (!peers[i].basetime)
449 continue; // Shutdown peer! Skip them.
450
451 if (peers[i].uptodate)
452 have_peers = 1;
453
454 if (!peers[i].uptodate)
455 config->cluster_iam_uptodate = 0; // Start fast heartbeats
456 }
457
458 #ifdef BGP
459 // master lost all slaves, need to handle traffic ourself
460 if (bgp_configured && had_peers && !have_peers)
461 bgp_enable_routing(1);
462 else if (bgp_configured && !had_peers && have_peers)
463 bgp_enable_routing(0);
464 #endif /* BGP */
465 }
466
467 //
468 // Check that we have a master. If it's been too
469 // long since we heard from a master then hold an election.
470 //
471 void cluster_check_master(void)
472 {
473 int i, count, tcount, high_unique_id = 0;
474 int last_free = 0;
475 clockt t = TIME;
476 static int probed = 0;
477
478 if (config->cluster_iam_master)
479 return; // Only runs on the slaves...
480
481 // If the master is late (missed 2 hearbeats by a second and a
482 // hair) it may be that the switch has dropped us from the
483 // multicast group, try unicasting one probe to the master
484 // which will hopefully respond with a unicast heartbeat that
485 // will allow us to limp along until the querier next runs.
486 if (TIME > (config->cluster_last_hb + 2 * config->cluster_hb_interval + 11))
487 {
488 if (!probed && config->cluster_master_address)
489 {
490 probed = 1;
491 log(1, 0, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
492 0.1 * (TIME - (config->cluster_last_hb + config->cluster_hb_interval)));
493
494 peer_send_message(config->cluster_master_address,
495 C_LASTSEEN, config->cluster_seq_number, NULL, 0);
496 }
497 } else { // We got a recent heartbeat; reset the probe flag.
498 probed = 0;
499 }
500
501 if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout))
502 return; // Everything's ok!
503
504 config->cluster_last_hb = TIME + 1; // Just the one election thanks.
505
506 log(0,0,0,0, "Master timed out! Holding election...\n");
507
508 for (i = 0; i < num_peers; i++)
509 {
510 if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
511 continue; // Stale peer! Skip them.
512
513 if (!peers[i].basetime)
514 continue; // Shutdown peer! Skip them.
515
516 if (peers[i].basetime < basetime) {
517 log(1,0,0,0, "Expecting %s to become master\n", inet_toa(peers[i].peer) );
518 return; // They'll win the election. Get out of here.
519 }
520
521 if (peers[i].basetime == basetime &&
522 peers[i].peer > my_address) {
523 log(1,0,0,0, "Expecting %s to become master\n", inet_toa(peers[i].peer) );
524 return; // They'll win the election. Wait for them to come up.
525 }
526 }
527
528 // Wow. it's been ages since I last heard a heartbeat
529 // and I'm better than an of my peers so it's time
530 // to become a master!!!
531
532 config->cluster_iam_master = 1;
533 config->cluster_master_address = 0;
534
535 log(0,0,0,0, "I am declaring myself the master!\n");
536
537 if (config->cluster_seq_number == -1)
538 config->cluster_seq_number = 0;
539
540 //
541 // Go through and mark all the tunnels as defined.
542 // Count the highest used tunnel number as well.
543 //
544 config->cluster_highest_tunnelid = 0;
545 for (i = 0, tcount = 0; i < MAXTUNNEL; ++i) {
546 if (tunnel[i].state == TUNNELUNDEF)
547 tunnel[i].state = TUNNELFREE;
548
549 if (tunnel[i].state != TUNNELFREE && i > config->cluster_highest_tunnelid)
550 config->cluster_highest_tunnelid = i;
551 }
552
553 //
554 // Go through and mark all the sessions as being defined.
555 // reset the idle timeouts.
556 // add temporary byte counters to permanent ones.
557 // Re-string the free list.
558 // Find the ID of the highest session.
559 last_free = 0;
560 high_unique_id = 0;
561 config->cluster_highest_sessionid = 0;
562 for (i = 0, count = 0; i < MAXSESSION; ++i) {
563 if (session[i].tunnel == T_UNDEF) {
564 session[i].tunnel = T_FREE;
565 ++count;
566 }
567
568 if (session[i].tunnel == T_FREE) { // Unused session. Add to free list.
569 session[last_free].next = i;
570 session[i].next = 0;
571 last_free = i;
572 }
573
574 // Reset all the idle timeouts..
575 session[i].last_packet = time_now;
576
577 // Accumulate un-sent byte counters.
578 session[i].cin += sess_count[i].cin;
579 session[i].cout += sess_count[i].cout;
580 session[i].total_cin += sess_count[i].cin;
581 session[i].total_cout += sess_count[i].cout;
582
583 sess_count[i].cin = sess_count[i].cout = 0;
584
585 session[i].radius = 0; // Reset authentication as the radius blocks aren't up to date.
586
587 if (session[i].unique_id >= high_unique_id) // This is different to the index into the session table!!!
588 high_unique_id = session[i].unique_id+1;
589
590
591 session[i].tbf_in = session[i].tbf_out = 0; // Remove stale pointers from old master.
592 throttle_session(i, session[i].throttle);
593
594 // I'm unsure about this. --mo
595 // It's potentially a good thing, but it could send a
596 // LOT of packets.
597 // if (session[i].throttle)
598 // cluster_send_session(s); // Tell the slaves about the new tbf indexes.
599
600 if (session[i].tunnel != T_FREE && i > config->cluster_highest_sessionid)
601 config->cluster_highest_sessionid = i;
602
603 }
604
605 session[last_free].next = 0; // End of chain.
606 last_id = high_unique_id; // Keep track of the highest used session ID.
607
608 become_master();
609
610 rebuild_address_pool();
611
612 // If we're not the very first master, this is a big issue!
613 if(count>0)
614 log(0,0,0,0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count);
615
616 config->cluster_undefined_sessions = 0;
617 config->cluster_undefined_tunnels = 0;
618 config->cluster_iam_uptodate = 1; // assume all peers are up-to-date
619
620 // FIXME. We need to fix up the tunnel control message
621 // queue here! There's a number of other variables we
622 // should also update.
623 }
624
625
626 //
627 // Check that our session table is validly matching what the
628 // master has in mind.
629 //
630 // In particular, if we have too many sessions marked 'undefined'
631 // we fix it up here, and we ensure that the 'first free session'
632 // pointer is valid.
633 //
634 static void cluster_check_sessions(int highsession, int freesession_ptr, int hightunnel)
635 {
636 int i;
637
638 sessionfree = freesession_ptr; // Keep the freesession ptr valid.
639
640 if (config->cluster_iam_uptodate)
641 return;
642
643 if (highsession > config->cluster_undefined_sessions && hightunnel > config->cluster_undefined_tunnels)
644 return;
645
646 // Clear out defined sessions, counting the number of
647 // undefs remaining.
648 config->cluster_undefined_sessions = 0;
649 for (i = 1 ; i < MAXSESSION; ++i) {
650 if (i > highsession) {
651 session[i].tunnel = 0; // Defined.
652 continue;
653 }
654 if (session[i].tunnel != T_UNDEF)
655 continue;
656 ++config->cluster_undefined_sessions;
657 }
658
659 // Clear out defined tunnels, counting the number of
660 // undefs remaining.
661 config->cluster_undefined_tunnels = 0;
662 for (i = 1 ; i < MAXTUNNEL; ++i) {
663 if (i > hightunnel) {
664 tunnel[i].state = TUNNELFREE; // Defined.
665 continue;
666 }
667 if (tunnel[i].state != TUNNELUNDEF)
668 continue;
669 ++config->cluster_undefined_tunnels;
670 }
671
672
673 if (config->cluster_undefined_sessions || config->cluster_undefined_tunnels) {
674 log(2,0,0,0, "Cleared undefined sessions/tunnels. %d sess (high %d), %d tunn (high %d)\n",
675 config->cluster_undefined_sessions, highsession, config->cluster_undefined_tunnels, hightunnel);
676 return;
677 }
678
679 // Are we up to date?
680
681 if (!config->cluster_iam_uptodate)
682 cluster_uptodate();
683 }
684
685 int hb_add_type(char **p, int type, int id)
686 {
687 switch (type) {
688 case C_CSESSION: { // Compressed C_SESSION.
689 u8 c[sizeof(sessiont) * 2]; // Bigger than worst case.
690 u8 *d = (u8 *) &session[id];
691 u8 *orig = d;
692 int size;
693
694 size = rle_compress( &d, sizeof(sessiont), c, sizeof(c) );
695
696 // Did we compress the full structure, and is the size actually
697 // reduced??
698 if ( (d - orig) == sizeof(sessiont) && size < sizeof(sessiont) ) {
699 add_type(p, C_CSESSION, id, (char*) c, size);
700 break;
701 }
702 // Failed to compress : Fall through.
703 }
704 case C_SESSION: add_type(p, C_SESSION, id,
705 (char*) &session[id], sizeof(sessiont));
706 break;
707
708 case C_CTUNNEL: { // Compressed C_TUNNEL
709 u8 c[sizeof(tunnelt) * 2]; // Bigger than worst case.
710 u8 *d = (u8 *) &tunnel[id];
711 u8 *orig = d;
712 int size;
713
714 size = rle_compress( &d, sizeof(tunnelt), c, sizeof(c) );
715
716 // Did we compress the full structure, and is the size actually
717 // reduced??
718 if ( (d - orig) == sizeof(tunnelt) && size < sizeof(tunnelt) ) {
719 add_type(p, C_CTUNNEL, id, c, size);
720 break;
721 }
722 // Failed to compress : Fall through.
723 }
724 case C_TUNNEL: add_type(p, C_TUNNEL, id,
725 (char*) &tunnel[id], sizeof(tunnelt));
726 break;
727 default:
728 log(0,0,0,0, "Found an invalid type in heart queue! (%d)\n", type);
729 kill(0, SIGTERM);
730 exit(1);
731 }
732 return 0;
733 }
734
735 //
736 // Send a heartbeat, incidently sending out any queued changes..
737 //
738 void cluster_heartbeat()
739 {
740 int i, count = 0, tcount = 0;
741 char buff[MAX_HEART_SIZE + sizeof(heartt) + sizeof(int) ];
742 heartt h;
743 char *p = buff;
744
745 if (!config->cluster_iam_master) // Only the master does this.
746 return;
747
748 // Fill out the heartbeat header.
749 memset(&h, 0, sizeof(h));
750
751 h.version = HB_VERSION;
752 h.seq = config->cluster_seq_number;
753 h.basetime = basetime;
754 h.clusterid = config->bind_address; // Will this do??
755 h.basetime = basetime;
756 h.highsession = config->cluster_highest_sessionid;
757 h.freesession = sessionfree;
758 h.hightunnel = config->cluster_highest_tunnelid;
759 h.size_sess = sizeof(sessiont); // Just in case.
760 h.size_tunn = sizeof(tunnelt);
761 h.interval = config->cluster_hb_interval;
762 h.timeout = config->cluster_hb_timeout;
763
764 add_type(&p, C_HEARTBEAT, HB_VERSION, (char*) &h, sizeof(h));
765
766 for (i = 0; i < config->cluster_num_changes; ++i) {
767 hb_add_type(&p, cluster_changes[i].type, cluster_changes[i].id);
768 }
769
770 if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
771 log(0,0,0,0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", p - buff);
772 kill(0, SIGTERM);
773 exit(1);
774 }
775
776 //
777 // Fill out the packet with sessions from the session table...
778 // (not forgetting to leave space so we can get some tunnels in too )
779 while ( (p + sizeof(u32) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) {
780
781 if (!walk_session_number) // session #0 isn't valid.
782 ++walk_session_number;
783
784 if (count >= config->cluster_highest_sessionid) // If we're a small cluster, don't go wild.
785 break;
786
787 hb_add_type(&p, C_CSESSION, walk_session_number);
788 walk_session_number = (1+walk_session_number)%(config->cluster_highest_sessionid+1); // +1 avoids divide by zero.
789
790 ++count; // Count the number of extra sessions we're sending.
791 }
792
793 //
794 // Fill out the packet with tunnels from the tunnel table...
795 // This effectively means we walk the tunnel table more quickly
796 // than the session table. This is good because stuffing up a
797 // tunnel is a much bigger deal than stuffing up a session.
798 //
799 while ( (p + sizeof(u32) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {
800
801 if (!walk_tunnel_number) // tunnel #0 isn't valid.
802 ++walk_tunnel_number;
803
804 if (tcount >= config->cluster_highest_tunnelid)
805 break;
806
807 hb_add_type(&p, C_CTUNNEL, walk_tunnel_number);
808 walk_tunnel_number = (1+walk_tunnel_number)%(config->cluster_highest_tunnelid+1); // +1 avoids divide by zero.
809
810 ++tcount;
811 }
812
813 //
814 // Did we do something wrong?
815 if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
816 log(0,0,0,0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", p - buff);
817 kill(0, SIGTERM);
818 exit(1);
819 }
820
821 log(3,0,0,0, "Sending heartbeat #%d with %d changes (%d x-sess, %d x-tunnels, %d highsess, %d hightun size %d)\n",
822 h.seq, config->cluster_num_changes, count, tcount, config->cluster_highest_sessionid,
823 config->cluster_highest_tunnelid, (p-buff));
824
825 config->cluster_num_changes = 0;
826
827 send_heartbeat(h.seq, buff, (p-buff) ); // Send out the heartbeat to the cluster, keeping a copy of it.
828
829 config->cluster_seq_number = (config->cluster_seq_number+1)%HB_MAX_SEQ; // Next seq number to use.
830 }
831
832 //
833 // A structure of type 'type' has changed; Add it to the queue to send.
834 //
835 int type_changed(int type, int id)
836 {
837 int i;
838
839 for (i = 0 ; i < config->cluster_num_changes ; ++i)
840 if ( cluster_changes[i].id == id &&
841 cluster_changes[i].type == type)
842 return 0; // Already marked for change.
843
844 cluster_changes[i].type = type;
845 cluster_changes[i].id = id;
846 ++config->cluster_num_changes;
847
848 if (config->cluster_num_changes > MAX_CHANGES)
849 cluster_heartbeat(); // flush now
850
851 return 1;
852 }
853
854
855 // A particular session has been changed!
856 int cluster_send_session(int sid)
857 {
858 if (!config->cluster_iam_master) {
859 log(0,0,sid,0, "I'm not a master, but I just tried to change a session!\n");
860 return -1;
861 }
862
863 return type_changed(C_CSESSION, sid);
864 }
865
866 // A particular tunnel has been changed!
867 int cluster_send_tunnel(int tid)
868 {
869 if (!config->cluster_iam_master) {
870 log(0,0,0,tid, "I'm not a master, but I just tried to change a tunnel!\n");
871 return -1;
872 }
873
874 return type_changed(C_CTUNNEL, tid);
875 }
876
877
878 //
879 // We're a master, and a slave has just told us that it's
880 // missed a packet. We'll resend it every packet since
881 // the last one it's seen.
882 //
883 int cluster_catchup_slave(int seq, u32 slave)
884 {
885 int s;
886 int diff;
887
888 log(1,0,0,0, "Slave %s sent LASTSEEN with seq %d\n", inet_toa(slave), seq);
889
890 diff = config->cluster_seq_number - seq; // How many packet do we need to send?
891 if (diff < 0)
892 diff += HB_MAX_SEQ;
893
894 if (diff >= HB_HISTORY_SIZE) { // Ouch. We don't have the packet to send it!
895 log(0,0,0,0, "A slaved asked for message %d when our seq number is %d. Killing it.\n",
896 seq, config->cluster_seq_number);
897 return peer_send_message(slave, C_KILL, seq, NULL, 0);// Kill the slave. Nothing else to do.
898 }
899
900 // Now resend every packet that it missed, in order.
901 while (seq != config->cluster_seq_number) {
902 s = seq%HB_HISTORY_SIZE;
903 if (seq != past_hearts[s].seq) {
904 log(0,0,0,0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n",
905 inet_toa(slave), seq, past_hearts[s].seq, s, config->cluster_seq_number);
906 return -1; // What to do here!?
907 }
908 peer_send_data(slave, past_hearts[s].data, past_hearts[s].size);
909 seq = (seq+1)%HB_MAX_SEQ; // Increment to next seq number.
910 }
911 return 0; // All good!
912 }
913
914 //
915 // We've heard from another peer! Add it to the list
916 // that we select from at election time.
917 //
918 int cluster_add_peer(u32 peer, time_t basetime, pingt *pp, int size)
919 {
920 int i;
921 u32 clusterid;
922 pingt p;
923
924 // Allow for backward compatability.
925 // Just the ping packet into a new structure to allow
926 // for the possibility that we might have received
927 // more or fewer elements than we were expecting.
928 if (size > sizeof(p))
929 size = sizeof(p);
930
931 memset( (void*) &p, 0, sizeof(p) );
932 memcpy( (void*) &p, (void*) pp, size);
933
934 clusterid = p.addr;
935 if (clusterid != config->bind_address)
936 {
937 // Is this for us?
938 log(4,0,0,0, "Skipping ping from %s (different cluster)\n", inet_toa(peer));
939 return 0;
940 }
941
942 for (i = 0; i < num_peers ; ++i)
943 {
944 if (peers[i].peer != peer)
945 continue;
946
947 // This peer already exists. Just update the timestamp.
948 peers[i].basetime = basetime;
949 peers[i].timestamp = TIME;
950 peers[i].uptodate = !p.undef;
951 break;
952 }
953
954 // Is this the master shutting down??
955 if (peer == config->cluster_master_address && !basetime) {
956 log(3,0,0,0, "Master %s shutting down...\n", inet_toa(config->cluster_master_address));
957 config->cluster_master_address = 0;
958 config->cluster_last_hb = 0; // Force an election.
959 cluster_check_master();
960 return 0;
961 }
962
963 if (i >= num_peers)
964 {
965 log(4,0,0,0, "Adding %s as a peer\n", inet_toa(peer));
966
967 // Not found. Is there a stale slot to re-use?
968 for (i = 0; i < num_peers ; ++i)
969 {
970 if (!peers[i].basetime) // Shutdown
971 break;
972
973 if ((peers[i].timestamp + config->cluster_hb_timeout * 10) < TIME) // Stale.
974 break;
975 }
976
977 if (i >= CLUSTER_MAX_SIZE)
978 {
979 // Too many peers!!
980 log(0,0,0,0, "Tried to add %s as a peer, but I already have %d of them!\n", inet_toa(peer), i);
981 return -1;
982 }
983
984 peers[i].peer = peer;
985 peers[i].basetime = basetime;
986 peers[i].timestamp = TIME;
987 peers[i].uptodate = !p.undef;
988 if (i == num_peers)
989 ++num_peers;
990
991 log(1,0,0,0, "Added %s as a new peer. Now %d peers\n", inet_toa(peer), num_peers);
992 }
993
994 return 1;
995 }
996
997 /* Handle the slave updating the byte counters for the master. */
998 //
999 // Note that we don't mark the session as dirty; We rely on
1000 // the slow table walk to propogate this back out to the slaves.
1001 //
1002 int cluster_handle_bytes(char * data, int size)
1003 {
1004 bytest * b;
1005
1006 b = (bytest*) data;
1007
1008 log(3,0,0,0, "Got byte counter update (size %d)\n", size);
1009
1010 /* Loop around, adding the byte
1011 counts to each of the sessions. */
1012
1013 while (size >= sizeof(*b) ) {
1014 if (b->sid > MAXSESSION) {
1015 log(0,0,0,0, "Got C_BYTES with session #%d!\n", b->sid);
1016 return -1; /* Abort processing */
1017 }
1018
1019 session[b->sid].total_cin += b->in;
1020 session[b->sid].total_cout += b->out;
1021
1022 session[b->sid].cin += b->in;
1023 session[b->sid].cout += b->out;
1024 session[b->sid].last_packet = time_now; // Reset idle timer!
1025
1026 size -= sizeof(*b);
1027 ++b;
1028 }
1029
1030 if (size != 0)
1031 log(0,0,0,0, "Got C_BYTES with %d bytes of trailing junk!\n", size);
1032
1033 return size;
1034 }
1035
1036 //
1037 // Handle receiving a session structure in a heartbeat packet.
1038 //
1039 static int cluster_recv_session(int more , u8 * p)
1040 {
1041 if (more >= MAXSESSION) {
1042 log(0,0,0,0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
1043 return -1;
1044 }
1045
1046 if (session[more].tunnel == T_UNDEF) {
1047 if (config->cluster_iam_uptodate) { // Sanity.
1048 log(0,0,0,0, "I thought I was uptodate but I just found an undefined session!\n");
1049 } else {
1050 --config->cluster_undefined_sessions;
1051 }
1052 }
1053
1054 load_session(more, (sessiont*) p); // Copy session into session table..
1055
1056 log(5,0,more,0, "Received session update (%d undef)\n", config->cluster_undefined_sessions);
1057
1058 if (!config->cluster_iam_uptodate)
1059 cluster_uptodate(); // Check to see if we're up to date.
1060
1061 return 0;
1062 }
1063
1064 static int cluster_recv_tunnel(int more, u8 *p)
1065 {
1066 if (more >= MAXTUNNEL) {
1067 log(0,0,0,0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
1068 return -1;
1069 }
1070
1071 if (tunnel[more].state == TUNNELUNDEF) {
1072 if (config->cluster_iam_uptodate) { // Sanity.
1073 log(0,0,0,0, "I thought I was uptodate but I just found an undefined tunnel!\n");
1074 } else {
1075 --config->cluster_undefined_tunnels;
1076 }
1077 }
1078
1079 memcpy(&tunnel[more], p, sizeof(tunnel[more]) );
1080
1081 //
1082 // Clear tunnel control messages. These are dynamically allocated.
1083 // If we get unlucky, this may cause the tunnel to drop!
1084 //
1085 tunnel[more].controls = tunnel[more].controle = NULL;
1086 tunnel[more].controlc = 0;
1087
1088 log(5,0,0,more, "Received tunnel update\n");
1089
1090 if (!config->cluster_iam_uptodate)
1091 cluster_uptodate(); // Check to see if we're up to date.
1092
1093 return 0;
1094 }
1095
1096
1097 //
1098 // Process a heartbeat..
1099 //
1100 static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32 addr)
1101 {
1102 heartt * h;
1103 int s = size - (p-data);
1104 int i, type;
1105
1106 #if HB_VERSION != 3
1107 # error "need to update cluster_process_heartbeat()"
1108 #endif
1109
1110 // we handle version 2+
1111 if (more < 2 || more > HB_VERSION) {
1112 log(0,0,0,0, "Received a heartbeat version that I don't support (%d)!\n", more);
1113 return -1; // Ignore it??
1114 }
1115
1116 // Ok. It's a heartbeat packet from a cluster master!
1117 if (s < sizeof(*h))
1118 goto shortpacket;
1119
1120 h = (heartt*) p;
1121 p += sizeof(*h);
1122 s -= sizeof(*h);
1123
1124 if (h->clusterid != config->bind_address)
1125 return -1; // It's not part of our cluster.
1126
1127 if (config->cluster_iam_master) { // Sanity...
1128 // Note that this MUST match the election process above!
1129
1130 log(0,0,0,0, "I just got a packet claiming to be from a master but _I_ am the master!\n");
1131 if (!h->basetime) {
1132 log(0,0,0,0, "Heartbeat from addr %s with zero basetime!\n", inet_toa(addr) );
1133 return -1; // Skip it.
1134 }
1135 if (basetime > h->basetime) {
1136 log(0,0,0,0, "They're (%s) an older master than me so I'm gone!\n", inet_toa(addr));
1137 kill(0, SIGTERM);
1138 exit(1);
1139 }
1140 if (basetime == h->basetime && my_address < addr) { // Tie breaker.
1141 log(0,0,0,0, "They're a higher IP address than me, so I'm gone!\n");
1142 kill(0, SIGTERM);
1143 exit(1);
1144 }
1145 return -1; // Skip it.
1146 }
1147
1148 if (config->cluster_seq_number == -1) // Don't have one. Just align to the master...
1149 config->cluster_seq_number = h->seq;
1150
1151 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1152
1153 if (config->cluster_seq_number != h->seq) { // Out of sequence heartbeat!
1154 log(1,0,0,0, "HB: Got seq# %d but was expecting %d. asking for resend.\n", h->seq, config->cluster_seq_number);
1155
1156 peer_send_message(addr, C_LASTSEEN, config->cluster_seq_number, NULL, 0);
1157
1158 config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
1159
1160 // Just drop the packet. The master will resend it as part of the catchup.
1161
1162 return 0;
1163 }
1164 // Save the packet in our buffer.
1165 // This is needed in case we become the master.
1166 config->cluster_seq_number = (h->seq+1)%HB_MAX_SEQ;
1167 i = h->seq % HB_HISTORY_SIZE;
1168 past_hearts[i].seq = h->seq;
1169 past_hearts[i].size = size;
1170 memcpy(&past_hearts[i].data, data, size); // Save it.
1171
1172
1173 // Check that we don't have too many undefined sessions, and
1174 // that the free session pointer is correct.
1175 cluster_check_sessions(h->highsession, h->freesession, h->hightunnel);
1176
1177 if (more > 2) // reserved section of heartt was not initialized prior to v3
1178 {
1179 if (h->interval != config->cluster_hb_interval)
1180 {
1181 log(2, 0, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
1182 h->interval, config->cluster_hb_interval);
1183
1184 config->cluster_hb_interval = h->interval;
1185 }
1186
1187 if (h->timeout != config->cluster_hb_timeout)
1188 {
1189 log(2, 0, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
1190 h->timeout, config->cluster_hb_timeout);
1191
1192 config->cluster_hb_timeout = h->timeout;
1193 }
1194 }
1195
1196 // Ok. process the packet...
1197 while ( s > 0) {
1198
1199 type = * ((u32*) p);
1200 p += sizeof(u32);
1201 s -= sizeof(u32);
1202
1203 more = * ((u32*) p);
1204 p += sizeof(u32);
1205 s -= sizeof(u32);
1206
1207 switch (type) {
1208 case C_CSESSION: { // Compressed session structure.
1209 u8 c [ sizeof(sessiont) + 2];
1210 int size;
1211 u8 * orig_p = p;
1212
1213 size = rle_decompress((u8 **) &p, s, c, sizeof(c) );
1214 s -= (p - orig_p);
1215
1216 if (size != sizeof(sessiont) ) { // Ouch! Very very bad!
1217 log(0,0,0,0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1218 // Now what? Should exit! No-longer up to date!
1219 break;
1220 }
1221
1222 cluster_recv_session(more, c);
1223 break;
1224 }
1225 case C_SESSION:
1226 if ( s < sizeof(session[more]))
1227 goto shortpacket;
1228
1229 cluster_recv_session(more, p);
1230
1231 p += sizeof(session[more]);
1232 s -= sizeof(session[more]);
1233 break;
1234
1235 case C_CTUNNEL: { // Compressed tunnel structure.
1236 u8 c [ sizeof(tunnelt) + 2];
1237 int size;
1238 u8 * orig_p = p;
1239
1240 size = rle_decompress( (u8 **) &p, s, c, sizeof(c) );
1241 s -= (p - orig_p);
1242
1243 if (size != sizeof(tunnelt) ) { // Ouch! Very very bad!
1244 log(0,0,0,0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1245 // Now what? Should exit! No-longer up to date!
1246 break;
1247 }
1248
1249 cluster_recv_tunnel(more, c);
1250 break;
1251
1252 }
1253 case C_TUNNEL:
1254 if ( s < sizeof(tunnel[more]))
1255 goto shortpacket;
1256
1257 cluster_recv_tunnel(more, p);
1258
1259 p += sizeof(tunnel[more]);
1260 s -= sizeof(tunnel[more]);
1261 break;
1262 default:
1263 log(0,0,0,0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type);
1264 return -1; // can't process any more of the packet!!
1265 }
1266 }
1267 if (config->cluster_master_address != addr)
1268 {
1269 char *str;
1270 str = strdup(inet_toa(config->cluster_master_address));
1271 log(0,0,0,0, "My master just changed from %s to %s!\n", str, inet_toa(addr));
1272 if (str) free(str);
1273 }
1274
1275 config->cluster_master_address = addr;
1276 config->cluster_last_hb = TIME; // Successfully received a heartbeat!
1277 return 0;
1278
1279 shortpacket:
1280 log(0,0,0,0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
1281 return -1;
1282 }
1283
1284 //
1285 // We got a packet on the cluster port!
1286 // Handle pings, lastseens, and heartbeats!
1287 //
1288 int processcluster(char * data, int size, u32 addr)
1289 {
1290 int type, more;
1291 char * p = data;
1292 int s = size;
1293
1294 if (addr == my_address)
1295 return -1; // Ignore it. Something looped back the multicast!
1296
1297 log(5,0,0,0, "Process cluster: %d bytes from %s\n", size, inet_toa(addr));
1298
1299 if (s <= 0) // Any data there??
1300 return -1;
1301
1302 if (s < 8)
1303 goto shortpacket;
1304
1305 type = * ((u32*) p);
1306 p += sizeof(u32);
1307 s -= sizeof(u32);
1308
1309 more = * ((u32*) p);
1310 p += sizeof(u32);
1311 s -= sizeof(u32);
1312
1313 switch (type) {
1314 case C_PING: // Update the peers table.
1315 return cluster_add_peer(addr, more, (pingt*)p, s);
1316
1317 case C_LASTSEEN: // Catch up a slave (slave missed a packet).
1318 return cluster_catchup_slave(more, addr);
1319
1320 case C_FORWARD: { // Forwarded control packet. pass off to processudp.
1321 struct sockaddr_in a;
1322 a.sin_addr.s_addr = more;
1323
1324 a.sin_port = * (int*) p;
1325 s -= sizeof(int);
1326 p += sizeof(int);
1327
1328 if (!config->cluster_iam_master) { // huh?
1329 log(0,0,0,0, "I'm not the master, but I got a C_FORWARD from %s?\n", inet_toa(addr));
1330 return -1;
1331 }
1332
1333 log(4,0,0,0, "Got a forwarded packet... (%s:%d)\n", inet_toa(more), a.sin_port);
1334 STAT(recv_forward);
1335 processudp(p, s, &a);
1336 return 0;
1337 }
1338 case C_THROTTLE: { // Receive a forwarded packet from a slave.
1339 if (!config->cluster_iam_master) {
1340 log(0,0,0,0, "I'm not the master, but I got a C_THROTTLE from %s?\n", inet_toa(addr));
1341 return -1;
1342 }
1343
1344 tbf_queue_packet(more, p, s); // The TBF id tells wether it goes in or out.
1345 return 0;
1346 }
1347 case C_GARDEN:
1348 // Receive a walled garden packet from a slave.
1349 if (!config->cluster_iam_master) {
1350 log(0,0,0,0, "I'm not the master, but I got a C_GARDEN from %s?\n", inet_toa(addr));
1351 return -1;
1352 }
1353
1354 tun_write(p, s);
1355 return 0;
1356
1357 case C_BYTES:
1358 return cluster_handle_bytes(p, s);
1359
1360 case C_KILL: // The master asked us to die!? (usually because we're too out of date).
1361 if (config->cluster_iam_master) {
1362 log(0,0,0,0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", inet_toa(addr), more);
1363 return -1;
1364 }
1365 if (more != config->cluster_seq_number) {
1366 log(0,0,0,0, "The master asked us to die but the seq number didn't match!?\n");
1367 return -1;
1368 }
1369
1370 if (addr != config->cluster_master_address) {
1371 log(0,0,0,0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%x)\n",
1372 inet_toa(addr), config->cluster_master_address);
1373 // We can only warn about it. The master might really have switched!
1374 }
1375
1376 log(0,0,0,0, "Received a valid C_KILL: I'm going to die now.\n");
1377 kill(0, SIGTERM);
1378 exit(0); // Lets be paranoid;
1379 return -1; // Just signalling the compiler.
1380
1381 case C_HEARTBEAT:
1382 log(4,0,0,0, "Got a heartbeat from %s\n", inet_toa(addr));
1383 return cluster_process_heartbeat(data, size, more, p, addr);
1384
1385 default:
1386 log(0,0,0,0, "Strange type packet received on cluster socket (%d)\n", type);
1387 return -1;
1388 }
1389 return 0;
1390
1391 shortpacket:
1392 log(0,0,0,0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n");
1393 return -1;
1394 }
1395
1396 //====================================================================================================
1397
1398 int cmd_show_cluster(struct cli_def *cli, char *command, char **argv, int argc)
1399 {
1400 int i;
1401
1402 if (CLI_HELP_REQUESTED)
1403 return CLI_HELP_NO_ARGS;
1404
1405 cli_print(cli, "Cluster status : %s", config->cluster_iam_master ? "Master" : "Slave" );
1406 cli_print(cli, "My address : %s", inet_toa(my_address));
1407 cli_print(cli, "VIP address : %s", inet_toa(config->bind_address));
1408 cli_print(cli, "Multicast address: %s", inet_toa(config->cluster_address));
1409 cli_print(cli, "Multicast i'face : %s", config->cluster_interface);
1410
1411 if (!config->cluster_iam_master) {
1412 cli_print(cli, "My master : %s (last heartbeat %.1f seconds old)",
1413 config->cluster_master_address ? inet_toa(config->cluster_master_address) : "Not defined",
1414 0.1 * (TIME - config->cluster_last_hb));
1415 cli_print(cli, "Uptodate : %s", config->cluster_iam_uptodate ? "Yes" : "No");
1416 cli_print(cli, "Next sequence number expected: %d", config->cluster_seq_number);
1417 cli_print(cli, "%d sessions undefined of %d", config->cluster_undefined_sessions, config->cluster_highest_sessionid);
1418 cli_print(cli, "%d tunnels undefined of %d", config->cluster_undefined_tunnels, config->cluster_highest_tunnelid);
1419 } else {
1420 cli_print(cli, "Next heartbeat # : %d", config->cluster_seq_number);
1421 cli_print(cli, "Highest session : %d", config->cluster_highest_sessionid);
1422 cli_print(cli, "Highest tunnel : %d", config->cluster_highest_tunnelid);
1423 cli_print(cli, "%d changes queued for sending", config->cluster_num_changes);
1424 }
1425 cli_print(cli, "%d peers.", num_peers);
1426
1427 if (num_peers)
1428 cli_print(cli, "%20s %10s %8s", "Address", "Basetime", "Age");
1429 for (i = 0; i < num_peers; ++i) {
1430 cli_print(cli, "%20s %10d %8d", inet_toa(peers[i].peer),
1431 peers[i].basetime, TIME - peers[i].timestamp);
1432 }
1433 return CLI_OK;
1434 }
1435
1436 //
1437 // Simple run-length-encoding compression.
1438 // Format is
1439 // 1 byte < 128 = count of non-zero bytes following. // Not legal to be zero.
1440 // n non-zero bytes;
1441 // or
1442 // 1 byte > 128 = (count - 128) run of zero bytes. //
1443 // repeat.
1444 // count == 0 indicates end of compressed stream.
1445 //
1446 // Compress from 'src' into 'dst'. return number of bytes
1447 // used from 'dst'.
1448 // Updates *src_p to indicate 1 past last bytes used.
1449 //
1450 // We could get an extra byte in the zero runs by storing (count-1)
1451 // but I'm playing it safe.
1452 //
1453 // Worst case is a 50% expansion in space required (trying to
1454 // compress { 0x00, 0x01 } * N )
1455 int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize)
1456 {
1457 int count;
1458 int orig_dsize = dsize;
1459 u8 * x,*src;
1460 src = *src_p;
1461
1462 while (ssize > 0 && dsize > 2) {
1463 count = 0;
1464 x = dst++; --dsize; // Reserve space for count byte..
1465
1466 if (*src) { // Copy a run of non-zero bytes.
1467 while (*src && count < 127 && ssize > 0 && dsize > 1) { // Count number of non-zero bytes.
1468 *dst++ = *src++;
1469 --dsize; --ssize;
1470 ++count;
1471 }
1472 *x = count; // Store number of non-zero bytes. Guarenteed to be non-zero!
1473
1474 } else { // Compress a run of zero bytes.
1475 while (*src == 0 && count < 127 && ssize > 0) {
1476 ++src;
1477 --ssize;
1478 ++count;
1479 }
1480 *x = count | 0x80 ;
1481 }
1482 }
1483
1484 *dst++ = 0x0; // Add Stop byte.
1485 --dsize;
1486
1487 *src_p = src;
1488 return (orig_dsize - dsize);
1489 }
1490
1491 //
1492 // Decompress the buffer into **p.
1493 // 'psize' is the size of the decompression buffer available.
1494 //
1495 // Returns the number of bytes decompressed.
1496 //
1497 // Decompresses from '*src_p' into 'dst'.
1498 // Return the number of dst bytes used.
1499 // Updates the 'src_p' pointer to point to the
1500 // first un-used byte.
1501 int rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize)
1502 {
1503 int count;
1504 int orig_dsize = dsize;
1505 char * src = *src_p;
1506
1507 while (ssize >0 && dsize > 0) { // While there's more to decompress, and there's room in the decompress buffer...
1508 count = *src++; --ssize; // get the count byte from the source.
1509 if (count == 0x0) // End marker reached? If so, finish.
1510 break;
1511
1512 if (count & 0x80) { // Decompress a run of zeros
1513 for (count &= 0x7f ; count > 0 && dsize > 0; --count) {
1514 *dst++ = 0x0;
1515 --dsize;
1516 }
1517 } else { // Copy run of non-zero bytes.
1518 for ( ; count > 0 && ssize && dsize; --count) { // Copy non-zero bytes across.
1519 *dst++ = *src++;
1520 --ssize; --dsize;
1521 }
1522 }
1523 }
1524 *src_p = src;
1525 return (orig_dsize - dsize);
1526 }