1 // L2TPNS Clustering Stuff
3 char const *cvs_id_cluster
= "$Id: cluster.c,v 1.51 2006/04/27 09:53:49 bodea Exp $";
12 #include <sys/socket.h>
13 #include <netinet/in.h>
14 #include <arpa/inet.h>
15 #include <sys/ioctl.h>
31 * All cluster packets have the same format.
33 * One or more instances of
35 * a 32 bit 'extra' data dependant on the 'type'.
36 * zero or more bytes of structure data, dependant on the type.
41 extern int cluster_sockfd
; // The filedescriptor for the cluster communications port.
43 in_addr_t my_address
= 0; // The network address of my ethernet port.
44 static int walk_session_number
= 0; // The next session to send when doing the slow table walk.
45 static int walk_bundle_number
= 0; // The next bundle to send when doing the slow table walk.
46 static int walk_tunnel_number
= 0; // The next tunnel to send when doing the slow table walk.
47 int forked
= 0; // Sanity check: CLI must not diddle with heartbeat table
49 #define MAX_HEART_SIZE (8192) // Maximum size of heartbeat packet. Must be less than max IP packet size :)
50 #define MAX_CHANGES (MAX_HEART_SIZE/(sizeof(sessiont) + sizeof(int) ) - 2) // Assumes a session is the biggest type!
55 } cluster_changes
[MAX_CHANGES
]; // Queue of changed structures that need to go out when next heartbeat.
60 uint8_t data
[MAX_HEART_SIZE
];
61 } past_hearts
[HB_HISTORY_SIZE
]; // Ring buffer of heartbeats that we've recently sent out. Needed so
62 // we can re-transmit if needed.
69 } peers
[CLUSTER_MAX_SIZE
]; // List of all the peers we've heard from.
70 static int num_peers
; // Number of peers in list.
72 static int rle_decompress(uint8_t **src_p
, int ssize
, uint8_t *dst
, int dsize
);
73 static int rle_compress(uint8_t **src_p
, int ssize
, uint8_t *dst
, int dsize
);
76 // Create a listening socket
78 // This joins the cluster multi-cast group.
82 struct sockaddr_in addr
;
83 struct sockaddr_in interface_addr
;
88 config
->cluster_undefined_sessions
= MAXSESSION
-1;
89 config
->cluster_undefined_bundles
= MAXBUNDLE
-1;
90 config
->cluster_undefined_tunnels
= MAXTUNNEL
-1;
92 if (!config
->cluster_address
)
94 if (!*config
->cluster_interface
)
97 cluster_sockfd
= socket(AF_INET
, SOCK_DGRAM
, IPPROTO_UDP
);
99 memset(&addr
, 0, sizeof(addr
));
100 addr
.sin_family
= AF_INET
;
101 addr
.sin_port
= htons(CLUSTERPORT
);
102 addr
.sin_addr
.s_addr
= INADDR_ANY
;
103 setsockopt(cluster_sockfd
, SOL_SOCKET
, SO_REUSEADDR
, &addr
, sizeof(addr
));
105 opt
= fcntl(cluster_sockfd
, F_GETFL
, 0);
106 fcntl(cluster_sockfd
, F_SETFL
, opt
| O_NONBLOCK
);
108 if (bind(cluster_sockfd
, (void *) &addr
, sizeof(addr
)) < 0)
110 LOG(0, 0, 0, "Failed to bind cluster socket: %s\n", strerror(errno
));
114 strcpy(ifr
.ifr_name
, config
->cluster_interface
);
115 if (ioctl(cluster_sockfd
, SIOCGIFADDR
, &ifr
) < 0)
117 LOG(0, 0, 0, "Failed to get interface address for (%s): %s\n", config
->cluster_interface
, strerror(errno
));
121 memcpy(&interface_addr
, &ifr
.ifr_addr
, sizeof(interface_addr
));
122 my_address
= interface_addr
.sin_addr
.s_addr
;
124 // Join multicast group.
125 mreq
.imr_multiaddr
.s_addr
= config
->cluster_address
;
126 mreq
.imr_interface
= interface_addr
.sin_addr
;
129 opt
= 0; // Turn off multicast loopback.
130 setsockopt(cluster_sockfd
, IPPROTO_IP
, IP_MULTICAST_LOOP
, &opt
, sizeof(opt
));
132 if (config
->cluster_mcast_ttl
!= 1)
135 if (config
->cluster_mcast_ttl
> 0)
136 ttl
= config
->cluster_mcast_ttl
< 256 ? config
->cluster_mcast_ttl
: 255;
138 setsockopt(cluster_sockfd
, IPPROTO_IP
, IP_MULTICAST_TTL
, &ttl
, sizeof(ttl
));
141 if (setsockopt(cluster_sockfd
, IPPROTO_IP
, IP_ADD_MEMBERSHIP
, &mreq
, sizeof(mreq
)) < 0)
143 LOG(0, 0, 0, "Failed to setsockopt (join mcast group): %s\n", strerror(errno
));
147 if (setsockopt(cluster_sockfd
, IPPROTO_IP
, IP_MULTICAST_IF
, &interface_addr
, sizeof(interface_addr
)) < 0)
149 LOG(0, 0, 0, "Failed to setsockopt (set mcast interface): %s\n", strerror(errno
));
153 config
->cluster_last_hb
= TIME
;
154 config
->cluster_seq_number
= -1;
156 return cluster_sockfd
;
161 // Send a chunk of data to the entire cluster (usually via the multicast
165 static int cluster_send_data(void *data
, int datalen
)
167 struct sockaddr_in addr
= {0};
169 if (!cluster_sockfd
) return -1;
170 if (!config
->cluster_address
) return 0;
172 addr
.sin_addr
.s_addr
= config
->cluster_address
;
173 addr
.sin_port
= htons(CLUSTERPORT
);
174 addr
.sin_family
= AF_INET
;
176 LOG(5, 0, 0, "Cluster send data: %d bytes\n", datalen
);
178 if (sendto(cluster_sockfd
, data
, datalen
, MSG_NOSIGNAL
, (void *) &addr
, sizeof(addr
)) < 0)
180 LOG(0, 0, 0, "sendto: %s\n", strerror(errno
));
188 // Add a chunk of data to a heartbeat packet.
189 // Maintains the format. Assumes that the caller
190 // has passed in a big enough buffer!
192 static void add_type(uint8_t **p
, int type
, int more
, uint8_t *data
, int size
)
194 *((uint32_t *) (*p
)) = type
;
195 *p
+= sizeof(uint32_t);
197 *((uint32_t *)(*p
)) = more
;
198 *p
+= sizeof(uint32_t);
200 if (data
&& size
> 0) {
201 memcpy(*p
, data
, size
);
206 // advertise our presence via BGP or gratuitous ARP
207 static void advertise_routes(void)
211 bgp_enable_routing(1);
214 if (config
->send_garp
)
215 send_garp(config
->bind_address
); // Start taking traffic.
218 // withdraw our routes (BGP only)
219 static void withdraw_routes(void)
223 bgp_enable_routing(0);
227 static void cluster_uptodate(void)
229 if (config
->cluster_iam_uptodate
)
232 if (config
->cluster_undefined_sessions
|| config
->cluster_undefined_tunnels
|| config
->cluster_undefined_bundles
)
235 config
->cluster_iam_uptodate
= 1;
237 LOG(0, 0, 0, "Now uptodate with master.\n");
242 // Send a unicast UDP packet to a peer with 'data' as the
245 static int peer_send_data(in_addr_t peer
, uint8_t *data
, int size
)
247 struct sockaddr_in addr
= {0};
249 if (!cluster_sockfd
) return -1;
250 if (!config
->cluster_address
) return 0;
255 addr
.sin_addr
.s_addr
= peer
;
256 addr
.sin_port
= htons(CLUSTERPORT
);
257 addr
.sin_family
= AF_INET
;
259 LOG_HEX(5, "Peer send", data
, size
);
261 if (sendto(cluster_sockfd
, data
, size
, MSG_NOSIGNAL
, (void *) &addr
, sizeof(addr
)) < 0)
263 LOG(0, 0, 0, "sendto: %s\n", strerror(errno
));
271 // Send a structured message to a peer with a single element of type 'type'.
273 static int peer_send_message(in_addr_t peer
, int type
, int more
, uint8_t *data
, int size
)
275 uint8_t buf
[65536]; // Vast overkill.
278 LOG(4, 0, 0, "Sending message to peer (type %d, more %d, size %d)\n", type
, more
, size
);
279 add_type(&p
, type
, more
, data
, size
);
281 return peer_send_data(peer
, buf
, (p
-buf
) );
284 // send a packet to the master
285 static int _forward_packet(uint8_t *data
, int size
, in_addr_t addr
, int port
, int type
)
287 uint8_t buf
[65536]; // Vast overkill.
290 if (!config
->cluster_master_address
) // No election has been held yet. Just skip it.
293 LOG(4, 0, 0, "Forwarding packet from %s to master (size %d)\n", fmtaddr(addr
, 0), size
);
296 add_type(&p
, type
, addr
, (uint8_t *) &port
, sizeof(port
)); // ick. should be uint16_t
297 memcpy(p
, data
, size
);
300 return peer_send_data(config
->cluster_master_address
, buf
, (p
- buf
));
304 // Forward a state changing packet to the master.
306 // The master just processes the payload as if it had
307 // received it off the tun device.
309 int master_forward_packet(uint8_t *data
, int size
, in_addr_t addr
, int port
)
311 return _forward_packet(data
, size
, addr
, port
, C_FORWARD
);
314 // Forward a DAE RADIUS packet to the master.
315 int master_forward_dae_packet(uint8_t *data
, int size
, in_addr_t addr
, int port
)
317 return _forward_packet(data
, size
, addr
, port
, C_FORWARD_DAE
);
321 // Forward a throttled packet to the master for handling.
323 // The master just drops the packet into the appropriate
324 // token bucket queue, and lets normal processing take care
327 int master_throttle_packet(int tbfid
, uint8_t *data
, int size
)
329 uint8_t buf
[65536]; // Vast overkill.
332 if (!config
->cluster_master_address
) // No election has been held yet. Just skip it.
335 LOG(4, 0, 0, "Throttling packet master (size %d, tbfid %d)\n", size
, tbfid
);
337 add_type(&p
, C_THROTTLE
, tbfid
, data
, size
);
339 return peer_send_data(config
->cluster_master_address
, buf
, (p
-buf
) );
344 // Forward a walled garden packet to the master for handling.
346 // The master just writes the packet straight to the tun
347 // device (where is will normally loop through the
348 // firewall rules, and come back in on the tun device)
350 // (Note that this must be called with the tun header
351 // as the start of the data).
352 int master_garden_packet(sessionidt s
, uint8_t *data
, int size
)
354 uint8_t buf
[65536]; // Vast overkill.
357 if (!config
->cluster_master_address
) // No election has been held yet. Just skip it.
360 LOG(4, 0, 0, "Walled garden packet to master (size %d)\n", size
);
362 add_type(&p
, C_GARDEN
, s
, data
, size
);
364 return peer_send_data(config
->cluster_master_address
, buf
, (p
-buf
));
369 // Send a chunk of data as a heartbeat..
370 // We save it in the history buffer as we do so.
372 static void send_heartbeat(int seq
, uint8_t *data
, int size
)
376 if (size
> sizeof(past_hearts
[0].data
))
378 LOG(0, 0, 0, "Tried to heartbeat something larger than the maximum packet!\n");
382 i
= seq
% HB_HISTORY_SIZE
;
383 past_hearts
[i
].seq
= seq
;
384 past_hearts
[i
].size
= size
;
385 memcpy(&past_hearts
[i
].data
, data
, size
); // Save it.
386 cluster_send_data(data
, size
);
390 // Send an 'i am alive' message to every machine in the cluster.
392 void cluster_send_ping(time_t basetime
)
394 uint8_t buff
[100 + sizeof(pingt
)];
398 if (config
->cluster_iam_master
&& basetime
) // We're heartbeating so no need to ping.
401 LOG(5, 0, 0, "Sending cluster ping...\n");
404 x
.addr
= config
->bind_address
;
405 x
.undef
= config
->cluster_undefined_sessions
+ config
->cluster_undefined_tunnels
;
406 x
.basetime
= basetime
;
408 add_type(&p
, C_PING
, basetime
, (uint8_t *) &x
, sizeof(x
));
409 cluster_send_data(buff
, (p
-buff
) );
413 // Walk the session counters looking for non-zero ones to send
414 // to the master. We send up to 600 of them at one time.
415 // We examine a maximum of 3000 sessions.
416 // (50k max session should mean that we normally
417 // examine the entire session table every 25 seconds).
419 #define MAX_B_RECS (600)
420 void master_update_counts(void)
423 bytest b
[MAX_B_RECS
+1];
425 if (config
->cluster_iam_master
) // Only happens on the slaves.
428 if (!config
->cluster_master_address
) // If we don't have a master, skip it for a while.
431 // C_BYTES format changed in 2.1.0 (cluster version 5)
432 // during upgrade from previous versions, hang onto our counters
433 // for a bit until the new master comes up
434 if (config
->cluster_last_hb_ver
< 5)
437 i
= MAX_B_RECS
* 5; // Examine max 3000 sessions;
438 if (config
->cluster_highest_sessionid
> i
)
439 i
= config
->cluster_highest_sessionid
;
441 for ( c
= 0; i
> 0 ; --i
) {
442 // Next session to look at.
443 walk_session_number
++;
444 if ( walk_session_number
> config
->cluster_highest_sessionid
)
445 walk_session_number
= 1;
447 if (!sess_local
[walk_session_number
].cin
&& !sess_local
[walk_session_number
].cout
)
448 continue; // Unchanged. Skip it.
450 b
[c
].sid
= walk_session_number
;
451 b
[c
].pin
= sess_local
[walk_session_number
].pin
;
452 b
[c
].pout
= sess_local
[walk_session_number
].pout
;
453 b
[c
].cin
= sess_local
[walk_session_number
].cin
;
454 b
[c
].cout
= sess_local
[walk_session_number
].cout
;
457 sess_local
[walk_session_number
].pin
= sess_local
[walk_session_number
].pout
= 0;
458 sess_local
[walk_session_number
].cin
= sess_local
[walk_session_number
].cout
= 0;
460 if (++c
> MAX_B_RECS
) // Send a max of 600 elements in a packet.
464 if (!c
) // Didn't find any that changes. Get out of here!
468 // Forward the data to the master.
469 LOG(4, 0, 0, "Sending byte counters to master (%d elements)\n", c
);
470 peer_send_message(config
->cluster_master_address
, C_BYTES
, c
, (uint8_t *) &b
, sizeof(b
[0]) * c
);
475 // On the master, check how our slaves are going. If
476 // one of them's not up-to-date we'll heartbeat faster.
477 // If we don't have any of them, then we need to turn
478 // on our own packet handling!
480 void cluster_check_slaves(void)
483 static int have_peers
= 0;
484 int had_peers
= have_peers
;
487 if (!config
->cluster_iam_master
)
488 return; // Only runs on the master...
490 config
->cluster_iam_uptodate
= 1; // cleared in loop below
492 for (i
= have_peers
= 0; i
< num_peers
; i
++)
494 if ((peers
[i
].timestamp
+ config
->cluster_hb_timeout
) < t
)
495 continue; // Stale peer! Skip them.
497 if (!peers
[i
].basetime
)
498 continue; // Shutdown peer! Skip them.
500 if (peers
[i
].uptodate
)
503 config
->cluster_iam_uptodate
= 0; // Start fast heartbeats
506 // in a cluster, withdraw/add routes when we get a peer/lose peers
507 if (have_peers
!= had_peers
)
509 if (had_peers
< config
->cluster_master_min_adv
&&
510 have_peers
>= config
->cluster_master_min_adv
)
513 else if (had_peers
>= config
->cluster_master_min_adv
&&
514 have_peers
< config
->cluster_master_min_adv
)
520 // Check that we have a master. If it's been too
521 // long since we heard from a master then hold an election.
523 void cluster_check_master(void)
525 int i
, count
, tcount
, bcount
, high_unique_id
= 0;
528 static int probed
= 0;
531 if (config
->cluster_iam_master
)
532 return; // Only runs on the slaves...
534 // If the master is late (missed 2 hearbeats by a second and a
535 // hair) it may be that the switch has dropped us from the
536 // multicast group, try unicasting probes to the master
537 // which will hopefully respond with a unicast heartbeat that
538 // will allow us to limp along until the querier next runs.
539 if (config
->cluster_master_address
540 && TIME
> (config
->cluster_last_hb
+ 2 * config
->cluster_hb_interval
+ 11))
542 if (!probed
|| (TIME
> (probed
+ 2 * config
->cluster_hb_interval
)))
545 LOG(1, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
546 0.1 * (TIME
- (config
->cluster_last_hb
+ config
->cluster_hb_interval
)));
548 peer_send_message(config
->cluster_master_address
,
549 C_LASTSEEN
, config
->cluster_seq_number
, NULL
, 0);
551 } else { // We got a recent heartbeat; reset the probe flag.
555 if (TIME
< (config
->cluster_last_hb
+ config
->cluster_hb_timeout
))
556 return; // Everything's ok!
558 config
->cluster_last_hb
= TIME
+ 1; // Just the one election thanks.
559 config
->cluster_master_address
= 0;
561 LOG(0, 0, 0, "Master timed out! Holding election...\n");
563 // In the process of shutting down, can't be master
567 for (i
= have_peers
= 0; i
< num_peers
; i
++)
569 if ((peers
[i
].timestamp
+ config
->cluster_hb_timeout
) < t
)
570 continue; // Stale peer! Skip them.
572 if (!peers
[i
].basetime
)
573 continue; // Shutdown peer! Skip them.
575 if (peers
[i
].basetime
< basetime
) {
576 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers
[i
].peer
, 0));
577 return; // They'll win the election. Get out of here.
580 if (peers
[i
].basetime
== basetime
&&
581 peers
[i
].peer
> my_address
) {
582 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers
[i
].peer
, 0));
583 return; // They'll win the election. Wait for them to come up.
586 if (peers
[i
].uptodate
)
590 // Wow. it's been ages since I last heard a heartbeat
591 // and I'm better than an of my peers so it's time
592 // to become a master!!!
594 config
->cluster_iam_master
= 1;
596 LOG(0, 0, 0, "I am declaring myself the master!\n");
598 if (have_peers
< config
->cluster_master_min_adv
)
603 if (config
->cluster_seq_number
== -1)
604 config
->cluster_seq_number
= 0;
607 // Go through and mark all the tunnels as defined.
608 // Count the highest used tunnel number as well.
610 config
->cluster_highest_tunnelid
= 0;
611 for (i
= 0, tcount
= 0; i
< MAXTUNNEL
; ++i
) {
612 if (tunnel
[i
].state
== TUNNELUNDEF
)
613 tunnel
[i
].state
= TUNNELFREE
;
615 if (tunnel
[i
].state
!= TUNNELFREE
&& i
> config
->cluster_highest_tunnelid
)
616 config
->cluster_highest_tunnelid
= i
;
620 // Go through and mark all the bundles as defined.
621 // Count the highest used bundle number as well.
623 config
->cluster_highest_bundleid
= 0;
624 for (i
= 0, bcount
= 0; i
< MAXBUNDLE
; ++i
) {
625 if (bundle
[i
].state
== BUNDLEUNDEF
)
626 bundle
[i
].state
= BUNDLEFREE
;
628 if (bundle
[i
].state
!= BUNDLEFREE
&& i
> config
->cluster_highest_bundleid
)
629 config
->cluster_highest_bundleid
= i
;
633 // Go through and mark all the sessions as being defined.
634 // reset the idle timeouts.
635 // add temporary byte counters to permanent ones.
636 // Re-string the free list.
637 // Find the ID of the highest session.
640 config
->cluster_highest_sessionid
= 0;
641 for (i
= 0, count
= 0; i
< MAXSESSION
; ++i
) {
642 if (session
[i
].tunnel
== T_UNDEF
) {
643 session
[i
].tunnel
= T_FREE
;
647 if (!session
[i
].opened
) { // Unused session. Add to free list.
648 memset(&session
[i
], 0, sizeof(session
[i
]));
649 session
[i
].tunnel
= T_FREE
;
650 session
[last_free
].next
= i
;
656 // Reset idle timeouts..
657 session
[i
].last_packet
= time_now
;
659 // Reset die relative to our uptime rather than the old master's
660 if (session
[i
].die
) session
[i
].die
= TIME
;
662 // Accumulate un-sent byte/packet counters.
663 increment_counter(&session
[i
].cin
, &session
[i
].cin_wrap
, sess_local
[i
].cin
);
664 increment_counter(&session
[i
].cout
, &session
[i
].cout_wrap
, sess_local
[i
].cout
);
665 session
[i
].cin_delta
+= sess_local
[i
].cin
;
666 session
[i
].cout_delta
+= sess_local
[i
].cout
;
668 session
[i
].pin
+= sess_local
[i
].pin
;
669 session
[i
].pout
+= sess_local
[i
].pout
;
671 sess_local
[i
].cin
= sess_local
[i
].cout
= 0;
672 sess_local
[i
].pin
= sess_local
[i
].pout
= 0;
674 sess_local
[i
].radius
= 0; // Reset authentication as the radius blocks aren't up to date.
676 if (session
[i
].unique_id
>= high_unique_id
) // This is different to the index into the session table!!!
677 high_unique_id
= session
[i
].unique_id
+1;
679 session
[i
].tbf_in
= session
[i
].tbf_out
= 0; // Remove stale pointers from old master.
680 throttle_session(i
, session
[i
].throttle_in
, session
[i
].throttle_out
);
682 config
->cluster_highest_sessionid
= i
;
685 session
[last_free
].next
= 0; // End of chain.
686 last_id
= high_unique_id
; // Keep track of the highest used session ID.
690 rebuild_address_pool();
692 // If we're not the very first master, this is a big issue!
694 LOG(0, 0, 0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count
);
696 config
->cluster_undefined_sessions
= 0;
697 config
->cluster_undefined_bundles
= 0;
698 config
->cluster_undefined_tunnels
= 0;
699 config
->cluster_iam_uptodate
= 1; // assume all peers are up-to-date
701 // FIXME. We need to fix up the tunnel control message
702 // queue here! There's a number of other variables we
703 // should also update.
708 // Check that our session table is validly matching what the
709 // master has in mind.
711 // In particular, if we have too many sessions marked 'undefined'
712 // we fix it up here, and we ensure that the 'first free session'
715 static void cluster_check_sessions(int highsession
, int freesession_ptr
, int highbundle
, int hightunnel
)
719 sessionfree
= freesession_ptr
; // Keep the freesession ptr valid.
721 if (config
->cluster_iam_uptodate
)
724 if (highsession
> config
->cluster_undefined_sessions
&& highbundle
> config
->cluster_undefined_bundles
&& hightunnel
> config
->cluster_undefined_tunnels
)
727 // Clear out defined sessions, counting the number of
729 config
->cluster_undefined_sessions
= 0;
730 for (i
= 1 ; i
< MAXSESSION
; ++i
) {
731 if (i
> highsession
) {
732 if (session
[i
].tunnel
== T_UNDEF
) session
[i
].tunnel
= T_FREE
; // Defined.
736 if (session
[i
].tunnel
== T_UNDEF
)
737 ++config
->cluster_undefined_sessions
;
740 // Clear out defined bundles, counting the number of
742 config
->cluster_undefined_bundles
= 0;
743 for (i
= 1 ; i
< MAXBUNDLE
; ++i
) {
744 if (i
> highbundle
) {
745 if (bundle
[i
].state
== BUNDLEUNDEF
) bundle
[i
].state
= BUNDLEFREE
; // Defined.
749 if (bundle
[i
].state
== BUNDLEUNDEF
)
750 ++config
->cluster_undefined_bundles
;
753 // Clear out defined tunnels, counting the number of
755 config
->cluster_undefined_tunnels
= 0;
756 for (i
= 1 ; i
< MAXTUNNEL
; ++i
) {
757 if (i
> hightunnel
) {
758 if (tunnel
[i
].state
== TUNNELUNDEF
) tunnel
[i
].state
= TUNNELFREE
; // Defined.
762 if (tunnel
[i
].state
== TUNNELUNDEF
)
763 ++config
->cluster_undefined_tunnels
;
767 if (config
->cluster_undefined_sessions
|| config
->cluster_undefined_tunnels
|| config
->cluster_undefined_bundles
) {
768 LOG(2, 0, 0, "Cleared undefined sessions/bundles/tunnels. %d sess (high %d), %d bund (high %d), %d tunn (high %d)\n",
769 config
->cluster_undefined_sessions
, highsession
, config
->cluster_undefined_bundles
, highbundle
, config
->cluster_undefined_tunnels
, hightunnel
);
773 // Are we up to date?
775 if (!config
->cluster_iam_uptodate
)
779 static int hb_add_type(uint8_t **p
, int type
, int id
)
782 case C_CSESSION
: { // Compressed C_SESSION.
783 uint8_t c
[sizeof(sessiont
) * 2]; // Bigger than worst case.
784 uint8_t *d
= (uint8_t *) &session
[id
];
788 size
= rle_compress( &d
, sizeof(sessiont
), c
, sizeof(c
) );
790 // Did we compress the full structure, and is the size actually
792 if ( (d
- orig
) == sizeof(sessiont
) && size
< sizeof(sessiont
) ) {
793 add_type(p
, C_CSESSION
, id
, c
, size
);
796 // Failed to compress : Fall through.
799 add_type(p
, C_SESSION
, id
, (uint8_t *) &session
[id
], sizeof(sessiont
));
802 case C_CBUNDLE
: { // Compressed C_BUNDLE
803 uint8_t c
[sizeof(bundlet
) * 2]; // Bigger than worst case.
804 uint8_t *d
= (uint8_t *) &bundle
[id
];
808 size
= rle_compress( &d
, sizeof(bundlet
), c
, sizeof(c
) );
810 // Did we compress the full structure, and is the size actually
812 if ( (d
- orig
) == sizeof(bundlet
) && size
< sizeof(bundlet
) ) {
813 add_type(p
, C_CBUNDLE
, id
, c
, size
);
816 // Failed to compress : Fall through.
820 add_type(p
, C_BUNDLE
, id
, (uint8_t *) &bundle
[id
], sizeof(bundlet
));
823 case C_CTUNNEL
: { // Compressed C_TUNNEL
824 uint8_t c
[sizeof(tunnelt
) * 2]; // Bigger than worst case.
825 uint8_t *d
= (uint8_t *) &tunnel
[id
];
829 size
= rle_compress( &d
, sizeof(tunnelt
), c
, sizeof(c
) );
831 // Did we compress the full structure, and is the size actually
833 if ( (d
- orig
) == sizeof(tunnelt
) && size
< sizeof(tunnelt
) ) {
834 add_type(p
, C_CTUNNEL
, id
, c
, size
);
837 // Failed to compress : Fall through.
840 add_type(p
, C_TUNNEL
, id
, (uint8_t *) &tunnel
[id
], sizeof(tunnelt
));
843 LOG(0, 0, 0, "Found an invalid type in heart queue! (%d)\n", type
);
851 // Send a heartbeat, incidently sending out any queued changes..
853 void cluster_heartbeat()
855 int i
, count
= 0, tcount
= 0, bcount
= 0;
856 uint8_t buff
[MAX_HEART_SIZE
+ sizeof(heartt
) + sizeof(int) ];
860 if (!config
->cluster_iam_master
) // Only the master does this.
863 config
->cluster_table_version
+= config
->cluster_num_changes
;
865 // Fill out the heartbeat header.
866 memset(&h
, 0, sizeof(h
));
868 h
.version
= HB_VERSION
;
869 h
.seq
= config
->cluster_seq_number
;
870 h
.basetime
= basetime
;
871 h
.clusterid
= config
->bind_address
; // Will this do??
872 h
.basetime
= basetime
;
873 h
.highsession
= config
->cluster_highest_sessionid
;
874 h
.freesession
= sessionfree
;
875 h
.hightunnel
= config
->cluster_highest_tunnelid
;
876 h
.highbundle
= config
->cluster_highest_bundleid
;
877 h
.size_sess
= sizeof(sessiont
); // Just in case.
878 h
.size_bund
= sizeof(bundlet
);
879 h
.size_tunn
= sizeof(tunnelt
);
880 h
.interval
= config
->cluster_hb_interval
;
881 h
.timeout
= config
->cluster_hb_timeout
;
882 h
.table_version
= config
->cluster_table_version
;
884 add_type(&p
, C_HEARTBEAT
, HB_VERSION
, (uint8_t *) &h
, sizeof(h
));
886 for (i
= 0; i
< config
->cluster_num_changes
; ++i
) {
887 hb_add_type(&p
, cluster_changes
[i
].type
, cluster_changes
[i
].id
);
890 if (p
> (buff
+ sizeof(buff
))) { // Did we somehow manage to overun the buffer?
891 LOG(0, 0, 0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", (int) (p
- buff
));
897 // Fill out the packet with sessions from the session table...
898 // (not forgetting to leave space so we can get some tunnels in too )
899 while ( (p
+ sizeof(uint32_t) * 2 + sizeof(sessiont
) * 2 ) < (buff
+ MAX_HEART_SIZE
) ) {
901 if (!walk_session_number
) // session #0 isn't valid.
902 ++walk_session_number
;
904 if (count
>= config
->cluster_highest_sessionid
) // If we're a small cluster, don't go wild.
907 hb_add_type(&p
, C_CSESSION
, walk_session_number
);
908 walk_session_number
= (1+walk_session_number
)%(config
->cluster_highest_sessionid
+1); // +1 avoids divide by zero.
910 ++count
; // Count the number of extra sessions we're sending.
914 // Fill out the packet with tunnels from the tunnel table...
915 // This effectively means we walk the tunnel table more quickly
916 // than the session table. This is good because stuffing up a
917 // tunnel is a much bigger deal than stuffing up a session.
919 while ( (p
+ sizeof(uint32_t) * 2 + sizeof(tunnelt
) ) < (buff
+ MAX_HEART_SIZE
) ) {
921 if (!walk_tunnel_number
) // tunnel #0 isn't valid.
922 ++walk_tunnel_number
;
924 if (tcount
>= config
->cluster_highest_tunnelid
)
927 hb_add_type(&p
, C_CTUNNEL
, walk_tunnel_number
);
928 walk_tunnel_number
= (1+walk_tunnel_number
)%(config
->cluster_highest_tunnelid
+1); // +1 avoids divide by zero.
934 // Fill out the packet with bundles from the bundle table...
935 while ( (p
+ sizeof(uint32_t) * 2 + sizeof(bundlet
) ) < (buff
+ MAX_HEART_SIZE
) ) {
937 if (!walk_bundle_number
) // bundle #0 isn't valid.
938 ++walk_bundle_number
;
940 if (bcount
>= config
->cluster_highest_bundleid
)
943 hb_add_type(&p
, C_CTUNNEL
, walk_bundle_number
);
944 walk_tunnel_number
= (1+walk_bundle_number
)%(config
->cluster_highest_bundleid
+1); // +1 avoids divide by zero.
950 // Did we do something wrong?
951 if (p
> (buff
+ sizeof(buff
))) { // Did we somehow manage to overun the buffer?
952 LOG(0, 0, 0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", (int) (p
- buff
));
957 LOG(3, 0, 0, "Sending v%d heartbeat #%d, change #%" PRIu64
" with %d changes "
958 "(%d x-sess, %d x-bundles, %d x-tunnels, %d highsess, %d highbund, %d hightun, size %d)\n",
959 HB_VERSION
, h
.seq
, h
.table_version
, config
->cluster_num_changes
,
960 count
, bcount
, tcount
, config
->cluster_highest_sessionid
, config
->cluster_highest_bundleid
,
961 config
->cluster_highest_tunnelid
, (int) (p
- buff
));
963 config
->cluster_num_changes
= 0;
965 send_heartbeat(h
.seq
, buff
, (p
-buff
) ); // Send out the heartbeat to the cluster, keeping a copy of it.
967 config
->cluster_seq_number
= (config
->cluster_seq_number
+1)%HB_MAX_SEQ
; // Next seq number to use.
971 // A structure of type 'type' has changed; Add it to the queue to send.
973 static int type_changed(int type
, int id
)
977 for (i
= 0 ; i
< config
->cluster_num_changes
; ++i
)
978 if ( cluster_changes
[i
].id
== id
&&
979 cluster_changes
[i
].type
== type
)
980 return 0; // Already marked for change.
982 cluster_changes
[i
].type
= type
;
983 cluster_changes
[i
].id
= id
;
984 ++config
->cluster_num_changes
;
986 if (config
->cluster_num_changes
> MAX_CHANGES
)
987 cluster_heartbeat(); // flush now
993 // A particular session has been changed!
994 int cluster_send_session(int sid
)
996 if (!config
->cluster_iam_master
) {
997 LOG(0, sid
, 0, "I'm not a master, but I just tried to change a session!\n");
1002 LOG(0, sid
, 0, "cluster_send_session called from child process!\n");
1006 return type_changed(C_CSESSION
, sid
);
1009 // A particular bundle has been changed!
1010 int cluster_send_bundle(int bid
)
1012 if (!config
->cluster_iam_master
) {
1013 LOG(0, 0, bid
, "I'm not a master, but I just tried to change a bundle!\n");
1017 return type_changed(C_CBUNDLE
, bid
);
1020 // A particular tunnel has been changed!
1021 int cluster_send_tunnel(int tid
)
1023 if (!config
->cluster_iam_master
) {
1024 LOG(0, 0, tid
, "I'm not a master, but I just tried to change a tunnel!\n");
1028 return type_changed(C_CTUNNEL
, tid
);
1033 // We're a master, and a slave has just told us that it's
1034 // missed a packet. We'll resend it every packet since
1035 // the last one it's seen.
1037 static int cluster_catchup_slave(int seq
, in_addr_t slave
)
1042 LOG(1, 0, 0, "Slave %s sent LASTSEEN with seq %d\n", fmtaddr(slave
, 0), seq
);
1043 if (!config
->cluster_iam_master
) {
1044 LOG(1, 0, 0, "Got LASTSEEN but I'm not a master! Redirecting it to %s.\n",
1045 fmtaddr(config
->cluster_master_address
, 0));
1047 peer_send_message(slave
, C_MASTER
, config
->cluster_master_address
, NULL
, 0);
1051 diff
= config
->cluster_seq_number
- seq
; // How many packet do we need to send?
1055 if (diff
>= HB_HISTORY_SIZE
) { // Ouch. We don't have the packet to send it!
1056 LOG(0, 0, 0, "A slave asked for message %d when our seq number is %d. Killing it.\n",
1057 seq
, config
->cluster_seq_number
);
1058 return peer_send_message(slave
, C_KILL
, seq
, NULL
, 0);// Kill the slave. Nothing else to do.
1061 LOG(1, 0, 0, "Sending %d catchup packets to slave %s\n", diff
, fmtaddr(slave
, 0) );
1063 // Now resend every packet that it missed, in order.
1064 while (seq
!= config
->cluster_seq_number
) {
1065 s
= seq
% HB_HISTORY_SIZE
;
1066 if (seq
!= past_hearts
[s
].seq
) {
1067 LOG(0, 0, 0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n",
1068 fmtaddr(slave
, 0), seq
, past_hearts
[s
].seq
, s
, config
->cluster_seq_number
);
1069 return -1; // What to do here!?
1071 peer_send_data(slave
, past_hearts
[s
].data
, past_hearts
[s
].size
);
1072 seq
= (seq
+1)%HB_MAX_SEQ
; // Increment to next seq number.
1074 return 0; // All good!
1078 // We've heard from another peer! Add it to the list
1079 // that we select from at election time.
1081 static int cluster_add_peer(in_addr_t peer
, time_t basetime
, pingt
*pp
, int size
)
1084 in_addr_t clusterid
;
1087 // Allow for backward compatability.
1088 // Just the ping packet into a new structure to allow
1089 // for the possibility that we might have received
1090 // more or fewer elements than we were expecting.
1091 if (size
> sizeof(p
))
1094 memset( (void *) &p
, 0, sizeof(p
) );
1095 memcpy( (void *) &p
, (void *) pp
, size
);
1098 if (clusterid
!= config
->bind_address
)
1101 LOG(4, 0, 0, "Skipping ping from %s (different cluster)\n", fmtaddr(peer
, 0));
1105 for (i
= 0; i
< num_peers
; ++i
)
1107 if (peers
[i
].peer
!= peer
)
1110 // This peer already exists. Just update the timestamp.
1111 peers
[i
].basetime
= basetime
;
1112 peers
[i
].timestamp
= TIME
;
1113 peers
[i
].uptodate
= !p
.undef
;
1117 // Is this the master shutting down??
1118 if (peer
== config
->cluster_master_address
) {
1119 LOG(3, 0, 0, "Master %s %s\n", fmtaddr(config
->cluster_master_address
, 0),
1120 basetime
? "has restarted!" : "shutting down...");
1122 config
->cluster_master_address
= 0;
1123 config
->cluster_last_hb
= 0; // Force an election.
1124 cluster_check_master();
1129 LOG(4, 0, 0, "Adding %s as a peer\n", fmtaddr(peer
, 0));
1131 // Not found. Is there a stale slot to re-use?
1132 for (i
= 0; i
< num_peers
; ++i
)
1134 if (!peers
[i
].basetime
) // Shutdown
1137 if ((peers
[i
].timestamp
+ config
->cluster_hb_timeout
* 10) < TIME
) // Stale.
1141 if (i
>= CLUSTER_MAX_SIZE
)
1144 LOG(0, 0, 0, "Tried to add %s as a peer, but I already have %d of them!\n", fmtaddr(peer
, 0), i
);
1148 peers
[i
].peer
= peer
;
1149 peers
[i
].basetime
= basetime
;
1150 peers
[i
].timestamp
= TIME
;
1151 peers
[i
].uptodate
= !p
.undef
;
1155 LOG(1, 0, 0, "Added %s as a new peer. Now %d peers\n", fmtaddr(peer
, 0), num_peers
);
1161 // A slave responds with C_MASTER when it gets a message which should have gone to a master.
1162 static int cluster_set_master(in_addr_t peer
, in_addr_t master
)
1164 if (config
->cluster_iam_master
) // Sanity...
1167 LOG(3, 0, 0, "Peer %s set the master to %s...\n", fmtaddr(peer
, 0),
1168 fmtaddr(master
, 1));
1170 config
->cluster_master_address
= master
;
1173 // catchup with new master
1174 peer_send_message(master
, C_LASTSEEN
, config
->cluster_seq_number
, NULL
, 0);
1176 // delay next election
1177 config
->cluster_last_hb
= TIME
;
1180 // run election (or reset "probed" if master was set)
1181 cluster_check_master();
1185 /* Handle the slave updating the byte counters for the master. */
1187 // Note that we don't mark the session as dirty; We rely on
1188 // the slow table walk to propogate this back out to the slaves.
1190 static int cluster_handle_bytes(uint8_t *data
, int size
)
1194 b
= (bytest
*) data
;
1196 LOG(3, 0, 0, "Got byte counter update (size %d)\n", size
);
1198 /* Loop around, adding the byte
1199 counts to each of the sessions. */
1201 while (size
>= sizeof(*b
) ) {
1202 if (b
->sid
> MAXSESSION
) {
1203 LOG(0, 0, 0, "Got C_BYTES with session #%d!\n", b
->sid
);
1204 return -1; /* Abort processing */
1207 session
[b
->sid
].pin
+= b
->pin
;
1208 session
[b
->sid
].pout
+= b
->pout
;
1210 increment_counter(&session
[b
->sid
].cin
, &session
[b
->sid
].cin_wrap
, b
->cin
);
1211 increment_counter(&session
[b
->sid
].cout
, &session
[b
->sid
].cout_wrap
, b
->cout
);
1213 session
[b
->sid
].cin_delta
+= b
->cin
;
1214 session
[b
->sid
].cout_delta
+= b
->cout
;
1217 session
[b
->sid
].last_packet
= time_now
; // Reset idle timer!
1224 LOG(0, 0, 0, "Got C_BYTES with %d bytes of trailing junk!\n", size
);
1230 // Handle receiving a session structure in a heartbeat packet.
1232 static int cluster_recv_session(int more
, uint8_t *p
)
1234 if (more
>= MAXSESSION
) {
1235 LOG(0, 0, 0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
1239 if (session
[more
].tunnel
== T_UNDEF
) {
1240 if (config
->cluster_iam_uptodate
) { // Sanity.
1241 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined session!\n");
1243 --config
->cluster_undefined_sessions
;
1247 load_session(more
, (sessiont
*) p
); // Copy session into session table..
1249 LOG(5, more
, 0, "Received session update (%d undef)\n", config
->cluster_undefined_sessions
);
1251 if (!config
->cluster_iam_uptodate
)
1252 cluster_uptodate(); // Check to see if we're up to date.
1257 static int cluster_recv_bundle(int more
, uint8_t *p
)
1259 if (more
>= MAXBUNDLE
) {
1260 LOG(0, 0, 0, "DANGER: Received a bundle id > MAXBUNDLE!\n");
1264 if (bundle
[more
].state
== BUNDLEUNDEF
) {
1265 if (config
->cluster_iam_uptodate
) { // Sanity.
1266 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined bundle!\n");
1268 --config
->cluster_undefined_bundles
;
1272 memcpy(&bundle
[more
], p
, sizeof(bundle
[more
]) );
1274 LOG(5, 0, more
, "Received bundle update\n");
1276 if (!config
->cluster_iam_uptodate
)
1277 cluster_uptodate(); // Check to see if we're up to date.
1282 static int cluster_recv_tunnel(int more
, uint8_t *p
)
1284 if (more
>= MAXTUNNEL
) {
1285 LOG(0, 0, 0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
1289 if (tunnel
[more
].state
== TUNNELUNDEF
) {
1290 if (config
->cluster_iam_uptodate
) { // Sanity.
1291 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined tunnel!\n");
1293 --config
->cluster_undefined_tunnels
;
1297 memcpy(&tunnel
[more
], p
, sizeof(tunnel
[more
]) );
1300 // Clear tunnel control messages. These are dynamically allocated.
1301 // If we get unlucky, this may cause the tunnel to drop!
1303 tunnel
[more
].controls
= tunnel
[more
].controle
= NULL
;
1304 tunnel
[more
].controlc
= 0;
1306 LOG(5, 0, more
, "Received tunnel update\n");
1308 if (!config
->cluster_iam_uptodate
)
1309 cluster_uptodate(); // Check to see if we're up to date.
1315 // pre v5 heartbeat session structure
1322 unsigned long unique_id
;
1329 uint32_t total_cout
;
1331 uint16_t throttle_in
;
1332 uint16_t throttle_out
;
1336 in_addr_t dns1
, dns2
;
1337 routet route
[MAXROUTE
];
1343 uint8_t reserved_old_snoop
;
1344 uint8_t walled_garden
;
1346 char random_vector
[MAXTEL
];
1347 int random_vector_length
;
1349 char called
[MAXTEL
];
1350 char calling
[MAXTEL
];
1351 uint32_t tx_connect_speed
;
1352 uint32_t rx_connect_speed
;
1354 #define SF_IPCP_ACKED 1 // Has this session seen an IPCP Ack?
1355 #define SF_LCP_ACKED 2 // LCP negotiated
1356 #define SF_CCP_ACKED 4 // CCP negotiated
1358 uint16_t snoop_port
;
1365 static uint8_t *convert_session(struct oldsession
*old
)
1367 static sessiont
new;
1370 memset(&new, 0, sizeof(new));
1372 new.next
= old
->next
;
1374 new.tunnel
= old
->tunnel
;
1375 new.flags
= old
->l2tp_flags
;
1377 new.ip_pool_index
= old
->ip_pool_index
;
1378 new.unique_id
= old
->unique_id
;
1379 new.magic
= old
->magic
;
1381 new.pout
= old
->pout
;
1382 new.cin
= old
->total_cin
;
1383 new.cout
= old
->total_cout
;
1384 new.cin_delta
= old
->cin
;
1385 new.cout_delta
= old
->cout
;
1386 new.throttle_in
= old
->throttle_in
;
1387 new.throttle_out
= old
->throttle_out
;
1388 new.filter_in
= old
->filter_in
;
1389 new.filter_out
= old
->filter_out
;
1391 new.opened
= old
->opened
;
1393 new.last_packet
= old
->last_packet
;
1394 new.dns1
= old
->dns1
;
1395 new.dns2
= old
->dns2
;
1396 new.tbf_in
= old
->tbf_in
;
1397 new.tbf_out
= old
->tbf_out
;
1398 new.random_vector_length
= old
->random_vector_length
;
1399 new.tx_connect_speed
= old
->tx_connect_speed
;
1400 new.rx_connect_speed
= old
->rx_connect_speed
;
1401 new.snoop_ip
= old
->snoop_ip
;
1402 new.snoop_port
= old
->snoop_port
;
1403 new.walled_garden
= old
->walled_garden
;
1405 memcpy(new.random_vector
, old
->random_vector
, sizeof(new.random_vector
));
1406 memcpy(new.user
, old
->user
, sizeof(new.user
));
1407 memcpy(new.called
, old
->called
, sizeof(new.called
));
1408 memcpy(new.calling
, old
->calling
, sizeof(new.calling
));
1410 for (i
= 0; i
< MAXROUTE
; i
++)
1411 memcpy(&new.route
[i
], &old
->route
[i
], sizeof(new.route
[i
]));
1415 new.ppp
.phase
= Establish
;
1416 if (old
->flags
& (SF_IPCP_ACKED
|SF_LCP_ACKED
))
1418 new.ppp
.phase
= Network
;
1419 new.ppp
.lcp
= Opened
;
1420 new.ppp
.ipcp
= (old
->flags
& SF_IPCP_ACKED
) ? Opened
: Starting
;
1421 new.ppp
.ccp
= (old
->flags
& SF_CCP_ACKED
) ? Opened
: Stopped
;
1424 // no PPPv6 in old session
1425 new.ppp
.ipv6cp
= Stopped
;
1428 return (uint8_t *) &new;
1432 // Process a heartbeat..
1434 // v3: added interval, timeout
1435 // v4: added table_version
1436 // v5: added ipv6, re-ordered session structure
1437 static int cluster_process_heartbeat(uint8_t *data
, int size
, int more
, uint8_t *p
, in_addr_t addr
)
1440 int s
= size
- (p
-data
);
1445 # error "need to update cluster_process_heartbeat()"
1448 // we handle versions 3 through 5
1449 if (hb_ver
< 3 || hb_ver
> HB_VERSION
) {
1450 LOG(0, 0, 0, "Received a heartbeat version that I don't support (%d)!\n", hb_ver
);
1451 return -1; // Ignore it??
1454 // Ok. It's a heartbeat packet from a cluster master!
1462 if (h
->clusterid
!= config
->bind_address
)
1463 return -1; // It's not part of our cluster.
1465 if (config
->cluster_iam_master
) { // Sanity...
1466 // Note that this MUST match the election process above!
1468 LOG(0, 0, 0, "I just got a heartbeat from master %s, but _I_ am the master!\n", fmtaddr(addr
, 0));
1470 LOG(0, 0, 0, "Heartbeat with zero basetime! Ignoring\n");
1471 return -1; // Skip it.
1475 if (h
->table_version
> config
->cluster_table_version
) {
1476 LOG(0, 0, 0, "They've seen more state changes (%" PRIu64
" vs my %" PRIu64
") so I'm gone!\n",
1477 h
->table_version
, config
->cluster_table_version
);
1482 if (h
->table_version
< config
->cluster_table_version
)
1486 if (basetime
> h
->basetime
) {
1487 LOG(0, 0, 0, "They're an older master than me so I'm gone!\n");
1492 if (basetime
< h
->basetime
)
1495 if (my_address
< addr
) { // Tie breaker.
1496 LOG(0, 0, 0, "They're a higher IP address than me, so I'm gone!\n");
1502 // Send it a unicast heartbeat to see give it a chance to die.
1503 // NOTE: It's actually safe to do seq-number - 1 without checking
1506 cluster_catchup_slave(config
->cluster_seq_number
- 1, addr
);
1508 return -1; // Skip it.
1512 // Try and guard against a stray master appearing.
1514 // Ignore heartbeats received from another master before the
1515 // timeout (less a smidgen) for the old master has elapsed.
1517 // Note that after a clean failover, the cluster_master_address
1518 // is cleared, so this doesn't run.
1520 if (config
->cluster_master_address
&& addr
!= config
->cluster_master_address
) {
1521 LOG(0, 0, 0, "Ignoring stray heartbeat from %s, current master %s has not yet timed out (last heartbeat %.1f seconds ago).\n",
1522 fmtaddr(addr
, 0), fmtaddr(config
->cluster_master_address
, 1),
1523 0.1 * (TIME
- config
->cluster_last_hb
));
1524 return -1; // ignore
1527 if (config
->cluster_seq_number
== -1) // Don't have one. Just align to the master...
1528 config
->cluster_seq_number
= h
->seq
;
1530 config
->cluster_last_hb
= TIME
; // Reset to ensure that we don't become master!!
1531 config
->cluster_last_hb_ver
= hb_ver
; // remember what cluster version the master is using
1533 if (config
->cluster_seq_number
!= h
->seq
) { // Out of sequence heartbeat!
1534 static int lastseen_seq
= 0;
1535 static time_t lastseen_time
= 0;
1537 // limit to once per second for a particular seq#
1538 int ask
= (config
->cluster_seq_number
!= lastseen_seq
|| time_now
!= lastseen_time
);
1540 LOG(1, 0, 0, "HB: Got seq# %d but was expecting %d. %s.\n",
1541 h
->seq
, config
->cluster_seq_number
,
1542 ask
? "Asking for resend" : "Ignoring");
1546 lastseen_seq
= config
->cluster_seq_number
;
1547 lastseen_time
= time_now
;
1548 peer_send_message(addr
, C_LASTSEEN
, config
->cluster_seq_number
, NULL
, 0);
1551 config
->cluster_last_hb
= TIME
; // Reset to ensure that we don't become master!!
1553 // Just drop the packet. The master will resend it as part of the catchup.
1557 // Save the packet in our buffer.
1558 // This is needed in case we become the master.
1559 config
->cluster_seq_number
= (h
->seq
+1)%HB_MAX_SEQ
;
1560 i
= h
->seq
% HB_HISTORY_SIZE
;
1561 past_hearts
[i
].seq
= h
->seq
;
1562 past_hearts
[i
].size
= size
;
1563 memcpy(&past_hearts
[i
].data
, data
, size
); // Save it.
1566 // Check that we don't have too many undefined sessions, and
1567 // that the free session pointer is correct.
1568 cluster_check_sessions(h
->highsession
, h
->freesession
, h
->highbundle
, h
->hightunnel
);
1570 if (h
->interval
!= config
->cluster_hb_interval
)
1572 LOG(2, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
1573 h
->interval
, config
->cluster_hb_interval
);
1575 config
->cluster_hb_interval
= h
->interval
;
1578 if (h
->timeout
!= config
->cluster_hb_timeout
)
1580 LOG(2, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
1581 h
->timeout
, config
->cluster_hb_timeout
);
1583 config
->cluster_hb_timeout
= h
->timeout
;
1586 // Ok. process the packet...
1589 type
= *((uint32_t *) p
);
1590 p
+= sizeof(uint32_t);
1591 s
-= sizeof(uint32_t);
1593 more
= *((uint32_t *) p
);
1594 p
+= sizeof(uint32_t);
1595 s
-= sizeof(uint32_t);
1598 case C_CSESSION
: { // Compressed session structure.
1599 uint8_t c
[ sizeof(sessiont
) + 2];
1601 uint8_t *orig_p
= p
;
1603 size
= rle_decompress((uint8_t **) &p
, s
, c
, sizeof(c
) );
1606 // session struct changed with v5
1609 if (size
!= sizeof(struct oldsession
)) {
1610 LOG(0, 0, 0, "DANGER: Received a v%d CSESSION that didn't decompress correctly!\n", hb_ver
);
1611 // Now what? Should exit! No-longer up to date!
1614 cluster_recv_session(more
, convert_session((struct oldsession
*) c
));
1618 if (size
!= sizeof(sessiont
) ) { // Ouch! Very very bad!
1619 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1620 // Now what? Should exit! No-longer up to date!
1624 cluster_recv_session(more
, c
);
1630 if (s
< sizeof(struct oldsession
))
1633 cluster_recv_session(more
, convert_session((struct oldsession
*) p
));
1635 p
+= sizeof(struct oldsession
);
1636 s
-= sizeof(struct oldsession
);
1640 if ( s
< sizeof(session
[more
]))
1643 cluster_recv_session(more
, p
);
1645 p
+= sizeof(session
[more
]);
1646 s
-= sizeof(session
[more
]);
1649 case C_CTUNNEL
: { // Compressed tunnel structure.
1650 uint8_t c
[ sizeof(tunnelt
) + 2];
1652 uint8_t *orig_p
= p
;
1654 size
= rle_decompress((uint8_t **) &p
, s
, c
, sizeof(c
));
1657 if (size
!= sizeof(tunnelt
) ) { // Ouch! Very very bad!
1658 LOG(0, 0, 0, "DANGER: Received a CTUNNEL that didn't decompress correctly!\n");
1659 // Now what? Should exit! No-longer up to date!
1663 cluster_recv_tunnel(more
, c
);
1668 if ( s
< sizeof(tunnel
[more
]))
1671 cluster_recv_tunnel(more
, p
);
1673 p
+= sizeof(tunnel
[more
]);
1674 s
-= sizeof(tunnel
[more
]);
1677 case C_CBUNDLE
: { // Compressed bundle structure.
1678 uint8_t c
[ sizeof(bundlet
) + 2];
1680 uint8_t *orig_p
= p
;
1682 size
= rle_decompress((uint8_t **) &p
, s
, c
, sizeof(c
));
1685 if (size
!= sizeof(bundlet
) ) { // Ouch! Very very bad!
1686 LOG(0, 0, 0, "DANGER: Received a CBUNDLE that didn't decompress correctly!\n");
1687 // Now what? Should exit! No-longer up to date!
1691 cluster_recv_bundle(more
, c
);
1696 if ( s
< sizeof(bundle
[more
]))
1699 cluster_recv_bundle(more
, p
);
1701 p
+= sizeof(bundle
[more
]);
1702 s
-= sizeof(bundle
[more
]);
1705 LOG(0, 0, 0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type
);
1706 return -1; // can't process any more of the packet!!
1710 if (config
->cluster_master_address
!= addr
)
1712 LOG(0, 0, 0, "My master just changed from %s to %s!\n",
1713 fmtaddr(config
->cluster_master_address
, 0), fmtaddr(addr
, 1));
1715 config
->cluster_master_address
= addr
;
1718 config
->cluster_last_hb
= TIME
; // Successfully received a heartbeat!
1719 config
->cluster_table_version
= h
->table_version
;
1723 LOG(0, 0, 0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
1728 // We got a packet on the cluster port!
1729 // Handle pings, lastseens, and heartbeats!
1731 int processcluster(uint8_t *data
, int size
, in_addr_t addr
)
1737 if (addr
== my_address
)
1738 return -1; // Ignore it. Something looped back the multicast!
1740 LOG(5, 0, 0, "Process cluster: %d bytes from %s\n", size
, fmtaddr(addr
, 0));
1742 if (s
<= 0) // Any data there??
1748 type
= *((uint32_t *) p
);
1749 p
+= sizeof(uint32_t);
1750 s
-= sizeof(uint32_t);
1752 more
= *((uint32_t *) p
);
1753 p
+= sizeof(uint32_t);
1754 s
-= sizeof(uint32_t);
1758 case C_PING
: // Update the peers table.
1759 return cluster_add_peer(addr
, more
, (pingt
*) p
, s
);
1761 case C_MASTER
: // Our master is wrong
1762 return cluster_set_master(addr
, more
);
1764 case C_LASTSEEN
: // Catch up a slave (slave missed a packet).
1765 return cluster_catchup_slave(more
, addr
);
1767 case C_FORWARD
: // Forwarded control packet. pass off to processudp.
1768 case C_FORWARD_DAE
: // Forwarded DAE packet. pass off to processdae.
1769 if (!config
->cluster_iam_master
)
1771 LOG(0, 0, 0, "I'm not the master, but I got a C_FORWARD_%s from %s?\n",
1772 type
== C_FORWARD_DAE
? "_DAE" : "", fmtaddr(addr
, 0));
1778 struct sockaddr_in a
;
1779 a
.sin_addr
.s_addr
= more
;
1781 a
.sin_port
= *(int *) p
;
1785 LOG(4, 0, 0, "Got a forwarded %spacket... (%s:%d)\n",
1786 type
== C_FORWARD_DAE
? "DAE " : "", fmtaddr(more
, 0), a
.sin_port
);
1789 if (type
== C_FORWARD_DAE
)
1791 struct in_addr local
;
1792 local
.s_addr
= config
->bind_address
? config
->bind_address
: my_address
;
1793 processdae(p
, s
, &a
, sizeof(a
), &local
);
1796 processudp(p
, s
, &a
);
1801 case C_THROTTLE
: { // Receive a forwarded packet from a slave.
1802 if (!config
->cluster_iam_master
) {
1803 LOG(0, 0, 0, "I'm not the master, but I got a C_THROTTLE from %s?\n", fmtaddr(addr
, 0));
1807 tbf_queue_packet(more
, p
, s
); // The TBF id tells wether it goes in or out.
1811 // Receive a walled garden packet from a slave.
1812 if (!config
->cluster_iam_master
) {
1813 LOG(0, 0, 0, "I'm not the master, but I got a C_GARDEN from %s?\n", fmtaddr(addr
, 0));
1821 if (!config
->cluster_iam_master
) {
1822 LOG(0, 0, 0, "I'm not the master, but I got a C_BYTES from %s?\n", fmtaddr(addr
, 0));
1826 return cluster_handle_bytes(p
, s
);
1828 case C_KILL
: // The master asked us to die!? (usually because we're too out of date).
1829 if (config
->cluster_iam_master
) {
1830 LOG(0, 0, 0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", fmtaddr(addr
, 0), more
);
1833 if (more
!= config
->cluster_seq_number
) {
1834 LOG(0, 0, 0, "The master asked us to die but the seq number didn't match!?\n");
1838 if (addr
!= config
->cluster_master_address
) {
1839 LOG(0, 0, 0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%s)\n",
1840 fmtaddr(addr
, 0), fmtaddr(config
->cluster_master_address
, 1));
1841 // We can only warn about it. The master might really have switched!
1844 LOG(0, 0, 0, "Received a valid C_KILL: I'm going to die now.\n");
1846 exit(0); // Lets be paranoid;
1847 return -1; // Just signalling the compiler.
1850 LOG(4, 0, 0, "Got a heartbeat from %s\n", fmtaddr(addr
, 0));
1851 return cluster_process_heartbeat(data
, size
, more
, p
, addr
);
1854 LOG(0, 0, 0, "Strange type packet received on cluster socket (%d)\n", type
);
1860 LOG(0, 0, 0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n");
1864 //====================================================================================================
1866 int cmd_show_cluster(struct cli_def
*cli
, char *command
, char **argv
, int argc
)
1870 if (CLI_HELP_REQUESTED
)
1871 return CLI_HELP_NO_ARGS
;
1873 cli_print(cli
, "Cluster status : %s", config
->cluster_iam_master
? "Master" : "Slave" );
1874 cli_print(cli
, "My address : %s", fmtaddr(my_address
, 0));
1875 cli_print(cli
, "VIP address : %s", fmtaddr(config
->bind_address
, 0));
1876 cli_print(cli
, "Multicast address: %s", fmtaddr(config
->cluster_address
, 0));
1877 cli_print(cli
, "Multicast i'face : %s", config
->cluster_interface
);
1879 if (!config
->cluster_iam_master
) {
1880 cli_print(cli
, "My master : %s (last heartbeat %.1f seconds old)",
1881 config
->cluster_master_address
1882 ? fmtaddr(config
->cluster_master_address
, 0)
1884 0.1 * (TIME
- config
->cluster_last_hb
));
1885 cli_print(cli
, "Uptodate : %s", config
->cluster_iam_uptodate
? "Yes" : "No");
1886 cli_print(cli
, "Table version # : %" PRIu64
, config
->cluster_table_version
);
1887 cli_print(cli
, "Next sequence number expected: %d", config
->cluster_seq_number
);
1888 cli_print(cli
, "%d sessions undefined of %d", config
->cluster_undefined_sessions
, config
->cluster_highest_sessionid
);
1889 cli_print(cli
, "%d bundles undefined of %d", config
->cluster_undefined_bundles
, config
->cluster_highest_bundleid
);
1890 cli_print(cli
, "%d tunnels undefined of %d", config
->cluster_undefined_tunnels
, config
->cluster_highest_tunnelid
);
1892 cli_print(cli
, "Table version # : %" PRIu64
, config
->cluster_table_version
);
1893 cli_print(cli
, "Next heartbeat # : %d", config
->cluster_seq_number
);
1894 cli_print(cli
, "Highest session : %d", config
->cluster_highest_sessionid
);
1895 cli_print(cli
, "Highest bundle : %d", config
->cluster_highest_bundleid
);
1896 cli_print(cli
, "Highest tunnel : %d", config
->cluster_highest_tunnelid
);
1897 cli_print(cli
, "%d changes queued for sending", config
->cluster_num_changes
);
1899 cli_print(cli
, "%d peers.", num_peers
);
1902 cli_print(cli
, "%20s %10s %8s", "Address", "Basetime", "Age");
1903 for (i
= 0; i
< num_peers
; ++i
) {
1904 cli_print(cli
, "%20s %10u %8d", fmtaddr(peers
[i
].peer
, 0),
1905 peers
[i
].basetime
, TIME
- peers
[i
].timestamp
);
1911 // Simple run-length-encoding compression.
1913 // 1 byte < 128 = count of non-zero bytes following. // Not legal to be zero.
1914 // n non-zero bytes;
1916 // 1 byte > 128 = (count - 128) run of zero bytes. //
1918 // count == 0 indicates end of compressed stream.
1920 // Compress from 'src' into 'dst'. return number of bytes
1922 // Updates *src_p to indicate 1 past last bytes used.
1924 // We could get an extra byte in the zero runs by storing (count-1)
1925 // but I'm playing it safe.
1927 // Worst case is a 50% expansion in space required (trying to
1928 // compress { 0x00, 0x01 } * N )
1929 static int rle_compress(uint8_t **src_p
, int ssize
, uint8_t *dst
, int dsize
)
1932 int orig_dsize
= dsize
;
1936 while (ssize
> 0 && dsize
> 2) {
1938 x
= dst
++; --dsize
; // Reserve space for count byte..
1940 if (*src
) { // Copy a run of non-zero bytes.
1941 while (*src
&& count
< 127 && ssize
> 0 && dsize
> 1) { // Count number of non-zero bytes.
1946 *x
= count
; // Store number of non-zero bytes. Guarenteed to be non-zero!
1948 } else { // Compress a run of zero bytes.
1949 while (*src
== 0 && count
< 127 && ssize
> 0) {
1958 *dst
++ = 0x0; // Add Stop byte.
1962 return (orig_dsize
- dsize
);
1966 // Decompress the buffer into **p.
1967 // 'psize' is the size of the decompression buffer available.
1969 // Returns the number of bytes decompressed.
1971 // Decompresses from '*src_p' into 'dst'.
1972 // Return the number of dst bytes used.
1973 // Updates the 'src_p' pointer to point to the
1974 // first un-used byte.
1975 static int rle_decompress(uint8_t **src_p
, int ssize
, uint8_t *dst
, int dsize
)
1978 int orig_dsize
= dsize
;
1979 uint8_t *src
= *src_p
;
1981 while (ssize
>0 && dsize
> 0) { // While there's more to decompress, and there's room in the decompress buffer...
1982 count
= *src
++; --ssize
; // get the count byte from the source.
1983 if (count
== 0x0) // End marker reached? If so, finish.
1986 if (count
& 0x80) { // Decompress a run of zeros
1987 for (count
&= 0x7f ; count
> 0 && dsize
> 0; --count
) {
1991 } else { // Copy run of non-zero bytes.
1992 for ( ; count
> 0 && ssize
&& dsize
; --count
) { // Copy non-zero bytes across.
1999 return (orig_dsize
- dsize
);