1 // L2TPNS Clustering Stuff
10 #include <sys/socket.h>
11 #include <netinet/in.h>
12 #include <arpa/inet.h>
13 #include <sys/ioctl.h>
29 * All cluster packets have the same format.
31 * One or more instances of
33 * a 32 bit 'extra' data dependant on the 'type'.
34 * zero or more bytes of structure data, dependant on the type.
39 extern int cluster_sockfd
; // The filedescriptor for the cluster communications port.
41 in_addr_t my_address
= 0; // The network address of my ethernet port.
42 static int walk_session_number
= 0; // The next session to send when doing the slow table walk.
43 static int walk_bundle_number
= 0; // The next bundle to send when doing the slow table walk.
44 static int walk_tunnel_number
= 0; // The next tunnel to send when doing the slow table walk.
45 int forked
= 0; // Sanity check: CLI must not diddle with heartbeat table
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!
53 } cluster_changes
[MAX_CHANGES
]; // Queue of changed structures that need to go out when next heartbeat.
58 uint8_t 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.
67 } peers
[CLUSTER_MAX_SIZE
]; // List of all the peers we've heard from.
68 static int num_peers
; // Number of peers in list.
70 static int rle_decompress(uint8_t **src_p
, int ssize
, uint8_t *dst
, int dsize
);
71 static int rle_compress(uint8_t **src_p
, int ssize
, uint8_t *dst
, int dsize
);
74 // Create a listening socket
76 // This joins the cluster multi-cast group.
80 struct sockaddr_in addr
;
81 struct sockaddr_in interface_addr
;
86 config
->cluster_undefined_sessions
= MAXSESSION
-1;
87 config
->cluster_undefined_bundles
= MAXBUNDLE
-1;
88 config
->cluster_undefined_tunnels
= MAXTUNNEL
-1;
90 if (!config
->cluster_address
)
92 if (!*config
->cluster_interface
)
95 cluster_sockfd
= socket(AF_INET
, SOCK_DGRAM
, IPPROTO_UDP
);
97 memset(&addr
, 0, sizeof(addr
));
98 addr
.sin_family
= AF_INET
;
99 addr
.sin_port
= htons(CLUSTERPORT
);
100 addr
.sin_addr
.s_addr
= INADDR_ANY
;
101 setsockopt(cluster_sockfd
, SOL_SOCKET
, SO_REUSEADDR
, &addr
, sizeof(addr
));
103 opt
= fcntl(cluster_sockfd
, F_GETFL
, 0);
104 fcntl(cluster_sockfd
, F_SETFL
, opt
| O_NONBLOCK
);
106 if (bind(cluster_sockfd
, (void *) &addr
, sizeof(addr
)) < 0)
108 LOG(0, 0, 0, "Failed to bind cluster socket: %s\n", strerror(errno
));
112 strcpy(ifr
.ifr_name
, config
->cluster_interface
);
113 if (ioctl(cluster_sockfd
, SIOCGIFADDR
, &ifr
) < 0)
115 LOG(0, 0, 0, "Failed to get interface address for (%s): %s\n", config
->cluster_interface
, strerror(errno
));
119 memcpy(&interface_addr
, &ifr
.ifr_addr
, sizeof(interface_addr
));
120 my_address
= interface_addr
.sin_addr
.s_addr
;
122 // Join multicast group.
123 mreq
.imr_multiaddr
.s_addr
= config
->cluster_address
;
124 mreq
.imr_interface
= interface_addr
.sin_addr
;
127 opt
= 0; // Turn off multicast loopback.
128 setsockopt(cluster_sockfd
, IPPROTO_IP
, IP_MULTICAST_LOOP
, &opt
, sizeof(opt
));
130 if (config
->cluster_mcast_ttl
!= 1)
133 if (config
->cluster_mcast_ttl
> 0)
134 ttl
= config
->cluster_mcast_ttl
< 256 ? config
->cluster_mcast_ttl
: 255;
136 setsockopt(cluster_sockfd
, IPPROTO_IP
, IP_MULTICAST_TTL
, &ttl
, sizeof(ttl
));
139 if (setsockopt(cluster_sockfd
, IPPROTO_IP
, IP_ADD_MEMBERSHIP
, &mreq
, sizeof(mreq
)) < 0)
141 LOG(0, 0, 0, "Failed to setsockopt (join mcast group): %s\n", strerror(errno
));
145 if (setsockopt(cluster_sockfd
, IPPROTO_IP
, IP_MULTICAST_IF
, &interface_addr
, sizeof(interface_addr
)) < 0)
147 LOG(0, 0, 0, "Failed to setsockopt (set mcast interface): %s\n", strerror(errno
));
151 config
->cluster_last_hb
= TIME
;
152 config
->cluster_seq_number
= -1;
154 return cluster_sockfd
;
159 // Send a chunk of data to the entire cluster (usually via the multicast
163 static int cluster_send_data(void *data
, int datalen
)
165 struct sockaddr_in addr
= {0};
167 if (!cluster_sockfd
) return -1;
168 if (!config
->cluster_address
) return 0;
170 addr
.sin_addr
.s_addr
= config
->cluster_address
;
171 addr
.sin_port
= htons(CLUSTERPORT
);
172 addr
.sin_family
= AF_INET
;
174 LOG(5, 0, 0, "Cluster send data: %d bytes\n", datalen
);
176 if (sendto(cluster_sockfd
, data
, datalen
, MSG_NOSIGNAL
, (void *) &addr
, sizeof(addr
)) < 0)
178 LOG(0, 0, 0, "sendto: %s\n", strerror(errno
));
186 // Add a chunk of data to a heartbeat packet.
187 // Maintains the format. Assumes that the caller
188 // has passed in a big enough buffer!
190 static void add_type(uint8_t **p
, int type
, int more
, uint8_t *data
, int size
)
192 *((uint32_t *) (*p
)) = type
;
193 *p
+= sizeof(uint32_t);
195 *((uint32_t *)(*p
)) = more
;
196 *p
+= sizeof(uint32_t);
198 if (data
&& size
> 0) {
199 memcpy(*p
, data
, size
);
204 // advertise our presence via BGP or gratuitous ARP
205 static void advertise_routes(void)
209 bgp_enable_routing(1);
212 if (config
->send_garp
)
213 send_garp(config
->bind_address
); // Start taking traffic.
216 // withdraw our routes (BGP only)
217 static void withdraw_routes(void)
221 bgp_enable_routing(0);
225 static void cluster_uptodate(void)
227 if (config
->cluster_iam_uptodate
)
230 if (config
->cluster_undefined_sessions
|| config
->cluster_undefined_tunnels
|| config
->cluster_undefined_bundles
)
233 config
->cluster_iam_uptodate
= 1;
235 LOG(0, 0, 0, "Now uptodate with master.\n");
240 // Send a unicast UDP packet to a peer with 'data' as the
243 static int peer_send_data(in_addr_t peer
, uint8_t *data
, int size
)
245 struct sockaddr_in addr
= {0};
247 if (!cluster_sockfd
) return -1;
248 if (!config
->cluster_address
) return 0;
253 addr
.sin_addr
.s_addr
= peer
;
254 addr
.sin_port
= htons(CLUSTERPORT
);
255 addr
.sin_family
= AF_INET
;
257 LOG_HEX(5, "Peer send", data
, size
);
259 if (sendto(cluster_sockfd
, data
, size
, MSG_NOSIGNAL
, (void *) &addr
, sizeof(addr
)) < 0)
261 LOG(0, 0, 0, "sendto: %s\n", strerror(errno
));
269 // Send a structured message to a peer with a single element of type 'type'.
271 static int peer_send_message(in_addr_t peer
, int type
, int more
, uint8_t *data
, int size
)
273 uint8_t buf
[65536]; // Vast overkill.
276 LOG(4, 0, 0, "Sending message to peer (type %d, more %d, size %d)\n", type
, more
, size
);
277 add_type(&p
, type
, more
, data
, size
);
279 return peer_send_data(peer
, buf
, (p
-buf
) );
282 // send a packet to the master
283 static int _forward_packet(uint8_t *data
, int size
, in_addr_t addr
, int port
, int type
)
285 uint8_t buf
[65536]; // Vast overkill.
288 if (!config
->cluster_master_address
) // No election has been held yet. Just skip it.
291 LOG(4, 0, 0, "Forwarding packet from %s to master (size %d)\n", fmtaddr(addr
, 0), size
);
294 add_type(&p
, type
, addr
, (uint8_t *) &port
, sizeof(port
)); // ick. should be uint16_t
295 memcpy(p
, data
, size
);
298 return peer_send_data(config
->cluster_master_address
, buf
, (p
- buf
));
302 // Forward a state changing packet to the master.
304 // The master just processes the payload as if it had
305 // received it off the tun device.
307 int master_forward_packet(uint8_t *data
, int size
, in_addr_t addr
, int port
)
309 return _forward_packet(data
, size
, addr
, port
, C_FORWARD
);
312 // Forward a DAE RADIUS packet to the master.
313 int master_forward_dae_packet(uint8_t *data
, int size
, in_addr_t addr
, int port
)
315 return _forward_packet(data
, size
, addr
, port
, C_FORWARD_DAE
);
319 // Forward a throttled packet to the master for handling.
321 // The master just drops the packet into the appropriate
322 // token bucket queue, and lets normal processing take care
325 int master_throttle_packet(int tbfid
, uint8_t *data
, int size
)
327 uint8_t buf
[65536]; // Vast overkill.
330 if (!config
->cluster_master_address
) // No election has been held yet. Just skip it.
333 LOG(4, 0, 0, "Throttling packet master (size %d, tbfid %d)\n", size
, tbfid
);
335 add_type(&p
, C_THROTTLE
, tbfid
, data
, size
);
337 return peer_send_data(config
->cluster_master_address
, buf
, (p
-buf
) );
342 // Forward a walled garden packet to the master for handling.
344 // The master just writes the packet straight to the tun
345 // device (where is will normally loop through the
346 // firewall rules, and come back in on the tun device)
348 // (Note that this must be called with the tun header
349 // as the start of the data).
350 int master_garden_packet(sessionidt s
, uint8_t *data
, int size
)
352 uint8_t buf
[65536]; // Vast overkill.
355 if (!config
->cluster_master_address
) // No election has been held yet. Just skip it.
358 LOG(4, 0, 0, "Walled garden packet to master (size %d)\n", size
);
360 add_type(&p
, C_GARDEN
, s
, data
, size
);
362 return peer_send_data(config
->cluster_master_address
, buf
, (p
-buf
));
367 // Send a chunk of data as a heartbeat..
368 // We save it in the history buffer as we do so.
370 static void send_heartbeat(int seq
, uint8_t *data
, int size
)
374 if (size
> sizeof(past_hearts
[0].data
))
376 LOG(0, 0, 0, "Tried to heartbeat something larger than the maximum packet!\n");
380 i
= seq
% HB_HISTORY_SIZE
;
381 past_hearts
[i
].seq
= seq
;
382 past_hearts
[i
].size
= size
;
383 memcpy(&past_hearts
[i
].data
, data
, size
); // Save it.
384 cluster_send_data(data
, size
);
388 // Send an 'i am alive' message to every machine in the cluster.
390 void cluster_send_ping(time_t basetime
)
392 uint8_t buff
[100 + sizeof(pingt
)];
396 if (config
->cluster_iam_master
&& basetime
) // We're heartbeating so no need to ping.
399 LOG(5, 0, 0, "Sending cluster ping...\n");
402 x
.addr
= config
->bind_address
;
403 x
.undef
= config
->cluster_undefined_sessions
+ config
->cluster_undefined_tunnels
+ config
->cluster_undefined_bundles
;
404 x
.basetime
= basetime
;
406 add_type(&p
, C_PING
, basetime
, (uint8_t *) &x
, sizeof(x
));
407 cluster_send_data(buff
, (p
-buff
) );
411 // Walk the session counters looking for non-zero ones to send
412 // to the master. We send up to 600 of them at one time.
413 // We examine a maximum of 3000 sessions.
414 // (50k max session should mean that we normally
415 // examine the entire session table every 25 seconds).
417 #define MAX_B_RECS (600)
418 void master_update_counts(void)
421 bytest b
[MAX_B_RECS
+1];
423 if (config
->cluster_iam_master
) // Only happens on the slaves.
426 if (!config
->cluster_master_address
) // If we don't have a master, skip it for a while.
429 i
= MAX_B_RECS
* 5; // Examine max 3000 sessions;
430 if (config
->cluster_highest_sessionid
> i
)
431 i
= config
->cluster_highest_sessionid
;
433 for ( c
= 0; i
> 0 ; --i
) {
434 // Next session to look at.
435 walk_session_number
++;
436 if ( walk_session_number
> config
->cluster_highest_sessionid
)
437 walk_session_number
= 1;
439 if (!sess_local
[walk_session_number
].cin
&& !sess_local
[walk_session_number
].cout
)
440 continue; // Unchanged. Skip it.
442 b
[c
].sid
= walk_session_number
;
443 b
[c
].pin
= sess_local
[walk_session_number
].pin
;
444 b
[c
].pout
= sess_local
[walk_session_number
].pout
;
445 b
[c
].cin
= sess_local
[walk_session_number
].cin
;
446 b
[c
].cout
= sess_local
[walk_session_number
].cout
;
449 sess_local
[walk_session_number
].pin
= sess_local
[walk_session_number
].pout
= 0;
450 sess_local
[walk_session_number
].cin
= sess_local
[walk_session_number
].cout
= 0;
452 if (++c
> MAX_B_RECS
) // Send a max of 600 elements in a packet.
456 if (!c
) // Didn't find any that changes. Get out of here!
460 // Forward the data to the master.
461 LOG(4, 0, 0, "Sending byte counters to master (%d elements)\n", c
);
462 peer_send_message(config
->cluster_master_address
, C_BYTES
, c
, (uint8_t *) &b
, sizeof(b
[0]) * c
);
467 // On the master, check how our slaves are going. If
468 // one of them's not up-to-date we'll heartbeat faster.
469 // If we don't have any of them, then we need to turn
470 // on our own packet handling!
472 void cluster_check_slaves(void)
475 static int have_peers
= 0;
476 int had_peers
= have_peers
;
479 if (!config
->cluster_iam_master
)
480 return; // Only runs on the master...
482 config
->cluster_iam_uptodate
= 1; // cleared in loop below
484 for (i
= have_peers
= 0; i
< num_peers
; i
++)
486 if ((peers
[i
].timestamp
+ config
->cluster_hb_timeout
) < t
)
487 continue; // Stale peer! Skip them.
489 if (!peers
[i
].basetime
)
490 continue; // Shutdown peer! Skip them.
492 if (peers
[i
].uptodate
)
495 config
->cluster_iam_uptodate
= 0; // Start fast heartbeats
498 // in a cluster, withdraw/add routes when we get a peer/lose peers
499 if (have_peers
!= had_peers
)
501 if (had_peers
< config
->cluster_master_min_adv
&&
502 have_peers
>= config
->cluster_master_min_adv
)
505 else if (had_peers
>= config
->cluster_master_min_adv
&&
506 have_peers
< config
->cluster_master_min_adv
)
512 // Check that we have a master. If it's been too
513 // long since we heard from a master then hold an election.
515 void cluster_check_master(void)
517 int i
, count
, high_unique_id
= 0;
520 static int probed
= 0;
523 if (config
->cluster_iam_master
)
524 return; // Only runs on the slaves...
526 // If the master is late (missed 2 hearbeats by a second and a
527 // hair) it may be that the switch has dropped us from the
528 // multicast group, try unicasting probes to the master
529 // which will hopefully respond with a unicast heartbeat that
530 // will allow us to limp along until the querier next runs.
531 if (config
->cluster_master_address
532 && TIME
> (config
->cluster_last_hb
+ 2 * config
->cluster_hb_interval
+ 11))
534 if (!probed
|| (TIME
> (probed
+ 2 * config
->cluster_hb_interval
)))
537 LOG(1, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
538 0.1 * (TIME
- (config
->cluster_last_hb
+ config
->cluster_hb_interval
)));
540 peer_send_message(config
->cluster_master_address
,
541 C_LASTSEEN
, config
->cluster_seq_number
, NULL
, 0);
543 } else { // We got a recent heartbeat; reset the probe flag.
547 if (TIME
< (config
->cluster_last_hb
+ config
->cluster_hb_timeout
))
548 return; // Everything's ok!
550 config
->cluster_last_hb
= TIME
+ 1; // Just the one election thanks.
551 config
->cluster_master_address
= 0;
553 LOG(0, 0, 0, "Master timed out! Holding election...\n");
555 // In the process of shutting down, can't be master
559 for (i
= have_peers
= 0; i
< num_peers
; i
++)
561 if ((peers
[i
].timestamp
+ config
->cluster_hb_timeout
) < t
)
562 continue; // Stale peer! Skip them.
564 if (!peers
[i
].basetime
)
565 continue; // Shutdown peer! Skip them.
567 if (peers
[i
].basetime
< basetime
) {
568 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers
[i
].peer
, 0));
569 return; // They'll win the election. Get out of here.
572 if (peers
[i
].basetime
== basetime
&&
573 peers
[i
].peer
> my_address
) {
574 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers
[i
].peer
, 0));
575 return; // They'll win the election. Wait for them to come up.
578 if (peers
[i
].uptodate
)
582 // Wow. it's been ages since I last heard a heartbeat
583 // and I'm better than an of my peers so it's time
584 // to become a master!!!
586 config
->cluster_iam_master
= 1;
588 LOG(0, 0, 0, "I am declaring myself the master!\n");
590 if (have_peers
< config
->cluster_master_min_adv
)
595 if (config
->cluster_seq_number
== -1)
596 config
->cluster_seq_number
= 0;
599 // Go through and mark all the tunnels as defined.
600 // Count the highest used tunnel number as well.
602 config
->cluster_highest_tunnelid
= 0;
603 for (i
= 0; i
< MAXTUNNEL
; ++i
) {
604 if (tunnel
[i
].state
== TUNNELUNDEF
)
605 tunnel
[i
].state
= TUNNELFREE
;
607 if (tunnel
[i
].state
!= TUNNELFREE
&& i
> config
->cluster_highest_tunnelid
)
608 config
->cluster_highest_tunnelid
= i
;
612 // Go through and mark all the bundles as defined.
613 // Count the highest used bundle number as well.
615 config
->cluster_highest_bundleid
= 0;
616 for (i
= 0; i
< MAXBUNDLE
; ++i
) {
617 if (bundle
[i
].state
== BUNDLEUNDEF
)
618 bundle
[i
].state
= BUNDLEFREE
;
620 if (bundle
[i
].state
!= BUNDLEFREE
&& i
> config
->cluster_highest_bundleid
)
621 config
->cluster_highest_bundleid
= i
;
625 // Go through and mark all the sessions as being defined.
626 // reset the idle timeouts.
627 // add temporary byte counters to permanent ones.
628 // Re-string the free list.
629 // Find the ID of the highest session.
632 config
->cluster_highest_sessionid
= 0;
633 for (i
= 0, count
= 0; i
< MAXSESSION
; ++i
) {
634 if (session
[i
].tunnel
== T_UNDEF
) {
635 session
[i
].tunnel
= T_FREE
;
639 if (!session
[i
].opened
) { // Unused session. Add to free list.
640 memset(&session
[i
], 0, sizeof(session
[i
]));
641 session
[i
].tunnel
= T_FREE
;
642 session
[last_free
].next
= i
;
648 // Reset idle timeouts..
649 session
[i
].last_packet
= session
[i
].last_data
= time_now
;
651 // Reset die relative to our uptime rather than the old master's
652 if (session
[i
].die
) session
[i
].die
= TIME
;
654 // Accumulate un-sent byte/packet counters.
655 increment_counter(&session
[i
].cin
, &session
[i
].cin_wrap
, sess_local
[i
].cin
);
656 increment_counter(&session
[i
].cout
, &session
[i
].cout_wrap
, sess_local
[i
].cout
);
657 session
[i
].cin_delta
+= sess_local
[i
].cin
;
658 session
[i
].cout_delta
+= sess_local
[i
].cout
;
660 session
[i
].pin
+= sess_local
[i
].pin
;
661 session
[i
].pout
+= sess_local
[i
].pout
;
663 sess_local
[i
].cin
= sess_local
[i
].cout
= 0;
664 sess_local
[i
].pin
= sess_local
[i
].pout
= 0;
666 sess_local
[i
].radius
= 0; // Reset authentication as the radius blocks aren't up to date.
668 if (session
[i
].unique_id
>= high_unique_id
) // This is different to the index into the session table!!!
669 high_unique_id
= session
[i
].unique_id
+1;
671 session
[i
].tbf_in
= session
[i
].tbf_out
= 0; // Remove stale pointers from old master.
672 throttle_session(i
, session
[i
].throttle_in
, session
[i
].throttle_out
);
674 config
->cluster_highest_sessionid
= i
;
677 session
[last_free
].next
= 0; // End of chain.
678 last_id
= high_unique_id
; // Keep track of the highest used session ID.
682 rebuild_address_pool();
684 // If we're not the very first master, this is a big issue!
686 LOG(0, 0, 0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count
);
688 config
->cluster_undefined_sessions
= 0;
689 config
->cluster_undefined_bundles
= 0;
690 config
->cluster_undefined_tunnels
= 0;
691 config
->cluster_iam_uptodate
= 1; // assume all peers are up-to-date
693 // FIXME. We need to fix up the tunnel control message
694 // queue here! There's a number of other variables we
695 // should also update.
700 // Check that our session table is validly matching what the
701 // master has in mind.
703 // In particular, if we have too many sessions marked 'undefined'
704 // we fix it up here, and we ensure that the 'first free session'
707 static void cluster_check_sessions(int highsession
, int freesession_ptr
, int highbundle
, int hightunnel
)
711 sessionfree
= freesession_ptr
; // Keep the freesession ptr valid.
713 if (config
->cluster_iam_uptodate
)
716 if (highsession
> config
->cluster_undefined_sessions
&& highbundle
> config
->cluster_undefined_bundles
&& hightunnel
> config
->cluster_undefined_tunnels
)
719 // Clear out defined sessions, counting the number of
721 config
->cluster_undefined_sessions
= 0;
722 for (i
= 1 ; i
< MAXSESSION
; ++i
) {
723 if (i
> highsession
) {
724 if (session
[i
].tunnel
== T_UNDEF
) session
[i
].tunnel
= T_FREE
; // Defined.
728 if (session
[i
].tunnel
== T_UNDEF
)
729 ++config
->cluster_undefined_sessions
;
732 // Clear out defined bundles, counting the number of
734 config
->cluster_undefined_bundles
= 0;
735 for (i
= 1 ; i
< MAXBUNDLE
; ++i
) {
736 if (i
> highbundle
) {
737 if (bundle
[i
].state
== BUNDLEUNDEF
) bundle
[i
].state
= BUNDLEFREE
; // Defined.
741 if (bundle
[i
].state
== BUNDLEUNDEF
)
742 ++config
->cluster_undefined_bundles
;
745 // Clear out defined tunnels, counting the number of
747 config
->cluster_undefined_tunnels
= 0;
748 for (i
= 1 ; i
< MAXTUNNEL
; ++i
) {
749 if (i
> hightunnel
) {
750 if (tunnel
[i
].state
== TUNNELUNDEF
) tunnel
[i
].state
= TUNNELFREE
; // Defined.
754 if (tunnel
[i
].state
== TUNNELUNDEF
)
755 ++config
->cluster_undefined_tunnels
;
759 if (config
->cluster_undefined_sessions
|| config
->cluster_undefined_tunnels
|| config
->cluster_undefined_bundles
) {
760 LOG(2, 0, 0, "Cleared undefined sessions/bundles/tunnels. %d sess (high %d), %d bund (high %d), %d tunn (high %d)\n",
761 config
->cluster_undefined_sessions
, highsession
, config
->cluster_undefined_bundles
, highbundle
, config
->cluster_undefined_tunnels
, hightunnel
);
765 // Are we up to date?
767 if (!config
->cluster_iam_uptodate
)
771 static int hb_add_type(uint8_t **p
, int type
, int id
)
774 case C_CSESSION
: { // Compressed C_SESSION.
775 uint8_t c
[sizeof(sessiont
) * 2]; // Bigger than worst case.
776 uint8_t *d
= (uint8_t *) &session
[id
];
780 size
= rle_compress( &d
, sizeof(sessiont
), c
, sizeof(c
) );
782 // Did we compress the full structure, and is the size actually
784 if ( (d
- orig
) == sizeof(sessiont
) && size
< sizeof(sessiont
) ) {
785 add_type(p
, C_CSESSION
, id
, c
, size
);
788 // Failed to compress : Fall through.
791 add_type(p
, C_SESSION
, id
, (uint8_t *) &session
[id
], sizeof(sessiont
));
794 case C_CBUNDLE
: { // Compressed C_BUNDLE
795 uint8_t c
[sizeof(bundlet
) * 2]; // Bigger than worst case.
796 uint8_t *d
= (uint8_t *) &bundle
[id
];
800 size
= rle_compress( &d
, sizeof(bundlet
), c
, sizeof(c
) );
802 // Did we compress the full structure, and is the size actually
804 if ( (d
- orig
) == sizeof(bundlet
) && size
< sizeof(bundlet
) ) {
805 add_type(p
, C_CBUNDLE
, id
, c
, size
);
808 // Failed to compress : Fall through.
812 add_type(p
, C_BUNDLE
, id
, (uint8_t *) &bundle
[id
], sizeof(bundlet
));
815 case C_CTUNNEL
: { // Compressed C_TUNNEL
816 uint8_t c
[sizeof(tunnelt
) * 2]; // Bigger than worst case.
817 uint8_t *d
= (uint8_t *) &tunnel
[id
];
821 size
= rle_compress( &d
, sizeof(tunnelt
), c
, sizeof(c
) );
823 // Did we compress the full structure, and is the size actually
825 if ( (d
- orig
) == sizeof(tunnelt
) && size
< sizeof(tunnelt
) ) {
826 add_type(p
, C_CTUNNEL
, id
, c
, size
);
829 // Failed to compress : Fall through.
832 add_type(p
, C_TUNNEL
, id
, (uint8_t *) &tunnel
[id
], sizeof(tunnelt
));
835 LOG(0, 0, 0, "Found an invalid type in heart queue! (%d)\n", type
);
843 // Send a heartbeat, incidently sending out any queued changes..
845 void cluster_heartbeat()
847 int i
, count
= 0, tcount
= 0, bcount
= 0;
848 uint8_t buff
[MAX_HEART_SIZE
+ sizeof(heartt
) + sizeof(int) ];
852 if (!config
->cluster_iam_master
) // Only the master does this.
855 config
->cluster_table_version
+= config
->cluster_num_changes
;
857 // Fill out the heartbeat header.
858 memset(&h
, 0, sizeof(h
));
860 h
.version
= HB_VERSION
;
861 h
.seq
= config
->cluster_seq_number
;
862 h
.basetime
= basetime
;
863 h
.clusterid
= config
->bind_address
; // Will this do??
864 h
.basetime
= basetime
;
865 h
.highsession
= config
->cluster_highest_sessionid
;
866 h
.freesession
= sessionfree
;
867 h
.hightunnel
= config
->cluster_highest_tunnelid
;
868 h
.highbundle
= config
->cluster_highest_bundleid
;
869 h
.size_sess
= sizeof(sessiont
); // Just in case.
870 h
.size_bund
= sizeof(bundlet
);
871 h
.size_tunn
= sizeof(tunnelt
);
872 h
.interval
= config
->cluster_hb_interval
;
873 h
.timeout
= config
->cluster_hb_timeout
;
874 h
.table_version
= config
->cluster_table_version
;
876 add_type(&p
, C_HEARTBEAT
, HB_VERSION
, (uint8_t *) &h
, sizeof(h
));
878 for (i
= 0; i
< config
->cluster_num_changes
; ++i
) {
879 hb_add_type(&p
, cluster_changes
[i
].type
, cluster_changes
[i
].id
);
882 if (p
> (buff
+ sizeof(buff
))) { // Did we somehow manage to overun the buffer?
883 LOG(0, 0, 0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", (int) (p
- buff
));
889 // Fill out the packet with sessions from the session table...
890 // (not forgetting to leave space so we can get some tunnels in too )
891 while ( (p
+ sizeof(uint32_t) * 2 + sizeof(sessiont
) * 2 ) < (buff
+ MAX_HEART_SIZE
) ) {
893 if (!walk_session_number
) // session #0 isn't valid.
894 ++walk_session_number
;
896 if (count
>= config
->cluster_highest_sessionid
) // If we're a small cluster, don't go wild.
899 hb_add_type(&p
, C_CSESSION
, walk_session_number
);
900 walk_session_number
= (1+walk_session_number
)%(config
->cluster_highest_sessionid
+1); // +1 avoids divide by zero.
902 ++count
; // Count the number of extra sessions we're sending.
906 // Fill out the packet with tunnels from the tunnel table...
907 // This effectively means we walk the tunnel table more quickly
908 // than the session table. This is good because stuffing up a
909 // tunnel is a much bigger deal than stuffing up a session.
911 while ( (p
+ sizeof(uint32_t) * 2 + sizeof(tunnelt
) ) < (buff
+ MAX_HEART_SIZE
) ) {
913 if (!walk_tunnel_number
) // tunnel #0 isn't valid.
914 ++walk_tunnel_number
;
916 if (tcount
>= config
->cluster_highest_tunnelid
)
919 hb_add_type(&p
, C_CTUNNEL
, walk_tunnel_number
);
920 walk_tunnel_number
= (1+walk_tunnel_number
)%(config
->cluster_highest_tunnelid
+1); // +1 avoids divide by zero.
926 // Fill out the packet with bundles from the bundle table...
927 while ( (p
+ sizeof(uint32_t) * 2 + sizeof(bundlet
) ) < (buff
+ MAX_HEART_SIZE
) ) {
929 if (!walk_bundle_number
) // bundle #0 isn't valid.
930 ++walk_bundle_number
;
932 if (bcount
>= config
->cluster_highest_bundleid
)
935 hb_add_type(&p
, C_CBUNDLE
, walk_bundle_number
);
936 walk_bundle_number
= (1+walk_bundle_number
)%(config
->cluster_highest_bundleid
+1); // +1 avoids divide by zero.
941 // Did we do something wrong?
942 if (p
> (buff
+ sizeof(buff
))) { // Did we somehow manage to overun the buffer?
943 LOG(0, 0, 0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", (int) (p
- buff
));
948 LOG(3, 0, 0, "Sending v%d heartbeat #%d, change #%" PRIu64
" with %d changes "
949 "(%d x-sess, %d x-bundles, %d x-tunnels, %d highsess, %d highbund, %d hightun, size %d)\n",
950 HB_VERSION
, h
.seq
, h
.table_version
, config
->cluster_num_changes
,
951 count
, bcount
, tcount
, config
->cluster_highest_sessionid
, config
->cluster_highest_bundleid
,
952 config
->cluster_highest_tunnelid
, (int) (p
- buff
));
954 config
->cluster_num_changes
= 0;
956 send_heartbeat(h
.seq
, buff
, (p
-buff
) ); // Send out the heartbeat to the cluster, keeping a copy of it.
958 config
->cluster_seq_number
= (config
->cluster_seq_number
+1)%HB_MAX_SEQ
; // Next seq number to use.
962 // A structure of type 'type' has changed; Add it to the queue to send.
964 static int type_changed(int type
, int id
)
968 for (i
= 0 ; i
< config
->cluster_num_changes
; ++i
)
969 if ( cluster_changes
[i
].id
== id
&&
970 cluster_changes
[i
].type
== type
)
971 return 0; // Already marked for change.
973 cluster_changes
[i
].type
= type
;
974 cluster_changes
[i
].id
= id
;
975 ++config
->cluster_num_changes
;
977 if (config
->cluster_num_changes
> MAX_CHANGES
)
978 cluster_heartbeat(); // flush now
984 // A particular session has been changed!
985 int cluster_send_session(int sid
)
987 if (!config
->cluster_iam_master
) {
988 LOG(0, sid
, 0, "I'm not a master, but I just tried to change a session!\n");
993 LOG(0, sid
, 0, "cluster_send_session called from child process!\n");
997 return type_changed(C_CSESSION
, sid
);
1000 // A particular bundle has been changed!
1001 int cluster_send_bundle(int bid
)
1003 if (!config
->cluster_iam_master
) {
1004 LOG(0, 0, bid
, "I'm not a master, but I just tried to change a bundle!\n");
1008 return type_changed(C_CBUNDLE
, bid
);
1011 // A particular tunnel has been changed!
1012 int cluster_send_tunnel(int tid
)
1014 if (!config
->cluster_iam_master
) {
1015 LOG(0, 0, tid
, "I'm not a master, but I just tried to change a tunnel!\n");
1019 return type_changed(C_CTUNNEL
, tid
);
1024 // We're a master, and a slave has just told us that it's
1025 // missed a packet. We'll resend it every packet since
1026 // the last one it's seen.
1028 static int cluster_catchup_slave(int seq
, in_addr_t slave
)
1033 LOG(1, 0, 0, "Slave %s sent LASTSEEN with seq %d\n", fmtaddr(slave
, 0), seq
);
1034 if (!config
->cluster_iam_master
) {
1035 LOG(1, 0, 0, "Got LASTSEEN but I'm not a master! Redirecting it to %s.\n",
1036 fmtaddr(config
->cluster_master_address
, 0));
1038 peer_send_message(slave
, C_MASTER
, config
->cluster_master_address
, NULL
, 0);
1042 diff
= config
->cluster_seq_number
- seq
; // How many packet do we need to send?
1046 if (diff
>= HB_HISTORY_SIZE
) { // Ouch. We don't have the packet to send it!
1047 LOG(0, 0, 0, "A slave asked for message %d when our seq number is %d. Killing it.\n",
1048 seq
, config
->cluster_seq_number
);
1049 return peer_send_message(slave
, C_KILL
, seq
, NULL
, 0);// Kill the slave. Nothing else to do.
1052 LOG(1, 0, 0, "Sending %d catchup packets to slave %s\n", diff
, fmtaddr(slave
, 0) );
1054 // Now resend every packet that it missed, in order.
1055 while (seq
!= config
->cluster_seq_number
) {
1056 s
= seq
% HB_HISTORY_SIZE
;
1057 if (seq
!= past_hearts
[s
].seq
) {
1058 LOG(0, 0, 0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n",
1059 fmtaddr(slave
, 0), seq
, past_hearts
[s
].seq
, s
, config
->cluster_seq_number
);
1060 return -1; // What to do here!?
1062 peer_send_data(slave
, past_hearts
[s
].data
, past_hearts
[s
].size
);
1063 seq
= (seq
+1)%HB_MAX_SEQ
; // Increment to next seq number.
1065 return 0; // All good!
1069 // We've heard from another peer! Add it to the list
1070 // that we select from at election time.
1072 static int cluster_add_peer(in_addr_t peer
, time_t basetime
, pingt
*pp
, int size
)
1075 in_addr_t clusterid
;
1078 // Allow for backward compatability.
1079 // Just the ping packet into a new structure to allow
1080 // for the possibility that we might have received
1081 // more or fewer elements than we were expecting.
1082 if (size
> sizeof(p
))
1085 memset( (void *) &p
, 0, sizeof(p
) );
1086 memcpy( (void *) &p
, (void *) pp
, size
);
1089 if (clusterid
!= config
->bind_address
)
1092 LOG(4, 0, 0, "Skipping ping from %s (different cluster)\n", fmtaddr(peer
, 0));
1096 for (i
= 0; i
< num_peers
; ++i
)
1098 if (peers
[i
].peer
!= peer
)
1101 // This peer already exists. Just update the timestamp.
1102 peers
[i
].basetime
= basetime
;
1103 peers
[i
].timestamp
= TIME
;
1104 peers
[i
].uptodate
= !p
.undef
;
1108 // Is this the master shutting down??
1109 if (peer
== config
->cluster_master_address
) {
1110 LOG(3, 0, 0, "Master %s %s\n", fmtaddr(config
->cluster_master_address
, 0),
1111 basetime
? "has restarted!" : "shutting down...");
1113 config
->cluster_master_address
= 0;
1114 config
->cluster_last_hb
= 0; // Force an election.
1115 cluster_check_master();
1120 LOG(4, 0, 0, "Adding %s as a peer\n", fmtaddr(peer
, 0));
1122 // Not found. Is there a stale slot to re-use?
1123 for (i
= 0; i
< num_peers
; ++i
)
1125 if (!peers
[i
].basetime
) // Shutdown
1128 if ((peers
[i
].timestamp
+ config
->cluster_hb_timeout
* 10) < TIME
) // Stale.
1132 if (i
>= CLUSTER_MAX_SIZE
)
1135 LOG(0, 0, 0, "Tried to add %s as a peer, but I already have %d of them!\n", fmtaddr(peer
, 0), i
);
1139 peers
[i
].peer
= peer
;
1140 peers
[i
].basetime
= basetime
;
1141 peers
[i
].timestamp
= TIME
;
1142 peers
[i
].uptodate
= !p
.undef
;
1146 LOG(1, 0, 0, "Added %s as a new peer. Now %d peers\n", fmtaddr(peer
, 0), num_peers
);
1152 // A slave responds with C_MASTER when it gets a message which should have gone to a master.
1153 static int cluster_set_master(in_addr_t peer
, in_addr_t master
)
1155 if (config
->cluster_iam_master
) // Sanity...
1158 LOG(3, 0, 0, "Peer %s set the master to %s...\n", fmtaddr(peer
, 0),
1159 fmtaddr(master
, 1));
1161 config
->cluster_master_address
= master
;
1164 // catchup with new master
1165 peer_send_message(master
, C_LASTSEEN
, config
->cluster_seq_number
, NULL
, 0);
1167 // delay next election
1168 config
->cluster_last_hb
= TIME
;
1171 // run election (or reset "probed" if master was set)
1172 cluster_check_master();
1176 /* Handle the slave updating the byte counters for the master. */
1178 // Note that we don't mark the session as dirty; We rely on
1179 // the slow table walk to propogate this back out to the slaves.
1181 static int cluster_handle_bytes(uint8_t *data
, int size
)
1185 b
= (bytest
*) data
;
1187 LOG(3, 0, 0, "Got byte counter update (size %d)\n", size
);
1189 /* Loop around, adding the byte
1190 counts to each of the sessions. */
1192 while (size
>= sizeof(*b
) ) {
1193 if (b
->sid
> MAXSESSION
) {
1194 LOG(0, 0, 0, "Got C_BYTES with session #%d!\n", b
->sid
);
1195 return -1; /* Abort processing */
1198 session
[b
->sid
].pin
+= b
->pin
;
1199 session
[b
->sid
].pout
+= b
->pout
;
1201 increment_counter(&session
[b
->sid
].cin
, &session
[b
->sid
].cin_wrap
, b
->cin
);
1202 increment_counter(&session
[b
->sid
].cout
, &session
[b
->sid
].cout_wrap
, b
->cout
);
1204 session
[b
->sid
].cin_delta
+= b
->cin
;
1205 session
[b
->sid
].cout_delta
+= b
->cout
;
1208 session
[b
->sid
].last_packet
= session
[b
->sid
].last_data
= time_now
;
1210 session
[b
->sid
].last_data
= time_now
;
1217 LOG(0, 0, 0, "Got C_BYTES with %d bytes of trailing junk!\n", size
);
1223 // Handle receiving a session structure in a heartbeat packet.
1225 static int cluster_recv_session(int more
, uint8_t *p
)
1227 if (more
>= MAXSESSION
) {
1228 LOG(0, 0, 0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
1232 if (session
[more
].tunnel
== T_UNDEF
) {
1233 if (config
->cluster_iam_uptodate
) { // Sanity.
1234 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined session!\n");
1236 --config
->cluster_undefined_sessions
;
1240 load_session(more
, (sessiont
*) p
); // Copy session into session table..
1242 LOG(5, more
, 0, "Received session update (%d undef)\n", config
->cluster_undefined_sessions
);
1244 if (!config
->cluster_iam_uptodate
)
1245 cluster_uptodate(); // Check to see if we're up to date.
1250 static int cluster_recv_bundle(int more
, uint8_t *p
)
1252 if (more
>= MAXBUNDLE
) {
1253 LOG(0, 0, 0, "DANGER: Received a bundle id > MAXBUNDLE!\n");
1257 if (bundle
[more
].state
== BUNDLEUNDEF
) {
1258 if (config
->cluster_iam_uptodate
) { // Sanity.
1259 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined bundle!\n");
1261 --config
->cluster_undefined_bundles
;
1265 memcpy(&bundle
[more
], p
, sizeof(bundle
[more
]) );
1267 LOG(5, 0, more
, "Received bundle update\n");
1269 if (!config
->cluster_iam_uptodate
)
1270 cluster_uptodate(); // Check to see if we're up to date.
1275 static int cluster_recv_tunnel(int more
, uint8_t *p
)
1277 if (more
>= MAXTUNNEL
) {
1278 LOG(0, 0, 0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
1282 if (tunnel
[more
].state
== TUNNELUNDEF
) {
1283 if (config
->cluster_iam_uptodate
) { // Sanity.
1284 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined tunnel!\n");
1286 --config
->cluster_undefined_tunnels
;
1290 memcpy(&tunnel
[more
], p
, sizeof(tunnel
[more
]) );
1293 // Clear tunnel control messages. These are dynamically allocated.
1294 // If we get unlucky, this may cause the tunnel to drop!
1296 tunnel
[more
].controls
= tunnel
[more
].controle
= NULL
;
1297 tunnel
[more
].controlc
= 0;
1299 LOG(5, 0, more
, "Received tunnel update\n");
1301 if (!config
->cluster_iam_uptodate
)
1302 cluster_uptodate(); // Check to see if we're up to date.
1308 // pre v6 heartbeat session structure
1329 uint32_t cin_wrap
, cout_wrap
;
1330 uint32_t cin_delta
, cout_delta
;
1331 uint16_t throttle_in
;
1332 uint16_t throttle_out
;
1338 uint32_t session_timeout
;
1339 uint32_t idle_timeout
;
1342 in_addr_t dns1
, dns2
;
1343 routet route
[MAXROUTE
];
1346 int random_vector_length
;
1347 uint8_t random_vector
[MAXTEL
];
1349 char called
[MAXTEL
];
1350 char calling
[MAXTEL
];
1351 uint32_t tx_connect_speed
;
1352 uint32_t rx_connect_speed
;
1359 uint16_t snoop_port
;
1360 uint8_t walled_garden
;
1361 uint8_t ipv6prefixlen
;
1362 struct in6_addr ipv6route
;
1363 char reserved_3
[11];
1366 static uint8_t *convert_session(struct oldsession
*old
)
1368 static sessiont
new;
1371 memset(&new, 0, sizeof(new));
1373 new.next
= old
->next
;
1375 new.tunnel
= old
->tunnel
;
1376 new.flags
= old
->flags
;
1377 new.ppp
.phase
= old
->ppp
.phase
;
1378 new.ppp
.lcp
= old
->ppp
.lcp
;
1379 new.ppp
.ipcp
= old
->ppp
.ipcp
;
1380 new.ppp
.ipv6cp
= old
->ppp
.ipv6cp
;
1381 new.ppp
.ccp
= old
->ppp
.ccp
;
1383 new.ip_pool_index
= old
->ip_pool_index
;
1384 new.unique_id
= old
->unique_id
;
1385 new.magic
= old
->magic
;
1387 new.pout
= old
->pout
;
1389 new.cout
= old
->cout
;
1390 new.cin_wrap
= old
->cin_wrap
;
1391 new.cout_wrap
= old
->cout_wrap
;
1392 new.cin_delta
= old
->cin_delta
;
1393 new.cout_delta
= old
->cout_delta
;
1394 new.throttle_in
= old
->throttle_in
;
1395 new.throttle_out
= old
->throttle_out
;
1396 new.filter_in
= old
->filter_in
;
1397 new.filter_out
= old
->filter_out
;
1399 new.opened
= old
->opened
;
1401 new.session_timeout
= old
->session_timeout
;
1402 new.idle_timeout
= old
->idle_timeout
;
1403 new.last_packet
= old
->last_packet
;
1404 new.last_data
= old
->last_data
;
1405 new.dns1
= old
->dns1
;
1406 new.dns2
= old
->dns2
;
1407 new.tbf_in
= old
->tbf_in
;
1408 new.tbf_out
= old
->tbf_out
;
1409 new.random_vector_length
= old
->random_vector_length
;
1410 new.tx_connect_speed
= old
->tx_connect_speed
;
1411 new.rx_connect_speed
= old
->rx_connect_speed
;
1412 new.timeout
= old
->timeout
;
1413 new.mrru
= old
->mrru
;
1414 new.mssf
= old
->mssf
;
1415 new.epdis
= old
->epdis
;
1416 new.bundle
= old
->bundle
;
1417 new.snoop_ip
= old
->snoop_ip
;
1418 new.snoop_port
= old
->snoop_port
;
1419 new.walled_garden
= old
->walled_garden
;
1420 new.ipv6prefixlen
= old
->ipv6prefixlen
;
1421 new.ipv6route
= old
->ipv6route
;
1423 memcpy(new.random_vector
, old
->random_vector
, sizeof(new.random_vector
));
1424 memcpy(new.user
, old
->user
, sizeof(new.user
));
1425 memcpy(new.called
, old
->called
, sizeof(new.called
));
1426 memcpy(new.calling
, old
->calling
, sizeof(new.calling
));
1428 for (i
= 0; i
< MAXROUTE
; i
++)
1429 memcpy(&new.route
[i
], &old
->route
[i
], sizeof(new.route
[i
]));
1431 return (uint8_t *) &new;
1435 // Process a heartbeat..
1437 // v6: added RADIUS class attribute, re-ordered session structure
1438 static int cluster_process_heartbeat(uint8_t *data
, int size
, int more
, uint8_t *p
, in_addr_t addr
)
1441 int s
= size
- (p
-data
);
1446 # error "need to update cluster_process_heartbeat()"
1449 // we handle versions 5 through 6
1450 if (hb_ver
< 5 || hb_ver
> HB_VERSION
) {
1451 LOG(0, 0, 0, "Received a heartbeat version that I don't support (%d)!\n", hb_ver
);
1452 return -1; // Ignore it??
1455 if (size
> sizeof(past_hearts
[0].data
)) {
1456 LOG(0, 0, 0, "Received an oversize heartbeat from %s (%d)!\n", fmtaddr(addr
, 0), size
);
1467 if (h
->clusterid
!= config
->bind_address
)
1468 return -1; // It's not part of our cluster.
1470 if (config
->cluster_iam_master
) { // Sanity...
1471 // Note that this MUST match the election process above!
1473 LOG(0, 0, 0, "I just got a heartbeat from master %s, but _I_ am the master!\n", fmtaddr(addr
, 0));
1475 LOG(0, 0, 0, "Heartbeat with zero basetime! Ignoring\n");
1476 return -1; // Skip it.
1479 if (h
->table_version
> config
->cluster_table_version
) {
1480 LOG(0, 0, 0, "They've seen more state changes (%" PRIu64
" vs my %" PRIu64
") so I'm gone!\n",
1481 h
->table_version
, config
->cluster_table_version
);
1487 if (h
->table_version
< config
->cluster_table_version
)
1490 if (basetime
> h
->basetime
) {
1491 LOG(0, 0, 0, "They're an older master than me so I'm gone!\n");
1496 if (basetime
< h
->basetime
)
1499 if (my_address
< addr
) { // Tie breaker.
1500 LOG(0, 0, 0, "They're a higher IP address than me, so I'm gone!\n");
1506 // Send it a unicast heartbeat to see give it a chance to die.
1507 // NOTE: It's actually safe to do seq-number - 1 without checking
1510 cluster_catchup_slave(config
->cluster_seq_number
- 1, addr
);
1512 return -1; // Skip it.
1516 // Try and guard against a stray master appearing.
1518 // Ignore heartbeats received from another master before the
1519 // timeout (less a smidgen) for the old master has elapsed.
1521 // Note that after a clean failover, the cluster_master_address
1522 // is cleared, so this doesn't run.
1524 if (config
->cluster_master_address
&& addr
!= config
->cluster_master_address
) {
1525 LOG(0, 0, 0, "Ignoring stray heartbeat from %s, current master %s has not yet timed out (last heartbeat %.1f seconds ago).\n",
1526 fmtaddr(addr
, 0), fmtaddr(config
->cluster_master_address
, 1),
1527 0.1 * (TIME
- config
->cluster_last_hb
));
1528 return -1; // ignore
1531 if (config
->cluster_seq_number
== -1) // Don't have one. Just align to the master...
1532 config
->cluster_seq_number
= h
->seq
;
1534 config
->cluster_last_hb
= TIME
; // Reset to ensure that we don't become master!!
1535 config
->cluster_last_hb_ver
= hb_ver
; // remember what cluster version the master is using
1537 if (config
->cluster_seq_number
!= h
->seq
) { // Out of sequence heartbeat!
1538 static int lastseen_seq
= 0;
1539 static time_t lastseen_time
= 0;
1541 // limit to once per second for a particular seq#
1542 int ask
= (config
->cluster_seq_number
!= lastseen_seq
|| time_now
!= lastseen_time
);
1544 LOG(1, 0, 0, "HB: Got seq# %d but was expecting %d. %s.\n",
1545 h
->seq
, config
->cluster_seq_number
,
1546 ask
? "Asking for resend" : "Ignoring");
1550 lastseen_seq
= config
->cluster_seq_number
;
1551 lastseen_time
= time_now
;
1552 peer_send_message(addr
, C_LASTSEEN
, config
->cluster_seq_number
, NULL
, 0);
1555 config
->cluster_last_hb
= TIME
; // Reset to ensure that we don't become master!!
1557 // Just drop the packet. The master will resend it as part of the catchup.
1561 // Save the packet in our buffer.
1562 // This is needed in case we become the master.
1563 config
->cluster_seq_number
= (h
->seq
+1)%HB_MAX_SEQ
;
1564 i
= h
->seq
% HB_HISTORY_SIZE
;
1565 past_hearts
[i
].seq
= h
->seq
;
1566 past_hearts
[i
].size
= size
;
1567 memcpy(&past_hearts
[i
].data
, data
, size
); // Save it.
1570 // Check that we don't have too many undefined sessions, and
1571 // that the free session pointer is correct.
1572 cluster_check_sessions(h
->highsession
, h
->freesession
, h
->highbundle
, h
->hightunnel
);
1574 if (h
->interval
!= config
->cluster_hb_interval
)
1576 LOG(2, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
1577 h
->interval
, config
->cluster_hb_interval
);
1579 config
->cluster_hb_interval
= h
->interval
;
1582 if (h
->timeout
!= config
->cluster_hb_timeout
)
1584 LOG(2, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
1585 h
->timeout
, config
->cluster_hb_timeout
);
1587 config
->cluster_hb_timeout
= h
->timeout
;
1590 // Ok. process the packet...
1593 type
= *((uint32_t *) p
);
1594 p
+= sizeof(uint32_t);
1595 s
-= sizeof(uint32_t);
1597 more
= *((uint32_t *) p
);
1598 p
+= sizeof(uint32_t);
1599 s
-= sizeof(uint32_t);
1602 case C_CSESSION
: { // Compressed session structure.
1603 uint8_t c
[ sizeof(sessiont
) + 2];
1605 uint8_t *orig_p
= p
;
1607 size
= rle_decompress((uint8_t **) &p
, s
, c
, sizeof(c
) );
1610 // session struct changed with v5
1613 if (size
!= sizeof(struct oldsession
)) {
1614 LOG(0, 0, 0, "DANGER: Received a v%d CSESSION that didn't decompress correctly!\n", hb_ver
);
1615 // Now what? Should exit! No-longer up to date!
1618 cluster_recv_session(more
, convert_session((struct oldsession
*) c
));
1622 if (size
!= sizeof(sessiont
) ) { // Ouch! Very very bad!
1623 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1624 // Now what? Should exit! No-longer up to date!
1628 cluster_recv_session(more
, c
);
1634 if (s
< sizeof(struct oldsession
))
1637 cluster_recv_session(more
, convert_session((struct oldsession
*) p
));
1639 p
+= sizeof(struct oldsession
);
1640 s
-= sizeof(struct oldsession
);
1644 if ( s
< sizeof(session
[more
]))
1647 cluster_recv_session(more
, p
);
1649 p
+= sizeof(session
[more
]);
1650 s
-= sizeof(session
[more
]);
1653 case C_CTUNNEL
: { // Compressed tunnel structure.
1654 uint8_t c
[ sizeof(tunnelt
) + 2];
1656 uint8_t *orig_p
= p
;
1658 size
= rle_decompress((uint8_t **) &p
, s
, c
, sizeof(c
));
1661 if (size
!= sizeof(tunnelt
) ) { // Ouch! Very very bad!
1662 LOG(0, 0, 0, "DANGER: Received a CTUNNEL that didn't decompress correctly!\n");
1663 // Now what? Should exit! No-longer up to date!
1667 cluster_recv_tunnel(more
, c
);
1672 if ( s
< sizeof(tunnel
[more
]))
1675 cluster_recv_tunnel(more
, p
);
1677 p
+= sizeof(tunnel
[more
]);
1678 s
-= sizeof(tunnel
[more
]);
1681 case C_CBUNDLE
: { // Compressed bundle structure.
1682 uint8_t c
[ sizeof(bundlet
) + 2];
1684 uint8_t *orig_p
= p
;
1686 size
= rle_decompress((uint8_t **) &p
, s
, c
, sizeof(c
));
1689 if (size
!= sizeof(bundlet
) ) { // Ouch! Very very bad!
1690 LOG(0, 0, 0, "DANGER: Received a CBUNDLE that didn't decompress correctly!\n");
1691 // Now what? Should exit! No-longer up to date!
1695 cluster_recv_bundle(more
, c
);
1700 if ( s
< sizeof(bundle
[more
]))
1703 cluster_recv_bundle(more
, p
);
1705 p
+= sizeof(bundle
[more
]);
1706 s
-= sizeof(bundle
[more
]);
1709 LOG(0, 0, 0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type
);
1710 return -1; // can't process any more of the packet!!
1714 if (config
->cluster_master_address
!= addr
)
1716 LOG(0, 0, 0, "My master just changed from %s to %s!\n",
1717 fmtaddr(config
->cluster_master_address
, 0), fmtaddr(addr
, 1));
1719 config
->cluster_master_address
= addr
;
1722 config
->cluster_last_hb
= TIME
; // Successfully received a heartbeat!
1723 config
->cluster_table_version
= h
->table_version
;
1727 LOG(0, 0, 0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
1732 // We got a packet on the cluster port!
1733 // Handle pings, lastseens, and heartbeats!
1735 int processcluster(uint8_t *data
, int size
, in_addr_t addr
)
1741 if (addr
== my_address
)
1742 return -1; // Ignore it. Something looped back the multicast!
1744 LOG(5, 0, 0, "Process cluster: %d bytes from %s\n", size
, fmtaddr(addr
, 0));
1746 if (s
<= 0) // Any data there??
1752 type
= *((uint32_t *) p
);
1753 p
+= sizeof(uint32_t);
1754 s
-= sizeof(uint32_t);
1756 more
= *((uint32_t *) p
);
1757 p
+= sizeof(uint32_t);
1758 s
-= sizeof(uint32_t);
1762 case C_PING
: // Update the peers table.
1763 return cluster_add_peer(addr
, more
, (pingt
*) p
, s
);
1765 case C_MASTER
: // Our master is wrong
1766 return cluster_set_master(addr
, more
);
1768 case C_LASTSEEN
: // Catch up a slave (slave missed a packet).
1769 return cluster_catchup_slave(more
, addr
);
1771 case C_FORWARD
: // Forwarded control packet. pass off to processudp.
1772 case C_FORWARD_DAE
: // Forwarded DAE packet. pass off to processdae.
1773 if (!config
->cluster_iam_master
)
1775 LOG(0, 0, 0, "I'm not the master, but I got a C_FORWARD%s from %s?\n",
1776 type
== C_FORWARD_DAE
? "_DAE" : "", fmtaddr(addr
, 0));
1782 struct sockaddr_in a
;
1783 a
.sin_addr
.s_addr
= more
;
1785 a
.sin_port
= *(int *) p
;
1789 LOG(4, 0, 0, "Got a forwarded %spacket... (%s:%d)\n",
1790 type
== C_FORWARD_DAE
? "DAE " : "", fmtaddr(more
, 0), a
.sin_port
);
1793 if (type
== C_FORWARD_DAE
)
1795 struct in_addr local
;
1796 local
.s_addr
= config
->bind_address
? config
->bind_address
: my_address
;
1797 processdae(p
, s
, &a
, sizeof(a
), &local
);
1800 processudp(p
, s
, &a
);
1805 case C_THROTTLE
: { // Receive a forwarded packet from a slave.
1806 if (!config
->cluster_iam_master
) {
1807 LOG(0, 0, 0, "I'm not the master, but I got a C_THROTTLE from %s?\n", fmtaddr(addr
, 0));
1811 tbf_queue_packet(more
, p
, s
); // The TBF id tells wether it goes in or out.
1815 // Receive a walled garden packet from a slave.
1816 if (!config
->cluster_iam_master
) {
1817 LOG(0, 0, 0, "I'm not the master, but I got a C_GARDEN from %s?\n", fmtaddr(addr
, 0));
1825 if (!config
->cluster_iam_master
) {
1826 LOG(0, 0, 0, "I'm not the master, but I got a C_BYTES from %s?\n", fmtaddr(addr
, 0));
1830 return cluster_handle_bytes(p
, s
);
1832 case C_KILL
: // The master asked us to die!? (usually because we're too out of date).
1833 if (config
->cluster_iam_master
) {
1834 LOG(0, 0, 0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", fmtaddr(addr
, 0), more
);
1837 if (more
!= config
->cluster_seq_number
) {
1838 LOG(0, 0, 0, "The master asked us to die but the seq number didn't match!?\n");
1842 if (addr
!= config
->cluster_master_address
) {
1843 LOG(0, 0, 0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%s)\n",
1844 fmtaddr(addr
, 0), fmtaddr(config
->cluster_master_address
, 1));
1845 // We can only warn about it. The master might really have switched!
1848 LOG(0, 0, 0, "Received a valid C_KILL: I'm going to die now.\n");
1850 exit(0); // Lets be paranoid;
1851 return -1; // Just signalling the compiler.
1854 LOG(4, 0, 0, "Got a heartbeat from %s\n", fmtaddr(addr
, 0));
1855 return cluster_process_heartbeat(data
, size
, more
, p
, addr
);
1858 LOG(0, 0, 0, "Strange type packet received on cluster socket (%d)\n", type
);
1864 LOG(0, 0, 0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n");
1868 //====================================================================================================
1870 int cmd_show_cluster(struct cli_def
*cli
, char *command
, char **argv
, int argc
)
1874 if (CLI_HELP_REQUESTED
)
1875 return CLI_HELP_NO_ARGS
;
1877 cli_print(cli
, "Cluster status : %s", config
->cluster_iam_master
? "Master" : "Slave" );
1878 cli_print(cli
, "My address : %s", fmtaddr(my_address
, 0));
1879 cli_print(cli
, "VIP address : %s", fmtaddr(config
->bind_address
, 0));
1880 cli_print(cli
, "Multicast address: %s", fmtaddr(config
->cluster_address
, 0));
1881 cli_print(cli
, "Multicast i'face : %s", config
->cluster_interface
);
1883 if (!config
->cluster_iam_master
) {
1884 cli_print(cli
, "My master : %s (last heartbeat %.1f seconds old)",
1885 config
->cluster_master_address
1886 ? fmtaddr(config
->cluster_master_address
, 0)
1888 0.1 * (TIME
- config
->cluster_last_hb
));
1889 cli_print(cli
, "Uptodate : %s", config
->cluster_iam_uptodate
? "Yes" : "No");
1890 cli_print(cli
, "Table version # : %" PRIu64
, config
->cluster_table_version
);
1891 cli_print(cli
, "Next sequence number expected: %d", config
->cluster_seq_number
);
1892 cli_print(cli
, "%d sessions undefined of %d", config
->cluster_undefined_sessions
, config
->cluster_highest_sessionid
);
1893 cli_print(cli
, "%d bundles undefined of %d", config
->cluster_undefined_bundles
, config
->cluster_highest_bundleid
);
1894 cli_print(cli
, "%d tunnels undefined of %d", config
->cluster_undefined_tunnels
, config
->cluster_highest_tunnelid
);
1896 cli_print(cli
, "Table version # : %" PRIu64
, config
->cluster_table_version
);
1897 cli_print(cli
, "Next heartbeat # : %d", config
->cluster_seq_number
);
1898 cli_print(cli
, "Highest session : %d", config
->cluster_highest_sessionid
);
1899 cli_print(cli
, "Highest bundle : %d", config
->cluster_highest_bundleid
);
1900 cli_print(cli
, "Highest tunnel : %d", config
->cluster_highest_tunnelid
);
1901 cli_print(cli
, "%d changes queued for sending", config
->cluster_num_changes
);
1903 cli_print(cli
, "%d peers.", num_peers
);
1906 cli_print(cli
, "%20s %10s %8s", "Address", "Basetime", "Age");
1907 for (i
= 0; i
< num_peers
; ++i
) {
1908 cli_print(cli
, "%20s %10u %8d", fmtaddr(peers
[i
].peer
, 0),
1909 peers
[i
].basetime
, TIME
- peers
[i
].timestamp
);
1915 // Simple run-length-encoding compression.
1917 // 1 byte < 128 = count of non-zero bytes following. // Not legal to be zero.
1918 // n non-zero bytes;
1920 // 1 byte > 128 = (count - 128) run of zero bytes. //
1922 // count == 0 indicates end of compressed stream.
1924 // Compress from 'src' into 'dst'. return number of bytes
1926 // Updates *src_p to indicate 1 past last bytes used.
1928 // We could get an extra byte in the zero runs by storing (count-1)
1929 // but I'm playing it safe.
1931 // Worst case is a 50% expansion in space required (trying to
1932 // compress { 0x00, 0x01 } * N )
1933 static int rle_compress(uint8_t **src_p
, int ssize
, uint8_t *dst
, int dsize
)
1936 int orig_dsize
= dsize
;
1940 while (ssize
> 0 && dsize
> 2) {
1942 x
= dst
++; --dsize
; // Reserve space for count byte..
1944 if (*src
) { // Copy a run of non-zero bytes.
1945 while (*src
&& count
< 127 && ssize
> 0 && dsize
> 1) { // Count number of non-zero bytes.
1950 *x
= count
; // Store number of non-zero bytes. Guarenteed to be non-zero!
1952 } else { // Compress a run of zero bytes.
1953 while (*src
== 0 && count
< 127 && ssize
> 0) {
1962 *dst
++ = 0x0; // Add Stop byte.
1966 return (orig_dsize
- dsize
);
1970 // Decompress the buffer into **p.
1971 // 'psize' is the size of the decompression buffer available.
1973 // Returns the number of bytes decompressed.
1975 // Decompresses from '*src_p' into 'dst'.
1976 // Return the number of dst bytes used.
1977 // Updates the 'src_p' pointer to point to the
1978 // first un-used byte.
1979 static int rle_decompress(uint8_t **src_p
, int ssize
, uint8_t *dst
, int dsize
)
1982 int orig_dsize
= dsize
;
1983 uint8_t *src
= *src_p
;
1985 while (ssize
>0 && dsize
> 0) { // While there's more to decompress, and there's room in the decompress buffer...
1986 count
= *src
++; --ssize
; // get the count byte from the source.
1987 if (count
== 0x0) // End marker reached? If so, finish.
1990 if (count
& 0x80) { // Decompress a run of zeros
1991 for (count
&= 0x7f ; count
> 0 && dsize
> 0; --count
) {
1995 } else { // Copy run of non-zero bytes.
1996 for ( ; count
> 0 && ssize
&& dsize
; --count
) { // Copy non-zero bytes across.
2003 return (orig_dsize
- dsize
);