1 // L2TPNS Clustering Stuff
3 char const *cvs_id_cluster
= "$Id: cluster.c,v 1.41 2005-06-04 15:42:35 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_tunnel_number
= 0; // The next tunnel to send when doing the slow table walk.
46 int forked
= 0; // Sanity check: CLI must not diddle with heartbeat table
48 #define MAX_HEART_SIZE (8192) // Maximum size of heartbeat packet. Must be less than max IP packet size :)
49 #define MAX_CHANGES (MAX_HEART_SIZE/(sizeof(sessiont) + sizeof(int) ) - 2) // Assumes a session is the biggest type!
54 } cluster_changes
[MAX_CHANGES
]; // Queue of changed structures that need to go out when next heartbeat.
59 char data
[MAX_HEART_SIZE
];
60 } past_hearts
[HB_HISTORY_SIZE
]; // Ring buffer of heartbeats that we've recently sent out. Needed so
61 // we can re-transmit if needed.
68 } peers
[CLUSTER_MAX_SIZE
]; // List of all the peers we've heard from.
69 static int num_peers
; // Number of peers in list.
71 static int rle_decompress(uint8_t **src_p
, int ssize
, uint8_t *dst
, int dsize
);
72 static int rle_compress(uint8_t **src_p
, int ssize
, uint8_t *dst
, int dsize
);
75 // Create a listening socket
77 // This joins the cluster multi-cast group.
81 struct sockaddr_in addr
;
82 struct sockaddr_in interface_addr
;
87 config
->cluster_undefined_sessions
= MAXSESSION
-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 (setsockopt(cluster_sockfd
, IPPROTO_IP
, IP_ADD_MEMBERSHIP
, &mreq
, sizeof(mreq
)) < 0)
132 LOG(0, 0, 0, "Failed to setsockopt (join mcast group): %s\n", strerror(errno
));
136 if (setsockopt(cluster_sockfd
, IPPROTO_IP
, IP_MULTICAST_IF
, &interface_addr
, sizeof(interface_addr
)) < 0)
138 LOG(0, 0, 0, "Failed to setsockopt (set mcast interface): %s\n", strerror(errno
));
142 config
->cluster_last_hb
= TIME
;
143 config
->cluster_seq_number
= -1;
145 return cluster_sockfd
;
150 // Send a chunk of data to the entire cluster (usually via the multicast
154 static int cluster_send_data(void *data
, int datalen
)
156 struct sockaddr_in addr
= {0};
158 if (!cluster_sockfd
) return -1;
159 if (!config
->cluster_address
) return 0;
161 addr
.sin_addr
.s_addr
= config
->cluster_address
;
162 addr
.sin_port
= htons(CLUSTERPORT
);
163 addr
.sin_family
= AF_INET
;
165 LOG(5, 0, 0, "Cluster send data: %d bytes\n", datalen
);
167 if (sendto(cluster_sockfd
, data
, datalen
, MSG_NOSIGNAL
, (void *) &addr
, sizeof(addr
)) < 0)
169 LOG(0, 0, 0, "sendto: %s\n", strerror(errno
));
177 // Add a chunk of data to a heartbeat packet.
178 // Maintains the format. Assumes that the caller
179 // has passed in a big enough buffer!
181 static void add_type(char **p
, int type
, int more
, char *data
, int size
)
183 *((uint32_t *) (*p
)) = type
;
184 *p
+= sizeof(uint32_t);
186 *((uint32_t *)(*p
)) = more
;
187 *p
+= sizeof(uint32_t);
189 if (data
&& size
> 0) {
190 memcpy(*p
, data
, size
);
195 // advertise our presence via BGP or gratuitous ARP
196 static void advertise_routes(void)
200 bgp_enable_routing(1);
203 if (config
->send_garp
)
204 send_garp(config
->bind_address
); // Start taking traffic.
207 // withdraw our routes (BGP only)
208 static void withdraw_routes(void)
212 bgp_enable_routing(0);
216 static void cluster_uptodate(void)
218 if (config
->cluster_iam_uptodate
)
221 if (config
->cluster_undefined_sessions
|| config
->cluster_undefined_tunnels
)
224 config
->cluster_iam_uptodate
= 1;
226 LOG(0, 0, 0, "Now uptodate with master.\n");
231 // Send a unicast UDP packet to a peer with 'data' as the
234 static int peer_send_data(in_addr_t peer
, char *data
, int size
)
236 struct sockaddr_in addr
= {0};
238 if (!cluster_sockfd
) return -1;
239 if (!config
->cluster_address
) return 0;
244 addr
.sin_addr
.s_addr
= peer
;
245 addr
.sin_port
= htons(CLUSTERPORT
);
246 addr
.sin_family
= AF_INET
;
248 LOG_HEX(5, "Peer send", data
, size
);
250 if (sendto(cluster_sockfd
, data
, size
, MSG_NOSIGNAL
, (void *) &addr
, sizeof(addr
)) < 0)
252 LOG(0, 0, 0, "sendto: %s\n", strerror(errno
));
260 // Send a structured message to a peer with a single element of type 'type'.
262 static int peer_send_message(in_addr_t peer
, int type
, int more
, char *data
, int size
)
264 char buf
[65536]; // Vast overkill.
267 LOG(4, 0, 0, "Sending message to peer (type %d, more %d, size %d)\n", type
, more
, size
);
268 add_type(&p
, type
, more
, data
, size
);
270 return peer_send_data(peer
, buf
, (p
-buf
) );
274 // Forward a state changing packet to the master.
276 // The master just processes the payload as if it had
277 // received it off the tun device.
279 int master_forward_packet(char *data
, int size
, in_addr_t addr
, int port
)
281 char buf
[65536]; // Vast overkill.
284 if (!config
->cluster_master_address
) // No election has been held yet. Just skip it.
287 LOG(4, 0, 0, "Forwarding packet from %s to master (size %d)\n", fmtaddr(addr
, 0), size
);
290 add_type(&p
, C_FORWARD
, addr
, (char *) &port
, sizeof(port
)); // ick. should be uint16_t
291 memcpy(p
, data
, size
);
294 return peer_send_data(config
->cluster_master_address
, buf
, (p
- buf
));
298 // Forward a throttled packet to the master for handling.
300 // The master just drops the packet into the appropriate
301 // token bucket queue, and lets normal processing take care
304 int master_throttle_packet(int tbfid
, char *data
, int size
)
306 char buf
[65536]; // Vast overkill.
309 if (!config
->cluster_master_address
) // No election has been held yet. Just skip it.
312 LOG(4, 0, 0, "Throttling packet master (size %d, tbfid %d)\n", size
, tbfid
);
314 add_type(&p
, C_THROTTLE
, tbfid
, data
, size
);
316 return peer_send_data(config
->cluster_master_address
, buf
, (p
-buf
) );
321 // Forward a walled garden packet to the master for handling.
323 // The master just writes the packet straight to the tun
324 // device (where is will normally loop through the
325 // firewall rules, and come back in on the tun device)
327 // (Note that this must be called with the tun header
328 // as the start of the data).
329 int master_garden_packet(sessionidt s
, char *data
, int size
)
331 char buf
[65536]; // Vast overkill.
334 if (!config
->cluster_master_address
) // No election has been held yet. Just skip it.
337 LOG(4, 0, 0, "Walled garden packet to master (size %d)\n", size
);
339 add_type(&p
, C_GARDEN
, s
, data
, size
);
341 return peer_send_data(config
->cluster_master_address
, buf
, (p
-buf
));
346 // Send a chunk of data as a heartbeat..
347 // We save it in the history buffer as we do so.
349 static void send_heartbeat(int seq
, char *data
, int size
)
353 if (size
> sizeof(past_hearts
[0].data
))
355 LOG(0, 0, 0, "Tried to heartbeat something larger than the maximum packet!\n");
359 i
= seq
% HB_HISTORY_SIZE
;
360 past_hearts
[i
].seq
= seq
;
361 past_hearts
[i
].size
= size
;
362 memcpy(&past_hearts
[i
].data
, data
, size
); // Save it.
363 cluster_send_data(data
, size
);
367 // Send an 'i am alive' message to every machine in the cluster.
369 void cluster_send_ping(time_t basetime
)
371 char buff
[100 + sizeof(pingt
)];
375 if (config
->cluster_iam_master
&& basetime
) // We're heartbeating so no need to ping.
378 LOG(5, 0, 0, "Sending cluster ping...\n");
381 x
.addr
= config
->bind_address
;
382 x
.undef
= config
->cluster_undefined_sessions
+ config
->cluster_undefined_tunnels
;
383 x
.basetime
= basetime
;
385 add_type(&p
, C_PING
, basetime
, (char *) &x
, sizeof(x
));
386 cluster_send_data(buff
, (p
-buff
) );
390 // Walk the session counters looking for non-zero ones to send
391 // to the master. We send up to 600 of them at one time.
392 // We examine a maximum of 3000 sessions.
393 // (50k max session should mean that we normally
394 // examine the entire session table every 25 seconds).
396 #define MAX_B_RECS (600)
397 void master_update_counts(void)
400 bytest b
[MAX_B_RECS
+1];
402 if (config
->cluster_iam_master
) // Only happens on the slaves.
405 if (!config
->cluster_master_address
) // If we don't have a master, skip it for a while.
408 // C_BYTES format changed in 2.1.0 (cluster version 5)
409 // during upgrade from previous versions, hang onto our counters
410 // for a bit until the new master comes up
411 if (config
->cluster_last_hb_ver
< 5)
414 i
= MAX_B_RECS
* 5; // Examine max 3000 sessions;
415 if (config
->cluster_highest_sessionid
> i
)
416 i
= config
->cluster_highest_sessionid
;
418 for ( c
= 0; i
> 0 ; --i
) {
419 // Next session to look at.
420 walk_session_number
++;
421 if ( walk_session_number
> config
->cluster_highest_sessionid
)
422 walk_session_number
= 1;
424 if (!sess_local
[walk_session_number
].cin
&& !sess_local
[walk_session_number
].cout
)
425 continue; // Unchanged. Skip it.
427 b
[c
].sid
= walk_session_number
;
428 b
[c
].pin
= sess_local
[walk_session_number
].pin
;
429 b
[c
].pout
= sess_local
[walk_session_number
].pout
;
430 b
[c
].cin
= sess_local
[walk_session_number
].cin
;
431 b
[c
].cout
= sess_local
[walk_session_number
].cout
;
434 sess_local
[walk_session_number
].pin
= sess_local
[walk_session_number
].pout
= 0;
435 sess_local
[walk_session_number
].cin
= sess_local
[walk_session_number
].cout
= 0;
437 if (++c
> MAX_B_RECS
) // Send a max of 600 elements in a packet.
441 if (!c
) // Didn't find any that changes. Get out of here!
445 // Forward the data to the master.
446 LOG(4, 0, 0, "Sending byte counters to master (%d elements)\n", c
);
447 peer_send_message(config
->cluster_master_address
, C_BYTES
, c
, (char *) &b
, sizeof(b
[0]) * c
);
452 // On the master, check how our slaves are going. If
453 // one of them's not up-to-date we'll heartbeat faster.
454 // If we don't have any of them, then we need to turn
455 // on our own packet handling!
457 void cluster_check_slaves(void)
460 static int have_peers
= 0;
461 int had_peers
= have_peers
;
464 if (!config
->cluster_iam_master
)
465 return; // Only runs on the master...
467 config
->cluster_iam_uptodate
= 1; // cleared in loop below
469 for (i
= have_peers
= 0; i
< num_peers
; i
++)
471 if ((peers
[i
].timestamp
+ config
->cluster_hb_timeout
) < t
)
472 continue; // Stale peer! Skip them.
474 if (!peers
[i
].basetime
)
475 continue; // Shutdown peer! Skip them.
477 if (peers
[i
].uptodate
)
480 config
->cluster_iam_uptodate
= 0; // Start fast heartbeats
483 // in a cluster, withdraw/add routes when we get a peer/lose peers
484 if (have_peers
!= had_peers
)
486 if (had_peers
< config
->cluster_master_min_adv
&&
487 have_peers
>= config
->cluster_master_min_adv
)
490 else if (had_peers
>= config
->cluster_master_min_adv
&&
491 have_peers
< config
->cluster_master_min_adv
)
497 // Check that we have a master. If it's been too
498 // long since we heard from a master then hold an election.
500 void cluster_check_master(void)
502 int i
, count
, tcount
, high_unique_id
= 0;
505 static int probed
= 0;
508 if (config
->cluster_iam_master
)
509 return; // Only runs on the slaves...
511 // If the master is late (missed 2 hearbeats by a second and a
512 // hair) it may be that the switch has dropped us from the
513 // multicast group, try unicasting probes to the master
514 // which will hopefully respond with a unicast heartbeat that
515 // will allow us to limp along until the querier next runs.
516 if (config
->cluster_master_address
517 && TIME
> (config
->cluster_last_hb
+ 2 * config
->cluster_hb_interval
+ 11))
519 if (!probed
|| (TIME
> (probed
+ 2 * config
->cluster_hb_interval
)))
522 LOG(1, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
523 0.1 * (TIME
- (config
->cluster_last_hb
+ config
->cluster_hb_interval
)));
525 peer_send_message(config
->cluster_master_address
,
526 C_LASTSEEN
, config
->cluster_seq_number
, NULL
, 0);
528 } else { // We got a recent heartbeat; reset the probe flag.
532 if (TIME
< (config
->cluster_last_hb
+ config
->cluster_hb_timeout
))
533 return; // Everything's ok!
535 config
->cluster_last_hb
= TIME
+ 1; // Just the one election thanks.
537 LOG(0, 0, 0, "Master timed out! Holding election...\n");
539 // In the process of shutting down, can't be master
543 for (i
= have_peers
= 0; i
< num_peers
; i
++)
545 if ((peers
[i
].timestamp
+ config
->cluster_hb_timeout
) < t
)
546 continue; // Stale peer! Skip them.
548 if (!peers
[i
].basetime
)
549 continue; // Shutdown peer! Skip them.
551 if (peers
[i
].basetime
< basetime
) {
552 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers
[i
].peer
, 0));
553 return; // They'll win the election. Get out of here.
556 if (peers
[i
].basetime
== basetime
&&
557 peers
[i
].peer
> my_address
) {
558 LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers
[i
].peer
, 0));
559 return; // They'll win the election. Wait for them to come up.
562 if (peers
[i
].uptodate
)
566 // Wow. it's been ages since I last heard a heartbeat
567 // and I'm better than an of my peers so it's time
568 // to become a master!!!
570 config
->cluster_iam_master
= 1;
571 config
->cluster_master_address
= 0;
573 LOG(0, 0, 0, "I am declaring myself the master!\n");
575 if (have_peers
< config
->cluster_master_min_adv
)
580 if (config
->cluster_seq_number
== -1)
581 config
->cluster_seq_number
= 0;
584 // Go through and mark all the tunnels as defined.
585 // Count the highest used tunnel number as well.
587 config
->cluster_highest_tunnelid
= 0;
588 for (i
= 0, tcount
= 0; i
< MAXTUNNEL
; ++i
) {
589 if (tunnel
[i
].state
== TUNNELUNDEF
)
590 tunnel
[i
].state
= TUNNELFREE
;
592 if (tunnel
[i
].state
!= TUNNELFREE
&& i
> config
->cluster_highest_tunnelid
)
593 config
->cluster_highest_tunnelid
= i
;
597 // Go through and mark all the sessions as being defined.
598 // reset the idle timeouts.
599 // add temporary byte counters to permanent ones.
600 // Re-string the free list.
601 // Find the ID of the highest session.
604 config
->cluster_highest_sessionid
= 0;
605 for (i
= 0, count
= 0; i
< MAXSESSION
; ++i
) {
606 if (session
[i
].tunnel
== T_UNDEF
) {
607 session
[i
].tunnel
= T_FREE
;
611 if (!session
[i
].opened
) { // Unused session. Add to free list.
612 memset(&session
[i
], 0, sizeof(session
[i
]));
613 session
[i
].tunnel
= T_FREE
;
614 session
[last_free
].next
= i
;
620 // Reset idle timeouts..
621 session
[i
].last_packet
= time_now
;
623 // Reset die relative to our uptime rather than the old master's
624 if (session
[i
].die
) session
[i
].die
= TIME
;
626 // Accumulate un-sent byte/packet counters.
627 increment_counter(&session
[i
].cin
, &session
[i
].cin_wrap
, sess_local
[i
].cin
);
628 increment_counter(&session
[i
].cout
, &session
[i
].cout_wrap
, sess_local
[i
].cout
);
629 session
[i
].cin_delta
+= sess_local
[i
].cin
;
630 session
[i
].cout_delta
+= sess_local
[i
].cout
;
632 session
[i
].pin
+= sess_local
[i
].pin
;
633 session
[i
].pout
+= sess_local
[i
].pout
;
635 sess_local
[i
].cin
= sess_local
[i
].cout
= 0;
636 sess_local
[i
].pin
= sess_local
[i
].pout
= 0;
638 sess_local
[i
].radius
= 0; // Reset authentication as the radius blocks aren't up to date.
640 if (session
[i
].unique_id
>= high_unique_id
) // This is different to the index into the session table!!!
641 high_unique_id
= session
[i
].unique_id
+1;
643 session
[i
].tbf_in
= session
[i
].tbf_out
= 0; // Remove stale pointers from old master.
644 throttle_session(i
, session
[i
].throttle_in
, session
[i
].throttle_out
);
646 config
->cluster_highest_sessionid
= i
;
649 session
[last_free
].next
= 0; // End of chain.
650 last_id
= high_unique_id
; // Keep track of the highest used session ID.
654 rebuild_address_pool();
656 // If we're not the very first master, this is a big issue!
658 LOG(0, 0, 0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count
);
660 config
->cluster_undefined_sessions
= 0;
661 config
->cluster_undefined_tunnels
= 0;
662 config
->cluster_iam_uptodate
= 1; // assume all peers are up-to-date
664 // FIXME. We need to fix up the tunnel control message
665 // queue here! There's a number of other variables we
666 // should also update.
671 // Check that our session table is validly matching what the
672 // master has in mind.
674 // In particular, if we have too many sessions marked 'undefined'
675 // we fix it up here, and we ensure that the 'first free session'
678 static void cluster_check_sessions(int highsession
, int freesession_ptr
, int hightunnel
)
682 sessionfree
= freesession_ptr
; // Keep the freesession ptr valid.
684 if (config
->cluster_iam_uptodate
)
687 if (highsession
> config
->cluster_undefined_sessions
&& hightunnel
> config
->cluster_undefined_tunnels
)
690 // Clear out defined sessions, counting the number of
692 config
->cluster_undefined_sessions
= 0;
693 for (i
= 1 ; i
< MAXSESSION
; ++i
) {
694 if (i
> highsession
) {
695 if (session
[i
].tunnel
== T_UNDEF
) session
[i
].tunnel
= T_FREE
; // Defined.
699 if (session
[i
].tunnel
== T_UNDEF
)
700 ++config
->cluster_undefined_sessions
;
703 // Clear out defined tunnels, counting the number of
705 config
->cluster_undefined_tunnels
= 0;
706 for (i
= 1 ; i
< MAXTUNNEL
; ++i
) {
707 if (i
> hightunnel
) {
708 if (tunnel
[i
].state
== TUNNELUNDEF
) tunnel
[i
].state
= TUNNELFREE
; // Defined.
712 if (tunnel
[i
].state
== TUNNELUNDEF
)
713 ++config
->cluster_undefined_tunnels
;
717 if (config
->cluster_undefined_sessions
|| config
->cluster_undefined_tunnels
) {
718 LOG(2, 0, 0, "Cleared undefined sessions/tunnels. %d sess (high %d), %d tunn (high %d)\n",
719 config
->cluster_undefined_sessions
, highsession
, config
->cluster_undefined_tunnels
, hightunnel
);
723 // Are we up to date?
725 if (!config
->cluster_iam_uptodate
)
729 static int hb_add_type(char **p
, int type
, int id
)
732 case C_CSESSION
: { // Compressed C_SESSION.
733 uint8_t c
[sizeof(sessiont
) * 2]; // Bigger than worst case.
734 uint8_t *d
= (uint8_t *) &session
[id
];
738 size
= rle_compress( &d
, sizeof(sessiont
), c
, sizeof(c
) );
740 // Did we compress the full structure, and is the size actually
742 if ( (d
- orig
) == sizeof(sessiont
) && size
< sizeof(sessiont
) ) {
743 add_type(p
, C_CSESSION
, id
, (char *) c
, size
);
746 // Failed to compress : Fall through.
748 case C_SESSION
: add_type(p
, C_SESSION
, id
,
749 (char *) &session
[id
], sizeof(sessiont
));
752 case C_CTUNNEL
: { // Compressed C_TUNNEL
753 uint8_t c
[sizeof(tunnelt
) * 2]; // Bigger than worst case.
754 uint8_t *d
= (uint8_t *) &tunnel
[id
];
758 size
= rle_compress( &d
, sizeof(tunnelt
), c
, sizeof(c
) );
760 // Did we compress the full structure, and is the size actually
762 if ( (d
- orig
) == sizeof(tunnelt
) && size
< sizeof(tunnelt
) ) {
763 add_type(p
, C_CTUNNEL
, id
, c
, size
);
766 // Failed to compress : Fall through.
768 case C_TUNNEL
: add_type(p
, C_TUNNEL
, id
,
769 (char *) &tunnel
[id
], sizeof(tunnelt
));
772 LOG(0, 0, 0, "Found an invalid type in heart queue! (%d)\n", type
);
780 // Send a heartbeat, incidently sending out any queued changes..
782 void cluster_heartbeat()
784 int i
, count
= 0, tcount
= 0;
785 char buff
[MAX_HEART_SIZE
+ sizeof(heartt
) + sizeof(int) ];
789 if (!config
->cluster_iam_master
) // Only the master does this.
792 config
->cluster_table_version
+= config
->cluster_num_changes
;
794 // Fill out the heartbeat header.
795 memset(&h
, 0, sizeof(h
));
797 h
.version
= HB_VERSION
;
798 h
.seq
= config
->cluster_seq_number
;
799 h
.basetime
= basetime
;
800 h
.clusterid
= config
->bind_address
; // Will this do??
801 h
.basetime
= basetime
;
802 h
.highsession
= config
->cluster_highest_sessionid
;
803 h
.freesession
= sessionfree
;
804 h
.hightunnel
= config
->cluster_highest_tunnelid
;
805 h
.size_sess
= sizeof(sessiont
); // Just in case.
806 h
.size_tunn
= sizeof(tunnelt
);
807 h
.interval
= config
->cluster_hb_interval
;
808 h
.timeout
= config
->cluster_hb_timeout
;
809 h
.table_version
= config
->cluster_table_version
;
811 add_type(&p
, C_HEARTBEAT
, HB_VERSION
, (char *) &h
, sizeof(h
));
813 for (i
= 0; i
< config
->cluster_num_changes
; ++i
) {
814 hb_add_type(&p
, cluster_changes
[i
].type
, cluster_changes
[i
].id
);
817 if (p
> (buff
+ sizeof(buff
))) { // Did we somehow manage to overun the buffer?
818 LOG(0, 0, 0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", (int) (p
- buff
));
824 // Fill out the packet with sessions from the session table...
825 // (not forgetting to leave space so we can get some tunnels in too )
826 while ( (p
+ sizeof(uint32_t) * 2 + sizeof(sessiont
) * 2 ) < (buff
+ MAX_HEART_SIZE
) ) {
828 if (!walk_session_number
) // session #0 isn't valid.
829 ++walk_session_number
;
831 if (count
>= config
->cluster_highest_sessionid
) // If we're a small cluster, don't go wild.
834 hb_add_type(&p
, C_CSESSION
, walk_session_number
);
835 walk_session_number
= (1+walk_session_number
)%(config
->cluster_highest_sessionid
+1); // +1 avoids divide by zero.
837 ++count
; // Count the number of extra sessions we're sending.
841 // Fill out the packet with tunnels from the tunnel table...
842 // This effectively means we walk the tunnel table more quickly
843 // than the session table. This is good because stuffing up a
844 // tunnel is a much bigger deal than stuffing up a session.
846 while ( (p
+ sizeof(uint32_t) * 2 + sizeof(tunnelt
) ) < (buff
+ MAX_HEART_SIZE
) ) {
848 if (!walk_tunnel_number
) // tunnel #0 isn't valid.
849 ++walk_tunnel_number
;
851 if (tcount
>= config
->cluster_highest_tunnelid
)
854 hb_add_type(&p
, C_CTUNNEL
, walk_tunnel_number
);
855 walk_tunnel_number
= (1+walk_tunnel_number
)%(config
->cluster_highest_tunnelid
+1); // +1 avoids divide by zero.
861 // Did we do something wrong?
862 if (p
> (buff
+ sizeof(buff
))) { // Did we somehow manage to overun the buffer?
863 LOG(0, 0, 0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", (int) (p
- buff
));
868 LOG(3, 0, 0, "Sending v%d heartbeat #%d, change #%" PRIu64
" with %d changes "
869 "(%d x-sess, %d x-tunnels, %d highsess, %d hightun, size %d)\n",
870 HB_VERSION
, h
.seq
, h
.table_version
, config
->cluster_num_changes
,
871 count
, tcount
, config
->cluster_highest_sessionid
,
872 config
->cluster_highest_tunnelid
, (int) (p
- buff
));
874 config
->cluster_num_changes
= 0;
876 send_heartbeat(h
.seq
, buff
, (p
-buff
) ); // Send out the heartbeat to the cluster, keeping a copy of it.
878 config
->cluster_seq_number
= (config
->cluster_seq_number
+1)%HB_MAX_SEQ
; // Next seq number to use.
882 // A structure of type 'type' has changed; Add it to the queue to send.
884 static int type_changed(int type
, int id
)
888 for (i
= 0 ; i
< config
->cluster_num_changes
; ++i
)
889 if ( cluster_changes
[i
].id
== id
&&
890 cluster_changes
[i
].type
== type
)
891 return 0; // Already marked for change.
893 cluster_changes
[i
].type
= type
;
894 cluster_changes
[i
].id
= id
;
895 ++config
->cluster_num_changes
;
897 if (config
->cluster_num_changes
> MAX_CHANGES
)
898 cluster_heartbeat(); // flush now
904 // A particular session has been changed!
905 int cluster_send_session(int sid
)
907 if (!config
->cluster_iam_master
) {
908 LOG(0, sid
, 0, "I'm not a master, but I just tried to change a session!\n");
913 LOG(0, sid
, 0, "cluster_send_session called from child process!\n");
917 return type_changed(C_CSESSION
, sid
);
920 // A particular tunnel has been changed!
921 int cluster_send_tunnel(int tid
)
923 if (!config
->cluster_iam_master
) {
924 LOG(0, 0, tid
, "I'm not a master, but I just tried to change a tunnel!\n");
928 return type_changed(C_CTUNNEL
, tid
);
933 // We're a master, and a slave has just told us that it's
934 // missed a packet. We'll resend it every packet since
935 // the last one it's seen.
937 static int cluster_catchup_slave(int seq
, in_addr_t slave
)
942 LOG(1, 0, 0, "Slave %s sent LASTSEEN with seq %d\n", fmtaddr(slave
, 0), seq
);
943 if (!config
->cluster_iam_master
) {
944 LOG(1, 0, 0, "Got LASTSEEN but I'm not a master! Redirecting it to %s.\n",
945 fmtaddr(config
->cluster_master_address
, 0));
947 peer_send_message(slave
, C_MASTER
, config
->cluster_master_address
, NULL
, 0);
951 diff
= config
->cluster_seq_number
- seq
; // How many packet do we need to send?
955 if (diff
>= HB_HISTORY_SIZE
) { // Ouch. We don't have the packet to send it!
956 LOG(0, 0, 0, "A slave asked for message %d when our seq number is %d. Killing it.\n",
957 seq
, config
->cluster_seq_number
);
958 return peer_send_message(slave
, C_KILL
, seq
, NULL
, 0);// Kill the slave. Nothing else to do.
961 LOG(1, 0, 0, "Sending %d catchup packets to slave %s\n", diff
, fmtaddr(slave
, 0) );
963 // Now resend every packet that it missed, in order.
964 while (seq
!= config
->cluster_seq_number
) {
965 s
= seq
% HB_HISTORY_SIZE
;
966 if (seq
!= past_hearts
[s
].seq
) {
967 LOG(0, 0, 0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n",
968 fmtaddr(slave
, 0), seq
, past_hearts
[s
].seq
, s
, config
->cluster_seq_number
);
969 return -1; // What to do here!?
971 peer_send_data(slave
, past_hearts
[s
].data
, past_hearts
[s
].size
);
972 seq
= (seq
+1)%HB_MAX_SEQ
; // Increment to next seq number.
974 return 0; // All good!
978 // We've heard from another peer! Add it to the list
979 // that we select from at election time.
981 static int cluster_add_peer(in_addr_t peer
, time_t basetime
, pingt
*pp
, int size
)
987 // Allow for backward compatability.
988 // Just the ping packet into a new structure to allow
989 // for the possibility that we might have received
990 // more or fewer elements than we were expecting.
991 if (size
> sizeof(p
))
994 memset( (void *) &p
, 0, sizeof(p
) );
995 memcpy( (void *) &p
, (void *) pp
, size
);
998 if (clusterid
!= config
->bind_address
)
1001 LOG(4, 0, 0, "Skipping ping from %s (different cluster)\n", fmtaddr(peer
, 0));
1005 for (i
= 0; i
< num_peers
; ++i
)
1007 if (peers
[i
].peer
!= peer
)
1010 // This peer already exists. Just update the timestamp.
1011 peers
[i
].basetime
= basetime
;
1012 peers
[i
].timestamp
= TIME
;
1013 peers
[i
].uptodate
= !p
.undef
;
1017 // Is this the master shutting down??
1018 if (peer
== config
->cluster_master_address
) {
1019 LOG(3, 0, 0, "Master %s %s\n", fmtaddr(config
->cluster_master_address
, 0),
1020 basetime
? "has restarted!" : "shutting down...");
1022 config
->cluster_master_address
= 0;
1023 config
->cluster_last_hb
= 0; // Force an election.
1024 cluster_check_master();
1030 LOG(4, 0, 0, "Adding %s as a peer\n", fmtaddr(peer
, 0));
1032 // Not found. Is there a stale slot to re-use?
1033 for (i
= 0; i
< num_peers
; ++i
)
1035 if (!peers
[i
].basetime
) // Shutdown
1038 if ((peers
[i
].timestamp
+ config
->cluster_hb_timeout
* 10) < TIME
) // Stale.
1042 if (i
>= CLUSTER_MAX_SIZE
)
1045 LOG(0, 0, 0, "Tried to add %s as a peer, but I already have %d of them!\n", fmtaddr(peer
, 0), i
);
1049 peers
[i
].peer
= peer
;
1050 peers
[i
].basetime
= basetime
;
1051 peers
[i
].timestamp
= TIME
;
1052 peers
[i
].uptodate
= !p
.undef
;
1056 LOG(1, 0, 0, "Added %s as a new peer. Now %d peers\n", fmtaddr(peer
, 0), num_peers
);
1062 // A slave responds with C_MASTER when it gets a message which should have gone to a master.
1063 static int cluster_set_master(in_addr_t peer
, in_addr_t master
)
1065 if (config
->cluster_iam_master
) // Sanity...
1068 LOG(3, 0, 0, "Peer %s set the master to %s...\n", fmtaddr(peer
, 0),
1069 fmtaddr(master
, 1));
1071 config
->cluster_master_address
= master
;
1072 cluster_check_master();
1076 /* Handle the slave updating the byte counters for the master. */
1078 // Note that we don't mark the session as dirty; We rely on
1079 // the slow table walk to propogate this back out to the slaves.
1081 static int cluster_handle_bytes(char *data
, int size
)
1085 b
= (bytest
*) data
;
1087 LOG(3, 0, 0, "Got byte counter update (size %d)\n", size
);
1089 /* Loop around, adding the byte
1090 counts to each of the sessions. */
1092 while (size
>= sizeof(*b
) ) {
1093 if (b
->sid
> MAXSESSION
) {
1094 LOG(0, 0, 0, "Got C_BYTES with session #%d!\n", b
->sid
);
1095 return -1; /* Abort processing */
1098 session
[b
->sid
].pin
+= b
->pin
;
1099 session
[b
->sid
].pout
+= b
->pout
;
1101 increment_counter(&session
[b
->sid
].cin
, &session
[b
->sid
].cin_wrap
, b
->cin
);
1102 increment_counter(&session
[b
->sid
].cout
, &session
[b
->sid
].cout_wrap
, b
->cout
);
1104 session
[b
->sid
].cin_delta
+= b
->cin
;
1105 session
[b
->sid
].cout_delta
+= b
->cout
;
1108 session
[b
->sid
].last_packet
= time_now
; // Reset idle timer!
1115 LOG(0, 0, 0, "Got C_BYTES with %d bytes of trailing junk!\n", size
);
1121 // Handle receiving a session structure in a heartbeat packet.
1123 static int cluster_recv_session(int more
, uint8_t *p
)
1125 if (more
>= MAXSESSION
) {
1126 LOG(0, 0, 0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
1130 if (session
[more
].tunnel
== T_UNDEF
) {
1131 if (config
->cluster_iam_uptodate
) { // Sanity.
1132 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined session!\n");
1134 --config
->cluster_undefined_sessions
;
1138 load_session(more
, (sessiont
*) p
); // Copy session into session table..
1140 LOG(5, more
, 0, "Received session update (%d undef)\n", config
->cluster_undefined_sessions
);
1142 if (!config
->cluster_iam_uptodate
)
1143 cluster_uptodate(); // Check to see if we're up to date.
1148 static int cluster_recv_tunnel(int more
, uint8_t *p
)
1150 if (more
>= MAXTUNNEL
) {
1151 LOG(0, 0, 0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
1155 if (tunnel
[more
].state
== TUNNELUNDEF
) {
1156 if (config
->cluster_iam_uptodate
) { // Sanity.
1157 LOG(0, 0, 0, "I thought I was uptodate but I just found an undefined tunnel!\n");
1159 --config
->cluster_undefined_tunnels
;
1163 memcpy(&tunnel
[more
], p
, sizeof(tunnel
[more
]) );
1166 // Clear tunnel control messages. These are dynamically allocated.
1167 // If we get unlucky, this may cause the tunnel to drop!
1169 tunnel
[more
].controls
= tunnel
[more
].controle
= NULL
;
1170 tunnel
[more
].controlc
= 0;
1172 LOG(5, 0, more
, "Received tunnel update\n");
1174 if (!config
->cluster_iam_uptodate
)
1175 cluster_uptodate(); // Check to see if we're up to date.
1181 // pre v5 heartbeat session structure
1188 unsigned long unique_id
;
1195 uint32_t total_cout
;
1197 uint16_t throttle_in
;
1198 uint16_t throttle_out
;
1202 in_addr_t dns1
, dns2
;
1203 routet route
[MAXROUTE
];
1209 uint8_t reserved_old_snoop
;
1210 uint8_t walled_garden
;
1212 char random_vector
[MAXTEL
];
1213 int random_vector_length
;
1215 char called
[MAXTEL
];
1216 char calling
[MAXTEL
];
1217 uint32_t tx_connect_speed
;
1218 uint32_t rx_connect_speed
;
1221 uint16_t snoop_port
;
1228 static uint8_t *convert_session(struct oldsession
*old
)
1230 static sessiont
new;
1233 memset(&new, 0, sizeof(new));
1235 new.next
= old
->next
;
1237 new.tunnel
= old
->tunnel
;
1238 new.l2tp_flags
= old
->l2tp_flags
;
1239 new.flags
= old
->flags
;
1241 new.ip_pool_index
= old
->ip_pool_index
;
1242 new.unique_id
= old
->unique_id
;
1245 new.magic
= old
->magic
;
1247 new.pout
= old
->pout
;
1248 new.cin
= old
->total_cin
;
1249 new.cout
= old
->total_cout
;
1250 new.cin_delta
= old
->cin
;
1251 new.cout_delta
= old
->cout
;
1252 new.throttle_in
= old
->throttle_in
;
1253 new.throttle_out
= old
->throttle_out
;
1254 new.filter_in
= old
->filter_in
;
1255 new.filter_out
= old
->filter_out
;
1257 new.opened
= old
->opened
;
1259 new.last_packet
= old
->last_packet
;
1260 new.dns1
= old
->dns1
;
1261 new.dns2
= old
->dns2
;
1262 new.tbf_in
= old
->tbf_in
;
1263 new.tbf_out
= old
->tbf_out
;
1264 new.random_vector_length
= old
->random_vector_length
;
1265 new.tx_connect_speed
= old
->tx_connect_speed
;
1266 new.rx_connect_speed
= old
->rx_connect_speed
;
1267 new.snoop_ip
= old
->snoop_ip
;
1268 new.snoop_port
= old
->snoop_port
;
1269 new.walled_garden
= old
->walled_garden
;
1271 memcpy(new.random_vector
, old
->random_vector
, sizeof(new.random_vector
));
1272 memcpy(new.user
, old
->user
, sizeof(new.user
));
1273 memcpy(new.called
, old
->called
, sizeof(new.called
));
1274 memcpy(new.calling
, old
->calling
, sizeof(new.calling
));
1276 for (i
= 0; i
< MAXROUTE
; i
++)
1277 memcpy(&new.route
[i
], &old
->route
[i
], sizeof(new.route
[i
]));
1279 return (uint8_t *) &new;
1283 // Process a heartbeat..
1285 // v3: added interval, timeout
1286 // v4: added table_version
1287 // v5: added ipv6, re-ordered session structure
1288 static int cluster_process_heartbeat(uint8_t *data
, int size
, int more
, uint8_t *p
, in_addr_t addr
)
1291 int s
= size
- (p
-data
);
1296 # error "need to update cluster_process_heartbeat()"
1299 // we handle versions 3 through 5
1300 if (hb_ver
< 3 || hb_ver
> HB_VERSION
) {
1301 LOG(0, 0, 0, "Received a heartbeat version that I don't support (%d)!\n", hb_ver
);
1302 return -1; // Ignore it??
1305 // Ok. It's a heartbeat packet from a cluster master!
1313 if (h
->clusterid
!= config
->bind_address
)
1314 return -1; // It's not part of our cluster.
1316 if (config
->cluster_iam_master
) { // Sanity...
1317 // Note that this MUST match the election process above!
1319 LOG(0, 0, 0, "I just got a heartbeat from master %s, but _I_ am the master!\n", fmtaddr(addr
, 0));
1321 LOG(0, 0, 0, "Heartbeat with zero basetime! Ignoring\n");
1322 return -1; // Skip it.
1326 if (h
->table_version
> config
->cluster_table_version
) {
1327 LOG(0, 0, 0, "They've seen more state changes (%" PRIu64
" vs my %" PRIu64
") so I'm gone!\n",
1328 h
->table_version
, config
->cluster_table_version
);
1333 if (h
->table_version
< config
->cluster_table_version
)
1337 if (basetime
> h
->basetime
) {
1338 LOG(0, 0, 0, "They're an older master than me so I'm gone!\n");
1343 if (basetime
< h
->basetime
)
1346 if (my_address
< addr
) { // Tie breaker.
1347 LOG(0, 0, 0, "They're a higher IP address than me, so I'm gone!\n");
1353 // Send it a unicast heartbeat to see give it a chance to die.
1354 // NOTE: It's actually safe to do seq-number - 1 without checking
1357 cluster_catchup_slave(config
->cluster_seq_number
- 1, addr
);
1359 return -1; // Skip it.
1363 // Try and guard against a stray master appearing.
1365 // Ignore heartbeats received from another master before the
1366 // timeout (less a smidgen) for the old master has elapsed.
1368 // Note that after a clean failover, the cluster_master_address
1369 // is cleared, so this doesn't run.
1371 if (config
->cluster_master_address
&& addr
!= config
->cluster_master_address
1372 && (config
->cluster_last_hb
+ config
->cluster_hb_timeout
- 11) > TIME
) {
1373 LOG(0, 0, 0, "Ignoring stray heartbeat from %s, current master %s has not yet timed out (last heartbeat %.1f seconds ago).\n",
1374 fmtaddr(addr
, 0), fmtaddr(config
->cluster_master_address
, 1),
1375 0.1 * (TIME
- config
->cluster_last_hb
));
1376 return -1; // ignore
1379 if (config
->cluster_seq_number
== -1) // Don't have one. Just align to the master...
1380 config
->cluster_seq_number
= h
->seq
;
1382 config
->cluster_last_hb
= TIME
; // Reset to ensure that we don't become master!!
1383 config
->cluster_last_hb_ver
= hb_ver
; // remember what cluster version the master is using
1385 if (config
->cluster_seq_number
!= h
->seq
) { // Out of sequence heartbeat!
1386 static int lastseen_seq
= 0;
1387 static time_t lastseen_time
= 0;
1389 // limit to once per second for a particular seq#
1390 int ask
= (config
->cluster_seq_number
!= lastseen_seq
|| time_now
!= lastseen_time
);
1392 LOG(1, 0, 0, "HB: Got seq# %d but was expecting %d. %s.\n",
1393 h
->seq
, config
->cluster_seq_number
,
1394 ask
? "Asking for resend" : "Ignoring");
1398 lastseen_seq
= config
->cluster_seq_number
;
1399 lastseen_time
= time_now
;
1400 peer_send_message(addr
, C_LASTSEEN
, config
->cluster_seq_number
, NULL
, 0);
1403 config
->cluster_last_hb
= TIME
; // Reset to ensure that we don't become master!!
1405 // Just drop the packet. The master will resend it as part of the catchup.
1409 // Save the packet in our buffer.
1410 // This is needed in case we become the master.
1411 config
->cluster_seq_number
= (h
->seq
+1)%HB_MAX_SEQ
;
1412 i
= h
->seq
% HB_HISTORY_SIZE
;
1413 past_hearts
[i
].seq
= h
->seq
;
1414 past_hearts
[i
].size
= size
;
1415 memcpy(&past_hearts
[i
].data
, data
, size
); // Save it.
1418 // Check that we don't have too many undefined sessions, and
1419 // that the free session pointer is correct.
1420 cluster_check_sessions(h
->highsession
, h
->freesession
, h
->hightunnel
);
1422 if (h
->interval
!= config
->cluster_hb_interval
)
1424 LOG(2, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
1425 h
->interval
, config
->cluster_hb_interval
);
1427 config
->cluster_hb_interval
= h
->interval
;
1430 if (h
->timeout
!= config
->cluster_hb_timeout
)
1432 LOG(2, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
1433 h
->timeout
, config
->cluster_hb_timeout
);
1435 config
->cluster_hb_timeout
= h
->timeout
;
1438 // Ok. process the packet...
1441 type
= *((uint32_t *) p
);
1442 p
+= sizeof(uint32_t);
1443 s
-= sizeof(uint32_t);
1445 more
= *((uint32_t *) p
);
1446 p
+= sizeof(uint32_t);
1447 s
-= sizeof(uint32_t);
1450 case C_CSESSION
: { // Compressed session structure.
1451 uint8_t c
[ sizeof(sessiont
) + 2];
1453 uint8_t *orig_p
= p
;
1455 size
= rle_decompress((uint8_t **) &p
, s
, c
, sizeof(c
) );
1458 // session struct changed with v5
1461 if (size
!= sizeof(struct oldsession
)) {
1462 LOG(0, 0, 0, "DANGER: Received a v%d CSESSION that didn't decompress correctly!\n", hb_ver
);
1463 // Now what? Should exit! No-longer up to date!
1466 cluster_recv_session(more
, convert_session((struct oldsession
*) c
));
1470 if (size
!= sizeof(sessiont
) ) { // Ouch! Very very bad!
1471 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1472 // Now what? Should exit! No-longer up to date!
1476 cluster_recv_session(more
, c
);
1482 if (s
< sizeof(struct oldsession
))
1485 cluster_recv_session(more
, convert_session((struct oldsession
*) p
));
1487 p
+= sizeof(struct oldsession
);
1488 s
-= sizeof(struct oldsession
);
1492 if ( s
< sizeof(session
[more
]))
1495 cluster_recv_session(more
, p
);
1497 p
+= sizeof(session
[more
]);
1498 s
-= sizeof(session
[more
]);
1501 case C_CTUNNEL
: { // Compressed tunnel structure.
1502 uint8_t c
[ sizeof(tunnelt
) + 2];
1504 uint8_t *orig_p
= p
;
1506 size
= rle_decompress((uint8_t **) &p
, s
, c
, sizeof(c
));
1509 if (size
!= sizeof(tunnelt
) ) { // Ouch! Very very bad!
1510 LOG(0, 0, 0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1511 // Now what? Should exit! No-longer up to date!
1515 cluster_recv_tunnel(more
, c
);
1520 if ( s
< sizeof(tunnel
[more
]))
1523 cluster_recv_tunnel(more
, p
);
1525 p
+= sizeof(tunnel
[more
]);
1526 s
-= sizeof(tunnel
[more
]);
1529 LOG(0, 0, 0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type
);
1530 return -1; // can't process any more of the packet!!
1534 if (config
->cluster_master_address
!= addr
)
1536 LOG(0, 0, 0, "My master just changed from %s to %s!\n",
1537 fmtaddr(config
->cluster_master_address
, 0), fmtaddr(addr
, 1));
1539 config
->cluster_master_address
= addr
;
1542 config
->cluster_last_hb
= TIME
; // Successfully received a heartbeat!
1543 config
->cluster_table_version
= h
->table_version
;
1547 LOG(0, 0, 0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
1552 // We got a packet on the cluster port!
1553 // Handle pings, lastseens, and heartbeats!
1555 int processcluster(char *data
, int size
, in_addr_t addr
)
1561 if (addr
== my_address
)
1562 return -1; // Ignore it. Something looped back the multicast!
1564 LOG(5, 0, 0, "Process cluster: %d bytes from %s\n", size
, fmtaddr(addr
, 0));
1566 if (s
<= 0) // Any data there??
1572 type
= *((uint32_t *) p
);
1573 p
+= sizeof(uint32_t);
1574 s
-= sizeof(uint32_t);
1576 more
= *((uint32_t *) p
);
1577 p
+= sizeof(uint32_t);
1578 s
-= sizeof(uint32_t);
1581 case C_PING
: // Update the peers table.
1582 return cluster_add_peer(addr
, more
, (pingt
*) p
, s
);
1584 case C_MASTER
: // Our master is wrong
1585 return cluster_set_master(addr
, more
);
1587 case C_LASTSEEN
: // Catch up a slave (slave missed a packet).
1588 return cluster_catchup_slave(more
, addr
);
1590 case C_FORWARD
: { // Forwarded control packet. pass off to processudp.
1591 struct sockaddr_in a
;
1592 a
.sin_addr
.s_addr
= more
;
1594 a
.sin_port
= *(int *) p
;
1598 if (!config
->cluster_iam_master
) { // huh?
1599 LOG(0, 0, 0, "I'm not the master, but I got a C_FORWARD from %s?\n", fmtaddr(addr
, 0));
1603 LOG(4, 0, 0, "Got a forwarded packet... (%s:%d)\n", fmtaddr(more
, 0), a
.sin_port
);
1605 processudp(p
, s
, &a
);
1608 case C_THROTTLE
: { // Receive a forwarded packet from a slave.
1609 if (!config
->cluster_iam_master
) {
1610 LOG(0, 0, 0, "I'm not the master, but I got a C_THROTTLE from %s?\n", fmtaddr(addr
, 0));
1614 tbf_queue_packet(more
, p
, s
); // The TBF id tells wether it goes in or out.
1618 // Receive a walled garden packet from a slave.
1619 if (!config
->cluster_iam_master
) {
1620 LOG(0, 0, 0, "I'm not the master, but I got a C_GARDEN from %s?\n", fmtaddr(addr
, 0));
1628 if (!config
->cluster_iam_master
) {
1629 LOG(0, 0, 0, "I'm not the master, but I got a C_BYTES from %s?\n", fmtaddr(addr
, 0));
1633 return cluster_handle_bytes(p
, s
);
1635 case C_KILL
: // The master asked us to die!? (usually because we're too out of date).
1636 if (config
->cluster_iam_master
) {
1637 LOG(0, 0, 0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", fmtaddr(addr
, 0), more
);
1640 if (more
!= config
->cluster_seq_number
) {
1641 LOG(0, 0, 0, "The master asked us to die but the seq number didn't match!?\n");
1645 if (addr
!= config
->cluster_master_address
) {
1646 LOG(0, 0, 0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%s)\n",
1647 fmtaddr(addr
, 0), fmtaddr(config
->cluster_master_address
, 1));
1648 // We can only warn about it. The master might really have switched!
1651 LOG(0, 0, 0, "Received a valid C_KILL: I'm going to die now.\n");
1653 exit(0); // Lets be paranoid;
1654 return -1; // Just signalling the compiler.
1657 LOG(4, 0, 0, "Got a heartbeat from %s\n", fmtaddr(addr
, 0));
1658 return cluster_process_heartbeat(data
, size
, more
, p
, addr
);
1661 LOG(0, 0, 0, "Strange type packet received on cluster socket (%d)\n", type
);
1667 LOG(0, 0, 0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n");
1671 //====================================================================================================
1673 int cmd_show_cluster(struct cli_def
*cli
, char *command
, char **argv
, int argc
)
1677 if (CLI_HELP_REQUESTED
)
1678 return CLI_HELP_NO_ARGS
;
1680 cli_print(cli
, "Cluster status : %s", config
->cluster_iam_master
? "Master" : "Slave" );
1681 cli_print(cli
, "My address : %s", fmtaddr(my_address
, 0));
1682 cli_print(cli
, "VIP address : %s", fmtaddr(config
->bind_address
, 0));
1683 cli_print(cli
, "Multicast address: %s", fmtaddr(config
->cluster_address
, 0));
1684 cli_print(cli
, "Multicast i'face : %s", config
->cluster_interface
);
1686 if (!config
->cluster_iam_master
) {
1687 cli_print(cli
, "My master : %s (last heartbeat %.1f seconds old)",
1688 config
->cluster_master_address
1689 ? fmtaddr(config
->cluster_master_address
, 0)
1691 0.1 * (TIME
- config
->cluster_last_hb
));
1692 cli_print(cli
, "Uptodate : %s", config
->cluster_iam_uptodate
? "Yes" : "No");
1693 cli_print(cli
, "Table version # : %" PRIu64
, config
->cluster_table_version
);
1694 cli_print(cli
, "Next sequence number expected: %d", config
->cluster_seq_number
);
1695 cli_print(cli
, "%d sessions undefined of %d", config
->cluster_undefined_sessions
, config
->cluster_highest_sessionid
);
1696 cli_print(cli
, "%d tunnels undefined of %d", config
->cluster_undefined_tunnels
, config
->cluster_highest_tunnelid
);
1698 cli_print(cli
, "Table version # : %" PRIu64
, config
->cluster_table_version
);
1699 cli_print(cli
, "Next heartbeat # : %d", config
->cluster_seq_number
);
1700 cli_print(cli
, "Highest session : %d", config
->cluster_highest_sessionid
);
1701 cli_print(cli
, "Highest tunnel : %d", config
->cluster_highest_tunnelid
);
1702 cli_print(cli
, "%d changes queued for sending", config
->cluster_num_changes
);
1704 cli_print(cli
, "%d peers.", num_peers
);
1707 cli_print(cli
, "%20s %10s %8s", "Address", "Basetime", "Age");
1708 for (i
= 0; i
< num_peers
; ++i
) {
1709 cli_print(cli
, "%20s %10u %8d", fmtaddr(peers
[i
].peer
, 0),
1710 peers
[i
].basetime
, TIME
- peers
[i
].timestamp
);
1716 // Simple run-length-encoding compression.
1718 // 1 byte < 128 = count of non-zero bytes following. // Not legal to be zero.
1719 // n non-zero bytes;
1721 // 1 byte > 128 = (count - 128) run of zero bytes. //
1723 // count == 0 indicates end of compressed stream.
1725 // Compress from 'src' into 'dst'. return number of bytes
1727 // Updates *src_p to indicate 1 past last bytes used.
1729 // We could get an extra byte in the zero runs by storing (count-1)
1730 // but I'm playing it safe.
1732 // Worst case is a 50% expansion in space required (trying to
1733 // compress { 0x00, 0x01 } * N )
1734 static int rle_compress(uint8_t **src_p
, int ssize
, uint8_t *dst
, int dsize
)
1737 int orig_dsize
= dsize
;
1741 while (ssize
> 0 && dsize
> 2) {
1743 x
= dst
++; --dsize
; // Reserve space for count byte..
1745 if (*src
) { // Copy a run of non-zero bytes.
1746 while (*src
&& count
< 127 && ssize
> 0 && dsize
> 1) { // Count number of non-zero bytes.
1751 *x
= count
; // Store number of non-zero bytes. Guarenteed to be non-zero!
1753 } else { // Compress a run of zero bytes.
1754 while (*src
== 0 && count
< 127 && ssize
> 0) {
1763 *dst
++ = 0x0; // Add Stop byte.
1767 return (orig_dsize
- dsize
);
1771 // Decompress the buffer into **p.
1772 // 'psize' is the size of the decompression buffer available.
1774 // Returns the number of bytes decompressed.
1776 // Decompresses from '*src_p' into 'dst'.
1777 // Return the number of dst bytes used.
1778 // Updates the 'src_p' pointer to point to the
1779 // first un-used byte.
1780 static int rle_decompress(uint8_t **src_p
, int ssize
, uint8_t *dst
, int dsize
)
1783 int orig_dsize
= dsize
;
1786 while (ssize
>0 && dsize
> 0) { // While there's more to decompress, and there's room in the decompress buffer...
1787 count
= *src
++; --ssize
; // get the count byte from the source.
1788 if (count
== 0x0) // End marker reached? If so, finish.
1791 if (count
& 0x80) { // Decompress a run of zeros
1792 for (count
&= 0x7f ; count
> 0 && dsize
> 0; --count
) {
1796 } else { // Copy run of non-zero bytes.
1797 for ( ; count
> 0 && ssize
&& dsize
; --count
) { // Copy non-zero bytes across.
1804 return (orig_dsize
- dsize
);