1 // L2TPNS Clustering Stuff
3 char const *cvs_id_cluster
= "$Id: cluster.c,v 1.18 2004-11-16 07:54:32 bodea Exp $";
8 #include <sys/socket.h>
9 #include <netinet/in.h>
10 #include <arpa/inet.h>
11 #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 int cluster_sockfd
= 0; // The filedescriptor for the cluster communications port.
43 ipt my_address
= 0; // The network address of my ethernet port.
44 static int walk_session_number
= 0; // The next session to send when doing the slow table walk.
45 static int walk_tunnel_number
= 0; // The next tunnel to send when doing the slow table walk.
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 char data
[MAX_HEART_SIZE
];
59 } past_hearts
[HB_HISTORY_SIZE
]; // Ring buffer of heartbeats that we've recently sent out. Needed so
60 // we can re-transmit if needed.
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(u8
** src_p
, int ssize
, u8
*dst
, int dsize
);
71 static int rle_compress(u8
** src_p
, int ssize
, u8
*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_tunnels
= MAXTUNNEL
-1;
89 if (!config
->cluster_address
)
91 if (!*config
->cluster_interface
)
94 cluster_sockfd
= socket(AF_INET
, SOCK_DGRAM
, IPPROTO_UDP
);
96 memset(&addr
, 0, sizeof(addr
));
97 addr
.sin_family
= AF_INET
;
98 addr
.sin_port
= htons(CLUSTERPORT
);
99 addr
.sin_addr
.s_addr
= INADDR_ANY
;
100 setsockopt(cluster_sockfd
, SOL_SOCKET
, SO_REUSEADDR
, &addr
, sizeof(addr
));
102 if (bind(cluster_sockfd
, (void *) &addr
, sizeof(addr
)) < 0)
104 LOG(0, 0, 0, 0, "Failed to bind cluster socket: %s\n", strerror(errno
));
108 strcpy(ifr
.ifr_name
, config
->cluster_interface
);
109 if (ioctl(cluster_sockfd
, SIOCGIFADDR
, &ifr
) < 0)
111 LOG(0, 0, 0, 0, "Failed to get interface address for (%s): %s\n", config
->cluster_interface
, strerror(errno
));
115 memcpy(&interface_addr
, &ifr
.ifr_addr
, sizeof(interface_addr
));
116 my_address
= interface_addr
.sin_addr
.s_addr
;
118 // Join multicast group.
119 mreq
.imr_multiaddr
.s_addr
= config
->cluster_address
;
120 mreq
.imr_interface
= interface_addr
.sin_addr
;
123 opt
= 0; // Turn off multicast loopback.
124 setsockopt(cluster_sockfd
, IPPROTO_IP
, IP_MULTICAST_LOOP
, &opt
, sizeof(opt
));
126 if (setsockopt(cluster_sockfd
, IPPROTO_IP
, IP_ADD_MEMBERSHIP
, &mreq
, sizeof(mreq
)) < 0)
128 LOG(0, 0, 0, 0, "Failed to setsockopt (join mcast group): %s\n", strerror(errno
));
132 if (setsockopt (cluster_sockfd
, IPPROTO_IP
, IP_MULTICAST_IF
, &interface_addr
, sizeof(interface_addr
)) < 0)
134 LOG(0, 0, 0, 0, "Failed to setsockopt (set mcast interface): %s\n", strerror(errno
));
138 config
->cluster_last_hb
= TIME
;
139 config
->cluster_seq_number
= -1;
141 return cluster_sockfd
;
146 // Send a chunk of data to the entire cluster (usually via the multicast
150 static int cluster_send_data(void *data
, int datalen
)
152 struct sockaddr_in addr
= {0};
154 if (!cluster_sockfd
) return -1;
155 if (!config
->cluster_address
) return 0;
157 addr
.sin_addr
.s_addr
= config
->cluster_address
;
158 addr
.sin_port
= htons(CLUSTERPORT
);
159 addr
.sin_family
= AF_INET
;
161 LOG(5,0,0,0, "Cluster send data: %d bytes\n", datalen
);
163 if (sendto(cluster_sockfd
, data
, datalen
, MSG_NOSIGNAL
, (void *) &addr
, sizeof(addr
)) < 0)
165 LOG(0, 0, 0, 0, "sendto: %s\n", strerror(errno
));
173 // Add a chunk of data to a heartbeat packet.
174 // Maintains the format. Assumes that the caller
175 // has passed in a big enough buffer!
177 static void add_type(char ** p
, int type
, int more
, char * data
, int size
)
179 * ( (u32
*)(*p
) ) = type
;
182 * ( (u32
*)(*p
) ) = more
;
185 if (data
&& size
> 0) {
186 memcpy(*p
, data
, size
);
191 // advertise our presence via BGP or gratuitous ARP
192 static void advertise(void)
196 bgp_enable_routing(1);
199 if (config
->send_garp
)
200 send_garp(config
->bind_address
); // Start taking traffic.
203 static void cluster_uptodate(void)
205 if (config
->cluster_iam_uptodate
)
208 if (config
->cluster_undefined_sessions
|| config
->cluster_undefined_tunnels
)
211 config
->cluster_iam_uptodate
= 1;
213 LOG(0,0,0,0, "Now uptodate with master.\n");
218 // Send a unicast UDP packet to a peer with 'data' as the
221 static int peer_send_data(u32 peer
, char * data
, int size
)
223 struct sockaddr_in addr
= {0};
225 if (!cluster_sockfd
) return -1;
226 if (!config
->cluster_address
) return 0;
231 addr
.sin_addr
.s_addr
= peer
;
232 addr
.sin_port
= htons(CLUSTERPORT
);
233 addr
.sin_family
= AF_INET
;
235 LOG_HEX(5, "Peer send", data
, size
);
237 if (sendto(cluster_sockfd
, data
, size
, MSG_NOSIGNAL
, (void *) &addr
, sizeof(addr
)) < 0)
239 LOG(0, 0, 0, 0, "sendto: %s\n", strerror(errno
));
247 // Send a structured message to a peer with a single element of type 'type'.
249 static int peer_send_message(u32 peer
, int type
, int more
, char * data
, int size
)
251 char buf
[65536]; // Vast overkill.
254 LOG(4,0,0,0, "Sending message to peer (type %d, more %d, size %d)\n", type
, more
, size
);
255 add_type(&p
, type
, more
, data
, size
);
257 return peer_send_data(peer
, buf
, (p
-buf
) );
261 // Forward a state changing packet to the master.
263 // The master just processes the payload as if it had
264 // received it off the tun device.
266 int master_forward_packet(char *data
, int size
, u32 addr
, int port
)
268 char buf
[65536]; // Vast overkill.
271 if (!config
->cluster_master_address
) // No election has been held yet. Just skip it.
274 LOG(4,0,0,0, "Forwarding packet from %s to master (size %d)\n", inet_toa(addr
), size
);
277 add_type(&p
, C_FORWARD
, addr
, (char*) &port
, sizeof(port
) );
278 memcpy(p
, data
, size
);
281 return peer_send_data(config
->cluster_master_address
, buf
, (p
-buf
) );
286 // Forward a throttled packet to the master for handling.
288 // The master just drops the packet into the appropriate
289 // token bucket queue, and lets normal processing take care
292 int master_throttle_packet(int tbfid
, char *data
, int size
)
294 char buf
[65536]; // Vast overkill.
297 if (!config
->cluster_master_address
) // No election has been held yet. Just skip it.
300 LOG(4,0,0,0, "Throttling packet master (size %d, tbfid %d)\n", size
, tbfid
);
302 add_type(&p
, C_THROTTLE
, tbfid
, data
, size
);
304 return peer_send_data(config
->cluster_master_address
, buf
, (p
-buf
) );
309 // Forward a walled garden packet to the master for handling.
311 // The master just writes the packet straight to the tun
312 // device (where is will normally loop through the
313 // firewall rules, and come back in on the tun device)
315 // (Note that this must be called with the tun header
316 // as the start of the data).
317 int master_garden_packet(sessionidt s
, char *data
, int size
)
319 char buf
[65536]; // Vast overkill.
322 if (!config
->cluster_master_address
) // No election has been held yet. Just skip it.
325 LOG(4,0,0,0, "Walled garden packet to master (size %d)\n", size
);
327 add_type(&p
, C_GARDEN
, s
, data
, size
);
329 return peer_send_data(config
->cluster_master_address
, buf
, (p
-buf
));
334 // Send a chunk of data as a heartbeat..
335 // We save it in the history buffer as we do so.
337 static void send_heartbeat(int seq
, char * data
, int size
)
341 if (size
> sizeof(past_hearts
[0].data
))
343 LOG(0,0,0,0, "Tried to heartbeat something larger than the maximum packet!\n");
347 i
= seq
% HB_HISTORY_SIZE
;
348 past_hearts
[i
].seq
= seq
;
349 past_hearts
[i
].size
= size
;
350 memcpy(&past_hearts
[i
].data
, data
, size
); // Save it.
351 cluster_send_data(data
, size
);
355 // Send an 'i am alive' message to every machine in the cluster.
357 void cluster_send_ping(time_t basetime
)
359 char buff
[100 + sizeof(pingt
)];
363 if (config
->cluster_iam_master
&& basetime
) // We're heartbeating so no need to ping.
366 LOG(5,0,0,0, "Sending cluster ping...\n");
369 x
.addr
= config
->bind_address
;
370 x
.undef
= config
->cluster_undefined_sessions
+ config
->cluster_undefined_tunnels
;
371 x
.basetime
= basetime
;
373 add_type(&p
, C_PING
, basetime
, (char *) &x
, sizeof(x
));
374 cluster_send_data(buff
, (p
-buff
) );
378 // Walk the session counters looking for non-zero ones to send
379 // to the master. We send up to 600 of them at one time.
380 // We examine a maximum of 3000 sessions.
381 // (50k max session should mean that we normally
382 // examine the entire session table every 25 seconds).
384 #define MAX_B_RECS (600)
385 void master_update_counts(void)
388 bytest b
[MAX_B_RECS
+1];
390 if (config
->cluster_iam_master
) // Only happens on the slaves.
393 if (!config
->cluster_master_address
) // If we don't have a master, skip it for a while.
396 i
= MAX_B_RECS
* 5; // Examine max 2000 sessions;
397 if (config
->cluster_highest_sessionid
> i
)
398 i
= config
->cluster_highest_sessionid
;
400 for ( c
= 0; i
> 0 ; --i
) {
401 // Next session to look at.
402 walk_session_number
++;
403 if ( walk_session_number
> config
->cluster_highest_sessionid
)
404 walk_session_number
= 1;
406 if (!sess_count
[walk_session_number
].cin
&& !sess_count
[walk_session_number
].cout
)
407 continue; // Unused. Skip it.
409 b
[c
].sid
= walk_session_number
;
410 b
[c
].in
= sess_count
[walk_session_number
].cin
;
411 b
[c
].out
= sess_count
[walk_session_number
].cout
;
413 if (++c
> MAX_B_RECS
) // Send a max of 400 elements in a packet.
417 sess_count
[walk_session_number
].cin
= sess_count
[walk_session_number
].cout
= 0;
420 if (!c
) // Didn't find any that changes. Get out of here!
424 // Forward the data to the master.
425 LOG(4,0,0,0, "Sending byte counters to master (%d elements)\n", c
);
426 peer_send_message(config
->cluster_master_address
, C_BYTES
, c
, (char*) &b
, sizeof(b
[0]) * c
);
431 // On the master, check how our slaves are going. If
432 // one of them's not up-to-date we'll heartbeat faster.
433 // If we don't have any of them, then we need to turn
434 // on our own packet handling!
436 void cluster_check_slaves(void)
439 static int have_peers
= 0;
440 int had_peers
= have_peers
;
443 if (!config
->cluster_iam_master
)
444 return; // Only runs on the master...
446 config
->cluster_iam_uptodate
= 1; // cleared in loop below
448 for (i
= have_peers
= 0; i
< num_peers
; i
++)
450 if ((peers
[i
].timestamp
+ config
->cluster_hb_timeout
) < t
)
451 continue; // Stale peer! Skip them.
453 if (!peers
[i
].basetime
)
454 continue; // Shutdown peer! Skip them.
456 if (peers
[i
].uptodate
)
459 if (!peers
[i
].uptodate
)
460 config
->cluster_iam_uptodate
= 0; // Start fast heartbeats
464 // in a cluster, withdraw/add routes when we get a peer/lose all peers
465 if (bgp_configured
&& have_peers
!= had_peers
)
466 bgp_enable_routing(!have_peers
);
471 // Check that we have a master. If it's been too
472 // long since we heard from a master then hold an election.
474 void cluster_check_master(void)
476 int i
, count
, tcount
, high_unique_id
= 0;
479 static int probed
= 0;
481 if (config
->cluster_iam_master
)
482 return; // Only runs on the slaves...
484 // If the master is late (missed 2 hearbeats by a second and a
485 // hair) it may be that the switch has dropped us from the
486 // multicast group, try unicasting one probe to the master
487 // which will hopefully respond with a unicast heartbeat that
488 // will allow us to limp along until the querier next runs.
489 if (TIME
> (config
->cluster_last_hb
+ 2 * config
->cluster_hb_interval
+ 11))
491 if (!probed
&& config
->cluster_master_address
)
494 LOG(1, 0, 0, 0, "Heartbeat from master %.1fs late, probing...\n",
495 0.1 * (TIME
- (config
->cluster_last_hb
+ config
->cluster_hb_interval
)));
497 peer_send_message(config
->cluster_master_address
,
498 C_LASTSEEN
, config
->cluster_seq_number
, NULL
, 0);
500 } else { // We got a recent heartbeat; reset the probe flag.
504 if (TIME
< (config
->cluster_last_hb
+ config
->cluster_hb_timeout
))
505 return; // Everything's ok!
507 config
->cluster_last_hb
= TIME
+ 1; // Just the one election thanks.
509 LOG(0,0,0,0, "Master timed out! Holding election...\n");
511 for (i
= 0; i
< num_peers
; i
++)
513 if ((peers
[i
].timestamp
+ config
->cluster_hb_timeout
) < t
)
514 continue; // Stale peer! Skip them.
516 if (!peers
[i
].basetime
)
517 continue; // Shutdown peer! Skip them.
519 if (peers
[i
].basetime
< basetime
) {
520 LOG(1,0,0,0, "Expecting %s to become master\n", inet_toa(peers
[i
].peer
) );
521 return; // They'll win the election. Get out of here.
524 if (peers
[i
].basetime
== basetime
&&
525 peers
[i
].peer
> my_address
) {
526 LOG(1,0,0,0, "Expecting %s to become master\n", inet_toa(peers
[i
].peer
) );
527 return; // They'll win the election. Wait for them to come up.
531 // Wow. it's been ages since I last heard a heartbeat
532 // and I'm better than an of my peers so it's time
533 // to become a master!!!
535 config
->cluster_iam_master
= 1;
536 config
->cluster_master_address
= 0;
538 LOG(0,0,0,0, "I am declaring myself the master!\n");
540 if (config
->cluster_seq_number
== -1)
541 config
->cluster_seq_number
= 0;
544 // Go through and mark all the tunnels as defined.
545 // Count the highest used tunnel number as well.
547 config
->cluster_highest_tunnelid
= 0;
548 for (i
= 0, tcount
= 0; i
< MAXTUNNEL
; ++i
) {
549 if (tunnel
[i
].state
== TUNNELUNDEF
)
550 tunnel
[i
].state
= TUNNELFREE
;
552 if (tunnel
[i
].state
!= TUNNELFREE
&& i
> config
->cluster_highest_tunnelid
)
553 config
->cluster_highest_tunnelid
= i
;
557 // Go through and mark all the sessions as being defined.
558 // reset the idle timeouts.
559 // add temporary byte counters to permanent ones.
560 // Re-string the free list.
561 // Find the ID of the highest session.
564 config
->cluster_highest_sessionid
= 0;
565 for (i
= 0, count
= 0; i
< MAXSESSION
; ++i
) {
566 if (session
[i
].tunnel
== T_UNDEF
) {
567 session
[i
].tunnel
= T_FREE
;
571 if (session
[i
].tunnel
== T_FREE
) { // Unused session. Add to free list.
572 session
[last_free
].next
= i
;
577 // Reset all the idle timeouts..
578 session
[i
].last_packet
= time_now
;
580 // Accumulate un-sent byte counters.
581 session
[i
].cin
+= sess_count
[i
].cin
;
582 session
[i
].cout
+= sess_count
[i
].cout
;
583 session
[i
].total_cin
+= sess_count
[i
].cin
;
584 session
[i
].total_cout
+= sess_count
[i
].cout
;
586 sess_count
[i
].cin
= sess_count
[i
].cout
= 0;
588 session
[i
].radius
= 0; // Reset authentication as the radius blocks aren't up to date.
590 if (session
[i
].unique_id
>= high_unique_id
) // This is different to the index into the session table!!!
591 high_unique_id
= session
[i
].unique_id
+1;
594 session
[i
].tbf_in
= session
[i
].tbf_out
= 0; // Remove stale pointers from old master.
595 throttle_session(i
, session
[i
].throttle_in
, session
[i
].throttle_out
);
597 if (session
[i
].tunnel
!= T_FREE
&& i
> config
->cluster_highest_sessionid
)
598 config
->cluster_highest_sessionid
= i
;
601 session
[last_free
].next
= 0; // End of chain.
602 last_id
= high_unique_id
; // Keep track of the highest used session ID.
606 rebuild_address_pool();
608 // If we're not the very first master, this is a big issue!
610 LOG(0,0,0,0, "Warning: Fixed %d uninitialized sessions in becoming master!\n", count
);
612 config
->cluster_undefined_sessions
= 0;
613 config
->cluster_undefined_tunnels
= 0;
614 config
->cluster_iam_uptodate
= 1; // assume all peers are up-to-date
616 if (!num_peers
) // lone master
619 // FIXME. We need to fix up the tunnel control message
620 // queue here! There's a number of other variables we
621 // should also update.
626 // Check that our session table is validly matching what the
627 // master has in mind.
629 // In particular, if we have too many sessions marked 'undefined'
630 // we fix it up here, and we ensure that the 'first free session'
633 static void cluster_check_sessions(int highsession
, int freesession_ptr
, int hightunnel
)
637 sessionfree
= freesession_ptr
; // Keep the freesession ptr valid.
639 if (config
->cluster_iam_uptodate
)
642 if (highsession
> config
->cluster_undefined_sessions
&& hightunnel
> config
->cluster_undefined_tunnels
)
645 // Clear out defined sessions, counting the number of
647 config
->cluster_undefined_sessions
= 0;
648 for (i
= 1 ; i
< MAXSESSION
; ++i
) {
649 if (i
> highsession
) {
650 session
[i
].tunnel
= 0; // Defined.
653 if (session
[i
].tunnel
!= T_UNDEF
)
655 ++config
->cluster_undefined_sessions
;
658 // Clear out defined tunnels, counting the number of
660 config
->cluster_undefined_tunnels
= 0;
661 for (i
= 1 ; i
< MAXTUNNEL
; ++i
) {
662 if (i
> hightunnel
) {
663 tunnel
[i
].state
= TUNNELFREE
; // Defined.
666 if (tunnel
[i
].state
!= TUNNELUNDEF
)
668 ++config
->cluster_undefined_tunnels
;
672 if (config
->cluster_undefined_sessions
|| config
->cluster_undefined_tunnels
) {
673 LOG(2,0,0,0, "Cleared undefined sessions/tunnels. %d sess (high %d), %d tunn (high %d)\n",
674 config
->cluster_undefined_sessions
, highsession
, config
->cluster_undefined_tunnels
, hightunnel
);
678 // Are we up to date?
680 if (!config
->cluster_iam_uptodate
)
684 static int hb_add_type(char **p
, int type
, int id
)
687 case C_CSESSION
: { // Compressed C_SESSION.
688 u8 c
[sizeof(sessiont
) * 2]; // Bigger than worst case.
689 u8
*d
= (u8
*) &session
[id
];
693 size
= rle_compress( &d
, sizeof(sessiont
), c
, sizeof(c
) );
695 // Did we compress the full structure, and is the size actually
697 if ( (d
- orig
) == sizeof(sessiont
) && size
< sizeof(sessiont
) ) {
698 add_type(p
, C_CSESSION
, id
, (char*) c
, size
);
701 // Failed to compress : Fall through.
703 case C_SESSION
: add_type(p
, C_SESSION
, id
,
704 (char*) &session
[id
], sizeof(sessiont
));
707 case C_CTUNNEL
: { // Compressed C_TUNNEL
708 u8 c
[sizeof(tunnelt
) * 2]; // Bigger than worst case.
709 u8
*d
= (u8
*) &tunnel
[id
];
713 size
= rle_compress( &d
, sizeof(tunnelt
), c
, sizeof(c
) );
715 // Did we compress the full structure, and is the size actually
717 if ( (d
- orig
) == sizeof(tunnelt
) && size
< sizeof(tunnelt
) ) {
718 add_type(p
, C_CTUNNEL
, id
, c
, size
);
721 // Failed to compress : Fall through.
723 case C_TUNNEL
: add_type(p
, C_TUNNEL
, id
,
724 (char*) &tunnel
[id
], sizeof(tunnelt
));
727 LOG(0,0,0,0, "Found an invalid type in heart queue! (%d)\n", type
);
735 // Send a heartbeat, incidently sending out any queued changes..
737 void cluster_heartbeat()
739 int i
, count
= 0, tcount
= 0;
740 char buff
[MAX_HEART_SIZE
+ sizeof(heartt
) + sizeof(int) ];
744 if (!config
->cluster_iam_master
) // Only the master does this.
747 // Fill out the heartbeat header.
748 memset(&h
, 0, sizeof(h
));
750 h
.version
= HB_VERSION
;
751 h
.seq
= config
->cluster_seq_number
;
752 h
.basetime
= basetime
;
753 h
.clusterid
= config
->bind_address
; // Will this do??
754 h
.basetime
= basetime
;
755 h
.highsession
= config
->cluster_highest_sessionid
;
756 h
.freesession
= sessionfree
;
757 h
.hightunnel
= config
->cluster_highest_tunnelid
;
758 h
.size_sess
= sizeof(sessiont
); // Just in case.
759 h
.size_tunn
= sizeof(tunnelt
);
760 h
.interval
= config
->cluster_hb_interval
;
761 h
.timeout
= config
->cluster_hb_timeout
;
763 add_type(&p
, C_HEARTBEAT
, HB_VERSION
, (char*) &h
, sizeof(h
));
765 for (i
= 0; i
< config
->cluster_num_changes
; ++i
) {
766 hb_add_type(&p
, cluster_changes
[i
].type
, cluster_changes
[i
].id
);
769 if (p
> (buff
+ sizeof(buff
))) { // Did we somehow manage to overun the buffer?
770 LOG(0,0,0,0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", p
- buff
);
776 // Fill out the packet with sessions from the session table...
777 // (not forgetting to leave space so we can get some tunnels in too )
778 while ( (p
+ sizeof(u32
) * 2 + sizeof(sessiont
) * 2 ) < (buff
+ MAX_HEART_SIZE
) ) {
780 if (!walk_session_number
) // session #0 isn't valid.
781 ++walk_session_number
;
783 if (count
>= config
->cluster_highest_sessionid
) // If we're a small cluster, don't go wild.
786 hb_add_type(&p
, C_CSESSION
, walk_session_number
);
787 walk_session_number
= (1+walk_session_number
)%(config
->cluster_highest_sessionid
+1); // +1 avoids divide by zero.
789 ++count
; // Count the number of extra sessions we're sending.
793 // Fill out the packet with tunnels from the tunnel table...
794 // This effectively means we walk the tunnel table more quickly
795 // than the session table. This is good because stuffing up a
796 // tunnel is a much bigger deal than stuffing up a session.
798 while ( (p
+ sizeof(u32
) * 2 + sizeof(tunnelt
) ) < (buff
+ MAX_HEART_SIZE
) ) {
800 if (!walk_tunnel_number
) // tunnel #0 isn't valid.
801 ++walk_tunnel_number
;
803 if (tcount
>= config
->cluster_highest_tunnelid
)
806 hb_add_type(&p
, C_CTUNNEL
, walk_tunnel_number
);
807 walk_tunnel_number
= (1+walk_tunnel_number
)%(config
->cluster_highest_tunnelid
+1); // +1 avoids divide by zero.
813 // Did we do something wrong?
814 if (p
> (buff
+ sizeof(buff
))) { // Did we somehow manage to overun the buffer?
815 LOG(0,0,0,0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", p
- buff
);
820 LOG(3,0,0,0, "Sending heartbeat #%d with %d changes (%d x-sess, %d x-tunnels, %d highsess, %d hightun, size %d)\n",
821 h
.seq
, config
->cluster_num_changes
, count
, tcount
, config
->cluster_highest_sessionid
,
822 config
->cluster_highest_tunnelid
, (p
-buff
));
824 config
->cluster_num_changes
= 0;
826 send_heartbeat(h
.seq
, buff
, (p
-buff
) ); // Send out the heartbeat to the cluster, keeping a copy of it.
828 config
->cluster_seq_number
= (config
->cluster_seq_number
+1)%HB_MAX_SEQ
; // Next seq number to use.
832 // A structure of type 'type' has changed; Add it to the queue to send.
834 static int type_changed(int type
, int id
)
838 for (i
= 0 ; i
< config
->cluster_num_changes
; ++i
)
839 if ( cluster_changes
[i
].id
== id
&&
840 cluster_changes
[i
].type
== type
)
841 return 0; // Already marked for change.
843 cluster_changes
[i
].type
= type
;
844 cluster_changes
[i
].id
= id
;
845 ++config
->cluster_num_changes
;
847 if (config
->cluster_num_changes
> MAX_CHANGES
)
848 cluster_heartbeat(); // flush now
854 // A particular session has been changed!
855 int cluster_send_session(int sid
)
857 if (!config
->cluster_iam_master
) {
858 LOG(0,0,sid
,0, "I'm not a master, but I just tried to change a session!\n");
862 return type_changed(C_CSESSION
, sid
);
865 // A particular tunnel has been changed!
866 int cluster_send_tunnel(int tid
)
868 if (!config
->cluster_iam_master
) {
869 LOG(0,0,0,tid
, "I'm not a master, but I just tried to change a tunnel!\n");
873 return type_changed(C_CTUNNEL
, tid
);
878 // We're a master, and a slave has just told us that it's
879 // missed a packet. We'll resend it every packet since
880 // the last one it's seen.
882 static int cluster_catchup_slave(int seq
, u32 slave
)
887 LOG(1,0,0,0, "Slave %s sent LASTSEEN with seq %d\n", inet_toa(slave
), seq
);
889 diff
= config
->cluster_seq_number
- seq
; // How many packet do we need to send?
893 if (diff
>= HB_HISTORY_SIZE
) { // Ouch. We don't have the packet to send it!
894 LOG(0,0,0,0, "A slaved asked for message %d when our seq number is %d. Killing it.\n",
895 seq
, config
->cluster_seq_number
);
896 return peer_send_message(slave
, C_KILL
, seq
, NULL
, 0);// Kill the slave. Nothing else to do.
899 // Now resend every packet that it missed, in order.
900 while (seq
!= config
->cluster_seq_number
) {
901 s
= seq
%HB_HISTORY_SIZE
;
902 if (seq
!= past_hearts
[s
].seq
) {
903 LOG(0,0,0,0, "Tried to re-send heartbeat for %s but %d doesn't match %d! (%d,%d)\n",
904 inet_toa(slave
), seq
, past_hearts
[s
].seq
, s
, config
->cluster_seq_number
);
905 return -1; // What to do here!?
907 peer_send_data(slave
, past_hearts
[s
].data
, past_hearts
[s
].size
);
908 seq
= (seq
+1)%HB_MAX_SEQ
; // Increment to next seq number.
910 return 0; // All good!
914 // We've heard from another peer! Add it to the list
915 // that we select from at election time.
917 static int cluster_add_peer(u32 peer
, time_t basetime
, pingt
*pp
, int size
)
923 // Allow for backward compatability.
924 // Just the ping packet into a new structure to allow
925 // for the possibility that we might have received
926 // more or fewer elements than we were expecting.
927 if (size
> sizeof(p
))
930 memset( (void*) &p
, 0, sizeof(p
) );
931 memcpy( (void*) &p
, (void*) pp
, size
);
934 if (clusterid
!= config
->bind_address
)
937 LOG(4,0,0,0, "Skipping ping from %s (different cluster)\n", inet_toa(peer
));
941 for (i
= 0; i
< num_peers
; ++i
)
943 if (peers
[i
].peer
!= peer
)
946 // This peer already exists. Just update the timestamp.
947 peers
[i
].basetime
= basetime
;
948 peers
[i
].timestamp
= TIME
;
949 peers
[i
].uptodate
= !p
.undef
;
953 // Is this the master shutting down??
954 if (peer
== config
->cluster_master_address
&& !basetime
) {
955 LOG(3,0,0,0, "Master %s shutting down...\n", inet_toa(config
->cluster_master_address
));
956 config
->cluster_master_address
= 0;
957 config
->cluster_last_hb
= 0; // Force an election.
958 cluster_check_master();
964 LOG(4,0,0,0, "Adding %s as a peer\n", inet_toa(peer
));
966 // Not found. Is there a stale slot to re-use?
967 for (i
= 0; i
< num_peers
; ++i
)
969 if (!peers
[i
].basetime
) // Shutdown
972 if ((peers
[i
].timestamp
+ config
->cluster_hb_timeout
* 10) < TIME
) // Stale.
976 if (i
>= CLUSTER_MAX_SIZE
)
979 LOG(0,0,0,0, "Tried to add %s as a peer, but I already have %d of them!\n", inet_toa(peer
), i
);
983 peers
[i
].peer
= peer
;
984 peers
[i
].basetime
= basetime
;
985 peers
[i
].timestamp
= TIME
;
986 peers
[i
].uptodate
= !p
.undef
;
990 LOG(1,0,0,0, "Added %s as a new peer. Now %d peers\n", inet_toa(peer
), num_peers
);
996 /* Handle the slave updating the byte counters for the master. */
998 // Note that we don't mark the session as dirty; We rely on
999 // the slow table walk to propogate this back out to the slaves.
1001 static int cluster_handle_bytes(char * data
, int size
)
1007 LOG(3,0,0,0, "Got byte counter update (size %d)\n", size
);
1009 /* Loop around, adding the byte
1010 counts to each of the sessions. */
1012 while (size
>= sizeof(*b
) ) {
1013 if (b
->sid
> MAXSESSION
) {
1014 LOG(0,0,0,0, "Got C_BYTES with session #%d!\n", b
->sid
);
1015 return -1; /* Abort processing */
1018 session
[b
->sid
].total_cin
+= b
->in
;
1019 session
[b
->sid
].total_cout
+= b
->out
;
1021 session
[b
->sid
].cin
+= b
->in
;
1022 session
[b
->sid
].cout
+= b
->out
;
1023 session
[b
->sid
].last_packet
= time_now
; // Reset idle timer!
1030 LOG(0,0,0,0, "Got C_BYTES with %d bytes of trailing junk!\n", size
);
1036 // Handle receiving a session structure in a heartbeat packet.
1038 static int cluster_recv_session(int more
, u8
* p
)
1040 if (more
>= MAXSESSION
) {
1041 LOG(0,0,0,0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
1045 if (session
[more
].tunnel
== T_UNDEF
) {
1046 if (config
->cluster_iam_uptodate
) { // Sanity.
1047 LOG(0,0,0,0, "I thought I was uptodate but I just found an undefined session!\n");
1049 --config
->cluster_undefined_sessions
;
1053 load_session(more
, (sessiont
*) p
); // Copy session into session table..
1055 LOG(5,0,more
,0, "Received session update (%d undef)\n", config
->cluster_undefined_sessions
);
1057 if (!config
->cluster_iam_uptodate
)
1058 cluster_uptodate(); // Check to see if we're up to date.
1063 static int cluster_recv_tunnel(int more
, u8
*p
)
1065 if (more
>= MAXTUNNEL
) {
1066 LOG(0,0,0,0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
1070 if (tunnel
[more
].state
== TUNNELUNDEF
) {
1071 if (config
->cluster_iam_uptodate
) { // Sanity.
1072 LOG(0,0,0,0, "I thought I was uptodate but I just found an undefined tunnel!\n");
1074 --config
->cluster_undefined_tunnels
;
1078 memcpy(&tunnel
[more
], p
, sizeof(tunnel
[more
]) );
1081 // Clear tunnel control messages. These are dynamically allocated.
1082 // If we get unlucky, this may cause the tunnel to drop!
1084 tunnel
[more
].controls
= tunnel
[more
].controle
= NULL
;
1085 tunnel
[more
].controlc
= 0;
1087 LOG(5,0,0,more
, "Received tunnel update\n");
1089 if (!config
->cluster_iam_uptodate
)
1090 cluster_uptodate(); // Check to see if we're up to date.
1097 // Process a heartbeat..
1099 static int cluster_process_heartbeat(u8
* data
, int size
, int more
, u8
* p
, u32 addr
)
1102 int s
= size
- (p
-data
);
1106 # error "need to update cluster_process_heartbeat()"
1109 // we handle version 2+
1110 if (more
< 2 || more
> HB_VERSION
) {
1111 LOG(0,0,0,0, "Received a heartbeat version that I don't support (%d)!\n", more
);
1112 return -1; // Ignore it??
1115 // Ok. It's a heartbeat packet from a cluster master!
1123 if (h
->clusterid
!= config
->bind_address
)
1124 return -1; // It's not part of our cluster.
1126 if (config
->cluster_iam_master
) { // Sanity...
1127 // Note that this MUST match the election process above!
1129 LOG(0,0,0,0, "I just got a packet claiming to be from a master but _I_ am the master!\n");
1131 LOG(0,0,0,0, "Heartbeat from addr %s with zero basetime!\n", inet_toa(addr
) );
1132 return -1; // Skip it.
1134 if (basetime
> h
->basetime
) {
1135 LOG(0,0,0,0, "They're (%s) an older master than me so I'm gone!\n", inet_toa(addr
));
1139 if (basetime
== h
->basetime
&& my_address
< addr
) { // Tie breaker.
1140 LOG(0,0,0,0, "They're a higher IP address than me, so I'm gone!\n");
1144 return -1; // Skip it.
1147 if (config
->cluster_seq_number
== -1) // Don't have one. Just align to the master...
1148 config
->cluster_seq_number
= h
->seq
;
1150 config
->cluster_last_hb
= TIME
; // Reset to ensure that we don't become master!!
1152 if (config
->cluster_seq_number
!= h
->seq
) { // Out of sequence heartbeat!
1153 LOG(1,0,0,0, "HB: Got seq# %d but was expecting %d. asking for resend.\n", h
->seq
, config
->cluster_seq_number
);
1155 peer_send_message(addr
, C_LASTSEEN
, config
->cluster_seq_number
, NULL
, 0);
1157 config
->cluster_last_hb
= TIME
; // Reset to ensure that we don't become master!!
1159 // Just drop the packet. The master will resend it as part of the catchup.
1163 // Save the packet in our buffer.
1164 // This is needed in case we become the master.
1165 config
->cluster_seq_number
= (h
->seq
+1)%HB_MAX_SEQ
;
1166 i
= h
->seq
% HB_HISTORY_SIZE
;
1167 past_hearts
[i
].seq
= h
->seq
;
1168 past_hearts
[i
].size
= size
;
1169 memcpy(&past_hearts
[i
].data
, data
, size
); // Save it.
1172 // Check that we don't have too many undefined sessions, and
1173 // that the free session pointer is correct.
1174 cluster_check_sessions(h
->highsession
, h
->freesession
, h
->hightunnel
);
1176 if (more
> 2) // reserved section of heartt was not initialized prior to v3
1178 if (h
->interval
!= config
->cluster_hb_interval
)
1180 LOG(2, 0, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
1181 h
->interval
, config
->cluster_hb_interval
);
1183 config
->cluster_hb_interval
= h
->interval
;
1186 if (h
->timeout
!= config
->cluster_hb_timeout
)
1188 LOG(2, 0, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
1189 h
->timeout
, config
->cluster_hb_timeout
);
1191 config
->cluster_hb_timeout
= h
->timeout
;
1195 // Ok. process the packet...
1198 type
= * ((u32
*) p
);
1202 more
= * ((u32
*) p
);
1207 case C_CSESSION
: { // Compressed session structure.
1208 u8 c
[ sizeof(sessiont
) + 2];
1212 size
= rle_decompress((u8
**) &p
, s
, c
, sizeof(c
) );
1215 if (size
!= sizeof(sessiont
) ) { // Ouch! Very very bad!
1216 LOG(0,0,0,0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1217 // Now what? Should exit! No-longer up to date!
1221 cluster_recv_session(more
, c
);
1225 if ( s
< sizeof(session
[more
]))
1228 cluster_recv_session(more
, p
);
1230 p
+= sizeof(session
[more
]);
1231 s
-= sizeof(session
[more
]);
1234 case C_CTUNNEL
: { // Compressed tunnel structure.
1235 u8 c
[ sizeof(tunnelt
) + 2];
1239 size
= rle_decompress( (u8
**) &p
, s
, c
, sizeof(c
) );
1242 if (size
!= sizeof(tunnelt
) ) { // Ouch! Very very bad!
1243 LOG(0,0,0,0, "DANGER: Received a CSESSION that didn't decompress correctly!\n");
1244 // Now what? Should exit! No-longer up to date!
1248 cluster_recv_tunnel(more
, c
);
1253 if ( s
< sizeof(tunnel
[more
]))
1256 cluster_recv_tunnel(more
, p
);
1258 p
+= sizeof(tunnel
[more
]);
1259 s
-= sizeof(tunnel
[more
]);
1262 LOG(0,0,0,0, "DANGER: I received a heartbeat element where I didn't understand the type! (%d)\n", type
);
1263 return -1; // can't process any more of the packet!!
1266 if (config
->cluster_master_address
!= addr
)
1269 str
= strdup(inet_toa(config
->cluster_master_address
));
1270 LOG(0,0,0,0, "My master just changed from %s to %s!\n", str
, inet_toa(addr
));
1274 config
->cluster_master_address
= addr
;
1275 config
->cluster_last_hb
= TIME
; // Successfully received a heartbeat!
1279 LOG(0,0,0,0, "I got an incomplete heartbeat packet! This means I'm probably out of sync!!\n");
1284 // We got a packet on the cluster port!
1285 // Handle pings, lastseens, and heartbeats!
1287 int processcluster(char * data
, int size
, u32 addr
)
1293 if (addr
== my_address
)
1294 return -1; // Ignore it. Something looped back the multicast!
1296 LOG(5,0,0,0, "Process cluster: %d bytes from %s\n", size
, inet_toa(addr
));
1298 if (s
<= 0) // Any data there??
1304 type
= * ((u32
*) p
);
1308 more
= * ((u32
*) p
);
1313 case C_PING
: // Update the peers table.
1314 return cluster_add_peer(addr
, more
, (pingt
*)p
, s
);
1316 case C_LASTSEEN
: // Catch up a slave (slave missed a packet).
1317 return cluster_catchup_slave(more
, addr
);
1319 case C_FORWARD
: { // Forwarded control packet. pass off to processudp.
1320 struct sockaddr_in a
;
1321 a
.sin_addr
.s_addr
= more
;
1323 a
.sin_port
= * (int*) p
;
1327 if (!config
->cluster_iam_master
) { // huh?
1328 LOG(0,0,0,0, "I'm not the master, but I got a C_FORWARD from %s?\n", inet_toa(addr
));
1332 LOG(4,0,0,0, "Got a forwarded packet... (%s:%d)\n", inet_toa(more
), a
.sin_port
);
1334 processudp(p
, s
, &a
);
1337 case C_THROTTLE
: { // Receive a forwarded packet from a slave.
1338 if (!config
->cluster_iam_master
) {
1339 LOG(0,0,0,0, "I'm not the master, but I got a C_THROTTLE from %s?\n", inet_toa(addr
));
1343 tbf_queue_packet(more
, p
, s
); // The TBF id tells wether it goes in or out.
1347 // Receive a walled garden packet from a slave.
1348 if (!config
->cluster_iam_master
) {
1349 LOG(0,0,0,0, "I'm not the master, but I got a C_GARDEN from %s?\n", inet_toa(addr
));
1357 return cluster_handle_bytes(p
, s
);
1359 case C_KILL
: // The master asked us to die!? (usually because we're too out of date).
1360 if (config
->cluster_iam_master
) {
1361 LOG(0,0,0,0, "_I_ am master, but I received a C_KILL from %s! (Seq# %d)\n", inet_toa(addr
), more
);
1364 if (more
!= config
->cluster_seq_number
) {
1365 LOG(0,0,0,0, "The master asked us to die but the seq number didn't match!?\n");
1369 if (addr
!= config
->cluster_master_address
) {
1370 LOG(0,0,0,0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%x)\n",
1371 inet_toa(addr
), config
->cluster_master_address
);
1372 // We can only warn about it. The master might really have switched!
1375 LOG(0,0,0,0, "Received a valid C_KILL: I'm going to die now.\n");
1377 exit(0); // Lets be paranoid;
1378 return -1; // Just signalling the compiler.
1381 LOG(4,0,0,0, "Got a heartbeat from %s\n", inet_toa(addr
));
1382 return cluster_process_heartbeat(data
, size
, more
, p
, addr
);
1385 LOG(0,0,0,0, "Strange type packet received on cluster socket (%d)\n", type
);
1391 LOG(0,0,0,0, "I got a _short_ cluster heartbeat packet! This means I'm probably out of sync!!\n");
1395 //====================================================================================================
1397 int cmd_show_cluster(struct cli_def
*cli
, char *command
, char **argv
, int argc
)
1401 if (CLI_HELP_REQUESTED
)
1402 return CLI_HELP_NO_ARGS
;
1404 cli_print(cli
, "Cluster status : %s", config
->cluster_iam_master
? "Master" : "Slave" );
1405 cli_print(cli
, "My address : %s", inet_toa(my_address
));
1406 cli_print(cli
, "VIP address : %s", inet_toa(config
->bind_address
));
1407 cli_print(cli
, "Multicast address: %s", inet_toa(config
->cluster_address
));
1408 cli_print(cli
, "Multicast i'face : %s", config
->cluster_interface
);
1410 if (!config
->cluster_iam_master
) {
1411 cli_print(cli
, "My master : %s (last heartbeat %.1f seconds old)",
1412 config
->cluster_master_address
? inet_toa(config
->cluster_master_address
) : "Not defined",
1413 0.1 * (TIME
- config
->cluster_last_hb
));
1414 cli_print(cli
, "Uptodate : %s", config
->cluster_iam_uptodate
? "Yes" : "No");
1415 cli_print(cli
, "Next sequence number expected: %d", config
->cluster_seq_number
);
1416 cli_print(cli
, "%d sessions undefined of %d", config
->cluster_undefined_sessions
, config
->cluster_highest_sessionid
);
1417 cli_print(cli
, "%d tunnels undefined of %d", config
->cluster_undefined_tunnels
, config
->cluster_highest_tunnelid
);
1419 cli_print(cli
, "Next heartbeat # : %d", config
->cluster_seq_number
);
1420 cli_print(cli
, "Highest session : %d", config
->cluster_highest_sessionid
);
1421 cli_print(cli
, "Highest tunnel : %d", config
->cluster_highest_tunnelid
);
1422 cli_print(cli
, "%d changes queued for sending", config
->cluster_num_changes
);
1424 cli_print(cli
, "%d peers.", num_peers
);
1427 cli_print(cli
, "%20s %10s %8s", "Address", "Basetime", "Age");
1428 for (i
= 0; i
< num_peers
; ++i
) {
1429 cli_print(cli
, "%20s %10d %8d", inet_toa(peers
[i
].peer
),
1430 peers
[i
].basetime
, TIME
- peers
[i
].timestamp
);
1436 // Simple run-length-encoding compression.
1438 // 1 byte < 128 = count of non-zero bytes following. // Not legal to be zero.
1439 // n non-zero bytes;
1441 // 1 byte > 128 = (count - 128) run of zero bytes. //
1443 // count == 0 indicates end of compressed stream.
1445 // Compress from 'src' into 'dst'. return number of bytes
1447 // Updates *src_p to indicate 1 past last bytes used.
1449 // We could get an extra byte in the zero runs by storing (count-1)
1450 // but I'm playing it safe.
1452 // Worst case is a 50% expansion in space required (trying to
1453 // compress { 0x00, 0x01 } * N )
1454 static int rle_compress(u8
** src_p
, int ssize
, u8
*dst
, int dsize
)
1457 int orig_dsize
= dsize
;
1461 while (ssize
> 0 && dsize
> 2) {
1463 x
= dst
++; --dsize
; // Reserve space for count byte..
1465 if (*src
) { // Copy a run of non-zero bytes.
1466 while (*src
&& count
< 127 && ssize
> 0 && dsize
> 1) { // Count number of non-zero bytes.
1471 *x
= count
; // Store number of non-zero bytes. Guarenteed to be non-zero!
1473 } else { // Compress a run of zero bytes.
1474 while (*src
== 0 && count
< 127 && ssize
> 0) {
1483 *dst
++ = 0x0; // Add Stop byte.
1487 return (orig_dsize
- dsize
);
1491 // Decompress the buffer into **p.
1492 // 'psize' is the size of the decompression buffer available.
1494 // Returns the number of bytes decompressed.
1496 // Decompresses from '*src_p' into 'dst'.
1497 // Return the number of dst bytes used.
1498 // Updates the 'src_p' pointer to point to the
1499 // first un-used byte.
1500 static int rle_decompress(u8
** src_p
, int ssize
, u8
*dst
, int dsize
)
1503 int orig_dsize
= dsize
;
1504 char * src
= *src_p
;
1506 while (ssize
>0 && dsize
> 0) { // While there's more to decompress, and there's room in the decompress buffer...
1507 count
= *src
++; --ssize
; // get the count byte from the source.
1508 if (count
== 0x0) // End marker reached? If so, finish.
1511 if (count
& 0x80) { // Decompress a run of zeros
1512 for (count
&= 0x7f ; count
> 0 && dsize
> 0; --count
) {
1516 } else { // Copy run of non-zero bytes.
1517 for ( ; count
> 0 && ssize
&& dsize
; --count
) { // Copy non-zero bytes across.
1524 return (orig_dsize
- dsize
);