4 Copyright (C) Andrew Tridgell 2006
6 This program is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 3 of the License, or
9 (at your option) any later version.
11 This program is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with this program; if not, see <http://www.gnu.org/licenses/>.
20 #ifndef _CTDB_PROTOCOL_H
21 #define _CTDB_PROTOCOL_H
23 /* location of daemon socket */
24 #define CTDB_PATH "/tmp/ctdb.socket"
26 /* default ctdb port number */
27 #define CTDB_PORT 4379
29 /* we must align packets to ensure ctdb works on all architectures (eg. sparc) */
30 #define CTDB_DS_ALIGNMENT 8
33 #define CTDB_NULL_FUNC 0xFF000001
34 #define CTDB_FETCH_FUNC 0xFF000002
35 #define CTDB_FETCH_WITH_HEADER_FUNC 0xFF000003
44 #define CTDB_IMMEDIATE_MIGRATION 0x00000001
45 #define CTDB_CALL_FLAG_VACUUM_MIGRATION 0x00000002
46 #define CTDB_WANT_READONLY 0x00000004
51 structure passed to a ctdb call backend function
53 struct ctdb_call_info {
54 TDB_DATA key; /* record key */
55 struct ctdb_ltdb_header *header;
56 TDB_DATA record_data; /* current data in the record */
57 TDB_DATA *new_data; /* optionally updated record data */
58 TDB_DATA *call_data; /* optionally passed from caller */
59 TDB_DATA *reply_data; /* optionally returned by function */
60 uint32_t status; /* optional reply status - defaults to zero */
63 #define CTDB_ERR_INVALID 1
64 #define CTDB_ERR_NOMEM 2
69 #define CTDB_FLAG_TORTURE (1<<1)
72 a message handler ID meaning "give me all messages"
74 #define CTDB_SRVID_ALL (~(uint64_t)0)
79 #define CTDB_SRVID_RECOVERY 0xF100000000000000LL
82 a message handler ID meaning that the cluster has been reconfigured
84 #define CTDB_SRVID_RECONFIGURE 0xF200000000000000LL
87 a message handler ID meaning that an IP address has been released
89 #define CTDB_SRVID_RELEASE_IP 0xF300000000000000LL
92 a message handler ID meaning that an IP address has been taken
94 #define CTDB_SRVID_TAKE_IP 0xF301000000000000LL
97 a message ID to set the node flags in the recovery daemon
99 #define CTDB_SRVID_SET_NODE_FLAGS 0xF400000000000000LL
102 a message ID to ask the recovery daemon to update the expected node
103 assignment for a public ip
105 #define CTDB_SRVID_RECD_UPDATE_IP 0xF500000000000000LL
108 a message to tell the recovery daemon to fetch a set of records
110 #define CTDB_SRVID_VACUUM_FETCH 0xF700000000000000LL
113 a message to tell the recovery daemon to write a talloc memdump
116 #define CTDB_SRVID_MEM_DUMP 0xF800000000000000LL
119 a message ID to get the recovery daemon to push the node flags out
121 #define CTDB_SRVID_PUSH_NODE_FLAGS 0xF900000000000000LL
124 a message ID to get the recovery daemon to reload the nodes file
126 #define CTDB_SRVID_RELOAD_NODES 0xFA00000000000000LL
129 a message ID to get the recovery daemon to perform a takeover run
131 #define CTDB_SRVID_TAKEOVER_RUN 0xFB00000000000000LL
133 /* A message id to ask the recovery daemon to temporarily disable the
136 #define CTDB_SRVID_DISABLE_IP_CHECK 0xFC00000000000000LL
138 /* A dummy port used for sending back ipreallocate resposnes to the main
141 #define CTDB_SRVID_TAKEOVER_RUN_RESPONSE 0xFD00000000000000LL
143 /* A range of ports reserved for registering a PID (top 8 bits)
144 * All ports matching the 8 top bits are reserved for exclusive use by
145 * registering a SRVID that matches the process-id of the requesting process
147 #define CTDB_SRVID_PID_RANGE 0x0000000000000000LL
149 /* A range of ports reserved for samba (top 8 bits)
150 * All ports matching the 8 top bits are reserved for exclusive use by
153 #define CTDB_SRVID_SAMBA_NOTIFY 0xFE00000000000000LL
154 #define CTDB_SRVID_SAMBA_RANGE 0xFE00000000000000LL
156 /* A range of ports reserved for a CTDB NFS server (top 8 bits)
157 * All ports matching the 8 top bits are reserved for exclusive use by
160 #define CTDB_SRVID_NFSD_RANGE 0xEE00000000000000LL
162 /* A range of ports reserved for a CTDB ISCSI server (top 8 bits)
163 * All ports matching the 8 top bits are reserved for exclusive use by
166 #define CTDB_SRVID_ISCSID_RANGE 0xDE00000000000000LL
168 /* A range of ports reserved for testing (top 8 bits)
169 * All ports matching the 8 top bits are reserved for exclusive use by
172 #define CTDB_SRVID_TEST_RANGE 0xCE00000000000000LL
174 /* Range of ports reserved for traversals */
175 #define CTDB_SRVID_TRAVERSE_RANGE 0xBE00000000000000LL
177 /* used on the domain socket, send a pdu to the local daemon */
178 #define CTDB_CURRENT_NODE 0xF0000001
179 /* send a broadcast to all nodes in the cluster, active or not */
180 #define CTDB_BROADCAST_ALL 0xF0000002
181 /* send a broadcast to all nodes in the current vnn map */
182 #define CTDB_BROADCAST_VNNMAP 0xF0000003
183 /* send a broadcast to all connected nodes */
184 #define CTDB_BROADCAST_CONNECTED 0xF0000004
186 /* the key used for transaction locking on persistent databases */
187 #define CTDB_TRANSACTION_LOCK_KEY "__transaction_lock__"
189 /* the key used to store persistent db sequence number */
190 #define CTDB_DB_SEQNUM_KEY "__db_sequence_number__"
192 #define MONITOR_SCRIPT_OK 0
193 #define MONITOR_SCRIPT_TIMEOUT 1
195 #define MAX_SCRIPT_NAME 31
196 #define MAX_SCRIPT_OUTPUT 511
197 struct ctdb_script_wire {
198 char name[MAX_SCRIPT_NAME+1];
199 struct timeval start;
200 struct timeval finished;
202 char output[MAX_SCRIPT_OUTPUT+1];
205 struct ctdb_scripts_wire {
206 uint32_t num_scripts;
207 struct ctdb_script_wire scripts[1];
210 /* different calls to event scripts. */
211 enum ctdb_eventscript_call {
212 CTDB_EVENT_INIT, /* CTDB starting up: no args */
213 CTDB_EVENT_SETUP, /* CTDB starting up after transport is readdy: no args. */
214 CTDB_EVENT_STARTUP, /* CTDB starting up after initial recovery: no args. */
215 CTDB_EVENT_START_RECOVERY, /* CTDB recovery starting: no args. */
216 CTDB_EVENT_RECOVERED, /* CTDB recovery finished: no args. */
217 CTDB_EVENT_TAKE_IP, /* IP taken: interface, IP address, netmask bits. */
218 CTDB_EVENT_RELEASE_IP, /* IP released: interface, IP address, netmask bits. */
219 CTDB_EVENT_STOPPED, /* This node is stopped: no args. */
220 CTDB_EVENT_MONITOR, /* Please check if service is healthy: no args. */
221 CTDB_EVENT_STATUS, /* Report service status: no args. */
222 CTDB_EVENT_SHUTDOWN, /* CTDB shutting down: no args. */
223 CTDB_EVENT_RELOAD, /* magic */
224 CTDB_EVENT_UPDATE_IP, /* IP updating: old interface, new interface, IP address, netmask bits. */
225 CTDB_EVENT_IPREALLOCATED, /* when a takeover_run() completes */
229 /* Mapping from enum to names. */
230 extern const char *ctdb_eventscript_call_names[];
235 enum ctdb_operation {
238 CTDB_REQ_DMASTER = 2,
239 CTDB_REPLY_DMASTER = 3,
240 CTDB_REPLY_ERROR = 4,
241 CTDB_REQ_MESSAGE = 5,
243 CTDB_REQ_CONTROL = 7,
244 CTDB_REPLY_CONTROL = 8,
245 CTDB_REQ_KEEPALIVE = 9,
248 #define CTDB_MAGIC 0x43544442 /* CTDB */
249 #define CTDB_VERSION 1
251 enum ctdb_controls {CTDB_CONTROL_PROCESS_EXISTS = 0,
252 CTDB_CONTROL_STATISTICS = 1,
254 CTDB_CONTROL_PING = 3,
255 CTDB_CONTROL_GETDBPATH = 4,
256 CTDB_CONTROL_GETVNNMAP = 5,
257 CTDB_CONTROL_SETVNNMAP = 6,
258 CTDB_CONTROL_GET_DEBUG = 7,
259 CTDB_CONTROL_SET_DEBUG = 8,
260 CTDB_CONTROL_GET_DBMAP = 9,
261 CTDB_CONTROL_GET_NODEMAPv4 = 10, /* obsolete */
262 CTDB_CONTROL_SET_DMASTER = 11,
264 CTDB_CONTROL_PULL_DB = 13,
265 CTDB_CONTROL_PUSH_DB = 14,
266 CTDB_CONTROL_GET_RECMODE = 15,
267 CTDB_CONTROL_SET_RECMODE = 16,
268 CTDB_CONTROL_STATISTICS_RESET = 17,
269 CTDB_CONTROL_DB_ATTACH = 18,
270 CTDB_CONTROL_SET_CALL = 19,
271 CTDB_CONTROL_TRAVERSE_START = 20,
272 CTDB_CONTROL_TRAVERSE_ALL = 21,
273 CTDB_CONTROL_TRAVERSE_DATA = 22,
274 CTDB_CONTROL_REGISTER_SRVID = 23,
275 CTDB_CONTROL_DEREGISTER_SRVID = 24,
276 CTDB_CONTROL_GET_DBNAME = 25,
277 CTDB_CONTROL_ENABLE_SEQNUM = 26,
278 CTDB_CONTROL_UPDATE_SEQNUM = 27,
280 CTDB_CONTROL_DUMP_MEMORY = 29,
281 CTDB_CONTROL_GET_PID = 30,
282 CTDB_CONTROL_GET_RECMASTER = 31,
283 CTDB_CONTROL_SET_RECMASTER = 32,
284 CTDB_CONTROL_FREEZE = 33,
285 CTDB_CONTROL_THAW = 34,
286 CTDB_CONTROL_GET_PNN = 35,
287 CTDB_CONTROL_SHUTDOWN = 36,
288 CTDB_CONTROL_GET_MONMODE = 37,
293 CTDB_CONTROL_TAKEOVER_IPv4 = 42, /* obsolete */
294 CTDB_CONTROL_RELEASE_IPv4 = 43, /* obsolete */
295 CTDB_CONTROL_TCP_CLIENT = 44,
296 CTDB_CONTROL_TCP_ADD = 45,
297 CTDB_CONTROL_TCP_REMOVE = 46,
298 CTDB_CONTROL_STARTUP = 47,
299 CTDB_CONTROL_SET_TUNABLE = 48,
300 CTDB_CONTROL_GET_TUNABLE = 49,
301 CTDB_CONTROL_LIST_TUNABLES = 50,
302 CTDB_CONTROL_GET_PUBLIC_IPSv4 = 51, /* obsolete */
303 CTDB_CONTROL_MODIFY_FLAGS = 52,
304 CTDB_CONTROL_GET_ALL_TUNABLES = 53,
305 CTDB_CONTROL_KILL_TCP = 54,
306 CTDB_CONTROL_GET_TCP_TICKLE_LIST = 55,
307 CTDB_CONTROL_SET_TCP_TICKLE_LIST = 56,
308 CTDB_CONTROL_REGISTER_SERVER_ID = 57,
309 CTDB_CONTROL_UNREGISTER_SERVER_ID = 58,
310 CTDB_CONTROL_CHECK_SERVER_ID = 59,
311 CTDB_CONTROL_GET_SERVER_ID_LIST = 60,
312 CTDB_CONTROL_DB_ATTACH_PERSISTENT = 61,
313 CTDB_CONTROL_PERSISTENT_STORE = 62,
314 CTDB_CONTROL_UPDATE_RECORD = 63,
315 CTDB_CONTROL_SEND_GRATIOUS_ARP = 64,
316 CTDB_CONTROL_TRANSACTION_START = 65,
317 CTDB_CONTROL_TRANSACTION_COMMIT = 66,
318 CTDB_CONTROL_WIPE_DATABASE = 67,
320 CTDB_CONTROL_UPTIME = 69,
321 CTDB_CONTROL_START_RECOVERY = 70,
322 CTDB_CONTROL_END_RECOVERY = 71,
323 CTDB_CONTROL_RELOAD_NODES_FILE = 72,
325 CTDB_CONTROL_TRY_DELETE_RECORDS = 74,
326 CTDB_CONTROL_ENABLE_MONITOR = 75,
327 CTDB_CONTROL_DISABLE_MONITOR = 76,
328 CTDB_CONTROL_ADD_PUBLIC_IP = 77,
329 CTDB_CONTROL_DEL_PUBLIC_IP = 78,
330 CTDB_CONTROL_RUN_EVENTSCRIPTS = 79,
331 CTDB_CONTROL_GET_CAPABILITIES = 80,
332 CTDB_CONTROL_START_PERSISTENT_UPDATE = 81,
333 CTDB_CONTROL_CANCEL_PERSISTENT_UPDATE= 82,
334 CTDB_CONTROL_TRANS2_COMMIT = 83,
335 CTDB_CONTROL_TRANS2_FINISHED = 84,
336 CTDB_CONTROL_TRANS2_ERROR = 85,
337 CTDB_CONTROL_TRANS2_COMMIT_RETRY = 86,
338 CTDB_CONTROL_RECD_PING = 87,
339 CTDB_CONTROL_RELEASE_IP = 88,
340 CTDB_CONTROL_TAKEOVER_IP = 89,
341 CTDB_CONTROL_GET_PUBLIC_IPS = 90,
342 CTDB_CONTROL_GET_NODEMAP = 91,
343 CTDB_CONTROL_GET_EVENT_SCRIPT_STATUS = 96,
344 CTDB_CONTROL_TRAVERSE_KILL = 97,
345 CTDB_CONTROL_RECD_RECLOCK_LATENCY = 98,
346 CTDB_CONTROL_GET_RECLOCK_FILE = 99,
347 CTDB_CONTROL_SET_RECLOCK_FILE = 100,
348 CTDB_CONTROL_STOP_NODE = 101,
349 CTDB_CONTROL_CONTINUE_NODE = 102,
350 CTDB_CONTROL_SET_NATGWSTATE = 103,
351 CTDB_CONTROL_SET_LMASTERROLE = 104,
352 CTDB_CONTROL_SET_RECMASTERROLE = 105,
353 CTDB_CONTROL_ENABLE_SCRIPT = 107,
354 CTDB_CONTROL_DISABLE_SCRIPT = 108,
355 CTDB_CONTROL_SET_BAN_STATE = 109,
356 CTDB_CONTROL_GET_BAN_STATE = 110,
357 CTDB_CONTROL_SET_DB_PRIORITY = 111,
358 CTDB_CONTROL_GET_DB_PRIORITY = 112,
359 CTDB_CONTROL_TRANSACTION_CANCEL = 113,
360 CTDB_CONTROL_REGISTER_NOTIFY = 114,
361 CTDB_CONTROL_DEREGISTER_NOTIFY = 115,
362 CTDB_CONTROL_TRANS2_ACTIVE = 116,
363 CTDB_CONTROL_GET_LOG = 117,
364 CTDB_CONTROL_CLEAR_LOG = 118,
365 CTDB_CONTROL_TRANS3_COMMIT = 119,
366 CTDB_CONTROL_GET_DB_SEQNUM = 120,
367 CTDB_CONTROL_DB_SET_HEALTHY = 121,
368 CTDB_CONTROL_DB_GET_HEALTH = 122,
369 CTDB_CONTROL_GET_PUBLIC_IP_INFO = 123,
370 CTDB_CONTROL_GET_IFACES = 124,
371 CTDB_CONTROL_SET_IFACE_LINK_STATE = 125,
372 CTDB_CONTROL_TCP_ADD_DELAYED_UPDATE = 126,
373 CTDB_CONTROL_GET_STAT_HISTORY = 127,
374 CTDB_CONTROL_SCHEDULE_FOR_DELETION = 128,
375 CTDB_CONTROL_SET_DB_READONLY = 129,
376 CTDB_CONTROL_CHECK_SRVIDS = 130,
377 CTDB_CONTROL_TRAVERSE_START_EXT = 131,
383 struct ctdb_req_header {
386 uint32_t ctdb_version;
394 struct ctdb_req_call {
395 struct ctdb_req_header hdr;
401 uint32_t calldatalen;
402 uint8_t data[1]; /* key[] followed by calldata[] */
405 struct ctdb_reply_call {
406 struct ctdb_req_header hdr;
412 struct ctdb_reply_error {
413 struct ctdb_req_header hdr;
419 struct ctdb_req_dmaster {
420 struct ctdb_req_header hdr;
429 struct ctdb_reply_dmaster {
430 struct ctdb_req_header hdr;
438 struct ctdb_req_message {
439 struct ctdb_req_header hdr;
445 struct ctdb_req_getdbpath {
446 struct ctdb_req_header hdr;
450 struct ctdb_reply_getdbpath {
451 struct ctdb_req_header hdr;
456 struct ctdb_req_control {
457 struct ctdb_req_header hdr;
462 #define CTDB_CTRL_FLAG_NOREPLY 1
463 #define CTDB_CTRL_FLAG_OPCODE_SPECIFIC 0xFFFF0000
469 struct ctdb_reply_control {
470 struct ctdb_req_header hdr;
477 struct ctdb_req_keepalive {
478 struct ctdb_req_header hdr;
482 /* types of failures possible from TRANS2_COMMIT */
483 enum ctdb_trans2_commit_error {
484 CTDB_TRANS2_COMMIT_SUCCESS=0, /* all nodes committed successfully */
485 CTDB_TRANS2_COMMIT_TIMEOUT=1, /* at least one node timed out */
486 CTDB_TRANS2_COMMIT_ALLFAIL=2, /* all nodes failed the commit */
487 CTDB_TRANS2_COMMIT_SOMEFAIL=3 /* some nodes failed the commit, some allowed it */
491 the extended header for records in the ltdb
493 struct ctdb_ltdb_header {
497 #define CTDB_REC_FLAG_DEFAULT 0x00000000
498 #define CTDB_REC_FLAG_MIGRATED_WITH_DATA 0x00010000
499 #define CTDB_REC_FLAG_VACUUM_MIGRATED 0x00020000
500 #define CTDB_REC_FLAG_AUTOMATIC 0x00040000
501 #define CTDB_REC_RO_HAVE_DELEGATIONS 0x01000000
502 #define CTDB_REC_RO_HAVE_READONLY 0x02000000
503 #define CTDB_REC_RO_REVOKING_READONLY 0x04000000
504 #define CTDB_REC_RO_REVOKE_COMPLETE 0x08000000
510 definitions for different socket structures
512 typedef struct sockaddr_in ctdb_addr_in;
513 typedef struct sockaddr_in6 ctdb_addr_in6;
521 A structure describing a single node, its flags and its address
523 struct ctdb_node_and_flags {
531 Structure used for a nodemap.
532 The nodemap is the structure containing a list of all nodes
533 known to the cluster and their associated flags.
535 struct ctdb_node_map {
537 struct ctdb_node_and_flags nodes[1];
543 #define NODE_FLAGS_DISCONNECTED 0x00000001 /* node isn't connected */
544 #define NODE_FLAGS_UNHEALTHY 0x00000002 /* monitoring says node is unhealthy */
545 #define NODE_FLAGS_PERMANENTLY_DISABLED 0x00000004 /* administrator has disabled node */
546 #define NODE_FLAGS_BANNED 0x00000008 /* recovery daemon has banned the node */
547 #define NODE_FLAGS_DELETED 0x00000010 /* this node has been deleted */
548 #define NODE_FLAGS_STOPPED 0x00000020 /* this node has been stopped */
549 #define NODE_FLAGS_DISABLED (NODE_FLAGS_UNHEALTHY|NODE_FLAGS_PERMANENTLY_DISABLED)
550 #define NODE_FLAGS_INACTIVE (NODE_FLAGS_DELETED|NODE_FLAGS_DISCONNECTED|NODE_FLAGS_BANNED|NODE_FLAGS_STOPPED)
553 struct ctdb_public_ip {
558 struct ctdb_all_public_ips {
560 struct ctdb_public_ip ips[1];
564 struct latency_counter {
572 structure used to pass record data between the child and parent
574 struct ctdb_rec_data {
582 struct ctdb_traverse_start {
588 struct ctdb_traverse_start_ext {
592 bool withemptyrecords;
596 ctdb statistics information
598 struct ctdb_statistics {
599 uint32_t num_clients;
602 uint32_t client_packets_sent;
603 uint32_t client_packets_recv;
604 uint32_t node_packets_sent;
605 uint32_t node_packets_recv;
606 uint32_t keepalive_packets_sent;
607 uint32_t keepalive_packets_recv;
611 uint32_t req_dmaster;
612 uint32_t reply_dmaster;
613 uint32_t reply_error;
614 uint32_t req_message;
615 uint32_t req_control;
616 uint32_t reply_control;
620 uint32_t req_message;
621 uint32_t req_control;
629 struct latency_counter ctdbd;
630 struct latency_counter recd;
632 uint32_t total_calls;
633 uint32_t pending_calls;
634 uint32_t lockwait_calls;
635 uint32_t pending_lockwait_calls;
636 uint32_t childwrite_calls;
637 uint32_t pending_childwrite_calls;
638 uint32_t memory_used;
639 uint32_t __last_counter; /* hack for control_statistics_all */
640 uint32_t max_hop_count;
641 struct latency_counter call_latency;
642 struct latency_counter lockwait_latency;
643 struct latency_counter childwrite_latency;
644 uint32_t num_recoveries;
645 struct timeval statistics_start_time;
646 struct timeval statistics_current_time;
650 * wire format for statistics history
652 struct ctdb_statistics_wire {
654 struct ctdb_statistics stats[1];
658 * wire format for interface list
661 #define CTDB_IFACE_SIZE IFNAMSIZ
663 #define CTDB_IFACE_SIZE 16
666 struct ctdb_iface_info {
667 char name[CTDB_IFACE_SIZE+2];
672 struct ctdb_ifaces_list {
674 struct ctdb_iface_info ifaces[1];