2 ctdb main protocol code
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/>.
21 #include "lib/tdb/include/tdb.h"
22 #include "lib/events/events.h"
23 #include "lib/util/dlinklist.h"
24 #include "system/network.h"
25 #include "system/filesys.h"
26 #include "../include/ctdb_private.h"
29 choose the transport we will use
31 int ctdb_set_transport(struct ctdb_context *ctdb, const char *transport)
33 ctdb->transport = talloc_strdup(ctdb, transport);
34 CTDB_NO_MEMORY(ctdb, ctdb->transport);
40 Check whether an ip is a valid node ip
41 Returns the node id for this ip address or -1
43 int ctdb_ip_to_nodeid(struct ctdb_context *ctdb, const char *nodeip)
47 for (nodeid=0;nodeid<ctdb->num_nodes;nodeid++) {
48 if (ctdb->nodes[nodeid]->flags & NODE_FLAGS_DELETED) {
51 if (!strcmp(ctdb->nodes[nodeid]->address.address, nodeip)) {
60 choose the recovery lock file
62 int ctdb_set_recovery_lock_file(struct ctdb_context *ctdb, const char *file)
64 if (ctdb->recovery_lock_file != NULL) {
65 talloc_free(ctdb->recovery_lock_file);
66 ctdb->recovery_lock_file = NULL;
70 DEBUG(DEBUG_ALERT,("Recovery lock file set to \"\". Disabling recovery lock checking\n"));
71 ctdb->tunable.verify_recovery_lock = 0;
75 ctdb->recovery_lock_file = talloc_strdup(ctdb, file);
76 CTDB_NO_MEMORY(ctdb, ctdb->recovery_lock_file);
82 set the directory for the local databases
84 int ctdb_set_tdb_dir(struct ctdb_context *ctdb, const char *dir)
86 ctdb->db_directory = talloc_strdup(ctdb, dir);
87 if (ctdb->db_directory == NULL) {
94 set the directory for the persistent databases
96 int ctdb_set_tdb_dir_persistent(struct ctdb_context *ctdb, const char *dir)
98 ctdb->db_directory_persistent = talloc_strdup(ctdb, dir);
99 if (ctdb->db_directory_persistent == NULL) {
106 add a node to the list of nodes
108 static int ctdb_add_node(struct ctdb_context *ctdb, char *nstr)
110 struct ctdb_node *node, **nodep;
112 nodep = talloc_realloc(ctdb, ctdb->nodes, struct ctdb_node *, ctdb->num_nodes+1);
113 CTDB_NO_MEMORY(ctdb, nodep);
116 nodep = &ctdb->nodes[ctdb->num_nodes];
117 (*nodep) = talloc_zero(ctdb->nodes, struct ctdb_node);
118 CTDB_NO_MEMORY(ctdb, *nodep);
121 if (ctdb_parse_address(ctdb, node, nstr, &node->address) != 0) {
125 node->name = talloc_asprintf(node, "%s:%u",
126 node->address.address,
128 /* this assumes that the nodes are kept in sorted order, and no gaps */
129 node->pnn = ctdb->num_nodes;
131 /* nodes start out disconnected and unhealthy */
132 node->flags = (NODE_FLAGS_DISCONNECTED | NODE_FLAGS_UNHEALTHY);
134 if (ctdb->address.address &&
135 ctdb_same_address(&ctdb->address, &node->address)) {
136 /* for automatic binding to interfaces, see tcp_connect.c */
137 ctdb->pnn = node->pnn;
138 node->flags &= ~NODE_FLAGS_DISCONNECTED;
140 /* do we start out in DISABLED mode? */
141 if (ctdb->start_as_disabled != 0) {
142 DEBUG(DEBUG_INFO, ("This node is configured to start in DISABLED state\n"));
143 node->flags |= NODE_FLAGS_DISABLED;
145 /* do we start out in STOPPED mode? */
146 if (ctdb->start_as_stopped != 0) {
147 DEBUG(DEBUG_INFO, ("This node is configured to start in STOPPED state\n"));
148 node->flags |= NODE_FLAGS_STOPPED;
153 node->dead_count = 0;
159 add an entry for a "deleted" node to the list of nodes.
160 a "deleted" node is a node that is commented out from the nodes file.
161 this is used to prevent that subsequent nodes in the nodes list
162 change their pnn value if a node is "delete" by commenting it out and then
163 using "ctdb reloadnodes" at runtime.
165 static int ctdb_add_deleted_node(struct ctdb_context *ctdb)
167 struct ctdb_node *node, **nodep;
169 nodep = talloc_realloc(ctdb, ctdb->nodes, struct ctdb_node *, ctdb->num_nodes+1);
170 CTDB_NO_MEMORY(ctdb, nodep);
173 nodep = &ctdb->nodes[ctdb->num_nodes];
174 (*nodep) = talloc_zero(ctdb->nodes, struct ctdb_node);
175 CTDB_NO_MEMORY(ctdb, *nodep);
178 if (ctdb_parse_address(ctdb, node, "0.0.0.0", &node->address) != 0) {
179 DEBUG(DEBUG_ERR,("Failed to setup deleted node %d\n", ctdb->num_nodes));
183 node->name = talloc_strdup(node, "0.0.0.0:0");
185 /* this assumes that the nodes are kept in sorted order, and no gaps */
186 node->pnn = ctdb->num_nodes;
188 /* this node is permanently deleted/disconnected */
189 node->flags = NODE_FLAGS_DELETED|NODE_FLAGS_DISCONNECTED;
192 node->dead_count = 0;
199 setup the node list from a file
201 int ctdb_set_nlist(struct ctdb_context *ctdb, const char *nlist)
205 int i, j, num_present;
207 talloc_free(ctdb->nodes);
211 lines = file_lines_load(nlist, &nlines, ctdb);
213 ctdb_set_error(ctdb, "Failed to load nlist '%s'\n", nlist);
216 while (nlines > 0 && strcmp(lines[nlines-1], "") == 0) {
221 for (i=0; i < nlines; i++) {
225 /* strip leading spaces */
226 while((*node == ' ') || (*node == '\t')) {
230 if (ctdb_add_deleted_node(ctdb) != 0) {
236 if (strcmp(node, "") == 0) {
239 if (ctdb_add_node(ctdb, node) != 0) {
246 /* initialize the vnn mapping table now that we have the nodes list,
247 skipping any deleted nodes
249 ctdb->vnn_map = talloc(ctdb, struct ctdb_vnn_map);
250 CTDB_NO_MEMORY(ctdb, ctdb->vnn_map);
252 ctdb->vnn_map->generation = INVALID_GENERATION;
253 ctdb->vnn_map->size = num_present;
254 ctdb->vnn_map->map = talloc_array(ctdb->vnn_map, uint32_t, ctdb->vnn_map->size);
255 CTDB_NO_MEMORY(ctdb, ctdb->vnn_map->map);
257 for(i=0, j=0; i < ctdb->vnn_map->size; i++) {
258 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
261 ctdb->vnn_map->map[j] = i;
271 setup the local node address
273 int ctdb_set_address(struct ctdb_context *ctdb, const char *address)
275 if (ctdb_parse_address(ctdb, ctdb, address, &ctdb->address) != 0) {
279 ctdb->name = talloc_asprintf(ctdb, "%s:%u",
280 ctdb->address.address,
287 return the number of active nodes
289 uint32_t ctdb_get_num_active_nodes(struct ctdb_context *ctdb)
293 for (i=0; i < ctdb->num_nodes; i++) {
294 if (!(ctdb->nodes[i]->flags & NODE_FLAGS_INACTIVE)) {
303 called when we need to process a packet. This can be a requeued packet
304 after a lockwait, or a real packet from another node
306 void ctdb_input_pkt(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
310 /* place the packet as a child of the tmp_ctx. We then use
311 talloc_free() below to free it. If any of the calls want
312 to keep it, then they will steal it somewhere else, and the
313 talloc_free() will only free the tmp_ctx */
314 tmp_ctx = talloc_new(ctdb);
315 talloc_steal(tmp_ctx, hdr);
317 DEBUG(DEBUG_DEBUG,(__location__ " ctdb request %u of type %u length %u from "
318 "node %u to %u\n", hdr->reqid, hdr->operation, hdr->length,
319 hdr->srcnode, hdr->destnode));
321 switch (hdr->operation) {
323 case CTDB_REPLY_CALL:
324 case CTDB_REQ_DMASTER:
325 case CTDB_REPLY_DMASTER:
326 /* we dont allow these calls when banned */
327 if (ctdb->nodes[ctdb->pnn]->flags & NODE_FLAGS_BANNED) {
328 DEBUG(DEBUG_DEBUG,(__location__ " ctdb operation %u"
330 " length %u from node %u to %u while node"
332 hdr->operation, hdr->reqid,
334 hdr->srcnode, hdr->destnode));
338 /* for ctdb_call inter-node operations verify that the
339 remote node that sent us the call is running in the
340 same generation instance as this node
342 if (ctdb->vnn_map->generation != hdr->generation) {
343 DEBUG(DEBUG_DEBUG,(__location__ " ctdb operation %u"
345 " length %u from node %u to %u had an"
346 " invalid generation id:%u while our"
347 " generation id is:%u\n",
348 hdr->operation, hdr->reqid,
350 hdr->srcnode, hdr->destnode,
351 hdr->generation, ctdb->vnn_map->generation));
356 switch (hdr->operation) {
358 ctdb->statistics.node.req_call++;
359 ctdb_request_call(ctdb, hdr);
362 case CTDB_REPLY_CALL:
363 ctdb->statistics.node.reply_call++;
364 ctdb_reply_call(ctdb, hdr);
367 case CTDB_REPLY_ERROR:
368 ctdb->statistics.node.reply_error++;
369 ctdb_reply_error(ctdb, hdr);
372 case CTDB_REQ_DMASTER:
373 ctdb->statistics.node.req_dmaster++;
374 ctdb_request_dmaster(ctdb, hdr);
377 case CTDB_REPLY_DMASTER:
378 ctdb->statistics.node.reply_dmaster++;
379 ctdb_reply_dmaster(ctdb, hdr);
382 case CTDB_REQ_MESSAGE:
383 ctdb->statistics.node.req_message++;
384 ctdb_request_message(ctdb, hdr);
387 case CTDB_REQ_CONTROL:
388 ctdb->statistics.node.req_control++;
389 ctdb_request_control(ctdb, hdr);
392 case CTDB_REPLY_CONTROL:
393 ctdb->statistics.node.reply_control++;
394 ctdb_reply_control(ctdb, hdr);
397 case CTDB_REQ_KEEPALIVE:
398 ctdb->statistics.keepalive_packets_recv++;
402 DEBUG(DEBUG_CRIT,("%s: Packet with unknown operation %u\n",
403 __location__, hdr->operation));
408 talloc_free(tmp_ctx);
413 called by the transport layer when a node is dead
415 void ctdb_node_dead(struct ctdb_node *node)
417 if (node->flags & NODE_FLAGS_DISCONNECTED) {
418 DEBUG(DEBUG_INFO,("%s: node %s is already marked disconnected: %u connected\n",
419 node->ctdb->name, node->name,
420 node->ctdb->num_connected));
423 node->ctdb->num_connected--;
424 node->flags |= NODE_FLAGS_DISCONNECTED | NODE_FLAGS_UNHEALTHY;
426 node->dead_count = 0;
428 DEBUG(DEBUG_NOTICE,("%s: node %s is dead: %u connected\n",
429 node->ctdb->name, node->name, node->ctdb->num_connected));
430 ctdb_daemon_cancel_controls(node->ctdb, node);
432 if (node->ctdb->methods == NULL) {
433 DEBUG(DEBUG_ERR,(__location__ " Can not restart transport while shutting down daemon.\n"));
437 node->ctdb->methods->restart(node);
441 called by the transport layer when a node is connected
443 void ctdb_node_connected(struct ctdb_node *node)
445 if (!(node->flags & NODE_FLAGS_DISCONNECTED)) {
446 DEBUG(DEBUG_INFO,("%s: node %s is already marked connected: %u connected\n",
447 node->ctdb->name, node->name,
448 node->ctdb->num_connected));
451 node->ctdb->num_connected++;
452 node->dead_count = 0;
453 node->flags &= ~NODE_FLAGS_DISCONNECTED;
454 node->flags |= NODE_FLAGS_UNHEALTHY;
455 DEBUG(DEBUG_INFO,("%s: connected to %s - %u connected\n",
456 node->ctdb->name, node->name, node->ctdb->num_connected));
460 struct ctdb_context *ctdb;
461 struct ctdb_req_header *hdr;
466 trigered when a deferred packet is due
468 static void queue_next_trigger(struct event_context *ev, struct timed_event *te,
469 struct timeval t, void *private_data)
471 struct queue_next *q = talloc_get_type(private_data, struct queue_next);
472 ctdb_input_pkt(q->ctdb, q->hdr);
477 defer a packet, so it is processed on the next event loop
478 this is used for sending packets to ourselves
480 static void ctdb_defer_packet(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
482 struct queue_next *q;
483 q = talloc(ctdb, struct queue_next);
485 DEBUG(DEBUG_ERR,(__location__ " Failed to allocate deferred packet\n"));
489 q->hdr = talloc_memdup(ctdb, hdr, hdr->length);
490 if (q->hdr == NULL) {
491 DEBUG(DEBUG_ERR,("Error copying deferred packet to self\n"));
495 /* use this to put packets directly into our recv function */
496 ctdb_input_pkt(q->ctdb, q->hdr);
498 event_add_timed(ctdb->ev, q, timeval_zero(), queue_next_trigger, q);
504 broadcast a packet to all nodes
506 static void ctdb_broadcast_packet_all(struct ctdb_context *ctdb,
507 struct ctdb_req_header *hdr)
510 for (i=0; i < ctdb->num_nodes; i++) {
511 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
514 hdr->destnode = ctdb->nodes[i]->pnn;
515 ctdb_queue_packet(ctdb, hdr);
520 broadcast a packet to all nodes in the current vnnmap
522 static void ctdb_broadcast_packet_vnnmap(struct ctdb_context *ctdb,
523 struct ctdb_req_header *hdr)
526 for (i=0;i<ctdb->vnn_map->size;i++) {
527 hdr->destnode = ctdb->vnn_map->map[i];
528 ctdb_queue_packet(ctdb, hdr);
533 broadcast a packet to all connected nodes
535 static void ctdb_broadcast_packet_connected(struct ctdb_context *ctdb,
536 struct ctdb_req_header *hdr)
539 for (i=0; i < ctdb->num_nodes; i++) {
540 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
543 if (!(ctdb->nodes[i]->flags & NODE_FLAGS_DISCONNECTED)) {
544 hdr->destnode = ctdb->nodes[i]->pnn;
545 ctdb_queue_packet(ctdb, hdr);
551 queue a packet or die
553 void ctdb_queue_packet(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
555 struct ctdb_node *node;
557 switch (hdr->destnode) {
558 case CTDB_BROADCAST_ALL:
559 ctdb_broadcast_packet_all(ctdb, hdr);
561 case CTDB_BROADCAST_VNNMAP:
562 ctdb_broadcast_packet_vnnmap(ctdb, hdr);
564 case CTDB_BROADCAST_CONNECTED:
565 ctdb_broadcast_packet_connected(ctdb, hdr);
569 ctdb->statistics.node_packets_sent++;
571 if (!ctdb_validate_pnn(ctdb, hdr->destnode)) {
572 DEBUG(DEBUG_CRIT,(__location__ " cant send to node %u that does not exist\n",
577 node = ctdb->nodes[hdr->destnode];
579 if (node->flags & NODE_FLAGS_DELETED) {
580 DEBUG(DEBUG_ERR, (__location__ " Can not queue packet to DELETED node %d\n", hdr->destnode));
584 if (node->pnn == ctdb->pnn) {
585 ctdb_defer_packet(ctdb, hdr);
587 if (ctdb->methods == NULL) {
588 DEBUG(DEBUG_ALERT, (__location__ " Can not queue packet. Transport is DOWN\n"));
593 if (ctdb->methods->queue_pkt(node, (uint8_t *)hdr, hdr->length) != 0) {
594 ctdb_fatal(ctdb, "Unable to queue packet\n");
603 a valgrind hack to allow us to get opcode specific backtraces
604 very ugly, and relies on no compiler optimisation!
606 void ctdb_queue_packet_opcode(struct ctdb_context *ctdb, struct ctdb_req_header *hdr, unsigned opcode)
609 #define DO_OP(x) case x: ctdb_queue_packet(ctdb, hdr); break
711 ctdb_queue_packet(ctdb, hdr);