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/>.
22 #include "lib/util/dlinklist.h"
23 #include "system/network.h"
24 #include "system/filesys.h"
25 #include "../include/ctdb_private.h"
28 choose the transport we will use
30 int ctdb_set_transport(struct ctdb_context *ctdb, const char *transport)
32 ctdb->transport = talloc_strdup(ctdb, transport);
33 CTDB_NO_MEMORY(ctdb, ctdb->transport);
39 Check whether an ip is a valid node ip
40 Returns the node id for this ip address or -1
42 int ctdb_ip_to_nodeid(struct ctdb_context *ctdb, const char *nodeip)
46 for (nodeid=0;nodeid<ctdb->num_nodes;nodeid++) {
47 if (ctdb->nodes[nodeid]->flags & NODE_FLAGS_DELETED) {
50 if (!strcmp(ctdb->nodes[nodeid]->address.address, nodeip)) {
59 choose the recovery lock file
61 int ctdb_set_recovery_lock_file(struct ctdb_context *ctdb, const char *file)
63 if (ctdb->recovery_lock_file != NULL) {
64 talloc_free(ctdb->recovery_lock_file);
65 ctdb->recovery_lock_file = NULL;
69 DEBUG(DEBUG_ALERT,("Recovery lock file set to \"\". Disabling recovery lock checking\n"));
70 ctdb->tunable.verify_recovery_lock = 0;
74 ctdb->recovery_lock_file = talloc_strdup(ctdb, file);
75 CTDB_NO_MEMORY(ctdb, ctdb->recovery_lock_file);
81 add a node to the list of nodes
83 static int ctdb_add_node(struct ctdb_context *ctdb, char *nstr)
85 struct ctdb_node *node, **nodep;
87 nodep = talloc_realloc(ctdb, ctdb->nodes, struct ctdb_node *, ctdb->num_nodes+1);
88 CTDB_NO_MEMORY(ctdb, nodep);
91 nodep = &ctdb->nodes[ctdb->num_nodes];
92 (*nodep) = talloc_zero(ctdb->nodes, struct ctdb_node);
93 CTDB_NO_MEMORY(ctdb, *nodep);
96 if (ctdb_parse_address(ctdb, node, nstr, &node->address) != 0) {
100 node->name = talloc_asprintf(node, "%s:%u",
101 node->address.address,
103 /* this assumes that the nodes are kept in sorted order, and no gaps */
104 node->pnn = ctdb->num_nodes;
106 /* nodes start out disconnected and unhealthy */
107 node->flags = (NODE_FLAGS_DISCONNECTED | NODE_FLAGS_UNHEALTHY);
109 if (ctdb->address.address &&
110 ctdb_same_address(&ctdb->address, &node->address)) {
111 /* for automatic binding to interfaces, see tcp_connect.c */
112 ctdb->pnn = node->pnn;
116 node->dead_count = 0;
122 add an entry for a "deleted" node to the list of nodes.
123 a "deleted" node is a node that is commented out from the nodes file.
124 this is used to prevent that subsequent nodes in the nodes list
125 change their pnn value if a node is "delete" by commenting it out and then
126 using "ctdb reloadnodes" at runtime.
128 static int ctdb_add_deleted_node(struct ctdb_context *ctdb)
130 struct ctdb_node *node, **nodep;
132 nodep = talloc_realloc(ctdb, ctdb->nodes, struct ctdb_node *, ctdb->num_nodes+1);
133 CTDB_NO_MEMORY(ctdb, nodep);
136 nodep = &ctdb->nodes[ctdb->num_nodes];
137 (*nodep) = talloc_zero(ctdb->nodes, struct ctdb_node);
138 CTDB_NO_MEMORY(ctdb, *nodep);
141 if (ctdb_parse_address(ctdb, node, "0.0.0.0", &node->address) != 0) {
142 DEBUG(DEBUG_ERR,("Failed to setup deleted node %d\n", ctdb->num_nodes));
146 node->name = talloc_strdup(node, "0.0.0.0:0");
148 /* this assumes that the nodes are kept in sorted order, and no gaps */
149 node->pnn = ctdb->num_nodes;
151 /* this node is permanently deleted/disconnected */
152 node->flags = NODE_FLAGS_DELETED|NODE_FLAGS_DISCONNECTED;
155 node->dead_count = 0;
162 setup the node list from a file
164 static int ctdb_set_nlist(struct ctdb_context *ctdb, const char *nlist)
168 int i, j, num_present;
170 talloc_free(ctdb->nodes);
174 lines = file_lines_load(nlist, &nlines, ctdb);
176 ctdb_set_error(ctdb, "Failed to load nlist '%s'\n", nlist);
179 while (nlines > 0 && strcmp(lines[nlines-1], "") == 0) {
184 for (i=0; i < nlines; i++) {
188 /* strip leading spaces */
189 while((*node == ' ') || (*node == '\t')) {
193 if (ctdb_add_deleted_node(ctdb) != 0) {
199 if (strcmp(node, "") == 0) {
202 if (ctdb_add_node(ctdb, node) != 0) {
209 /* initialize the vnn mapping table now that we have the nodes list,
210 skipping any deleted nodes
212 ctdb->vnn_map = talloc(ctdb, struct ctdb_vnn_map);
213 CTDB_NO_MEMORY(ctdb, ctdb->vnn_map);
215 ctdb->vnn_map->generation = INVALID_GENERATION;
216 ctdb->vnn_map->size = num_present;
217 ctdb->vnn_map->map = talloc_array(ctdb->vnn_map, uint32_t, ctdb->vnn_map->size);
218 CTDB_NO_MEMORY(ctdb, ctdb->vnn_map->map);
220 for(i=0, j=0; i < ctdb->vnn_map->size; i++) {
221 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
224 ctdb->vnn_map->map[j] = i;
232 void ctdb_load_nodes_file(struct ctdb_context *ctdb)
236 ret = ctdb_set_nlist(ctdb, ctdb->nodes_file);
238 DEBUG(DEBUG_ALERT,("ctdb_set_nlist failed - %s\n", ctdb_errstr(ctdb)));
244 setup the local node address
246 int ctdb_set_address(struct ctdb_context *ctdb, const char *address)
248 if (ctdb_parse_address(ctdb, ctdb, address, &ctdb->address) != 0) {
252 ctdb->name = talloc_asprintf(ctdb, "%s:%u",
253 ctdb->address.address,
260 return the number of active nodes
262 uint32_t ctdb_get_num_active_nodes(struct ctdb_context *ctdb)
266 for (i=0; i < ctdb->num_nodes; i++) {
267 if (!(ctdb->nodes[i]->flags & NODE_FLAGS_INACTIVE)) {
276 called when we need to process a packet. This can be a requeued packet
277 after a lockwait, or a real packet from another node
279 void ctdb_input_pkt(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
283 /* place the packet as a child of the tmp_ctx. We then use
284 talloc_free() below to free it. If any of the calls want
285 to keep it, then they will steal it somewhere else, and the
286 talloc_free() will only free the tmp_ctx */
287 tmp_ctx = talloc_new(ctdb);
288 talloc_steal(tmp_ctx, hdr);
290 DEBUG(DEBUG_DEBUG,(__location__ " ctdb request %u of type %u length %u from "
291 "node %u to %u\n", hdr->reqid, hdr->operation, hdr->length,
292 hdr->srcnode, hdr->destnode));
294 switch (hdr->operation) {
296 case CTDB_REPLY_CALL:
297 case CTDB_REQ_DMASTER:
298 case CTDB_REPLY_DMASTER:
299 /* we dont allow these calls when banned */
300 if (ctdb->nodes[ctdb->pnn]->flags & NODE_FLAGS_BANNED) {
301 DEBUG(DEBUG_DEBUG,(__location__ " ctdb operation %u"
303 " length %u from node %u to %u while node"
305 hdr->operation, hdr->reqid,
307 hdr->srcnode, hdr->destnode));
311 /* for ctdb_call inter-node operations verify that the
312 remote node that sent us the call is running in the
313 same generation instance as this node
315 if (ctdb->vnn_map->generation != hdr->generation) {
316 DEBUG(DEBUG_DEBUG,(__location__ " ctdb operation %u"
318 " length %u from node %u to %u had an"
319 " invalid generation id:%u while our"
320 " generation id is:%u\n",
321 hdr->operation, hdr->reqid,
323 hdr->srcnode, hdr->destnode,
324 hdr->generation, ctdb->vnn_map->generation));
329 switch (hdr->operation) {
331 CTDB_INCREMENT_STAT(ctdb, node.req_call);
332 ctdb_request_call(ctdb, hdr);
335 case CTDB_REPLY_CALL:
336 CTDB_INCREMENT_STAT(ctdb, node.reply_call);
337 ctdb_reply_call(ctdb, hdr);
340 case CTDB_REPLY_ERROR:
341 CTDB_INCREMENT_STAT(ctdb, node.reply_error);
342 ctdb_reply_error(ctdb, hdr);
345 case CTDB_REQ_DMASTER:
346 CTDB_INCREMENT_STAT(ctdb, node.req_dmaster);
347 ctdb_request_dmaster(ctdb, hdr);
350 case CTDB_REPLY_DMASTER:
351 CTDB_INCREMENT_STAT(ctdb, node.reply_dmaster);
352 ctdb_reply_dmaster(ctdb, hdr);
355 case CTDB_REQ_MESSAGE:
356 CTDB_INCREMENT_STAT(ctdb, node.req_message);
357 ctdb_request_message(ctdb, hdr);
360 case CTDB_REQ_CONTROL:
361 CTDB_INCREMENT_STAT(ctdb, node.req_control);
362 ctdb_request_control(ctdb, hdr);
365 case CTDB_REPLY_CONTROL:
366 CTDB_INCREMENT_STAT(ctdb, node.reply_control);
367 ctdb_reply_control(ctdb, hdr);
370 case CTDB_REQ_KEEPALIVE:
371 CTDB_INCREMENT_STAT(ctdb, keepalive_packets_recv);
375 DEBUG(DEBUG_CRIT,("%s: Packet with unknown operation %u\n",
376 __location__, hdr->operation));
381 talloc_free(tmp_ctx);
386 called by the transport layer when a node is dead
388 void ctdb_node_dead(struct ctdb_node *node)
390 if (node->flags & NODE_FLAGS_DISCONNECTED) {
391 DEBUG(DEBUG_INFO,("%s: node %s is already marked disconnected: %u connected\n",
392 node->ctdb->name, node->name,
393 node->ctdb->num_connected));
396 node->ctdb->num_connected--;
397 node->flags |= NODE_FLAGS_DISCONNECTED | NODE_FLAGS_UNHEALTHY;
399 node->dead_count = 0;
401 DEBUG(DEBUG_NOTICE,("%s: node %s is dead: %u connected\n",
402 node->ctdb->name, node->name, node->ctdb->num_connected));
403 ctdb_daemon_cancel_controls(node->ctdb, node);
405 if (node->ctdb->methods == NULL) {
406 DEBUG(DEBUG_ERR,(__location__ " Can not restart transport while shutting down daemon.\n"));
410 node->ctdb->methods->restart(node);
414 called by the transport layer when a node is connected
416 void ctdb_node_connected(struct ctdb_node *node)
418 if (!(node->flags & NODE_FLAGS_DISCONNECTED)) {
419 DEBUG(DEBUG_INFO,("%s: node %s is already marked connected: %u connected\n",
420 node->ctdb->name, node->name,
421 node->ctdb->num_connected));
424 node->ctdb->num_connected++;
425 node->dead_count = 0;
426 node->flags &= ~NODE_FLAGS_DISCONNECTED;
427 node->flags |= NODE_FLAGS_UNHEALTHY;
429 ("%s: connected to %s - %u connected\n",
430 node->ctdb->name, node->name, node->ctdb->num_connected));
434 struct ctdb_context *ctdb;
435 struct ctdb_req_header *hdr;
440 triggered when a deferred packet is due
442 static void queue_next_trigger(struct event_context *ev, struct timed_event *te,
443 struct timeval t, void *private_data)
445 struct queue_next *q = talloc_get_type(private_data, struct queue_next);
446 ctdb_input_pkt(q->ctdb, q->hdr);
451 defer a packet, so it is processed on the next event loop
452 this is used for sending packets to ourselves
454 static void ctdb_defer_packet(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
456 struct queue_next *q;
457 q = talloc(ctdb, struct queue_next);
459 DEBUG(DEBUG_ERR,(__location__ " Failed to allocate deferred packet\n"));
463 q->hdr = talloc_memdup(ctdb, hdr, hdr->length);
464 if (q->hdr == NULL) {
465 DEBUG(DEBUG_ERR,("Error copying deferred packet to self\n"));
469 /* use this to put packets directly into our recv function */
470 ctdb_input_pkt(q->ctdb, q->hdr);
472 event_add_timed(ctdb->ev, q, timeval_zero(), queue_next_trigger, q);
478 broadcast a packet to all nodes
480 static void ctdb_broadcast_packet_all(struct ctdb_context *ctdb,
481 struct ctdb_req_header *hdr)
484 for (i=0; i < ctdb->num_nodes; i++) {
485 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
488 hdr->destnode = ctdb->nodes[i]->pnn;
489 ctdb_queue_packet(ctdb, hdr);
494 broadcast a packet to all nodes in the current vnnmap
496 static void ctdb_broadcast_packet_vnnmap(struct ctdb_context *ctdb,
497 struct ctdb_req_header *hdr)
500 for (i=0;i<ctdb->vnn_map->size;i++) {
501 hdr->destnode = ctdb->vnn_map->map[i];
502 ctdb_queue_packet(ctdb, hdr);
507 broadcast a packet to all connected nodes
509 static void ctdb_broadcast_packet_connected(struct ctdb_context *ctdb,
510 struct ctdb_req_header *hdr)
513 for (i=0; i < ctdb->num_nodes; i++) {
514 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
517 if (!(ctdb->nodes[i]->flags & NODE_FLAGS_DISCONNECTED)) {
518 hdr->destnode = ctdb->nodes[i]->pnn;
519 ctdb_queue_packet(ctdb, hdr);
525 queue a packet or die
527 void ctdb_queue_packet(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
529 struct ctdb_node *node;
531 switch (hdr->destnode) {
532 case CTDB_BROADCAST_ALL:
533 ctdb_broadcast_packet_all(ctdb, hdr);
535 case CTDB_BROADCAST_VNNMAP:
536 ctdb_broadcast_packet_vnnmap(ctdb, hdr);
538 case CTDB_BROADCAST_CONNECTED:
539 ctdb_broadcast_packet_connected(ctdb, hdr);
543 CTDB_INCREMENT_STAT(ctdb, node_packets_sent);
545 if (!ctdb_validate_pnn(ctdb, hdr->destnode)) {
546 DEBUG(DEBUG_CRIT,(__location__ " cant send to node %u that does not exist\n",
551 node = ctdb->nodes[hdr->destnode];
553 if (node->flags & NODE_FLAGS_DELETED) {
554 DEBUG(DEBUG_ERR, (__location__ " Can not queue packet to DELETED node %d\n", hdr->destnode));
558 if (node->pnn == ctdb->pnn) {
559 ctdb_defer_packet(ctdb, hdr);
563 if (ctdb->methods == NULL) {
564 DEBUG(DEBUG_ALERT, (__location__ " Can not queue packet. "
565 "Transport is DOWN\n"));
570 if (ctdb->methods->queue_pkt(node, (uint8_t *)hdr, hdr->length) != 0) {
571 ctdb_fatal(ctdb, "Unable to queue packet\n");
579 a valgrind hack to allow us to get opcode specific backtraces
580 very ugly, and relies on no compiler optimisation!
582 void ctdb_queue_packet_opcode(struct ctdb_context *ctdb, struct ctdb_req_header *hdr, unsigned opcode)
585 #define DO_OP(x) case x: ctdb_queue_packet(ctdb, hdr); break
687 ctdb_queue_packet(ctdb, hdr);