2 ctdb main protocol code
4 Copyright (C) Andrew Tridgell 2006
6 This library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Lesser General Public
8 License as published by the Free Software Foundation; either
9 version 2 of the License, or (at your option) any later version.
11 This library 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 GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public
17 License along with this library; if not, write to the Free Software
18 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 #include "lib/tdb/include/tdb.h"
23 #include "lib/events/events.h"
24 #include "lib/util/dlinklist.h"
25 #include "system/network.h"
26 #include "system/filesys.h"
27 #include "../include/ctdb_private.h"
30 choose the transport we will use
32 int ctdb_set_transport(struct ctdb_context *ctdb, const char *transport)
34 ctdb->transport = talloc_strdup(ctdb, transport);
39 choose the logfile location
41 int ctdb_set_logfile(struct ctdb_context *ctdb, const char *logfile)
43 ctdb->logfile = talloc_strdup(ctdb, logfile);
44 if (ctdb->logfile != NULL) {
47 fd = open(ctdb->logfile, O_WRONLY|O_APPEND|O_CREAT, 0666);
63 void ctdb_set_flags(struct ctdb_context *ctdb, unsigned flags)
71 void ctdb_clear_flags(struct ctdb_context *ctdb, unsigned flags)
73 ctdb->flags &= ~flags;
77 set max acess count before a dmaster migration
79 void ctdb_set_max_lacount(struct ctdb_context *ctdb, unsigned count)
81 ctdb->max_lacount = count;
85 set the directory for the local databases
87 int ctdb_set_tdb_dir(struct ctdb_context *ctdb, const char *dir)
90 ctdb->db_directory = talloc_asprintf(ctdb, "ctdb-%u", ctdb_get_vnn(ctdb));
92 ctdb->db_directory = talloc_strdup(ctdb, dir);
94 if (ctdb->db_directory == NULL) {
101 add a node to the list of active nodes
103 static int ctdb_add_node(struct ctdb_context *ctdb, char *nstr)
105 struct ctdb_node *node, **nodep;
107 nodep = talloc_realloc(ctdb, ctdb->nodes, struct ctdb_node *, ctdb->num_nodes+1);
108 CTDB_NO_MEMORY(ctdb, nodep);
111 nodep = &ctdb->nodes[ctdb->num_nodes];
112 (*nodep) = talloc_zero(ctdb->nodes, struct ctdb_node);
113 CTDB_NO_MEMORY(ctdb, *nodep);
116 if (ctdb_parse_address(ctdb, node, nstr, &node->address) != 0) {
120 node->name = talloc_asprintf(node, "%s:%u",
121 node->address.address,
123 /* for now we just set the vnn to the line in the file - this
125 node->vnn = ctdb->num_nodes;
127 if (ctdb_same_address(&ctdb->address, &node->address)) {
128 ctdb->vnn = node->vnn;
129 node->flags |= NODE_FLAGS_CONNECTED;
138 setup the node list from a file
140 int ctdb_set_nlist(struct ctdb_context *ctdb, const char *nlist)
146 lines = file_lines_load(nlist, &nlines, ctdb);
148 ctdb_set_error(ctdb, "Failed to load nlist '%s'\n", nlist);
152 for (i=0;i<nlines;i++) {
153 if (ctdb_add_node(ctdb, lines[i]) != 0) {
164 setup the local node address
166 int ctdb_set_address(struct ctdb_context *ctdb, const char *address)
168 if (ctdb_parse_address(ctdb, ctdb, address, &ctdb->address) != 0) {
172 ctdb->name = talloc_asprintf(ctdb, "%s:%u",
173 ctdb->address.address,
180 setup the local socket name
182 int ctdb_set_socketname(struct ctdb_context *ctdb, const char *socketname)
184 ctdb->daemon.name = talloc_strdup(ctdb, socketname);
188 return the vnn of this node
190 uint32_t ctdb_get_vnn(struct ctdb_context *ctdb)
196 return the number of nodes
198 uint32_t ctdb_get_num_nodes(struct ctdb_context *ctdb)
200 return ctdb->num_nodes;
205 called by the transport layer when a packet comes in
207 void ctdb_recv_pkt(struct ctdb_context *ctdb, uint8_t *data, uint32_t length)
209 struct ctdb_req_header *hdr = (struct ctdb_req_header *)data;
212 ctdb->status.node_packets_recv++;
214 /* place the packet as a child of the tmp_ctx. We then use
215 talloc_free() below to free it. If any of the calls want
216 to keep it, then they will steal it somewhere else, and the
217 talloc_free() will only free the tmp_ctx */
218 tmp_ctx = talloc_new(ctdb);
219 talloc_steal(tmp_ctx, hdr);
221 if (length < sizeof(*hdr)) {
222 ctdb_set_error(ctdb, "Bad packet length %d\n", length);
225 if (length != hdr->length) {
226 ctdb_set_error(ctdb, "Bad header length %d expected %d\n",
227 hdr->length, length);
231 if (hdr->ctdb_magic != CTDB_MAGIC) {
232 ctdb_set_error(ctdb, "Non CTDB packet rejected\n");
236 if (hdr->ctdb_version != CTDB_VERSION) {
237 ctdb_set_error(ctdb, "Bad CTDB version 0x%x rejected\n", hdr->ctdb_version);
241 DEBUG(3,(__location__ " ctdb request %d of type %d length %d from "
242 "node %d to %d\n", hdr->reqid, hdr->operation, hdr->length,
243 hdr->srcnode, hdr->destnode));
245 switch (hdr->operation) {
247 /* verify that the remote node that sent us the call
248 is running in the same generation instance as this node
250 if (ctdb->vnn_map->generation != hdr->generation) {
251 DEBUG(0,(__location__ " ctdb request %d of type"
252 " %d length %d from node %d to %d had an"
253 " invalid generation id:%d while our"
254 " generation id is:%d\n",
255 hdr->reqid, hdr->operation, hdr->length,
256 hdr->srcnode, hdr->destnode,
257 ctdb->vnn_map->generation,
261 ctdb->status.count.req_call++;
262 ctdb_request_call(ctdb, hdr);
265 case CTDB_REPLY_CALL:
266 ctdb->status.count.reply_call++;
267 ctdb_reply_call(ctdb, hdr);
270 case CTDB_REPLY_ERROR:
271 ctdb->status.count.reply_error++;
272 ctdb_reply_error(ctdb, hdr);
275 case CTDB_REQ_DMASTER:
276 ctdb->status.count.req_dmaster++;
277 ctdb_request_dmaster(ctdb, hdr);
280 case CTDB_REPLY_DMASTER:
281 ctdb->status.count.reply_dmaster++;
282 ctdb_reply_dmaster(ctdb, hdr);
285 case CTDB_REQ_MESSAGE:
286 ctdb->status.count.req_message++;
287 ctdb_request_message(ctdb, hdr);
290 case CTDB_REQ_FINISHED:
291 ctdb->status.count.req_finished++;
292 ctdb_request_finished(ctdb, hdr);
295 case CTDB_REQ_CONTROL:
296 ctdb->status.count.req_control++;
297 ctdb_request_control(ctdb, hdr);
300 case CTDB_REPLY_CONTROL:
301 ctdb->status.count.reply_control++;
302 ctdb_reply_control(ctdb, hdr);
306 DEBUG(0,("%s: Packet with unknown operation %d\n",
307 __location__, hdr->operation));
312 talloc_free(tmp_ctx);
316 called by the transport layer when a packet comes in
318 void ctdb_recv_raw_pkt(void *p, uint8_t *data, uint32_t length)
320 struct ctdb_context *ctdb = talloc_get_type(p, struct ctdb_context);
321 ctdb_recv_pkt(ctdb, data, length);
325 called by the transport layer when a node is dead
327 static void ctdb_node_dead(struct ctdb_node *node)
329 node->ctdb->num_connected--;
330 node->flags &= ~NODE_FLAGS_CONNECTED;
331 DEBUG(1,("%s: node %s is dead: %d connected\n",
332 node->ctdb->name, node->name, node->ctdb->num_connected));
336 called by the transport layer when a node is connected
338 static void ctdb_node_connected(struct ctdb_node *node)
340 node->ctdb->num_connected++;
341 node->flags |= NODE_FLAGS_CONNECTED;
342 DEBUG(1,("%s: connected to %s - %d connected\n",
343 node->ctdb->name, node->name, node->ctdb->num_connected));
347 wait for all nodes to be connected
349 void ctdb_daemon_connect_wait(struct ctdb_context *ctdb)
351 int expected = ctdb->num_nodes - 1;
352 if (ctdb->flags & CTDB_FLAG_SELF_CONNECT) {
355 while (ctdb->num_connected != expected) {
356 DEBUG(3,("ctdb_connect_wait: waiting for %d nodes (have %d)\n",
357 expected, ctdb->num_connected));
358 event_loop_once(ctdb->ev);
360 DEBUG(3,("ctdb_connect_wait: got all %d nodes\n", expected));
364 struct ctdb_context *ctdb;
365 struct ctdb_req_header *hdr;
370 trigered when a deferred packet is due
372 static void queue_next_trigger(struct event_context *ev, struct timed_event *te,
373 struct timeval t, void *private_data)
375 struct queue_next *q = talloc_get_type(private_data, struct queue_next);
376 ctdb_recv_pkt(q->ctdb, (uint8_t *)q->hdr, q->hdr->length);
381 defer a packet, so it is processed on the next event loop
382 this is used for sending packets to ourselves
384 static void ctdb_defer_packet(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
386 struct queue_next *q;
387 q = talloc(ctdb, struct queue_next);
389 DEBUG(0,(__location__ " Failed to allocate deferred packet\n"));
393 q->hdr = talloc_memdup(ctdb, hdr, hdr->length);
394 if (q->hdr == NULL) {
395 DEBUG(0,("Error copying deferred packet to self\n"));
399 /* use this to put packets directly into our recv function */
400 ctdb_recv_pkt(q->ctdb, (uint8_t *)q->hdr, q->hdr->length);
403 event_add_timed(ctdb->ev, q, timeval_zero(), queue_next_trigger, q);
409 broadcast a packet to all nodes
411 static void ctdb_broadcast_packet(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
414 for (i=0;i<ctdb_get_num_nodes(ctdb);i++) {
415 hdr->destnode = ctdb->nodes[i]->vnn;
416 ctdb_queue_packet(ctdb, hdr);
421 queue a packet or die
423 void ctdb_queue_packet(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
425 struct ctdb_node *node;
427 if (hdr->destnode == CTDB_BROADCAST_VNN) {
428 ctdb_broadcast_packet(ctdb, hdr);
432 ctdb->status.node_packets_sent++;
434 if (!ctdb_validate_vnn(ctdb, hdr->destnode)) {
435 DEBUG(0,(__location__ " cant send to node %u that does not exist\n",
440 node = ctdb->nodes[hdr->destnode];
442 if (hdr->destnode == ctdb->vnn && !(ctdb->flags & CTDB_FLAG_SELF_CONNECT)) {
443 ctdb_defer_packet(ctdb, hdr);
444 } else if (ctdb->methods->queue_pkt(node, (uint8_t *)hdr, hdr->length) != 0) {
445 ctdb_fatal(ctdb, "Unable to queue packet\n");
450 static const struct ctdb_upcalls ctdb_upcalls = {
451 .recv_pkt = ctdb_recv_pkt,
452 .node_dead = ctdb_node_dead,
453 .node_connected = ctdb_node_connected
457 initialise the ctdb daemon.
459 NOTE: In current code the daemon does not fork. This is for testing purposes only
460 and to simplify the code.
462 struct ctdb_context *ctdb_init(struct event_context *ev)
464 struct ctdb_context *ctdb;
466 ctdb = talloc_zero(ev, struct ctdb_context);
468 ctdb->upcalls = &ctdb_upcalls;
469 ctdb->idr = idr_init(ctdb);
470 ctdb->max_lacount = CTDB_DEFAULT_MAX_LACOUNT;