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 3 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 int ctdb_tcp_init(struct ctdb_context *ctdb);
36 int ctdb_ibw_init(struct ctdb_context *ctdb);
37 #endif /* USE_INFINIBAND */
39 if (strcmp(transport, "tcp") == 0) {
40 return ctdb_tcp_init(ctdb);
43 if (strcmp(transport, "ib") == 0) {
44 return ctdb_ibw_init(ctdb);
46 #endif /* USE_INFINIBAND */
48 ctdb_set_error(ctdb, "Unknown transport '%s'\n", transport);
55 void ctdb_set_flags(struct ctdb_context *ctdb, unsigned flags)
63 void ctdb_clear_flags(struct ctdb_context *ctdb, unsigned flags)
65 ctdb->flags &= ~flags;
69 set max acess count before a dmaster migration
71 void ctdb_set_max_lacount(struct ctdb_context *ctdb, unsigned count)
73 ctdb->max_lacount = count;
77 set the directory for the local databases
79 int ctdb_set_tdb_dir(struct ctdb_context *ctdb, const char *dir)
82 ctdb->db_directory = talloc_asprintf(ctdb, "ctdb-%u", ctdb_get_vnn(ctdb));
84 ctdb->db_directory = talloc_strdup(ctdb, dir);
86 if (ctdb->db_directory == NULL) {
93 add a node to the list of active nodes
95 static int ctdb_add_node(struct ctdb_context *ctdb, char *nstr)
97 struct ctdb_node *node, **nodep;
99 nodep = talloc_realloc(ctdb, ctdb->nodes, struct ctdb_node *, ctdb->num_nodes+1);
100 CTDB_NO_MEMORY(ctdb, nodep);
103 nodep = &ctdb->nodes[ctdb->num_nodes];
104 (*nodep) = talloc_zero(ctdb->nodes, struct ctdb_node);
105 CTDB_NO_MEMORY(ctdb, *nodep);
108 if (ctdb_parse_address(ctdb, node, nstr, &node->address) != 0) {
112 node->name = talloc_asprintf(node, "%s:%u",
113 node->address.address,
115 /* for now we just set the vnn to the line in the file - this
117 node->vnn = ctdb->num_nodes;
119 if (ctdb->methods->add_node(node) != 0) {
124 if (ctdb_same_address(&ctdb->address, &node->address)) {
125 ctdb->vnn = node->vnn;
134 setup the node list from a file
136 int ctdb_set_nlist(struct ctdb_context *ctdb, const char *nlist)
142 lines = file_lines_load(nlist, &nlines, ctdb);
144 ctdb_set_error(ctdb, "Failed to load nlist '%s'\n", nlist);
148 for (i=0;i<nlines;i++) {
149 if (ctdb_add_node(ctdb, lines[i]) != 0) {
160 setup the local node address
162 int ctdb_set_address(struct ctdb_context *ctdb, const char *address)
164 if (ctdb_parse_address(ctdb, ctdb, address, &ctdb->address) != 0) {
168 ctdb->name = talloc_asprintf(ctdb, "%s:%u",
169 ctdb->address.address,
175 add a node to the list of active nodes
177 int ctdb_set_call(struct ctdb_db_context *ctdb_db, ctdb_fn_t fn, int id)
179 struct ctdb_registered_call *call;
181 call = talloc(ctdb_db, struct ctdb_registered_call);
185 DLIST_ADD(ctdb_db->calls, call);
190 return the vnn of this node
192 uint32_t ctdb_get_vnn(struct ctdb_context *ctdb)
198 return the number of nodes
200 uint32_t ctdb_get_num_nodes(struct ctdb_context *ctdb)
202 return ctdb->num_nodes;
207 called by the transport layer when a packet comes in
209 void ctdb_recv_pkt(struct ctdb_context *ctdb, uint8_t *data, uint32_t length)
211 struct ctdb_req_header *hdr = (struct ctdb_req_header *)data;
214 ctdb->status.node_packets_recv++;
216 /* place the packet as a child of the tmp_ctx. We then use
217 talloc_free() below to free it. If any of the calls want
218 to keep it, then they will steal it somewhere else, and the
219 talloc_free() will only free the tmp_ctx */
220 tmp_ctx = talloc_new(ctdb);
221 talloc_steal(tmp_ctx, hdr);
223 if (length < sizeof(*hdr)) {
224 ctdb_set_error(ctdb, "Bad packet length %d\n", length);
227 if (length != hdr->length) {
228 ctdb_set_error(ctdb, "Bad header length %d expected %d\n",
229 hdr->length, length);
233 if (hdr->ctdb_magic != CTDB_MAGIC) {
234 ctdb_set_error(ctdb, "Non CTDB packet rejected\n");
238 if (hdr->ctdb_version != CTDB_VERSION) {
239 ctdb_set_error(ctdb, "Bad CTDB version 0x%x rejected\n", hdr->ctdb_version);
243 DEBUG(3,(__location__ " ctdb request %d of type %d length %d from "
244 "node %d to %d\n", hdr->reqid, hdr->operation, hdr->length,
245 hdr->srcnode, hdr->destnode));
247 switch (hdr->operation) {
249 ctdb->status.count.req_call++;
250 ctdb_request_call(ctdb, hdr);
253 case CTDB_REPLY_CALL:
254 ctdb->status.count.reply_call++;
255 ctdb_reply_call(ctdb, hdr);
258 case CTDB_REPLY_ERROR:
259 ctdb->status.count.reply_error++;
260 ctdb_reply_error(ctdb, hdr);
263 case CTDB_REPLY_REDIRECT:
264 ctdb->status.count.reply_redirect++;
265 ctdb_reply_redirect(ctdb, hdr);
268 case CTDB_REQ_DMASTER:
269 ctdb->status.count.req_dmaster++;
270 ctdb_request_dmaster(ctdb, hdr);
273 case CTDB_REPLY_DMASTER:
274 ctdb->status.count.reply_dmaster++;
275 ctdb_reply_dmaster(ctdb, hdr);
278 case CTDB_REQ_MESSAGE:
279 ctdb->status.count.req_message++;
280 ctdb_request_message(ctdb, hdr);
283 case CTDB_REQ_FINISHED:
284 ctdb->status.count.req_finished++;
285 ctdb_request_finished(ctdb, hdr);
289 DEBUG(0,("%s: Packet with unknown operation %d\n",
290 __location__, hdr->operation));
295 talloc_free(tmp_ctx);
299 called by the transport layer when a packet comes in
301 void ctdb_recv_raw_pkt(void *p, uint8_t *data, uint32_t length)
303 struct ctdb_context *ctdb = talloc_get_type(p, struct ctdb_context);
304 ctdb_recv_pkt(ctdb, data, length);
308 called by the transport layer when a node is dead
310 static void ctdb_node_dead(struct ctdb_node *node)
312 node->ctdb->num_connected--;
313 DEBUG(1,("%s: node %s is dead: %d connected\n",
314 node->ctdb->name, node->name, node->ctdb->num_connected));
318 called by the transport layer when a node is connected
320 static void ctdb_node_connected(struct ctdb_node *node)
322 node->ctdb->num_connected++;
323 DEBUG(1,("%s: connected to %s - %d connected\n",
324 node->ctdb->name, node->name, node->ctdb->num_connected));
328 wait for all nodes to be connected
330 void ctdb_daemon_connect_wait(struct ctdb_context *ctdb)
332 int expected = ctdb->num_nodes - 1;
333 if (ctdb->flags & CTDB_FLAG_SELF_CONNECT) {
336 while (ctdb->num_connected != expected) {
337 DEBUG(3,("ctdb_connect_wait: waiting for %d nodes (have %d)\n",
338 expected, ctdb->num_connected));
339 event_loop_once(ctdb->ev);
341 DEBUG(3,("ctdb_connect_wait: got all %d nodes\n", expected));
345 struct ctdb_context *ctdb;
346 struct ctdb_req_header *hdr;
351 trigered when a deferred packet is due
353 static void queue_next_trigger(struct event_context *ev, struct timed_event *te,
354 struct timeval t, void *private_data)
356 struct queue_next *q = talloc_get_type(private_data, struct queue_next);
357 ctdb_recv_pkt(q->ctdb, (uint8_t *)q->hdr, q->hdr->length);
362 defer a packet, so it is processed on the next event loop
363 this is used for sending packets to ourselves
365 static void ctdb_defer_packet(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
367 struct queue_next *q;
368 q = talloc(ctdb, struct queue_next);
370 DEBUG(0,(__location__ " Failed to allocate deferred packet\n"));
374 q->hdr = talloc_memdup(ctdb, hdr, hdr->length);
375 if (q->hdr == NULL) {
376 DEBUG(0,("Error copying deferred packet to self\n"));
379 event_add_timed(ctdb->ev, q, timeval_zero(), queue_next_trigger, q);
383 queue a packet or die
385 void ctdb_queue_packet(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
387 struct ctdb_node *node;
388 ctdb->status.node_packets_sent++;
389 node = ctdb->nodes[hdr->destnode];
390 if (hdr->destnode == ctdb->vnn && !(ctdb->flags & CTDB_FLAG_SELF_CONNECT)) {
391 ctdb_defer_packet(ctdb, hdr);
392 } else if (ctdb->methods->queue_pkt(node, (uint8_t *)hdr, hdr->length) != 0) {
393 ctdb_fatal(ctdb, "Unable to queue packet\n");
398 static const struct ctdb_upcalls ctdb_upcalls = {
399 .recv_pkt = ctdb_recv_pkt,
400 .node_dead = ctdb_node_dead,
401 .node_connected = ctdb_node_connected
405 initialise the ctdb daemon.
407 NOTE: In current code the daemon does not fork. This is for testing purposes only
408 and to simplify the code.
410 struct ctdb_context *ctdb_init(struct event_context *ev)
412 struct ctdb_context *ctdb;
414 ctdb = talloc_zero(ev, struct ctdb_context);
416 ctdb->upcalls = &ctdb_upcalls;
417 ctdb->idr = idr_init(ctdb);
418 ctdb->max_lacount = CTDB_DEFAULT_MAX_LACOUNT;