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/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 "system/wait.h"
28 #include "../include/ctdb.h"
29 #include "../include/ctdb_private.h"
30 #include <sys/socket.h>
32 static void daemon_incoming_packet(void *, struct ctdb_req_header *);
34 static void print_exit_message(void)
36 DEBUG(DEBUG_NOTICE,("CTDB daemon shutting down\n"));
40 /* called when the "startup" event script has finished */
41 static void ctdb_start_transport(struct ctdb_context *ctdb)
43 if (ctdb->methods == NULL) {
44 DEBUG(DEBUG_ALERT,(__location__ " startup event finished but transport is DOWN.\n"));
45 ctdb_fatal(ctdb, "transport is not initialized but startup completed");
48 /* start the transport running */
49 if (ctdb->methods->start(ctdb) != 0) {
50 DEBUG(DEBUG_ALERT,("transport failed to start!\n"));
51 ctdb_fatal(ctdb, "transport failed to start");
54 /* start the recovery daemon process */
55 if (ctdb_start_recoverd(ctdb) != 0) {
56 DEBUG(DEBUG_ALERT,("Failed to start recovery daemon\n"));
60 /* Make sure we log something when the daemon terminates */
61 atexit(print_exit_message);
63 /* start monitoring for connected/disconnected nodes */
64 ctdb_start_keepalive(ctdb);
66 /* start monitoring for node health */
67 ctdb_start_monitoring(ctdb);
69 /* start periodic update of tcp tickle lists */
70 ctdb_start_tcp_tickle_update(ctdb);
72 /* start listening for recovery daemon pings */
73 ctdb_control_recd_ping(ctdb);
76 static void block_signal(int signum)
80 memset(&act, 0, sizeof(act));
82 act.sa_handler = SIG_IGN;
83 sigemptyset(&act.sa_mask);
84 sigaddset(&act.sa_mask, signum);
85 sigaction(signum, &act, NULL);
90 send a packet to a client
92 static int daemon_queue_send(struct ctdb_client *client, struct ctdb_req_header *hdr)
94 client->ctdb->statistics.client_packets_sent++;
95 if (hdr->operation == CTDB_REQ_MESSAGE) {
96 if (ctdb_queue_length(client->queue) > client->ctdb->tunable.max_queue_depth_drop_msg) {
97 DEBUG(DEBUG_ERR,("Drop CTDB_REQ_MESSAGE to client. Queue full.\n"));
101 return ctdb_queue_send(client->queue, (uint8_t *)hdr, hdr->length);
105 message handler for when we are in daemon mode. This redirects the message
108 static void daemon_message_handler(struct ctdb_context *ctdb, uint64_t srvid,
109 TDB_DATA data, void *private_data)
111 struct ctdb_client *client = talloc_get_type(private_data, struct ctdb_client);
112 struct ctdb_req_message *r;
115 /* construct a message to send to the client containing the data */
116 len = offsetof(struct ctdb_req_message, data) + data.dsize;
117 r = ctdbd_allocate_pkt(ctdb, ctdb, CTDB_REQ_MESSAGE,
118 len, struct ctdb_req_message);
119 CTDB_NO_MEMORY_VOID(ctdb, r);
121 talloc_set_name_const(r, "req_message packet");
124 r->datalen = data.dsize;
125 memcpy(&r->data[0], data.dptr, data.dsize);
127 daemon_queue_send(client, &r->hdr);
134 this is called when the ctdb daemon received a ctdb request to
135 set the srvid from the client
137 int daemon_register_message_handler(struct ctdb_context *ctdb, uint32_t client_id, uint64_t srvid)
139 struct ctdb_client *client = ctdb_reqid_find(ctdb, client_id, struct ctdb_client);
141 if (client == NULL) {
142 DEBUG(DEBUG_ERR,("Bad client_id in daemon_request_register_message_handler\n"));
145 res = ctdb_register_message_handler(ctdb, client, srvid, daemon_message_handler, client);
147 DEBUG(DEBUG_ERR,(__location__ " Failed to register handler %llu in daemon\n",
148 (unsigned long long)srvid));
150 DEBUG(DEBUG_INFO,(__location__ " Registered message handler for srvid=%llu\n",
151 (unsigned long long)srvid));
154 /* this is a hack for Samba - we now know the pid of the Samba client */
155 if ((srvid & 0xFFFFFFFF) == srvid &&
156 kill(srvid, 0) == 0) {
158 DEBUG(DEBUG_INFO,(__location__ " Registered PID %u for client %u\n",
159 (unsigned)client->pid, client_id));
165 this is called when the ctdb daemon received a ctdb request to
166 remove a srvid from the client
168 int daemon_deregister_message_handler(struct ctdb_context *ctdb, uint32_t client_id, uint64_t srvid)
170 struct ctdb_client *client = ctdb_reqid_find(ctdb, client_id, struct ctdb_client);
171 if (client == NULL) {
172 DEBUG(DEBUG_ERR,("Bad client_id in daemon_request_deregister_message_handler\n"));
175 return ctdb_deregister_message_handler(ctdb, srvid, client);
180 destroy a ctdb_client
182 static int ctdb_client_destructor(struct ctdb_client *client)
184 struct ctdb_db_context *ctdb_db;
186 ctdb_takeover_client_destructor_hook(client);
187 ctdb_reqid_remove(client->ctdb, client->client_id);
188 if (client->ctdb->statistics.num_clients) {
189 client->ctdb->statistics.num_clients--;
192 if (client->num_persistent_updates != 0) {
193 DEBUG(DEBUG_ERR,(__location__ " Client disconnecting with %u persistent updates in flight. Starting recovery\n", client->num_persistent_updates));
194 client->ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
196 ctdb_db = find_ctdb_db(client->ctdb, client->db_id);
198 DEBUG(DEBUG_ERR, (__location__ " client exit while transaction "
199 "commit active. Forcing recovery.\n"));
200 client->ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
201 ctdb_db->transaction_active = false;
209 this is called when the ctdb daemon received a ctdb request message
210 from a local client over the unix domain socket
212 static void daemon_request_message_from_client(struct ctdb_client *client,
213 struct ctdb_req_message *c)
218 /* maybe the message is for another client on this node */
219 if (ctdb_get_pnn(client->ctdb)==c->hdr.destnode) {
220 ctdb_request_message(client->ctdb, (struct ctdb_req_header *)c);
224 /* its for a remote node */
225 data.dptr = &c->data[0];
226 data.dsize = c->datalen;
227 res = ctdb_daemon_send_message(client->ctdb, c->hdr.destnode,
230 DEBUG(DEBUG_ERR,(__location__ " Failed to send message to remote node %u\n",
236 struct daemon_call_state {
237 struct ctdb_client *client;
239 struct ctdb_call *call;
240 struct timeval start_time;
244 complete a call from a client
246 static void daemon_call_from_client_callback(struct ctdb_call_state *state)
248 struct daemon_call_state *dstate = talloc_get_type(state->async.private_data,
249 struct daemon_call_state);
250 struct ctdb_reply_call *r;
253 struct ctdb_client *client = dstate->client;
254 struct ctdb_db_context *ctdb_db = state->ctdb_db;
256 talloc_steal(client, dstate);
257 talloc_steal(dstate, dstate->call);
259 res = ctdb_daemon_call_recv(state, dstate->call);
261 DEBUG(DEBUG_ERR, (__location__ " ctdbd_call_recv() returned error\n"));
262 if (client->ctdb->statistics.pending_calls > 0) {
263 client->ctdb->statistics.pending_calls--;
265 ctdb_latency(ctdb_db, "call_from_client_cb 1", &client->ctdb->statistics.max_call_latency, dstate->start_time);
269 length = offsetof(struct ctdb_reply_call, data) + dstate->call->reply_data.dsize;
270 r = ctdbd_allocate_pkt(client->ctdb, dstate, CTDB_REPLY_CALL,
271 length, struct ctdb_reply_call);
273 DEBUG(DEBUG_ERR, (__location__ " Failed to allocate reply_call in ctdb daemon\n"));
274 if (client->ctdb->statistics.pending_calls > 0) {
275 client->ctdb->statistics.pending_calls--;
277 ctdb_latency(ctdb_db, "call_from_client_cb 2", &client->ctdb->statistics.max_call_latency, dstate->start_time);
280 r->hdr.reqid = dstate->reqid;
281 r->datalen = dstate->call->reply_data.dsize;
282 memcpy(&r->data[0], dstate->call->reply_data.dptr, r->datalen);
284 res = daemon_queue_send(client, &r->hdr);
286 DEBUG(DEBUG_ERR, (__location__ " Failed to queue packet from daemon to client\n"));
288 ctdb_latency(ctdb_db, "call_from_client_cb 3", &client->ctdb->statistics.max_call_latency, dstate->start_time);
290 if (client->ctdb->statistics.pending_calls > 0) {
291 client->ctdb->statistics.pending_calls--;
295 struct ctdb_daemon_packet_wrap {
296 struct ctdb_context *ctdb;
301 a wrapper to catch disconnected clients
303 static void daemon_incoming_packet_wrap(void *p, struct ctdb_req_header *hdr)
305 struct ctdb_client *client;
306 struct ctdb_daemon_packet_wrap *w = talloc_get_type(p,
307 struct ctdb_daemon_packet_wrap);
309 DEBUG(DEBUG_CRIT,(__location__ " Bad packet type '%s'\n", talloc_get_name(p)));
313 client = ctdb_reqid_find(w->ctdb, w->client_id, struct ctdb_client);
314 if (client == NULL) {
315 DEBUG(DEBUG_ERR,(__location__ " Packet for disconnected client %u\n",
323 daemon_incoming_packet(client, hdr);
328 this is called when the ctdb daemon received a ctdb request call
329 from a local client over the unix domain socket
331 static void daemon_request_call_from_client(struct ctdb_client *client,
332 struct ctdb_req_call *c)
334 struct ctdb_call_state *state;
335 struct ctdb_db_context *ctdb_db;
336 struct daemon_call_state *dstate;
337 struct ctdb_call *call;
338 struct ctdb_ltdb_header header;
341 struct ctdb_context *ctdb = client->ctdb;
342 struct ctdb_daemon_packet_wrap *w;
344 ctdb->statistics.total_calls++;
345 if (client->ctdb->statistics.pending_calls > 0) {
346 ctdb->statistics.pending_calls++;
349 ctdb_db = find_ctdb_db(client->ctdb, c->db_id);
351 DEBUG(DEBUG_ERR, (__location__ " Unknown database in request. db_id==0x%08x",
353 if (client->ctdb->statistics.pending_calls > 0) {
354 ctdb->statistics.pending_calls--;
360 key.dsize = c->keylen;
362 w = talloc(ctdb, struct ctdb_daemon_packet_wrap);
363 CTDB_NO_MEMORY_VOID(ctdb, w);
366 w->client_id = client->client_id;
368 ret = ctdb_ltdb_lock_fetch_requeue(ctdb_db, key, &header,
369 (struct ctdb_req_header *)c, &data,
370 daemon_incoming_packet_wrap, w, True);
372 /* will retry later */
373 if (client->ctdb->statistics.pending_calls > 0) {
374 ctdb->statistics.pending_calls--;
382 DEBUG(DEBUG_ERR,(__location__ " Unable to fetch record\n"));
383 if (client->ctdb->statistics.pending_calls > 0) {
384 ctdb->statistics.pending_calls--;
389 dstate = talloc(client, struct daemon_call_state);
390 if (dstate == NULL) {
391 ctdb_ltdb_unlock(ctdb_db, key);
392 DEBUG(DEBUG_ERR,(__location__ " Unable to allocate dstate\n"));
393 if (client->ctdb->statistics.pending_calls > 0) {
394 ctdb->statistics.pending_calls--;
398 dstate->start_time = timeval_current();
399 dstate->client = client;
400 dstate->reqid = c->hdr.reqid;
401 talloc_steal(dstate, data.dptr);
403 call = dstate->call = talloc_zero(dstate, struct ctdb_call);
405 ctdb_ltdb_unlock(ctdb_db, key);
406 DEBUG(DEBUG_ERR,(__location__ " Unable to allocate call\n"));
407 if (client->ctdb->statistics.pending_calls > 0) {
408 ctdb->statistics.pending_calls--;
410 ctdb_latency(ctdb_db, "call_from_client 1", &ctdb->statistics.max_call_latency, dstate->start_time);
414 call->call_id = c->callid;
416 call->call_data.dptr = c->data + c->keylen;
417 call->call_data.dsize = c->calldatalen;
418 call->flags = c->flags;
420 if (header.dmaster == ctdb->pnn) {
421 state = ctdb_call_local_send(ctdb_db, call, &header, &data);
423 state = ctdb_daemon_call_send_remote(ctdb_db, call, &header);
426 ctdb_ltdb_unlock(ctdb_db, key);
429 DEBUG(DEBUG_ERR,(__location__ " Unable to setup call send\n"));
430 if (client->ctdb->statistics.pending_calls > 0) {
431 ctdb->statistics.pending_calls--;
433 ctdb_latency(ctdb_db, "call_from_client 2", &ctdb->statistics.max_call_latency, dstate->start_time);
436 talloc_steal(state, dstate);
437 talloc_steal(client, state);
439 state->async.fn = daemon_call_from_client_callback;
440 state->async.private_data = dstate;
444 static void daemon_request_control_from_client(struct ctdb_client *client,
445 struct ctdb_req_control *c);
447 /* data contains a packet from the client */
448 static void daemon_incoming_packet(void *p, struct ctdb_req_header *hdr)
450 struct ctdb_client *client = talloc_get_type(p, struct ctdb_client);
452 struct ctdb_context *ctdb = client->ctdb;
454 /* place the packet as a child of a tmp_ctx. We then use
455 talloc_free() below to free it. If any of the calls want
456 to keep it, then they will steal it somewhere else, and the
457 talloc_free() will be a no-op */
458 tmp_ctx = talloc_new(client);
459 talloc_steal(tmp_ctx, hdr);
461 if (hdr->ctdb_magic != CTDB_MAGIC) {
462 ctdb_set_error(client->ctdb, "Non CTDB packet rejected in daemon\n");
466 if (hdr->ctdb_version != CTDB_VERSION) {
467 ctdb_set_error(client->ctdb, "Bad CTDB version 0x%x rejected in daemon\n", hdr->ctdb_version);
471 switch (hdr->operation) {
473 ctdb->statistics.client.req_call++;
474 daemon_request_call_from_client(client, (struct ctdb_req_call *)hdr);
477 case CTDB_REQ_MESSAGE:
478 ctdb->statistics.client.req_message++;
479 daemon_request_message_from_client(client, (struct ctdb_req_message *)hdr);
482 case CTDB_REQ_CONTROL:
483 ctdb->statistics.client.req_control++;
484 daemon_request_control_from_client(client, (struct ctdb_req_control *)hdr);
488 DEBUG(DEBUG_CRIT,(__location__ " daemon: unrecognized operation %u\n",
493 talloc_free(tmp_ctx);
497 called when the daemon gets a incoming packet
499 static void ctdb_daemon_read_cb(uint8_t *data, size_t cnt, void *args)
501 struct ctdb_client *client = talloc_get_type(args, struct ctdb_client);
502 struct ctdb_req_header *hdr;
509 client->ctdb->statistics.client_packets_recv++;
511 if (cnt < sizeof(*hdr)) {
512 ctdb_set_error(client->ctdb, "Bad packet length %u in daemon\n",
516 hdr = (struct ctdb_req_header *)data;
517 if (cnt != hdr->length) {
518 ctdb_set_error(client->ctdb, "Bad header length %u expected %u\n in daemon",
519 (unsigned)hdr->length, (unsigned)cnt);
523 if (hdr->ctdb_magic != CTDB_MAGIC) {
524 ctdb_set_error(client->ctdb, "Non CTDB packet rejected\n");
528 if (hdr->ctdb_version != CTDB_VERSION) {
529 ctdb_set_error(client->ctdb, "Bad CTDB version 0x%x rejected in daemon\n", hdr->ctdb_version);
533 DEBUG(DEBUG_DEBUG,(__location__ " client request %u of type %u length %u from "
534 "node %u to %u\n", hdr->reqid, hdr->operation, hdr->length,
535 hdr->srcnode, hdr->destnode));
537 /* it is the responsibility of the incoming packet function to free 'data' */
538 daemon_incoming_packet(client, hdr);
541 static void ctdb_accept_client(struct event_context *ev, struct fd_event *fde,
542 uint16_t flags, void *private_data)
544 struct sockaddr_un addr;
547 struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
548 struct ctdb_client *client;
550 struct peercred_struct cr;
551 socklen_t crl = sizeof(struct peercred_struct);
554 socklen_t crl = sizeof(struct ucred);
557 memset(&addr, 0, sizeof(addr));
559 fd = accept(ctdb->daemon.sd, (struct sockaddr *)&addr, &len);
565 set_close_on_exec(fd);
567 DEBUG(DEBUG_DEBUG,(__location__ " Created SOCKET FD:%d to connected child\n", fd));
569 client = talloc_zero(ctdb, struct ctdb_client);
571 if (getsockopt(fd, SOL_SOCKET, SO_PEERID, &cr, &crl) == 0) {
573 if (getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &cr, &crl) == 0) {
575 talloc_asprintf(client, "struct ctdb_client: pid:%u", (unsigned)cr.pid);
580 client->client_id = ctdb_reqid_new(ctdb, client);
581 ctdb->statistics.num_clients++;
583 client->queue = ctdb_queue_setup(ctdb, client, fd, CTDB_DS_ALIGNMENT,
584 ctdb_daemon_read_cb, client);
586 talloc_set_destructor(client, ctdb_client_destructor);
592 create a unix domain socket and bind it
593 return a file descriptor open on the socket
595 static int ux_socket_bind(struct ctdb_context *ctdb)
597 struct sockaddr_un addr;
599 ctdb->daemon.sd = socket(AF_UNIX, SOCK_STREAM, 0);
600 if (ctdb->daemon.sd == -1) {
604 set_close_on_exec(ctdb->daemon.sd);
605 set_nonblocking(ctdb->daemon.sd);
607 memset(&addr, 0, sizeof(addr));
608 addr.sun_family = AF_UNIX;
609 strncpy(addr.sun_path, ctdb->daemon.name, sizeof(addr.sun_path));
611 if (bind(ctdb->daemon.sd, (struct sockaddr *)&addr, sizeof(addr)) == -1) {
612 DEBUG(DEBUG_CRIT,("Unable to bind on ctdb socket '%s'\n", ctdb->daemon.name));
616 if (chown(ctdb->daemon.name, geteuid(), getegid()) != 0 ||
617 chmod(ctdb->daemon.name, 0700) != 0) {
618 DEBUG(DEBUG_CRIT,("Unable to secure ctdb socket '%s', ctdb->daemon.name\n", ctdb->daemon.name));
623 if (listen(ctdb->daemon.sd, 100) != 0) {
624 DEBUG(DEBUG_CRIT,("Unable to listen on ctdb socket '%s'\n", ctdb->daemon.name));
631 close(ctdb->daemon.sd);
632 ctdb->daemon.sd = -1;
636 static void sig_child_handler(struct event_context *ev,
637 struct signal_event *se, int signum, int count,
641 // struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
646 pid = waitpid(-1, &status, WNOHANG);
648 DEBUG(DEBUG_ERR, (__location__ " waitpid() returned error. errno:%d\n", errno));
652 DEBUG(DEBUG_DEBUG, ("SIGCHLD from %d\n", (int)pid));
658 start the protocol going as a daemon
660 int ctdb_start_daemon(struct ctdb_context *ctdb, bool do_fork)
663 struct fd_event *fde;
664 const char *domain_socket_name;
665 struct signal_event *se;
667 /* get rid of any old sockets */
668 unlink(ctdb->daemon.name);
670 /* create a unix domain stream socket to listen to */
671 res = ux_socket_bind(ctdb);
673 DEBUG(DEBUG_ALERT,(__location__ " Failed to open CTDB unix domain socket\n"));
677 if (do_fork && fork()) {
681 tdb_reopen_all(False);
686 if (open("/dev/null", O_RDONLY) != 0) {
687 DEBUG(DEBUG_ALERT,(__location__ " Failed to setup stdin on /dev/null\n"));
691 block_signal(SIGPIPE);
693 if (ctdb->do_setsched) {
694 /* try to set us up as realtime */
695 ctdb_set_scheduler(ctdb);
698 /* ensure the socket is deleted on exit of the daemon */
699 domain_socket_name = talloc_strdup(talloc_autofree_context(), ctdb->daemon.name);
700 if (domain_socket_name == NULL) {
701 DEBUG(DEBUG_ALERT,(__location__ " talloc_strdup failed.\n"));
705 ctdb->ev = event_context_init(NULL);
707 ctdb_set_child_logging(ctdb);
709 /* force initial recovery for election */
710 ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
712 if (strcmp(ctdb->transport, "tcp") == 0) {
713 int ctdb_tcp_init(struct ctdb_context *);
714 ret = ctdb_tcp_init(ctdb);
716 #ifdef USE_INFINIBAND
717 if (strcmp(ctdb->transport, "ib") == 0) {
718 int ctdb_ibw_init(struct ctdb_context *);
719 ret = ctdb_ibw_init(ctdb);
723 DEBUG(DEBUG_ERR,("Failed to initialise transport '%s'\n", ctdb->transport));
727 if (ctdb->methods == NULL) {
728 DEBUG(DEBUG_ALERT,(__location__ " Can not initialize transport. ctdb->methods is NULL\n"));
729 ctdb_fatal(ctdb, "transport is unavailable. can not initialize.");
732 /* initialise the transport */
733 if (ctdb->methods->initialise(ctdb) != 0) {
734 ctdb_fatal(ctdb, "transport failed to initialise");
737 /* attach to any existing persistent databases */
738 if (ctdb_attach_persistent(ctdb) != 0) {
739 ctdb_fatal(ctdb, "Failed to attach to persistent databases\n");
742 /* start frozen, then let the first election sort things out */
743 if (ctdb_blocking_freeze(ctdb)) {
744 ctdb_fatal(ctdb, "Failed to get initial freeze\n");
747 /* now start accepting clients, only can do this once frozen */
748 fde = event_add_fd(ctdb->ev, ctdb, ctdb->daemon.sd,
749 EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
750 ctdb_accept_client, ctdb);
752 /* tell all other nodes we've just started up */
753 ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_ALL,
754 0, CTDB_CONTROL_STARTUP, 0,
755 CTDB_CTRL_FLAG_NOREPLY,
756 tdb_null, NULL, NULL);
758 /* release any IPs we hold from previous runs of the daemon */
759 ctdb_release_all_ips(ctdb);
761 /* start the transport going */
762 ctdb_start_transport(ctdb);
764 /* set up a handler to pick up sigchld */
765 se = event_add_signal(ctdb->ev, ctdb,
770 DEBUG(DEBUG_CRIT,("Failed to set up signal handler for SIGCHLD\n"));
774 /* go into a wait loop to allow other nodes to complete */
775 event_loop_wait(ctdb->ev);
777 DEBUG(DEBUG_CRIT,("event_loop_wait() returned. this should not happen\n"));
782 allocate a packet for use in daemon<->daemon communication
784 struct ctdb_req_header *_ctdb_transport_allocate(struct ctdb_context *ctdb,
786 enum ctdb_operation operation,
787 size_t length, size_t slength,
791 struct ctdb_req_header *hdr;
793 length = MAX(length, slength);
794 size = (length+(CTDB_DS_ALIGNMENT-1)) & ~(CTDB_DS_ALIGNMENT-1);
796 if (ctdb->methods == NULL) {
797 DEBUG(DEBUG_ERR,(__location__ " Unable to allocate transport packet for operation %u of length %u. Transport is DOWN.\n",
798 operation, (unsigned)length));
802 hdr = (struct ctdb_req_header *)ctdb->methods->allocate_pkt(mem_ctx, size);
804 DEBUG(DEBUG_ERR,("Unable to allocate transport packet for operation %u of length %u\n",
805 operation, (unsigned)length));
808 talloc_set_name_const(hdr, type);
809 memset(hdr, 0, slength);
810 hdr->length = length;
811 hdr->operation = operation;
812 hdr->ctdb_magic = CTDB_MAGIC;
813 hdr->ctdb_version = CTDB_VERSION;
814 hdr->generation = ctdb->vnn_map->generation;
815 hdr->srcnode = ctdb->pnn;
820 struct daemon_control_state {
821 struct daemon_control_state *next, *prev;
822 struct ctdb_client *client;
823 struct ctdb_req_control *c;
825 struct ctdb_node *node;
829 callback when a control reply comes in
831 static void daemon_control_callback(struct ctdb_context *ctdb,
832 int32_t status, TDB_DATA data,
833 const char *errormsg,
836 struct daemon_control_state *state = talloc_get_type(private_data,
837 struct daemon_control_state);
838 struct ctdb_client *client = state->client;
839 struct ctdb_reply_control *r;
842 /* construct a message to send to the client containing the data */
843 len = offsetof(struct ctdb_reply_control, data) + data.dsize;
845 len += strlen(errormsg);
847 r = ctdbd_allocate_pkt(ctdb, state, CTDB_REPLY_CONTROL, len,
848 struct ctdb_reply_control);
849 CTDB_NO_MEMORY_VOID(ctdb, r);
851 r->hdr.reqid = state->reqid;
853 r->datalen = data.dsize;
855 memcpy(&r->data[0], data.dptr, data.dsize);
857 r->errorlen = strlen(errormsg);
858 memcpy(&r->data[r->datalen], errormsg, r->errorlen);
861 daemon_queue_send(client, &r->hdr);
867 fail all pending controls to a disconnected node
869 void ctdb_daemon_cancel_controls(struct ctdb_context *ctdb, struct ctdb_node *node)
871 struct daemon_control_state *state;
872 while ((state = node->pending_controls)) {
873 DLIST_REMOVE(node->pending_controls, state);
874 daemon_control_callback(ctdb, (uint32_t)-1, tdb_null,
875 "node is disconnected", state);
880 destroy a daemon_control_state
882 static int daemon_control_destructor(struct daemon_control_state *state)
885 DLIST_REMOVE(state->node->pending_controls, state);
891 this is called when the ctdb daemon received a ctdb request control
892 from a local client over the unix domain socket
894 static void daemon_request_control_from_client(struct ctdb_client *client,
895 struct ctdb_req_control *c)
899 struct daemon_control_state *state;
900 TALLOC_CTX *tmp_ctx = talloc_new(client);
902 if (c->hdr.destnode == CTDB_CURRENT_NODE) {
903 c->hdr.destnode = client->ctdb->pnn;
906 state = talloc(client, struct daemon_control_state);
907 CTDB_NO_MEMORY_VOID(client->ctdb, state);
909 state->client = client;
910 state->c = talloc_steal(state, c);
911 state->reqid = c->hdr.reqid;
912 if (ctdb_validate_pnn(client->ctdb, c->hdr.destnode)) {
913 state->node = client->ctdb->nodes[c->hdr.destnode];
914 DLIST_ADD(state->node->pending_controls, state);
919 talloc_set_destructor(state, daemon_control_destructor);
921 if (c->flags & CTDB_CTRL_FLAG_NOREPLY) {
922 talloc_steal(tmp_ctx, state);
925 data.dptr = &c->data[0];
926 data.dsize = c->datalen;
927 res = ctdb_daemon_send_control(client->ctdb, c->hdr.destnode,
928 c->srvid, c->opcode, client->client_id,
930 data, daemon_control_callback,
933 DEBUG(DEBUG_ERR,(__location__ " Failed to send control to remote node %u\n",
937 talloc_free(tmp_ctx);
941 register a call function
943 int ctdb_daemon_set_call(struct ctdb_context *ctdb, uint32_t db_id,
944 ctdb_fn_t fn, int id)
946 struct ctdb_registered_call *call;
947 struct ctdb_db_context *ctdb_db;
949 ctdb_db = find_ctdb_db(ctdb, db_id);
950 if (ctdb_db == NULL) {
954 call = talloc(ctdb_db, struct ctdb_registered_call);
958 DLIST_ADD(ctdb_db->calls, call);
965 this local messaging handler is ugly, but is needed to prevent
966 recursion in ctdb_send_message() when the destination node is the
967 same as the source node
969 struct ctdb_local_message {
970 struct ctdb_context *ctdb;
975 static void ctdb_local_message_trigger(struct event_context *ev, struct timed_event *te,
976 struct timeval t, void *private_data)
978 struct ctdb_local_message *m = talloc_get_type(private_data,
979 struct ctdb_local_message);
982 res = ctdb_dispatch_message(m->ctdb, m->srvid, m->data);
984 DEBUG(DEBUG_ERR, (__location__ " Failed to dispatch message for srvid=%llu\n",
985 (unsigned long long)m->srvid));
990 static int ctdb_local_message(struct ctdb_context *ctdb, uint64_t srvid, TDB_DATA data)
992 struct ctdb_local_message *m;
993 m = talloc(ctdb, struct ctdb_local_message);
994 CTDB_NO_MEMORY(ctdb, m);
999 m->data.dptr = talloc_memdup(m, m->data.dptr, m->data.dsize);
1000 if (m->data.dptr == NULL) {
1005 /* this needs to be done as an event to prevent recursion */
1006 event_add_timed(ctdb->ev, m, timeval_zero(), ctdb_local_message_trigger, m);
1013 int ctdb_daemon_send_message(struct ctdb_context *ctdb, uint32_t pnn,
1014 uint64_t srvid, TDB_DATA data)
1016 struct ctdb_req_message *r;
1019 if (ctdb->methods == NULL) {
1020 DEBUG(DEBUG_ERR,(__location__ " Failed to send message. Transport is DOWN\n"));
1024 /* see if this is a message to ourselves */
1025 if (pnn == ctdb->pnn) {
1026 return ctdb_local_message(ctdb, srvid, data);
1029 len = offsetof(struct ctdb_req_message, data) + data.dsize;
1030 r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REQ_MESSAGE, len,
1031 struct ctdb_req_message);
1032 CTDB_NO_MEMORY(ctdb, r);
1034 r->hdr.destnode = pnn;
1036 r->datalen = data.dsize;
1037 memcpy(&r->data[0], data.dptr, data.dsize);
1039 ctdb_queue_packet(ctdb, &r->hdr);
1047 struct ctdb_client_notify_list {
1048 struct ctdb_client_notify_list *next, *prev;
1049 struct ctdb_context *ctdb;
1055 static int ctdb_client_notify_destructor(struct ctdb_client_notify_list *nl)
1059 DEBUG(DEBUG_ERR,("Sending client notify message for srvid:%llu\n", (unsigned long long)nl->srvid));
1061 ret = ctdb_daemon_send_message(nl->ctdb, CTDB_BROADCAST_CONNECTED, (unsigned long long)nl->srvid, nl->data);
1063 DEBUG(DEBUG_ERR,("Failed to send client notify message\n"));
1069 int32_t ctdb_control_register_notify(struct ctdb_context *ctdb, uint32_t client_id, TDB_DATA indata)
1071 struct ctdb_client_notify_register *notify = (struct ctdb_client_notify_register *)indata.dptr;
1072 struct ctdb_client *client = ctdb_reqid_find(ctdb, client_id, struct ctdb_client);
1073 struct ctdb_client_notify_list *nl;
1075 DEBUG(DEBUG_ERR,("Register srvid %llu for client %d\n", (unsigned long long)notify->srvid, client_id));
1077 if (indata.dsize < offsetof(struct ctdb_client_notify_register, notify_data)) {
1078 DEBUG(DEBUG_ERR,(__location__ " Too little data in control : %d\n", (int)indata.dsize));
1082 if (indata.dsize != (notify->len + offsetof(struct ctdb_client_notify_register, notify_data))) {
1083 DEBUG(DEBUG_ERR,(__location__ " Wrong amount of data in control. Got %d, expected %d\n", (int)indata.dsize, (int)(notify->len + offsetof(struct ctdb_client_notify_register, notify_data))));
1088 if (client == NULL) {
1089 DEBUG(DEBUG_ERR,(__location__ " Could not find client parent structure. You can not send this control to a remote node\n"));
1093 for(nl=client->notify; nl; nl=nl->next) {
1094 if (nl->srvid == notify->srvid) {
1099 DEBUG(DEBUG_ERR,(__location__ " Notification for srvid:%llu already exists for this client\n", (unsigned long long)notify->srvid));
1103 nl = talloc(client, struct ctdb_client_notify_list);
1104 CTDB_NO_MEMORY(ctdb, nl);
1106 nl->srvid = notify->srvid;
1107 nl->data.dsize = notify->len;
1108 nl->data.dptr = talloc_size(nl, nl->data.dsize);
1109 CTDB_NO_MEMORY(ctdb, nl->data.dptr);
1110 memcpy(nl->data.dptr, notify->notify_data, nl->data.dsize);
1112 DLIST_ADD(client->notify, nl);
1113 talloc_set_destructor(nl, ctdb_client_notify_destructor);
1118 int32_t ctdb_control_deregister_notify(struct ctdb_context *ctdb, uint32_t client_id, TDB_DATA indata)
1120 struct ctdb_client_notify_deregister *notify = (struct ctdb_client_notify_deregister *)indata.dptr;
1121 struct ctdb_client *client = ctdb_reqid_find(ctdb, client_id, struct ctdb_client);
1122 struct ctdb_client_notify_list *nl;
1124 DEBUG(DEBUG_ERR,("Deregister srvid %llu for client %d\n", (unsigned long long)notify->srvid, client_id));
1126 if (client == NULL) {
1127 DEBUG(DEBUG_ERR,(__location__ " Could not find client parent structure. You can not send this control to a remote node\n"));
1131 for(nl=client->notify; nl; nl=nl->next) {
1132 if (nl->srvid == notify->srvid) {
1137 DEBUG(DEBUG_ERR,(__location__ " No notification for srvid:%llu found for this client\n", (unsigned long long)notify->srvid));
1141 DLIST_REMOVE(client->notify, nl);
1142 talloc_set_destructor(nl, NULL);