Fix Bug 4240. Thanks to Karolin <ks@sernet.de> for bugging :-)
[Samba.git] / source / lib / ctdbd_conn.c
blob46936c5f39bd62f1d5b7de7b0692bb056beb6bfb
1 /*
2 Unix SMB/CIFS implementation.
3 Samba internal messaging functions
4 Copyright (C) 2007 by Volker Lendecke
5 Copyright (C) 2007 by Andrew Tridgell
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program. If not, see <http://www.gnu.org/licenses/>.
21 #include "includes.h"
23 #ifdef CLUSTER_SUPPORT
25 #include "librpc/gen_ndr/messaging.h"
26 #include "librpc/gen_ndr/ndr_messaging.h"
28 /* paths to these include files come from --with-ctdb= in configure */
29 #include "ctdb.h"
30 #include "ctdb_private.h"
32 struct ctdbd_connection {
33 struct messaging_context *msg_ctx;
34 uint32 reqid;
35 uint32 our_vnn;
36 uint64 rand_srvid;
37 struct packet_context *pkt;
38 struct fd_event *fde;
40 void (*release_ip_handler)(const char *ip_addr, void *private_data);
41 void *release_ip_priv;
44 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
45 uint32_t vnn, uint32 opcode,
46 uint64_t srvid, TDB_DATA data,
47 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
48 int *cstatus);
51 * exit on fatal communications errors with the ctdbd daemon
53 static void cluster_fatal(const char *why)
55 DEBUG(0,("cluster fatal event: %s - exiting immediately\n", why));
56 /* we don't use smb_panic() as we don't want to delay to write
57 a core file. We need to release this process id immediately
58 so that someone else can take over without getting sharing
59 violations */
60 _exit(0);
66 static void ctdb_packet_dump(struct ctdb_req_header *hdr)
68 if (DEBUGLEVEL < 10) {
69 return;
71 DEBUGADD(10, ("len=%d, magic=%x, vers=%d, gen=%d, op=%d, reqid=%d\n",
72 (int)hdr->length, (int)hdr->ctdb_magic,
73 (int)hdr->ctdb_version, (int)hdr->generation,
74 (int)hdr->operation, (int)hdr->reqid));
78 * Register a srvid with ctdbd
80 static NTSTATUS register_with_ctdbd(struct ctdbd_connection *conn,
81 uint64_t srvid)
84 int cstatus;
85 return ctdbd_control(conn, CTDB_CURRENT_NODE,
86 CTDB_CONTROL_REGISTER_SRVID, srvid,
87 tdb_null, NULL, NULL, &cstatus);
91 * get our vnn from the cluster
93 static NTSTATUS get_cluster_vnn(struct ctdbd_connection *conn, uint32 *vnn)
95 int32_t cstatus=-1;
96 NTSTATUS status;
97 status = ctdbd_control(conn,
98 CTDB_CURRENT_NODE, CTDB_CONTROL_GET_PNN, 0,
99 tdb_null, NULL, NULL, &cstatus);
100 if (!NT_STATUS_IS_OK(status)) {
101 cluster_fatal("ctdbd_control failed\n");
103 *vnn = (uint32_t)cstatus;
104 return status;
107 uint32 ctdbd_vnn(const struct ctdbd_connection *conn)
109 return conn->our_vnn;
113 * Get us a ctdb connection
116 static NTSTATUS ctdbd_connect(TALLOC_CTX *mem_ctx,
117 struct packet_context **presult)
119 struct packet_context *result;
120 const char *sockname = lp_ctdbd_socket();
121 struct sockaddr_un addr;
122 int fd;
124 if (!sockname || !*sockname) {
125 sockname = CTDB_PATH;
128 fd = socket(AF_UNIX, SOCK_STREAM, 0);
129 if (fd == -1) {
130 DEBUG(3, ("Could not create socket: %s\n", strerror(errno)));
131 return map_nt_error_from_unix(errno);
134 ZERO_STRUCT(addr);
135 addr.sun_family = AF_UNIX;
136 strncpy(addr.sun_path, sockname, sizeof(addr.sun_path));
138 if (sys_connect(fd, (struct sockaddr *)&addr) == -1) {
139 DEBUG(0, ("connect(%s) failed: %s\n", sockname,
140 strerror(errno)));
141 close(fd);
142 return map_nt_error_from_unix(errno);
145 if (!(result = packet_init(mem_ctx, fd))) {
146 close(fd);
147 return NT_STATUS_NO_MEMORY;
150 *presult = result;
151 return NT_STATUS_OK;
155 * Do we have a complete ctdb packet in the queue?
158 static bool ctdb_req_complete(const struct data_blob *data,
159 size_t *length,
160 void *private_data)
162 uint32 msglen;
164 if (data->length < sizeof(msglen)) {
165 return False;
168 msglen = *((uint32 *)data->data);
170 DEBUG(10, ("msglen = %d\n", msglen));
172 if (msglen < sizeof(struct ctdb_req_header)) {
173 DEBUG(0, ("Got invalid msglen: %d, expected at least %d for "
174 "the req_header\n", (int)msglen,
175 (int)sizeof(struct ctdb_req_header)));
176 cluster_fatal("ctdbd protocol error\n");
179 if (data->length >= msglen) {
180 *length = msglen;
181 return True;
184 return False;
188 * State necessary to defer an incoming message while we are waiting for a
189 * ctdb reply.
192 struct deferred_msg_state {
193 struct messaging_context *msg_ctx;
194 struct messaging_rec *rec;
198 * Timed event handler for the deferred message
201 static void deferred_message_dispatch(struct event_context *event_ctx,
202 struct timed_event *te,
203 const struct timeval *now,
204 void *private_data)
206 struct deferred_msg_state *state = talloc_get_type_abort(
207 private_data, struct deferred_msg_state);
209 messaging_dispatch_rec(state->msg_ctx, state->rec);
210 TALLOC_FREE(state);
211 TALLOC_FREE(te);
214 struct req_pull_state {
215 TALLOC_CTX *mem_ctx;
216 DATA_BLOB req;
220 * Pull a ctdb request out of the incoming packet queue
223 static NTSTATUS ctdb_req_pull(const struct data_blob *data,
224 void *private_data)
226 struct req_pull_state *state = (struct req_pull_state *)private_data;
228 state->req = data_blob_talloc(state->mem_ctx, data->data,
229 data->length);
230 if (state->req.data == NULL) {
231 return NT_STATUS_NO_MEMORY;
233 return NT_STATUS_OK;
237 * Fetch a messaging_rec from an incoming ctdb style message
240 static struct messaging_rec *ctdb_pull_messaging_rec(TALLOC_CTX *mem_ctx,
241 size_t overall_length,
242 struct ctdb_req_message *msg)
244 struct messaging_rec *result;
245 DATA_BLOB blob;
246 enum ndr_err_code ndr_err;
248 if ((overall_length < offsetof(struct ctdb_req_message, data))
249 || (overall_length
250 < offsetof(struct ctdb_req_message, data) + msg->datalen)) {
252 cluster_fatal("got invalid msg length");
255 if (!(result = TALLOC_P(mem_ctx, struct messaging_rec))) {
256 DEBUG(0, ("talloc failed\n"));
257 return NULL;
260 blob = data_blob_const(msg->data, msg->datalen);
262 ndr_err = ndr_pull_struct_blob(
263 &blob, result, result,
264 (ndr_pull_flags_fn_t)ndr_pull_messaging_rec);
266 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
267 DEBUG(0, ("ndr_pull_struct_blob failed: %s\n",
268 ndr_errstr(ndr_err)));
269 TALLOC_FREE(result);
270 return NULL;
273 if (DEBUGLEVEL >= 10) {
274 DEBUG(10, ("ctdb_pull_messaging_rec:\n"));
275 NDR_PRINT_DEBUG(messaging_rec, result);
278 return result;
282 * Read a full ctdbd request. If we have a messaging context, defer incoming
283 * messages that might come in between.
286 static NTSTATUS ctdb_read_req(struct ctdbd_connection *conn, uint32 reqid,
287 TALLOC_CTX *mem_ctx, void *result)
289 struct ctdb_req_header *hdr;
290 struct req_pull_state state;
291 NTSTATUS status;
293 again:
295 status = packet_fd_read_sync(conn->pkt);
297 if (NT_STATUS_EQUAL(status, NT_STATUS_NETWORK_BUSY)) {
298 /* EAGAIN */
299 goto again;
300 } else if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
301 /* EAGAIN */
302 goto again;
305 if (!NT_STATUS_IS_OK(status)) {
306 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
307 cluster_fatal("ctdbd died\n");
310 next_pkt:
312 ZERO_STRUCT(state);
313 state.mem_ctx = mem_ctx;
315 if (!packet_handler(conn->pkt, ctdb_req_complete, ctdb_req_pull,
316 &state, &status)) {
318 * Not enough data
320 DEBUG(10, ("not enough data from ctdb socket, retrying\n"));
321 goto again;
324 if (!NT_STATUS_IS_OK(status)) {
325 DEBUG(0, ("Could not read packet: %s\n", nt_errstr(status)));
326 cluster_fatal("ctdbd died\n");
329 hdr = (struct ctdb_req_header *)state.req.data;
331 DEBUG(10, ("Received ctdb packet\n"));
332 ctdb_packet_dump(hdr);
334 if (hdr->operation == CTDB_REQ_MESSAGE) {
335 struct timed_event *evt;
336 struct deferred_msg_state *msg_state;
337 struct ctdb_req_message *msg = (struct ctdb_req_message *)hdr;
339 if (conn->msg_ctx == NULL) {
340 DEBUG(1, ("Got a message without having a msg ctx, "
341 "dropping msg %llu\n",
342 (long long unsigned)msg->srvid));
343 goto next_pkt;
346 if ((conn->release_ip_handler != NULL)
347 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
348 /* must be dispatched immediately */
349 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
350 conn->release_ip_handler((const char *)msg->data,
351 conn->release_ip_priv);
352 TALLOC_FREE(hdr);
353 goto next_pkt;
356 if (!(msg_state = TALLOC_P(NULL, struct deferred_msg_state))) {
357 DEBUG(0, ("talloc failed\n"));
358 TALLOC_FREE(hdr);
359 goto next_pkt;
362 if (!(msg_state->rec = ctdb_pull_messaging_rec(
363 msg_state, state.req.length, msg))) {
364 DEBUG(0, ("ctdbd_pull_messaging_rec failed\n"));
365 TALLOC_FREE(msg_state);
366 TALLOC_FREE(hdr);
367 goto next_pkt;
370 TALLOC_FREE(hdr);
372 msg_state->msg_ctx = conn->msg_ctx;
375 * We're waiting for a call reply, but an async message has
376 * crossed. Defer dispatching to the toplevel event loop.
378 evt = event_add_timed(conn->msg_ctx->event_ctx,
379 conn->msg_ctx->event_ctx,
380 timeval_zero(),
381 "deferred_message_dispatch",
382 deferred_message_dispatch,
383 msg_state);
384 if (evt == NULL) {
385 DEBUG(0, ("event_add_timed failed\n"));
386 TALLOC_FREE(msg_state);
387 TALLOC_FREE(hdr);
388 goto next_pkt;
391 goto next_pkt;
394 if (hdr->reqid != reqid) {
395 /* we got the wrong reply */
396 DEBUG(0,("Discarding mismatched ctdb reqid %u should have "
397 "been %u\n", hdr->reqid, reqid));
398 TALLOC_FREE(hdr);
399 goto again;
402 *((void **)result) = talloc_move(mem_ctx, &hdr);
404 return NT_STATUS_OK;
408 * Get us a ctdbd connection
411 NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
412 struct ctdbd_connection **pconn)
414 struct ctdbd_connection *conn;
415 NTSTATUS status;
417 if (!(conn = TALLOC_ZERO_P(mem_ctx, struct ctdbd_connection))) {
418 DEBUG(0, ("talloc failed\n"));
419 return NT_STATUS_NO_MEMORY;
422 status = ctdbd_connect(conn, &conn->pkt);
424 if (!NT_STATUS_IS_OK(status)) {
425 DEBUG(10, ("ctdbd_connect failed: %s\n", nt_errstr(status)));
426 goto fail;
429 status = get_cluster_vnn(conn, &conn->our_vnn);
431 if (!NT_STATUS_IS_OK(status)) {
432 DEBUG(10, ("get_cluster_vnn failed: %s\n", nt_errstr(status)));
433 goto fail;
436 generate_random_buffer((unsigned char *)&conn->rand_srvid,
437 sizeof(conn->rand_srvid));
439 status = register_with_ctdbd(conn, conn->rand_srvid);
441 if (!NT_STATUS_IS_OK(status)) {
442 DEBUG(5, ("Could not register random srvid: %s\n",
443 nt_errstr(status)));
444 goto fail;
447 *pconn = conn;
448 return NT_STATUS_OK;
450 fail:
451 TALLOC_FREE(conn);
452 return status;
456 * Get us a ctdbd connection and register us as a process
459 NTSTATUS ctdbd_messaging_connection(TALLOC_CTX *mem_ctx,
460 struct ctdbd_connection **pconn)
462 struct ctdbd_connection *conn;
463 NTSTATUS status;
465 status = ctdbd_init_connection(mem_ctx, &conn);
467 if (!NT_STATUS_IS_OK(status)) {
468 return status;
471 status = register_with_ctdbd(conn, (uint64_t)sys_getpid());
472 if (!NT_STATUS_IS_OK(status)) {
473 goto fail;
476 status = register_with_ctdbd(conn, MSG_SRVID_SAMBA);
477 if (!NT_STATUS_IS_OK(status)) {
478 goto fail;
481 *pconn = conn;
482 return NT_STATUS_OK;
484 fail:
485 TALLOC_FREE(conn);
486 return status;
490 * Packet handler to receive and handle a ctdb message
492 static NTSTATUS ctdb_handle_message(const struct data_blob *data,
493 void *private_data)
495 struct ctdbd_connection *conn = talloc_get_type_abort(
496 private_data, struct ctdbd_connection);
497 struct ctdb_req_message *msg;
498 struct messaging_rec *msg_rec;
500 msg = (struct ctdb_req_message *)data->data;
502 if (msg->hdr.operation != CTDB_REQ_MESSAGE) {
503 DEBUG(0, ("Received async msg of type %u, discarding\n",
504 msg->hdr.operation));
505 return NT_STATUS_INVALID_PARAMETER;
508 if ((conn->release_ip_handler != NULL)
509 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
510 /* must be dispatched immediately */
511 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
512 conn->release_ip_handler((const char *)msg->data,
513 conn->release_ip_priv);
514 return NT_STATUS_OK;
517 SMB_ASSERT(conn->msg_ctx != NULL);
519 if (msg->srvid == CTDB_SRVID_RECONFIGURE) {
520 DEBUG(0,("Got cluster reconfigure message\n"));
522 * when the cluster is reconfigured, we need to clean the brl
523 * database
525 messaging_send(conn->msg_ctx, procid_self(),
526 MSG_SMB_BRL_VALIDATE, &data_blob_null);
529 * it's possible that we have just rejoined the cluster after
530 * an outage. In that case our pending locks could have been
531 * removed from the lockdb, so retry them once more
533 message_send_all(conn->msg_ctx, MSG_SMB_UNLOCK, NULL, 0, NULL);
535 return NT_STATUS_OK;
539 /* only messages to our pid or the broadcast are valid here */
540 if (msg->srvid != sys_getpid() && msg->srvid != MSG_SRVID_SAMBA) {
541 DEBUG(0,("Got unexpected message with srvid=%llu\n",
542 (unsigned long long)msg->srvid));
543 return NT_STATUS_OK;
546 if (!(msg_rec = ctdb_pull_messaging_rec(NULL, data->length, msg))) {
547 DEBUG(10, ("ctdb_pull_messaging_rec failed\n"));
548 return NT_STATUS_NO_MEMORY;
551 messaging_dispatch_rec(conn->msg_ctx, msg_rec);
553 TALLOC_FREE(msg_rec);
554 return NT_STATUS_OK;
558 * The ctdbd socket is readable asynchronuously
561 static void ctdbd_socket_handler(struct event_context *event_ctx,
562 struct fd_event *event,
563 uint16 flags,
564 void *private_data)
566 struct ctdbd_connection *conn = talloc_get_type_abort(
567 private_data, struct ctdbd_connection);
569 NTSTATUS status;
571 status = packet_fd_read(conn->pkt);
573 if (!NT_STATUS_IS_OK(status)) {
574 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
575 cluster_fatal("ctdbd died\n");
578 while (packet_handler(conn->pkt, ctdb_req_complete,
579 ctdb_handle_message, conn, &status)) {
580 if (!NT_STATUS_IS_OK(status)) {
581 DEBUG(10, ("could not handle incoming message: %s\n",
582 nt_errstr(status)));
588 * Prepare a ctdbd connection to receive messages
591 NTSTATUS ctdbd_register_msg_ctx(struct ctdbd_connection *conn,
592 struct messaging_context *msg_ctx)
594 SMB_ASSERT(conn->msg_ctx == NULL);
595 SMB_ASSERT(conn->fde == NULL);
597 if (!(conn->fde = event_add_fd(msg_ctx->event_ctx, conn,
598 packet_get_fd(conn->pkt),
599 EVENT_FD_READ,
600 ctdbd_socket_handler,
601 conn))) {
602 DEBUG(0, ("event_add_fd failed\n"));
603 return NT_STATUS_NO_MEMORY;
606 conn->msg_ctx = msg_ctx;
608 return NT_STATUS_OK;
612 * Send a messaging message across a ctdbd
615 NTSTATUS ctdbd_messaging_send(struct ctdbd_connection *conn,
616 uint32 dst_vnn, uint64 dst_srvid,
617 struct messaging_rec *msg)
619 struct ctdb_req_message r;
620 TALLOC_CTX *mem_ctx;
621 DATA_BLOB blob;
622 NTSTATUS status;
623 enum ndr_err_code ndr_err;
625 if (!(mem_ctx = talloc_init("ctdbd_messaging_send"))) {
626 DEBUG(0, ("talloc failed\n"));
627 return NT_STATUS_NO_MEMORY;
630 ndr_err = ndr_push_struct_blob(
631 &blob, mem_ctx, msg,
632 (ndr_push_flags_fn_t)ndr_push_messaging_rec);
634 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
635 DEBUG(0, ("ndr_push_struct_blob failed: %s\n",
636 ndr_errstr(ndr_err)));
637 status = ndr_map_error2ntstatus(ndr_err);
638 goto fail;
641 r.hdr.length = offsetof(struct ctdb_req_message, data) + blob.length;
642 r.hdr.ctdb_magic = CTDB_MAGIC;
643 r.hdr.ctdb_version = CTDB_VERSION;
644 r.hdr.generation = 1;
645 r.hdr.operation = CTDB_REQ_MESSAGE;
646 r.hdr.destnode = dst_vnn;
647 r.hdr.srcnode = conn->our_vnn;
648 r.hdr.reqid = 0;
649 r.srvid = dst_srvid;
650 r.datalen = blob.length;
652 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
653 ctdb_packet_dump(&r.hdr);
655 status = packet_send(
656 conn->pkt, 2,
657 data_blob_const(&r, offsetof(struct ctdb_req_message, data)),
658 blob);
660 if (!NT_STATUS_IS_OK(status)) {
661 DEBUG(0, ("packet_send failed: %s\n", nt_errstr(status)));
662 goto fail;
665 status = packet_flush(conn->pkt);
667 if (!NT_STATUS_IS_OK(status)) {
668 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
669 cluster_fatal("cluster dispatch daemon msg write error\n");
672 status = NT_STATUS_OK;
673 fail:
674 TALLOC_FREE(mem_ctx);
675 return status;
679 * send/recv a generic ctdb control message
681 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
682 uint32_t vnn, uint32 opcode,
683 uint64_t srvid, TDB_DATA data,
684 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
685 int *cstatus)
687 struct ctdb_req_control req;
688 struct ctdb_reply_control *reply = NULL;
689 struct ctdbd_connection *new_conn = NULL;
690 NTSTATUS status;
692 if (conn == NULL) {
693 status = ctdbd_init_connection(NULL, &new_conn);
695 if (!NT_STATUS_IS_OK(status)) {
696 DEBUG(10, ("Could not init temp connection: %s\n",
697 nt_errstr(status)));
698 goto fail;
701 conn = new_conn;
704 ZERO_STRUCT(req);
705 req.hdr.length = offsetof(struct ctdb_req_control, data) + data.dsize;
706 req.hdr.ctdb_magic = CTDB_MAGIC;
707 req.hdr.ctdb_version = CTDB_VERSION;
708 req.hdr.operation = CTDB_REQ_CONTROL;
709 req.hdr.reqid = ++conn->reqid;
710 req.hdr.destnode = vnn;
711 req.opcode = opcode;
712 req.srvid = srvid;
713 req.datalen = data.dsize;
715 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
716 ctdb_packet_dump(&req.hdr);
718 status = packet_send(
719 conn->pkt, 2,
720 data_blob_const(&req, offsetof(struct ctdb_req_control, data)),
721 data_blob_const(data.dptr, data.dsize));
723 if (!NT_STATUS_IS_OK(status)) {
724 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
725 goto fail;
728 status = packet_flush(conn->pkt);
730 if (!NT_STATUS_IS_OK(status)) {
731 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
732 cluster_fatal("cluster dispatch daemon control write error\n");
735 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
737 if (!NT_STATUS_IS_OK(status)) {
738 DEBUG(10, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
739 goto fail;
742 if (reply->hdr.operation != CTDB_REPLY_CONTROL) {
743 DEBUG(0, ("received invalid reply\n"));
744 goto fail;
747 if (outdata) {
748 if (!(outdata->dptr = (uint8 *)talloc_memdup(
749 mem_ctx, reply->data, reply->datalen))) {
750 TALLOC_FREE(reply);
751 return NT_STATUS_NO_MEMORY;
753 outdata->dsize = reply->datalen;
755 if (cstatus) {
756 (*cstatus) = reply->status;
759 status = NT_STATUS_OK;
761 fail:
762 TALLOC_FREE(new_conn);
763 TALLOC_FREE(reply);
764 return status;
768 * see if a remote process exists
770 bool ctdbd_process_exists(struct ctdbd_connection *conn, uint32 vnn, pid_t pid)
772 NTSTATUS status;
773 TDB_DATA data;
774 int32_t cstatus;
776 data.dptr = (uint8_t*)&pid;
777 data.dsize = sizeof(pid);
779 status = ctdbd_control(conn, vnn, CTDB_CONTROL_PROCESS_EXISTS, 0,
780 data, NULL, NULL, &cstatus);
781 if (!NT_STATUS_IS_OK(status)) {
782 DEBUG(0, (__location__ " ctdb_control for process_exists "
783 "failed\n"));
784 return False;
787 return cstatus == 0;
791 * Get a db path
793 char *ctdbd_dbpath(struct ctdbd_connection *conn,
794 TALLOC_CTX *mem_ctx, uint32_t db_id)
796 NTSTATUS status;
797 TDB_DATA data;
798 int32_t cstatus;
800 data.dptr = (uint8_t*)&db_id;
801 data.dsize = sizeof(db_id);
803 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
804 CTDB_CONTROL_GETDBPATH, 0, data,
805 mem_ctx, &data, &cstatus);
806 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
807 DEBUG(0,(__location__ " ctdb_control for getdbpath failed\n"));
808 return NULL;
811 return (char *)data.dptr;
815 * attach to a ctdb database
817 NTSTATUS ctdbd_db_attach(struct ctdbd_connection *conn,
818 const char *name, uint32_t *db_id, int tdb_flags)
820 NTSTATUS status;
821 TDB_DATA data;
822 int32_t cstatus;
823 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
825 data.dptr = (uint8_t*)name;
826 data.dsize = strlen(name)+1;
828 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
829 persistent
830 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
831 : CTDB_CONTROL_DB_ATTACH,
832 0, data, NULL, &data, &cstatus);
833 if (!NT_STATUS_IS_OK(status)) {
834 DEBUG(0, (__location__ " ctdb_control for db_attach "
835 "failed: %s\n", nt_errstr(status)));
836 return status;
839 if (cstatus != 0 || data.dsize != sizeof(uint32_t)) {
840 DEBUG(0,(__location__ " ctdb_control for db_attach failed\n"));
841 return NT_STATUS_INTERNAL_ERROR;
844 *db_id = *(uint32_t *)data.dptr;
845 talloc_free(data.dptr);
847 if (!(tdb_flags & TDB_SEQNUM)) {
848 return NT_STATUS_OK;
851 data.dptr = (uint8_t *)db_id;
852 data.dsize = sizeof(*db_id);
854 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
855 CTDB_CONTROL_ENABLE_SEQNUM, 0, data,
856 NULL, NULL, &cstatus);
857 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
858 DEBUG(0,(__location__ " ctdb_control for enable seqnum "
859 "failed\n"));
860 return NT_STATUS_IS_OK(status) ? NT_STATUS_INTERNAL_ERROR :
861 status;
864 return NT_STATUS_OK;
868 * force the migration of a record to this node
870 NTSTATUS ctdbd_migrate(struct ctdbd_connection *conn, uint32 db_id,
871 TDB_DATA key)
873 struct ctdb_req_call req;
874 struct ctdb_reply_call *reply;
875 NTSTATUS status;
877 ZERO_STRUCT(req);
879 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
880 req.hdr.ctdb_magic = CTDB_MAGIC;
881 req.hdr.ctdb_version = CTDB_VERSION;
882 req.hdr.operation = CTDB_REQ_CALL;
883 req.hdr.reqid = ++conn->reqid;
884 req.flags = CTDB_IMMEDIATE_MIGRATION;
885 req.callid = CTDB_NULL_FUNC;
886 req.db_id = db_id;
887 req.keylen = key.dsize;
889 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
890 ctdb_packet_dump(&req.hdr);
892 status = packet_send(
893 conn->pkt, 2,
894 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
895 data_blob_const(key.dptr, key.dsize));
897 if (!NT_STATUS_IS_OK(status)) {
898 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
899 return status;
902 status = packet_flush(conn->pkt);
904 if (!NT_STATUS_IS_OK(status)) {
905 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
906 cluster_fatal("cluster dispatch daemon control write error\n");
909 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
911 if (!NT_STATUS_IS_OK(status)) {
912 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
913 goto fail;
916 if (reply->hdr.operation != CTDB_REPLY_CALL) {
917 DEBUG(0, ("received invalid reply\n"));
918 status = NT_STATUS_INTERNAL_ERROR;
919 goto fail;
922 status = NT_STATUS_OK;
923 fail:
925 TALLOC_FREE(reply);
926 return status;
930 * remotely fetch a record without locking it or forcing a migration
932 NTSTATUS ctdbd_fetch(struct ctdbd_connection *conn, uint32 db_id,
933 TDB_DATA key, TALLOC_CTX *mem_ctx, TDB_DATA *data)
935 struct ctdb_req_call req;
936 struct ctdb_reply_call *reply;
937 NTSTATUS status;
939 ZERO_STRUCT(req);
941 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
942 req.hdr.ctdb_magic = CTDB_MAGIC;
943 req.hdr.ctdb_version = CTDB_VERSION;
944 req.hdr.operation = CTDB_REQ_CALL;
945 req.hdr.reqid = ++conn->reqid;
946 req.flags = 0;
947 req.callid = CTDB_FETCH_FUNC;
948 req.db_id = db_id;
949 req.keylen = key.dsize;
951 status = packet_send(
952 conn->pkt, 2,
953 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
954 data_blob_const(key.dptr, key.dsize));
956 if (!NT_STATUS_IS_OK(status)) {
957 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
958 return status;
961 status = packet_flush(conn->pkt);
963 if (!NT_STATUS_IS_OK(status)) {
964 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
965 cluster_fatal("cluster dispatch daemon control write error\n");
968 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
970 if (!NT_STATUS_IS_OK(status)) {
971 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
972 goto fail;
975 if (reply->hdr.operation != CTDB_REPLY_CALL) {
976 DEBUG(0, ("received invalid reply\n"));
977 status = NT_STATUS_INTERNAL_ERROR;
978 goto fail;
981 data->dsize = reply->datalen;
982 if (data->dsize == 0) {
983 data->dptr = NULL;
984 goto done;
987 data->dptr = (uint8 *)talloc_memdup(mem_ctx, &reply->data[0],
988 reply->datalen);
989 if (data->dptr == NULL) {
990 DEBUG(0, ("talloc failed\n"));
991 status = NT_STATUS_NO_MEMORY;
992 goto fail;
995 done:
996 status = NT_STATUS_OK;
997 fail:
998 TALLOC_FREE(reply);
999 return status;
1002 struct ctdbd_traverse_state {
1003 void (*fn)(TDB_DATA key, TDB_DATA data, void *private_data);
1004 void *private_data;
1008 * Handle a traverse record coming in on the ctdbd connection
1011 static NTSTATUS ctdb_traverse_handler(const struct data_blob *blob,
1012 void *private_data)
1014 struct ctdbd_traverse_state *state =
1015 (struct ctdbd_traverse_state *)private_data;
1017 struct ctdb_req_message *m;
1018 struct ctdb_rec_data *d;
1019 TDB_DATA key, data;
1021 m = (struct ctdb_req_message *)blob->data;
1023 if (blob->length < sizeof(*m) || m->hdr.length != blob->length) {
1024 DEBUG(0, ("Got invalid message of length %d\n",
1025 (int)blob->length));
1026 return NT_STATUS_UNEXPECTED_IO_ERROR;
1029 d = (struct ctdb_rec_data *)&m->data[0];
1030 if (m->datalen < sizeof(uint32_t) || m->datalen != d->length) {
1031 DEBUG(0, ("Got invalid traverse data of length %d\n",
1032 (int)m->datalen));
1033 return NT_STATUS_UNEXPECTED_IO_ERROR;
1036 key.dsize = d->keylen;
1037 key.dptr = &d->data[0];
1038 data.dsize = d->datalen;
1039 data.dptr = &d->data[d->keylen];
1041 if (key.dsize == 0 && data.dsize == 0) {
1042 /* end of traverse */
1043 return NT_STATUS_END_OF_FILE;
1046 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1047 DEBUG(0, ("Got invalid ltdb header length %d\n",
1048 (int)data.dsize));
1049 return NT_STATUS_UNEXPECTED_IO_ERROR;
1051 data.dsize -= sizeof(struct ctdb_ltdb_header);
1052 data.dptr += sizeof(struct ctdb_ltdb_header);
1054 if (state->fn) {
1055 state->fn(key, data, state->private_data);
1058 return NT_STATUS_OK;
1062 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1063 connection to ctdbd to avoid the hairy recursive and async problems with
1064 everything in-line.
1067 NTSTATUS ctdbd_traverse(uint32 db_id,
1068 void (*fn)(TDB_DATA key, TDB_DATA data,
1069 void *private_data),
1070 void *private_data)
1072 struct ctdbd_connection *conn;
1073 NTSTATUS status;
1075 TDB_DATA data;
1076 struct ctdb_traverse_start t;
1077 int cstatus;
1078 struct ctdbd_traverse_state state;
1080 status = ctdbd_init_connection(NULL, &conn);
1082 t.db_id = db_id;
1083 t.srvid = conn->rand_srvid;
1084 t.reqid = ++conn->reqid;
1086 data.dptr = (uint8_t *)&t;
1087 data.dsize = sizeof(t);
1089 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1090 CTDB_CONTROL_TRAVERSE_START, conn->rand_srvid,
1091 data, NULL, NULL, &cstatus);
1093 if (!NT_STATUS_IS_OK(status) || (cstatus != 0)) {
1095 DEBUG(0,("ctdbd_control failed: %s, %d\n", nt_errstr(status),
1096 cstatus));
1098 if (NT_STATUS_IS_OK(status)) {
1100 * We need a mapping here
1102 status = NT_STATUS_UNSUCCESSFUL;
1104 goto done;
1107 state.fn = fn;
1108 state.private_data = private_data;
1110 while (True) {
1112 status = NT_STATUS_OK;
1114 if (packet_handler(conn->pkt, ctdb_req_complete,
1115 ctdb_traverse_handler, &state, &status)) {
1117 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1118 status = NT_STATUS_OK;
1119 break;
1123 * There might be more in the queue
1125 continue;
1128 if (!NT_STATUS_IS_OK(status)) {
1129 break;
1132 status = packet_fd_read_sync(conn->pkt);
1134 if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
1136 * There might be more in the queue
1138 continue;
1141 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1142 status = NT_STATUS_OK;
1145 if (!NT_STATUS_IS_OK(status)) {
1146 DEBUG(0, ("packet_fd_read_sync failed: %s\n", nt_errstr(status)));
1147 cluster_fatal("ctdbd died\n");
1151 done:
1152 TALLOC_FREE(conn);
1153 return status;
1157 * Register us as a server for a particular tcp connection
1160 NTSTATUS ctdbd_register_ips(struct ctdbd_connection *conn,
1161 const struct sockaddr_in *server,
1162 const struct sockaddr_in *client,
1163 void (*release_ip_handler)(const char *ip_addr,
1164 void *private_data),
1165 void *private_data)
1167 struct ctdb_control_tcp p;
1168 TDB_DATA data;
1169 NTSTATUS status;
1172 * Only one connection so far
1174 SMB_ASSERT(conn->release_ip_handler == NULL);
1177 * We want to be told about IP releases
1180 status = register_with_ctdbd(conn, CTDB_SRVID_RELEASE_IP);
1181 if (!NT_STATUS_IS_OK(status)) {
1182 return status;
1185 p.dest = *server;
1186 p.src = *client;
1189 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1190 * can send an extra ack to trigger a reset for our client, so it
1191 * immediately reconnects
1193 data.dptr = (uint8_t *)&p;
1194 data.dsize = sizeof(p);
1196 return ctdbd_control(conn, CTDB_CURRENT_NODE,
1197 CTDB_CONTROL_TCP_CLIENT,
1198 CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL, NULL);
1202 * We want to handle reconfigure events
1204 NTSTATUS ctdbd_register_reconfigure(struct ctdbd_connection *conn)
1206 return register_with_ctdbd(conn, CTDB_SRVID_RECONFIGURE);
1210 persstent store. Used when we update a record in a persistent database
1212 NTSTATUS ctdbd_persistent_store(struct ctdbd_connection *conn, uint32_t db_id, TDB_DATA key, TDB_DATA data)
1214 int cstatus=0;
1215 struct ctdb_rec_data *rec;
1216 TDB_DATA recdata;
1217 size_t length;
1218 NTSTATUS status;
1220 length = offsetof(struct ctdb_rec_data, data) + key.dsize + data.dsize;
1222 rec = (struct ctdb_rec_data *)talloc_size(conn, length);
1223 NT_STATUS_HAVE_NO_MEMORY(rec);
1225 rec->length = length;
1226 rec->reqid = db_id;
1227 rec->keylen = key.dsize;
1228 rec->datalen= data.dsize;
1229 memcpy(&rec->data[0], key.dptr, key.dsize);
1230 memcpy(&rec->data[key.dsize], data.dptr, data.dsize);
1232 recdata.dptr = (uint8_t *)rec;
1233 recdata.dsize = length;
1235 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1236 CTDB_CONTROL_PERSISTENT_STORE,
1237 0, recdata, NULL, NULL, &cstatus);
1238 if (cstatus != 0) {
1239 return NT_STATUS_INTERNAL_DB_CORRUPTION;
1241 return status;
1245 #else
1247 NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
1248 struct ctdbd_connection **pconn)
1250 return NT_STATUS_NOT_IMPLEMENTED;
1253 #endif