s3:dbwrap_ctdb: fix db_ctdb_fetch_db_seqnum_from_db() when NT_STATUS_NOT_FOUND.
[Samba/ekacnet.git] / source3 / lib / ctdbd_conn.c
blob743594d43b2740819665f609271d62619e95c362
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, uint32_t flags, 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, 0,
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, 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(1, ("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 uint8_t *buf, size_t available,
159 size_t *length,
160 void *private_data)
162 uint32 msglen;
164 if (available < sizeof(msglen)) {
165 return False;
168 msglen = *((uint32 *)buf);
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 (available < msglen) {
180 return false;
183 *length = msglen;
184 return true;
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 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(uint8_t *buf, size_t length,
224 void *private_data)
226 struct req_pull_state *state = (struct req_pull_state *)private_data;
228 state->req.data = talloc_move(state->mem_ctx, &buf);
229 state->req.length = length;
230 return NT_STATUS_OK;
234 * Fetch a messaging_rec from an incoming ctdb style message
237 static struct messaging_rec *ctdb_pull_messaging_rec(TALLOC_CTX *mem_ctx,
238 size_t overall_length,
239 struct ctdb_req_message *msg)
241 struct messaging_rec *result;
242 DATA_BLOB blob;
243 enum ndr_err_code ndr_err;
245 if ((overall_length < offsetof(struct ctdb_req_message, data))
246 || (overall_length
247 < offsetof(struct ctdb_req_message, data) + msg->datalen)) {
249 cluster_fatal("got invalid msg length");
252 if (!(result = TALLOC_P(mem_ctx, struct messaging_rec))) {
253 DEBUG(0, ("talloc failed\n"));
254 return NULL;
257 blob = data_blob_const(msg->data, msg->datalen);
259 ndr_err = ndr_pull_struct_blob(
260 &blob, result, NULL, result,
261 (ndr_pull_flags_fn_t)ndr_pull_messaging_rec);
263 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
264 DEBUG(0, ("ndr_pull_struct_blob failed: %s\n",
265 ndr_errstr(ndr_err)));
266 TALLOC_FREE(result);
267 return NULL;
270 if (DEBUGLEVEL >= 10) {
271 DEBUG(10, ("ctdb_pull_messaging_rec:\n"));
272 NDR_PRINT_DEBUG(messaging_rec, result);
275 return result;
278 static NTSTATUS ctdb_packet_fd_read_sync(struct packet_context *ctx)
280 struct timeval timeout;
281 struct timeval *ptimeout;
283 timeout = timeval_set(lp_ctdb_timeout(), 0);
284 ptimeout = (timeout.tv_sec != 0) ? &timeout : NULL;
286 return packet_fd_read_sync(ctx, ptimeout);
290 * Read a full ctdbd request. If we have a messaging context, defer incoming
291 * messages that might come in between.
294 static NTSTATUS ctdb_read_req(struct ctdbd_connection *conn, uint32 reqid,
295 TALLOC_CTX *mem_ctx, void *result)
297 struct ctdb_req_header *hdr;
298 struct req_pull_state state;
299 NTSTATUS status;
301 again:
303 status = ctdb_packet_fd_read_sync(conn->pkt);
305 if (NT_STATUS_EQUAL(status, NT_STATUS_NETWORK_BUSY)) {
306 /* EAGAIN */
307 goto again;
308 } else if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
309 /* EAGAIN */
310 goto again;
313 if (!NT_STATUS_IS_OK(status)) {
314 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
315 cluster_fatal("ctdbd died\n");
318 next_pkt:
320 ZERO_STRUCT(state);
321 state.mem_ctx = mem_ctx;
323 if (!packet_handler(conn->pkt, ctdb_req_complete, ctdb_req_pull,
324 &state, &status)) {
326 * Not enough data
328 DEBUG(10, ("not enough data from ctdb socket, retrying\n"));
329 goto again;
332 if (!NT_STATUS_IS_OK(status)) {
333 DEBUG(0, ("Could not read packet: %s\n", nt_errstr(status)));
334 cluster_fatal("ctdbd died\n");
337 hdr = (struct ctdb_req_header *)state.req.data;
339 DEBUG(10, ("Received ctdb packet\n"));
340 ctdb_packet_dump(hdr);
342 if (hdr->operation == CTDB_REQ_MESSAGE) {
343 struct timed_event *evt;
344 struct deferred_msg_state *msg_state;
345 struct ctdb_req_message *msg = (struct ctdb_req_message *)hdr;
347 if (conn->msg_ctx == NULL) {
348 DEBUG(1, ("Got a message without having a msg ctx, "
349 "dropping msg %llu\n",
350 (long long unsigned)msg->srvid));
351 goto next_pkt;
354 if ((conn->release_ip_handler != NULL)
355 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
356 /* must be dispatched immediately */
357 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
358 conn->release_ip_handler((const char *)msg->data,
359 conn->release_ip_priv);
360 TALLOC_FREE(hdr);
361 goto next_pkt;
364 if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
365 || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)) {
367 DEBUG(1, ("ctdb_read_req: Got %s message\n",
368 (msg->srvid == CTDB_SRVID_RECONFIGURE)
369 ? "cluster reconfigure" : "SAMBA_NOTIFY"));
371 messaging_send(conn->msg_ctx, procid_self(),
372 MSG_SMB_BRL_VALIDATE, &data_blob_null);
373 messaging_send(conn->msg_ctx, procid_self(),
374 MSG_DBWRAP_G_LOCK_RETRY,
375 &data_blob_null);
376 TALLOC_FREE(hdr);
377 goto next_pkt;
380 if (!(msg_state = TALLOC_P(talloc_autofree_context(), struct deferred_msg_state))) {
381 DEBUG(0, ("talloc failed\n"));
382 TALLOC_FREE(hdr);
383 goto next_pkt;
386 if (!(msg_state->rec = ctdb_pull_messaging_rec(
387 msg_state, state.req.length, msg))) {
388 DEBUG(0, ("ctdbd_pull_messaging_rec failed\n"));
389 TALLOC_FREE(msg_state);
390 TALLOC_FREE(hdr);
391 goto next_pkt;
394 TALLOC_FREE(hdr);
396 msg_state->msg_ctx = conn->msg_ctx;
399 * We're waiting for a call reply, but an async message has
400 * crossed. Defer dispatching to the toplevel event loop.
402 evt = event_add_timed(conn->msg_ctx->event_ctx,
403 conn->msg_ctx->event_ctx,
404 timeval_zero(),
405 deferred_message_dispatch,
406 msg_state);
407 if (evt == NULL) {
408 DEBUG(0, ("event_add_timed failed\n"));
409 TALLOC_FREE(msg_state);
410 TALLOC_FREE(hdr);
411 goto next_pkt;
414 goto next_pkt;
417 if (hdr->reqid != reqid) {
418 /* we got the wrong reply */
419 DEBUG(0,("Discarding mismatched ctdb reqid %u should have "
420 "been %u\n", hdr->reqid, reqid));
421 TALLOC_FREE(hdr);
422 goto again;
425 *((void **)result) = talloc_move(mem_ctx, &hdr);
427 return NT_STATUS_OK;
431 * Get us a ctdbd connection
434 NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
435 struct ctdbd_connection **pconn)
437 struct ctdbd_connection *conn;
438 NTSTATUS status;
440 if (!(conn = TALLOC_ZERO_P(mem_ctx, struct ctdbd_connection))) {
441 DEBUG(0, ("talloc failed\n"));
442 return NT_STATUS_NO_MEMORY;
445 status = ctdbd_connect(conn, &conn->pkt);
447 if (!NT_STATUS_IS_OK(status)) {
448 DEBUG(10, ("ctdbd_connect failed: %s\n", nt_errstr(status)));
449 goto fail;
452 status = get_cluster_vnn(conn, &conn->our_vnn);
454 if (!NT_STATUS_IS_OK(status)) {
455 DEBUG(10, ("get_cluster_vnn failed: %s\n", nt_errstr(status)));
456 goto fail;
459 generate_random_buffer((unsigned char *)&conn->rand_srvid,
460 sizeof(conn->rand_srvid));
462 status = register_with_ctdbd(conn, conn->rand_srvid);
464 if (!NT_STATUS_IS_OK(status)) {
465 DEBUG(5, ("Could not register random srvid: %s\n",
466 nt_errstr(status)));
467 goto fail;
470 *pconn = conn;
471 return NT_STATUS_OK;
473 fail:
474 TALLOC_FREE(conn);
475 return status;
479 * Get us a ctdbd connection and register us as a process
482 NTSTATUS ctdbd_messaging_connection(TALLOC_CTX *mem_ctx,
483 struct ctdbd_connection **pconn)
485 struct ctdbd_connection *conn;
486 NTSTATUS status;
488 status = ctdbd_init_connection(mem_ctx, &conn);
490 if (!NT_STATUS_IS_OK(status)) {
491 return status;
494 status = register_with_ctdbd(conn, (uint64_t)sys_getpid());
495 if (!NT_STATUS_IS_OK(status)) {
496 goto fail;
499 status = register_with_ctdbd(conn, MSG_SRVID_SAMBA);
500 if (!NT_STATUS_IS_OK(status)) {
501 goto fail;
504 status = register_with_ctdbd(conn, CTDB_SRVID_SAMBA_NOTIFY);
505 if (!NT_STATUS_IS_OK(status)) {
506 goto fail;
509 *pconn = conn;
510 return NT_STATUS_OK;
512 fail:
513 TALLOC_FREE(conn);
514 return status;
517 struct messaging_context *ctdb_conn_msg_ctx(struct ctdbd_connection *conn)
519 return conn->msg_ctx;
523 * Packet handler to receive and handle a ctdb message
525 static NTSTATUS ctdb_handle_message(uint8_t *buf, size_t length,
526 void *private_data)
528 struct ctdbd_connection *conn = talloc_get_type_abort(
529 private_data, struct ctdbd_connection);
530 struct ctdb_req_message *msg;
531 struct messaging_rec *msg_rec;
533 msg = (struct ctdb_req_message *)buf;
535 if (msg->hdr.operation != CTDB_REQ_MESSAGE) {
536 DEBUG(0, ("Received async msg of type %u, discarding\n",
537 msg->hdr.operation));
538 TALLOC_FREE(buf);
539 return NT_STATUS_INVALID_PARAMETER;
542 if ((conn->release_ip_handler != NULL)
543 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
544 /* must be dispatched immediately */
545 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
546 conn->release_ip_handler((const char *)msg->data,
547 conn->release_ip_priv);
548 TALLOC_FREE(buf);
549 return NT_STATUS_OK;
552 SMB_ASSERT(conn->msg_ctx != NULL);
554 if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
555 || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)){
556 DEBUG(0,("Got cluster reconfigure message\n"));
558 * when the cluster is reconfigured or someone of the
559 * family has passed away (SAMBA_NOTIFY), we need to
560 * clean the brl database
562 messaging_send(conn->msg_ctx, procid_self(),
563 MSG_SMB_BRL_VALIDATE, &data_blob_null);
565 messaging_send(conn->msg_ctx, procid_self(),
566 MSG_DBWRAP_G_LOCK_RETRY,
567 &data_blob_null);
569 TALLOC_FREE(buf);
570 return NT_STATUS_OK;
573 /* only messages to our pid or the broadcast are valid here */
574 if (msg->srvid != sys_getpid() && msg->srvid != MSG_SRVID_SAMBA) {
575 DEBUG(0,("Got unexpected message with srvid=%llu\n",
576 (unsigned long long)msg->srvid));
577 TALLOC_FREE(buf);
578 return NT_STATUS_OK;
581 if (!(msg_rec = ctdb_pull_messaging_rec(NULL, length, msg))) {
582 DEBUG(10, ("ctdb_pull_messaging_rec failed\n"));
583 TALLOC_FREE(buf);
584 return NT_STATUS_NO_MEMORY;
587 messaging_dispatch_rec(conn->msg_ctx, msg_rec);
589 TALLOC_FREE(msg_rec);
590 TALLOC_FREE(buf);
591 return NT_STATUS_OK;
595 * The ctdbd socket is readable asynchronuously
598 static void ctdbd_socket_handler(struct event_context *event_ctx,
599 struct fd_event *event,
600 uint16 flags,
601 void *private_data)
603 struct ctdbd_connection *conn = talloc_get_type_abort(
604 private_data, struct ctdbd_connection);
606 NTSTATUS status;
608 status = packet_fd_read(conn->pkt);
610 if (!NT_STATUS_IS_OK(status)) {
611 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
612 cluster_fatal("ctdbd died\n");
615 while (packet_handler(conn->pkt, ctdb_req_complete,
616 ctdb_handle_message, conn, &status)) {
617 if (!NT_STATUS_IS_OK(status)) {
618 DEBUG(10, ("could not handle incoming message: %s\n",
619 nt_errstr(status)));
625 * Prepare a ctdbd connection to receive messages
628 NTSTATUS ctdbd_register_msg_ctx(struct ctdbd_connection *conn,
629 struct messaging_context *msg_ctx)
631 SMB_ASSERT(conn->msg_ctx == NULL);
632 SMB_ASSERT(conn->fde == NULL);
634 if (!(conn->fde = event_add_fd(msg_ctx->event_ctx, conn,
635 packet_get_fd(conn->pkt),
636 EVENT_FD_READ,
637 ctdbd_socket_handler,
638 conn))) {
639 DEBUG(0, ("event_add_fd failed\n"));
640 return NT_STATUS_NO_MEMORY;
643 conn->msg_ctx = msg_ctx;
645 return NT_STATUS_OK;
649 * Send a messaging message across a ctdbd
652 NTSTATUS ctdbd_messaging_send(struct ctdbd_connection *conn,
653 uint32 dst_vnn, uint64 dst_srvid,
654 struct messaging_rec *msg)
656 struct ctdb_req_message r;
657 TALLOC_CTX *mem_ctx;
658 DATA_BLOB blob;
659 NTSTATUS status;
660 enum ndr_err_code ndr_err;
662 if (!(mem_ctx = talloc_init("ctdbd_messaging_send"))) {
663 DEBUG(0, ("talloc failed\n"));
664 return NT_STATUS_NO_MEMORY;
667 ndr_err = ndr_push_struct_blob(
668 &blob, mem_ctx, NULL, msg,
669 (ndr_push_flags_fn_t)ndr_push_messaging_rec);
671 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
672 DEBUG(0, ("ndr_push_struct_blob failed: %s\n",
673 ndr_errstr(ndr_err)));
674 status = ndr_map_error2ntstatus(ndr_err);
675 goto fail;
678 r.hdr.length = offsetof(struct ctdb_req_message, data) + blob.length;
679 r.hdr.ctdb_magic = CTDB_MAGIC;
680 r.hdr.ctdb_version = CTDB_VERSION;
681 r.hdr.generation = 1;
682 r.hdr.operation = CTDB_REQ_MESSAGE;
683 r.hdr.destnode = dst_vnn;
684 r.hdr.srcnode = conn->our_vnn;
685 r.hdr.reqid = 0;
686 r.srvid = dst_srvid;
687 r.datalen = blob.length;
689 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
690 ctdb_packet_dump(&r.hdr);
692 status = packet_send(
693 conn->pkt, 2,
694 data_blob_const(&r, offsetof(struct ctdb_req_message, data)),
695 blob);
697 if (!NT_STATUS_IS_OK(status)) {
698 DEBUG(0, ("packet_send failed: %s\n", nt_errstr(status)));
699 goto fail;
702 status = packet_flush(conn->pkt);
704 if (!NT_STATUS_IS_OK(status)) {
705 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
706 cluster_fatal("cluster dispatch daemon msg write error\n");
709 status = NT_STATUS_OK;
710 fail:
711 TALLOC_FREE(mem_ctx);
712 return status;
716 * send/recv a generic ctdb control message
718 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
719 uint32_t vnn, uint32 opcode,
720 uint64_t srvid, uint32_t flags,
721 TDB_DATA data,
722 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
723 int *cstatus)
725 struct ctdb_req_control req;
726 struct ctdb_reply_control *reply = NULL;
727 struct ctdbd_connection *new_conn = NULL;
728 NTSTATUS status;
730 /* the samba3 ctdb code can't handle NOREPLY yet */
731 flags &= ~CTDB_CTRL_FLAG_NOREPLY;
733 if (conn == NULL) {
734 status = ctdbd_init_connection(NULL, &new_conn);
736 if (!NT_STATUS_IS_OK(status)) {
737 DEBUG(10, ("Could not init temp connection: %s\n",
738 nt_errstr(status)));
739 goto fail;
742 conn = new_conn;
745 ZERO_STRUCT(req);
746 req.hdr.length = offsetof(struct ctdb_req_control, data) + data.dsize;
747 req.hdr.ctdb_magic = CTDB_MAGIC;
748 req.hdr.ctdb_version = CTDB_VERSION;
749 req.hdr.operation = CTDB_REQ_CONTROL;
750 req.hdr.reqid = ++conn->reqid;
751 req.hdr.destnode = vnn;
752 req.opcode = opcode;
753 req.srvid = srvid;
754 req.datalen = data.dsize;
756 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
757 ctdb_packet_dump(&req.hdr);
759 status = packet_send(
760 conn->pkt, 2,
761 data_blob_const(&req, offsetof(struct ctdb_req_control, data)),
762 data_blob_const(data.dptr, data.dsize));
764 if (!NT_STATUS_IS_OK(status)) {
765 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
766 goto fail;
769 status = packet_flush(conn->pkt);
771 if (!NT_STATUS_IS_OK(status)) {
772 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
773 cluster_fatal("cluster dispatch daemon control write error\n");
776 if (flags & CTDB_CTRL_FLAG_NOREPLY) {
777 TALLOC_FREE(new_conn);
778 return NT_STATUS_OK;
781 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
783 if (!NT_STATUS_IS_OK(status)) {
784 DEBUG(10, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
785 goto fail;
788 if (reply->hdr.operation != CTDB_REPLY_CONTROL) {
789 DEBUG(0, ("received invalid reply\n"));
790 goto fail;
793 if (outdata) {
794 if (!(outdata->dptr = (uint8 *)talloc_memdup(
795 mem_ctx, reply->data, reply->datalen))) {
796 TALLOC_FREE(reply);
797 return NT_STATUS_NO_MEMORY;
799 outdata->dsize = reply->datalen;
801 if (cstatus) {
802 (*cstatus) = reply->status;
805 status = NT_STATUS_OK;
807 fail:
808 TALLOC_FREE(new_conn);
809 TALLOC_FREE(reply);
810 return status;
814 * see if a remote process exists
816 bool ctdbd_process_exists(struct ctdbd_connection *conn, uint32 vnn, pid_t pid)
818 NTSTATUS status;
819 TDB_DATA data;
820 int32_t cstatus;
822 data.dptr = (uint8_t*)&pid;
823 data.dsize = sizeof(pid);
825 status = ctdbd_control(conn, vnn, CTDB_CONTROL_PROCESS_EXISTS, 0, 0,
826 data, NULL, NULL, &cstatus);
827 if (!NT_STATUS_IS_OK(status)) {
828 DEBUG(0, (__location__ " ctdb_control for process_exists "
829 "failed\n"));
830 return False;
833 return cstatus == 0;
837 * Get a db path
839 char *ctdbd_dbpath(struct ctdbd_connection *conn,
840 TALLOC_CTX *mem_ctx, uint32_t db_id)
842 NTSTATUS status;
843 TDB_DATA data;
844 int32_t cstatus;
846 data.dptr = (uint8_t*)&db_id;
847 data.dsize = sizeof(db_id);
849 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
850 CTDB_CONTROL_GETDBPATH, 0, 0, data,
851 mem_ctx, &data, &cstatus);
852 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
853 DEBUG(0,(__location__ " ctdb_control for getdbpath failed\n"));
854 return NULL;
857 return (char *)data.dptr;
861 * attach to a ctdb database
863 NTSTATUS ctdbd_db_attach(struct ctdbd_connection *conn,
864 const char *name, uint32_t *db_id, int tdb_flags)
866 NTSTATUS status;
867 TDB_DATA data;
868 int32_t cstatus;
869 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
871 data.dptr = (uint8_t*)name;
872 data.dsize = strlen(name)+1;
874 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
875 persistent
876 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
877 : CTDB_CONTROL_DB_ATTACH,
878 0, 0, data, NULL, &data, &cstatus);
879 if (!NT_STATUS_IS_OK(status)) {
880 DEBUG(0, (__location__ " ctdb_control for db_attach "
881 "failed: %s\n", nt_errstr(status)));
882 return status;
885 if (cstatus != 0 || data.dsize != sizeof(uint32_t)) {
886 DEBUG(0,(__location__ " ctdb_control for db_attach failed\n"));
887 return NT_STATUS_INTERNAL_ERROR;
890 *db_id = *(uint32_t *)data.dptr;
891 talloc_free(data.dptr);
893 if (!(tdb_flags & TDB_SEQNUM)) {
894 return NT_STATUS_OK;
897 data.dptr = (uint8_t *)db_id;
898 data.dsize = sizeof(*db_id);
900 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
901 CTDB_CONTROL_ENABLE_SEQNUM, 0, 0, data,
902 NULL, NULL, &cstatus);
903 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
904 DEBUG(0,(__location__ " ctdb_control for enable seqnum "
905 "failed\n"));
906 return NT_STATUS_IS_OK(status) ? NT_STATUS_INTERNAL_ERROR :
907 status;
910 return NT_STATUS_OK;
914 * force the migration of a record to this node
916 NTSTATUS ctdbd_migrate(struct ctdbd_connection *conn, uint32 db_id,
917 TDB_DATA key)
919 struct ctdb_req_call req;
920 struct ctdb_reply_call *reply;
921 NTSTATUS status;
923 ZERO_STRUCT(req);
925 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
926 req.hdr.ctdb_magic = CTDB_MAGIC;
927 req.hdr.ctdb_version = CTDB_VERSION;
928 req.hdr.operation = CTDB_REQ_CALL;
929 req.hdr.reqid = ++conn->reqid;
930 req.flags = CTDB_IMMEDIATE_MIGRATION;
931 req.callid = CTDB_NULL_FUNC;
932 req.db_id = db_id;
933 req.keylen = key.dsize;
935 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
936 ctdb_packet_dump(&req.hdr);
938 status = packet_send(
939 conn->pkt, 2,
940 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
941 data_blob_const(key.dptr, key.dsize));
943 if (!NT_STATUS_IS_OK(status)) {
944 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
945 return status;
948 status = packet_flush(conn->pkt);
950 if (!NT_STATUS_IS_OK(status)) {
951 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
952 cluster_fatal("cluster dispatch daemon control write error\n");
955 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
957 if (!NT_STATUS_IS_OK(status)) {
958 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
959 goto fail;
962 if (reply->hdr.operation != CTDB_REPLY_CALL) {
963 DEBUG(0, ("received invalid reply\n"));
964 status = NT_STATUS_INTERNAL_ERROR;
965 goto fail;
968 status = NT_STATUS_OK;
969 fail:
971 TALLOC_FREE(reply);
972 return status;
976 * remotely fetch a record without locking it or forcing a migration
978 NTSTATUS ctdbd_fetch(struct ctdbd_connection *conn, uint32 db_id,
979 TDB_DATA key, TALLOC_CTX *mem_ctx, TDB_DATA *data)
981 struct ctdb_req_call req;
982 struct ctdb_reply_call *reply;
983 NTSTATUS status;
985 ZERO_STRUCT(req);
987 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
988 req.hdr.ctdb_magic = CTDB_MAGIC;
989 req.hdr.ctdb_version = CTDB_VERSION;
990 req.hdr.operation = CTDB_REQ_CALL;
991 req.hdr.reqid = ++conn->reqid;
992 req.flags = 0;
993 req.callid = CTDB_FETCH_FUNC;
994 req.db_id = db_id;
995 req.keylen = key.dsize;
997 status = packet_send(
998 conn->pkt, 2,
999 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
1000 data_blob_const(key.dptr, key.dsize));
1002 if (!NT_STATUS_IS_OK(status)) {
1003 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
1004 return status;
1007 status = packet_flush(conn->pkt);
1009 if (!NT_STATUS_IS_OK(status)) {
1010 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
1011 cluster_fatal("cluster dispatch daemon control write error\n");
1014 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
1016 if (!NT_STATUS_IS_OK(status)) {
1017 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
1018 goto fail;
1021 if (reply->hdr.operation != CTDB_REPLY_CALL) {
1022 DEBUG(0, ("received invalid reply\n"));
1023 status = NT_STATUS_INTERNAL_ERROR;
1024 goto fail;
1027 data->dsize = reply->datalen;
1028 if (data->dsize == 0) {
1029 data->dptr = NULL;
1030 goto done;
1033 data->dptr = (uint8 *)talloc_memdup(mem_ctx, &reply->data[0],
1034 reply->datalen);
1035 if (data->dptr == NULL) {
1036 DEBUG(0, ("talloc failed\n"));
1037 status = NT_STATUS_NO_MEMORY;
1038 goto fail;
1041 done:
1042 status = NT_STATUS_OK;
1043 fail:
1044 TALLOC_FREE(reply);
1045 return status;
1048 struct ctdbd_traverse_state {
1049 void (*fn)(TDB_DATA key, TDB_DATA data, void *private_data);
1050 void *private_data;
1054 * Handle a traverse record coming in on the ctdbd connection
1057 static NTSTATUS ctdb_traverse_handler(uint8_t *buf, size_t length,
1058 void *private_data)
1060 struct ctdbd_traverse_state *state =
1061 (struct ctdbd_traverse_state *)private_data;
1063 struct ctdb_req_message *m;
1064 struct ctdb_rec_data *d;
1065 TDB_DATA key, data;
1067 m = (struct ctdb_req_message *)buf;
1069 if (length < sizeof(*m) || m->hdr.length != length) {
1070 DEBUG(0, ("Got invalid message of length %d\n", (int)length));
1071 TALLOC_FREE(buf);
1072 return NT_STATUS_UNEXPECTED_IO_ERROR;
1075 d = (struct ctdb_rec_data *)&m->data[0];
1076 if (m->datalen < sizeof(uint32_t) || m->datalen != d->length) {
1077 DEBUG(0, ("Got invalid traverse data of length %d\n",
1078 (int)m->datalen));
1079 TALLOC_FREE(buf);
1080 return NT_STATUS_UNEXPECTED_IO_ERROR;
1083 key.dsize = d->keylen;
1084 key.dptr = &d->data[0];
1085 data.dsize = d->datalen;
1086 data.dptr = &d->data[d->keylen];
1088 if (key.dsize == 0 && data.dsize == 0) {
1089 /* end of traverse */
1090 return NT_STATUS_END_OF_FILE;
1093 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1094 DEBUG(0, ("Got invalid ltdb header length %d\n",
1095 (int)data.dsize));
1096 TALLOC_FREE(buf);
1097 return NT_STATUS_UNEXPECTED_IO_ERROR;
1099 data.dsize -= sizeof(struct ctdb_ltdb_header);
1100 data.dptr += sizeof(struct ctdb_ltdb_header);
1102 if (state->fn) {
1103 state->fn(key, data, state->private_data);
1106 TALLOC_FREE(buf);
1107 return NT_STATUS_OK;
1111 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1112 connection to ctdbd to avoid the hairy recursive and async problems with
1113 everything in-line.
1116 NTSTATUS ctdbd_traverse(uint32 db_id,
1117 void (*fn)(TDB_DATA key, TDB_DATA data,
1118 void *private_data),
1119 void *private_data)
1121 struct ctdbd_connection *conn;
1122 NTSTATUS status;
1124 TDB_DATA data;
1125 struct ctdb_traverse_start t;
1126 int cstatus;
1127 struct ctdbd_traverse_state state;
1129 status = ctdbd_init_connection(NULL, &conn);
1130 if (!NT_STATUS_IS_OK(status)) {
1131 DEBUG(0, ("ctdbd_init_connection failed: %s\n",
1132 nt_errstr(status)));
1133 return status;
1136 t.db_id = db_id;
1137 t.srvid = conn->rand_srvid;
1138 t.reqid = ++conn->reqid;
1140 data.dptr = (uint8_t *)&t;
1141 data.dsize = sizeof(t);
1143 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1144 CTDB_CONTROL_TRAVERSE_START, conn->rand_srvid, 0,
1145 data, NULL, NULL, &cstatus);
1147 if (!NT_STATUS_IS_OK(status) || (cstatus != 0)) {
1149 DEBUG(0,("ctdbd_control failed: %s, %d\n", nt_errstr(status),
1150 cstatus));
1152 if (NT_STATUS_IS_OK(status)) {
1154 * We need a mapping here
1156 status = NT_STATUS_UNSUCCESSFUL;
1158 goto done;
1161 state.fn = fn;
1162 state.private_data = private_data;
1164 while (True) {
1166 status = NT_STATUS_OK;
1168 if (packet_handler(conn->pkt, ctdb_req_complete,
1169 ctdb_traverse_handler, &state, &status)) {
1171 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1172 status = NT_STATUS_OK;
1173 break;
1177 * There might be more in the queue
1179 continue;
1182 if (!NT_STATUS_IS_OK(status)) {
1183 break;
1186 status = ctdb_packet_fd_read_sync(conn->pkt);
1188 if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
1190 * There might be more in the queue
1192 continue;
1195 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1196 status = NT_STATUS_OK;
1197 break;
1200 if (!NT_STATUS_IS_OK(status)) {
1201 DEBUG(0, ("packet_fd_read_sync failed: %s\n", nt_errstr(status)));
1202 cluster_fatal("ctdbd died\n");
1206 done:
1207 TALLOC_FREE(conn);
1208 return status;
1212 This is used to canonicalize a ctdb_sock_addr structure.
1214 static void smbd_ctdb_canonicalize_ip(const struct sockaddr_storage *in,
1215 struct sockaddr_storage *out)
1217 memcpy(out, in, sizeof (*out));
1219 #ifdef HAVE_IPV6
1220 if (in->ss_family == AF_INET6) {
1221 const char prefix[12] = { 0,0,0,0,0,0,0,0,0,0,0xff,0xff };
1222 const struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)in;
1223 struct sockaddr_in *out4 = (struct sockaddr_in *)out;
1224 if (memcmp(&in6->sin6_addr, prefix, 12) == 0) {
1225 memset(out, 0, sizeof(*out));
1226 #ifdef HAVE_SOCK_SIN_LEN
1227 out4->sin_len = sizeof(*out);
1228 #endif
1229 out4->sin_family = AF_INET;
1230 out4->sin_port = in6->sin6_port;
1231 memcpy(&out4->sin_addr, &in6->sin6_addr.s6_addr32[3], 4);
1234 #endif
1238 * Register us as a server for a particular tcp connection
1241 NTSTATUS ctdbd_register_ips(struct ctdbd_connection *conn,
1242 const struct sockaddr_storage *_server,
1243 const struct sockaddr_storage *_client,
1244 void (*release_ip_handler)(const char *ip_addr,
1245 void *private_data),
1246 void *private_data)
1249 * we still use ctdb_control_tcp for ipv4
1250 * because we want to work against older ctdb
1251 * versions at runtime
1253 struct ctdb_control_tcp p4;
1254 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1255 struct ctdb_control_tcp_addr p;
1256 #endif
1257 TDB_DATA data;
1258 NTSTATUS status;
1259 struct sockaddr_storage client;
1260 struct sockaddr_storage server;
1263 * Only one connection so far
1265 SMB_ASSERT(conn->release_ip_handler == NULL);
1267 smbd_ctdb_canonicalize_ip(_client, &client);
1268 smbd_ctdb_canonicalize_ip(_server, &server);
1270 switch (client.ss_family) {
1271 case AF_INET:
1272 p4.dest = *(struct sockaddr_in *)&server;
1273 p4.src = *(struct sockaddr_in *)&client;
1274 data.dptr = (uint8_t *)&p4;
1275 data.dsize = sizeof(p4);
1276 break;
1277 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1278 case AF_INET6:
1279 p.dest.ip6 = *(struct sockaddr_in6 *)&server;
1280 p.src.ip6 = *(struct sockaddr_in6 *)&client;
1281 data.dptr = (uint8_t *)&p;
1282 data.dsize = sizeof(p);
1283 break;
1284 #endif
1285 default:
1286 return NT_STATUS_INTERNAL_ERROR;
1289 conn->release_ip_handler = release_ip_handler;
1292 * We want to be told about IP releases
1295 status = register_with_ctdbd(conn, CTDB_SRVID_RELEASE_IP);
1296 if (!NT_STATUS_IS_OK(status)) {
1297 return status;
1301 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1302 * can send an extra ack to trigger a reset for our client, so it
1303 * immediately reconnects
1305 return ctdbd_control(conn, CTDB_CURRENT_NODE,
1306 CTDB_CONTROL_TCP_CLIENT, 0,
1307 CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL, NULL);
1311 * We want to handle reconfigure events
1313 NTSTATUS ctdbd_register_reconfigure(struct ctdbd_connection *conn)
1315 return register_with_ctdbd(conn, CTDB_SRVID_RECONFIGURE);
1319 call a control on the local node
1321 NTSTATUS ctdbd_control_local(struct ctdbd_connection *conn, uint32 opcode,
1322 uint64_t srvid, uint32_t flags, TDB_DATA data,
1323 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
1324 int *cstatus)
1326 return ctdbd_control(conn, CTDB_CURRENT_NODE, opcode, srvid, flags, data, mem_ctx, outdata, cstatus);
1329 NTSTATUS ctdb_watch_us(struct ctdbd_connection *conn)
1331 struct ctdb_client_notify_register reg_data;
1332 size_t struct_len;
1333 NTSTATUS status;
1334 int cstatus;
1336 reg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1337 reg_data.len = 1;
1338 reg_data.notify_data[0] = 0;
1340 struct_len = offsetof(struct ctdb_client_notify_register,
1341 notify_data) + reg_data.len;
1343 status = ctdbd_control_local(
1344 conn, CTDB_CONTROL_REGISTER_NOTIFY, conn->rand_srvid, 0,
1345 make_tdb_data((uint8_t *)&reg_data, struct_len),
1346 NULL, NULL, &cstatus);
1347 if (!NT_STATUS_IS_OK(status)) {
1348 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1349 nt_errstr(status)));
1351 return status;
1354 NTSTATUS ctdb_unwatch(struct ctdbd_connection *conn)
1356 struct ctdb_client_notify_deregister dereg_data;
1357 NTSTATUS status;
1358 int cstatus;
1360 dereg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1362 status = ctdbd_control_local(
1363 conn, CTDB_CONTROL_DEREGISTER_NOTIFY, conn->rand_srvid, 0,
1364 make_tdb_data((uint8_t *)&dereg_data, sizeof(dereg_data)),
1365 NULL, NULL, &cstatus);
1366 if (!NT_STATUS_IS_OK(status)) {
1367 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1368 nt_errstr(status)));
1370 return status;
1373 #else
1375 NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
1376 struct ctdbd_connection **pconn)
1378 return NT_STATUS_NOT_IMPLEMENTED;
1381 #endif