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/>.
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 */
30 #include "ctdb_private.h"
32 struct ctdbd_connection
{
33 struct messaging_context
*msg_ctx
;
37 struct packet_context
*pkt
;
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
,
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
66 static void ctdb_packet_dump(struct ctdb_req_header
*hdr
)
68 if (DEBUGLEVEL
< 10) {
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
,
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
)
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
;
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
;
124 if (!sockname
|| !*sockname
) {
125 sockname
= CTDB_PATH
;
128 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
130 DEBUG(3, ("Could not create socket: %s\n", strerror(errno
)));
131 return map_nt_error_from_unix(errno
);
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
,
142 return map_nt_error_from_unix(errno
);
145 if (!(result
= packet_init(mem_ctx
, fd
))) {
147 return NT_STATUS_NO_MEMORY
;
155 * Do we have a complete ctdb packet in the queue?
158 static bool ctdb_req_complete(const struct data_blob
*data
,
164 if (data
->length
< sizeof(msglen
)) {
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
) {
188 * State necessary to defer an incoming message while we are waiting for a
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
,
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
);
214 struct req_pull_state
{
220 * Pull a ctdb request out of the incoming packet queue
223 static NTSTATUS
ctdb_req_pull(const struct data_blob
*data
,
226 struct req_pull_state
*state
= (struct req_pull_state
*)private_data
;
228 state
->req
= data_blob_talloc(state
->mem_ctx
, data
->data
,
230 if (state
->req
.data
== NULL
) {
231 return NT_STATUS_NO_MEMORY
;
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
;
246 enum ndr_err_code ndr_err
;
248 if ((overall_length
< offsetof(struct ctdb_req_message
, data
))
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"));
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
)));
273 if (DEBUGLEVEL
>= 10) {
274 DEBUG(10, ("ctdb_pull_messaging_rec:\n"));
275 NDR_PRINT_DEBUG(messaging_rec
, 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
;
295 status
= packet_fd_read_sync(conn
->pkt
);
297 if (NT_STATUS_EQUAL(status
, NT_STATUS_NETWORK_BUSY
)) {
300 } else if (NT_STATUS_EQUAL(status
, NT_STATUS_RETRY
)) {
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");
313 state
.mem_ctx
= mem_ctx
;
315 if (!packet_handler(conn
->pkt
, ctdb_req_complete
, ctdb_req_pull
,
320 DEBUG(10, ("not enough data from ctdb socket, retrying\n"));
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
));
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
);
356 if (msg
->srvid
== CTDB_SRVID_RECONFIGURE
) {
357 DEBUG(0,("Got cluster reconfigure message in ctdb_read_req\n"));
358 messaging_send(conn
->msg_ctx
, procid_self(),
359 MSG_SMB_BRL_VALIDATE
, &data_blob_null
);
364 if (!(msg_state
= TALLOC_P(NULL
, struct deferred_msg_state
))) {
365 DEBUG(0, ("talloc failed\n"));
370 if (!(msg_state
->rec
= ctdb_pull_messaging_rec(
371 msg_state
, state
.req
.length
, msg
))) {
372 DEBUG(0, ("ctdbd_pull_messaging_rec failed\n"));
373 TALLOC_FREE(msg_state
);
380 msg_state
->msg_ctx
= conn
->msg_ctx
;
383 * We're waiting for a call reply, but an async message has
384 * crossed. Defer dispatching to the toplevel event loop.
386 evt
= event_add_timed(conn
->msg_ctx
->event_ctx
,
387 conn
->msg_ctx
->event_ctx
,
389 "deferred_message_dispatch",
390 deferred_message_dispatch
,
393 DEBUG(0, ("event_add_timed failed\n"));
394 TALLOC_FREE(msg_state
);
402 if (hdr
->reqid
!= reqid
) {
403 /* we got the wrong reply */
404 DEBUG(0,("Discarding mismatched ctdb reqid %u should have "
405 "been %u\n", hdr
->reqid
, reqid
));
410 *((void **)result
) = talloc_move(mem_ctx
, &hdr
);
416 * Get us a ctdbd connection
419 NTSTATUS
ctdbd_init_connection(TALLOC_CTX
*mem_ctx
,
420 struct ctdbd_connection
**pconn
)
422 struct ctdbd_connection
*conn
;
425 if (!(conn
= TALLOC_ZERO_P(mem_ctx
, struct ctdbd_connection
))) {
426 DEBUG(0, ("talloc failed\n"));
427 return NT_STATUS_NO_MEMORY
;
430 status
= ctdbd_connect(conn
, &conn
->pkt
);
432 if (!NT_STATUS_IS_OK(status
)) {
433 DEBUG(10, ("ctdbd_connect failed: %s\n", nt_errstr(status
)));
437 status
= get_cluster_vnn(conn
, &conn
->our_vnn
);
439 if (!NT_STATUS_IS_OK(status
)) {
440 DEBUG(10, ("get_cluster_vnn failed: %s\n", nt_errstr(status
)));
444 generate_random_buffer((unsigned char *)&conn
->rand_srvid
,
445 sizeof(conn
->rand_srvid
));
447 status
= register_with_ctdbd(conn
, conn
->rand_srvid
);
449 if (!NT_STATUS_IS_OK(status
)) {
450 DEBUG(5, ("Could not register random srvid: %s\n",
464 * Get us a ctdbd connection and register us as a process
467 NTSTATUS
ctdbd_messaging_connection(TALLOC_CTX
*mem_ctx
,
468 struct ctdbd_connection
**pconn
)
470 struct ctdbd_connection
*conn
;
473 status
= ctdbd_init_connection(mem_ctx
, &conn
);
475 if (!NT_STATUS_IS_OK(status
)) {
479 status
= register_with_ctdbd(conn
, (uint64_t)sys_getpid());
480 if (!NT_STATUS_IS_OK(status
)) {
484 status
= register_with_ctdbd(conn
, MSG_SRVID_SAMBA
);
485 if (!NT_STATUS_IS_OK(status
)) {
498 * Packet handler to receive and handle a ctdb message
500 static NTSTATUS
ctdb_handle_message(const struct data_blob
*data
,
503 struct ctdbd_connection
*conn
= talloc_get_type_abort(
504 private_data
, struct ctdbd_connection
);
505 struct ctdb_req_message
*msg
;
506 struct messaging_rec
*msg_rec
;
508 msg
= (struct ctdb_req_message
*)data
->data
;
510 if (msg
->hdr
.operation
!= CTDB_REQ_MESSAGE
) {
511 DEBUG(0, ("Received async msg of type %u, discarding\n",
512 msg
->hdr
.operation
));
513 return NT_STATUS_INVALID_PARAMETER
;
516 if ((conn
->release_ip_handler
!= NULL
)
517 && (msg
->srvid
== CTDB_SRVID_RELEASE_IP
)) {
518 /* must be dispatched immediately */
519 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
520 conn
->release_ip_handler((const char *)msg
->data
,
521 conn
->release_ip_priv
);
525 SMB_ASSERT(conn
->msg_ctx
!= NULL
);
527 if (msg
->srvid
== CTDB_SRVID_RECONFIGURE
) {
528 DEBUG(0,("Got cluster reconfigure message\n"));
530 * when the cluster is reconfigured, we need to clean the brl
533 messaging_send(conn
->msg_ctx
, procid_self(),
534 MSG_SMB_BRL_VALIDATE
, &data_blob_null
);
537 * it's possible that we have just rejoined the cluster after
538 * an outage. In that case our pending locks could have been
539 * removed from the lockdb, so retry them once more
541 message_send_all(conn
->msg_ctx
, MSG_SMB_UNLOCK
, NULL
, 0, NULL
);
547 /* only messages to our pid or the broadcast are valid here */
548 if (msg
->srvid
!= sys_getpid() && msg
->srvid
!= MSG_SRVID_SAMBA
) {
549 DEBUG(0,("Got unexpected message with srvid=%llu\n",
550 (unsigned long long)msg
->srvid
));
554 if (!(msg_rec
= ctdb_pull_messaging_rec(NULL
, data
->length
, msg
))) {
555 DEBUG(10, ("ctdb_pull_messaging_rec failed\n"));
556 return NT_STATUS_NO_MEMORY
;
559 messaging_dispatch_rec(conn
->msg_ctx
, msg_rec
);
561 TALLOC_FREE(msg_rec
);
566 * The ctdbd socket is readable asynchronuously
569 static void ctdbd_socket_handler(struct event_context
*event_ctx
,
570 struct fd_event
*event
,
574 struct ctdbd_connection
*conn
= talloc_get_type_abort(
575 private_data
, struct ctdbd_connection
);
579 status
= packet_fd_read(conn
->pkt
);
581 if (!NT_STATUS_IS_OK(status
)) {
582 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status
)));
583 cluster_fatal("ctdbd died\n");
586 while (packet_handler(conn
->pkt
, ctdb_req_complete
,
587 ctdb_handle_message
, conn
, &status
)) {
588 if (!NT_STATUS_IS_OK(status
)) {
589 DEBUG(10, ("could not handle incoming message: %s\n",
596 * Prepare a ctdbd connection to receive messages
599 NTSTATUS
ctdbd_register_msg_ctx(struct ctdbd_connection
*conn
,
600 struct messaging_context
*msg_ctx
)
602 SMB_ASSERT(conn
->msg_ctx
== NULL
);
603 SMB_ASSERT(conn
->fde
== NULL
);
605 if (!(conn
->fde
= event_add_fd(msg_ctx
->event_ctx
, conn
,
606 packet_get_fd(conn
->pkt
),
608 ctdbd_socket_handler
,
610 DEBUG(0, ("event_add_fd failed\n"));
611 return NT_STATUS_NO_MEMORY
;
614 conn
->msg_ctx
= msg_ctx
;
620 * Send a messaging message across a ctdbd
623 NTSTATUS
ctdbd_messaging_send(struct ctdbd_connection
*conn
,
624 uint32 dst_vnn
, uint64 dst_srvid
,
625 struct messaging_rec
*msg
)
627 struct ctdb_req_message r
;
631 enum ndr_err_code ndr_err
;
633 if (!(mem_ctx
= talloc_init("ctdbd_messaging_send"))) {
634 DEBUG(0, ("talloc failed\n"));
635 return NT_STATUS_NO_MEMORY
;
638 ndr_err
= ndr_push_struct_blob(
640 (ndr_push_flags_fn_t
)ndr_push_messaging_rec
);
642 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err
)) {
643 DEBUG(0, ("ndr_push_struct_blob failed: %s\n",
644 ndr_errstr(ndr_err
)));
645 status
= ndr_map_error2ntstatus(ndr_err
);
649 r
.hdr
.length
= offsetof(struct ctdb_req_message
, data
) + blob
.length
;
650 r
.hdr
.ctdb_magic
= CTDB_MAGIC
;
651 r
.hdr
.ctdb_version
= CTDB_VERSION
;
652 r
.hdr
.generation
= 1;
653 r
.hdr
.operation
= CTDB_REQ_MESSAGE
;
654 r
.hdr
.destnode
= dst_vnn
;
655 r
.hdr
.srcnode
= conn
->our_vnn
;
658 r
.datalen
= blob
.length
;
660 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
661 ctdb_packet_dump(&r
.hdr
);
663 status
= packet_send(
665 data_blob_const(&r
, offsetof(struct ctdb_req_message
, data
)),
668 if (!NT_STATUS_IS_OK(status
)) {
669 DEBUG(0, ("packet_send failed: %s\n", nt_errstr(status
)));
673 status
= packet_flush(conn
->pkt
);
675 if (!NT_STATUS_IS_OK(status
)) {
676 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status
)));
677 cluster_fatal("cluster dispatch daemon msg write error\n");
680 status
= NT_STATUS_OK
;
682 TALLOC_FREE(mem_ctx
);
687 * send/recv a generic ctdb control message
689 static NTSTATUS
ctdbd_control(struct ctdbd_connection
*conn
,
690 uint32_t vnn
, uint32 opcode
,
691 uint64_t srvid
, uint32_t flags
,
693 TALLOC_CTX
*mem_ctx
, TDB_DATA
*outdata
,
696 struct ctdb_req_control req
;
697 struct ctdb_reply_control
*reply
= NULL
;
698 struct ctdbd_connection
*new_conn
= NULL
;
701 /* the samba3 ctdb code can't handle NOREPLY yet */
702 flags
&= ~CTDB_CTRL_FLAG_NOREPLY
;
705 status
= ctdbd_init_connection(NULL
, &new_conn
);
707 if (!NT_STATUS_IS_OK(status
)) {
708 DEBUG(10, ("Could not init temp connection: %s\n",
717 req
.hdr
.length
= offsetof(struct ctdb_req_control
, data
) + data
.dsize
;
718 req
.hdr
.ctdb_magic
= CTDB_MAGIC
;
719 req
.hdr
.ctdb_version
= CTDB_VERSION
;
720 req
.hdr
.operation
= CTDB_REQ_CONTROL
;
721 req
.hdr
.reqid
= ++conn
->reqid
;
722 req
.hdr
.destnode
= vnn
;
725 req
.datalen
= data
.dsize
;
727 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
728 ctdb_packet_dump(&req
.hdr
);
730 status
= packet_send(
732 data_blob_const(&req
, offsetof(struct ctdb_req_control
, data
)),
733 data_blob_const(data
.dptr
, data
.dsize
));
735 if (!NT_STATUS_IS_OK(status
)) {
736 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status
)));
740 status
= packet_flush(conn
->pkt
);
742 if (!NT_STATUS_IS_OK(status
)) {
743 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status
)));
744 cluster_fatal("cluster dispatch daemon control write error\n");
747 if (flags
& CTDB_CTRL_FLAG_NOREPLY
) {
748 TALLOC_FREE(new_conn
);
752 status
= ctdb_read_req(conn
, req
.hdr
.reqid
, NULL
, (void *)&reply
);
754 if (!NT_STATUS_IS_OK(status
)) {
755 DEBUG(10, ("ctdb_read_req failed: %s\n", nt_errstr(status
)));
759 if (reply
->hdr
.operation
!= CTDB_REPLY_CONTROL
) {
760 DEBUG(0, ("received invalid reply\n"));
765 if (!(outdata
->dptr
= (uint8
*)talloc_memdup(
766 mem_ctx
, reply
->data
, reply
->datalen
))) {
768 return NT_STATUS_NO_MEMORY
;
770 outdata
->dsize
= reply
->datalen
;
773 (*cstatus
) = reply
->status
;
776 status
= NT_STATUS_OK
;
779 TALLOC_FREE(new_conn
);
785 * see if a remote process exists
787 bool ctdbd_process_exists(struct ctdbd_connection
*conn
, uint32 vnn
, pid_t pid
)
793 data
.dptr
= (uint8_t*)&pid
;
794 data
.dsize
= sizeof(pid
);
796 status
= ctdbd_control(conn
, vnn
, CTDB_CONTROL_PROCESS_EXISTS
, 0, 0,
797 data
, NULL
, NULL
, &cstatus
);
798 if (!NT_STATUS_IS_OK(status
)) {
799 DEBUG(0, (__location__
" ctdb_control for process_exists "
810 char *ctdbd_dbpath(struct ctdbd_connection
*conn
,
811 TALLOC_CTX
*mem_ctx
, uint32_t db_id
)
817 data
.dptr
= (uint8_t*)&db_id
;
818 data
.dsize
= sizeof(db_id
);
820 status
= ctdbd_control(conn
, CTDB_CURRENT_NODE
,
821 CTDB_CONTROL_GETDBPATH
, 0, 0, data
,
822 mem_ctx
, &data
, &cstatus
);
823 if (!NT_STATUS_IS_OK(status
) || cstatus
!= 0) {
824 DEBUG(0,(__location__
" ctdb_control for getdbpath failed\n"));
828 return (char *)data
.dptr
;
832 * attach to a ctdb database
834 NTSTATUS
ctdbd_db_attach(struct ctdbd_connection
*conn
,
835 const char *name
, uint32_t *db_id
, int tdb_flags
)
840 bool persistent
= (tdb_flags
& TDB_CLEAR_IF_FIRST
) == 0;
842 data
.dptr
= (uint8_t*)name
;
843 data
.dsize
= strlen(name
)+1;
845 status
= ctdbd_control(conn
, CTDB_CURRENT_NODE
,
847 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
848 : CTDB_CONTROL_DB_ATTACH
,
849 0, 0, data
, NULL
, &data
, &cstatus
);
850 if (!NT_STATUS_IS_OK(status
)) {
851 DEBUG(0, (__location__
" ctdb_control for db_attach "
852 "failed: %s\n", nt_errstr(status
)));
856 if (cstatus
!= 0 || data
.dsize
!= sizeof(uint32_t)) {
857 DEBUG(0,(__location__
" ctdb_control for db_attach failed\n"));
858 return NT_STATUS_INTERNAL_ERROR
;
861 *db_id
= *(uint32_t *)data
.dptr
;
862 talloc_free(data
.dptr
);
864 if (!(tdb_flags
& TDB_SEQNUM
)) {
868 data
.dptr
= (uint8_t *)db_id
;
869 data
.dsize
= sizeof(*db_id
);
871 status
= ctdbd_control(conn
, CTDB_CURRENT_NODE
,
872 CTDB_CONTROL_ENABLE_SEQNUM
, 0, 0, data
,
873 NULL
, NULL
, &cstatus
);
874 if (!NT_STATUS_IS_OK(status
) || cstatus
!= 0) {
875 DEBUG(0,(__location__
" ctdb_control for enable seqnum "
877 return NT_STATUS_IS_OK(status
) ? NT_STATUS_INTERNAL_ERROR
:
885 * force the migration of a record to this node
887 NTSTATUS
ctdbd_migrate(struct ctdbd_connection
*conn
, uint32 db_id
,
890 struct ctdb_req_call req
;
891 struct ctdb_reply_call
*reply
;
896 req
.hdr
.length
= offsetof(struct ctdb_req_call
, data
) + key
.dsize
;
897 req
.hdr
.ctdb_magic
= CTDB_MAGIC
;
898 req
.hdr
.ctdb_version
= CTDB_VERSION
;
899 req
.hdr
.operation
= CTDB_REQ_CALL
;
900 req
.hdr
.reqid
= ++conn
->reqid
;
901 req
.flags
= CTDB_IMMEDIATE_MIGRATION
;
902 req
.callid
= CTDB_NULL_FUNC
;
904 req
.keylen
= key
.dsize
;
906 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
907 ctdb_packet_dump(&req
.hdr
);
909 status
= packet_send(
911 data_blob_const(&req
, offsetof(struct ctdb_req_call
, data
)),
912 data_blob_const(key
.dptr
, key
.dsize
));
914 if (!NT_STATUS_IS_OK(status
)) {
915 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status
)));
919 status
= packet_flush(conn
->pkt
);
921 if (!NT_STATUS_IS_OK(status
)) {
922 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status
)));
923 cluster_fatal("cluster dispatch daemon control write error\n");
926 status
= ctdb_read_req(conn
, req
.hdr
.reqid
, NULL
, (void *)&reply
);
928 if (!NT_STATUS_IS_OK(status
)) {
929 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status
)));
933 if (reply
->hdr
.operation
!= CTDB_REPLY_CALL
) {
934 DEBUG(0, ("received invalid reply\n"));
935 status
= NT_STATUS_INTERNAL_ERROR
;
939 status
= NT_STATUS_OK
;
947 * remotely fetch a record without locking it or forcing a migration
949 NTSTATUS
ctdbd_fetch(struct ctdbd_connection
*conn
, uint32 db_id
,
950 TDB_DATA key
, TALLOC_CTX
*mem_ctx
, TDB_DATA
*data
)
952 struct ctdb_req_call req
;
953 struct ctdb_reply_call
*reply
;
958 req
.hdr
.length
= offsetof(struct ctdb_req_call
, data
) + key
.dsize
;
959 req
.hdr
.ctdb_magic
= CTDB_MAGIC
;
960 req
.hdr
.ctdb_version
= CTDB_VERSION
;
961 req
.hdr
.operation
= CTDB_REQ_CALL
;
962 req
.hdr
.reqid
= ++conn
->reqid
;
964 req
.callid
= CTDB_FETCH_FUNC
;
966 req
.keylen
= key
.dsize
;
968 status
= packet_send(
970 data_blob_const(&req
, offsetof(struct ctdb_req_call
, data
)),
971 data_blob_const(key
.dptr
, key
.dsize
));
973 if (!NT_STATUS_IS_OK(status
)) {
974 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status
)));
978 status
= packet_flush(conn
->pkt
);
980 if (!NT_STATUS_IS_OK(status
)) {
981 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status
)));
982 cluster_fatal("cluster dispatch daemon control write error\n");
985 status
= ctdb_read_req(conn
, req
.hdr
.reqid
, NULL
, (void *)&reply
);
987 if (!NT_STATUS_IS_OK(status
)) {
988 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status
)));
992 if (reply
->hdr
.operation
!= CTDB_REPLY_CALL
) {
993 DEBUG(0, ("received invalid reply\n"));
994 status
= NT_STATUS_INTERNAL_ERROR
;
998 data
->dsize
= reply
->datalen
;
999 if (data
->dsize
== 0) {
1004 data
->dptr
= (uint8
*)talloc_memdup(mem_ctx
, &reply
->data
[0],
1006 if (data
->dptr
== NULL
) {
1007 DEBUG(0, ("talloc failed\n"));
1008 status
= NT_STATUS_NO_MEMORY
;
1013 status
= NT_STATUS_OK
;
1019 struct ctdbd_traverse_state
{
1020 void (*fn
)(TDB_DATA key
, TDB_DATA data
, void *private_data
);
1025 * Handle a traverse record coming in on the ctdbd connection
1028 static NTSTATUS
ctdb_traverse_handler(const struct data_blob
*blob
,
1031 struct ctdbd_traverse_state
*state
=
1032 (struct ctdbd_traverse_state
*)private_data
;
1034 struct ctdb_req_message
*m
;
1035 struct ctdb_rec_data
*d
;
1038 m
= (struct ctdb_req_message
*)blob
->data
;
1040 if (blob
->length
< sizeof(*m
) || m
->hdr
.length
!= blob
->length
) {
1041 DEBUG(0, ("Got invalid message of length %d\n",
1042 (int)blob
->length
));
1043 return NT_STATUS_UNEXPECTED_IO_ERROR
;
1046 d
= (struct ctdb_rec_data
*)&m
->data
[0];
1047 if (m
->datalen
< sizeof(uint32_t) || m
->datalen
!= d
->length
) {
1048 DEBUG(0, ("Got invalid traverse data of length %d\n",
1050 return NT_STATUS_UNEXPECTED_IO_ERROR
;
1053 key
.dsize
= d
->keylen
;
1054 key
.dptr
= &d
->data
[0];
1055 data
.dsize
= d
->datalen
;
1056 data
.dptr
= &d
->data
[d
->keylen
];
1058 if (key
.dsize
== 0 && data
.dsize
== 0) {
1059 /* end of traverse */
1060 return NT_STATUS_END_OF_FILE
;
1063 if (data
.dsize
< sizeof(struct ctdb_ltdb_header
)) {
1064 DEBUG(0, ("Got invalid ltdb header length %d\n",
1066 return NT_STATUS_UNEXPECTED_IO_ERROR
;
1068 data
.dsize
-= sizeof(struct ctdb_ltdb_header
);
1069 data
.dptr
+= sizeof(struct ctdb_ltdb_header
);
1072 state
->fn(key
, data
, state
->private_data
);
1075 return NT_STATUS_OK
;
1079 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1080 connection to ctdbd to avoid the hairy recursive and async problems with
1084 NTSTATUS
ctdbd_traverse(uint32 db_id
,
1085 void (*fn
)(TDB_DATA key
, TDB_DATA data
,
1086 void *private_data
),
1089 struct ctdbd_connection
*conn
;
1093 struct ctdb_traverse_start t
;
1095 struct ctdbd_traverse_state state
;
1097 status
= ctdbd_init_connection(NULL
, &conn
);
1100 t
.srvid
= conn
->rand_srvid
;
1101 t
.reqid
= ++conn
->reqid
;
1103 data
.dptr
= (uint8_t *)&t
;
1104 data
.dsize
= sizeof(t
);
1106 status
= ctdbd_control(conn
, CTDB_CURRENT_NODE
,
1107 CTDB_CONTROL_TRAVERSE_START
, conn
->rand_srvid
, 0,
1108 data
, NULL
, NULL
, &cstatus
);
1110 if (!NT_STATUS_IS_OK(status
) || (cstatus
!= 0)) {
1112 DEBUG(0,("ctdbd_control failed: %s, %d\n", nt_errstr(status
),
1115 if (NT_STATUS_IS_OK(status
)) {
1117 * We need a mapping here
1119 status
= NT_STATUS_UNSUCCESSFUL
;
1125 state
.private_data
= private_data
;
1129 status
= NT_STATUS_OK
;
1131 if (packet_handler(conn
->pkt
, ctdb_req_complete
,
1132 ctdb_traverse_handler
, &state
, &status
)) {
1134 if (NT_STATUS_EQUAL(status
, NT_STATUS_END_OF_FILE
)) {
1135 status
= NT_STATUS_OK
;
1140 * There might be more in the queue
1145 if (!NT_STATUS_IS_OK(status
)) {
1149 status
= packet_fd_read_sync(conn
->pkt
);
1151 if (NT_STATUS_EQUAL(status
, NT_STATUS_RETRY
)) {
1153 * There might be more in the queue
1158 if (NT_STATUS_EQUAL(status
, NT_STATUS_END_OF_FILE
)) {
1159 status
= NT_STATUS_OK
;
1162 if (!NT_STATUS_IS_OK(status
)) {
1163 DEBUG(0, ("packet_fd_read_sync failed: %s\n", nt_errstr(status
)));
1164 cluster_fatal("ctdbd died\n");
1174 * Register us as a server for a particular tcp connection
1177 NTSTATUS
ctdbd_register_ips(struct ctdbd_connection
*conn
,
1178 const struct sockaddr_in
*server
,
1179 const struct sockaddr_in
*client
,
1180 void (*release_ip_handler
)(const char *ip_addr
,
1181 void *private_data
),
1184 struct ctdb_control_tcp p
;
1189 * Only one connection so far
1191 SMB_ASSERT(conn
->release_ip_handler
== NULL
);
1193 conn
->release_ip_handler
= release_ip_handler
;
1196 * We want to be told about IP releases
1199 status
= register_with_ctdbd(conn
, CTDB_SRVID_RELEASE_IP
);
1200 if (!NT_STATUS_IS_OK(status
)) {
1208 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1209 * can send an extra ack to trigger a reset for our client, so it
1210 * immediately reconnects
1212 data
.dptr
= (uint8_t *)&p
;
1213 data
.dsize
= sizeof(p
);
1215 return ctdbd_control(conn
, CTDB_CURRENT_NODE
,
1216 CTDB_CONTROL_TCP_CLIENT
, 0,
1217 CTDB_CTRL_FLAG_NOREPLY
, data
, NULL
, NULL
, NULL
);
1221 * We want to handle reconfigure events
1223 NTSTATUS
ctdbd_register_reconfigure(struct ctdbd_connection
*conn
)
1225 return register_with_ctdbd(conn
, CTDB_SRVID_RECONFIGURE
);
1229 call a control on the local node
1231 NTSTATUS
ctdbd_control_local(struct ctdbd_connection
*conn
, uint32 opcode
,
1232 uint64_t srvid
, uint32_t flags
, TDB_DATA data
,
1233 TALLOC_CTX
*mem_ctx
, TDB_DATA
*outdata
,
1236 return ctdbd_control(conn
, CTDB_CURRENT_NODE
, opcode
, srvid
, flags
, data
, mem_ctx
, outdata
, cstatus
);
1241 NTSTATUS
ctdbd_init_connection(TALLOC_CTX
*mem_ctx
,
1242 struct ctdbd_connection
**pconn
)
1244 return NT_STATUS_NOT_IMPLEMENTED
;