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/>.
24 #ifdef CLUSTER_SUPPORT
26 #include "ctdbd_conn.h"
27 #include "ctdb_packet.h"
31 * It is not possible to include ctdb.h and tdb_compat.h (included via
32 * some other include above) without warnings. This fixes those
40 #ifdef typesafe_cb_preargs
41 #undef typesafe_cb_preargs
44 #ifdef typesafe_cb_postargs
45 #undef typesafe_cb_postargs
48 /* paths to these include files come from --with-ctdb= in configure */
51 #include "ctdb_private.h"
53 struct ctdbd_connection
{
54 struct messaging_context
*msg_ctx
;
58 struct ctdb_packet_context
*pkt
;
61 void (*release_ip_handler
)(const char *ip_addr
, void *private_data
);
62 void *release_ip_priv
;
65 static uint32_t ctdbd_next_reqid(struct ctdbd_connection
*conn
)
68 if (conn
->reqid
== 0) {
74 static NTSTATUS
ctdbd_control(struct ctdbd_connection
*conn
,
75 uint32_t vnn
, uint32 opcode
,
76 uint64_t srvid
, uint32_t flags
, TDB_DATA data
,
77 TALLOC_CTX
*mem_ctx
, TDB_DATA
*outdata
,
81 * exit on fatal communications errors with the ctdbd daemon
83 static void cluster_fatal(const char *why
)
85 DEBUG(0,("cluster fatal event: %s - exiting immediately\n", why
));
86 /* we don't use smb_panic() as we don't want to delay to write
87 a core file. We need to release this process id immediately
88 so that someone else can take over without getting sharing
96 static void ctdb_packet_dump(struct ctdb_req_header
*hdr
)
98 if (DEBUGLEVEL
< 10) {
101 DEBUGADD(10, ("len=%d, magic=%x, vers=%d, gen=%d, op=%d, reqid=%d\n",
102 (int)hdr
->length
, (int)hdr
->ctdb_magic
,
103 (int)hdr
->ctdb_version
, (int)hdr
->generation
,
104 (int)hdr
->operation
, (int)hdr
->reqid
));
108 * Register a srvid with ctdbd
110 static NTSTATUS
register_with_ctdbd(struct ctdbd_connection
*conn
,
115 return ctdbd_control(conn
, CTDB_CURRENT_NODE
,
116 CTDB_CONTROL_REGISTER_SRVID
, srvid
, 0,
117 tdb_null
, NULL
, NULL
, &cstatus
);
121 * get our vnn from the cluster
123 static NTSTATUS
get_cluster_vnn(struct ctdbd_connection
*conn
, uint32
*vnn
)
127 status
= ctdbd_control(conn
,
128 CTDB_CURRENT_NODE
, CTDB_CONTROL_GET_PNN
, 0, 0,
129 tdb_null
, NULL
, NULL
, &cstatus
);
130 if (!NT_STATUS_IS_OK(status
)) {
131 cluster_fatal("ctdbd_control failed\n");
133 *vnn
= (uint32_t)cstatus
;
138 * Are we active (i.e. not banned or stopped?)
140 static bool ctdbd_working(struct ctdbd_connection
*conn
, uint32_t vnn
)
145 struct ctdb_node_map
*m
;
146 uint32_t failure_flags
;
150 status
= ctdbd_control(conn
, CTDB_CURRENT_NODE
,
151 CTDB_CONTROL_GET_NODEMAP
, 0, 0,
152 tdb_null
, talloc_tos(), &outdata
, &cstatus
);
153 if (!NT_STATUS_IS_OK(status
)) {
154 cluster_fatal("ctdbd_control failed\n");
156 if ((cstatus
!= 0) || (outdata
.dptr
== NULL
)) {
157 DEBUG(2, ("Received invalid ctdb data\n"));
161 m
= (struct ctdb_node_map
*)outdata
.dptr
;
163 for (i
=0; i
<m
->num
; i
++) {
164 if (vnn
== m
->nodes
[i
].pnn
) {
170 DEBUG(2, ("Did not find ourselves (node %d) in nodemap\n",
175 failure_flags
= NODE_FLAGS_BANNED
| NODE_FLAGS_DISCONNECTED
176 | NODE_FLAGS_PERMANENTLY_DISABLED
| NODE_FLAGS_STOPPED
;
178 if ((m
->nodes
[i
].flags
& failure_flags
) != 0) {
179 DEBUG(2, ("Node has status %x, not active\n",
180 (int)m
->nodes
[i
].flags
));
186 TALLOC_FREE(outdata
.dptr
);
190 uint32
ctdbd_vnn(const struct ctdbd_connection
*conn
)
192 return conn
->our_vnn
;
196 * Get us a ctdb connection
199 static NTSTATUS
ctdbd_connect(TALLOC_CTX
*mem_ctx
,
200 struct ctdb_packet_context
**presult
)
202 struct ctdb_packet_context
*result
;
203 const char *sockname
= lp_ctdbd_socket();
204 struct sockaddr_un addr
;
207 if (!sockname
|| !*sockname
) {
208 sockname
= CTDB_PATH
;
211 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
213 DEBUG(3, ("Could not create socket: %s\n", strerror(errno
)));
214 return map_nt_error_from_unix(errno
);
218 addr
.sun_family
= AF_UNIX
;
219 strncpy(addr
.sun_path
, sockname
, sizeof(addr
.sun_path
));
221 if (sys_connect(fd
, (struct sockaddr
*)(void *)&addr
) == -1) {
222 DEBUG(1, ("connect(%s) failed: %s\n", sockname
,
225 return map_nt_error_from_unix(errno
);
228 if (!(result
= ctdb_packet_init(mem_ctx
, fd
))) {
230 return NT_STATUS_NO_MEMORY
;
238 * Do we have a complete ctdb packet in the queue?
241 static bool ctdb_req_complete(const uint8_t *buf
, size_t available
,
247 if (available
< sizeof(msglen
)) {
251 msglen
= *((uint32
*)buf
);
253 DEBUG(10, ("msglen = %d\n", msglen
));
255 if (msglen
< sizeof(struct ctdb_req_header
)) {
256 DEBUG(0, ("Got invalid msglen: %d, expected at least %d for "
257 "the req_header\n", (int)msglen
,
258 (int)sizeof(struct ctdb_req_header
)));
259 cluster_fatal("ctdbd protocol error\n");
262 if (available
< msglen
) {
271 * State necessary to defer an incoming message while we are waiting for a
275 struct deferred_msg_state
{
276 struct messaging_context
*msg_ctx
;
277 struct messaging_rec
*rec
;
281 * Timed event handler for the deferred message
284 static void deferred_message_dispatch(struct event_context
*event_ctx
,
285 struct timed_event
*te
,
289 struct deferred_msg_state
*state
= talloc_get_type_abort(
290 private_data
, struct deferred_msg_state
);
292 messaging_dispatch_rec(state
->msg_ctx
, state
->rec
);
297 struct req_pull_state
{
303 * Pull a ctdb request out of the incoming ctdb_packet queue
306 static NTSTATUS
ctdb_req_pull(uint8_t *buf
, size_t length
,
309 struct req_pull_state
*state
= (struct req_pull_state
*)private_data
;
311 state
->req
.data
= talloc_move(state
->mem_ctx
, &buf
);
312 state
->req
.length
= length
;
317 * Fetch a messaging_rec from an incoming ctdb style message
320 static struct messaging_rec
*ctdb_pull_messaging_rec(TALLOC_CTX
*mem_ctx
,
321 size_t overall_length
,
322 struct ctdb_req_message
*msg
)
324 struct messaging_rec
*result
;
326 enum ndr_err_code ndr_err
;
328 if ((overall_length
< offsetof(struct ctdb_req_message
, data
))
330 < offsetof(struct ctdb_req_message
, data
) + msg
->datalen
)) {
332 cluster_fatal("got invalid msg length");
335 if (!(result
= talloc(mem_ctx
, struct messaging_rec
))) {
336 DEBUG(0, ("talloc failed\n"));
340 blob
= data_blob_const(msg
->data
, msg
->datalen
);
342 ndr_err
= ndr_pull_struct_blob(
343 &blob
, result
, result
,
344 (ndr_pull_flags_fn_t
)ndr_pull_messaging_rec
);
346 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err
)) {
347 DEBUG(0, ("ndr_pull_struct_blob failed: %s\n",
348 ndr_errstr(ndr_err
)));
353 if (DEBUGLEVEL
>= 10) {
354 DEBUG(10, ("ctdb_pull_messaging_rec:\n"));
355 NDR_PRINT_DEBUG(messaging_rec
, result
);
361 static NTSTATUS
ctdb_packet_fd_read_sync(struct ctdb_packet_context
*ctx
)
363 int timeout
= lp_ctdb_timeout();
368 return ctdb_packet_fd_read_sync_timeout(ctx
, timeout
);
372 * Read a full ctdbd request. If we have a messaging context, defer incoming
373 * messages that might come in between.
376 static NTSTATUS
ctdb_read_req(struct ctdbd_connection
*conn
, uint32 reqid
,
377 TALLOC_CTX
*mem_ctx
, void *result
)
379 struct ctdb_req_header
*hdr
;
380 struct req_pull_state state
;
385 state
.mem_ctx
= mem_ctx
;
387 while (!ctdb_packet_handler(conn
->pkt
, ctdb_req_complete
,
388 ctdb_req_pull
, &state
, &status
)) {
392 status
= ctdb_packet_fd_read_sync(conn
->pkt
);
394 if (NT_STATUS_EQUAL(status
, NT_STATUS_NETWORK_BUSY
)) {
397 } else if (NT_STATUS_EQUAL(status
, NT_STATUS_RETRY
)) {
402 if (!NT_STATUS_IS_OK(status
)) {
403 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status
)));
404 cluster_fatal("ctdbd died\n");
408 if (!NT_STATUS_IS_OK(status
)) {
409 DEBUG(0, ("Could not read ctdb_packet: %s\n", nt_errstr(status
)));
410 cluster_fatal("ctdbd died\n");
413 hdr
= (struct ctdb_req_header
*)state
.req
.data
;
415 DEBUG(10, ("Received ctdb packet\n"));
416 ctdb_packet_dump(hdr
);
418 if (hdr
->operation
== CTDB_REQ_MESSAGE
) {
419 struct timed_event
*evt
;
420 struct deferred_msg_state
*msg_state
;
421 struct ctdb_req_message
*msg
= (struct ctdb_req_message
*)hdr
;
423 if (conn
->msg_ctx
== NULL
) {
424 DEBUG(1, ("Got a message without having a msg ctx, "
425 "dropping msg %llu\n",
426 (long long unsigned)msg
->srvid
));
430 if ((conn
->release_ip_handler
!= NULL
)
431 && (msg
->srvid
== CTDB_SRVID_RELEASE_IP
)) {
432 /* must be dispatched immediately */
433 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
434 conn
->release_ip_handler((const char *)msg
->data
,
435 conn
->release_ip_priv
);
440 if ((msg
->srvid
== CTDB_SRVID_RECONFIGURE
)
441 || (msg
->srvid
== CTDB_SRVID_SAMBA_NOTIFY
)) {
443 DEBUG(1, ("ctdb_read_req: Got %s message\n",
444 (msg
->srvid
== CTDB_SRVID_RECONFIGURE
)
445 ? "cluster reconfigure" : "SAMBA_NOTIFY"));
447 messaging_send(conn
->msg_ctx
,
448 messaging_server_id(conn
->msg_ctx
),
449 MSG_SMB_BRL_VALIDATE
, &data_blob_null
);
450 messaging_send(conn
->msg_ctx
,
451 messaging_server_id(conn
->msg_ctx
),
452 MSG_DBWRAP_G_LOCK_RETRY
,
458 msg_state
= talloc(NULL
, struct deferred_msg_state
);
459 if (msg_state
== NULL
) {
460 DEBUG(0, ("talloc failed\n"));
465 if (!(msg_state
->rec
= ctdb_pull_messaging_rec(
466 msg_state
, state
.req
.length
, msg
))) {
467 DEBUG(0, ("ctdbd_pull_messaging_rec failed\n"));
468 TALLOC_FREE(msg_state
);
475 msg_state
->msg_ctx
= conn
->msg_ctx
;
478 * We're waiting for a call reply, but an async message has
479 * crossed. Defer dispatching to the toplevel event loop.
481 evt
= event_add_timed(conn
->msg_ctx
->event_ctx
,
482 conn
->msg_ctx
->event_ctx
,
484 deferred_message_dispatch
,
487 DEBUG(0, ("event_add_timed failed\n"));
488 TALLOC_FREE(msg_state
);
496 if ((reqid
!= 0) && (hdr
->reqid
!= reqid
)) {
497 /* we got the wrong reply */
498 DEBUG(0,("Discarding mismatched ctdb reqid %u should have "
499 "been %u\n", hdr
->reqid
, reqid
));
504 *((void **)result
) = talloc_move(mem_ctx
, &hdr
);
510 * Get us a ctdbd connection
513 static NTSTATUS
ctdbd_init_connection(TALLOC_CTX
*mem_ctx
,
514 struct ctdbd_connection
**pconn
)
516 struct ctdbd_connection
*conn
;
519 if (!(conn
= talloc_zero(mem_ctx
, struct ctdbd_connection
))) {
520 DEBUG(0, ("talloc failed\n"));
521 return NT_STATUS_NO_MEMORY
;
524 status
= ctdbd_connect(conn
, &conn
->pkt
);
526 if (!NT_STATUS_IS_OK(status
)) {
527 DEBUG(10, ("ctdbd_connect failed: %s\n", nt_errstr(status
)));
531 status
= get_cluster_vnn(conn
, &conn
->our_vnn
);
533 if (!NT_STATUS_IS_OK(status
)) {
534 DEBUG(10, ("get_cluster_vnn failed: %s\n", nt_errstr(status
)));
538 if (!ctdbd_working(conn
, conn
->our_vnn
)) {
539 DEBUG(2, ("Node is not working, can not connect\n"));
540 status
= NT_STATUS_INTERNAL_DB_ERROR
;
544 generate_random_buffer((unsigned char *)&conn
->rand_srvid
,
545 sizeof(conn
->rand_srvid
));
547 status
= register_with_ctdbd(conn
, conn
->rand_srvid
);
549 if (!NT_STATUS_IS_OK(status
)) {
550 DEBUG(5, ("Could not register random srvid: %s\n",
564 * Get us a ctdbd connection and register us as a process
567 NTSTATUS
ctdbd_messaging_connection(TALLOC_CTX
*mem_ctx
,
568 struct ctdbd_connection
**pconn
)
570 struct ctdbd_connection
*conn
;
573 status
= ctdbd_init_connection(mem_ctx
, &conn
);
575 if (!NT_STATUS_IS_OK(status
)) {
579 status
= register_with_ctdbd(conn
, (uint64_t)sys_getpid());
580 if (!NT_STATUS_IS_OK(status
)) {
584 status
= register_with_ctdbd(conn
, MSG_SRVID_SAMBA
);
585 if (!NT_STATUS_IS_OK(status
)) {
589 status
= register_with_ctdbd(conn
, CTDB_SRVID_SAMBA_NOTIFY
);
590 if (!NT_STATUS_IS_OK(status
)) {
602 struct messaging_context
*ctdb_conn_msg_ctx(struct ctdbd_connection
*conn
)
604 return conn
->msg_ctx
;
607 int ctdbd_conn_get_fd(struct ctdbd_connection
*conn
)
609 return ctdb_packet_get_fd(conn
->pkt
);
613 * Packet handler to receive and handle a ctdb message
615 static NTSTATUS
ctdb_handle_message(uint8_t *buf
, size_t length
,
618 struct ctdbd_connection
*conn
= talloc_get_type_abort(
619 private_data
, struct ctdbd_connection
);
620 struct ctdb_req_message
*msg
;
621 struct messaging_rec
*msg_rec
;
623 msg
= (struct ctdb_req_message
*)buf
;
625 if (msg
->hdr
.operation
!= CTDB_REQ_MESSAGE
) {
626 DEBUG(0, ("Received async msg of type %u, discarding\n",
627 msg
->hdr
.operation
));
629 return NT_STATUS_INVALID_PARAMETER
;
632 if ((conn
->release_ip_handler
!= NULL
)
633 && (msg
->srvid
== CTDB_SRVID_RELEASE_IP
)) {
634 /* must be dispatched immediately */
635 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
636 conn
->release_ip_handler((const char *)msg
->data
,
637 conn
->release_ip_priv
);
642 SMB_ASSERT(conn
->msg_ctx
!= NULL
);
644 if ((msg
->srvid
== CTDB_SRVID_RECONFIGURE
)
645 || (msg
->srvid
== CTDB_SRVID_SAMBA_NOTIFY
)){
646 DEBUG(0,("Got cluster reconfigure message\n"));
648 * when the cluster is reconfigured or someone of the
649 * family has passed away (SAMBA_NOTIFY), we need to
650 * clean the brl database
652 messaging_send(conn
->msg_ctx
,
653 messaging_server_id(conn
->msg_ctx
),
654 MSG_SMB_BRL_VALIDATE
, &data_blob_null
);
656 messaging_send(conn
->msg_ctx
,
657 messaging_server_id(conn
->msg_ctx
),
658 MSG_DBWRAP_G_LOCK_RETRY
,
665 /* only messages to our pid or the broadcast are valid here */
666 if (msg
->srvid
!= sys_getpid() && msg
->srvid
!= MSG_SRVID_SAMBA
) {
667 DEBUG(0,("Got unexpected message with srvid=%llu\n",
668 (unsigned long long)msg
->srvid
));
673 if (!(msg_rec
= ctdb_pull_messaging_rec(NULL
, length
, msg
))) {
674 DEBUG(10, ("ctdb_pull_messaging_rec failed\n"));
676 return NT_STATUS_NO_MEMORY
;
679 messaging_dispatch_rec(conn
->msg_ctx
, msg_rec
);
681 TALLOC_FREE(msg_rec
);
687 * The ctdbd socket is readable asynchronuously
690 static void ctdbd_socket_handler(struct event_context
*event_ctx
,
691 struct fd_event
*event
,
695 struct ctdbd_connection
*conn
= talloc_get_type_abort(
696 private_data
, struct ctdbd_connection
);
700 status
= ctdb_packet_fd_read(conn
->pkt
);
702 if (!NT_STATUS_IS_OK(status
)) {
703 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status
)));
704 cluster_fatal("ctdbd died\n");
707 while (ctdb_packet_handler(conn
->pkt
, ctdb_req_complete
,
708 ctdb_handle_message
, conn
, &status
)) {
709 if (!NT_STATUS_IS_OK(status
)) {
710 DEBUG(10, ("could not handle incoming message: %s\n",
717 * Prepare a ctdbd connection to receive messages
720 NTSTATUS
ctdbd_register_msg_ctx(struct ctdbd_connection
*conn
,
721 struct messaging_context
*msg_ctx
)
723 SMB_ASSERT(conn
->msg_ctx
== NULL
);
724 SMB_ASSERT(conn
->fde
== NULL
);
726 if (!(conn
->fde
= event_add_fd(msg_ctx
->event_ctx
, conn
,
727 ctdb_packet_get_fd(conn
->pkt
),
729 ctdbd_socket_handler
,
731 DEBUG(0, ("event_add_fd failed\n"));
732 return NT_STATUS_NO_MEMORY
;
735 conn
->msg_ctx
= msg_ctx
;
741 * Send a messaging message across a ctdbd
744 NTSTATUS
ctdbd_messaging_send(struct ctdbd_connection
*conn
,
745 uint32 dst_vnn
, uint64 dst_srvid
,
746 struct messaging_rec
*msg
)
748 struct ctdb_req_message r
;
752 enum ndr_err_code ndr_err
;
754 if (!(mem_ctx
= talloc_init("ctdbd_messaging_send"))) {
755 DEBUG(0, ("talloc failed\n"));
756 return NT_STATUS_NO_MEMORY
;
759 ndr_err
= ndr_push_struct_blob(
761 (ndr_push_flags_fn_t
)ndr_push_messaging_rec
);
763 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err
)) {
764 DEBUG(0, ("ndr_push_struct_blob failed: %s\n",
765 ndr_errstr(ndr_err
)));
766 status
= ndr_map_error2ntstatus(ndr_err
);
770 r
.hdr
.length
= offsetof(struct ctdb_req_message
, data
) + blob
.length
;
771 r
.hdr
.ctdb_magic
= CTDB_MAGIC
;
772 r
.hdr
.ctdb_version
= CTDB_VERSION
;
773 r
.hdr
.generation
= 1;
774 r
.hdr
.operation
= CTDB_REQ_MESSAGE
;
775 r
.hdr
.destnode
= dst_vnn
;
776 r
.hdr
.srcnode
= conn
->our_vnn
;
779 r
.datalen
= blob
.length
;
781 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
782 ctdb_packet_dump(&r
.hdr
);
784 status
= ctdb_packet_send(
786 data_blob_const(&r
, offsetof(struct ctdb_req_message
, data
)),
789 if (!NT_STATUS_IS_OK(status
)) {
790 DEBUG(0, ("ctdb_packet_send failed: %s\n", nt_errstr(status
)));
794 status
= ctdb_packet_flush(conn
->pkt
);
796 if (!NT_STATUS_IS_OK(status
)) {
797 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status
)));
798 cluster_fatal("cluster dispatch daemon msg write error\n");
801 status
= NT_STATUS_OK
;
803 TALLOC_FREE(mem_ctx
);
808 * send/recv a generic ctdb control message
810 static NTSTATUS
ctdbd_control(struct ctdbd_connection
*conn
,
811 uint32_t vnn
, uint32 opcode
,
812 uint64_t srvid
, uint32_t flags
,
814 TALLOC_CTX
*mem_ctx
, TDB_DATA
*outdata
,
817 struct ctdb_req_control req
;
818 struct ctdb_reply_control
*reply
= NULL
;
819 struct ctdbd_connection
*new_conn
= NULL
;
823 status
= ctdbd_init_connection(NULL
, &new_conn
);
825 if (!NT_STATUS_IS_OK(status
)) {
826 DEBUG(10, ("Could not init temp connection: %s\n",
835 req
.hdr
.length
= offsetof(struct ctdb_req_control
, data
) + data
.dsize
;
836 req
.hdr
.ctdb_magic
= CTDB_MAGIC
;
837 req
.hdr
.ctdb_version
= CTDB_VERSION
;
838 req
.hdr
.operation
= CTDB_REQ_CONTROL
;
839 req
.hdr
.reqid
= ctdbd_next_reqid(conn
);
840 req
.hdr
.destnode
= vnn
;
843 req
.datalen
= data
.dsize
;
846 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
847 ctdb_packet_dump(&req
.hdr
);
849 status
= ctdb_packet_send(
851 data_blob_const(&req
, offsetof(struct ctdb_req_control
, data
)),
852 data_blob_const(data
.dptr
, data
.dsize
));
854 if (!NT_STATUS_IS_OK(status
)) {
855 DEBUG(3, ("ctdb_packet_send failed: %s\n", nt_errstr(status
)));
859 status
= ctdb_packet_flush(conn
->pkt
);
861 if (!NT_STATUS_IS_OK(status
)) {
862 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status
)));
863 cluster_fatal("cluster dispatch daemon control write error\n");
866 if (flags
& CTDB_CTRL_FLAG_NOREPLY
) {
867 TALLOC_FREE(new_conn
);
874 status
= ctdb_read_req(conn
, req
.hdr
.reqid
, NULL
, (void *)&reply
);
876 if (!NT_STATUS_IS_OK(status
)) {
877 DEBUG(10, ("ctdb_read_req failed: %s\n", nt_errstr(status
)));
881 if (reply
->hdr
.operation
!= CTDB_REPLY_CONTROL
) {
882 DEBUG(0, ("received invalid reply\n"));
887 if (!(outdata
->dptr
= (uint8
*)talloc_memdup(
888 mem_ctx
, reply
->data
, reply
->datalen
))) {
890 return NT_STATUS_NO_MEMORY
;
892 outdata
->dsize
= reply
->datalen
;
895 (*cstatus
) = reply
->status
;
898 status
= NT_STATUS_OK
;
901 TALLOC_FREE(new_conn
);
907 * see if a remote process exists
909 bool ctdbd_process_exists(struct ctdbd_connection
*conn
, uint32 vnn
, pid_t pid
)
917 if (!ctdb_processes_exist(conn
, &id
, 1, &result
)) {
918 DEBUG(10, ("ctdb_processes_exist failed\n"));
924 bool ctdb_processes_exist(struct ctdbd_connection
*conn
,
925 const struct server_id
*pids
, int num_pids
,
928 TALLOC_CTX
*frame
= talloc_stackframe();
934 reqids
= talloc_array(talloc_tos(), uint32_t, num_pids
);
935 if (reqids
== NULL
) {
939 for (i
=0; i
<num_pids
; i
++) {
940 struct ctdb_req_control req
;
944 reqids
[i
] = ctdbd_next_reqid(conn
);
949 * pids[i].pid is uint64_t, scale down to pid_t which
950 * is the wire protocol towards ctdb.
954 DEBUG(10, ("Requesting PID %d/%d, reqid=%d\n",
955 (int)pids
[i
].vnn
, (int)pid
,
958 req
.hdr
.length
= offsetof(struct ctdb_req_control
, data
);
959 req
.hdr
.length
+= sizeof(pid
);
960 req
.hdr
.ctdb_magic
= CTDB_MAGIC
;
961 req
.hdr
.ctdb_version
= CTDB_VERSION
;
962 req
.hdr
.operation
= CTDB_REQ_CONTROL
;
963 req
.hdr
.reqid
= reqids
[i
];
964 req
.hdr
.destnode
= pids
[i
].vnn
;
965 req
.opcode
= CTDB_CONTROL_PROCESS_EXISTS
;
967 req
.datalen
= sizeof(pid
);
970 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
971 ctdb_packet_dump(&req
.hdr
);
973 status
= ctdb_packet_send(
976 &req
, offsetof(struct ctdb_req_control
, data
)),
977 data_blob_const(&pid
, sizeof(pid
)));
978 if (!NT_STATUS_IS_OK(status
)) {
979 DEBUG(10, ("ctdb_packet_send failed: %s\n",
985 status
= ctdb_packet_flush(conn
->pkt
);
986 if (!NT_STATUS_IS_OK(status
)) {
987 DEBUG(10, ("ctdb_packet_flush failed: %s\n",
994 while (num_received
< num_pids
) {
995 struct ctdb_reply_control
*reply
= NULL
;
998 status
= ctdb_read_req(conn
, 0, talloc_tos(), (void *)&reply
);
999 if (!NT_STATUS_IS_OK(status
)) {
1000 DEBUG(10, ("ctdb_read_req failed: %s\n",
1001 nt_errstr(status
)));
1005 if (reply
->hdr
.operation
!= CTDB_REPLY_CONTROL
) {
1006 DEBUG(10, ("Received invalid reply\n"));
1010 reqid
= reply
->hdr
.reqid
;
1012 DEBUG(10, ("Received reqid %d\n", (int)reqid
));
1014 for (i
=0; i
<num_pids
; i
++) {
1015 if (reqid
== reqids
[i
]) {
1019 if (i
== num_pids
) {
1020 DEBUG(10, ("Received unknown record number %u\n",
1024 results
[i
] = ((reply
->status
) == 0);
1038 char *ctdbd_dbpath(struct ctdbd_connection
*conn
,
1039 TALLOC_CTX
*mem_ctx
, uint32_t db_id
)
1045 data
.dptr
= (uint8_t*)&db_id
;
1046 data
.dsize
= sizeof(db_id
);
1048 status
= ctdbd_control(conn
, CTDB_CURRENT_NODE
,
1049 CTDB_CONTROL_GETDBPATH
, 0, 0, data
,
1050 mem_ctx
, &data
, &cstatus
);
1051 if (!NT_STATUS_IS_OK(status
) || cstatus
!= 0) {
1052 DEBUG(0,(__location__
" ctdb_control for getdbpath failed\n"));
1056 return (char *)data
.dptr
;
1060 * attach to a ctdb database
1062 NTSTATUS
ctdbd_db_attach(struct ctdbd_connection
*conn
,
1063 const char *name
, uint32_t *db_id
, int tdb_flags
)
1068 bool persistent
= (tdb_flags
& TDB_CLEAR_IF_FIRST
) == 0;
1070 data
.dptr
= (uint8_t*)name
;
1071 data
.dsize
= strlen(name
)+1;
1073 status
= ctdbd_control(conn
, CTDB_CURRENT_NODE
,
1075 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
1076 : CTDB_CONTROL_DB_ATTACH
,
1077 tdb_flags
, 0, data
, NULL
, &data
, &cstatus
);
1078 if (!NT_STATUS_IS_OK(status
)) {
1079 DEBUG(0, (__location__
" ctdb_control for db_attach "
1080 "failed: %s\n", nt_errstr(status
)));
1084 if (cstatus
!= 0 || data
.dsize
!= sizeof(uint32_t)) {
1085 DEBUG(0,(__location__
" ctdb_control for db_attach failed\n"));
1086 return NT_STATUS_INTERNAL_ERROR
;
1089 *db_id
= *(uint32_t *)data
.dptr
;
1090 talloc_free(data
.dptr
);
1092 if (!(tdb_flags
& TDB_SEQNUM
)) {
1093 return NT_STATUS_OK
;
1096 data
.dptr
= (uint8_t *)db_id
;
1097 data
.dsize
= sizeof(*db_id
);
1099 status
= ctdbd_control(conn
, CTDB_CURRENT_NODE
,
1100 CTDB_CONTROL_ENABLE_SEQNUM
, 0, 0, data
,
1101 NULL
, NULL
, &cstatus
);
1102 if (!NT_STATUS_IS_OK(status
) || cstatus
!= 0) {
1103 DEBUG(0,(__location__
" ctdb_control for enable seqnum "
1105 return NT_STATUS_IS_OK(status
) ? NT_STATUS_INTERNAL_ERROR
:
1109 return NT_STATUS_OK
;
1113 * force the migration of a record to this node
1115 NTSTATUS
ctdbd_migrate(struct ctdbd_connection
*conn
, uint32 db_id
,
1118 struct ctdb_req_call req
;
1119 struct ctdb_reply_call
*reply
;
1124 req
.hdr
.length
= offsetof(struct ctdb_req_call
, data
) + key
.dsize
;
1125 req
.hdr
.ctdb_magic
= CTDB_MAGIC
;
1126 req
.hdr
.ctdb_version
= CTDB_VERSION
;
1127 req
.hdr
.operation
= CTDB_REQ_CALL
;
1128 req
.hdr
.reqid
= ctdbd_next_reqid(conn
);
1129 req
.flags
= CTDB_IMMEDIATE_MIGRATION
;
1130 req
.callid
= CTDB_NULL_FUNC
;
1132 req
.keylen
= key
.dsize
;
1134 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
1135 ctdb_packet_dump(&req
.hdr
);
1137 status
= ctdb_packet_send(
1139 data_blob_const(&req
, offsetof(struct ctdb_req_call
, data
)),
1140 data_blob_const(key
.dptr
, key
.dsize
));
1142 if (!NT_STATUS_IS_OK(status
)) {
1143 DEBUG(3, ("ctdb_packet_send failed: %s\n", nt_errstr(status
)));
1147 status
= ctdb_packet_flush(conn
->pkt
);
1149 if (!NT_STATUS_IS_OK(status
)) {
1150 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status
)));
1151 cluster_fatal("cluster dispatch daemon control write error\n");
1154 status
= ctdb_read_req(conn
, req
.hdr
.reqid
, NULL
, (void *)&reply
);
1156 if (!NT_STATUS_IS_OK(status
)) {
1157 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status
)));
1161 if (reply
->hdr
.operation
!= CTDB_REPLY_CALL
) {
1162 DEBUG(0, ("received invalid reply\n"));
1163 status
= NT_STATUS_INTERNAL_ERROR
;
1167 status
= NT_STATUS_OK
;
1175 * remotely fetch a record without locking it or forcing a migration
1177 NTSTATUS
ctdbd_fetch(struct ctdbd_connection
*conn
, uint32 db_id
,
1178 TDB_DATA key
, TALLOC_CTX
*mem_ctx
, TDB_DATA
*data
)
1180 struct ctdb_req_call req
;
1181 struct ctdb_reply_call
*reply
;
1186 req
.hdr
.length
= offsetof(struct ctdb_req_call
, data
) + key
.dsize
;
1187 req
.hdr
.ctdb_magic
= CTDB_MAGIC
;
1188 req
.hdr
.ctdb_version
= CTDB_VERSION
;
1189 req
.hdr
.operation
= CTDB_REQ_CALL
;
1190 req
.hdr
.reqid
= ctdbd_next_reqid(conn
);
1192 req
.callid
= CTDB_FETCH_FUNC
;
1194 req
.keylen
= key
.dsize
;
1196 status
= ctdb_packet_send(
1198 data_blob_const(&req
, offsetof(struct ctdb_req_call
, data
)),
1199 data_blob_const(key
.dptr
, key
.dsize
));
1201 if (!NT_STATUS_IS_OK(status
)) {
1202 DEBUG(3, ("ctdb_packet_send failed: %s\n", nt_errstr(status
)));
1206 status
= ctdb_packet_flush(conn
->pkt
);
1208 if (!NT_STATUS_IS_OK(status
)) {
1209 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status
)));
1210 cluster_fatal("cluster dispatch daemon control write error\n");
1213 status
= ctdb_read_req(conn
, req
.hdr
.reqid
, NULL
, (void *)&reply
);
1215 if (!NT_STATUS_IS_OK(status
)) {
1216 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status
)));
1220 if (reply
->hdr
.operation
!= CTDB_REPLY_CALL
) {
1221 DEBUG(0, ("received invalid reply\n"));
1222 status
= NT_STATUS_INTERNAL_ERROR
;
1226 data
->dsize
= reply
->datalen
;
1227 if (data
->dsize
== 0) {
1232 data
->dptr
= (uint8
*)talloc_memdup(mem_ctx
, &reply
->data
[0],
1234 if (data
->dptr
== NULL
) {
1235 DEBUG(0, ("talloc failed\n"));
1236 status
= NT_STATUS_NO_MEMORY
;
1241 status
= NT_STATUS_OK
;
1247 struct ctdbd_traverse_state
{
1248 void (*fn
)(TDB_DATA key
, TDB_DATA data
, void *private_data
);
1253 * Handle a traverse record coming in on the ctdbd connection
1256 static NTSTATUS
ctdb_traverse_handler(uint8_t *buf
, size_t length
,
1259 struct ctdbd_traverse_state
*state
=
1260 (struct ctdbd_traverse_state
*)private_data
;
1262 struct ctdb_req_message
*m
;
1263 struct ctdb_rec_data
*d
;
1266 m
= (struct ctdb_req_message
*)buf
;
1268 if (length
< sizeof(*m
) || m
->hdr
.length
!= length
) {
1269 DEBUG(0, ("Got invalid message of length %d\n", (int)length
));
1271 return NT_STATUS_UNEXPECTED_IO_ERROR
;
1274 d
= (struct ctdb_rec_data
*)&m
->data
[0];
1275 if (m
->datalen
< sizeof(uint32_t) || m
->datalen
!= d
->length
) {
1276 DEBUG(0, ("Got invalid traverse data of length %d\n",
1279 return NT_STATUS_UNEXPECTED_IO_ERROR
;
1282 key
.dsize
= d
->keylen
;
1283 key
.dptr
= &d
->data
[0];
1284 data
.dsize
= d
->datalen
;
1285 data
.dptr
= &d
->data
[d
->keylen
];
1287 if (key
.dsize
== 0 && data
.dsize
== 0) {
1288 /* end of traverse */
1289 return NT_STATUS_END_OF_FILE
;
1292 if (data
.dsize
< sizeof(struct ctdb_ltdb_header
)) {
1293 DEBUG(0, ("Got invalid ltdb header length %d\n",
1296 return NT_STATUS_UNEXPECTED_IO_ERROR
;
1298 data
.dsize
-= sizeof(struct ctdb_ltdb_header
);
1299 data
.dptr
+= sizeof(struct ctdb_ltdb_header
);
1302 state
->fn(key
, data
, state
->private_data
);
1306 return NT_STATUS_OK
;
1310 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1311 connection to ctdbd to avoid the hairy recursive and async problems with
1315 NTSTATUS
ctdbd_traverse(uint32 db_id
,
1316 void (*fn
)(TDB_DATA key
, TDB_DATA data
,
1317 void *private_data
),
1320 struct ctdbd_connection
*conn
;
1324 struct ctdb_traverse_start t
;
1326 struct ctdbd_traverse_state state
;
1328 status
= ctdbd_init_connection(NULL
, &conn
);
1329 if (!NT_STATUS_IS_OK(status
)) {
1330 DEBUG(0, ("ctdbd_init_connection failed: %s\n",
1331 nt_errstr(status
)));
1336 t
.srvid
= conn
->rand_srvid
;
1337 t
.reqid
= ctdbd_next_reqid(conn
);
1339 data
.dptr
= (uint8_t *)&t
;
1340 data
.dsize
= sizeof(t
);
1342 status
= ctdbd_control(conn
, CTDB_CURRENT_NODE
,
1343 CTDB_CONTROL_TRAVERSE_START
, conn
->rand_srvid
, 0,
1344 data
, NULL
, NULL
, &cstatus
);
1346 if (!NT_STATUS_IS_OK(status
) || (cstatus
!= 0)) {
1348 DEBUG(0,("ctdbd_control failed: %s, %d\n", nt_errstr(status
),
1351 if (NT_STATUS_IS_OK(status
)) {
1353 * We need a mapping here
1355 status
= NT_STATUS_UNSUCCESSFUL
;
1361 state
.private_data
= private_data
;
1365 status
= NT_STATUS_OK
;
1367 if (ctdb_packet_handler(conn
->pkt
, ctdb_req_complete
,
1368 ctdb_traverse_handler
, &state
, &status
)) {
1370 if (NT_STATUS_EQUAL(status
, NT_STATUS_END_OF_FILE
)) {
1371 status
= NT_STATUS_OK
;
1376 * There might be more in the queue
1381 if (!NT_STATUS_IS_OK(status
)) {
1385 status
= ctdb_packet_fd_read_sync(conn
->pkt
);
1387 if (NT_STATUS_EQUAL(status
, NT_STATUS_RETRY
)) {
1389 * There might be more in the queue
1394 if (NT_STATUS_EQUAL(status
, NT_STATUS_END_OF_FILE
)) {
1395 status
= NT_STATUS_OK
;
1399 if (!NT_STATUS_IS_OK(status
)) {
1400 DEBUG(0, ("ctdb_packet_fd_read_sync failed: %s\n", nt_errstr(status
)));
1401 cluster_fatal("ctdbd died\n");
1411 This is used to canonicalize a ctdb_sock_addr structure.
1413 static void smbd_ctdb_canonicalize_ip(const struct sockaddr_storage
*in
,
1414 struct sockaddr_storage
*out
)
1416 memcpy(out
, in
, sizeof (*out
));
1419 if (in
->ss_family
== AF_INET6
) {
1420 const char prefix
[12] = { 0,0,0,0,0,0,0,0,0,0,0xff,0xff };
1421 const struct sockaddr_in6
*in6
= (struct sockaddr_in6
*)in
;
1422 struct sockaddr_in
*out4
= (struct sockaddr_in
*)out
;
1423 if (memcmp(&in6
->sin6_addr
, prefix
, 12) == 0) {
1424 memset(out
, 0, sizeof(*out
));
1425 #ifdef HAVE_SOCK_SIN_LEN
1426 out4
->sin_len
= sizeof(*out
);
1428 out4
->sin_family
= AF_INET
;
1429 out4
->sin_port
= in6
->sin6_port
;
1430 memcpy(&out4
->sin_addr
, &in6
->sin6_addr
.s6_addr32
[3], 4);
1437 * Register us as a server for a particular tcp connection
1440 NTSTATUS
ctdbd_register_ips(struct ctdbd_connection
*conn
,
1441 const struct sockaddr_storage
*_server
,
1442 const struct sockaddr_storage
*_client
,
1443 void (*release_ip_handler
)(const char *ip_addr
,
1444 void *private_data
),
1448 * we still use ctdb_control_tcp for ipv4
1449 * because we want to work against older ctdb
1450 * versions at runtime
1452 struct ctdb_control_tcp p4
;
1453 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1454 struct ctdb_control_tcp_addr p
;
1458 struct sockaddr_storage client
;
1459 struct sockaddr_storage server
;
1462 * Only one connection so far
1464 SMB_ASSERT(conn
->release_ip_handler
== NULL
);
1466 smbd_ctdb_canonicalize_ip(_client
, &client
);
1467 smbd_ctdb_canonicalize_ip(_server
, &server
);
1469 switch (client
.ss_family
) {
1471 memcpy(&p4
.dest
, &server
, sizeof(p4
.dest
));
1472 memcpy(&p4
.src
, &client
, sizeof(p4
.src
));
1473 data
.dptr
= (uint8_t *)&p4
;
1474 data
.dsize
= sizeof(p4
);
1476 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1478 memcpy(&p
.dest
.ip6
, &server
, sizeof(p
.dest
.ip6
));
1479 memcpy(&p
.src
.ip6
, &client
, sizeof(p
.src
.ip6
));
1480 data
.dptr
= (uint8_t *)&p
;
1481 data
.dsize
= sizeof(p
);
1485 return NT_STATUS_INTERNAL_ERROR
;
1488 conn
->release_ip_handler
= release_ip_handler
;
1489 conn
->release_ip_priv
= private_data
;
1492 * We want to be told about IP releases
1495 status
= register_with_ctdbd(conn
, CTDB_SRVID_RELEASE_IP
);
1496 if (!NT_STATUS_IS_OK(status
)) {
1501 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1502 * can send an extra ack to trigger a reset for our client, so it
1503 * immediately reconnects
1505 return ctdbd_control(conn
, CTDB_CURRENT_NODE
,
1506 CTDB_CONTROL_TCP_CLIENT
, 0,
1507 CTDB_CTRL_FLAG_NOREPLY
, data
, NULL
, NULL
, NULL
);
1511 * We want to handle reconfigure events
1513 NTSTATUS
ctdbd_register_reconfigure(struct ctdbd_connection
*conn
)
1515 return register_with_ctdbd(conn
, CTDB_SRVID_RECONFIGURE
);
1519 call a control on the local node
1521 NTSTATUS
ctdbd_control_local(struct ctdbd_connection
*conn
, uint32 opcode
,
1522 uint64_t srvid
, uint32_t flags
, TDB_DATA data
,
1523 TALLOC_CTX
*mem_ctx
, TDB_DATA
*outdata
,
1526 return ctdbd_control(conn
, CTDB_CURRENT_NODE
, opcode
, srvid
, flags
, data
, mem_ctx
, outdata
, cstatus
);
1529 NTSTATUS
ctdb_watch_us(struct ctdbd_connection
*conn
)
1531 struct ctdb_client_notify_register reg_data
;
1536 reg_data
.srvid
= CTDB_SRVID_SAMBA_NOTIFY
;
1538 reg_data
.notify_data
[0] = 0;
1540 struct_len
= offsetof(struct ctdb_client_notify_register
,
1541 notify_data
) + reg_data
.len
;
1543 status
= ctdbd_control_local(
1544 conn
, CTDB_CONTROL_REGISTER_NOTIFY
, conn
->rand_srvid
, 0,
1545 make_tdb_data((uint8_t *)®_data
, struct_len
),
1546 NULL
, NULL
, &cstatus
);
1547 if (!NT_STATUS_IS_OK(status
)) {
1548 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1549 nt_errstr(status
)));
1554 NTSTATUS
ctdb_unwatch(struct ctdbd_connection
*conn
)
1556 struct ctdb_client_notify_deregister dereg_data
;
1560 dereg_data
.srvid
= CTDB_SRVID_SAMBA_NOTIFY
;
1562 status
= ctdbd_control_local(
1563 conn
, CTDB_CONTROL_DEREGISTER_NOTIFY
, conn
->rand_srvid
, 0,
1564 make_tdb_data((uint8_t *)&dereg_data
, sizeof(dereg_data
)),
1565 NULL
, NULL
, &cstatus
);
1566 if (!NT_STATUS_IS_OK(status
)) {
1567 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1568 nt_errstr(status
)));