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 #include "ctdbd_conn.h"
25 #include "system/select.h"
26 #include "lib/util/sys_rw_data.h"
27 #include "lib/util/iov_buf.h"
31 /* paths to these include files come from --with-ctdb= in configure */
33 #include "ctdb_private.h"
35 struct ctdbd_srvid_cb
{
37 int (*cb
)(uint32_t src_vnn
, uint32_t dst_vnn
,
39 const uint8_t *msg
, size_t msglen
,
44 struct ctdbd_connection
{
45 const char *sockname
; /* Needed in ctdbd_traverse */
46 struct messaging_context
*msg_ctx
;
50 struct ctdbd_srvid_cb
*callbacks
;
52 struct tevent_fd
*fde
;
56 static uint32_t ctdbd_next_reqid(struct ctdbd_connection
*conn
)
59 if (conn
->reqid
== 0) {
65 static int ctdbd_control(struct ctdbd_connection
*conn
,
66 uint32_t vnn
, uint32_t opcode
,
67 uint64_t srvid
, uint32_t flags
,
69 TALLOC_CTX
*mem_ctx
, TDB_DATA
*outdata
,
73 * exit on fatal communications errors with the ctdbd daemon
75 static void cluster_fatal(const char *why
)
77 DEBUG(0,("cluster fatal event: %s - exiting immediately\n", why
));
78 /* we don't use smb_panic() as we don't want to delay to write
79 a core file. We need to release this process id immediately
80 so that someone else can take over without getting sharing
88 static void ctdb_packet_dump(struct ctdb_req_header
*hdr
)
90 if (DEBUGLEVEL
< 11) {
93 DEBUGADD(11, ("len=%d, magic=%x, vers=%d, gen=%d, op=%d, reqid=%d\n",
94 (int)hdr
->length
, (int)hdr
->ctdb_magic
,
95 (int)hdr
->ctdb_version
, (int)hdr
->generation
,
96 (int)hdr
->operation
, (int)hdr
->reqid
));
100 * Register a srvid with ctdbd
102 int register_with_ctdbd(struct ctdbd_connection
*conn
, uint64_t srvid
,
103 int (*cb
)(uint32_t src_vnn
, uint32_t dst_vnn
,
105 const uint8_t *msg
, size_t msglen
,
111 size_t num_callbacks
;
112 struct ctdbd_srvid_cb
*tmp
;
114 ret
= ctdbd_control(conn
, CTDB_CURRENT_NODE
,
115 CTDB_CONTROL_REGISTER_SRVID
, srvid
, 0,
116 tdb_null
, NULL
, NULL
, &cstatus
);
121 num_callbacks
= talloc_array_length(conn
->callbacks
);
123 tmp
= talloc_realloc(conn
, conn
->callbacks
, struct ctdbd_srvid_cb
,
128 conn
->callbacks
= tmp
;
130 conn
->callbacks
[num_callbacks
] = (struct ctdbd_srvid_cb
) {
131 .srvid
= srvid
, .cb
= cb
, .private_data
= private_data
137 static int ctdbd_msg_call_back(struct ctdbd_connection
*conn
,
138 struct ctdb_req_message_old
*msg
)
141 size_t i
, num_callbacks
;
143 msg_len
= msg
->hdr
.length
;
144 if (msg_len
< offsetof(struct ctdb_req_message_old
, data
)) {
145 DEBUG(10, ("%s: len %u too small\n", __func__
,
149 msg_len
-= offsetof(struct ctdb_req_message_old
, data
);
151 if (msg_len
< msg
->datalen
) {
152 DEBUG(10, ("%s: msg_len=%u < msg->datalen=%u\n", __func__
,
153 (unsigned)msg_len
, (unsigned)msg
->datalen
));
157 num_callbacks
= talloc_array_length(conn
->callbacks
);
159 for (i
=0; i
<num_callbacks
; i
++) {
160 struct ctdbd_srvid_cb
*cb
= &conn
->callbacks
[i
];
162 if ((cb
->srvid
== msg
->srvid
) && (cb
->cb
!= NULL
)) {
165 ret
= cb
->cb(msg
->hdr
.srcnode
, msg
->hdr
.destnode
,
166 msg
->srvid
, msg
->data
, msg
->datalen
,
177 * get our vnn from the cluster
179 static int get_cluster_vnn(struct ctdbd_connection
*conn
, uint32_t *vnn
)
183 ret
= ctdbd_control(conn
,
184 CTDB_CURRENT_NODE
, CTDB_CONTROL_GET_PNN
, 0, 0,
185 tdb_null
, NULL
, NULL
, &cstatus
);
187 DEBUG(1, ("ctdbd_control failed: %s\n", strerror(ret
)));
190 *vnn
= (uint32_t)cstatus
;
195 * Are we active (i.e. not banned or stopped?)
197 static bool ctdbd_working(struct ctdbd_connection
*conn
, uint32_t vnn
)
201 struct ctdb_node_map_old
*m
;
202 uint32_t failure_flags
;
206 ret
= ctdbd_control(conn
, CTDB_CURRENT_NODE
,
207 CTDB_CONTROL_GET_NODEMAP
, 0, 0,
208 tdb_null
, talloc_tos(), &outdata
, &cstatus
);
210 DEBUG(1, ("ctdbd_control failed: %s\n", strerror(ret
)));
213 if ((cstatus
!= 0) || (outdata
.dptr
== NULL
)) {
214 DEBUG(2, ("Received invalid ctdb data\n"));
218 m
= (struct ctdb_node_map_old
*)outdata
.dptr
;
220 for (i
=0; i
<m
->num
; i
++) {
221 if (vnn
== m
->nodes
[i
].pnn
) {
227 DEBUG(2, ("Did not find ourselves (node %d) in nodemap\n",
232 failure_flags
= NODE_FLAGS_BANNED
| NODE_FLAGS_DISCONNECTED
233 | NODE_FLAGS_PERMANENTLY_DISABLED
| NODE_FLAGS_STOPPED
;
235 if ((m
->nodes
[i
].flags
& failure_flags
) != 0) {
236 DEBUG(2, ("Node has status %x, not active\n",
237 (int)m
->nodes
[i
].flags
));
243 TALLOC_FREE(outdata
.dptr
);
247 uint32_t ctdbd_vnn(const struct ctdbd_connection
*conn
)
249 return conn
->our_vnn
;
253 * Get us a ctdb connection
256 static int ctdbd_connect(const char *sockname
, int *pfd
)
258 struct sockaddr_un addr
= { 0, };
263 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
266 DEBUG(3, ("Could not create socket: %s\n", strerror(err
)));
270 addr
.sun_family
= AF_UNIX
;
272 namelen
= strlcpy(addr
.sun_path
, sockname
, sizeof(addr
.sun_path
));
273 if (namelen
>= sizeof(addr
.sun_path
)) {
274 DEBUG(3, ("%s: Socket name too long: %s\n", __func__
,
280 salen
= sizeof(struct sockaddr_un
);
282 if (connect(fd
, (struct sockaddr
*)(void *)&addr
, salen
) == -1) {
284 DEBUG(1, ("connect(%s) failed: %s\n", sockname
,
294 static int ctdb_read_packet(int fd
, int timeout
, TALLOC_CTX
*mem_ctx
,
295 struct ctdb_req_header
**result
)
297 struct ctdb_req_header
*req
;
303 ret
= poll_intr_one_fd(fd
, POLLIN
, timeout
, &revents
);
315 nread
= read_data(fd
, &msglen
, sizeof(msglen
));
323 if (msglen
< sizeof(struct ctdb_req_header
)) {
327 req
= talloc_size(mem_ctx
, msglen
);
331 talloc_set_name_const(req
, "struct ctdb_req_header");
333 req
->length
= msglen
;
335 nread
= read_data(fd
, ((char *)req
) + sizeof(msglen
),
336 msglen
- sizeof(msglen
));
351 * Read a full ctdbd request. If we have a messaging context, defer incoming
352 * messages that might come in between.
355 static int ctdb_read_req(struct ctdbd_connection
*conn
, uint32_t reqid
,
356 TALLOC_CTX
*mem_ctx
, struct ctdb_req_header
**result
)
358 struct ctdb_req_header
*hdr
;
363 ret
= ctdb_read_packet(conn
->fd
, conn
->timeout
, mem_ctx
, &hdr
);
365 DEBUG(0, ("ctdb_read_packet failed: %s\n", strerror(ret
)));
366 cluster_fatal("ctdbd died\n");
369 DEBUG(11, ("Received ctdb packet\n"));
370 ctdb_packet_dump(hdr
);
372 if (hdr
->operation
== CTDB_REQ_MESSAGE
) {
373 struct ctdb_req_message_old
*msg
= (struct ctdb_req_message_old
*)hdr
;
375 if (conn
->msg_ctx
== NULL
) {
376 DEBUG(1, ("Got a message without having a msg ctx, "
377 "dropping msg %llu\n",
378 (long long unsigned)msg
->srvid
));
383 ret
= ctdbd_msg_call_back(conn
, msg
);
393 if ((reqid
!= 0) && (hdr
->reqid
!= reqid
)) {
394 /* we got the wrong reply */
395 DEBUG(0,("Discarding mismatched ctdb reqid %u should have "
396 "been %u\n", hdr
->reqid
, reqid
));
401 *result
= talloc_move(mem_ctx
, &hdr
);
406 static int ctdbd_connection_destructor(struct ctdbd_connection
*c
)
416 * Get us a ctdbd connection
419 static int ctdbd_init_connection(TALLOC_CTX
*mem_ctx
,
420 const char *sockname
, int timeout
,
421 struct ctdbd_connection
**pconn
)
423 struct ctdbd_connection
*conn
;
426 if (!(conn
= talloc_zero(mem_ctx
, struct ctdbd_connection
))) {
427 DEBUG(0, ("talloc failed\n"));
431 conn
->sockname
= talloc_strdup(conn
, sockname
);
432 if (conn
->sockname
== NULL
) {
433 DBG_ERR("talloc failed\n");
438 conn
->timeout
= timeout
;
440 if (conn
->timeout
== 0) {
444 ret
= ctdbd_connect(conn
->sockname
, &conn
->fd
);
446 DEBUG(1, ("ctdbd_connect failed: %s\n", strerror(ret
)));
449 talloc_set_destructor(conn
, ctdbd_connection_destructor
);
451 ret
= get_cluster_vnn(conn
, &conn
->our_vnn
);
454 DEBUG(10, ("get_cluster_vnn failed: %s\n", strerror(ret
)));
458 if (!ctdbd_working(conn
, conn
->our_vnn
)) {
459 DEBUG(2, ("Node is not working, can not connect\n"));
464 generate_random_buffer((unsigned char *)&conn
->rand_srvid
,
465 sizeof(conn
->rand_srvid
));
467 ret
= register_with_ctdbd(conn
, conn
->rand_srvid
, NULL
, NULL
);
470 DEBUG(5, ("Could not register random srvid: %s\n",
484 * Get us a ctdbd connection and register us as a process
487 int ctdbd_messaging_connection(TALLOC_CTX
*mem_ctx
,
488 const char *sockname
, int timeout
,
489 struct ctdbd_connection
**pconn
)
491 struct ctdbd_connection
*conn
;
494 ret
= ctdbd_init_connection(mem_ctx
, sockname
, timeout
, &conn
);
500 ret
= register_with_ctdbd(conn
, MSG_SRVID_SAMBA
, NULL
, NULL
);
513 struct messaging_context
*ctdb_conn_msg_ctx(struct ctdbd_connection
*conn
)
515 return conn
->msg_ctx
;
518 int ctdbd_conn_get_fd(struct ctdbd_connection
*conn
)
524 * Packet handler to receive and handle a ctdb message
526 static int ctdb_handle_message(struct ctdbd_connection
*conn
,
527 struct ctdb_req_header
*hdr
)
529 struct ctdb_req_message_old
*msg
;
531 if (hdr
->operation
!= CTDB_REQ_MESSAGE
) {
532 DEBUG(0, ("Received async msg of type %u, discarding\n",
537 msg
= (struct ctdb_req_message_old
*)hdr
;
539 ctdbd_msg_call_back(conn
, msg
);
545 * The ctdbd socket is readable asynchronuously
548 static void ctdbd_socket_handler(struct tevent_context
*event_ctx
,
549 struct tevent_fd
*event
,
553 struct ctdbd_connection
*conn
= talloc_get_type_abort(
554 private_data
, struct ctdbd_connection
);
555 struct ctdb_req_header
*hdr
= NULL
;
558 ret
= ctdb_read_packet(conn
->fd
, conn
->timeout
, talloc_tos(), &hdr
);
560 DEBUG(0, ("ctdb_read_packet failed: %s\n", strerror(ret
)));
561 cluster_fatal("ctdbd died\n");
564 ret
= ctdb_handle_message(conn
, hdr
);
569 DEBUG(10, ("could not handle incoming message: %s\n",
575 * Prepare a ctdbd connection to receive messages
578 int ctdbd_register_msg_ctx(struct ctdbd_connection
*conn
,
579 struct messaging_context
*msg_ctx
,
580 struct tevent_context
*ev
)
582 SMB_ASSERT(conn
->msg_ctx
== NULL
);
583 SMB_ASSERT(conn
->fde
== NULL
);
585 conn
->fde
= tevent_add_fd(ev
, conn
, conn
->fd
, TEVENT_FD_READ
,
586 ctdbd_socket_handler
, conn
);
587 if (conn
->fde
== NULL
) {
588 DEBUG(0, ("event_add_fd failed\n"));
592 conn
->msg_ctx
= msg_ctx
;
597 int ctdbd_messaging_send_iov(struct ctdbd_connection
*conn
,
598 uint32_t dst_vnn
, uint64_t dst_srvid
,
599 const struct iovec
*iov
, int iovlen
)
601 struct ctdb_req_message_old r
;
602 struct iovec iov2
[iovlen
+1];
603 size_t buflen
= iov_buflen(iov
, iovlen
);
606 r
.hdr
.length
= offsetof(struct ctdb_req_message_old
, data
) + buflen
;
607 r
.hdr
.ctdb_magic
= CTDB_MAGIC
;
608 r
.hdr
.ctdb_version
= CTDB_PROTOCOL
;
609 r
.hdr
.generation
= 1;
610 r
.hdr
.operation
= CTDB_REQ_MESSAGE
;
611 r
.hdr
.destnode
= dst_vnn
;
612 r
.hdr
.srcnode
= conn
->our_vnn
;
617 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
618 ctdb_packet_dump(&r
.hdr
);
620 iov2
[0].iov_base
= &r
;
621 iov2
[0].iov_len
= offsetof(struct ctdb_req_message_old
, data
);
622 memcpy(&iov2
[1], iov
, iovlen
* sizeof(struct iovec
));
624 nwritten
= write_data_iov(conn
->fd
, iov2
, iovlen
+1);
625 if (nwritten
== -1) {
626 DEBUG(3, ("write_data_iov failed: %s\n", strerror(errno
)));
627 cluster_fatal("cluster dispatch daemon msg write error\n");
634 * send/recv a generic ctdb control message
636 static int ctdbd_control(struct ctdbd_connection
*conn
,
637 uint32_t vnn
, uint32_t opcode
,
638 uint64_t srvid
, uint32_t flags
,
640 TALLOC_CTX
*mem_ctx
, TDB_DATA
*outdata
,
643 struct ctdb_req_control_old req
;
644 struct ctdb_req_header
*hdr
;
645 struct ctdb_reply_control_old
*reply
= NULL
;
651 req
.hdr
.length
= offsetof(struct ctdb_req_control_old
, data
) + data
.dsize
;
652 req
.hdr
.ctdb_magic
= CTDB_MAGIC
;
653 req
.hdr
.ctdb_version
= CTDB_PROTOCOL
;
654 req
.hdr
.operation
= CTDB_REQ_CONTROL
;
655 req
.hdr
.reqid
= ctdbd_next_reqid(conn
);
656 req
.hdr
.destnode
= vnn
;
659 req
.datalen
= data
.dsize
;
662 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
663 ctdb_packet_dump(&req
.hdr
);
665 iov
[0].iov_base
= &req
;
666 iov
[0].iov_len
= offsetof(struct ctdb_req_control_old
, data
);
667 iov
[1].iov_base
= data
.dptr
;
668 iov
[1].iov_len
= data
.dsize
;
670 nwritten
= write_data_iov(conn
->fd
, iov
, ARRAY_SIZE(iov
));
671 if (nwritten
== -1) {
672 DEBUG(3, ("write_data_iov failed: %s\n", strerror(errno
)));
673 cluster_fatal("cluster dispatch daemon msg write error\n");
676 if (flags
& CTDB_CTRL_FLAG_NOREPLY
) {
683 ret
= ctdb_read_req(conn
, req
.hdr
.reqid
, NULL
, &hdr
);
685 DEBUG(10, ("ctdb_read_req failed: %s\n", strerror(ret
)));
689 if (hdr
->operation
!= CTDB_REPLY_CONTROL
) {
690 DEBUG(0, ("received invalid reply\n"));
694 reply
= (struct ctdb_reply_control_old
*)hdr
;
697 if (!(outdata
->dptr
= (uint8_t *)talloc_memdup(
698 mem_ctx
, reply
->data
, reply
->datalen
))) {
702 outdata
->dsize
= reply
->datalen
;
705 (*cstatus
) = reply
->status
;
713 * see if a remote process exists
715 bool ctdbd_process_exists(struct ctdbd_connection
*conn
, uint32_t vnn
, pid_t pid
)
723 if (!ctdb_processes_exist(conn
, &id
, 1, &result
)) {
724 DEBUG(10, ("ctdb_processes_exist failed\n"));
730 bool ctdb_processes_exist(struct ctdbd_connection
*conn
,
731 const struct server_id
*pids
, int num_pids
,
734 TALLOC_CTX
*frame
= talloc_stackframe();
739 reqids
= talloc_array(talloc_tos(), uint32_t, num_pids
);
740 if (reqids
== NULL
) {
744 for (i
=0; i
<num_pids
; i
++) {
745 struct ctdb_req_control_old req
;
751 reqids
[i
] = ctdbd_next_reqid(conn
);
756 * pids[i].pid is uint64_t, scale down to pid_t which
757 * is the wire protocol towards ctdb.
761 DEBUG(10, ("Requesting PID %d/%d, reqid=%d\n",
762 (int)pids
[i
].vnn
, (int)pid
,
765 req
.hdr
.length
= offsetof(struct ctdb_req_control_old
, data
);
766 req
.hdr
.length
+= sizeof(pid
);
767 req
.hdr
.ctdb_magic
= CTDB_MAGIC
;
768 req
.hdr
.ctdb_version
= CTDB_PROTOCOL
;
769 req
.hdr
.operation
= CTDB_REQ_CONTROL
;
770 req
.hdr
.reqid
= reqids
[i
];
771 req
.hdr
.destnode
= pids
[i
].vnn
;
772 req
.opcode
= CTDB_CONTROL_PROCESS_EXISTS
;
774 req
.datalen
= sizeof(pid
);
777 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
778 ctdb_packet_dump(&req
.hdr
);
780 iov
[0].iov_base
= &req
;
781 iov
[0].iov_len
= offsetof(struct ctdb_req_control_old
, data
);
782 iov
[1].iov_base
= &pid
;
783 iov
[1].iov_len
= sizeof(pid
);
785 nwritten
= write_data_iov(conn
->fd
, iov
, ARRAY_SIZE(iov
));
786 if (nwritten
== -1) {
787 DEBUG(10, ("write_data_iov failed: %s\n",
795 while (num_received
< num_pids
) {
796 struct ctdb_req_header
*hdr
;
797 struct ctdb_reply_control_old
*reply
;
801 ret
= ctdb_read_req(conn
, 0, talloc_tos(), &hdr
);
803 DEBUG(10, ("ctdb_read_req failed: %s\n",
808 if (hdr
->operation
!= CTDB_REPLY_CONTROL
) {
809 DEBUG(10, ("Received invalid reply\n"));
812 reply
= (struct ctdb_reply_control_old
*)hdr
;
814 reqid
= reply
->hdr
.reqid
;
816 DEBUG(10, ("Received reqid %d\n", (int)reqid
));
818 for (i
=0; i
<num_pids
; i
++) {
819 if (reqid
== reqids
[i
]) {
824 DEBUG(10, ("Received unknown record number %u\n",
828 results
[i
] = ((reply
->status
) == 0);
842 char *ctdbd_dbpath(struct ctdbd_connection
*conn
,
843 TALLOC_CTX
*mem_ctx
, uint32_t db_id
)
847 TDB_DATA rdata
= {0};
850 data
.dptr
= (uint8_t*)&db_id
;
851 data
.dsize
= sizeof(db_id
);
853 ret
= ctdbd_control(conn
, CTDB_CURRENT_NODE
,
854 CTDB_CONTROL_GETDBPATH
, 0, 0, data
,
855 mem_ctx
, &rdata
, &cstatus
);
856 if ((ret
!= 0) || cstatus
!= 0) {
857 DEBUG(0, (__location__
" ctdb_control for getdbpath failed: %s\n",
862 return (char *)rdata
.dptr
;
866 * attach to a ctdb database
868 int ctdbd_db_attach(struct ctdbd_connection
*conn
,
869 const char *name
, uint32_t *db_id
, int tdb_flags
)
874 bool persistent
= (tdb_flags
& TDB_CLEAR_IF_FIRST
) == 0;
876 data
= string_term_tdb_data(name
);
878 ret
= ctdbd_control(conn
, CTDB_CURRENT_NODE
,
880 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
881 : CTDB_CONTROL_DB_ATTACH
,
882 tdb_flags
, 0, data
, NULL
, &data
, &cstatus
);
884 DEBUG(0, (__location__
" ctdb_control for db_attach "
885 "failed: %s\n", strerror(ret
)));
889 if (cstatus
!= 0 || data
.dsize
!= sizeof(uint32_t)) {
890 DEBUG(0,(__location__
" ctdb_control for db_attach failed\n"));
894 *db_id
= *(uint32_t *)data
.dptr
;
895 talloc_free(data
.dptr
);
897 if (!(tdb_flags
& TDB_SEQNUM
)) {
901 data
.dptr
= (uint8_t *)db_id
;
902 data
.dsize
= sizeof(*db_id
);
904 ret
= ctdbd_control(conn
, CTDB_CURRENT_NODE
,
905 CTDB_CONTROL_ENABLE_SEQNUM
, 0, 0, data
,
906 NULL
, NULL
, &cstatus
);
907 if ((ret
!= 0) || cstatus
!= 0) {
908 DEBUG(0, (__location__
" ctdb_control for enable seqnum "
909 "failed: %s\n", strerror(ret
)));
910 return (ret
== 0) ? EIO
: ret
;
917 * force the migration of a record to this node
919 int ctdbd_migrate(struct ctdbd_connection
*conn
, uint32_t db_id
, TDB_DATA key
)
921 struct ctdb_req_call_old req
;
922 struct ctdb_req_header
*hdr
;
929 req
.hdr
.length
= offsetof(struct ctdb_req_call_old
, data
) + key
.dsize
;
930 req
.hdr
.ctdb_magic
= CTDB_MAGIC
;
931 req
.hdr
.ctdb_version
= CTDB_PROTOCOL
;
932 req
.hdr
.operation
= CTDB_REQ_CALL
;
933 req
.hdr
.reqid
= ctdbd_next_reqid(conn
);
934 req
.flags
= CTDB_IMMEDIATE_MIGRATION
;
935 req
.callid
= CTDB_NULL_FUNC
;
937 req
.keylen
= key
.dsize
;
939 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
940 ctdb_packet_dump(&req
.hdr
);
942 iov
[0].iov_base
= &req
;
943 iov
[0].iov_len
= offsetof(struct ctdb_req_call_old
, data
);
944 iov
[1].iov_base
= key
.dptr
;
945 iov
[1].iov_len
= key
.dsize
;
947 nwritten
= write_data_iov(conn
->fd
, iov
, ARRAY_SIZE(iov
));
948 if (nwritten
== -1) {
949 DEBUG(3, ("write_data_iov failed: %s\n", strerror(errno
)));
950 cluster_fatal("cluster dispatch daemon msg write error\n");
953 ret
= ctdb_read_req(conn
, req
.hdr
.reqid
, NULL
, &hdr
);
955 DEBUG(10, ("ctdb_read_req failed: %s\n", strerror(ret
)));
959 if (hdr
->operation
!= CTDB_REPLY_CALL
) {
960 DEBUG(0, ("received invalid reply\n"));
971 * Fetch a record and parse it
973 int ctdbd_parse(struct ctdbd_connection
*conn
, uint32_t db_id
,
974 TDB_DATA key
, bool local_copy
,
975 void (*parser
)(TDB_DATA key
, TDB_DATA data
,
979 struct ctdb_req_call_old req
;
980 struct ctdb_req_header
*hdr
= NULL
;
981 struct ctdb_reply_call_old
*reply
;
987 flags
= local_copy
? CTDB_WANT_READONLY
: 0;
991 req
.hdr
.length
= offsetof(struct ctdb_req_call_old
, data
) + key
.dsize
;
992 req
.hdr
.ctdb_magic
= CTDB_MAGIC
;
993 req
.hdr
.ctdb_version
= CTDB_PROTOCOL
;
994 req
.hdr
.operation
= CTDB_REQ_CALL
;
995 req
.hdr
.reqid
= ctdbd_next_reqid(conn
);
997 req
.callid
= CTDB_FETCH_FUNC
;
999 req
.keylen
= key
.dsize
;
1001 iov
[0].iov_base
= &req
;
1002 iov
[0].iov_len
= offsetof(struct ctdb_req_call_old
, data
);
1003 iov
[1].iov_base
= key
.dptr
;
1004 iov
[1].iov_len
= key
.dsize
;
1006 nwritten
= write_data_iov(conn
->fd
, iov
, ARRAY_SIZE(iov
));
1007 if (nwritten
== -1) {
1008 DEBUG(3, ("write_data_iov failed: %s\n", strerror(errno
)));
1009 cluster_fatal("cluster dispatch daemon msg write error\n");
1012 ret
= ctdb_read_req(conn
, req
.hdr
.reqid
, NULL
, &hdr
);
1014 DEBUG(10, ("ctdb_read_req failed: %s\n", strerror(ret
)));
1018 if ((hdr
== NULL
) || (hdr
->operation
!= CTDB_REPLY_CALL
)) {
1019 DEBUG(0, ("received invalid reply\n"));
1023 reply
= (struct ctdb_reply_call_old
*)hdr
;
1025 if (reply
->datalen
== 0) {
1027 * Treat an empty record as non-existing
1033 parser(key
, make_tdb_data(&reply
->data
[0], reply
->datalen
),
1043 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1044 connection to ctdbd to avoid the hairy recursive and async problems with
1048 int ctdbd_traverse(struct ctdbd_connection
*master
, uint32_t db_id
,
1049 void (*fn
)(TDB_DATA key
, TDB_DATA data
,
1050 void *private_data
),
1053 struct ctdbd_connection
*conn
;
1056 struct ctdb_traverse_start t
;
1060 ret
= ctdbd_init_connection(NULL
, master
->sockname
, master
->timeout
,
1064 DEBUG(0, ("ctdbd_init_connection failed: %s\n",
1070 t
.srvid
= conn
->rand_srvid
;
1071 t
.reqid
= ctdbd_next_reqid(conn
);
1073 data
.dptr
= (uint8_t *)&t
;
1074 data
.dsize
= sizeof(t
);
1076 ret
= ctdbd_control(conn
, CTDB_CURRENT_NODE
,
1077 CTDB_CONTROL_TRAVERSE_START
, conn
->rand_srvid
,
1078 0, data
, NULL
, NULL
, &cstatus
);
1080 if ((ret
!= 0) || (cstatus
!= 0)) {
1081 DEBUG(0,("ctdbd_control failed: %s, %d\n", strerror(ret
),
1086 * We need a mapping here
1095 struct ctdb_req_header
*hdr
= NULL
;
1096 struct ctdb_req_message_old
*m
;
1097 struct ctdb_rec_data_old
*d
;
1099 ret
= ctdb_read_packet(conn
->fd
, conn
->timeout
, conn
, &hdr
);
1101 DEBUG(0, ("ctdb_read_packet failed: %s\n",
1103 cluster_fatal("ctdbd died\n");
1106 if (hdr
->operation
!= CTDB_REQ_MESSAGE
) {
1107 DEBUG(0, ("Got operation %u, expected a message\n",
1108 (unsigned)hdr
->operation
));
1113 m
= (struct ctdb_req_message_old
*)hdr
;
1114 d
= (struct ctdb_rec_data_old
*)&m
->data
[0];
1115 if (m
->datalen
< sizeof(uint32_t) || m
->datalen
!= d
->length
) {
1116 DEBUG(0, ("Got invalid traverse data of length %d\n",
1122 key
.dsize
= d
->keylen
;
1123 key
.dptr
= &d
->data
[0];
1124 data
.dsize
= d
->datalen
;
1125 data
.dptr
= &d
->data
[d
->keylen
];
1127 if (key
.dsize
== 0 && data
.dsize
== 0) {
1128 /* end of traverse */
1133 if (data
.dsize
< sizeof(struct ctdb_ltdb_header
)) {
1134 DEBUG(0, ("Got invalid ltdb header length %d\n",
1139 data
.dsize
-= sizeof(struct ctdb_ltdb_header
);
1140 data
.dptr
+= sizeof(struct ctdb_ltdb_header
);
1143 fn(key
, data
, private_data
);
1150 This is used to canonicalize a ctdb_sock_addr structure.
1152 static void smbd_ctdb_canonicalize_ip(const struct sockaddr_storage
*in
,
1153 struct sockaddr_storage
*out
)
1155 memcpy(out
, in
, sizeof (*out
));
1158 if (in
->ss_family
== AF_INET6
) {
1159 const char prefix
[12] = { 0,0,0,0,0,0,0,0,0,0,0xff,0xff };
1160 const struct sockaddr_in6
*in6
=
1161 (const struct sockaddr_in6
*)in
;
1162 struct sockaddr_in
*out4
= (struct sockaddr_in
*)out
;
1163 if (memcmp(&in6
->sin6_addr
, prefix
, 12) == 0) {
1164 memset(out
, 0, sizeof(*out
));
1165 #ifdef HAVE_SOCK_SIN_LEN
1166 out4
->sin_len
= sizeof(*out
);
1168 out4
->sin_family
= AF_INET
;
1169 out4
->sin_port
= in6
->sin6_port
;
1170 memcpy(&out4
->sin_addr
, &in6
->sin6_addr
.s6_addr
[12], 4);
1177 * Register us as a server for a particular tcp connection
1180 int ctdbd_register_ips(struct ctdbd_connection
*conn
,
1181 const struct sockaddr_storage
*_server
,
1182 const struct sockaddr_storage
*_client
,
1183 int (*cb
)(uint32_t src_vnn
, uint32_t dst_vnn
,
1185 const uint8_t *msg
, size_t msglen
,
1186 void *private_data
),
1189 struct ctdb_connection p
;
1190 TDB_DATA data
= { .dptr
= (uint8_t *)&p
, .dsize
= sizeof(p
) };
1192 struct sockaddr_storage client
;
1193 struct sockaddr_storage server
;
1196 * Only one connection so far
1199 smbd_ctdb_canonicalize_ip(_client
, &client
);
1200 smbd_ctdb_canonicalize_ip(_server
, &server
);
1202 switch (client
.ss_family
) {
1204 memcpy(&p
.dst
.ip
, &server
, sizeof(p
.dst
.ip
));
1205 memcpy(&p
.src
.ip
, &client
, sizeof(p
.src
.ip
));
1208 memcpy(&p
.dst
.ip6
, &server
, sizeof(p
.dst
.ip6
));
1209 memcpy(&p
.src
.ip6
, &client
, sizeof(p
.src
.ip6
));
1216 * We want to be told about IP releases
1219 ret
= register_with_ctdbd(conn
, CTDB_SRVID_RELEASE_IP
,
1226 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1227 * can send an extra ack to trigger a reset for our client, so it
1228 * immediately reconnects
1230 ret
= ctdbd_control(conn
, CTDB_CURRENT_NODE
,
1231 CTDB_CONTROL_TCP_CLIENT
, 0,
1232 CTDB_CTRL_FLAG_NOREPLY
, data
, NULL
, NULL
,
1241 call a control on the local node
1243 int ctdbd_control_local(struct ctdbd_connection
*conn
, uint32_t opcode
,
1244 uint64_t srvid
, uint32_t flags
, TDB_DATA data
,
1245 TALLOC_CTX
*mem_ctx
, TDB_DATA
*outdata
,
1248 return ctdbd_control(conn
, CTDB_CURRENT_NODE
, opcode
, srvid
, flags
, data
,
1249 mem_ctx
, outdata
, cstatus
);
1252 int ctdb_watch_us(struct ctdbd_connection
*conn
)
1254 struct ctdb_notify_data_old reg_data
;
1259 reg_data
.srvid
= CTDB_SRVID_SAMBA_NOTIFY
;
1261 reg_data
.notify_data
[0] = 0;
1263 struct_len
= offsetof(struct ctdb_notify_data_old
,
1264 notify_data
) + reg_data
.len
;
1266 ret
= ctdbd_control_local(
1267 conn
, CTDB_CONTROL_REGISTER_NOTIFY
, conn
->rand_srvid
, 0,
1268 make_tdb_data((uint8_t *)®_data
, struct_len
),
1269 NULL
, NULL
, &cstatus
);
1271 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1277 int ctdb_unwatch(struct ctdbd_connection
*conn
)
1279 uint64_t srvid
= CTDB_SRVID_SAMBA_NOTIFY
;
1283 ret
= ctdbd_control_local(
1284 conn
, CTDB_CONTROL_DEREGISTER_NOTIFY
, conn
->rand_srvid
, 0,
1285 make_tdb_data((uint8_t *)&srvid
, sizeof(srvid
)),
1286 NULL
, NULL
, &cstatus
);
1288 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1294 int ctdbd_probe(const char *sockname
, int timeout
)
1297 * Do a very early check if ctdbd is around to avoid an abort and core
1300 struct ctdbd_connection
*conn
= NULL
;
1303 ret
= ctdbd_messaging_connection(talloc_tos(), sockname
, timeout
,
1307 * We only care if we can connect.