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_state = TALLOC_P(NULL, struct deferred_msg_state))) {
357 DEBUG(0, ("talloc failed\n"));
362 if (!(msg_state->rec = ctdb_pull_messaging_rec(
363 msg_state, state.req.length, msg))) {
364 DEBUG(0, ("ctdbd_pull_messaging_rec failed\n"));
365 TALLOC_FREE(msg_state);
372 msg_state->msg_ctx = conn->msg_ctx;
375 * We're waiting for a call reply, but an async message has
376 * crossed. Defer dispatching to the toplevel event loop.
378 evt = event_add_timed(conn->msg_ctx->event_ctx,
379 conn->msg_ctx->event_ctx,
381 "deferred_message_dispatch",
382 deferred_message_dispatch,
385 DEBUG(0, ("event_add_timed failed\n"));
386 TALLOC_FREE(msg_state);
394 if (hdr->reqid != reqid) {
395 /* we got the wrong reply */
396 DEBUG(0,("Discarding mismatched ctdb reqid %u should have "
397 "been %u\n", hdr->reqid, reqid));
402 *((void **)result) = talloc_move(mem_ctx, &hdr);
408 * Get us a ctdbd connection
411 NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
412 struct ctdbd_connection **pconn)
414 struct ctdbd_connection *conn;
417 if (!(conn = TALLOC_ZERO_P(mem_ctx, struct ctdbd_connection))) {
418 DEBUG(0, ("talloc failed\n"));
419 return NT_STATUS_NO_MEMORY;
422 status = ctdbd_connect(conn, &conn->pkt);
424 if (!NT_STATUS_IS_OK(status)) {
425 DEBUG(10, ("ctdbd_connect failed: %s\n", nt_errstr(status)));
429 status = get_cluster_vnn(conn, &conn->our_vnn);
431 if (!NT_STATUS_IS_OK(status)) {
432 DEBUG(10, ("get_cluster_vnn failed: %s\n", nt_errstr(status)));
436 generate_random_buffer((unsigned char *)&conn->rand_srvid,
437 sizeof(conn->rand_srvid));
439 status = register_with_ctdbd(conn, conn->rand_srvid);
441 if (!NT_STATUS_IS_OK(status)) {
442 DEBUG(5, ("Could not register random srvid: %s\n",
456 * Get us a ctdbd connection and register us as a process
459 NTSTATUS ctdbd_messaging_connection(TALLOC_CTX *mem_ctx,
460 struct ctdbd_connection **pconn)
462 struct ctdbd_connection *conn;
465 status = ctdbd_init_connection(mem_ctx, &conn);
467 if (!NT_STATUS_IS_OK(status)) {
471 status = register_with_ctdbd(conn, (uint64_t)sys_getpid());
472 if (!NT_STATUS_IS_OK(status)) {
476 status = register_with_ctdbd(conn, MSG_SRVID_SAMBA);
477 if (!NT_STATUS_IS_OK(status)) {
490 * Packet handler to receive and handle a ctdb message
492 static NTSTATUS ctdb_handle_message(const struct data_blob *data,
495 struct ctdbd_connection *conn = talloc_get_type_abort(
496 private_data, struct ctdbd_connection);
497 struct ctdb_req_message *msg;
498 struct messaging_rec *msg_rec;
500 msg = (struct ctdb_req_message *)data->data;
502 if (msg->hdr.operation != CTDB_REQ_MESSAGE) {
503 DEBUG(0, ("Received async msg of type %u, discarding\n",
504 msg->hdr.operation));
505 return NT_STATUS_INVALID_PARAMETER;
508 if ((conn->release_ip_handler != NULL)
509 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
510 /* must be dispatched immediately */
511 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
512 conn->release_ip_handler((const char *)msg->data,
513 conn->release_ip_priv);
517 SMB_ASSERT(conn->msg_ctx != NULL);
519 if (msg->srvid == CTDB_SRVID_RECONFIGURE) {
520 DEBUG(0,("Got cluster reconfigure message\n"));
522 * when the cluster is reconfigured, we need to clean the brl
525 messaging_send(conn->msg_ctx, procid_self(),
526 MSG_SMB_BRL_VALIDATE, &data_blob_null);
529 * it's possible that we have just rejoined the cluster after
530 * an outage. In that case our pending locks could have been
531 * removed from the lockdb, so retry them once more
533 message_send_all(conn->msg_ctx, MSG_SMB_UNLOCK, NULL, 0, NULL);
539 /* only messages to our pid or the broadcast are valid here */
540 if (msg->srvid != sys_getpid() && msg->srvid != MSG_SRVID_SAMBA) {
541 DEBUG(0,("Got unexpected message with srvid=%llu\n",
542 (unsigned long long)msg->srvid));
546 if (!(msg_rec = ctdb_pull_messaging_rec(NULL, data->length, msg))) {
547 DEBUG(10, ("ctdb_pull_messaging_rec failed\n"));
548 return NT_STATUS_NO_MEMORY;
551 messaging_dispatch_rec(conn->msg_ctx, msg_rec);
553 TALLOC_FREE(msg_rec);
558 * The ctdbd socket is readable asynchronuously
561 static void ctdbd_socket_handler(struct event_context *event_ctx,
562 struct fd_event *event,
566 struct ctdbd_connection *conn = talloc_get_type_abort(
567 private_data, struct ctdbd_connection);
571 status = packet_fd_read(conn->pkt);
573 if (!NT_STATUS_IS_OK(status)) {
574 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
575 cluster_fatal("ctdbd died\n");
578 while (packet_handler(conn->pkt, ctdb_req_complete,
579 ctdb_handle_message, conn, &status)) {
580 if (!NT_STATUS_IS_OK(status)) {
581 DEBUG(10, ("could not handle incoming message: %s\n",
588 * Prepare a ctdbd connection to receive messages
591 NTSTATUS ctdbd_register_msg_ctx(struct ctdbd_connection *conn,
592 struct messaging_context *msg_ctx)
594 SMB_ASSERT(conn->msg_ctx == NULL);
595 SMB_ASSERT(conn->fde == NULL);
597 if (!(conn->fde = event_add_fd(msg_ctx->event_ctx, conn,
598 packet_get_fd(conn->pkt),
600 ctdbd_socket_handler,
602 DEBUG(0, ("event_add_fd failed\n"));
603 return NT_STATUS_NO_MEMORY;
606 conn->msg_ctx = msg_ctx;
612 * Send a messaging message across a ctdbd
615 NTSTATUS ctdbd_messaging_send(struct ctdbd_connection *conn,
616 uint32 dst_vnn, uint64 dst_srvid,
617 struct messaging_rec *msg)
619 struct ctdb_req_message r;
623 enum ndr_err_code ndr_err;
625 if (!(mem_ctx = talloc_init("ctdbd_messaging_send"))) {
626 DEBUG(0, ("talloc failed\n"));
627 return NT_STATUS_NO_MEMORY;
630 ndr_err = ndr_push_struct_blob(
632 (ndr_push_flags_fn_t)ndr_push_messaging_rec);
634 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
635 DEBUG(0, ("ndr_push_struct_blob failed: %s\n",
636 ndr_errstr(ndr_err)));
637 status = ndr_map_error2ntstatus(ndr_err);
641 r.hdr.length = offsetof(struct ctdb_req_message, data) + blob.length;
642 r.hdr.ctdb_magic = CTDB_MAGIC;
643 r.hdr.ctdb_version = CTDB_VERSION;
644 r.hdr.generation = 1;
645 r.hdr.operation = CTDB_REQ_MESSAGE;
646 r.hdr.destnode = dst_vnn;
647 r.hdr.srcnode = conn->our_vnn;
650 r.datalen = blob.length;
652 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
653 ctdb_packet_dump(&r.hdr);
655 status = packet_send(
657 data_blob_const(&r, offsetof(struct ctdb_req_message, data)),
660 if (!NT_STATUS_IS_OK(status)) {
661 DEBUG(0, ("packet_send failed: %s\n", nt_errstr(status)));
665 status = packet_flush(conn->pkt);
667 if (!NT_STATUS_IS_OK(status)) {
668 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
669 cluster_fatal("cluster dispatch daemon msg write error\n");
672 status = NT_STATUS_OK;
674 TALLOC_FREE(mem_ctx);
679 * send/recv a generic ctdb control message
681 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
682 uint32_t vnn, uint32 opcode,
683 uint64_t srvid, uint32_t flags,
685 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
688 struct ctdb_req_control req;
689 struct ctdb_reply_control *reply = NULL;
690 struct ctdbd_connection *new_conn = NULL;
694 status = ctdbd_init_connection(NULL, &new_conn);
696 if (!NT_STATUS_IS_OK(status)) {
697 DEBUG(10, ("Could not init temp connection: %s\n",
706 req.hdr.length = offsetof(struct ctdb_req_control, data) + data.dsize;
707 req.hdr.ctdb_magic = CTDB_MAGIC;
708 req.hdr.ctdb_version = CTDB_VERSION;
709 req.hdr.operation = CTDB_REQ_CONTROL;
710 req.hdr.reqid = ++conn->reqid;
711 req.hdr.destnode = vnn;
714 req.datalen = data.dsize;
716 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
717 ctdb_packet_dump(&req.hdr);
719 status = packet_send(
721 data_blob_const(&req, offsetof(struct ctdb_req_control, data)),
722 data_blob_const(data.dptr, data.dsize));
724 if (!NT_STATUS_IS_OK(status)) {
725 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
729 status = packet_flush(conn->pkt);
731 if (!NT_STATUS_IS_OK(status)) {
732 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
733 cluster_fatal("cluster dispatch daemon control write error\n");
736 if (flags & CTDB_CTRL_FLAG_NOREPLY) {
737 TALLOC_FREE(new_conn);
741 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
743 if (!NT_STATUS_IS_OK(status)) {
744 DEBUG(10, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
748 if (reply->hdr.operation != CTDB_REPLY_CONTROL) {
749 DEBUG(0, ("received invalid reply\n"));
754 if (!(outdata->dptr = (uint8 *)talloc_memdup(
755 mem_ctx, reply->data, reply->datalen))) {
757 return NT_STATUS_NO_MEMORY;
759 outdata->dsize = reply->datalen;
762 (*cstatus) = reply->status;
765 status = NT_STATUS_OK;
768 TALLOC_FREE(new_conn);
774 * see if a remote process exists
776 bool ctdbd_process_exists(struct ctdbd_connection *conn, uint32 vnn, pid_t pid)
782 data.dptr = (uint8_t*)&pid;
783 data.dsize = sizeof(pid);
785 status = ctdbd_control(conn, vnn, CTDB_CONTROL_PROCESS_EXISTS, 0, 0,
786 data, NULL, NULL, &cstatus);
787 if (!NT_STATUS_IS_OK(status)) {
788 DEBUG(0, (__location__ " ctdb_control for process_exists "
799 char *ctdbd_dbpath(struct ctdbd_connection *conn,
800 TALLOC_CTX *mem_ctx, uint32_t db_id)
806 data.dptr = (uint8_t*)&db_id;
807 data.dsize = sizeof(db_id);
809 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
810 CTDB_CONTROL_GETDBPATH, 0, 0, data,
811 mem_ctx, &data, &cstatus);
812 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
813 DEBUG(0,(__location__ " ctdb_control for getdbpath failed\n"));
817 return (char *)data.dptr;
821 * attach to a ctdb database
823 NTSTATUS ctdbd_db_attach(struct ctdbd_connection *conn,
824 const char *name, uint32_t *db_id, int tdb_flags)
829 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
831 data.dptr = (uint8_t*)name;
832 data.dsize = strlen(name)+1;
834 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
836 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
837 : CTDB_CONTROL_DB_ATTACH,
838 0, 0, data, NULL, &data, &cstatus);
839 if (!NT_STATUS_IS_OK(status)) {
840 DEBUG(0, (__location__ " ctdb_control for db_attach "
841 "failed: %s\n", nt_errstr(status)));
845 if (cstatus != 0 || data.dsize != sizeof(uint32_t)) {
846 DEBUG(0,(__location__ " ctdb_control for db_attach failed\n"));
847 return NT_STATUS_INTERNAL_ERROR;
850 *db_id = *(uint32_t *)data.dptr;
851 talloc_free(data.dptr);
853 if (!(tdb_flags & TDB_SEQNUM)) {
857 data.dptr = (uint8_t *)db_id;
858 data.dsize = sizeof(*db_id);
860 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
861 CTDB_CONTROL_ENABLE_SEQNUM, 0, 0, data,
862 NULL, NULL, &cstatus);
863 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
864 DEBUG(0,(__location__ " ctdb_control for enable seqnum "
866 return NT_STATUS_IS_OK(status) ? NT_STATUS_INTERNAL_ERROR :
874 * force the migration of a record to this node
876 NTSTATUS ctdbd_migrate(struct ctdbd_connection *conn, uint32 db_id,
879 struct ctdb_req_call req;
880 struct ctdb_reply_call *reply;
885 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
886 req.hdr.ctdb_magic = CTDB_MAGIC;
887 req.hdr.ctdb_version = CTDB_VERSION;
888 req.hdr.operation = CTDB_REQ_CALL;
889 req.hdr.reqid = ++conn->reqid;
890 req.flags = CTDB_IMMEDIATE_MIGRATION;
891 req.callid = CTDB_NULL_FUNC;
893 req.keylen = key.dsize;
895 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
896 ctdb_packet_dump(&req.hdr);
898 status = packet_send(
900 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
901 data_blob_const(key.dptr, key.dsize));
903 if (!NT_STATUS_IS_OK(status)) {
904 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
908 status = packet_flush(conn->pkt);
910 if (!NT_STATUS_IS_OK(status)) {
911 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
912 cluster_fatal("cluster dispatch daemon control write error\n");
915 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
917 if (!NT_STATUS_IS_OK(status)) {
918 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
922 if (reply->hdr.operation != CTDB_REPLY_CALL) {
923 DEBUG(0, ("received invalid reply\n"));
924 status = NT_STATUS_INTERNAL_ERROR;
928 status = NT_STATUS_OK;
936 * remotely fetch a record without locking it or forcing a migration
938 NTSTATUS ctdbd_fetch(struct ctdbd_connection *conn, uint32 db_id,
939 TDB_DATA key, TALLOC_CTX *mem_ctx, TDB_DATA *data)
941 struct ctdb_req_call req;
942 struct ctdb_reply_call *reply;
947 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
948 req.hdr.ctdb_magic = CTDB_MAGIC;
949 req.hdr.ctdb_version = CTDB_VERSION;
950 req.hdr.operation = CTDB_REQ_CALL;
951 req.hdr.reqid = ++conn->reqid;
953 req.callid = CTDB_FETCH_FUNC;
955 req.keylen = key.dsize;
957 status = packet_send(
959 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
960 data_blob_const(key.dptr, key.dsize));
962 if (!NT_STATUS_IS_OK(status)) {
963 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
967 status = packet_flush(conn->pkt);
969 if (!NT_STATUS_IS_OK(status)) {
970 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
971 cluster_fatal("cluster dispatch daemon control write error\n");
974 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
976 if (!NT_STATUS_IS_OK(status)) {
977 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
981 if (reply->hdr.operation != CTDB_REPLY_CALL) {
982 DEBUG(0, ("received invalid reply\n"));
983 status = NT_STATUS_INTERNAL_ERROR;
987 data->dsize = reply->datalen;
988 if (data->dsize == 0) {
993 data->dptr = (uint8 *)talloc_memdup(mem_ctx, &reply->data[0],
995 if (data->dptr == NULL) {
996 DEBUG(0, ("talloc failed\n"));
997 status = NT_STATUS_NO_MEMORY;
1002 status = NT_STATUS_OK;
1008 struct ctdbd_traverse_state {
1009 void (*fn)(TDB_DATA key, TDB_DATA data, void *private_data);
1014 * Handle a traverse record coming in on the ctdbd connection
1017 static NTSTATUS ctdb_traverse_handler(const struct data_blob *blob,
1020 struct ctdbd_traverse_state *state =
1021 (struct ctdbd_traverse_state *)private_data;
1023 struct ctdb_req_message *m;
1024 struct ctdb_rec_data *d;
1027 m = (struct ctdb_req_message *)blob->data;
1029 if (blob->length < sizeof(*m) || m->hdr.length != blob->length) {
1030 DEBUG(0, ("Got invalid message of length %d\n",
1031 (int)blob->length));
1032 return NT_STATUS_UNEXPECTED_IO_ERROR;
1035 d = (struct ctdb_rec_data *)&m->data[0];
1036 if (m->datalen < sizeof(uint32_t) || m->datalen != d->length) {
1037 DEBUG(0, ("Got invalid traverse data of length %d\n",
1039 return NT_STATUS_UNEXPECTED_IO_ERROR;
1042 key.dsize = d->keylen;
1043 key.dptr = &d->data[0];
1044 data.dsize = d->datalen;
1045 data.dptr = &d->data[d->keylen];
1047 if (key.dsize == 0 && data.dsize == 0) {
1048 /* end of traverse */
1049 return NT_STATUS_END_OF_FILE;
1052 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1053 DEBUG(0, ("Got invalid ltdb header length %d\n",
1055 return NT_STATUS_UNEXPECTED_IO_ERROR;
1057 data.dsize -= sizeof(struct ctdb_ltdb_header);
1058 data.dptr += sizeof(struct ctdb_ltdb_header);
1061 state->fn(key, data, state->private_data);
1064 return NT_STATUS_OK;
1068 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1069 connection to ctdbd to avoid the hairy recursive and async problems with
1073 NTSTATUS ctdbd_traverse(uint32 db_id,
1074 void (*fn)(TDB_DATA key, TDB_DATA data,
1075 void *private_data),
1078 struct ctdbd_connection *conn;
1082 struct ctdb_traverse_start t;
1084 struct ctdbd_traverse_state state;
1086 status = ctdbd_init_connection(NULL, &conn);
1089 t.srvid = conn->rand_srvid;
1090 t.reqid = ++conn->reqid;
1092 data.dptr = (uint8_t *)&t;
1093 data.dsize = sizeof(t);
1095 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1096 CTDB_CONTROL_TRAVERSE_START, conn->rand_srvid, 0,
1097 data, NULL, NULL, &cstatus);
1099 if (!NT_STATUS_IS_OK(status) || (cstatus != 0)) {
1101 DEBUG(0,("ctdbd_control failed: %s, %d\n", nt_errstr(status),
1104 if (NT_STATUS_IS_OK(status)) {
1106 * We need a mapping here
1108 status = NT_STATUS_UNSUCCESSFUL;
1114 state.private_data = private_data;
1118 status = NT_STATUS_OK;
1120 if (packet_handler(conn->pkt, ctdb_req_complete,
1121 ctdb_traverse_handler, &state, &status)) {
1123 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1124 status = NT_STATUS_OK;
1129 * There might be more in the queue
1134 if (!NT_STATUS_IS_OK(status)) {
1138 status = packet_fd_read_sync(conn->pkt);
1140 if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
1142 * There might be more in the queue
1147 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1148 status = NT_STATUS_OK;
1151 if (!NT_STATUS_IS_OK(status)) {
1152 DEBUG(0, ("packet_fd_read_sync failed: %s\n", nt_errstr(status)));
1153 cluster_fatal("ctdbd died\n");
1163 * Register us as a server for a particular tcp connection
1166 NTSTATUS ctdbd_register_ips(struct ctdbd_connection *conn,
1167 const struct sockaddr_in *server,
1168 const struct sockaddr_in *client,
1169 void (*release_ip_handler)(const char *ip_addr,
1170 void *private_data),
1173 struct ctdb_control_tcp p;
1178 * Only one connection so far
1180 SMB_ASSERT(conn->release_ip_handler == NULL);
1183 * We want to be told about IP releases
1186 status = register_with_ctdbd(conn, CTDB_SRVID_RELEASE_IP);
1187 if (!NT_STATUS_IS_OK(status)) {
1195 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1196 * can send an extra ack to trigger a reset for our client, so it
1197 * immediately reconnects
1199 data.dptr = (uint8_t *)&p;
1200 data.dsize = sizeof(p);
1202 return ctdbd_control(conn, CTDB_CURRENT_NODE,
1203 CTDB_CONTROL_TCP_CLIENT, 0,
1204 CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL, NULL);
1208 * We want to handle reconfigure events
1210 NTSTATUS ctdbd_register_reconfigure(struct ctdbd_connection *conn)
1212 return register_with_ctdbd(conn, CTDB_SRVID_RECONFIGURE);
1216 persistent call. Used to start, store or cancel persistent updates.
1218 static NTSTATUS ctdbd_persistent_call(struct ctdbd_connection *conn, uint32_t opcode, uint32_t db_id, TDB_DATA key, TDB_DATA data)
1221 struct ctdb_rec_data *rec;
1226 length = offsetof(struct ctdb_rec_data, data) + key.dsize + data.dsize;
1228 rec = (struct ctdb_rec_data *)talloc_size(conn, length);
1229 NT_STATUS_HAVE_NO_MEMORY(rec);
1231 rec->length = length;
1233 rec->keylen = key.dsize;
1234 rec->datalen= data.dsize;
1235 memcpy(&rec->data[0], key.dptr, key.dsize);
1236 memcpy(&rec->data[key.dsize], data.dptr, data.dsize);
1238 recdata.dptr = (uint8_t *)rec;
1239 recdata.dsize = length;
1241 status = ctdbd_control(conn, CTDB_CURRENT_NODE, opcode,
1242 0, 0, recdata, NULL, NULL, &cstatus);
1244 return NT_STATUS_INTERNAL_DB_CORRUPTION;
1250 persistent store. Used when we update a record in a persistent database
1252 NTSTATUS ctdbd_persistent_store(struct ctdbd_connection *conn, uint32_t db_id, TDB_DATA key, TDB_DATA data)
1254 return ctdbd_persistent_call(conn,
1255 CTDB_CONTROL_PERSISTENT_STORE,
1260 tell the ctdb daemon that we are starting a persistent update operation.
1261 If we terminate/disconnect from the daemon without first performing
1262 either a persistent_store or a cancel ctdbd will perform recovery.
1264 NTSTATUS ctdbd_start_persistent_update(struct ctdbd_connection *conn, uint32_t db_id, TDB_DATA key, TDB_DATA data)
1266 return ctdbd_persistent_call(conn,
1267 CTDB_CONTROL_START_PERSISTENT_UPDATE,
1273 Cancel a persistent update operation. This is used if we have started a
1274 persistent update but we want to abort it before we have made changes to
1277 NTSTATUS ctdbd_cancel_persistent_update(struct ctdbd_connection *conn, uint32_t db_id, TDB_DATA key, TDB_DATA data)
1279 return ctdbd_persistent_call(conn,
1280 CTDB_CONTROL_CANCEL_PERSISTENT_UPDATE,
1285 call a control on the local node
1287 NTSTATUS ctdbd_control_local(struct ctdbd_connection *conn, uint32 opcode,
1288 uint64_t srvid, uint32_t flags, TDB_DATA data,
1289 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
1292 return ctdbd_control(conn, CTDB_CURRENT_NODE, opcode, srvid, flags, data, mem_ctx, outdata, cstatus);
1300 NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
1301 struct ctdbd_connection **pconn)
1303 return NT_STATUS_NOT_IMPLEMENTED;