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 uint8_t *buf, size_t available,
164 if (available < sizeof(msglen)) {
168 msglen = *((uint32 *)buf);
170 DEBUG(10, ("msglen = %d\n", msglen));
172 if (msglen < sizeof(struct ctdb_req_header)) {
173 DEBUG(0, ("Got invalid msglen: %d, expected at least %d for "
174 "the req_header\n", (int)msglen,
175 (int)sizeof(struct ctdb_req_header)));
176 cluster_fatal("ctdbd protocol error\n");
179 if (available < msglen) {
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,
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(uint8_t *buf, size_t length,
226 struct req_pull_state *state = (struct req_pull_state *)private_data;
228 state->req.data = talloc_move(state->mem_ctx, &buf);
229 state->req.length = length;
234 * Fetch a messaging_rec from an incoming ctdb style message
237 static struct messaging_rec *ctdb_pull_messaging_rec(TALLOC_CTX *mem_ctx,
238 size_t overall_length,
239 struct ctdb_req_message *msg)
241 struct messaging_rec *result;
243 enum ndr_err_code ndr_err;
245 if ((overall_length < offsetof(struct ctdb_req_message, data))
247 < offsetof(struct ctdb_req_message, data) + msg->datalen)) {
249 cluster_fatal("got invalid msg length");
252 if (!(result = TALLOC_P(mem_ctx, struct messaging_rec))) {
253 DEBUG(0, ("talloc failed\n"));
257 blob = data_blob_const(msg->data, msg->datalen);
259 ndr_err = ndr_pull_struct_blob(
260 &blob, result, NULL, result,
261 (ndr_pull_flags_fn_t)ndr_pull_messaging_rec);
263 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
264 DEBUG(0, ("ndr_pull_struct_blob failed: %s\n",
265 ndr_errstr(ndr_err)));
270 if (DEBUGLEVEL >= 10) {
271 DEBUG(10, ("ctdb_pull_messaging_rec:\n"));
272 NDR_PRINT_DEBUG(messaging_rec, result);
278 static NTSTATUS ctdb_packet_fd_read_sync(struct packet_context *ctx)
280 struct timeval timeout;
281 struct timeval *ptimeout;
283 timeout = timeval_set(lp_ctdb_timeout(), 0);
284 ptimeout = (timeout.tv_sec != 0) ? &timeout : NULL;
286 return packet_fd_read_sync(ctx, ptimeout);
290 * Read a full ctdbd request. If we have a messaging context, defer incoming
291 * messages that might come in between.
294 static NTSTATUS ctdb_read_req(struct ctdbd_connection *conn, uint32 reqid,
295 TALLOC_CTX *mem_ctx, void *result)
297 struct ctdb_req_header *hdr;
298 struct req_pull_state state;
303 status = ctdb_packet_fd_read_sync(conn->pkt);
305 if (NT_STATUS_EQUAL(status, NT_STATUS_NETWORK_BUSY)) {
308 } else if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
313 if (!NT_STATUS_IS_OK(status)) {
314 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
315 cluster_fatal("ctdbd died\n");
321 state.mem_ctx = mem_ctx;
323 if (!packet_handler(conn->pkt, ctdb_req_complete, ctdb_req_pull,
328 DEBUG(10, ("not enough data from ctdb socket, retrying\n"));
332 if (!NT_STATUS_IS_OK(status)) {
333 DEBUG(0, ("Could not read packet: %s\n", nt_errstr(status)));
334 cluster_fatal("ctdbd died\n");
337 hdr = (struct ctdb_req_header *)state.req.data;
339 DEBUG(10, ("Received ctdb packet\n"));
340 ctdb_packet_dump(hdr);
342 if (hdr->operation == CTDB_REQ_MESSAGE) {
343 struct timed_event *evt;
344 struct deferred_msg_state *msg_state;
345 struct ctdb_req_message *msg = (struct ctdb_req_message *)hdr;
347 if (conn->msg_ctx == NULL) {
348 DEBUG(1, ("Got a message without having a msg ctx, "
349 "dropping msg %llu\n",
350 (long long unsigned)msg->srvid));
354 if ((conn->release_ip_handler != NULL)
355 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
356 /* must be dispatched immediately */
357 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
358 conn->release_ip_handler((const char *)msg->data,
359 conn->release_ip_priv);
364 if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
365 || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)) {
367 DEBUG(1, ("ctdb_read_req: Got %s message\n",
368 (msg->srvid == CTDB_SRVID_RECONFIGURE)
369 ? "cluster reconfigure" : "SAMBA_NOTIFY"));
371 messaging_send(conn->msg_ctx, procid_self(),
372 MSG_SMB_BRL_VALIDATE, &data_blob_null);
373 messaging_send(conn->msg_ctx, procid_self(),
374 MSG_DBWRAP_G_LOCK_RETRY,
380 if (!(msg_state = TALLOC_P(talloc_autofree_context(), struct deferred_msg_state))) {
381 DEBUG(0, ("talloc failed\n"));
386 if (!(msg_state->rec = ctdb_pull_messaging_rec(
387 msg_state, state.req.length, msg))) {
388 DEBUG(0, ("ctdbd_pull_messaging_rec failed\n"));
389 TALLOC_FREE(msg_state);
396 msg_state->msg_ctx = conn->msg_ctx;
399 * We're waiting for a call reply, but an async message has
400 * crossed. Defer dispatching to the toplevel event loop.
402 evt = event_add_timed(conn->msg_ctx->event_ctx,
403 conn->msg_ctx->event_ctx,
405 deferred_message_dispatch,
408 DEBUG(0, ("event_add_timed failed\n"));
409 TALLOC_FREE(msg_state);
417 if (hdr->reqid != reqid) {
418 /* we got the wrong reply */
419 DEBUG(0,("Discarding mismatched ctdb reqid %u should have "
420 "been %u\n", hdr->reqid, reqid));
425 *((void **)result) = talloc_move(mem_ctx, &hdr);
431 * Get us a ctdbd connection
434 NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
435 struct ctdbd_connection **pconn)
437 struct ctdbd_connection *conn;
440 if (!(conn = TALLOC_ZERO_P(mem_ctx, struct ctdbd_connection))) {
441 DEBUG(0, ("talloc failed\n"));
442 return NT_STATUS_NO_MEMORY;
445 status = ctdbd_connect(conn, &conn->pkt);
447 if (!NT_STATUS_IS_OK(status)) {
448 DEBUG(10, ("ctdbd_connect failed: %s\n", nt_errstr(status)));
452 status = get_cluster_vnn(conn, &conn->our_vnn);
454 if (!NT_STATUS_IS_OK(status)) {
455 DEBUG(10, ("get_cluster_vnn failed: %s\n", nt_errstr(status)));
459 generate_random_buffer((unsigned char *)&conn->rand_srvid,
460 sizeof(conn->rand_srvid));
462 status = register_with_ctdbd(conn, conn->rand_srvid);
464 if (!NT_STATUS_IS_OK(status)) {
465 DEBUG(5, ("Could not register random srvid: %s\n",
479 * Get us a ctdbd connection and register us as a process
482 NTSTATUS ctdbd_messaging_connection(TALLOC_CTX *mem_ctx,
483 struct ctdbd_connection **pconn)
485 struct ctdbd_connection *conn;
488 status = ctdbd_init_connection(mem_ctx, &conn);
490 if (!NT_STATUS_IS_OK(status)) {
494 status = register_with_ctdbd(conn, (uint64_t)sys_getpid());
495 if (!NT_STATUS_IS_OK(status)) {
499 status = register_with_ctdbd(conn, MSG_SRVID_SAMBA);
500 if (!NT_STATUS_IS_OK(status)) {
504 status = register_with_ctdbd(conn, CTDB_SRVID_SAMBA_NOTIFY);
505 if (!NT_STATUS_IS_OK(status)) {
518 * Packet handler to receive and handle a ctdb message
520 static NTSTATUS ctdb_handle_message(uint8_t *buf, size_t length,
523 struct ctdbd_connection *conn = talloc_get_type_abort(
524 private_data, struct ctdbd_connection);
525 struct ctdb_req_message *msg;
526 struct messaging_rec *msg_rec;
528 msg = (struct ctdb_req_message *)buf;
530 if (msg->hdr.operation != CTDB_REQ_MESSAGE) {
531 DEBUG(0, ("Received async msg of type %u, discarding\n",
532 msg->hdr.operation));
534 return NT_STATUS_INVALID_PARAMETER;
537 if ((conn->release_ip_handler != NULL)
538 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
539 /* must be dispatched immediately */
540 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
541 conn->release_ip_handler((const char *)msg->data,
542 conn->release_ip_priv);
547 SMB_ASSERT(conn->msg_ctx != NULL);
549 if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
550 || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)){
551 DEBUG(0,("Got cluster reconfigure message\n"));
553 * when the cluster is reconfigured or someone of the
554 * family has passed away (SAMBA_NOTIFY), we need to
555 * clean the brl database
557 messaging_send(conn->msg_ctx, procid_self(),
558 MSG_SMB_BRL_VALIDATE, &data_blob_null);
560 messaging_send(conn->msg_ctx, procid_self(),
561 MSG_DBWRAP_G_LOCK_RETRY,
568 /* only messages to our pid or the broadcast are valid here */
569 if (msg->srvid != sys_getpid() && msg->srvid != MSG_SRVID_SAMBA) {
570 DEBUG(0,("Got unexpected message with srvid=%llu\n",
571 (unsigned long long)msg->srvid));
576 if (!(msg_rec = ctdb_pull_messaging_rec(NULL, length, msg))) {
577 DEBUG(10, ("ctdb_pull_messaging_rec failed\n"));
579 return NT_STATUS_NO_MEMORY;
582 messaging_dispatch_rec(conn->msg_ctx, msg_rec);
584 TALLOC_FREE(msg_rec);
590 * The ctdbd socket is readable asynchronuously
593 static void ctdbd_socket_handler(struct event_context *event_ctx,
594 struct fd_event *event,
598 struct ctdbd_connection *conn = talloc_get_type_abort(
599 private_data, struct ctdbd_connection);
603 status = packet_fd_read(conn->pkt);
605 if (!NT_STATUS_IS_OK(status)) {
606 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
607 cluster_fatal("ctdbd died\n");
610 while (packet_handler(conn->pkt, ctdb_req_complete,
611 ctdb_handle_message, conn, &status)) {
612 if (!NT_STATUS_IS_OK(status)) {
613 DEBUG(10, ("could not handle incoming message: %s\n",
620 * Prepare a ctdbd connection to receive messages
623 NTSTATUS ctdbd_register_msg_ctx(struct ctdbd_connection *conn,
624 struct messaging_context *msg_ctx)
626 SMB_ASSERT(conn->msg_ctx == NULL);
627 SMB_ASSERT(conn->fde == NULL);
629 if (!(conn->fde = event_add_fd(msg_ctx->event_ctx, conn,
630 packet_get_fd(conn->pkt),
632 ctdbd_socket_handler,
634 DEBUG(0, ("event_add_fd failed\n"));
635 return NT_STATUS_NO_MEMORY;
638 conn->msg_ctx = msg_ctx;
644 * Send a messaging message across a ctdbd
647 NTSTATUS ctdbd_messaging_send(struct ctdbd_connection *conn,
648 uint32 dst_vnn, uint64 dst_srvid,
649 struct messaging_rec *msg)
651 struct ctdb_req_message r;
655 enum ndr_err_code ndr_err;
657 if (!(mem_ctx = talloc_init("ctdbd_messaging_send"))) {
658 DEBUG(0, ("talloc failed\n"));
659 return NT_STATUS_NO_MEMORY;
662 ndr_err = ndr_push_struct_blob(
663 &blob, mem_ctx, NULL, msg,
664 (ndr_push_flags_fn_t)ndr_push_messaging_rec);
666 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
667 DEBUG(0, ("ndr_push_struct_blob failed: %s\n",
668 ndr_errstr(ndr_err)));
669 status = ndr_map_error2ntstatus(ndr_err);
673 r.hdr.length = offsetof(struct ctdb_req_message, data) + blob.length;
674 r.hdr.ctdb_magic = CTDB_MAGIC;
675 r.hdr.ctdb_version = CTDB_VERSION;
676 r.hdr.generation = 1;
677 r.hdr.operation = CTDB_REQ_MESSAGE;
678 r.hdr.destnode = dst_vnn;
679 r.hdr.srcnode = conn->our_vnn;
682 r.datalen = blob.length;
684 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
685 ctdb_packet_dump(&r.hdr);
687 status = packet_send(
689 data_blob_const(&r, offsetof(struct ctdb_req_message, data)),
692 if (!NT_STATUS_IS_OK(status)) {
693 DEBUG(0, ("packet_send failed: %s\n", nt_errstr(status)));
697 status = packet_flush(conn->pkt);
699 if (!NT_STATUS_IS_OK(status)) {
700 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
701 cluster_fatal("cluster dispatch daemon msg write error\n");
704 status = NT_STATUS_OK;
706 TALLOC_FREE(mem_ctx);
711 * send/recv a generic ctdb control message
713 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
714 uint32_t vnn, uint32 opcode,
715 uint64_t srvid, uint32_t flags,
717 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
720 struct ctdb_req_control req;
721 struct ctdb_reply_control *reply = NULL;
722 struct ctdbd_connection *new_conn = NULL;
725 /* the samba3 ctdb code can't handle NOREPLY yet */
726 flags &= ~CTDB_CTRL_FLAG_NOREPLY;
729 status = ctdbd_init_connection(NULL, &new_conn);
731 if (!NT_STATUS_IS_OK(status)) {
732 DEBUG(10, ("Could not init temp connection: %s\n",
741 req.hdr.length = offsetof(struct ctdb_req_control, data) + data.dsize;
742 req.hdr.ctdb_magic = CTDB_MAGIC;
743 req.hdr.ctdb_version = CTDB_VERSION;
744 req.hdr.operation = CTDB_REQ_CONTROL;
745 req.hdr.reqid = ++conn->reqid;
746 req.hdr.destnode = vnn;
749 req.datalen = data.dsize;
751 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
752 ctdb_packet_dump(&req.hdr);
754 status = packet_send(
756 data_blob_const(&req, offsetof(struct ctdb_req_control, data)),
757 data_blob_const(data.dptr, data.dsize));
759 if (!NT_STATUS_IS_OK(status)) {
760 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
764 status = packet_flush(conn->pkt);
766 if (!NT_STATUS_IS_OK(status)) {
767 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
768 cluster_fatal("cluster dispatch daemon control write error\n");
771 if (flags & CTDB_CTRL_FLAG_NOREPLY) {
772 TALLOC_FREE(new_conn);
776 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
778 if (!NT_STATUS_IS_OK(status)) {
779 DEBUG(10, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
783 if (reply->hdr.operation != CTDB_REPLY_CONTROL) {
784 DEBUG(0, ("received invalid reply\n"));
789 if (!(outdata->dptr = (uint8 *)talloc_memdup(
790 mem_ctx, reply->data, reply->datalen))) {
792 return NT_STATUS_NO_MEMORY;
794 outdata->dsize = reply->datalen;
797 (*cstatus) = reply->status;
800 status = NT_STATUS_OK;
803 TALLOC_FREE(new_conn);
809 * see if a remote process exists
811 bool ctdbd_process_exists(struct ctdbd_connection *conn, uint32 vnn, pid_t pid)
817 data.dptr = (uint8_t*)&pid;
818 data.dsize = sizeof(pid);
820 status = ctdbd_control(conn, vnn, CTDB_CONTROL_PROCESS_EXISTS, 0, 0,
821 data, NULL, NULL, &cstatus);
822 if (!NT_STATUS_IS_OK(status)) {
823 DEBUG(0, (__location__ " ctdb_control for process_exists "
834 char *ctdbd_dbpath(struct ctdbd_connection *conn,
835 TALLOC_CTX *mem_ctx, uint32_t db_id)
841 data.dptr = (uint8_t*)&db_id;
842 data.dsize = sizeof(db_id);
844 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
845 CTDB_CONTROL_GETDBPATH, 0, 0, data,
846 mem_ctx, &data, &cstatus);
847 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
848 DEBUG(0,(__location__ " ctdb_control for getdbpath failed\n"));
852 return (char *)data.dptr;
856 * attach to a ctdb database
858 NTSTATUS ctdbd_db_attach(struct ctdbd_connection *conn,
859 const char *name, uint32_t *db_id, int tdb_flags)
864 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
866 data.dptr = (uint8_t*)name;
867 data.dsize = strlen(name)+1;
869 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
871 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
872 : CTDB_CONTROL_DB_ATTACH,
873 0, 0, data, NULL, &data, &cstatus);
874 if (!NT_STATUS_IS_OK(status)) {
875 DEBUG(0, (__location__ " ctdb_control for db_attach "
876 "failed: %s\n", nt_errstr(status)));
880 if (cstatus != 0 || data.dsize != sizeof(uint32_t)) {
881 DEBUG(0,(__location__ " ctdb_control for db_attach failed\n"));
882 return NT_STATUS_INTERNAL_ERROR;
885 *db_id = *(uint32_t *)data.dptr;
886 talloc_free(data.dptr);
888 if (!(tdb_flags & TDB_SEQNUM)) {
892 data.dptr = (uint8_t *)db_id;
893 data.dsize = sizeof(*db_id);
895 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
896 CTDB_CONTROL_ENABLE_SEQNUM, 0, 0, data,
897 NULL, NULL, &cstatus);
898 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
899 DEBUG(0,(__location__ " ctdb_control for enable seqnum "
901 return NT_STATUS_IS_OK(status) ? NT_STATUS_INTERNAL_ERROR :
909 * force the migration of a record to this node
911 NTSTATUS ctdbd_migrate(struct ctdbd_connection *conn, uint32 db_id,
914 struct ctdb_req_call req;
915 struct ctdb_reply_call *reply;
920 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
921 req.hdr.ctdb_magic = CTDB_MAGIC;
922 req.hdr.ctdb_version = CTDB_VERSION;
923 req.hdr.operation = CTDB_REQ_CALL;
924 req.hdr.reqid = ++conn->reqid;
925 req.flags = CTDB_IMMEDIATE_MIGRATION;
926 req.callid = CTDB_NULL_FUNC;
928 req.keylen = key.dsize;
930 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
931 ctdb_packet_dump(&req.hdr);
933 status = packet_send(
935 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
936 data_blob_const(key.dptr, key.dsize));
938 if (!NT_STATUS_IS_OK(status)) {
939 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
943 status = packet_flush(conn->pkt);
945 if (!NT_STATUS_IS_OK(status)) {
946 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
947 cluster_fatal("cluster dispatch daemon control write error\n");
950 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
952 if (!NT_STATUS_IS_OK(status)) {
953 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
957 if (reply->hdr.operation != CTDB_REPLY_CALL) {
958 DEBUG(0, ("received invalid reply\n"));
959 status = NT_STATUS_INTERNAL_ERROR;
963 status = NT_STATUS_OK;
971 * remotely fetch a record without locking it or forcing a migration
973 NTSTATUS ctdbd_fetch(struct ctdbd_connection *conn, uint32 db_id,
974 TDB_DATA key, TALLOC_CTX *mem_ctx, TDB_DATA *data)
976 struct ctdb_req_call req;
977 struct ctdb_reply_call *reply;
982 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
983 req.hdr.ctdb_magic = CTDB_MAGIC;
984 req.hdr.ctdb_version = CTDB_VERSION;
985 req.hdr.operation = CTDB_REQ_CALL;
986 req.hdr.reqid = ++conn->reqid;
988 req.callid = CTDB_FETCH_FUNC;
990 req.keylen = key.dsize;
992 status = packet_send(
994 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
995 data_blob_const(key.dptr, key.dsize));
997 if (!NT_STATUS_IS_OK(status)) {
998 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
1002 status = packet_flush(conn->pkt);
1004 if (!NT_STATUS_IS_OK(status)) {
1005 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
1006 cluster_fatal("cluster dispatch daemon control write error\n");
1009 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
1011 if (!NT_STATUS_IS_OK(status)) {
1012 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
1016 if (reply->hdr.operation != CTDB_REPLY_CALL) {
1017 DEBUG(0, ("received invalid reply\n"));
1018 status = NT_STATUS_INTERNAL_ERROR;
1022 data->dsize = reply->datalen;
1023 if (data->dsize == 0) {
1028 data->dptr = (uint8 *)talloc_memdup(mem_ctx, &reply->data[0],
1030 if (data->dptr == NULL) {
1031 DEBUG(0, ("talloc failed\n"));
1032 status = NT_STATUS_NO_MEMORY;
1037 status = NT_STATUS_OK;
1043 struct ctdbd_traverse_state {
1044 void (*fn)(TDB_DATA key, TDB_DATA data, void *private_data);
1049 * Handle a traverse record coming in on the ctdbd connection
1052 static NTSTATUS ctdb_traverse_handler(uint8_t *buf, size_t length,
1055 struct ctdbd_traverse_state *state =
1056 (struct ctdbd_traverse_state *)private_data;
1058 struct ctdb_req_message *m;
1059 struct ctdb_rec_data *d;
1062 m = (struct ctdb_req_message *)buf;
1064 if (length < sizeof(*m) || m->hdr.length != length) {
1065 DEBUG(0, ("Got invalid message of length %d\n", (int)length));
1067 return NT_STATUS_UNEXPECTED_IO_ERROR;
1070 d = (struct ctdb_rec_data *)&m->data[0];
1071 if (m->datalen < sizeof(uint32_t) || m->datalen != d->length) {
1072 DEBUG(0, ("Got invalid traverse data of length %d\n",
1075 return NT_STATUS_UNEXPECTED_IO_ERROR;
1078 key.dsize = d->keylen;
1079 key.dptr = &d->data[0];
1080 data.dsize = d->datalen;
1081 data.dptr = &d->data[d->keylen];
1083 if (key.dsize == 0 && data.dsize == 0) {
1084 /* end of traverse */
1085 return NT_STATUS_END_OF_FILE;
1088 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1089 DEBUG(0, ("Got invalid ltdb header length %d\n",
1092 return NT_STATUS_UNEXPECTED_IO_ERROR;
1094 data.dsize -= sizeof(struct ctdb_ltdb_header);
1095 data.dptr += sizeof(struct ctdb_ltdb_header);
1098 state->fn(key, data, state->private_data);
1102 return NT_STATUS_OK;
1106 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1107 connection to ctdbd to avoid the hairy recursive and async problems with
1111 NTSTATUS ctdbd_traverse(uint32 db_id,
1112 void (*fn)(TDB_DATA key, TDB_DATA data,
1113 void *private_data),
1116 struct ctdbd_connection *conn;
1120 struct ctdb_traverse_start t;
1122 struct ctdbd_traverse_state state;
1124 status = ctdbd_init_connection(NULL, &conn);
1125 if (!NT_STATUS_IS_OK(status)) {
1126 DEBUG(0, ("ctdbd_init_connection failed: %s\n",
1127 nt_errstr(status)));
1132 t.srvid = conn->rand_srvid;
1133 t.reqid = ++conn->reqid;
1135 data.dptr = (uint8_t *)&t;
1136 data.dsize = sizeof(t);
1138 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1139 CTDB_CONTROL_TRAVERSE_START, conn->rand_srvid, 0,
1140 data, NULL, NULL, &cstatus);
1142 if (!NT_STATUS_IS_OK(status) || (cstatus != 0)) {
1144 DEBUG(0,("ctdbd_control failed: %s, %d\n", nt_errstr(status),
1147 if (NT_STATUS_IS_OK(status)) {
1149 * We need a mapping here
1151 status = NT_STATUS_UNSUCCESSFUL;
1157 state.private_data = private_data;
1161 status = NT_STATUS_OK;
1163 if (packet_handler(conn->pkt, ctdb_req_complete,
1164 ctdb_traverse_handler, &state, &status)) {
1166 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1167 status = NT_STATUS_OK;
1172 * There might be more in the queue
1177 if (!NT_STATUS_IS_OK(status)) {
1181 status = ctdb_packet_fd_read_sync(conn->pkt);
1183 if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
1185 * There might be more in the queue
1190 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1191 status = NT_STATUS_OK;
1195 if (!NT_STATUS_IS_OK(status)) {
1196 DEBUG(0, ("packet_fd_read_sync failed: %s\n", nt_errstr(status)));
1197 cluster_fatal("ctdbd died\n");
1207 This is used to canonicalize a ctdb_sock_addr structure.
1209 static void smbd_ctdb_canonicalize_ip(const struct sockaddr_storage *in,
1210 struct sockaddr_storage *out)
1212 memcpy(out, in, sizeof (*out));
1215 if (in->ss_family == AF_INET6) {
1216 const char prefix[12] = { 0,0,0,0,0,0,0,0,0,0,0xff,0xff };
1217 const struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)in;
1218 struct sockaddr_in *out4 = (struct sockaddr_in *)out;
1219 if (memcmp(&in6->sin6_addr, prefix, 12) == 0) {
1220 memset(out, 0, sizeof(*out));
1221 #ifdef HAVE_SOCK_SIN_LEN
1222 out4->sin_len = sizeof(*out);
1224 out4->sin_family = AF_INET;
1225 out4->sin_port = in6->sin6_port;
1226 memcpy(&out4->sin_addr, &in6->sin6_addr.s6_addr32[3], 4);
1233 * Register us as a server for a particular tcp connection
1236 NTSTATUS ctdbd_register_ips(struct ctdbd_connection *conn,
1237 const struct sockaddr_storage *_server,
1238 const struct sockaddr_storage *_client,
1239 void (*release_ip_handler)(const char *ip_addr,
1240 void *private_data),
1244 * we still use ctdb_control_tcp for ipv4
1245 * because we want to work against older ctdb
1246 * versions at runtime
1248 struct ctdb_control_tcp p4;
1249 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1250 struct ctdb_control_tcp_addr p;
1254 struct sockaddr_storage client;
1255 struct sockaddr_storage server;
1258 * Only one connection so far
1260 SMB_ASSERT(conn->release_ip_handler == NULL);
1262 smbd_ctdb_canonicalize_ip(_client, &client);
1263 smbd_ctdb_canonicalize_ip(_server, &server);
1265 switch (client.ss_family) {
1267 p4.dest = *(struct sockaddr_in *)&server;
1268 p4.src = *(struct sockaddr_in *)&client;
1269 data.dptr = (uint8_t *)&p4;
1270 data.dsize = sizeof(p4);
1272 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1274 p.dest.ip6 = *(struct sockaddr_in6 *)&server;
1275 p.src.ip6 = *(struct sockaddr_in6 *)&client;
1276 data.dptr = (uint8_t *)&p;
1277 data.dsize = sizeof(p);
1281 return NT_STATUS_INTERNAL_ERROR;
1284 conn->release_ip_handler = release_ip_handler;
1287 * We want to be told about IP releases
1290 status = register_with_ctdbd(conn, CTDB_SRVID_RELEASE_IP);
1291 if (!NT_STATUS_IS_OK(status)) {
1296 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1297 * can send an extra ack to trigger a reset for our client, so it
1298 * immediately reconnects
1300 return ctdbd_control(conn, CTDB_CURRENT_NODE,
1301 CTDB_CONTROL_TCP_CLIENT, 0,
1302 CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL, NULL);
1306 * We want to handle reconfigure events
1308 NTSTATUS ctdbd_register_reconfigure(struct ctdbd_connection *conn)
1310 return register_with_ctdbd(conn, CTDB_SRVID_RECONFIGURE);
1314 call a control on the local node
1316 NTSTATUS ctdbd_control_local(struct ctdbd_connection *conn, uint32 opcode,
1317 uint64_t srvid, uint32_t flags, TDB_DATA data,
1318 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
1321 return ctdbd_control(conn, CTDB_CURRENT_NODE, opcode, srvid, flags, data, mem_ctx, outdata, cstatus);
1324 NTSTATUS ctdb_watch_us(struct ctdbd_connection *conn)
1326 struct ctdb_client_notify_register reg_data;
1331 reg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1333 reg_data.notify_data[0] = 0;
1335 struct_len = offsetof(struct ctdb_client_notify_register,
1336 notify_data) + reg_data.len;
1338 status = ctdbd_control_local(
1339 conn, CTDB_CONTROL_REGISTER_NOTIFY, conn->rand_srvid, 0,
1340 make_tdb_data((uint8_t *)®_data, struct_len),
1341 NULL, NULL, &cstatus);
1342 if (!NT_STATUS_IS_OK(status)) {
1343 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1344 nt_errstr(status)));
1349 NTSTATUS ctdb_unwatch(struct ctdbd_connection *conn)
1351 struct ctdb_client_notify_deregister dereg_data;
1355 dereg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1357 status = ctdbd_control_local(
1358 conn, CTDB_CONTROL_DEREGISTER_NOTIFY, conn->rand_srvid, 0,
1359 make_tdb_data((uint8_t *)&dereg_data, sizeof(dereg_data)),
1360 NULL, NULL, &cstatus);
1361 if (!NT_STATUS_IS_OK(status)) {
1362 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1363 nt_errstr(status)));
1370 NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
1371 struct ctdbd_connection **pconn)
1373 return NT_STATUS_NOT_IMPLEMENTED;