s4:rootdse LDB module - remove unused variable
[Samba/gebeck_regimport.git] / source3 / lib / ctdbd_conn.c
blobdee477f3c525f2924d88e6a2e37fc2e5f2b0018b
1 /*
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/>.
21 #include "includes.h"
23 #ifdef CLUSTER_SUPPORT
25 #include "librpc/gen_ndr/messaging.h"
26 #include "librpc/gen_ndr/ndr_messaging.h"
27 #include "ctdbd_conn.h"
29 /* paths to these include files come from --with-ctdb= in configure */
30 #include "ctdb.h"
31 #include "ctdb_private.h"
33 struct ctdbd_connection {
34 struct messaging_context *msg_ctx;
35 uint32 reqid;
36 uint32 our_vnn;
37 uint64 rand_srvid;
38 struct packet_context *pkt;
39 struct fd_event *fde;
41 void (*release_ip_handler)(const char *ip_addr, void *private_data);
42 void *release_ip_priv;
45 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
46 uint32_t vnn, uint32 opcode,
47 uint64_t srvid, uint32_t flags, TDB_DATA data,
48 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
49 int *cstatus);
52 * exit on fatal communications errors with the ctdbd daemon
54 static void cluster_fatal(const char *why)
56 DEBUG(0,("cluster fatal event: %s - exiting immediately\n", why));
57 /* we don't use smb_panic() as we don't want to delay to write
58 a core file. We need to release this process id immediately
59 so that someone else can take over without getting sharing
60 violations */
61 _exit(1);
67 static void ctdb_packet_dump(struct ctdb_req_header *hdr)
69 if (DEBUGLEVEL < 10) {
70 return;
72 DEBUGADD(10, ("len=%d, magic=%x, vers=%d, gen=%d, op=%d, reqid=%d\n",
73 (int)hdr->length, (int)hdr->ctdb_magic,
74 (int)hdr->ctdb_version, (int)hdr->generation,
75 (int)hdr->operation, (int)hdr->reqid));
79 * Register a srvid with ctdbd
81 static NTSTATUS register_with_ctdbd(struct ctdbd_connection *conn,
82 uint64_t srvid)
85 int cstatus;
86 return ctdbd_control(conn, CTDB_CURRENT_NODE,
87 CTDB_CONTROL_REGISTER_SRVID, srvid, 0,
88 tdb_null, NULL, NULL, &cstatus);
92 * get our vnn from the cluster
94 static NTSTATUS get_cluster_vnn(struct ctdbd_connection *conn, uint32 *vnn)
96 int32_t cstatus=-1;
97 NTSTATUS status;
98 status = ctdbd_control(conn,
99 CTDB_CURRENT_NODE, CTDB_CONTROL_GET_PNN, 0, 0,
100 tdb_null, NULL, NULL, &cstatus);
101 if (!NT_STATUS_IS_OK(status)) {
102 cluster_fatal("ctdbd_control failed\n");
104 *vnn = (uint32_t)cstatus;
105 return status;
109 * Are we active (i.e. not banned or stopped?)
111 static bool ctdbd_working(struct ctdbd_connection *conn, uint32_t vnn)
113 int32_t cstatus=-1;
114 NTSTATUS status;
115 TDB_DATA outdata;
116 struct ctdb_node_map *m;
117 uint32_t failure_flags;
118 bool ret = false;
119 int i;
121 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
122 CTDB_CONTROL_GET_NODEMAP, 0, 0,
123 tdb_null, talloc_tos(), &outdata, &cstatus);
124 if (!NT_STATUS_IS_OK(status)) {
125 cluster_fatal("ctdbd_control failed\n");
127 if ((cstatus != 0) || (outdata.dptr == NULL)) {
128 DEBUG(2, ("Received invalid ctdb data\n"));
129 return false;
132 m = (struct ctdb_node_map *)outdata.dptr;
134 for (i=0; i<m->num; i++) {
135 if (vnn == m->nodes[i].pnn) {
136 break;
140 if (i == m->num) {
141 DEBUG(2, ("Did not find ourselves (node %d) in nodemap\n",
142 (int)vnn));
143 goto fail;
146 failure_flags = NODE_FLAGS_BANNED | NODE_FLAGS_DISCONNECTED
147 | NODE_FLAGS_PERMANENTLY_DISABLED | NODE_FLAGS_STOPPED;
149 if ((m->nodes[i].flags & failure_flags) != 0) {
150 DEBUG(2, ("Node has status %x, not active\n",
151 (int)m->nodes[i].flags));
152 goto fail;
155 ret = true;
156 fail:
157 TALLOC_FREE(outdata.dptr);
158 return ret;;
161 uint32 ctdbd_vnn(const struct ctdbd_connection *conn)
163 return conn->our_vnn;
167 * Get us a ctdb connection
170 static NTSTATUS ctdbd_connect(TALLOC_CTX *mem_ctx,
171 struct packet_context **presult)
173 struct packet_context *result;
174 const char *sockname = lp_ctdbd_socket();
175 struct sockaddr_un addr;
176 int fd;
178 if (!sockname || !*sockname) {
179 sockname = CTDB_PATH;
182 fd = socket(AF_UNIX, SOCK_STREAM, 0);
183 if (fd == -1) {
184 DEBUG(3, ("Could not create socket: %s\n", strerror(errno)));
185 return map_nt_error_from_unix(errno);
188 ZERO_STRUCT(addr);
189 addr.sun_family = AF_UNIX;
190 strncpy(addr.sun_path, sockname, sizeof(addr.sun_path));
192 if (sys_connect(fd, (struct sockaddr *)(void *)&addr) == -1) {
193 DEBUG(1, ("connect(%s) failed: %s\n", sockname,
194 strerror(errno)));
195 close(fd);
196 return map_nt_error_from_unix(errno);
199 if (!(result = packet_init(mem_ctx, fd))) {
200 close(fd);
201 return NT_STATUS_NO_MEMORY;
204 *presult = result;
205 return NT_STATUS_OK;
209 * Do we have a complete ctdb packet in the queue?
212 static bool ctdb_req_complete(const uint8_t *buf, size_t available,
213 size_t *length,
214 void *private_data)
216 uint32 msglen;
218 if (available < sizeof(msglen)) {
219 return False;
222 msglen = *((uint32 *)buf);
224 DEBUG(10, ("msglen = %d\n", msglen));
226 if (msglen < sizeof(struct ctdb_req_header)) {
227 DEBUG(0, ("Got invalid msglen: %d, expected at least %d for "
228 "the req_header\n", (int)msglen,
229 (int)sizeof(struct ctdb_req_header)));
230 cluster_fatal("ctdbd protocol error\n");
233 if (available < msglen) {
234 return false;
237 *length = msglen;
238 return true;
242 * State necessary to defer an incoming message while we are waiting for a
243 * ctdb reply.
246 struct deferred_msg_state {
247 struct messaging_context *msg_ctx;
248 struct messaging_rec *rec;
252 * Timed event handler for the deferred message
255 static void deferred_message_dispatch(struct event_context *event_ctx,
256 struct timed_event *te,
257 struct timeval now,
258 void *private_data)
260 struct deferred_msg_state *state = talloc_get_type_abort(
261 private_data, struct deferred_msg_state);
263 messaging_dispatch_rec(state->msg_ctx, state->rec);
264 TALLOC_FREE(state);
265 TALLOC_FREE(te);
268 struct req_pull_state {
269 TALLOC_CTX *mem_ctx;
270 DATA_BLOB req;
274 * Pull a ctdb request out of the incoming packet queue
277 static NTSTATUS ctdb_req_pull(uint8_t *buf, size_t length,
278 void *private_data)
280 struct req_pull_state *state = (struct req_pull_state *)private_data;
282 state->req.data = talloc_move(state->mem_ctx, &buf);
283 state->req.length = length;
284 return NT_STATUS_OK;
288 * Fetch a messaging_rec from an incoming ctdb style message
291 static struct messaging_rec *ctdb_pull_messaging_rec(TALLOC_CTX *mem_ctx,
292 size_t overall_length,
293 struct ctdb_req_message *msg)
295 struct messaging_rec *result;
296 DATA_BLOB blob;
297 enum ndr_err_code ndr_err;
299 if ((overall_length < offsetof(struct ctdb_req_message, data))
300 || (overall_length
301 < offsetof(struct ctdb_req_message, data) + msg->datalen)) {
303 cluster_fatal("got invalid msg length");
306 if (!(result = TALLOC_P(mem_ctx, struct messaging_rec))) {
307 DEBUG(0, ("talloc failed\n"));
308 return NULL;
311 blob = data_blob_const(msg->data, msg->datalen);
313 ndr_err = ndr_pull_struct_blob(
314 &blob, result, result,
315 (ndr_pull_flags_fn_t)ndr_pull_messaging_rec);
317 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
318 DEBUG(0, ("ndr_pull_struct_blob failed: %s\n",
319 ndr_errstr(ndr_err)));
320 TALLOC_FREE(result);
321 return NULL;
324 if (DEBUGLEVEL >= 10) {
325 DEBUG(10, ("ctdb_pull_messaging_rec:\n"));
326 NDR_PRINT_DEBUG(messaging_rec, result);
329 return result;
332 static NTSTATUS ctdb_packet_fd_read_sync(struct packet_context *ctx)
334 struct timeval timeout;
335 struct timeval *ptimeout;
337 timeout = timeval_set(lp_ctdb_timeout(), 0);
338 ptimeout = (timeout.tv_sec != 0) ? &timeout : NULL;
340 return packet_fd_read_sync(ctx, ptimeout);
344 * Read a full ctdbd request. If we have a messaging context, defer incoming
345 * messages that might come in between.
348 static NTSTATUS ctdb_read_req(struct ctdbd_connection *conn, uint32 reqid,
349 TALLOC_CTX *mem_ctx, void *result)
351 struct ctdb_req_header *hdr;
352 struct req_pull_state state;
353 NTSTATUS status;
355 again:
357 status = ctdb_packet_fd_read_sync(conn->pkt);
359 if (NT_STATUS_EQUAL(status, NT_STATUS_NETWORK_BUSY)) {
360 /* EAGAIN */
361 goto again;
362 } else if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
363 /* EAGAIN */
364 goto again;
367 if (!NT_STATUS_IS_OK(status)) {
368 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
369 cluster_fatal("ctdbd died\n");
372 next_pkt:
374 ZERO_STRUCT(state);
375 state.mem_ctx = mem_ctx;
377 if (!packet_handler(conn->pkt, ctdb_req_complete, ctdb_req_pull,
378 &state, &status)) {
380 * Not enough data
382 DEBUG(10, ("not enough data from ctdb socket, retrying\n"));
383 goto again;
386 if (!NT_STATUS_IS_OK(status)) {
387 DEBUG(0, ("Could not read packet: %s\n", nt_errstr(status)));
388 cluster_fatal("ctdbd died\n");
391 hdr = (struct ctdb_req_header *)state.req.data;
393 DEBUG(10, ("Received ctdb packet\n"));
394 ctdb_packet_dump(hdr);
396 if (hdr->operation == CTDB_REQ_MESSAGE) {
397 struct timed_event *evt;
398 struct deferred_msg_state *msg_state;
399 struct ctdb_req_message *msg = (struct ctdb_req_message *)hdr;
401 if (conn->msg_ctx == NULL) {
402 DEBUG(1, ("Got a message without having a msg ctx, "
403 "dropping msg %llu\n",
404 (long long unsigned)msg->srvid));
405 goto next_pkt;
408 if ((conn->release_ip_handler != NULL)
409 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
410 /* must be dispatched immediately */
411 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
412 conn->release_ip_handler((const char *)msg->data,
413 conn->release_ip_priv);
414 TALLOC_FREE(hdr);
415 goto next_pkt;
418 if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
419 || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)) {
421 DEBUG(1, ("ctdb_read_req: Got %s message\n",
422 (msg->srvid == CTDB_SRVID_RECONFIGURE)
423 ? "cluster reconfigure" : "SAMBA_NOTIFY"));
425 messaging_send(conn->msg_ctx,
426 messaging_server_id(conn->msg_ctx),
427 MSG_SMB_BRL_VALIDATE, &data_blob_null);
428 messaging_send(conn->msg_ctx,
429 messaging_server_id(conn->msg_ctx),
430 MSG_DBWRAP_G_LOCK_RETRY,
431 &data_blob_null);
432 TALLOC_FREE(hdr);
433 goto next_pkt;
436 msg_state = TALLOC_P(NULL, struct deferred_msg_state);
437 if (msg_state == NULL) {
438 DEBUG(0, ("talloc failed\n"));
439 TALLOC_FREE(hdr);
440 goto next_pkt;
443 if (!(msg_state->rec = ctdb_pull_messaging_rec(
444 msg_state, state.req.length, msg))) {
445 DEBUG(0, ("ctdbd_pull_messaging_rec failed\n"));
446 TALLOC_FREE(msg_state);
447 TALLOC_FREE(hdr);
448 goto next_pkt;
451 TALLOC_FREE(hdr);
453 msg_state->msg_ctx = conn->msg_ctx;
456 * We're waiting for a call reply, but an async message has
457 * crossed. Defer dispatching to the toplevel event loop.
459 evt = event_add_timed(conn->msg_ctx->event_ctx,
460 conn->msg_ctx->event_ctx,
461 timeval_zero(),
462 deferred_message_dispatch,
463 msg_state);
464 if (evt == NULL) {
465 DEBUG(0, ("event_add_timed failed\n"));
466 TALLOC_FREE(msg_state);
467 TALLOC_FREE(hdr);
468 goto next_pkt;
471 goto next_pkt;
474 if (hdr->reqid != reqid) {
475 /* we got the wrong reply */
476 DEBUG(0,("Discarding mismatched ctdb reqid %u should have "
477 "been %u\n", hdr->reqid, reqid));
478 TALLOC_FREE(hdr);
479 goto again;
482 *((void **)result) = talloc_move(mem_ctx, &hdr);
484 return NT_STATUS_OK;
488 * Get us a ctdbd connection
491 static NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
492 struct ctdbd_connection **pconn)
494 struct ctdbd_connection *conn;
495 NTSTATUS status;
497 if (!(conn = TALLOC_ZERO_P(mem_ctx, struct ctdbd_connection))) {
498 DEBUG(0, ("talloc failed\n"));
499 return NT_STATUS_NO_MEMORY;
502 status = ctdbd_connect(conn, &conn->pkt);
504 if (!NT_STATUS_IS_OK(status)) {
505 DEBUG(10, ("ctdbd_connect failed: %s\n", nt_errstr(status)));
506 goto fail;
509 status = get_cluster_vnn(conn, &conn->our_vnn);
511 if (!NT_STATUS_IS_OK(status)) {
512 DEBUG(10, ("get_cluster_vnn failed: %s\n", nt_errstr(status)));
513 goto fail;
516 if (!ctdbd_working(conn, conn->our_vnn)) {
517 DEBUG(2, ("Node is not working, can not connect\n"));
518 goto fail;
521 generate_random_buffer((unsigned char *)&conn->rand_srvid,
522 sizeof(conn->rand_srvid));
524 status = register_with_ctdbd(conn, conn->rand_srvid);
526 if (!NT_STATUS_IS_OK(status)) {
527 DEBUG(5, ("Could not register random srvid: %s\n",
528 nt_errstr(status)));
529 goto fail;
532 *pconn = conn;
533 return NT_STATUS_OK;
535 fail:
536 TALLOC_FREE(conn);
537 return status;
541 * Get us a ctdbd connection and register us as a process
544 NTSTATUS ctdbd_messaging_connection(TALLOC_CTX *mem_ctx,
545 struct ctdbd_connection **pconn)
547 struct ctdbd_connection *conn;
548 NTSTATUS status;
550 status = ctdbd_init_connection(mem_ctx, &conn);
552 if (!NT_STATUS_IS_OK(status)) {
553 return status;
556 status = register_with_ctdbd(conn, (uint64_t)sys_getpid());
557 if (!NT_STATUS_IS_OK(status)) {
558 goto fail;
561 status = register_with_ctdbd(conn, MSG_SRVID_SAMBA);
562 if (!NT_STATUS_IS_OK(status)) {
563 goto fail;
566 status = register_with_ctdbd(conn, CTDB_SRVID_SAMBA_NOTIFY);
567 if (!NT_STATUS_IS_OK(status)) {
568 goto fail;
571 *pconn = conn;
572 return NT_STATUS_OK;
574 fail:
575 TALLOC_FREE(conn);
576 return status;
579 struct messaging_context *ctdb_conn_msg_ctx(struct ctdbd_connection *conn)
581 return conn->msg_ctx;
584 int ctdbd_conn_get_fd(struct ctdbd_connection *conn)
586 return packet_get_fd(conn->pkt);
590 * Packet handler to receive and handle a ctdb message
592 static NTSTATUS ctdb_handle_message(uint8_t *buf, size_t length,
593 void *private_data)
595 struct ctdbd_connection *conn = talloc_get_type_abort(
596 private_data, struct ctdbd_connection);
597 struct ctdb_req_message *msg;
598 struct messaging_rec *msg_rec;
600 msg = (struct ctdb_req_message *)buf;
602 if (msg->hdr.operation != CTDB_REQ_MESSAGE) {
603 DEBUG(0, ("Received async msg of type %u, discarding\n",
604 msg->hdr.operation));
605 TALLOC_FREE(buf);
606 return NT_STATUS_INVALID_PARAMETER;
609 if ((conn->release_ip_handler != NULL)
610 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
611 /* must be dispatched immediately */
612 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
613 conn->release_ip_handler((const char *)msg->data,
614 conn->release_ip_priv);
615 TALLOC_FREE(buf);
616 return NT_STATUS_OK;
619 SMB_ASSERT(conn->msg_ctx != NULL);
621 if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
622 || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)){
623 DEBUG(0,("Got cluster reconfigure message\n"));
625 * when the cluster is reconfigured or someone of the
626 * family has passed away (SAMBA_NOTIFY), we need to
627 * clean the brl database
629 messaging_send(conn->msg_ctx,
630 messaging_server_id(conn->msg_ctx),
631 MSG_SMB_BRL_VALIDATE, &data_blob_null);
633 messaging_send(conn->msg_ctx,
634 messaging_server_id(conn->msg_ctx),
635 MSG_DBWRAP_G_LOCK_RETRY,
636 &data_blob_null);
638 TALLOC_FREE(buf);
639 return NT_STATUS_OK;
642 /* only messages to our pid or the broadcast are valid here */
643 if (msg->srvid != sys_getpid() && msg->srvid != MSG_SRVID_SAMBA) {
644 DEBUG(0,("Got unexpected message with srvid=%llu\n",
645 (unsigned long long)msg->srvid));
646 TALLOC_FREE(buf);
647 return NT_STATUS_OK;
650 if (!(msg_rec = ctdb_pull_messaging_rec(NULL, length, msg))) {
651 DEBUG(10, ("ctdb_pull_messaging_rec failed\n"));
652 TALLOC_FREE(buf);
653 return NT_STATUS_NO_MEMORY;
656 messaging_dispatch_rec(conn->msg_ctx, msg_rec);
658 TALLOC_FREE(msg_rec);
659 TALLOC_FREE(buf);
660 return NT_STATUS_OK;
664 * The ctdbd socket is readable asynchronuously
667 static void ctdbd_socket_handler(struct event_context *event_ctx,
668 struct fd_event *event,
669 uint16 flags,
670 void *private_data)
672 struct ctdbd_connection *conn = talloc_get_type_abort(
673 private_data, struct ctdbd_connection);
675 NTSTATUS status;
677 status = packet_fd_read(conn->pkt);
679 if (!NT_STATUS_IS_OK(status)) {
680 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
681 cluster_fatal("ctdbd died\n");
684 while (packet_handler(conn->pkt, ctdb_req_complete,
685 ctdb_handle_message, conn, &status)) {
686 if (!NT_STATUS_IS_OK(status)) {
687 DEBUG(10, ("could not handle incoming message: %s\n",
688 nt_errstr(status)));
694 * Prepare a ctdbd connection to receive messages
697 NTSTATUS ctdbd_register_msg_ctx(struct ctdbd_connection *conn,
698 struct messaging_context *msg_ctx)
700 SMB_ASSERT(conn->msg_ctx == NULL);
701 SMB_ASSERT(conn->fde == NULL);
703 if (!(conn->fde = event_add_fd(msg_ctx->event_ctx, conn,
704 packet_get_fd(conn->pkt),
705 EVENT_FD_READ,
706 ctdbd_socket_handler,
707 conn))) {
708 DEBUG(0, ("event_add_fd failed\n"));
709 return NT_STATUS_NO_MEMORY;
712 conn->msg_ctx = msg_ctx;
714 return NT_STATUS_OK;
718 * Send a messaging message across a ctdbd
721 NTSTATUS ctdbd_messaging_send(struct ctdbd_connection *conn,
722 uint32 dst_vnn, uint64 dst_srvid,
723 struct messaging_rec *msg)
725 struct ctdb_req_message r;
726 TALLOC_CTX *mem_ctx;
727 DATA_BLOB blob;
728 NTSTATUS status;
729 enum ndr_err_code ndr_err;
731 if (!(mem_ctx = talloc_init("ctdbd_messaging_send"))) {
732 DEBUG(0, ("talloc failed\n"));
733 return NT_STATUS_NO_MEMORY;
736 ndr_err = ndr_push_struct_blob(
737 &blob, mem_ctx, msg,
738 (ndr_push_flags_fn_t)ndr_push_messaging_rec);
740 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
741 DEBUG(0, ("ndr_push_struct_blob failed: %s\n",
742 ndr_errstr(ndr_err)));
743 status = ndr_map_error2ntstatus(ndr_err);
744 goto fail;
747 r.hdr.length = offsetof(struct ctdb_req_message, data) + blob.length;
748 r.hdr.ctdb_magic = CTDB_MAGIC;
749 r.hdr.ctdb_version = CTDB_VERSION;
750 r.hdr.generation = 1;
751 r.hdr.operation = CTDB_REQ_MESSAGE;
752 r.hdr.destnode = dst_vnn;
753 r.hdr.srcnode = conn->our_vnn;
754 r.hdr.reqid = 0;
755 r.srvid = dst_srvid;
756 r.datalen = blob.length;
758 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
759 ctdb_packet_dump(&r.hdr);
761 status = packet_send(
762 conn->pkt, 2,
763 data_blob_const(&r, offsetof(struct ctdb_req_message, data)),
764 blob);
766 if (!NT_STATUS_IS_OK(status)) {
767 DEBUG(0, ("packet_send failed: %s\n", nt_errstr(status)));
768 goto fail;
771 status = packet_flush(conn->pkt);
773 if (!NT_STATUS_IS_OK(status)) {
774 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
775 cluster_fatal("cluster dispatch daemon msg write error\n");
778 status = NT_STATUS_OK;
779 fail:
780 TALLOC_FREE(mem_ctx);
781 return status;
785 * send/recv a generic ctdb control message
787 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
788 uint32_t vnn, uint32 opcode,
789 uint64_t srvid, uint32_t flags,
790 TDB_DATA data,
791 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
792 int *cstatus)
794 struct ctdb_req_control req;
795 struct ctdb_reply_control *reply = NULL;
796 struct ctdbd_connection *new_conn = NULL;
797 NTSTATUS status;
799 /* the samba3 ctdb code can't handle NOREPLY yet */
800 flags &= ~CTDB_CTRL_FLAG_NOREPLY;
802 if (conn == NULL) {
803 status = ctdbd_init_connection(NULL, &new_conn);
805 if (!NT_STATUS_IS_OK(status)) {
806 DEBUG(10, ("Could not init temp connection: %s\n",
807 nt_errstr(status)));
808 goto fail;
811 conn = new_conn;
814 ZERO_STRUCT(req);
815 req.hdr.length = offsetof(struct ctdb_req_control, data) + data.dsize;
816 req.hdr.ctdb_magic = CTDB_MAGIC;
817 req.hdr.ctdb_version = CTDB_VERSION;
818 req.hdr.operation = CTDB_REQ_CONTROL;
819 req.hdr.reqid = ++conn->reqid;
820 req.hdr.destnode = vnn;
821 req.opcode = opcode;
822 req.srvid = srvid;
823 req.datalen = data.dsize;
825 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
826 ctdb_packet_dump(&req.hdr);
828 status = packet_send(
829 conn->pkt, 2,
830 data_blob_const(&req, offsetof(struct ctdb_req_control, data)),
831 data_blob_const(data.dptr, data.dsize));
833 if (!NT_STATUS_IS_OK(status)) {
834 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
835 goto fail;
838 status = packet_flush(conn->pkt);
840 if (!NT_STATUS_IS_OK(status)) {
841 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
842 cluster_fatal("cluster dispatch daemon control write error\n");
845 if (flags & CTDB_CTRL_FLAG_NOREPLY) {
846 TALLOC_FREE(new_conn);
847 return NT_STATUS_OK;
850 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
852 if (!NT_STATUS_IS_OK(status)) {
853 DEBUG(10, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
854 goto fail;
857 if (reply->hdr.operation != CTDB_REPLY_CONTROL) {
858 DEBUG(0, ("received invalid reply\n"));
859 goto fail;
862 if (outdata) {
863 if (!(outdata->dptr = (uint8 *)talloc_memdup(
864 mem_ctx, reply->data, reply->datalen))) {
865 TALLOC_FREE(reply);
866 return NT_STATUS_NO_MEMORY;
868 outdata->dsize = reply->datalen;
870 if (cstatus) {
871 (*cstatus) = reply->status;
874 status = NT_STATUS_OK;
876 fail:
877 TALLOC_FREE(new_conn);
878 TALLOC_FREE(reply);
879 return status;
883 * see if a remote process exists
885 bool ctdbd_process_exists(struct ctdbd_connection *conn, uint32 vnn, pid_t pid)
887 NTSTATUS status;
888 TDB_DATA data;
889 int32_t cstatus;
891 data.dptr = (uint8_t*)&pid;
892 data.dsize = sizeof(pid);
894 status = ctdbd_control(conn, vnn, CTDB_CONTROL_PROCESS_EXISTS, 0, 0,
895 data, NULL, NULL, &cstatus);
896 if (!NT_STATUS_IS_OK(status)) {
897 DEBUG(0, (__location__ " ctdb_control for process_exists "
898 "failed\n"));
899 return False;
902 return cstatus == 0;
906 * Get a db path
908 char *ctdbd_dbpath(struct ctdbd_connection *conn,
909 TALLOC_CTX *mem_ctx, uint32_t db_id)
911 NTSTATUS status;
912 TDB_DATA data;
913 int32_t cstatus;
915 data.dptr = (uint8_t*)&db_id;
916 data.dsize = sizeof(db_id);
918 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
919 CTDB_CONTROL_GETDBPATH, 0, 0, data,
920 mem_ctx, &data, &cstatus);
921 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
922 DEBUG(0,(__location__ " ctdb_control for getdbpath failed\n"));
923 return NULL;
926 return (char *)data.dptr;
930 * attach to a ctdb database
932 NTSTATUS ctdbd_db_attach(struct ctdbd_connection *conn,
933 const char *name, uint32_t *db_id, int tdb_flags)
935 NTSTATUS status;
936 TDB_DATA data;
937 int32_t cstatus;
938 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
940 data.dptr = (uint8_t*)name;
941 data.dsize = strlen(name)+1;
943 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
944 persistent
945 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
946 : CTDB_CONTROL_DB_ATTACH,
947 tdb_flags, 0, data, NULL, &data, &cstatus);
948 if (!NT_STATUS_IS_OK(status)) {
949 DEBUG(0, (__location__ " ctdb_control for db_attach "
950 "failed: %s\n", nt_errstr(status)));
951 return status;
954 if (cstatus != 0 || data.dsize != sizeof(uint32_t)) {
955 DEBUG(0,(__location__ " ctdb_control for db_attach failed\n"));
956 return NT_STATUS_INTERNAL_ERROR;
959 *db_id = *(uint32_t *)data.dptr;
960 talloc_free(data.dptr);
962 if (!(tdb_flags & TDB_SEQNUM)) {
963 return NT_STATUS_OK;
966 data.dptr = (uint8_t *)db_id;
967 data.dsize = sizeof(*db_id);
969 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
970 CTDB_CONTROL_ENABLE_SEQNUM, 0, 0, data,
971 NULL, NULL, &cstatus);
972 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
973 DEBUG(0,(__location__ " ctdb_control for enable seqnum "
974 "failed\n"));
975 return NT_STATUS_IS_OK(status) ? NT_STATUS_INTERNAL_ERROR :
976 status;
979 return NT_STATUS_OK;
983 * force the migration of a record to this node
985 NTSTATUS ctdbd_migrate(struct ctdbd_connection *conn, uint32 db_id,
986 TDB_DATA key)
988 struct ctdb_req_call req;
989 struct ctdb_reply_call *reply;
990 NTSTATUS status;
992 ZERO_STRUCT(req);
994 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
995 req.hdr.ctdb_magic = CTDB_MAGIC;
996 req.hdr.ctdb_version = CTDB_VERSION;
997 req.hdr.operation = CTDB_REQ_CALL;
998 req.hdr.reqid = ++conn->reqid;
999 req.flags = CTDB_IMMEDIATE_MIGRATION;
1000 req.callid = CTDB_NULL_FUNC;
1001 req.db_id = db_id;
1002 req.keylen = key.dsize;
1004 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
1005 ctdb_packet_dump(&req.hdr);
1007 status = packet_send(
1008 conn->pkt, 2,
1009 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
1010 data_blob_const(key.dptr, key.dsize));
1012 if (!NT_STATUS_IS_OK(status)) {
1013 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
1014 return status;
1017 status = packet_flush(conn->pkt);
1019 if (!NT_STATUS_IS_OK(status)) {
1020 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
1021 cluster_fatal("cluster dispatch daemon control write error\n");
1024 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
1026 if (!NT_STATUS_IS_OK(status)) {
1027 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
1028 goto fail;
1031 if (reply->hdr.operation != CTDB_REPLY_CALL) {
1032 DEBUG(0, ("received invalid reply\n"));
1033 status = NT_STATUS_INTERNAL_ERROR;
1034 goto fail;
1037 status = NT_STATUS_OK;
1038 fail:
1040 TALLOC_FREE(reply);
1041 return status;
1045 * remotely fetch a record without locking it or forcing a migration
1047 NTSTATUS ctdbd_fetch(struct ctdbd_connection *conn, uint32 db_id,
1048 TDB_DATA key, TALLOC_CTX *mem_ctx, TDB_DATA *data)
1050 struct ctdb_req_call req;
1051 struct ctdb_reply_call *reply;
1052 NTSTATUS status;
1054 ZERO_STRUCT(req);
1056 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
1057 req.hdr.ctdb_magic = CTDB_MAGIC;
1058 req.hdr.ctdb_version = CTDB_VERSION;
1059 req.hdr.operation = CTDB_REQ_CALL;
1060 req.hdr.reqid = ++conn->reqid;
1061 req.flags = 0;
1062 req.callid = CTDB_FETCH_FUNC;
1063 req.db_id = db_id;
1064 req.keylen = key.dsize;
1066 status = packet_send(
1067 conn->pkt, 2,
1068 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
1069 data_blob_const(key.dptr, key.dsize));
1071 if (!NT_STATUS_IS_OK(status)) {
1072 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
1073 return status;
1076 status = packet_flush(conn->pkt);
1078 if (!NT_STATUS_IS_OK(status)) {
1079 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
1080 cluster_fatal("cluster dispatch daemon control write error\n");
1083 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
1085 if (!NT_STATUS_IS_OK(status)) {
1086 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
1087 goto fail;
1090 if (reply->hdr.operation != CTDB_REPLY_CALL) {
1091 DEBUG(0, ("received invalid reply\n"));
1092 status = NT_STATUS_INTERNAL_ERROR;
1093 goto fail;
1096 data->dsize = reply->datalen;
1097 if (data->dsize == 0) {
1098 data->dptr = NULL;
1099 goto done;
1102 data->dptr = (uint8 *)talloc_memdup(mem_ctx, &reply->data[0],
1103 reply->datalen);
1104 if (data->dptr == NULL) {
1105 DEBUG(0, ("talloc failed\n"));
1106 status = NT_STATUS_NO_MEMORY;
1107 goto fail;
1110 done:
1111 status = NT_STATUS_OK;
1112 fail:
1113 TALLOC_FREE(reply);
1114 return status;
1117 struct ctdbd_traverse_state {
1118 void (*fn)(TDB_DATA key, TDB_DATA data, void *private_data);
1119 void *private_data;
1123 * Handle a traverse record coming in on the ctdbd connection
1126 static NTSTATUS ctdb_traverse_handler(uint8_t *buf, size_t length,
1127 void *private_data)
1129 struct ctdbd_traverse_state *state =
1130 (struct ctdbd_traverse_state *)private_data;
1132 struct ctdb_req_message *m;
1133 struct ctdb_rec_data *d;
1134 TDB_DATA key, data;
1136 m = (struct ctdb_req_message *)buf;
1138 if (length < sizeof(*m) || m->hdr.length != length) {
1139 DEBUG(0, ("Got invalid message of length %d\n", (int)length));
1140 TALLOC_FREE(buf);
1141 return NT_STATUS_UNEXPECTED_IO_ERROR;
1144 d = (struct ctdb_rec_data *)&m->data[0];
1145 if (m->datalen < sizeof(uint32_t) || m->datalen != d->length) {
1146 DEBUG(0, ("Got invalid traverse data of length %d\n",
1147 (int)m->datalen));
1148 TALLOC_FREE(buf);
1149 return NT_STATUS_UNEXPECTED_IO_ERROR;
1152 key.dsize = d->keylen;
1153 key.dptr = &d->data[0];
1154 data.dsize = d->datalen;
1155 data.dptr = &d->data[d->keylen];
1157 if (key.dsize == 0 && data.dsize == 0) {
1158 /* end of traverse */
1159 return NT_STATUS_END_OF_FILE;
1162 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1163 DEBUG(0, ("Got invalid ltdb header length %d\n",
1164 (int)data.dsize));
1165 TALLOC_FREE(buf);
1166 return NT_STATUS_UNEXPECTED_IO_ERROR;
1168 data.dsize -= sizeof(struct ctdb_ltdb_header);
1169 data.dptr += sizeof(struct ctdb_ltdb_header);
1171 if (state->fn) {
1172 state->fn(key, data, state->private_data);
1175 TALLOC_FREE(buf);
1176 return NT_STATUS_OK;
1180 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1181 connection to ctdbd to avoid the hairy recursive and async problems with
1182 everything in-line.
1185 NTSTATUS ctdbd_traverse(uint32 db_id,
1186 void (*fn)(TDB_DATA key, TDB_DATA data,
1187 void *private_data),
1188 void *private_data)
1190 struct ctdbd_connection *conn;
1191 NTSTATUS status;
1193 TDB_DATA data;
1194 struct ctdb_traverse_start t;
1195 int cstatus;
1196 struct ctdbd_traverse_state state;
1198 status = ctdbd_init_connection(NULL, &conn);
1199 if (!NT_STATUS_IS_OK(status)) {
1200 DEBUG(0, ("ctdbd_init_connection failed: %s\n",
1201 nt_errstr(status)));
1202 return status;
1205 t.db_id = db_id;
1206 t.srvid = conn->rand_srvid;
1207 t.reqid = ++conn->reqid;
1209 data.dptr = (uint8_t *)&t;
1210 data.dsize = sizeof(t);
1212 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1213 CTDB_CONTROL_TRAVERSE_START, conn->rand_srvid, 0,
1214 data, NULL, NULL, &cstatus);
1216 if (!NT_STATUS_IS_OK(status) || (cstatus != 0)) {
1218 DEBUG(0,("ctdbd_control failed: %s, %d\n", nt_errstr(status),
1219 cstatus));
1221 if (NT_STATUS_IS_OK(status)) {
1223 * We need a mapping here
1225 status = NT_STATUS_UNSUCCESSFUL;
1227 goto done;
1230 state.fn = fn;
1231 state.private_data = private_data;
1233 while (True) {
1235 status = NT_STATUS_OK;
1237 if (packet_handler(conn->pkt, ctdb_req_complete,
1238 ctdb_traverse_handler, &state, &status)) {
1240 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1241 status = NT_STATUS_OK;
1242 break;
1246 * There might be more in the queue
1248 continue;
1251 if (!NT_STATUS_IS_OK(status)) {
1252 break;
1255 status = ctdb_packet_fd_read_sync(conn->pkt);
1257 if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
1259 * There might be more in the queue
1261 continue;
1264 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1265 status = NT_STATUS_OK;
1266 break;
1269 if (!NT_STATUS_IS_OK(status)) {
1270 DEBUG(0, ("packet_fd_read_sync failed: %s\n", nt_errstr(status)));
1271 cluster_fatal("ctdbd died\n");
1275 done:
1276 TALLOC_FREE(conn);
1277 return status;
1281 This is used to canonicalize a ctdb_sock_addr structure.
1283 static void smbd_ctdb_canonicalize_ip(const struct sockaddr_storage *in,
1284 struct sockaddr_storage *out)
1286 memcpy(out, in, sizeof (*out));
1288 #ifdef HAVE_IPV6
1289 if (in->ss_family == AF_INET6) {
1290 const char prefix[12] = { 0,0,0,0,0,0,0,0,0,0,0xff,0xff };
1291 const struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)in;
1292 struct sockaddr_in *out4 = (struct sockaddr_in *)out;
1293 if (memcmp(&in6->sin6_addr, prefix, 12) == 0) {
1294 memset(out, 0, sizeof(*out));
1295 #ifdef HAVE_SOCK_SIN_LEN
1296 out4->sin_len = sizeof(*out);
1297 #endif
1298 out4->sin_family = AF_INET;
1299 out4->sin_port = in6->sin6_port;
1300 memcpy(&out4->sin_addr, &in6->sin6_addr.s6_addr32[3], 4);
1303 #endif
1307 * Register us as a server for a particular tcp connection
1310 NTSTATUS ctdbd_register_ips(struct ctdbd_connection *conn,
1311 const struct sockaddr_storage *_server,
1312 const struct sockaddr_storage *_client,
1313 void (*release_ip_handler)(const char *ip_addr,
1314 void *private_data),
1315 void *private_data)
1318 * we still use ctdb_control_tcp for ipv4
1319 * because we want to work against older ctdb
1320 * versions at runtime
1322 struct ctdb_control_tcp p4;
1323 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1324 struct ctdb_control_tcp_addr p;
1325 #endif
1326 TDB_DATA data;
1327 NTSTATUS status;
1328 struct sockaddr_storage client;
1329 struct sockaddr_storage server;
1332 * Only one connection so far
1334 SMB_ASSERT(conn->release_ip_handler == NULL);
1336 smbd_ctdb_canonicalize_ip(_client, &client);
1337 smbd_ctdb_canonicalize_ip(_server, &server);
1339 switch (client.ss_family) {
1340 case AF_INET:
1341 p4.dest = *(struct sockaddr_in *)(void *)&server;
1342 p4.src = *(struct sockaddr_in *)(void *)&client;
1343 data.dptr = (uint8_t *)&p4;
1344 data.dsize = sizeof(p4);
1345 break;
1346 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1347 case AF_INET6:
1348 p.dest.ip6 = *(struct sockaddr_in6 *)(void *)&server;
1349 p.src.ip6 = *(struct sockaddr_in6 *)(void *)&client;
1350 data.dptr = (uint8_t *)&p;
1351 data.dsize = sizeof(p);
1352 break;
1353 #endif
1354 default:
1355 return NT_STATUS_INTERNAL_ERROR;
1358 conn->release_ip_handler = release_ip_handler;
1361 * We want to be told about IP releases
1364 status = register_with_ctdbd(conn, CTDB_SRVID_RELEASE_IP);
1365 if (!NT_STATUS_IS_OK(status)) {
1366 return status;
1370 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1371 * can send an extra ack to trigger a reset for our client, so it
1372 * immediately reconnects
1374 return ctdbd_control(conn, CTDB_CURRENT_NODE,
1375 CTDB_CONTROL_TCP_CLIENT, 0,
1376 CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL, NULL);
1380 * We want to handle reconfigure events
1382 NTSTATUS ctdbd_register_reconfigure(struct ctdbd_connection *conn)
1384 return register_with_ctdbd(conn, CTDB_SRVID_RECONFIGURE);
1388 call a control on the local node
1390 NTSTATUS ctdbd_control_local(struct ctdbd_connection *conn, uint32 opcode,
1391 uint64_t srvid, uint32_t flags, TDB_DATA data,
1392 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
1393 int *cstatus)
1395 return ctdbd_control(conn, CTDB_CURRENT_NODE, opcode, srvid, flags, data, mem_ctx, outdata, cstatus);
1398 NTSTATUS ctdb_watch_us(struct ctdbd_connection *conn)
1400 struct ctdb_client_notify_register reg_data;
1401 size_t struct_len;
1402 NTSTATUS status;
1403 int cstatus;
1405 reg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1406 reg_data.len = 1;
1407 reg_data.notify_data[0] = 0;
1409 struct_len = offsetof(struct ctdb_client_notify_register,
1410 notify_data) + reg_data.len;
1412 status = ctdbd_control_local(
1413 conn, CTDB_CONTROL_REGISTER_NOTIFY, conn->rand_srvid, 0,
1414 make_tdb_data((uint8_t *)&reg_data, struct_len),
1415 NULL, NULL, &cstatus);
1416 if (!NT_STATUS_IS_OK(status)) {
1417 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1418 nt_errstr(status)));
1420 return status;
1423 NTSTATUS ctdb_unwatch(struct ctdbd_connection *conn)
1425 struct ctdb_client_notify_deregister dereg_data;
1426 NTSTATUS status;
1427 int cstatus;
1429 dereg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1431 status = ctdbd_control_local(
1432 conn, CTDB_CONTROL_DEREGISTER_NOTIFY, conn->rand_srvid, 0,
1433 make_tdb_data((uint8_t *)&dereg_data, sizeof(dereg_data)),
1434 NULL, NULL, &cstatus);
1435 if (!NT_STATUS_IS_OK(status)) {
1436 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1437 nt_errstr(status)));
1439 return status;
1442 #endif