s3-talloc Change TALLOC_ZERO_P() to talloc_zero()
[Samba.git] / source3 / lib / ctdbd_conn.c
blob79dc1f246aad68d1cfde89822a832ad537df7e40
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"
22 #include "util_tdb.h"
24 #ifdef CLUSTER_SUPPORT
26 #include "ctdbd_conn.h"
27 #include "ctdb_packet.h"
28 #include "messages.h"
30 /* paths to these include files come from --with-ctdb= in configure */
31 #include "ctdb.h"
32 #include "ctdb_private.h"
34 struct ctdbd_connection {
35 struct messaging_context *msg_ctx;
36 uint32 reqid;
37 uint32 our_vnn;
38 uint64 rand_srvid;
39 struct ctdb_packet_context *pkt;
40 struct fd_event *fde;
42 void (*release_ip_handler)(const char *ip_addr, void *private_data);
43 void *release_ip_priv;
46 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
47 uint32_t vnn, uint32 opcode,
48 uint64_t srvid, uint32_t flags, TDB_DATA data,
49 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
50 int *cstatus);
53 * exit on fatal communications errors with the ctdbd daemon
55 static void cluster_fatal(const char *why)
57 DEBUG(0,("cluster fatal event: %s - exiting immediately\n", why));
58 /* we don't use smb_panic() as we don't want to delay to write
59 a core file. We need to release this process id immediately
60 so that someone else can take over without getting sharing
61 violations */
62 _exit(1);
68 static void ctdb_packet_dump(struct ctdb_req_header *hdr)
70 if (DEBUGLEVEL < 10) {
71 return;
73 DEBUGADD(10, ("len=%d, magic=%x, vers=%d, gen=%d, op=%d, reqid=%d\n",
74 (int)hdr->length, (int)hdr->ctdb_magic,
75 (int)hdr->ctdb_version, (int)hdr->generation,
76 (int)hdr->operation, (int)hdr->reqid));
80 * Register a srvid with ctdbd
82 static NTSTATUS register_with_ctdbd(struct ctdbd_connection *conn,
83 uint64_t srvid)
86 int cstatus;
87 return ctdbd_control(conn, CTDB_CURRENT_NODE,
88 CTDB_CONTROL_REGISTER_SRVID, srvid, 0,
89 tdb_null, NULL, NULL, &cstatus);
93 * get our vnn from the cluster
95 static NTSTATUS get_cluster_vnn(struct ctdbd_connection *conn, uint32 *vnn)
97 int32_t cstatus=-1;
98 NTSTATUS status;
99 status = ctdbd_control(conn,
100 CTDB_CURRENT_NODE, CTDB_CONTROL_GET_PNN, 0, 0,
101 tdb_null, NULL, NULL, &cstatus);
102 if (!NT_STATUS_IS_OK(status)) {
103 cluster_fatal("ctdbd_control failed\n");
105 *vnn = (uint32_t)cstatus;
106 return status;
110 * Are we active (i.e. not banned or stopped?)
112 static bool ctdbd_working(struct ctdbd_connection *conn, uint32_t vnn)
114 int32_t cstatus=-1;
115 NTSTATUS status;
116 TDB_DATA outdata;
117 struct ctdb_node_map *m;
118 uint32_t failure_flags;
119 bool ret = false;
120 int i;
122 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
123 CTDB_CONTROL_GET_NODEMAP, 0, 0,
124 tdb_null, talloc_tos(), &outdata, &cstatus);
125 if (!NT_STATUS_IS_OK(status)) {
126 cluster_fatal("ctdbd_control failed\n");
128 if ((cstatus != 0) || (outdata.dptr == NULL)) {
129 DEBUG(2, ("Received invalid ctdb data\n"));
130 return false;
133 m = (struct ctdb_node_map *)outdata.dptr;
135 for (i=0; i<m->num; i++) {
136 if (vnn == m->nodes[i].pnn) {
137 break;
141 if (i == m->num) {
142 DEBUG(2, ("Did not find ourselves (node %d) in nodemap\n",
143 (int)vnn));
144 goto fail;
147 failure_flags = NODE_FLAGS_BANNED | NODE_FLAGS_DISCONNECTED
148 | NODE_FLAGS_PERMANENTLY_DISABLED | NODE_FLAGS_STOPPED;
150 if ((m->nodes[i].flags & failure_flags) != 0) {
151 DEBUG(2, ("Node has status %x, not active\n",
152 (int)m->nodes[i].flags));
153 goto fail;
156 ret = true;
157 fail:
158 TALLOC_FREE(outdata.dptr);
159 return ret;
162 uint32 ctdbd_vnn(const struct ctdbd_connection *conn)
164 return conn->our_vnn;
168 * Get us a ctdb connection
171 static NTSTATUS ctdbd_connect(TALLOC_CTX *mem_ctx,
172 struct ctdb_packet_context **presult)
174 struct ctdb_packet_context *result;
175 const char *sockname = lp_ctdbd_socket();
176 struct sockaddr_un addr;
177 int fd;
179 if (!sockname || !*sockname) {
180 sockname = CTDB_PATH;
183 fd = socket(AF_UNIX, SOCK_STREAM, 0);
184 if (fd == -1) {
185 DEBUG(3, ("Could not create socket: %s\n", strerror(errno)));
186 return map_nt_error_from_unix(errno);
189 ZERO_STRUCT(addr);
190 addr.sun_family = AF_UNIX;
191 strncpy(addr.sun_path, sockname, sizeof(addr.sun_path));
193 if (sys_connect(fd, (struct sockaddr *)(void *)&addr) == -1) {
194 DEBUG(1, ("connect(%s) failed: %s\n", sockname,
195 strerror(errno)));
196 close(fd);
197 return map_nt_error_from_unix(errno);
200 if (!(result = ctdb_packet_init(mem_ctx, fd))) {
201 close(fd);
202 return NT_STATUS_NO_MEMORY;
205 *presult = result;
206 return NT_STATUS_OK;
210 * Do we have a complete ctdb packet in the queue?
213 static bool ctdb_req_complete(const uint8_t *buf, size_t available,
214 size_t *length,
215 void *private_data)
217 uint32 msglen;
219 if (available < sizeof(msglen)) {
220 return False;
223 msglen = *((uint32 *)buf);
225 DEBUG(10, ("msglen = %d\n", msglen));
227 if (msglen < sizeof(struct ctdb_req_header)) {
228 DEBUG(0, ("Got invalid msglen: %d, expected at least %d for "
229 "the req_header\n", (int)msglen,
230 (int)sizeof(struct ctdb_req_header)));
231 cluster_fatal("ctdbd protocol error\n");
234 if (available < msglen) {
235 return false;
238 *length = msglen;
239 return true;
243 * State necessary to defer an incoming message while we are waiting for a
244 * ctdb reply.
247 struct deferred_msg_state {
248 struct messaging_context *msg_ctx;
249 struct messaging_rec *rec;
253 * Timed event handler for the deferred message
256 static void deferred_message_dispatch(struct event_context *event_ctx,
257 struct timed_event *te,
258 struct timeval now,
259 void *private_data)
261 struct deferred_msg_state *state = talloc_get_type_abort(
262 private_data, struct deferred_msg_state);
264 messaging_dispatch_rec(state->msg_ctx, state->rec);
265 TALLOC_FREE(state);
266 TALLOC_FREE(te);
269 struct req_pull_state {
270 TALLOC_CTX *mem_ctx;
271 DATA_BLOB req;
275 * Pull a ctdb request out of the incoming ctdb_packet queue
278 static NTSTATUS ctdb_req_pull(uint8_t *buf, size_t length,
279 void *private_data)
281 struct req_pull_state *state = (struct req_pull_state *)private_data;
283 state->req.data = talloc_move(state->mem_ctx, &buf);
284 state->req.length = length;
285 return NT_STATUS_OK;
289 * Fetch a messaging_rec from an incoming ctdb style message
292 static struct messaging_rec *ctdb_pull_messaging_rec(TALLOC_CTX *mem_ctx,
293 size_t overall_length,
294 struct ctdb_req_message *msg)
296 struct messaging_rec *result;
297 DATA_BLOB blob;
298 enum ndr_err_code ndr_err;
300 if ((overall_length < offsetof(struct ctdb_req_message, data))
301 || (overall_length
302 < offsetof(struct ctdb_req_message, data) + msg->datalen)) {
304 cluster_fatal("got invalid msg length");
307 if (!(result = talloc(mem_ctx, struct messaging_rec))) {
308 DEBUG(0, ("talloc failed\n"));
309 return NULL;
312 blob = data_blob_const(msg->data, msg->datalen);
314 ndr_err = ndr_pull_struct_blob(
315 &blob, result, result,
316 (ndr_pull_flags_fn_t)ndr_pull_messaging_rec);
318 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
319 DEBUG(0, ("ndr_pull_struct_blob failed: %s\n",
320 ndr_errstr(ndr_err)));
321 TALLOC_FREE(result);
322 return NULL;
325 if (DEBUGLEVEL >= 10) {
326 DEBUG(10, ("ctdb_pull_messaging_rec:\n"));
327 NDR_PRINT_DEBUG(messaging_rec, result);
330 return result;
333 static NTSTATUS ctdb_packet_fd_read_sync(struct ctdb_packet_context *ctx)
335 int timeout = lp_ctdb_timeout();
337 if (timeout == 0) {
338 timeout = -1;
340 return ctdb_packet_fd_read_sync_timeout(ctx, timeout);
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, ("ctdb_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 (!ctdb_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 ctdb_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(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(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 status = NT_STATUS_INTERNAL_DB_ERROR;
519 goto fail;
522 generate_random_buffer((unsigned char *)&conn->rand_srvid,
523 sizeof(conn->rand_srvid));
525 status = register_with_ctdbd(conn, conn->rand_srvid);
527 if (!NT_STATUS_IS_OK(status)) {
528 DEBUG(5, ("Could not register random srvid: %s\n",
529 nt_errstr(status)));
530 goto fail;
533 *pconn = conn;
534 return NT_STATUS_OK;
536 fail:
537 TALLOC_FREE(conn);
538 return status;
542 * Get us a ctdbd connection and register us as a process
545 NTSTATUS ctdbd_messaging_connection(TALLOC_CTX *mem_ctx,
546 struct ctdbd_connection **pconn)
548 struct ctdbd_connection *conn;
549 NTSTATUS status;
551 status = ctdbd_init_connection(mem_ctx, &conn);
553 if (!NT_STATUS_IS_OK(status)) {
554 return status;
557 status = register_with_ctdbd(conn, (uint64_t)sys_getpid());
558 if (!NT_STATUS_IS_OK(status)) {
559 goto fail;
562 status = register_with_ctdbd(conn, MSG_SRVID_SAMBA);
563 if (!NT_STATUS_IS_OK(status)) {
564 goto fail;
567 status = register_with_ctdbd(conn, CTDB_SRVID_SAMBA_NOTIFY);
568 if (!NT_STATUS_IS_OK(status)) {
569 goto fail;
572 *pconn = conn;
573 return NT_STATUS_OK;
575 fail:
576 TALLOC_FREE(conn);
577 return status;
580 struct messaging_context *ctdb_conn_msg_ctx(struct ctdbd_connection *conn)
582 return conn->msg_ctx;
585 int ctdbd_conn_get_fd(struct ctdbd_connection *conn)
587 return ctdb_packet_get_fd(conn->pkt);
591 * Packet handler to receive and handle a ctdb message
593 static NTSTATUS ctdb_handle_message(uint8_t *buf, size_t length,
594 void *private_data)
596 struct ctdbd_connection *conn = talloc_get_type_abort(
597 private_data, struct ctdbd_connection);
598 struct ctdb_req_message *msg;
599 struct messaging_rec *msg_rec;
601 msg = (struct ctdb_req_message *)buf;
603 if (msg->hdr.operation != CTDB_REQ_MESSAGE) {
604 DEBUG(0, ("Received async msg of type %u, discarding\n",
605 msg->hdr.operation));
606 TALLOC_FREE(buf);
607 return NT_STATUS_INVALID_PARAMETER;
610 if ((conn->release_ip_handler != NULL)
611 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
612 /* must be dispatched immediately */
613 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
614 conn->release_ip_handler((const char *)msg->data,
615 conn->release_ip_priv);
616 TALLOC_FREE(buf);
617 return NT_STATUS_OK;
620 SMB_ASSERT(conn->msg_ctx != NULL);
622 if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
623 || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)){
624 DEBUG(0,("Got cluster reconfigure message\n"));
626 * when the cluster is reconfigured or someone of the
627 * family has passed away (SAMBA_NOTIFY), we need to
628 * clean the brl database
630 messaging_send(conn->msg_ctx,
631 messaging_server_id(conn->msg_ctx),
632 MSG_SMB_BRL_VALIDATE, &data_blob_null);
634 messaging_send(conn->msg_ctx,
635 messaging_server_id(conn->msg_ctx),
636 MSG_DBWRAP_G_LOCK_RETRY,
637 &data_blob_null);
639 TALLOC_FREE(buf);
640 return NT_STATUS_OK;
643 /* only messages to our pid or the broadcast are valid here */
644 if (msg->srvid != sys_getpid() && msg->srvid != MSG_SRVID_SAMBA) {
645 DEBUG(0,("Got unexpected message with srvid=%llu\n",
646 (unsigned long long)msg->srvid));
647 TALLOC_FREE(buf);
648 return NT_STATUS_OK;
651 if (!(msg_rec = ctdb_pull_messaging_rec(NULL, length, msg))) {
652 DEBUG(10, ("ctdb_pull_messaging_rec failed\n"));
653 TALLOC_FREE(buf);
654 return NT_STATUS_NO_MEMORY;
657 messaging_dispatch_rec(conn->msg_ctx, msg_rec);
659 TALLOC_FREE(msg_rec);
660 TALLOC_FREE(buf);
661 return NT_STATUS_OK;
665 * The ctdbd socket is readable asynchronuously
668 static void ctdbd_socket_handler(struct event_context *event_ctx,
669 struct fd_event *event,
670 uint16 flags,
671 void *private_data)
673 struct ctdbd_connection *conn = talloc_get_type_abort(
674 private_data, struct ctdbd_connection);
676 NTSTATUS status;
678 status = ctdb_packet_fd_read(conn->pkt);
680 if (!NT_STATUS_IS_OK(status)) {
681 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
682 cluster_fatal("ctdbd died\n");
685 while (ctdb_packet_handler(conn->pkt, ctdb_req_complete,
686 ctdb_handle_message, conn, &status)) {
687 if (!NT_STATUS_IS_OK(status)) {
688 DEBUG(10, ("could not handle incoming message: %s\n",
689 nt_errstr(status)));
695 * Prepare a ctdbd connection to receive messages
698 NTSTATUS ctdbd_register_msg_ctx(struct ctdbd_connection *conn,
699 struct messaging_context *msg_ctx)
701 SMB_ASSERT(conn->msg_ctx == NULL);
702 SMB_ASSERT(conn->fde == NULL);
704 if (!(conn->fde = event_add_fd(msg_ctx->event_ctx, conn,
705 ctdb_packet_get_fd(conn->pkt),
706 EVENT_FD_READ,
707 ctdbd_socket_handler,
708 conn))) {
709 DEBUG(0, ("event_add_fd failed\n"));
710 return NT_STATUS_NO_MEMORY;
713 conn->msg_ctx = msg_ctx;
715 return NT_STATUS_OK;
719 * Send a messaging message across a ctdbd
722 NTSTATUS ctdbd_messaging_send(struct ctdbd_connection *conn,
723 uint32 dst_vnn, uint64 dst_srvid,
724 struct messaging_rec *msg)
726 struct ctdb_req_message r;
727 TALLOC_CTX *mem_ctx;
728 DATA_BLOB blob;
729 NTSTATUS status;
730 enum ndr_err_code ndr_err;
732 if (!(mem_ctx = talloc_init("ctdbd_messaging_send"))) {
733 DEBUG(0, ("talloc failed\n"));
734 return NT_STATUS_NO_MEMORY;
737 ndr_err = ndr_push_struct_blob(
738 &blob, mem_ctx, msg,
739 (ndr_push_flags_fn_t)ndr_push_messaging_rec);
741 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
742 DEBUG(0, ("ndr_push_struct_blob failed: %s\n",
743 ndr_errstr(ndr_err)));
744 status = ndr_map_error2ntstatus(ndr_err);
745 goto fail;
748 r.hdr.length = offsetof(struct ctdb_req_message, data) + blob.length;
749 r.hdr.ctdb_magic = CTDB_MAGIC;
750 r.hdr.ctdb_version = CTDB_VERSION;
751 r.hdr.generation = 1;
752 r.hdr.operation = CTDB_REQ_MESSAGE;
753 r.hdr.destnode = dst_vnn;
754 r.hdr.srcnode = conn->our_vnn;
755 r.hdr.reqid = 0;
756 r.srvid = dst_srvid;
757 r.datalen = blob.length;
759 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
760 ctdb_packet_dump(&r.hdr);
762 status = ctdb_packet_send(
763 conn->pkt, 2,
764 data_blob_const(&r, offsetof(struct ctdb_req_message, data)),
765 blob);
767 if (!NT_STATUS_IS_OK(status)) {
768 DEBUG(0, ("ctdb_packet_send failed: %s\n", nt_errstr(status)));
769 goto fail;
772 status = ctdb_packet_flush(conn->pkt);
774 if (!NT_STATUS_IS_OK(status)) {
775 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
776 cluster_fatal("cluster dispatch daemon msg write error\n");
779 status = NT_STATUS_OK;
780 fail:
781 TALLOC_FREE(mem_ctx);
782 return status;
786 * send/recv a generic ctdb control message
788 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
789 uint32_t vnn, uint32 opcode,
790 uint64_t srvid, uint32_t flags,
791 TDB_DATA data,
792 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
793 int *cstatus)
795 struct ctdb_req_control req;
796 struct ctdb_reply_control *reply = NULL;
797 struct ctdbd_connection *new_conn = NULL;
798 NTSTATUS status;
800 if (conn == NULL) {
801 status = ctdbd_init_connection(NULL, &new_conn);
803 if (!NT_STATUS_IS_OK(status)) {
804 DEBUG(10, ("Could not init temp connection: %s\n",
805 nt_errstr(status)));
806 goto fail;
809 conn = new_conn;
812 ZERO_STRUCT(req);
813 req.hdr.length = offsetof(struct ctdb_req_control, data) + data.dsize;
814 req.hdr.ctdb_magic = CTDB_MAGIC;
815 req.hdr.ctdb_version = CTDB_VERSION;
816 req.hdr.operation = CTDB_REQ_CONTROL;
817 req.hdr.reqid = ++conn->reqid;
818 req.hdr.destnode = vnn;
819 req.opcode = opcode;
820 req.srvid = srvid;
821 req.datalen = data.dsize;
822 req.flags = flags;
824 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
825 ctdb_packet_dump(&req.hdr);
827 status = ctdb_packet_send(
828 conn->pkt, 2,
829 data_blob_const(&req, offsetof(struct ctdb_req_control, data)),
830 data_blob_const(data.dptr, data.dsize));
832 if (!NT_STATUS_IS_OK(status)) {
833 DEBUG(3, ("ctdb_packet_send failed: %s\n", nt_errstr(status)));
834 goto fail;
837 status = ctdb_packet_flush(conn->pkt);
839 if (!NT_STATUS_IS_OK(status)) {
840 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
841 cluster_fatal("cluster dispatch daemon control write error\n");
844 if (flags & CTDB_CTRL_FLAG_NOREPLY) {
845 TALLOC_FREE(new_conn);
846 if (cstatus) {
847 *cstatus = 0;
849 return NT_STATUS_OK;
852 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
854 if (!NT_STATUS_IS_OK(status)) {
855 DEBUG(10, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
856 goto fail;
859 if (reply->hdr.operation != CTDB_REPLY_CONTROL) {
860 DEBUG(0, ("received invalid reply\n"));
861 goto fail;
864 if (outdata) {
865 if (!(outdata->dptr = (uint8 *)talloc_memdup(
866 mem_ctx, reply->data, reply->datalen))) {
867 TALLOC_FREE(reply);
868 return NT_STATUS_NO_MEMORY;
870 outdata->dsize = reply->datalen;
872 if (cstatus) {
873 (*cstatus) = reply->status;
876 status = NT_STATUS_OK;
878 fail:
879 TALLOC_FREE(new_conn);
880 TALLOC_FREE(reply);
881 return status;
885 * see if a remote process exists
887 bool ctdbd_process_exists(struct ctdbd_connection *conn, uint32 vnn, pid_t pid)
889 NTSTATUS status;
890 TDB_DATA data;
891 int32_t cstatus;
893 data.dptr = (uint8_t*)&pid;
894 data.dsize = sizeof(pid);
896 status = ctdbd_control(conn, vnn, CTDB_CONTROL_PROCESS_EXISTS, 0, 0,
897 data, NULL, NULL, &cstatus);
898 if (!NT_STATUS_IS_OK(status)) {
899 DEBUG(0, (__location__ " ctdb_control for process_exists "
900 "failed\n"));
901 return False;
904 return cstatus == 0;
908 * Get a db path
910 char *ctdbd_dbpath(struct ctdbd_connection *conn,
911 TALLOC_CTX *mem_ctx, uint32_t db_id)
913 NTSTATUS status;
914 TDB_DATA data;
915 int32_t cstatus;
917 data.dptr = (uint8_t*)&db_id;
918 data.dsize = sizeof(db_id);
920 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
921 CTDB_CONTROL_GETDBPATH, 0, 0, data,
922 mem_ctx, &data, &cstatus);
923 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
924 DEBUG(0,(__location__ " ctdb_control for getdbpath failed\n"));
925 return NULL;
928 return (char *)data.dptr;
932 * attach to a ctdb database
934 NTSTATUS ctdbd_db_attach(struct ctdbd_connection *conn,
935 const char *name, uint32_t *db_id, int tdb_flags)
937 NTSTATUS status;
938 TDB_DATA data;
939 int32_t cstatus;
940 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
942 data.dptr = (uint8_t*)name;
943 data.dsize = strlen(name)+1;
945 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
946 persistent
947 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
948 : CTDB_CONTROL_DB_ATTACH,
949 tdb_flags, 0, data, NULL, &data, &cstatus);
950 if (!NT_STATUS_IS_OK(status)) {
951 DEBUG(0, (__location__ " ctdb_control for db_attach "
952 "failed: %s\n", nt_errstr(status)));
953 return status;
956 if (cstatus != 0 || data.dsize != sizeof(uint32_t)) {
957 DEBUG(0,(__location__ " ctdb_control for db_attach failed\n"));
958 return NT_STATUS_INTERNAL_ERROR;
961 *db_id = *(uint32_t *)data.dptr;
962 talloc_free(data.dptr);
964 if (!(tdb_flags & TDB_SEQNUM)) {
965 return NT_STATUS_OK;
968 data.dptr = (uint8_t *)db_id;
969 data.dsize = sizeof(*db_id);
971 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
972 CTDB_CONTROL_ENABLE_SEQNUM, 0, 0, data,
973 NULL, NULL, &cstatus);
974 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
975 DEBUG(0,(__location__ " ctdb_control for enable seqnum "
976 "failed\n"));
977 return NT_STATUS_IS_OK(status) ? NT_STATUS_INTERNAL_ERROR :
978 status;
981 return NT_STATUS_OK;
985 * force the migration of a record to this node
987 NTSTATUS ctdbd_migrate(struct ctdbd_connection *conn, uint32 db_id,
988 TDB_DATA key)
990 struct ctdb_req_call req;
991 struct ctdb_reply_call *reply;
992 NTSTATUS status;
994 ZERO_STRUCT(req);
996 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
997 req.hdr.ctdb_magic = CTDB_MAGIC;
998 req.hdr.ctdb_version = CTDB_VERSION;
999 req.hdr.operation = CTDB_REQ_CALL;
1000 req.hdr.reqid = ++conn->reqid;
1001 req.flags = CTDB_IMMEDIATE_MIGRATION;
1002 req.callid = CTDB_NULL_FUNC;
1003 req.db_id = db_id;
1004 req.keylen = key.dsize;
1006 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
1007 ctdb_packet_dump(&req.hdr);
1009 status = ctdb_packet_send(
1010 conn->pkt, 2,
1011 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
1012 data_blob_const(key.dptr, key.dsize));
1014 if (!NT_STATUS_IS_OK(status)) {
1015 DEBUG(3, ("ctdb_packet_send failed: %s\n", nt_errstr(status)));
1016 return status;
1019 status = ctdb_packet_flush(conn->pkt);
1021 if (!NT_STATUS_IS_OK(status)) {
1022 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
1023 cluster_fatal("cluster dispatch daemon control write error\n");
1026 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
1028 if (!NT_STATUS_IS_OK(status)) {
1029 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
1030 goto fail;
1033 if (reply->hdr.operation != CTDB_REPLY_CALL) {
1034 DEBUG(0, ("received invalid reply\n"));
1035 status = NT_STATUS_INTERNAL_ERROR;
1036 goto fail;
1039 status = NT_STATUS_OK;
1040 fail:
1042 TALLOC_FREE(reply);
1043 return status;
1047 * remotely fetch a record without locking it or forcing a migration
1049 NTSTATUS ctdbd_fetch(struct ctdbd_connection *conn, uint32 db_id,
1050 TDB_DATA key, TALLOC_CTX *mem_ctx, TDB_DATA *data)
1052 struct ctdb_req_call req;
1053 struct ctdb_reply_call *reply;
1054 NTSTATUS status;
1056 ZERO_STRUCT(req);
1058 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
1059 req.hdr.ctdb_magic = CTDB_MAGIC;
1060 req.hdr.ctdb_version = CTDB_VERSION;
1061 req.hdr.operation = CTDB_REQ_CALL;
1062 req.hdr.reqid = ++conn->reqid;
1063 req.flags = 0;
1064 req.callid = CTDB_FETCH_FUNC;
1065 req.db_id = db_id;
1066 req.keylen = key.dsize;
1068 status = ctdb_packet_send(
1069 conn->pkt, 2,
1070 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
1071 data_blob_const(key.dptr, key.dsize));
1073 if (!NT_STATUS_IS_OK(status)) {
1074 DEBUG(3, ("ctdb_packet_send failed: %s\n", nt_errstr(status)));
1075 return status;
1078 status = ctdb_packet_flush(conn->pkt);
1080 if (!NT_STATUS_IS_OK(status)) {
1081 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
1082 cluster_fatal("cluster dispatch daemon control write error\n");
1085 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
1087 if (!NT_STATUS_IS_OK(status)) {
1088 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
1089 goto fail;
1092 if (reply->hdr.operation != CTDB_REPLY_CALL) {
1093 DEBUG(0, ("received invalid reply\n"));
1094 status = NT_STATUS_INTERNAL_ERROR;
1095 goto fail;
1098 data->dsize = reply->datalen;
1099 if (data->dsize == 0) {
1100 data->dptr = NULL;
1101 goto done;
1104 data->dptr = (uint8 *)talloc_memdup(mem_ctx, &reply->data[0],
1105 reply->datalen);
1106 if (data->dptr == NULL) {
1107 DEBUG(0, ("talloc failed\n"));
1108 status = NT_STATUS_NO_MEMORY;
1109 goto fail;
1112 done:
1113 status = NT_STATUS_OK;
1114 fail:
1115 TALLOC_FREE(reply);
1116 return status;
1119 struct ctdbd_traverse_state {
1120 void (*fn)(TDB_DATA key, TDB_DATA data, void *private_data);
1121 void *private_data;
1125 * Handle a traverse record coming in on the ctdbd connection
1128 static NTSTATUS ctdb_traverse_handler(uint8_t *buf, size_t length,
1129 void *private_data)
1131 struct ctdbd_traverse_state *state =
1132 (struct ctdbd_traverse_state *)private_data;
1134 struct ctdb_req_message *m;
1135 struct ctdb_rec_data *d;
1136 TDB_DATA key, data;
1138 m = (struct ctdb_req_message *)buf;
1140 if (length < sizeof(*m) || m->hdr.length != length) {
1141 DEBUG(0, ("Got invalid message of length %d\n", (int)length));
1142 TALLOC_FREE(buf);
1143 return NT_STATUS_UNEXPECTED_IO_ERROR;
1146 d = (struct ctdb_rec_data *)&m->data[0];
1147 if (m->datalen < sizeof(uint32_t) || m->datalen != d->length) {
1148 DEBUG(0, ("Got invalid traverse data of length %d\n",
1149 (int)m->datalen));
1150 TALLOC_FREE(buf);
1151 return NT_STATUS_UNEXPECTED_IO_ERROR;
1154 key.dsize = d->keylen;
1155 key.dptr = &d->data[0];
1156 data.dsize = d->datalen;
1157 data.dptr = &d->data[d->keylen];
1159 if (key.dsize == 0 && data.dsize == 0) {
1160 /* end of traverse */
1161 return NT_STATUS_END_OF_FILE;
1164 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1165 DEBUG(0, ("Got invalid ltdb header length %d\n",
1166 (int)data.dsize));
1167 TALLOC_FREE(buf);
1168 return NT_STATUS_UNEXPECTED_IO_ERROR;
1170 data.dsize -= sizeof(struct ctdb_ltdb_header);
1171 data.dptr += sizeof(struct ctdb_ltdb_header);
1173 if (state->fn) {
1174 state->fn(key, data, state->private_data);
1177 TALLOC_FREE(buf);
1178 return NT_STATUS_OK;
1182 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1183 connection to ctdbd to avoid the hairy recursive and async problems with
1184 everything in-line.
1187 NTSTATUS ctdbd_traverse(uint32 db_id,
1188 void (*fn)(TDB_DATA key, TDB_DATA data,
1189 void *private_data),
1190 void *private_data)
1192 struct ctdbd_connection *conn;
1193 NTSTATUS status;
1195 TDB_DATA data;
1196 struct ctdb_traverse_start t;
1197 int cstatus;
1198 struct ctdbd_traverse_state state;
1200 status = ctdbd_init_connection(NULL, &conn);
1201 if (!NT_STATUS_IS_OK(status)) {
1202 DEBUG(0, ("ctdbd_init_connection failed: %s\n",
1203 nt_errstr(status)));
1204 return status;
1207 t.db_id = db_id;
1208 t.srvid = conn->rand_srvid;
1209 t.reqid = ++conn->reqid;
1211 data.dptr = (uint8_t *)&t;
1212 data.dsize = sizeof(t);
1214 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1215 CTDB_CONTROL_TRAVERSE_START, conn->rand_srvid, 0,
1216 data, NULL, NULL, &cstatus);
1218 if (!NT_STATUS_IS_OK(status) || (cstatus != 0)) {
1220 DEBUG(0,("ctdbd_control failed: %s, %d\n", nt_errstr(status),
1221 cstatus));
1223 if (NT_STATUS_IS_OK(status)) {
1225 * We need a mapping here
1227 status = NT_STATUS_UNSUCCESSFUL;
1229 goto done;
1232 state.fn = fn;
1233 state.private_data = private_data;
1235 while (True) {
1237 status = NT_STATUS_OK;
1239 if (ctdb_packet_handler(conn->pkt, ctdb_req_complete,
1240 ctdb_traverse_handler, &state, &status)) {
1242 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1243 status = NT_STATUS_OK;
1244 break;
1248 * There might be more in the queue
1250 continue;
1253 if (!NT_STATUS_IS_OK(status)) {
1254 break;
1257 status = ctdb_packet_fd_read_sync(conn->pkt);
1259 if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
1261 * There might be more in the queue
1263 continue;
1266 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1267 status = NT_STATUS_OK;
1268 break;
1271 if (!NT_STATUS_IS_OK(status)) {
1272 DEBUG(0, ("ctdb_packet_fd_read_sync failed: %s\n", nt_errstr(status)));
1273 cluster_fatal("ctdbd died\n");
1277 done:
1278 TALLOC_FREE(conn);
1279 return status;
1283 This is used to canonicalize a ctdb_sock_addr structure.
1285 static void smbd_ctdb_canonicalize_ip(const struct sockaddr_storage *in,
1286 struct sockaddr_storage *out)
1288 memcpy(out, in, sizeof (*out));
1290 #ifdef HAVE_IPV6
1291 if (in->ss_family == AF_INET6) {
1292 const char prefix[12] = { 0,0,0,0,0,0,0,0,0,0,0xff,0xff };
1293 const struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)in;
1294 struct sockaddr_in *out4 = (struct sockaddr_in *)out;
1295 if (memcmp(&in6->sin6_addr, prefix, 12) == 0) {
1296 memset(out, 0, sizeof(*out));
1297 #ifdef HAVE_SOCK_SIN_LEN
1298 out4->sin_len = sizeof(*out);
1299 #endif
1300 out4->sin_family = AF_INET;
1301 out4->sin_port = in6->sin6_port;
1302 memcpy(&out4->sin_addr, &in6->sin6_addr.s6_addr32[3], 4);
1305 #endif
1309 * Register us as a server for a particular tcp connection
1312 NTSTATUS ctdbd_register_ips(struct ctdbd_connection *conn,
1313 const struct sockaddr_storage *_server,
1314 const struct sockaddr_storage *_client,
1315 void (*release_ip_handler)(const char *ip_addr,
1316 void *private_data),
1317 void *private_data)
1320 * we still use ctdb_control_tcp for ipv4
1321 * because we want to work against older ctdb
1322 * versions at runtime
1324 struct ctdb_control_tcp p4;
1325 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1326 struct ctdb_control_tcp_addr p;
1327 #endif
1328 TDB_DATA data;
1329 NTSTATUS status;
1330 struct sockaddr_storage client;
1331 struct sockaddr_storage server;
1334 * Only one connection so far
1336 SMB_ASSERT(conn->release_ip_handler == NULL);
1338 smbd_ctdb_canonicalize_ip(_client, &client);
1339 smbd_ctdb_canonicalize_ip(_server, &server);
1341 switch (client.ss_family) {
1342 case AF_INET:
1343 p4.dest = *(struct sockaddr_in *)(void *)&server;
1344 p4.src = *(struct sockaddr_in *)(void *)&client;
1345 data.dptr = (uint8_t *)&p4;
1346 data.dsize = sizeof(p4);
1347 break;
1348 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1349 case AF_INET6:
1350 p.dest.ip6 = *(struct sockaddr_in6 *)(void *)&server;
1351 p.src.ip6 = *(struct sockaddr_in6 *)(void *)&client;
1352 data.dptr = (uint8_t *)&p;
1353 data.dsize = sizeof(p);
1354 break;
1355 #endif
1356 default:
1357 return NT_STATUS_INTERNAL_ERROR;
1360 conn->release_ip_handler = release_ip_handler;
1362 * store the IP address of the server socket for later
1363 * comparison in release_ip()
1365 conn->release_ip_priv = private_data;
1368 * We want to be told about IP releases
1371 status = register_with_ctdbd(conn, CTDB_SRVID_RELEASE_IP);
1372 if (!NT_STATUS_IS_OK(status)) {
1373 return status;
1377 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1378 * can send an extra ack to trigger a reset for our client, so it
1379 * immediately reconnects
1381 return ctdbd_control(conn, CTDB_CURRENT_NODE,
1382 CTDB_CONTROL_TCP_CLIENT, 0,
1383 CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL, NULL);
1387 * We want to handle reconfigure events
1389 NTSTATUS ctdbd_register_reconfigure(struct ctdbd_connection *conn)
1391 return register_with_ctdbd(conn, CTDB_SRVID_RECONFIGURE);
1395 call a control on the local node
1397 NTSTATUS ctdbd_control_local(struct ctdbd_connection *conn, uint32 opcode,
1398 uint64_t srvid, uint32_t flags, TDB_DATA data,
1399 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
1400 int *cstatus)
1402 return ctdbd_control(conn, CTDB_CURRENT_NODE, opcode, srvid, flags, data, mem_ctx, outdata, cstatus);
1405 NTSTATUS ctdb_watch_us(struct ctdbd_connection *conn)
1407 struct ctdb_client_notify_register reg_data;
1408 size_t struct_len;
1409 NTSTATUS status;
1410 int cstatus;
1412 reg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1413 reg_data.len = 1;
1414 reg_data.notify_data[0] = 0;
1416 struct_len = offsetof(struct ctdb_client_notify_register,
1417 notify_data) + reg_data.len;
1419 status = ctdbd_control_local(
1420 conn, CTDB_CONTROL_REGISTER_NOTIFY, conn->rand_srvid, 0,
1421 make_tdb_data((uint8_t *)&reg_data, struct_len),
1422 NULL, NULL, &cstatus);
1423 if (!NT_STATUS_IS_OK(status)) {
1424 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1425 nt_errstr(status)));
1427 return status;
1430 NTSTATUS ctdb_unwatch(struct ctdbd_connection *conn)
1432 struct ctdb_client_notify_deregister dereg_data;
1433 NTSTATUS status;
1434 int cstatus;
1436 dereg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1438 status = ctdbd_control_local(
1439 conn, CTDB_CONTROL_DEREGISTER_NOTIFY, conn->rand_srvid, 0,
1440 make_tdb_data((uint8_t *)&dereg_data, sizeof(dereg_data)),
1441 NULL, NULL, &cstatus);
1442 if (!NT_STATUS_IS_OK(status)) {
1443 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1444 nt_errstr(status)));
1446 return status;
1449 #endif