packet.h API: The callback is now responsible to talloc_free() "buf"
[Samba.git] / source3 / lib / ctdbd_conn.c
blob75a513312e21f27d93f35f7d6af4f894f6f87f65
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"
28 /* paths to these include files come from --with-ctdb= in configure */
29 #include "ctdb.h"
30 #include "ctdb_private.h"
32 struct ctdbd_connection {
33 struct messaging_context *msg_ctx;
34 uint32 reqid;
35 uint32 our_vnn;
36 uint64 rand_srvid;
37 struct packet_context *pkt;
38 struct fd_event *fde;
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,
48 int *cstatus);
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
59 violations */
60 _exit(0);
66 static void ctdb_packet_dump(struct ctdb_req_header *hdr)
68 if (DEBUGLEVEL < 10) {
69 return;
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,
81 uint64_t srvid)
84 int cstatus;
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)
95 int32_t cstatus=-1;
96 NTSTATUS status;
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;
104 return status;
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;
122 int fd;
124 if (!sockname || !*sockname) {
125 sockname = CTDB_PATH;
128 fd = socket(AF_UNIX, SOCK_STREAM, 0);
129 if (fd == -1) {
130 DEBUG(3, ("Could not create socket: %s\n", strerror(errno)));
131 return map_nt_error_from_unix(errno);
134 ZERO_STRUCT(addr);
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,
140 strerror(errno)));
141 close(fd);
142 return map_nt_error_from_unix(errno);
145 if (!(result = packet_init(mem_ctx, fd))) {
146 close(fd);
147 return NT_STATUS_NO_MEMORY;
150 *presult = result;
151 return NT_STATUS_OK;
155 * Do we have a complete ctdb packet in the queue?
158 static bool ctdb_req_complete(const uint8_t *buf, size_t available,
159 size_t *length,
160 void *private_data)
162 uint32 msglen;
164 if (available < sizeof(msglen)) {
165 return False;
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) {
180 return false;
183 *length = msglen;
184 return true;
188 * State necessary to defer an incoming message while we are waiting for a
189 * ctdb reply.
192 struct deferred_msg_state {
193 struct messaging_context *msg_ctx;
194 struct messaging_rec *rec;
198 * Timed event handler for the deferred message
201 static void deferred_message_dispatch(struct event_context *event_ctx,
202 struct timed_event *te,
203 const struct timeval *now,
204 void *private_data)
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);
210 TALLOC_FREE(state);
211 TALLOC_FREE(te);
214 struct req_pull_state {
215 TALLOC_CTX *mem_ctx;
216 DATA_BLOB req;
220 * Pull a ctdb request out of the incoming packet queue
223 static NTSTATUS ctdb_req_pull(uint8_t *buf, size_t length,
224 void *private_data)
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;
230 return NT_STATUS_OK;
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;
242 DATA_BLOB blob;
243 enum ndr_err_code ndr_err;
245 if ((overall_length < offsetof(struct ctdb_req_message, data))
246 || (overall_length
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"));
254 return NULL;
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)));
266 TALLOC_FREE(result);
267 return NULL;
270 if (DEBUGLEVEL >= 10) {
271 DEBUG(10, ("ctdb_pull_messaging_rec:\n"));
272 NDR_PRINT_DEBUG(messaging_rec, result);
275 return result;
279 * Read a full ctdbd request. If we have a messaging context, defer incoming
280 * messages that might come in between.
283 static NTSTATUS ctdb_read_req(struct ctdbd_connection *conn, uint32 reqid,
284 TALLOC_CTX *mem_ctx, void *result)
286 struct ctdb_req_header *hdr;
287 struct req_pull_state state;
288 NTSTATUS status;
290 again:
292 status = packet_fd_read_sync(conn->pkt);
294 if (NT_STATUS_EQUAL(status, NT_STATUS_NETWORK_BUSY)) {
295 /* EAGAIN */
296 goto again;
297 } else if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
298 /* EAGAIN */
299 goto again;
302 if (!NT_STATUS_IS_OK(status)) {
303 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
304 cluster_fatal("ctdbd died\n");
307 next_pkt:
309 ZERO_STRUCT(state);
310 state.mem_ctx = mem_ctx;
312 if (!packet_handler(conn->pkt, ctdb_req_complete, ctdb_req_pull,
313 &state, &status)) {
315 * Not enough data
317 DEBUG(10, ("not enough data from ctdb socket, retrying\n"));
318 goto again;
321 if (!NT_STATUS_IS_OK(status)) {
322 DEBUG(0, ("Could not read packet: %s\n", nt_errstr(status)));
323 cluster_fatal("ctdbd died\n");
326 hdr = (struct ctdb_req_header *)state.req.data;
328 DEBUG(10, ("Received ctdb packet\n"));
329 ctdb_packet_dump(hdr);
331 if (hdr->operation == CTDB_REQ_MESSAGE) {
332 struct timed_event *evt;
333 struct deferred_msg_state *msg_state;
334 struct ctdb_req_message *msg = (struct ctdb_req_message *)hdr;
336 if (conn->msg_ctx == NULL) {
337 DEBUG(1, ("Got a message without having a msg ctx, "
338 "dropping msg %llu\n",
339 (long long unsigned)msg->srvid));
340 goto next_pkt;
343 if ((conn->release_ip_handler != NULL)
344 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
345 /* must be dispatched immediately */
346 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
347 conn->release_ip_handler((const char *)msg->data,
348 conn->release_ip_priv);
349 TALLOC_FREE(hdr);
350 goto next_pkt;
353 if (msg->srvid == CTDB_SRVID_RECONFIGURE) {
354 DEBUG(0,("Got cluster reconfigure message in ctdb_read_req\n"));
355 messaging_send(conn->msg_ctx, procid_self(),
356 MSG_SMB_BRL_VALIDATE, &data_blob_null);
357 TALLOC_FREE(hdr);
358 goto next_pkt;
361 if (!(msg_state = TALLOC_P(NULL, struct deferred_msg_state))) {
362 DEBUG(0, ("talloc failed\n"));
363 TALLOC_FREE(hdr);
364 goto next_pkt;
367 if (!(msg_state->rec = ctdb_pull_messaging_rec(
368 msg_state, state.req.length, msg))) {
369 DEBUG(0, ("ctdbd_pull_messaging_rec failed\n"));
370 TALLOC_FREE(msg_state);
371 TALLOC_FREE(hdr);
372 goto next_pkt;
375 TALLOC_FREE(hdr);
377 msg_state->msg_ctx = conn->msg_ctx;
380 * We're waiting for a call reply, but an async message has
381 * crossed. Defer dispatching to the toplevel event loop.
383 evt = event_add_timed(conn->msg_ctx->event_ctx,
384 conn->msg_ctx->event_ctx,
385 timeval_zero(),
386 "deferred_message_dispatch",
387 deferred_message_dispatch,
388 msg_state);
389 if (evt == NULL) {
390 DEBUG(0, ("event_add_timed failed\n"));
391 TALLOC_FREE(msg_state);
392 TALLOC_FREE(hdr);
393 goto next_pkt;
396 goto next_pkt;
399 if (hdr->reqid != reqid) {
400 /* we got the wrong reply */
401 DEBUG(0,("Discarding mismatched ctdb reqid %u should have "
402 "been %u\n", hdr->reqid, reqid));
403 TALLOC_FREE(hdr);
404 goto again;
407 *((void **)result) = talloc_move(mem_ctx, &hdr);
409 return NT_STATUS_OK;
413 * Get us a ctdbd connection
416 NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
417 struct ctdbd_connection **pconn)
419 struct ctdbd_connection *conn;
420 NTSTATUS status;
422 if (!(conn = TALLOC_ZERO_P(mem_ctx, struct ctdbd_connection))) {
423 DEBUG(0, ("talloc failed\n"));
424 return NT_STATUS_NO_MEMORY;
427 status = ctdbd_connect(conn, &conn->pkt);
429 if (!NT_STATUS_IS_OK(status)) {
430 DEBUG(10, ("ctdbd_connect failed: %s\n", nt_errstr(status)));
431 goto fail;
434 status = get_cluster_vnn(conn, &conn->our_vnn);
436 if (!NT_STATUS_IS_OK(status)) {
437 DEBUG(10, ("get_cluster_vnn failed: %s\n", nt_errstr(status)));
438 goto fail;
441 generate_random_buffer((unsigned char *)&conn->rand_srvid,
442 sizeof(conn->rand_srvid));
444 status = register_with_ctdbd(conn, conn->rand_srvid);
446 if (!NT_STATUS_IS_OK(status)) {
447 DEBUG(5, ("Could not register random srvid: %s\n",
448 nt_errstr(status)));
449 goto fail;
452 *pconn = conn;
453 return NT_STATUS_OK;
455 fail:
456 TALLOC_FREE(conn);
457 return status;
461 * Get us a ctdbd connection and register us as a process
464 NTSTATUS ctdbd_messaging_connection(TALLOC_CTX *mem_ctx,
465 struct ctdbd_connection **pconn)
467 struct ctdbd_connection *conn;
468 NTSTATUS status;
470 status = ctdbd_init_connection(mem_ctx, &conn);
472 if (!NT_STATUS_IS_OK(status)) {
473 return status;
476 status = register_with_ctdbd(conn, (uint64_t)sys_getpid());
477 if (!NT_STATUS_IS_OK(status)) {
478 goto fail;
481 status = register_with_ctdbd(conn, MSG_SRVID_SAMBA);
482 if (!NT_STATUS_IS_OK(status)) {
483 goto fail;
486 *pconn = conn;
487 return NT_STATUS_OK;
489 fail:
490 TALLOC_FREE(conn);
491 return status;
495 * Packet handler to receive and handle a ctdb message
497 static NTSTATUS ctdb_handle_message(uint8_t *buf, size_t length,
498 void *private_data)
500 struct ctdbd_connection *conn = talloc_get_type_abort(
501 private_data, struct ctdbd_connection);
502 struct ctdb_req_message *msg;
503 struct messaging_rec *msg_rec;
505 msg = (struct ctdb_req_message *)buf;
507 if (msg->hdr.operation != CTDB_REQ_MESSAGE) {
508 DEBUG(0, ("Received async msg of type %u, discarding\n",
509 msg->hdr.operation));
510 TALLOC_FREE(buf);
511 return NT_STATUS_INVALID_PARAMETER;
514 if ((conn->release_ip_handler != NULL)
515 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
516 /* must be dispatched immediately */
517 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
518 conn->release_ip_handler((const char *)msg->data,
519 conn->release_ip_priv);
520 TALLOC_FREE(buf);
521 return NT_STATUS_OK;
524 SMB_ASSERT(conn->msg_ctx != NULL);
526 if (msg->srvid == CTDB_SRVID_RECONFIGURE) {
527 DEBUG(0,("Got cluster reconfigure message\n"));
529 * when the cluster is reconfigured, we need to clean the brl
530 * database
532 messaging_send(conn->msg_ctx, procid_self(),
533 MSG_SMB_BRL_VALIDATE, &data_blob_null);
536 * it's possible that we have just rejoined the cluster after
537 * an outage. In that case our pending locks could have been
538 * removed from the lockdb, so retry them once more
540 message_send_all(conn->msg_ctx, MSG_SMB_UNLOCK, NULL, 0, NULL);
542 TALLOC_FREE(buf);
544 return NT_STATUS_OK;
548 /* only messages to our pid or the broadcast are valid here */
549 if (msg->srvid != sys_getpid() && msg->srvid != MSG_SRVID_SAMBA) {
550 DEBUG(0,("Got unexpected message with srvid=%llu\n",
551 (unsigned long long)msg->srvid));
552 TALLOC_FREE(buf);
553 return NT_STATUS_OK;
556 if (!(msg_rec = ctdb_pull_messaging_rec(NULL, length, msg))) {
557 DEBUG(10, ("ctdb_pull_messaging_rec failed\n"));
558 TALLOC_FREE(buf);
559 return NT_STATUS_NO_MEMORY;
562 messaging_dispatch_rec(conn->msg_ctx, msg_rec);
564 TALLOC_FREE(msg_rec);
565 TALLOC_FREE(buf);
566 return NT_STATUS_OK;
570 * The ctdbd socket is readable asynchronuously
573 static void ctdbd_socket_handler(struct event_context *event_ctx,
574 struct fd_event *event,
575 uint16 flags,
576 void *private_data)
578 struct ctdbd_connection *conn = talloc_get_type_abort(
579 private_data, struct ctdbd_connection);
581 NTSTATUS status;
583 status = packet_fd_read(conn->pkt);
585 if (!NT_STATUS_IS_OK(status)) {
586 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
587 cluster_fatal("ctdbd died\n");
590 while (packet_handler(conn->pkt, ctdb_req_complete,
591 ctdb_handle_message, conn, &status)) {
592 if (!NT_STATUS_IS_OK(status)) {
593 DEBUG(10, ("could not handle incoming message: %s\n",
594 nt_errstr(status)));
600 * Prepare a ctdbd connection to receive messages
603 NTSTATUS ctdbd_register_msg_ctx(struct ctdbd_connection *conn,
604 struct messaging_context *msg_ctx)
606 SMB_ASSERT(conn->msg_ctx == NULL);
607 SMB_ASSERT(conn->fde == NULL);
609 if (!(conn->fde = event_add_fd(msg_ctx->event_ctx, conn,
610 packet_get_fd(conn->pkt),
611 EVENT_FD_READ,
612 ctdbd_socket_handler,
613 conn))) {
614 DEBUG(0, ("event_add_fd failed\n"));
615 return NT_STATUS_NO_MEMORY;
618 conn->msg_ctx = msg_ctx;
620 return NT_STATUS_OK;
624 * Send a messaging message across a ctdbd
627 NTSTATUS ctdbd_messaging_send(struct ctdbd_connection *conn,
628 uint32 dst_vnn, uint64 dst_srvid,
629 struct messaging_rec *msg)
631 struct ctdb_req_message r;
632 TALLOC_CTX *mem_ctx;
633 DATA_BLOB blob;
634 NTSTATUS status;
635 enum ndr_err_code ndr_err;
637 if (!(mem_ctx = talloc_init("ctdbd_messaging_send"))) {
638 DEBUG(0, ("talloc failed\n"));
639 return NT_STATUS_NO_MEMORY;
642 ndr_err = ndr_push_struct_blob(
643 &blob, mem_ctx, NULL, msg,
644 (ndr_push_flags_fn_t)ndr_push_messaging_rec);
646 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
647 DEBUG(0, ("ndr_push_struct_blob failed: %s\n",
648 ndr_errstr(ndr_err)));
649 status = ndr_map_error2ntstatus(ndr_err);
650 goto fail;
653 r.hdr.length = offsetof(struct ctdb_req_message, data) + blob.length;
654 r.hdr.ctdb_magic = CTDB_MAGIC;
655 r.hdr.ctdb_version = CTDB_VERSION;
656 r.hdr.generation = 1;
657 r.hdr.operation = CTDB_REQ_MESSAGE;
658 r.hdr.destnode = dst_vnn;
659 r.hdr.srcnode = conn->our_vnn;
660 r.hdr.reqid = 0;
661 r.srvid = dst_srvid;
662 r.datalen = blob.length;
664 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
665 ctdb_packet_dump(&r.hdr);
667 status = packet_send(
668 conn->pkt, 2,
669 data_blob_const(&r, offsetof(struct ctdb_req_message, data)),
670 blob);
672 if (!NT_STATUS_IS_OK(status)) {
673 DEBUG(0, ("packet_send failed: %s\n", nt_errstr(status)));
674 goto fail;
677 status = packet_flush(conn->pkt);
679 if (!NT_STATUS_IS_OK(status)) {
680 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
681 cluster_fatal("cluster dispatch daemon msg write error\n");
684 status = NT_STATUS_OK;
685 fail:
686 TALLOC_FREE(mem_ctx);
687 return status;
691 * send/recv a generic ctdb control message
693 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
694 uint32_t vnn, uint32 opcode,
695 uint64_t srvid, uint32_t flags,
696 TDB_DATA data,
697 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
698 int *cstatus)
700 struct ctdb_req_control req;
701 struct ctdb_reply_control *reply = NULL;
702 struct ctdbd_connection *new_conn = NULL;
703 NTSTATUS status;
705 /* the samba3 ctdb code can't handle NOREPLY yet */
706 flags &= ~CTDB_CTRL_FLAG_NOREPLY;
708 if (conn == NULL) {
709 status = ctdbd_init_connection(NULL, &new_conn);
711 if (!NT_STATUS_IS_OK(status)) {
712 DEBUG(10, ("Could not init temp connection: %s\n",
713 nt_errstr(status)));
714 goto fail;
717 conn = new_conn;
720 ZERO_STRUCT(req);
721 req.hdr.length = offsetof(struct ctdb_req_control, data) + data.dsize;
722 req.hdr.ctdb_magic = CTDB_MAGIC;
723 req.hdr.ctdb_version = CTDB_VERSION;
724 req.hdr.operation = CTDB_REQ_CONTROL;
725 req.hdr.reqid = ++conn->reqid;
726 req.hdr.destnode = vnn;
727 req.opcode = opcode;
728 req.srvid = srvid;
729 req.datalen = data.dsize;
731 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
732 ctdb_packet_dump(&req.hdr);
734 status = packet_send(
735 conn->pkt, 2,
736 data_blob_const(&req, offsetof(struct ctdb_req_control, data)),
737 data_blob_const(data.dptr, data.dsize));
739 if (!NT_STATUS_IS_OK(status)) {
740 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
741 goto fail;
744 status = packet_flush(conn->pkt);
746 if (!NT_STATUS_IS_OK(status)) {
747 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
748 cluster_fatal("cluster dispatch daemon control write error\n");
751 if (flags & CTDB_CTRL_FLAG_NOREPLY) {
752 TALLOC_FREE(new_conn);
753 return NT_STATUS_OK;
756 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
758 if (!NT_STATUS_IS_OK(status)) {
759 DEBUG(10, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
760 goto fail;
763 if (reply->hdr.operation != CTDB_REPLY_CONTROL) {
764 DEBUG(0, ("received invalid reply\n"));
765 goto fail;
768 if (outdata) {
769 if (!(outdata->dptr = (uint8 *)talloc_memdup(
770 mem_ctx, reply->data, reply->datalen))) {
771 TALLOC_FREE(reply);
772 return NT_STATUS_NO_MEMORY;
774 outdata->dsize = reply->datalen;
776 if (cstatus) {
777 (*cstatus) = reply->status;
780 status = NT_STATUS_OK;
782 fail:
783 TALLOC_FREE(new_conn);
784 TALLOC_FREE(reply);
785 return status;
789 * see if a remote process exists
791 bool ctdbd_process_exists(struct ctdbd_connection *conn, uint32 vnn, pid_t pid)
793 NTSTATUS status;
794 TDB_DATA data;
795 int32_t cstatus;
797 data.dptr = (uint8_t*)&pid;
798 data.dsize = sizeof(pid);
800 status = ctdbd_control(conn, vnn, CTDB_CONTROL_PROCESS_EXISTS, 0, 0,
801 data, NULL, NULL, &cstatus);
802 if (!NT_STATUS_IS_OK(status)) {
803 DEBUG(0, (__location__ " ctdb_control for process_exists "
804 "failed\n"));
805 return False;
808 return cstatus == 0;
812 * Get a db path
814 char *ctdbd_dbpath(struct ctdbd_connection *conn,
815 TALLOC_CTX *mem_ctx, uint32_t db_id)
817 NTSTATUS status;
818 TDB_DATA data;
819 int32_t cstatus;
821 data.dptr = (uint8_t*)&db_id;
822 data.dsize = sizeof(db_id);
824 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
825 CTDB_CONTROL_GETDBPATH, 0, 0, data,
826 mem_ctx, &data, &cstatus);
827 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
828 DEBUG(0,(__location__ " ctdb_control for getdbpath failed\n"));
829 return NULL;
832 return (char *)data.dptr;
836 * attach to a ctdb database
838 NTSTATUS ctdbd_db_attach(struct ctdbd_connection *conn,
839 const char *name, uint32_t *db_id, int tdb_flags)
841 NTSTATUS status;
842 TDB_DATA data;
843 int32_t cstatus;
844 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
846 data.dptr = (uint8_t*)name;
847 data.dsize = strlen(name)+1;
849 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
850 persistent
851 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
852 : CTDB_CONTROL_DB_ATTACH,
853 0, 0, data, NULL, &data, &cstatus);
854 if (!NT_STATUS_IS_OK(status)) {
855 DEBUG(0, (__location__ " ctdb_control for db_attach "
856 "failed: %s\n", nt_errstr(status)));
857 return status;
860 if (cstatus != 0 || data.dsize != sizeof(uint32_t)) {
861 DEBUG(0,(__location__ " ctdb_control for db_attach failed\n"));
862 return NT_STATUS_INTERNAL_ERROR;
865 *db_id = *(uint32_t *)data.dptr;
866 talloc_free(data.dptr);
868 if (!(tdb_flags & TDB_SEQNUM)) {
869 return NT_STATUS_OK;
872 data.dptr = (uint8_t *)db_id;
873 data.dsize = sizeof(*db_id);
875 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
876 CTDB_CONTROL_ENABLE_SEQNUM, 0, 0, data,
877 NULL, NULL, &cstatus);
878 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
879 DEBUG(0,(__location__ " ctdb_control for enable seqnum "
880 "failed\n"));
881 return NT_STATUS_IS_OK(status) ? NT_STATUS_INTERNAL_ERROR :
882 status;
885 return NT_STATUS_OK;
889 * force the migration of a record to this node
891 NTSTATUS ctdbd_migrate(struct ctdbd_connection *conn, uint32 db_id,
892 TDB_DATA key)
894 struct ctdb_req_call req;
895 struct ctdb_reply_call *reply;
896 NTSTATUS status;
898 ZERO_STRUCT(req);
900 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
901 req.hdr.ctdb_magic = CTDB_MAGIC;
902 req.hdr.ctdb_version = CTDB_VERSION;
903 req.hdr.operation = CTDB_REQ_CALL;
904 req.hdr.reqid = ++conn->reqid;
905 req.flags = CTDB_IMMEDIATE_MIGRATION;
906 req.callid = CTDB_NULL_FUNC;
907 req.db_id = db_id;
908 req.keylen = key.dsize;
910 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
911 ctdb_packet_dump(&req.hdr);
913 status = packet_send(
914 conn->pkt, 2,
915 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
916 data_blob_const(key.dptr, key.dsize));
918 if (!NT_STATUS_IS_OK(status)) {
919 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
920 return status;
923 status = packet_flush(conn->pkt);
925 if (!NT_STATUS_IS_OK(status)) {
926 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
927 cluster_fatal("cluster dispatch daemon control write error\n");
930 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
932 if (!NT_STATUS_IS_OK(status)) {
933 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
934 goto fail;
937 if (reply->hdr.operation != CTDB_REPLY_CALL) {
938 DEBUG(0, ("received invalid reply\n"));
939 status = NT_STATUS_INTERNAL_ERROR;
940 goto fail;
943 status = NT_STATUS_OK;
944 fail:
946 TALLOC_FREE(reply);
947 return status;
951 * remotely fetch a record without locking it or forcing a migration
953 NTSTATUS ctdbd_fetch(struct ctdbd_connection *conn, uint32 db_id,
954 TDB_DATA key, TALLOC_CTX *mem_ctx, TDB_DATA *data)
956 struct ctdb_req_call req;
957 struct ctdb_reply_call *reply;
958 NTSTATUS status;
960 ZERO_STRUCT(req);
962 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
963 req.hdr.ctdb_magic = CTDB_MAGIC;
964 req.hdr.ctdb_version = CTDB_VERSION;
965 req.hdr.operation = CTDB_REQ_CALL;
966 req.hdr.reqid = ++conn->reqid;
967 req.flags = 0;
968 req.callid = CTDB_FETCH_FUNC;
969 req.db_id = db_id;
970 req.keylen = key.dsize;
972 status = packet_send(
973 conn->pkt, 2,
974 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
975 data_blob_const(key.dptr, key.dsize));
977 if (!NT_STATUS_IS_OK(status)) {
978 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
979 return status;
982 status = packet_flush(conn->pkt);
984 if (!NT_STATUS_IS_OK(status)) {
985 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
986 cluster_fatal("cluster dispatch daemon control write error\n");
989 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
991 if (!NT_STATUS_IS_OK(status)) {
992 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
993 goto fail;
996 if (reply->hdr.operation != CTDB_REPLY_CALL) {
997 DEBUG(0, ("received invalid reply\n"));
998 status = NT_STATUS_INTERNAL_ERROR;
999 goto fail;
1002 data->dsize = reply->datalen;
1003 if (data->dsize == 0) {
1004 data->dptr = NULL;
1005 goto done;
1008 data->dptr = (uint8 *)talloc_memdup(mem_ctx, &reply->data[0],
1009 reply->datalen);
1010 if (data->dptr == NULL) {
1011 DEBUG(0, ("talloc failed\n"));
1012 status = NT_STATUS_NO_MEMORY;
1013 goto fail;
1016 done:
1017 status = NT_STATUS_OK;
1018 fail:
1019 TALLOC_FREE(reply);
1020 return status;
1023 struct ctdbd_traverse_state {
1024 void (*fn)(TDB_DATA key, TDB_DATA data, void *private_data);
1025 void *private_data;
1029 * Handle a traverse record coming in on the ctdbd connection
1032 static NTSTATUS ctdb_traverse_handler(uint8_t *buf, size_t length,
1033 void *private_data)
1035 struct ctdbd_traverse_state *state =
1036 (struct ctdbd_traverse_state *)private_data;
1038 struct ctdb_req_message *m;
1039 struct ctdb_rec_data *d;
1040 TDB_DATA key, data;
1042 m = (struct ctdb_req_message *)buf;
1044 if (length < sizeof(*m) || m->hdr.length != length) {
1045 DEBUG(0, ("Got invalid message of length %d\n", (int)length));
1046 TALLOC_FREE(buf);
1047 return NT_STATUS_UNEXPECTED_IO_ERROR;
1050 d = (struct ctdb_rec_data *)&m->data[0];
1051 if (m->datalen < sizeof(uint32_t) || m->datalen != d->length) {
1052 DEBUG(0, ("Got invalid traverse data of length %d\n",
1053 (int)m->datalen));
1054 TALLOC_FREE(buf);
1055 return NT_STATUS_UNEXPECTED_IO_ERROR;
1058 key.dsize = d->keylen;
1059 key.dptr = &d->data[0];
1060 data.dsize = d->datalen;
1061 data.dptr = &d->data[d->keylen];
1063 if (key.dsize == 0 && data.dsize == 0) {
1064 /* end of traverse */
1065 return NT_STATUS_END_OF_FILE;
1068 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1069 DEBUG(0, ("Got invalid ltdb header length %d\n",
1070 (int)data.dsize));
1071 TALLOC_FREE(buf);
1072 return NT_STATUS_UNEXPECTED_IO_ERROR;
1074 data.dsize -= sizeof(struct ctdb_ltdb_header);
1075 data.dptr += sizeof(struct ctdb_ltdb_header);
1077 if (state->fn) {
1078 state->fn(key, data, state->private_data);
1081 TALLOC_FREE(buf);
1082 return NT_STATUS_OK;
1086 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1087 connection to ctdbd to avoid the hairy recursive and async problems with
1088 everything in-line.
1091 NTSTATUS ctdbd_traverse(uint32 db_id,
1092 void (*fn)(TDB_DATA key, TDB_DATA data,
1093 void *private_data),
1094 void *private_data)
1096 struct ctdbd_connection *conn;
1097 NTSTATUS status;
1099 TDB_DATA data;
1100 struct ctdb_traverse_start t;
1101 int cstatus;
1102 struct ctdbd_traverse_state state;
1104 status = ctdbd_init_connection(NULL, &conn);
1106 t.db_id = db_id;
1107 t.srvid = conn->rand_srvid;
1108 t.reqid = ++conn->reqid;
1110 data.dptr = (uint8_t *)&t;
1111 data.dsize = sizeof(t);
1113 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1114 CTDB_CONTROL_TRAVERSE_START, conn->rand_srvid, 0,
1115 data, NULL, NULL, &cstatus);
1117 if (!NT_STATUS_IS_OK(status) || (cstatus != 0)) {
1119 DEBUG(0,("ctdbd_control failed: %s, %d\n", nt_errstr(status),
1120 cstatus));
1122 if (NT_STATUS_IS_OK(status)) {
1124 * We need a mapping here
1126 status = NT_STATUS_UNSUCCESSFUL;
1128 goto done;
1131 state.fn = fn;
1132 state.private_data = private_data;
1134 while (True) {
1136 status = NT_STATUS_OK;
1138 if (packet_handler(conn->pkt, ctdb_req_complete,
1139 ctdb_traverse_handler, &state, &status)) {
1141 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1142 status = NT_STATUS_OK;
1143 break;
1147 * There might be more in the queue
1149 continue;
1152 if (!NT_STATUS_IS_OK(status)) {
1153 break;
1156 status = packet_fd_read_sync(conn->pkt);
1158 if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
1160 * There might be more in the queue
1162 continue;
1165 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1166 status = NT_STATUS_OK;
1169 if (!NT_STATUS_IS_OK(status)) {
1170 DEBUG(0, ("packet_fd_read_sync failed: %s\n", nt_errstr(status)));
1171 cluster_fatal("ctdbd died\n");
1175 done:
1176 TALLOC_FREE(conn);
1177 return status;
1181 * Register us as a server for a particular tcp connection
1184 NTSTATUS ctdbd_register_ips(struct ctdbd_connection *conn,
1185 const struct sockaddr *server,
1186 const struct sockaddr *client,
1187 void (*release_ip_handler)(const char *ip_addr,
1188 void *private_data),
1189 void *private_data)
1191 struct ctdb_control_tcp_vnn p;
1192 TDB_DATA data;
1193 NTSTATUS status;
1196 * Only one connection so far
1198 SMB_ASSERT(conn->release_ip_handler == NULL);
1200 switch (client->sa_family) {
1201 case AF_INET:
1202 p.dest.ip = *(struct sockaddr_in *)server;
1203 p.src.ip = *(struct sockaddr_in *)client;
1204 break;
1205 #ifdef HAVE_IPV6
1206 case AF_INET6:
1207 p.dest.ip6 = *(struct sockaddr_in6 *)server;
1208 p.src.ip6 = *(struct sockaddr_in6 *)client;
1209 break;
1210 #endif
1211 default:
1212 return NT_STATUS_INTERNAL_ERROR;
1215 conn->release_ip_handler = release_ip_handler;
1218 * We want to be told about IP releases
1221 status = register_with_ctdbd(conn, CTDB_SRVID_RELEASE_IP);
1222 if (!NT_STATUS_IS_OK(status)) {
1223 return status;
1227 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1228 * can send an extra ack to trigger a reset for our client, so it
1229 * immediately reconnects
1231 data.dptr = (uint8_t *)&p;
1232 data.dsize = sizeof(p);
1234 return ctdbd_control(conn, CTDB_CURRENT_NODE,
1235 CTDB_CONTROL_TCP_ADD, 0,
1236 CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL, NULL);
1240 * We want to handle reconfigure events
1242 NTSTATUS ctdbd_register_reconfigure(struct ctdbd_connection *conn)
1244 return register_with_ctdbd(conn, CTDB_SRVID_RECONFIGURE);
1248 call a control on the local node
1250 NTSTATUS ctdbd_control_local(struct ctdbd_connection *conn, uint32 opcode,
1251 uint64_t srvid, uint32_t flags, TDB_DATA data,
1252 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
1253 int *cstatus)
1255 return ctdbd_control(conn, CTDB_CURRENT_NODE, opcode, srvid, flags, data, mem_ctx, outdata, cstatus);
1258 #else
1260 NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
1261 struct ctdbd_connection **pconn)
1263 return NT_STATUS_NOT_IMPLEMENTED;
1266 #endif