s3-brlock: we don't need these MSG_SMB_UNLOCK calls now
[Samba/gbeck.git] / source3 / lib / ctdbd_conn.c
blob8ddb12a765bd296fdc42752b2df439552d0d71fd
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 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;
278 static NTSTATUS ctdb_packet_fd_read_sync(struct packet_context *ctx)
280 struct timeval timeout;
281 struct timeval *ptimeout;
283 timeout = timeval_set(lp_ctdb_timeout(), 0);
284 ptimeout = (timeout.tv_sec != 0) ? &timeout : NULL;
286 return packet_fd_read_sync(ctx, ptimeout);
290 * Read a full ctdbd request. If we have a messaging context, defer incoming
291 * messages that might come in between.
294 static NTSTATUS ctdb_read_req(struct ctdbd_connection *conn, uint32 reqid,
295 TALLOC_CTX *mem_ctx, void *result)
297 struct ctdb_req_header *hdr;
298 struct req_pull_state state;
299 NTSTATUS status;
301 again:
303 status = ctdb_packet_fd_read_sync(conn->pkt);
305 if (NT_STATUS_EQUAL(status, NT_STATUS_NETWORK_BUSY)) {
306 /* EAGAIN */
307 goto again;
308 } else if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
309 /* EAGAIN */
310 goto again;
313 if (!NT_STATUS_IS_OK(status)) {
314 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
315 cluster_fatal("ctdbd died\n");
318 next_pkt:
320 ZERO_STRUCT(state);
321 state.mem_ctx = mem_ctx;
323 if (!packet_handler(conn->pkt, ctdb_req_complete, ctdb_req_pull,
324 &state, &status)) {
326 * Not enough data
328 DEBUG(10, ("not enough data from ctdb socket, retrying\n"));
329 goto again;
332 if (!NT_STATUS_IS_OK(status)) {
333 DEBUG(0, ("Could not read packet: %s\n", nt_errstr(status)));
334 cluster_fatal("ctdbd died\n");
337 hdr = (struct ctdb_req_header *)state.req.data;
339 DEBUG(10, ("Received ctdb packet\n"));
340 ctdb_packet_dump(hdr);
342 if (hdr->operation == CTDB_REQ_MESSAGE) {
343 struct timed_event *evt;
344 struct deferred_msg_state *msg_state;
345 struct ctdb_req_message *msg = (struct ctdb_req_message *)hdr;
347 if (conn->msg_ctx == NULL) {
348 DEBUG(1, ("Got a message without having a msg ctx, "
349 "dropping msg %llu\n",
350 (long long unsigned)msg->srvid));
351 goto next_pkt;
354 if ((conn->release_ip_handler != NULL)
355 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
356 /* must be dispatched immediately */
357 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
358 conn->release_ip_handler((const char *)msg->data,
359 conn->release_ip_priv);
360 TALLOC_FREE(hdr);
361 goto next_pkt;
364 if (msg->srvid == CTDB_SRVID_RECONFIGURE) {
365 DEBUG(0,("Got cluster reconfigure message in ctdb_read_req\n"));
366 messaging_send(conn->msg_ctx, procid_self(),
367 MSG_SMB_BRL_VALIDATE, &data_blob_null);
368 TALLOC_FREE(hdr);
369 goto next_pkt;
372 if (!(msg_state = TALLOC_P(talloc_autofree_context(), struct deferred_msg_state))) {
373 DEBUG(0, ("talloc failed\n"));
374 TALLOC_FREE(hdr);
375 goto next_pkt;
378 if (!(msg_state->rec = ctdb_pull_messaging_rec(
379 msg_state, state.req.length, msg))) {
380 DEBUG(0, ("ctdbd_pull_messaging_rec failed\n"));
381 TALLOC_FREE(msg_state);
382 TALLOC_FREE(hdr);
383 goto next_pkt;
386 TALLOC_FREE(hdr);
388 msg_state->msg_ctx = conn->msg_ctx;
391 * We're waiting for a call reply, but an async message has
392 * crossed. Defer dispatching to the toplevel event loop.
394 evt = event_add_timed(conn->msg_ctx->event_ctx,
395 conn->msg_ctx->event_ctx,
396 timeval_zero(),
397 deferred_message_dispatch,
398 msg_state);
399 if (evt == NULL) {
400 DEBUG(0, ("event_add_timed failed\n"));
401 TALLOC_FREE(msg_state);
402 TALLOC_FREE(hdr);
403 goto next_pkt;
406 goto next_pkt;
409 if (hdr->reqid != reqid) {
410 /* we got the wrong reply */
411 DEBUG(0,("Discarding mismatched ctdb reqid %u should have "
412 "been %u\n", hdr->reqid, reqid));
413 TALLOC_FREE(hdr);
414 goto again;
417 *((void **)result) = talloc_move(mem_ctx, &hdr);
419 return NT_STATUS_OK;
423 * Get us a ctdbd connection
426 NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
427 struct ctdbd_connection **pconn)
429 struct ctdbd_connection *conn;
430 NTSTATUS status;
432 if (!(conn = TALLOC_ZERO_P(mem_ctx, struct ctdbd_connection))) {
433 DEBUG(0, ("talloc failed\n"));
434 return NT_STATUS_NO_MEMORY;
437 status = ctdbd_connect(conn, &conn->pkt);
439 if (!NT_STATUS_IS_OK(status)) {
440 DEBUG(10, ("ctdbd_connect failed: %s\n", nt_errstr(status)));
441 goto fail;
444 status = get_cluster_vnn(conn, &conn->our_vnn);
446 if (!NT_STATUS_IS_OK(status)) {
447 DEBUG(10, ("get_cluster_vnn failed: %s\n", nt_errstr(status)));
448 goto fail;
451 generate_random_buffer((unsigned char *)&conn->rand_srvid,
452 sizeof(conn->rand_srvid));
454 status = register_with_ctdbd(conn, conn->rand_srvid);
456 if (!NT_STATUS_IS_OK(status)) {
457 DEBUG(5, ("Could not register random srvid: %s\n",
458 nt_errstr(status)));
459 goto fail;
462 *pconn = conn;
463 return NT_STATUS_OK;
465 fail:
466 TALLOC_FREE(conn);
467 return status;
471 * Get us a ctdbd connection and register us as a process
474 NTSTATUS ctdbd_messaging_connection(TALLOC_CTX *mem_ctx,
475 struct ctdbd_connection **pconn)
477 struct ctdbd_connection *conn;
478 NTSTATUS status;
480 status = ctdbd_init_connection(mem_ctx, &conn);
482 if (!NT_STATUS_IS_OK(status)) {
483 return status;
486 status = register_with_ctdbd(conn, (uint64_t)sys_getpid());
487 if (!NT_STATUS_IS_OK(status)) {
488 goto fail;
491 status = register_with_ctdbd(conn, MSG_SRVID_SAMBA);
492 if (!NT_STATUS_IS_OK(status)) {
493 goto fail;
496 *pconn = conn;
497 return NT_STATUS_OK;
499 fail:
500 TALLOC_FREE(conn);
501 return status;
505 * Packet handler to receive and handle a ctdb message
507 static NTSTATUS ctdb_handle_message(uint8_t *buf, size_t length,
508 void *private_data)
510 struct ctdbd_connection *conn = talloc_get_type_abort(
511 private_data, struct ctdbd_connection);
512 struct ctdb_req_message *msg;
513 struct messaging_rec *msg_rec;
515 msg = (struct ctdb_req_message *)buf;
517 if (msg->hdr.operation != CTDB_REQ_MESSAGE) {
518 DEBUG(0, ("Received async msg of type %u, discarding\n",
519 msg->hdr.operation));
520 TALLOC_FREE(buf);
521 return NT_STATUS_INVALID_PARAMETER;
524 if ((conn->release_ip_handler != NULL)
525 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
526 /* must be dispatched immediately */
527 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
528 conn->release_ip_handler((const char *)msg->data,
529 conn->release_ip_priv);
530 TALLOC_FREE(buf);
531 return NT_STATUS_OK;
534 SMB_ASSERT(conn->msg_ctx != NULL);
536 if (msg->srvid == CTDB_SRVID_RECONFIGURE) {
537 DEBUG(0,("Got cluster reconfigure message\n"));
539 * when the cluster is reconfigured, we need to clean the brl
540 * database
542 messaging_send(conn->msg_ctx, procid_self(),
543 MSG_SMB_BRL_VALIDATE, &data_blob_null);
545 TALLOC_FREE(buf);
546 return NT_STATUS_OK;
549 /* only messages to our pid or the broadcast are valid here */
550 if (msg->srvid != sys_getpid() && msg->srvid != MSG_SRVID_SAMBA) {
551 DEBUG(0,("Got unexpected message with srvid=%llu\n",
552 (unsigned long long)msg->srvid));
553 TALLOC_FREE(buf);
554 return NT_STATUS_OK;
557 if (!(msg_rec = ctdb_pull_messaging_rec(NULL, length, msg))) {
558 DEBUG(10, ("ctdb_pull_messaging_rec failed\n"));
559 TALLOC_FREE(buf);
560 return NT_STATUS_NO_MEMORY;
563 messaging_dispatch_rec(conn->msg_ctx, msg_rec);
565 TALLOC_FREE(msg_rec);
566 TALLOC_FREE(buf);
567 return NT_STATUS_OK;
571 * The ctdbd socket is readable asynchronuously
574 static void ctdbd_socket_handler(struct event_context *event_ctx,
575 struct fd_event *event,
576 uint16 flags,
577 void *private_data)
579 struct ctdbd_connection *conn = talloc_get_type_abort(
580 private_data, struct ctdbd_connection);
582 NTSTATUS status;
584 status = packet_fd_read(conn->pkt);
586 if (!NT_STATUS_IS_OK(status)) {
587 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
588 cluster_fatal("ctdbd died\n");
591 while (packet_handler(conn->pkt, ctdb_req_complete,
592 ctdb_handle_message, conn, &status)) {
593 if (!NT_STATUS_IS_OK(status)) {
594 DEBUG(10, ("could not handle incoming message: %s\n",
595 nt_errstr(status)));
601 * Prepare a ctdbd connection to receive messages
604 NTSTATUS ctdbd_register_msg_ctx(struct ctdbd_connection *conn,
605 struct messaging_context *msg_ctx)
607 SMB_ASSERT(conn->msg_ctx == NULL);
608 SMB_ASSERT(conn->fde == NULL);
610 if (!(conn->fde = event_add_fd(msg_ctx->event_ctx, conn,
611 packet_get_fd(conn->pkt),
612 EVENT_FD_READ,
613 ctdbd_socket_handler,
614 conn))) {
615 DEBUG(0, ("event_add_fd failed\n"));
616 return NT_STATUS_NO_MEMORY;
619 conn->msg_ctx = msg_ctx;
621 return NT_STATUS_OK;
625 * Send a messaging message across a ctdbd
628 NTSTATUS ctdbd_messaging_send(struct ctdbd_connection *conn,
629 uint32 dst_vnn, uint64 dst_srvid,
630 struct messaging_rec *msg)
632 struct ctdb_req_message r;
633 TALLOC_CTX *mem_ctx;
634 DATA_BLOB blob;
635 NTSTATUS status;
636 enum ndr_err_code ndr_err;
638 if (!(mem_ctx = talloc_init("ctdbd_messaging_send"))) {
639 DEBUG(0, ("talloc failed\n"));
640 return NT_STATUS_NO_MEMORY;
643 ndr_err = ndr_push_struct_blob(
644 &blob, mem_ctx, NULL, msg,
645 (ndr_push_flags_fn_t)ndr_push_messaging_rec);
647 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
648 DEBUG(0, ("ndr_push_struct_blob failed: %s\n",
649 ndr_errstr(ndr_err)));
650 status = ndr_map_error2ntstatus(ndr_err);
651 goto fail;
654 r.hdr.length = offsetof(struct ctdb_req_message, data) + blob.length;
655 r.hdr.ctdb_magic = CTDB_MAGIC;
656 r.hdr.ctdb_version = CTDB_VERSION;
657 r.hdr.generation = 1;
658 r.hdr.operation = CTDB_REQ_MESSAGE;
659 r.hdr.destnode = dst_vnn;
660 r.hdr.srcnode = conn->our_vnn;
661 r.hdr.reqid = 0;
662 r.srvid = dst_srvid;
663 r.datalen = blob.length;
665 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
666 ctdb_packet_dump(&r.hdr);
668 status = packet_send(
669 conn->pkt, 2,
670 data_blob_const(&r, offsetof(struct ctdb_req_message, data)),
671 blob);
673 if (!NT_STATUS_IS_OK(status)) {
674 DEBUG(0, ("packet_send failed: %s\n", nt_errstr(status)));
675 goto fail;
678 status = packet_flush(conn->pkt);
680 if (!NT_STATUS_IS_OK(status)) {
681 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
682 cluster_fatal("cluster dispatch daemon msg write error\n");
685 status = NT_STATUS_OK;
686 fail:
687 TALLOC_FREE(mem_ctx);
688 return status;
692 * send/recv a generic ctdb control message
694 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
695 uint32_t vnn, uint32 opcode,
696 uint64_t srvid, uint32_t flags,
697 TDB_DATA data,
698 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
699 int *cstatus)
701 struct ctdb_req_control req;
702 struct ctdb_reply_control *reply = NULL;
703 struct ctdbd_connection *new_conn = NULL;
704 NTSTATUS status;
706 /* the samba3 ctdb code can't handle NOREPLY yet */
707 flags &= ~CTDB_CTRL_FLAG_NOREPLY;
709 if (conn == NULL) {
710 status = ctdbd_init_connection(NULL, &new_conn);
712 if (!NT_STATUS_IS_OK(status)) {
713 DEBUG(10, ("Could not init temp connection: %s\n",
714 nt_errstr(status)));
715 goto fail;
718 conn = new_conn;
721 ZERO_STRUCT(req);
722 req.hdr.length = offsetof(struct ctdb_req_control, data) + data.dsize;
723 req.hdr.ctdb_magic = CTDB_MAGIC;
724 req.hdr.ctdb_version = CTDB_VERSION;
725 req.hdr.operation = CTDB_REQ_CONTROL;
726 req.hdr.reqid = ++conn->reqid;
727 req.hdr.destnode = vnn;
728 req.opcode = opcode;
729 req.srvid = srvid;
730 req.datalen = data.dsize;
732 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
733 ctdb_packet_dump(&req.hdr);
735 status = packet_send(
736 conn->pkt, 2,
737 data_blob_const(&req, offsetof(struct ctdb_req_control, data)),
738 data_blob_const(data.dptr, data.dsize));
740 if (!NT_STATUS_IS_OK(status)) {
741 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
742 goto fail;
745 status = packet_flush(conn->pkt);
747 if (!NT_STATUS_IS_OK(status)) {
748 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
749 cluster_fatal("cluster dispatch daemon control write error\n");
752 if (flags & CTDB_CTRL_FLAG_NOREPLY) {
753 TALLOC_FREE(new_conn);
754 return NT_STATUS_OK;
757 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
759 if (!NT_STATUS_IS_OK(status)) {
760 DEBUG(10, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
761 goto fail;
764 if (reply->hdr.operation != CTDB_REPLY_CONTROL) {
765 DEBUG(0, ("received invalid reply\n"));
766 goto fail;
769 if (outdata) {
770 if (!(outdata->dptr = (uint8 *)talloc_memdup(
771 mem_ctx, reply->data, reply->datalen))) {
772 TALLOC_FREE(reply);
773 return NT_STATUS_NO_MEMORY;
775 outdata->dsize = reply->datalen;
777 if (cstatus) {
778 (*cstatus) = reply->status;
781 status = NT_STATUS_OK;
783 fail:
784 TALLOC_FREE(new_conn);
785 TALLOC_FREE(reply);
786 return status;
790 * see if a remote process exists
792 bool ctdbd_process_exists(struct ctdbd_connection *conn, uint32 vnn, pid_t pid)
794 NTSTATUS status;
795 TDB_DATA data;
796 int32_t cstatus;
798 data.dptr = (uint8_t*)&pid;
799 data.dsize = sizeof(pid);
801 status = ctdbd_control(conn, vnn, CTDB_CONTROL_PROCESS_EXISTS, 0, 0,
802 data, NULL, NULL, &cstatus);
803 if (!NT_STATUS_IS_OK(status)) {
804 DEBUG(0, (__location__ " ctdb_control for process_exists "
805 "failed\n"));
806 return False;
809 return cstatus == 0;
813 * Get a db path
815 char *ctdbd_dbpath(struct ctdbd_connection *conn,
816 TALLOC_CTX *mem_ctx, uint32_t db_id)
818 NTSTATUS status;
819 TDB_DATA data;
820 int32_t cstatus;
822 data.dptr = (uint8_t*)&db_id;
823 data.dsize = sizeof(db_id);
825 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
826 CTDB_CONTROL_GETDBPATH, 0, 0, data,
827 mem_ctx, &data, &cstatus);
828 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
829 DEBUG(0,(__location__ " ctdb_control for getdbpath failed\n"));
830 return NULL;
833 return (char *)data.dptr;
837 * attach to a ctdb database
839 NTSTATUS ctdbd_db_attach(struct ctdbd_connection *conn,
840 const char *name, uint32_t *db_id, int tdb_flags)
842 NTSTATUS status;
843 TDB_DATA data;
844 int32_t cstatus;
845 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
847 data.dptr = (uint8_t*)name;
848 data.dsize = strlen(name)+1;
850 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
851 persistent
852 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
853 : CTDB_CONTROL_DB_ATTACH,
854 0, 0, data, NULL, &data, &cstatus);
855 if (!NT_STATUS_IS_OK(status)) {
856 DEBUG(0, (__location__ " ctdb_control for db_attach "
857 "failed: %s\n", nt_errstr(status)));
858 return status;
861 if (cstatus != 0 || data.dsize != sizeof(uint32_t)) {
862 DEBUG(0,(__location__ " ctdb_control for db_attach failed\n"));
863 return NT_STATUS_INTERNAL_ERROR;
866 *db_id = *(uint32_t *)data.dptr;
867 talloc_free(data.dptr);
869 if (!(tdb_flags & TDB_SEQNUM)) {
870 return NT_STATUS_OK;
873 data.dptr = (uint8_t *)db_id;
874 data.dsize = sizeof(*db_id);
876 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
877 CTDB_CONTROL_ENABLE_SEQNUM, 0, 0, data,
878 NULL, NULL, &cstatus);
879 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
880 DEBUG(0,(__location__ " ctdb_control for enable seqnum "
881 "failed\n"));
882 return NT_STATUS_IS_OK(status) ? NT_STATUS_INTERNAL_ERROR :
883 status;
886 return NT_STATUS_OK;
890 * force the migration of a record to this node
892 NTSTATUS ctdbd_migrate(struct ctdbd_connection *conn, uint32 db_id,
893 TDB_DATA key)
895 struct ctdb_req_call req;
896 struct ctdb_reply_call *reply;
897 NTSTATUS status;
899 ZERO_STRUCT(req);
901 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
902 req.hdr.ctdb_magic = CTDB_MAGIC;
903 req.hdr.ctdb_version = CTDB_VERSION;
904 req.hdr.operation = CTDB_REQ_CALL;
905 req.hdr.reqid = ++conn->reqid;
906 req.flags = CTDB_IMMEDIATE_MIGRATION;
907 req.callid = CTDB_NULL_FUNC;
908 req.db_id = db_id;
909 req.keylen = key.dsize;
911 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
912 ctdb_packet_dump(&req.hdr);
914 status = packet_send(
915 conn->pkt, 2,
916 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
917 data_blob_const(key.dptr, key.dsize));
919 if (!NT_STATUS_IS_OK(status)) {
920 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
921 return status;
924 status = packet_flush(conn->pkt);
926 if (!NT_STATUS_IS_OK(status)) {
927 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
928 cluster_fatal("cluster dispatch daemon control write error\n");
931 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
933 if (!NT_STATUS_IS_OK(status)) {
934 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
935 goto fail;
938 if (reply->hdr.operation != CTDB_REPLY_CALL) {
939 DEBUG(0, ("received invalid reply\n"));
940 status = NT_STATUS_INTERNAL_ERROR;
941 goto fail;
944 status = NT_STATUS_OK;
945 fail:
947 TALLOC_FREE(reply);
948 return status;
952 * remotely fetch a record without locking it or forcing a migration
954 NTSTATUS ctdbd_fetch(struct ctdbd_connection *conn, uint32 db_id,
955 TDB_DATA key, TALLOC_CTX *mem_ctx, TDB_DATA *data)
957 struct ctdb_req_call req;
958 struct ctdb_reply_call *reply;
959 NTSTATUS status;
961 ZERO_STRUCT(req);
963 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
964 req.hdr.ctdb_magic = CTDB_MAGIC;
965 req.hdr.ctdb_version = CTDB_VERSION;
966 req.hdr.operation = CTDB_REQ_CALL;
967 req.hdr.reqid = ++conn->reqid;
968 req.flags = 0;
969 req.callid = CTDB_FETCH_FUNC;
970 req.db_id = db_id;
971 req.keylen = key.dsize;
973 status = packet_send(
974 conn->pkt, 2,
975 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
976 data_blob_const(key.dptr, key.dsize));
978 if (!NT_STATUS_IS_OK(status)) {
979 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
980 return status;
983 status = packet_flush(conn->pkt);
985 if (!NT_STATUS_IS_OK(status)) {
986 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
987 cluster_fatal("cluster dispatch daemon control write error\n");
990 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
992 if (!NT_STATUS_IS_OK(status)) {
993 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
994 goto fail;
997 if (reply->hdr.operation != CTDB_REPLY_CALL) {
998 DEBUG(0, ("received invalid reply\n"));
999 status = NT_STATUS_INTERNAL_ERROR;
1000 goto fail;
1003 data->dsize = reply->datalen;
1004 if (data->dsize == 0) {
1005 data->dptr = NULL;
1006 goto done;
1009 data->dptr = (uint8 *)talloc_memdup(mem_ctx, &reply->data[0],
1010 reply->datalen);
1011 if (data->dptr == NULL) {
1012 DEBUG(0, ("talloc failed\n"));
1013 status = NT_STATUS_NO_MEMORY;
1014 goto fail;
1017 done:
1018 status = NT_STATUS_OK;
1019 fail:
1020 TALLOC_FREE(reply);
1021 return status;
1024 struct ctdbd_traverse_state {
1025 void (*fn)(TDB_DATA key, TDB_DATA data, void *private_data);
1026 void *private_data;
1030 * Handle a traverse record coming in on the ctdbd connection
1033 static NTSTATUS ctdb_traverse_handler(uint8_t *buf, size_t length,
1034 void *private_data)
1036 struct ctdbd_traverse_state *state =
1037 (struct ctdbd_traverse_state *)private_data;
1039 struct ctdb_req_message *m;
1040 struct ctdb_rec_data *d;
1041 TDB_DATA key, data;
1043 m = (struct ctdb_req_message *)buf;
1045 if (length < sizeof(*m) || m->hdr.length != length) {
1046 DEBUG(0, ("Got invalid message of length %d\n", (int)length));
1047 TALLOC_FREE(buf);
1048 return NT_STATUS_UNEXPECTED_IO_ERROR;
1051 d = (struct ctdb_rec_data *)&m->data[0];
1052 if (m->datalen < sizeof(uint32_t) || m->datalen != d->length) {
1053 DEBUG(0, ("Got invalid traverse data of length %d\n",
1054 (int)m->datalen));
1055 TALLOC_FREE(buf);
1056 return NT_STATUS_UNEXPECTED_IO_ERROR;
1059 key.dsize = d->keylen;
1060 key.dptr = &d->data[0];
1061 data.dsize = d->datalen;
1062 data.dptr = &d->data[d->keylen];
1064 if (key.dsize == 0 && data.dsize == 0) {
1065 /* end of traverse */
1066 return NT_STATUS_END_OF_FILE;
1069 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1070 DEBUG(0, ("Got invalid ltdb header length %d\n",
1071 (int)data.dsize));
1072 TALLOC_FREE(buf);
1073 return NT_STATUS_UNEXPECTED_IO_ERROR;
1075 data.dsize -= sizeof(struct ctdb_ltdb_header);
1076 data.dptr += sizeof(struct ctdb_ltdb_header);
1078 if (state->fn) {
1079 state->fn(key, data, state->private_data);
1082 TALLOC_FREE(buf);
1083 return NT_STATUS_OK;
1087 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1088 connection to ctdbd to avoid the hairy recursive and async problems with
1089 everything in-line.
1092 NTSTATUS ctdbd_traverse(uint32 db_id,
1093 void (*fn)(TDB_DATA key, TDB_DATA data,
1094 void *private_data),
1095 void *private_data)
1097 struct ctdbd_connection *conn;
1098 NTSTATUS status;
1100 TDB_DATA data;
1101 struct ctdb_traverse_start t;
1102 int cstatus;
1103 struct ctdbd_traverse_state state;
1105 status = ctdbd_init_connection(NULL, &conn);
1106 if (!NT_STATUS_IS_OK(status)) {
1107 DEBUG(0, ("ctdbd_init_connection failed: %s\n",
1108 nt_errstr(status)));
1109 return status;
1112 t.db_id = db_id;
1113 t.srvid = conn->rand_srvid;
1114 t.reqid = ++conn->reqid;
1116 data.dptr = (uint8_t *)&t;
1117 data.dsize = sizeof(t);
1119 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1120 CTDB_CONTROL_TRAVERSE_START, conn->rand_srvid, 0,
1121 data, NULL, NULL, &cstatus);
1123 if (!NT_STATUS_IS_OK(status) || (cstatus != 0)) {
1125 DEBUG(0,("ctdbd_control failed: %s, %d\n", nt_errstr(status),
1126 cstatus));
1128 if (NT_STATUS_IS_OK(status)) {
1130 * We need a mapping here
1132 status = NT_STATUS_UNSUCCESSFUL;
1134 goto done;
1137 state.fn = fn;
1138 state.private_data = private_data;
1140 while (True) {
1142 status = NT_STATUS_OK;
1144 if (packet_handler(conn->pkt, ctdb_req_complete,
1145 ctdb_traverse_handler, &state, &status)) {
1147 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1148 status = NT_STATUS_OK;
1149 break;
1153 * There might be more in the queue
1155 continue;
1158 if (!NT_STATUS_IS_OK(status)) {
1159 break;
1162 status = ctdb_packet_fd_read_sync(conn->pkt);
1164 if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
1166 * There might be more in the queue
1168 continue;
1171 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1172 status = NT_STATUS_OK;
1173 break;
1176 if (!NT_STATUS_IS_OK(status)) {
1177 DEBUG(0, ("packet_fd_read_sync failed: %s\n", nt_errstr(status)));
1178 cluster_fatal("ctdbd died\n");
1182 done:
1183 TALLOC_FREE(conn);
1184 return status;
1188 This is used to canonicalize a ctdb_sock_addr structure.
1190 static void smbd_ctdb_canonicalize_ip(const struct sockaddr_storage *in,
1191 struct sockaddr_storage *out)
1193 memcpy(out, in, sizeof (*out));
1195 #ifdef HAVE_IPV6
1196 if (in->ss_family == AF_INET6) {
1197 const char prefix[12] = { 0,0,0,0,0,0,0,0,0,0,0xff,0xff };
1198 const struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)in;
1199 struct sockaddr_in *out4 = (struct sockaddr_in *)out;
1200 if (memcmp(&in6->sin6_addr, prefix, 12) == 0) {
1201 memset(out, 0, sizeof(*out));
1202 #ifdef HAVE_SOCK_SIN_LEN
1203 out4->sin_len = sizeof(*out);
1204 #endif
1205 out4->sin_family = AF_INET;
1206 out4->sin_port = in6->sin6_port;
1207 memcpy(&out4->sin_addr, &in6->sin6_addr.s6_addr32[3], 4);
1210 #endif
1214 * Register us as a server for a particular tcp connection
1217 NTSTATUS ctdbd_register_ips(struct ctdbd_connection *conn,
1218 const struct sockaddr_storage *_server,
1219 const struct sockaddr_storage *_client,
1220 void (*release_ip_handler)(const char *ip_addr,
1221 void *private_data),
1222 void *private_data)
1225 * we still use ctdb_control_tcp for ipv4
1226 * because we want to work against older ctdb
1227 * versions at runtime
1229 struct ctdb_control_tcp p4;
1230 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1231 struct ctdb_control_tcp_addr p;
1232 #endif
1233 TDB_DATA data;
1234 NTSTATUS status;
1235 struct sockaddr_storage client;
1236 struct sockaddr_storage server;
1239 * Only one connection so far
1241 SMB_ASSERT(conn->release_ip_handler == NULL);
1243 smbd_ctdb_canonicalize_ip(_client, &client);
1244 smbd_ctdb_canonicalize_ip(_server, &server);
1246 switch (client.ss_family) {
1247 case AF_INET:
1248 p4.dest = *(struct sockaddr_in *)&server;
1249 p4.src = *(struct sockaddr_in *)&client;
1250 data.dptr = (uint8_t *)&p4;
1251 data.dsize = sizeof(p4);
1252 break;
1253 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1254 case AF_INET6:
1255 p.dest.ip6 = *(struct sockaddr_in6 *)&server;
1256 p.src.ip6 = *(struct sockaddr_in6 *)&client;
1257 data.dptr = (uint8_t *)&p;
1258 data.dsize = sizeof(p);
1259 break;
1260 #endif
1261 default:
1262 return NT_STATUS_INTERNAL_ERROR;
1265 conn->release_ip_handler = release_ip_handler;
1268 * We want to be told about IP releases
1271 status = register_with_ctdbd(conn, CTDB_SRVID_RELEASE_IP);
1272 if (!NT_STATUS_IS_OK(status)) {
1273 return status;
1277 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1278 * can send an extra ack to trigger a reset for our client, so it
1279 * immediately reconnects
1281 return ctdbd_control(conn, CTDB_CURRENT_NODE,
1282 CTDB_CONTROL_TCP_CLIENT, 0,
1283 CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL, NULL);
1287 * We want to handle reconfigure events
1289 NTSTATUS ctdbd_register_reconfigure(struct ctdbd_connection *conn)
1291 return register_with_ctdbd(conn, CTDB_SRVID_RECONFIGURE);
1295 call a control on the local node
1297 NTSTATUS ctdbd_control_local(struct ctdbd_connection *conn, uint32 opcode,
1298 uint64_t srvid, uint32_t flags, TDB_DATA data,
1299 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
1300 int *cstatus)
1302 return ctdbd_control(conn, CTDB_CURRENT_NODE, opcode, srvid, flags, data, mem_ctx, outdata, cstatus);
1305 #else
1307 NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
1308 struct ctdbd_connection **pconn)
1310 return NT_STATUS_NOT_IMPLEMENTED;
1313 #endif