s3: Use cli_writeall instead of cli_write
[Samba.git] / source3 / lib / ctdbd_conn.c
blob21bd96d3436d197f3143a3bc5d49249388ea3147
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 "ctdbd_conn.h"
26 #include "packet.h"
27 #include "messages.h"
29 /* paths to these include files come from --with-ctdb= in configure */
30 #include "ctdb.h"
31 #include "ctdb_private.h"
33 struct ctdbd_connection {
34 struct messaging_context *msg_ctx;
35 uint32 reqid;
36 uint32 our_vnn;
37 uint64 rand_srvid;
38 struct packet_context *pkt;
39 struct fd_event *fde;
41 void (*release_ip_handler)(const char *ip_addr, void *private_data);
42 void *release_ip_priv;
45 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
46 uint32_t vnn, uint32 opcode,
47 uint64_t srvid, uint32_t flags, TDB_DATA data,
48 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
49 int *cstatus);
52 * exit on fatal communications errors with the ctdbd daemon
54 static void cluster_fatal(const char *why)
56 DEBUG(0,("cluster fatal event: %s - exiting immediately\n", why));
57 /* we don't use smb_panic() as we don't want to delay to write
58 a core file. We need to release this process id immediately
59 so that someone else can take over without getting sharing
60 violations */
61 _exit(1);
67 static void ctdb_packet_dump(struct ctdb_req_header *hdr)
69 if (DEBUGLEVEL < 10) {
70 return;
72 DEBUGADD(10, ("len=%d, magic=%x, vers=%d, gen=%d, op=%d, reqid=%d\n",
73 (int)hdr->length, (int)hdr->ctdb_magic,
74 (int)hdr->ctdb_version, (int)hdr->generation,
75 (int)hdr->operation, (int)hdr->reqid));
79 * Register a srvid with ctdbd
81 static NTSTATUS register_with_ctdbd(struct ctdbd_connection *conn,
82 uint64_t srvid)
85 int cstatus;
86 return ctdbd_control(conn, CTDB_CURRENT_NODE,
87 CTDB_CONTROL_REGISTER_SRVID, srvid, 0,
88 tdb_null, NULL, NULL, &cstatus);
92 * get our vnn from the cluster
94 static NTSTATUS get_cluster_vnn(struct ctdbd_connection *conn, uint32 *vnn)
96 int32_t cstatus=-1;
97 NTSTATUS status;
98 status = ctdbd_control(conn,
99 CTDB_CURRENT_NODE, CTDB_CONTROL_GET_PNN, 0, 0,
100 tdb_null, NULL, NULL, &cstatus);
101 if (!NT_STATUS_IS_OK(status)) {
102 cluster_fatal("ctdbd_control failed\n");
104 *vnn = (uint32_t)cstatus;
105 return status;
109 * Are we active (i.e. not banned or stopped?)
111 static bool ctdbd_working(struct ctdbd_connection *conn, uint32_t vnn)
113 int32_t cstatus=-1;
114 NTSTATUS status;
115 TDB_DATA outdata;
116 struct ctdb_node_map *m;
117 uint32_t failure_flags;
118 bool ret = false;
119 int i;
121 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
122 CTDB_CONTROL_GET_NODEMAP, 0, 0,
123 tdb_null, talloc_tos(), &outdata, &cstatus);
124 if (!NT_STATUS_IS_OK(status)) {
125 cluster_fatal("ctdbd_control failed\n");
127 if ((cstatus != 0) || (outdata.dptr == NULL)) {
128 DEBUG(2, ("Received invalid ctdb data\n"));
129 return false;
132 m = (struct ctdb_node_map *)outdata.dptr;
134 for (i=0; i<m->num; i++) {
135 if (vnn == m->nodes[i].pnn) {
136 break;
140 if (i == m->num) {
141 DEBUG(2, ("Did not find ourselves (node %d) in nodemap\n",
142 (int)vnn));
143 goto fail;
146 failure_flags = NODE_FLAGS_BANNED | NODE_FLAGS_DISCONNECTED
147 | NODE_FLAGS_PERMANENTLY_DISABLED | NODE_FLAGS_STOPPED;
149 if ((m->nodes[i].flags & failure_flags) != 0) {
150 DEBUG(2, ("Node has status %x, not active\n",
151 (int)m->nodes[i].flags));
152 goto fail;
155 ret = true;
156 fail:
157 TALLOC_FREE(outdata.dptr);
158 return ret;
161 uint32 ctdbd_vnn(const struct ctdbd_connection *conn)
163 return conn->our_vnn;
167 * Get us a ctdb connection
170 static NTSTATUS ctdbd_connect(TALLOC_CTX *mem_ctx,
171 struct packet_context **presult)
173 struct packet_context *result;
174 const char *sockname = lp_ctdbd_socket();
175 struct sockaddr_un addr;
176 int fd;
178 if (!sockname || !*sockname) {
179 sockname = CTDB_PATH;
182 fd = socket(AF_UNIX, SOCK_STREAM, 0);
183 if (fd == -1) {
184 DEBUG(3, ("Could not create socket: %s\n", strerror(errno)));
185 return map_nt_error_from_unix(errno);
188 ZERO_STRUCT(addr);
189 addr.sun_family = AF_UNIX;
190 strncpy(addr.sun_path, sockname, sizeof(addr.sun_path));
192 if (sys_connect(fd, (struct sockaddr *)(void *)&addr) == -1) {
193 DEBUG(1, ("connect(%s) failed: %s\n", sockname,
194 strerror(errno)));
195 close(fd);
196 return map_nt_error_from_unix(errno);
199 if (!(result = packet_init(mem_ctx, fd))) {
200 close(fd);
201 return NT_STATUS_NO_MEMORY;
204 *presult = result;
205 return NT_STATUS_OK;
209 * Do we have a complete ctdb packet in the queue?
212 static bool ctdb_req_complete(const uint8_t *buf, size_t available,
213 size_t *length,
214 void *private_data)
216 uint32 msglen;
218 if (available < sizeof(msglen)) {
219 return False;
222 msglen = *((uint32 *)buf);
224 DEBUG(10, ("msglen = %d\n", msglen));
226 if (msglen < sizeof(struct ctdb_req_header)) {
227 DEBUG(0, ("Got invalid msglen: %d, expected at least %d for "
228 "the req_header\n", (int)msglen,
229 (int)sizeof(struct ctdb_req_header)));
230 cluster_fatal("ctdbd protocol error\n");
233 if (available < msglen) {
234 return false;
237 *length = msglen;
238 return true;
242 * State necessary to defer an incoming message while we are waiting for a
243 * ctdb reply.
246 struct deferred_msg_state {
247 struct messaging_context *msg_ctx;
248 struct messaging_rec *rec;
252 * Timed event handler for the deferred message
255 static void deferred_message_dispatch(struct event_context *event_ctx,
256 struct timed_event *te,
257 struct timeval now,
258 void *private_data)
260 struct deferred_msg_state *state = talloc_get_type_abort(
261 private_data, struct deferred_msg_state);
263 messaging_dispatch_rec(state->msg_ctx, state->rec);
264 TALLOC_FREE(state);
265 TALLOC_FREE(te);
268 struct req_pull_state {
269 TALLOC_CTX *mem_ctx;
270 DATA_BLOB req;
274 * Pull a ctdb request out of the incoming packet queue
277 static NTSTATUS ctdb_req_pull(uint8_t *buf, size_t length,
278 void *private_data)
280 struct req_pull_state *state = (struct req_pull_state *)private_data;
282 state->req.data = talloc_move(state->mem_ctx, &buf);
283 state->req.length = length;
284 return NT_STATUS_OK;
288 * Fetch a messaging_rec from an incoming ctdb style message
291 static struct messaging_rec *ctdb_pull_messaging_rec(TALLOC_CTX *mem_ctx,
292 size_t overall_length,
293 struct ctdb_req_message *msg)
295 struct messaging_rec *result;
296 DATA_BLOB blob;
297 enum ndr_err_code ndr_err;
299 if ((overall_length < offsetof(struct ctdb_req_message, data))
300 || (overall_length
301 < offsetof(struct ctdb_req_message, data) + msg->datalen)) {
303 cluster_fatal("got invalid msg length");
306 if (!(result = TALLOC_P(mem_ctx, struct messaging_rec))) {
307 DEBUG(0, ("talloc failed\n"));
308 return NULL;
311 blob = data_blob_const(msg->data, msg->datalen);
313 ndr_err = ndr_pull_struct_blob(
314 &blob, result, result,
315 (ndr_pull_flags_fn_t)ndr_pull_messaging_rec);
317 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
318 DEBUG(0, ("ndr_pull_struct_blob failed: %s\n",
319 ndr_errstr(ndr_err)));
320 TALLOC_FREE(result);
321 return NULL;
324 if (DEBUGLEVEL >= 10) {
325 DEBUG(10, ("ctdb_pull_messaging_rec:\n"));
326 NDR_PRINT_DEBUG(messaging_rec, result);
329 return result;
332 static NTSTATUS ctdb_packet_fd_read_sync(struct packet_context *ctx)
334 int timeout = lp_ctdb_timeout();
336 if (timeout == 0) {
337 timeout = -1;
339 return packet_fd_read_sync(ctx, timeout);
343 * Read a full ctdbd request. If we have a messaging context, defer incoming
344 * messages that might come in between.
347 static NTSTATUS ctdb_read_req(struct ctdbd_connection *conn, uint32 reqid,
348 TALLOC_CTX *mem_ctx, void *result)
350 struct ctdb_req_header *hdr;
351 struct req_pull_state state;
352 NTSTATUS status;
354 again:
356 status = ctdb_packet_fd_read_sync(conn->pkt);
358 if (NT_STATUS_EQUAL(status, NT_STATUS_NETWORK_BUSY)) {
359 /* EAGAIN */
360 goto again;
361 } else if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
362 /* EAGAIN */
363 goto again;
366 if (!NT_STATUS_IS_OK(status)) {
367 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
368 cluster_fatal("ctdbd died\n");
371 next_pkt:
373 ZERO_STRUCT(state);
374 state.mem_ctx = mem_ctx;
376 if (!packet_handler(conn->pkt, ctdb_req_complete, ctdb_req_pull,
377 &state, &status)) {
379 * Not enough data
381 DEBUG(10, ("not enough data from ctdb socket, retrying\n"));
382 goto again;
385 if (!NT_STATUS_IS_OK(status)) {
386 DEBUG(0, ("Could not read packet: %s\n", nt_errstr(status)));
387 cluster_fatal("ctdbd died\n");
390 hdr = (struct ctdb_req_header *)state.req.data;
392 DEBUG(10, ("Received ctdb packet\n"));
393 ctdb_packet_dump(hdr);
395 if (hdr->operation == CTDB_REQ_MESSAGE) {
396 struct timed_event *evt;
397 struct deferred_msg_state *msg_state;
398 struct ctdb_req_message *msg = (struct ctdb_req_message *)hdr;
400 if (conn->msg_ctx == NULL) {
401 DEBUG(1, ("Got a message without having a msg ctx, "
402 "dropping msg %llu\n",
403 (long long unsigned)msg->srvid));
404 goto next_pkt;
407 if ((conn->release_ip_handler != NULL)
408 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
409 /* must be dispatched immediately */
410 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
411 conn->release_ip_handler((const char *)msg->data,
412 conn->release_ip_priv);
413 TALLOC_FREE(hdr);
414 goto next_pkt;
417 if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
418 || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)) {
420 DEBUG(1, ("ctdb_read_req: Got %s message\n",
421 (msg->srvid == CTDB_SRVID_RECONFIGURE)
422 ? "cluster reconfigure" : "SAMBA_NOTIFY"));
424 messaging_send(conn->msg_ctx,
425 messaging_server_id(conn->msg_ctx),
426 MSG_SMB_BRL_VALIDATE, &data_blob_null);
427 messaging_send(conn->msg_ctx,
428 messaging_server_id(conn->msg_ctx),
429 MSG_DBWRAP_G_LOCK_RETRY,
430 &data_blob_null);
431 TALLOC_FREE(hdr);
432 goto next_pkt;
435 msg_state = TALLOC_P(NULL, struct deferred_msg_state);
436 if (msg_state == NULL) {
437 DEBUG(0, ("talloc failed\n"));
438 TALLOC_FREE(hdr);
439 goto next_pkt;
442 if (!(msg_state->rec = ctdb_pull_messaging_rec(
443 msg_state, state.req.length, msg))) {
444 DEBUG(0, ("ctdbd_pull_messaging_rec failed\n"));
445 TALLOC_FREE(msg_state);
446 TALLOC_FREE(hdr);
447 goto next_pkt;
450 TALLOC_FREE(hdr);
452 msg_state->msg_ctx = conn->msg_ctx;
455 * We're waiting for a call reply, but an async message has
456 * crossed. Defer dispatching to the toplevel event loop.
458 evt = event_add_timed(conn->msg_ctx->event_ctx,
459 conn->msg_ctx->event_ctx,
460 timeval_zero(),
461 deferred_message_dispatch,
462 msg_state);
463 if (evt == NULL) {
464 DEBUG(0, ("event_add_timed failed\n"));
465 TALLOC_FREE(msg_state);
466 TALLOC_FREE(hdr);
467 goto next_pkt;
470 goto next_pkt;
473 if (hdr->reqid != reqid) {
474 /* we got the wrong reply */
475 DEBUG(0,("Discarding mismatched ctdb reqid %u should have "
476 "been %u\n", hdr->reqid, reqid));
477 TALLOC_FREE(hdr);
478 goto again;
481 *((void **)result) = talloc_move(mem_ctx, &hdr);
483 return NT_STATUS_OK;
487 * Get us a ctdbd connection
490 static NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
491 struct ctdbd_connection **pconn)
493 struct ctdbd_connection *conn;
494 NTSTATUS status;
496 if (!(conn = TALLOC_ZERO_P(mem_ctx, struct ctdbd_connection))) {
497 DEBUG(0, ("talloc failed\n"));
498 return NT_STATUS_NO_MEMORY;
501 status = ctdbd_connect(conn, &conn->pkt);
503 if (!NT_STATUS_IS_OK(status)) {
504 DEBUG(10, ("ctdbd_connect failed: %s\n", nt_errstr(status)));
505 goto fail;
508 status = get_cluster_vnn(conn, &conn->our_vnn);
510 if (!NT_STATUS_IS_OK(status)) {
511 DEBUG(10, ("get_cluster_vnn failed: %s\n", nt_errstr(status)));
512 goto fail;
515 if (!ctdbd_working(conn, conn->our_vnn)) {
516 DEBUG(2, ("Node is not working, can not connect\n"));
517 status = NT_STATUS_INTERNAL_DB_ERROR;
518 goto fail;
521 generate_random_buffer((unsigned char *)&conn->rand_srvid,
522 sizeof(conn->rand_srvid));
524 status = register_with_ctdbd(conn, conn->rand_srvid);
526 if (!NT_STATUS_IS_OK(status)) {
527 DEBUG(5, ("Could not register random srvid: %s\n",
528 nt_errstr(status)));
529 goto fail;
532 *pconn = conn;
533 return NT_STATUS_OK;
535 fail:
536 TALLOC_FREE(conn);
537 return status;
541 * Get us a ctdbd connection and register us as a process
544 NTSTATUS ctdbd_messaging_connection(TALLOC_CTX *mem_ctx,
545 struct ctdbd_connection **pconn)
547 struct ctdbd_connection *conn;
548 NTSTATUS status;
550 status = ctdbd_init_connection(mem_ctx, &conn);
552 if (!NT_STATUS_IS_OK(status)) {
553 return status;
556 status = register_with_ctdbd(conn, (uint64_t)sys_getpid());
557 if (!NT_STATUS_IS_OK(status)) {
558 goto fail;
561 status = register_with_ctdbd(conn, MSG_SRVID_SAMBA);
562 if (!NT_STATUS_IS_OK(status)) {
563 goto fail;
566 status = register_with_ctdbd(conn, CTDB_SRVID_SAMBA_NOTIFY);
567 if (!NT_STATUS_IS_OK(status)) {
568 goto fail;
571 *pconn = conn;
572 return NT_STATUS_OK;
574 fail:
575 TALLOC_FREE(conn);
576 return status;
579 struct messaging_context *ctdb_conn_msg_ctx(struct ctdbd_connection *conn)
581 return conn->msg_ctx;
584 int ctdbd_conn_get_fd(struct ctdbd_connection *conn)
586 return packet_get_fd(conn->pkt);
590 * Packet handler to receive and handle a ctdb message
592 static NTSTATUS ctdb_handle_message(uint8_t *buf, size_t length,
593 void *private_data)
595 struct ctdbd_connection *conn = talloc_get_type_abort(
596 private_data, struct ctdbd_connection);
597 struct ctdb_req_message *msg;
598 struct messaging_rec *msg_rec;
600 msg = (struct ctdb_req_message *)buf;
602 if (msg->hdr.operation != CTDB_REQ_MESSAGE) {
603 DEBUG(0, ("Received async msg of type %u, discarding\n",
604 msg->hdr.operation));
605 TALLOC_FREE(buf);
606 return NT_STATUS_INVALID_PARAMETER;
609 if ((conn->release_ip_handler != NULL)
610 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
611 /* must be dispatched immediately */
612 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
613 conn->release_ip_handler((const char *)msg->data,
614 conn->release_ip_priv);
615 TALLOC_FREE(buf);
616 return NT_STATUS_OK;
619 SMB_ASSERT(conn->msg_ctx != NULL);
621 if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
622 || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)){
623 DEBUG(0,("Got cluster reconfigure message\n"));
625 * when the cluster is reconfigured or someone of the
626 * family has passed away (SAMBA_NOTIFY), we need to
627 * clean the brl database
629 messaging_send(conn->msg_ctx,
630 messaging_server_id(conn->msg_ctx),
631 MSG_SMB_BRL_VALIDATE, &data_blob_null);
633 messaging_send(conn->msg_ctx,
634 messaging_server_id(conn->msg_ctx),
635 MSG_DBWRAP_G_LOCK_RETRY,
636 &data_blob_null);
638 TALLOC_FREE(buf);
639 return NT_STATUS_OK;
642 /* only messages to our pid or the broadcast are valid here */
643 if (msg->srvid != sys_getpid() && msg->srvid != MSG_SRVID_SAMBA) {
644 DEBUG(0,("Got unexpected message with srvid=%llu\n",
645 (unsigned long long)msg->srvid));
646 TALLOC_FREE(buf);
647 return NT_STATUS_OK;
650 if (!(msg_rec = ctdb_pull_messaging_rec(NULL, length, msg))) {
651 DEBUG(10, ("ctdb_pull_messaging_rec failed\n"));
652 TALLOC_FREE(buf);
653 return NT_STATUS_NO_MEMORY;
656 messaging_dispatch_rec(conn->msg_ctx, msg_rec);
658 TALLOC_FREE(msg_rec);
659 TALLOC_FREE(buf);
660 return NT_STATUS_OK;
664 * The ctdbd socket is readable asynchronuously
667 static void ctdbd_socket_handler(struct event_context *event_ctx,
668 struct fd_event *event,
669 uint16 flags,
670 void *private_data)
672 struct ctdbd_connection *conn = talloc_get_type_abort(
673 private_data, struct ctdbd_connection);
675 NTSTATUS status;
677 status = packet_fd_read(conn->pkt);
679 if (!NT_STATUS_IS_OK(status)) {
680 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
681 cluster_fatal("ctdbd died\n");
684 while (packet_handler(conn->pkt, ctdb_req_complete,
685 ctdb_handle_message, conn, &status)) {
686 if (!NT_STATUS_IS_OK(status)) {
687 DEBUG(10, ("could not handle incoming message: %s\n",
688 nt_errstr(status)));
694 * Prepare a ctdbd connection to receive messages
697 NTSTATUS ctdbd_register_msg_ctx(struct ctdbd_connection *conn,
698 struct messaging_context *msg_ctx)
700 SMB_ASSERT(conn->msg_ctx == NULL);
701 SMB_ASSERT(conn->fde == NULL);
703 if (!(conn->fde = event_add_fd(msg_ctx->event_ctx, conn,
704 packet_get_fd(conn->pkt),
705 EVENT_FD_READ,
706 ctdbd_socket_handler,
707 conn))) {
708 DEBUG(0, ("event_add_fd failed\n"));
709 return NT_STATUS_NO_MEMORY;
712 conn->msg_ctx = msg_ctx;
714 return NT_STATUS_OK;
718 * Send a messaging message across a ctdbd
721 NTSTATUS ctdbd_messaging_send(struct ctdbd_connection *conn,
722 uint32 dst_vnn, uint64 dst_srvid,
723 struct messaging_rec *msg)
725 struct ctdb_req_message r;
726 TALLOC_CTX *mem_ctx;
727 DATA_BLOB blob;
728 NTSTATUS status;
729 enum ndr_err_code ndr_err;
731 if (!(mem_ctx = talloc_init("ctdbd_messaging_send"))) {
732 DEBUG(0, ("talloc failed\n"));
733 return NT_STATUS_NO_MEMORY;
736 ndr_err = ndr_push_struct_blob(
737 &blob, mem_ctx, msg,
738 (ndr_push_flags_fn_t)ndr_push_messaging_rec);
740 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
741 DEBUG(0, ("ndr_push_struct_blob failed: %s\n",
742 ndr_errstr(ndr_err)));
743 status = ndr_map_error2ntstatus(ndr_err);
744 goto fail;
747 r.hdr.length = offsetof(struct ctdb_req_message, data) + blob.length;
748 r.hdr.ctdb_magic = CTDB_MAGIC;
749 r.hdr.ctdb_version = CTDB_VERSION;
750 r.hdr.generation = 1;
751 r.hdr.operation = CTDB_REQ_MESSAGE;
752 r.hdr.destnode = dst_vnn;
753 r.hdr.srcnode = conn->our_vnn;
754 r.hdr.reqid = 0;
755 r.srvid = dst_srvid;
756 r.datalen = blob.length;
758 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
759 ctdb_packet_dump(&r.hdr);
761 status = packet_send(
762 conn->pkt, 2,
763 data_blob_const(&r, offsetof(struct ctdb_req_message, data)),
764 blob);
766 if (!NT_STATUS_IS_OK(status)) {
767 DEBUG(0, ("packet_send failed: %s\n", nt_errstr(status)));
768 goto fail;
771 status = packet_flush(conn->pkt);
773 if (!NT_STATUS_IS_OK(status)) {
774 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
775 cluster_fatal("cluster dispatch daemon msg write error\n");
778 status = NT_STATUS_OK;
779 fail:
780 TALLOC_FREE(mem_ctx);
781 return status;
785 * send/recv a generic ctdb control message
787 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
788 uint32_t vnn, uint32 opcode,
789 uint64_t srvid, uint32_t flags,
790 TDB_DATA data,
791 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
792 int *cstatus)
794 struct ctdb_req_control req;
795 struct ctdb_reply_control *reply = NULL;
796 struct ctdbd_connection *new_conn = NULL;
797 NTSTATUS status;
799 /* the samba3 ctdb code can't handle NOREPLY yet */
800 flags &= ~CTDB_CTRL_FLAG_NOREPLY;
802 if (conn == NULL) {
803 status = ctdbd_init_connection(NULL, &new_conn);
805 if (!NT_STATUS_IS_OK(status)) {
806 DEBUG(10, ("Could not init temp connection: %s\n",
807 nt_errstr(status)));
808 goto fail;
811 conn = new_conn;
814 ZERO_STRUCT(req);
815 req.hdr.length = offsetof(struct ctdb_req_control, data) + data.dsize;
816 req.hdr.ctdb_magic = CTDB_MAGIC;
817 req.hdr.ctdb_version = CTDB_VERSION;
818 req.hdr.operation = CTDB_REQ_CONTROL;
819 req.hdr.reqid = ++conn->reqid;
820 req.hdr.destnode = vnn;
821 req.opcode = opcode;
822 req.srvid = srvid;
823 req.datalen = data.dsize;
824 req.flags = flags;
826 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
827 ctdb_packet_dump(&req.hdr);
829 status = packet_send(
830 conn->pkt, 2,
831 data_blob_const(&req, offsetof(struct ctdb_req_control, data)),
832 data_blob_const(data.dptr, data.dsize));
834 if (!NT_STATUS_IS_OK(status)) {
835 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
836 goto fail;
839 status = packet_flush(conn->pkt);
841 if (!NT_STATUS_IS_OK(status)) {
842 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
843 cluster_fatal("cluster dispatch daemon control write error\n");
846 if (flags & CTDB_CTRL_FLAG_NOREPLY) {
847 TALLOC_FREE(new_conn);
848 if (cstatus) {
849 *cstatus = 0;
851 return NT_STATUS_OK;
854 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
856 if (!NT_STATUS_IS_OK(status)) {
857 DEBUG(10, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
858 goto fail;
861 if (reply->hdr.operation != CTDB_REPLY_CONTROL) {
862 DEBUG(0, ("received invalid reply\n"));
863 goto fail;
866 if (outdata) {
867 if (!(outdata->dptr = (uint8 *)talloc_memdup(
868 mem_ctx, reply->data, reply->datalen))) {
869 TALLOC_FREE(reply);
870 return NT_STATUS_NO_MEMORY;
872 outdata->dsize = reply->datalen;
874 if (cstatus) {
875 (*cstatus) = reply->status;
878 status = NT_STATUS_OK;
880 fail:
881 TALLOC_FREE(new_conn);
882 TALLOC_FREE(reply);
883 return status;
887 * see if a remote process exists
889 bool ctdbd_process_exists(struct ctdbd_connection *conn, uint32 vnn, pid_t pid)
891 NTSTATUS status;
892 TDB_DATA data;
893 int32_t cstatus;
895 data.dptr = (uint8_t*)&pid;
896 data.dsize = sizeof(pid);
898 status = ctdbd_control(conn, vnn, CTDB_CONTROL_PROCESS_EXISTS, 0, 0,
899 data, NULL, NULL, &cstatus);
900 if (!NT_STATUS_IS_OK(status)) {
901 DEBUG(0, (__location__ " ctdb_control for process_exists "
902 "failed\n"));
903 return False;
906 return cstatus == 0;
910 * Get a db path
912 char *ctdbd_dbpath(struct ctdbd_connection *conn,
913 TALLOC_CTX *mem_ctx, uint32_t db_id)
915 NTSTATUS status;
916 TDB_DATA data;
917 int32_t cstatus;
919 data.dptr = (uint8_t*)&db_id;
920 data.dsize = sizeof(db_id);
922 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
923 CTDB_CONTROL_GETDBPATH, 0, 0, data,
924 mem_ctx, &data, &cstatus);
925 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
926 DEBUG(0,(__location__ " ctdb_control for getdbpath failed\n"));
927 return NULL;
930 return (char *)data.dptr;
934 * attach to a ctdb database
936 NTSTATUS ctdbd_db_attach(struct ctdbd_connection *conn,
937 const char *name, uint32_t *db_id, int tdb_flags)
939 NTSTATUS status;
940 TDB_DATA data;
941 int32_t cstatus;
942 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
944 data.dptr = (uint8_t*)name;
945 data.dsize = strlen(name)+1;
947 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
948 persistent
949 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
950 : CTDB_CONTROL_DB_ATTACH,
951 tdb_flags, 0, data, NULL, &data, &cstatus);
952 if (!NT_STATUS_IS_OK(status)) {
953 DEBUG(0, (__location__ " ctdb_control for db_attach "
954 "failed: %s\n", nt_errstr(status)));
955 return status;
958 if (cstatus != 0 || data.dsize != sizeof(uint32_t)) {
959 DEBUG(0,(__location__ " ctdb_control for db_attach failed\n"));
960 return NT_STATUS_INTERNAL_ERROR;
963 *db_id = *(uint32_t *)data.dptr;
964 talloc_free(data.dptr);
966 if (!(tdb_flags & TDB_SEQNUM)) {
967 return NT_STATUS_OK;
970 data.dptr = (uint8_t *)db_id;
971 data.dsize = sizeof(*db_id);
973 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
974 CTDB_CONTROL_ENABLE_SEQNUM, 0, 0, data,
975 NULL, NULL, &cstatus);
976 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
977 DEBUG(0,(__location__ " ctdb_control for enable seqnum "
978 "failed\n"));
979 return NT_STATUS_IS_OK(status) ? NT_STATUS_INTERNAL_ERROR :
980 status;
983 return NT_STATUS_OK;
987 * force the migration of a record to this node
989 NTSTATUS ctdbd_migrate(struct ctdbd_connection *conn, uint32 db_id,
990 TDB_DATA key)
992 struct ctdb_req_call req;
993 struct ctdb_reply_call *reply;
994 NTSTATUS status;
996 ZERO_STRUCT(req);
998 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
999 req.hdr.ctdb_magic = CTDB_MAGIC;
1000 req.hdr.ctdb_version = CTDB_VERSION;
1001 req.hdr.operation = CTDB_REQ_CALL;
1002 req.hdr.reqid = ++conn->reqid;
1003 req.flags = CTDB_IMMEDIATE_MIGRATION;
1004 req.callid = CTDB_NULL_FUNC;
1005 req.db_id = db_id;
1006 req.keylen = key.dsize;
1008 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
1009 ctdb_packet_dump(&req.hdr);
1011 status = packet_send(
1012 conn->pkt, 2,
1013 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
1014 data_blob_const(key.dptr, key.dsize));
1016 if (!NT_STATUS_IS_OK(status)) {
1017 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
1018 return status;
1021 status = packet_flush(conn->pkt);
1023 if (!NT_STATUS_IS_OK(status)) {
1024 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
1025 cluster_fatal("cluster dispatch daemon control write error\n");
1028 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
1030 if (!NT_STATUS_IS_OK(status)) {
1031 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
1032 goto fail;
1035 if (reply->hdr.operation != CTDB_REPLY_CALL) {
1036 DEBUG(0, ("received invalid reply\n"));
1037 status = NT_STATUS_INTERNAL_ERROR;
1038 goto fail;
1041 status = NT_STATUS_OK;
1042 fail:
1044 TALLOC_FREE(reply);
1045 return status;
1049 * remotely fetch a record without locking it or forcing a migration
1051 NTSTATUS ctdbd_fetch(struct ctdbd_connection *conn, uint32 db_id,
1052 TDB_DATA key, TALLOC_CTX *mem_ctx, TDB_DATA *data)
1054 struct ctdb_req_call req;
1055 struct ctdb_reply_call *reply;
1056 NTSTATUS status;
1058 ZERO_STRUCT(req);
1060 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
1061 req.hdr.ctdb_magic = CTDB_MAGIC;
1062 req.hdr.ctdb_version = CTDB_VERSION;
1063 req.hdr.operation = CTDB_REQ_CALL;
1064 req.hdr.reqid = ++conn->reqid;
1065 req.flags = 0;
1066 req.callid = CTDB_FETCH_FUNC;
1067 req.db_id = db_id;
1068 req.keylen = key.dsize;
1070 status = packet_send(
1071 conn->pkt, 2,
1072 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
1073 data_blob_const(key.dptr, key.dsize));
1075 if (!NT_STATUS_IS_OK(status)) {
1076 DEBUG(3, ("packet_send failed: %s\n", nt_errstr(status)));
1077 return status;
1080 status = packet_flush(conn->pkt);
1082 if (!NT_STATUS_IS_OK(status)) {
1083 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
1084 cluster_fatal("cluster dispatch daemon control write error\n");
1087 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
1089 if (!NT_STATUS_IS_OK(status)) {
1090 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
1091 goto fail;
1094 if (reply->hdr.operation != CTDB_REPLY_CALL) {
1095 DEBUG(0, ("received invalid reply\n"));
1096 status = NT_STATUS_INTERNAL_ERROR;
1097 goto fail;
1100 data->dsize = reply->datalen;
1101 if (data->dsize == 0) {
1102 data->dptr = NULL;
1103 goto done;
1106 data->dptr = (uint8 *)talloc_memdup(mem_ctx, &reply->data[0],
1107 reply->datalen);
1108 if (data->dptr == NULL) {
1109 DEBUG(0, ("talloc failed\n"));
1110 status = NT_STATUS_NO_MEMORY;
1111 goto fail;
1114 done:
1115 status = NT_STATUS_OK;
1116 fail:
1117 TALLOC_FREE(reply);
1118 return status;
1121 struct ctdbd_traverse_state {
1122 void (*fn)(TDB_DATA key, TDB_DATA data, void *private_data);
1123 void *private_data;
1127 * Handle a traverse record coming in on the ctdbd connection
1130 static NTSTATUS ctdb_traverse_handler(uint8_t *buf, size_t length,
1131 void *private_data)
1133 struct ctdbd_traverse_state *state =
1134 (struct ctdbd_traverse_state *)private_data;
1136 struct ctdb_req_message *m;
1137 struct ctdb_rec_data *d;
1138 TDB_DATA key, data;
1140 m = (struct ctdb_req_message *)buf;
1142 if (length < sizeof(*m) || m->hdr.length != length) {
1143 DEBUG(0, ("Got invalid message of length %d\n", (int)length));
1144 TALLOC_FREE(buf);
1145 return NT_STATUS_UNEXPECTED_IO_ERROR;
1148 d = (struct ctdb_rec_data *)&m->data[0];
1149 if (m->datalen < sizeof(uint32_t) || m->datalen != d->length) {
1150 DEBUG(0, ("Got invalid traverse data of length %d\n",
1151 (int)m->datalen));
1152 TALLOC_FREE(buf);
1153 return NT_STATUS_UNEXPECTED_IO_ERROR;
1156 key.dsize = d->keylen;
1157 key.dptr = &d->data[0];
1158 data.dsize = d->datalen;
1159 data.dptr = &d->data[d->keylen];
1161 if (key.dsize == 0 && data.dsize == 0) {
1162 /* end of traverse */
1163 return NT_STATUS_END_OF_FILE;
1166 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1167 DEBUG(0, ("Got invalid ltdb header length %d\n",
1168 (int)data.dsize));
1169 TALLOC_FREE(buf);
1170 return NT_STATUS_UNEXPECTED_IO_ERROR;
1172 data.dsize -= sizeof(struct ctdb_ltdb_header);
1173 data.dptr += sizeof(struct ctdb_ltdb_header);
1175 if (state->fn) {
1176 state->fn(key, data, state->private_data);
1179 TALLOC_FREE(buf);
1180 return NT_STATUS_OK;
1184 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1185 connection to ctdbd to avoid the hairy recursive and async problems with
1186 everything in-line.
1189 NTSTATUS ctdbd_traverse(uint32 db_id,
1190 void (*fn)(TDB_DATA key, TDB_DATA data,
1191 void *private_data),
1192 void *private_data)
1194 struct ctdbd_connection *conn;
1195 NTSTATUS status;
1197 TDB_DATA data;
1198 struct ctdb_traverse_start t;
1199 int cstatus;
1200 struct ctdbd_traverse_state state;
1202 status = ctdbd_init_connection(NULL, &conn);
1203 if (!NT_STATUS_IS_OK(status)) {
1204 DEBUG(0, ("ctdbd_init_connection failed: %s\n",
1205 nt_errstr(status)));
1206 return status;
1209 t.db_id = db_id;
1210 t.srvid = conn->rand_srvid;
1211 t.reqid = ++conn->reqid;
1213 data.dptr = (uint8_t *)&t;
1214 data.dsize = sizeof(t);
1216 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1217 CTDB_CONTROL_TRAVERSE_START, conn->rand_srvid, 0,
1218 data, NULL, NULL, &cstatus);
1220 if (!NT_STATUS_IS_OK(status) || (cstatus != 0)) {
1222 DEBUG(0,("ctdbd_control failed: %s, %d\n", nt_errstr(status),
1223 cstatus));
1225 if (NT_STATUS_IS_OK(status)) {
1227 * We need a mapping here
1229 status = NT_STATUS_UNSUCCESSFUL;
1231 goto done;
1234 state.fn = fn;
1235 state.private_data = private_data;
1237 while (True) {
1239 status = NT_STATUS_OK;
1241 if (packet_handler(conn->pkt, ctdb_req_complete,
1242 ctdb_traverse_handler, &state, &status)) {
1244 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1245 status = NT_STATUS_OK;
1246 break;
1250 * There might be more in the queue
1252 continue;
1255 if (!NT_STATUS_IS_OK(status)) {
1256 break;
1259 status = ctdb_packet_fd_read_sync(conn->pkt);
1261 if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
1263 * There might be more in the queue
1265 continue;
1268 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1269 status = NT_STATUS_OK;
1270 break;
1273 if (!NT_STATUS_IS_OK(status)) {
1274 DEBUG(0, ("packet_fd_read_sync failed: %s\n", nt_errstr(status)));
1275 cluster_fatal("ctdbd died\n");
1279 done:
1280 TALLOC_FREE(conn);
1281 return status;
1285 This is used to canonicalize a ctdb_sock_addr structure.
1287 static void smbd_ctdb_canonicalize_ip(const struct sockaddr_storage *in,
1288 struct sockaddr_storage *out)
1290 memcpy(out, in, sizeof (*out));
1292 #ifdef HAVE_IPV6
1293 if (in->ss_family == AF_INET6) {
1294 const char prefix[12] = { 0,0,0,0,0,0,0,0,0,0,0xff,0xff };
1295 const struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)in;
1296 struct sockaddr_in *out4 = (struct sockaddr_in *)out;
1297 if (memcmp(&in6->sin6_addr, prefix, 12) == 0) {
1298 memset(out, 0, sizeof(*out));
1299 #ifdef HAVE_SOCK_SIN_LEN
1300 out4->sin_len = sizeof(*out);
1301 #endif
1302 out4->sin_family = AF_INET;
1303 out4->sin_port = in6->sin6_port;
1304 memcpy(&out4->sin_addr, &in6->sin6_addr.s6_addr32[3], 4);
1307 #endif
1311 * Register us as a server for a particular tcp connection
1314 NTSTATUS ctdbd_register_ips(struct ctdbd_connection *conn,
1315 const struct sockaddr_storage *_server,
1316 const struct sockaddr_storage *_client,
1317 void (*release_ip_handler)(const char *ip_addr,
1318 void *private_data),
1319 void *private_data)
1322 * we still use ctdb_control_tcp for ipv4
1323 * because we want to work against older ctdb
1324 * versions at runtime
1326 struct ctdb_control_tcp p4;
1327 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1328 struct ctdb_control_tcp_addr p;
1329 #endif
1330 TDB_DATA data;
1331 NTSTATUS status;
1332 struct sockaddr_storage client;
1333 struct sockaddr_storage server;
1336 * Only one connection so far
1338 SMB_ASSERT(conn->release_ip_handler == NULL);
1340 smbd_ctdb_canonicalize_ip(_client, &client);
1341 smbd_ctdb_canonicalize_ip(_server, &server);
1343 switch (client.ss_family) {
1344 case AF_INET:
1345 p4.dest = *(struct sockaddr_in *)(void *)&server;
1346 p4.src = *(struct sockaddr_in *)(void *)&client;
1347 data.dptr = (uint8_t *)&p4;
1348 data.dsize = sizeof(p4);
1349 break;
1350 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1351 case AF_INET6:
1352 p.dest.ip6 = *(struct sockaddr_in6 *)(void *)&server;
1353 p.src.ip6 = *(struct sockaddr_in6 *)(void *)&client;
1354 data.dptr = (uint8_t *)&p;
1355 data.dsize = sizeof(p);
1356 break;
1357 #endif
1358 default:
1359 return NT_STATUS_INTERNAL_ERROR;
1362 conn->release_ip_handler = release_ip_handler;
1365 * We want to be told about IP releases
1368 status = register_with_ctdbd(conn, CTDB_SRVID_RELEASE_IP);
1369 if (!NT_STATUS_IS_OK(status)) {
1370 return status;
1374 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1375 * can send an extra ack to trigger a reset for our client, so it
1376 * immediately reconnects
1378 return ctdbd_control(conn, CTDB_CURRENT_NODE,
1379 CTDB_CONTROL_TCP_CLIENT, 0,
1380 CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL, NULL);
1384 * We want to handle reconfigure events
1386 NTSTATUS ctdbd_register_reconfigure(struct ctdbd_connection *conn)
1388 return register_with_ctdbd(conn, CTDB_SRVID_RECONFIGURE);
1392 call a control on the local node
1394 NTSTATUS ctdbd_control_local(struct ctdbd_connection *conn, uint32 opcode,
1395 uint64_t srvid, uint32_t flags, TDB_DATA data,
1396 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
1397 int *cstatus)
1399 return ctdbd_control(conn, CTDB_CURRENT_NODE, opcode, srvid, flags, data, mem_ctx, outdata, cstatus);
1402 NTSTATUS ctdb_watch_us(struct ctdbd_connection *conn)
1404 struct ctdb_client_notify_register reg_data;
1405 size_t struct_len;
1406 NTSTATUS status;
1407 int cstatus;
1409 reg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1410 reg_data.len = 1;
1411 reg_data.notify_data[0] = 0;
1413 struct_len = offsetof(struct ctdb_client_notify_register,
1414 notify_data) + reg_data.len;
1416 status = ctdbd_control_local(
1417 conn, CTDB_CONTROL_REGISTER_NOTIFY, conn->rand_srvid, 0,
1418 make_tdb_data((uint8_t *)&reg_data, struct_len),
1419 NULL, NULL, &cstatus);
1420 if (!NT_STATUS_IS_OK(status)) {
1421 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1422 nt_errstr(status)));
1424 return status;
1427 NTSTATUS ctdb_unwatch(struct ctdbd_connection *conn)
1429 struct ctdb_client_notify_deregister dereg_data;
1430 NTSTATUS status;
1431 int cstatus;
1433 dereg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1435 status = ctdbd_control_local(
1436 conn, CTDB_CONTROL_DEREGISTER_NOTIFY, conn->rand_srvid, 0,
1437 make_tdb_data((uint8_t *)&dereg_data, sizeof(dereg_data)),
1438 NULL, NULL, &cstatus);
1439 if (!NT_STATUS_IS_OK(status)) {
1440 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1441 nt_errstr(status)));
1443 return status;
1446 #endif