?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
smbd: Use full_path_tos() where appropriate
[Samba/wip.git] / source3 / lib / ctdbd_conn.c
blob201c70029050ff52cd18d4402f155be2037660e2
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
6
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.
11
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.
16
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/>.
19 */
20
21 #include "includes.h"
22 #include "util_tdb.h"
23 #include "serverid.h"
24 #include "ctdbd_conn.h"
25
26 #include "ctdb_packet.h"
27 #include "messages.h"
28
29 /*
30 * It is not possible to include ctdb.h and tdb_compat.h (included via
31 * some other include above) without warnings. This fixes those
32 * warnings.
33 */
34
35 #ifdef typesafe_cb
36 #undef typesafe_cb
37 #endif
38
39 #ifdef typesafe_cb_preargs
40 #undef typesafe_cb_preargs
41 #endif
42
43 #ifdef typesafe_cb_postargs
44 #undef typesafe_cb_postargs
45 #endif
46
47 /* paths to these include files come from --with-ctdb= in configure */
48
49 #include "ctdb.h"
50 #include "ctdb_private.h"
51
52 struct ctdbd_connection {
53 struct messaging_context *msg_ctx;
54 uint32_t reqid;
55 uint32_t our_vnn;
56 uint64_t rand_srvid;
57 struct ctdb_packet_context *pkt;
58 struct tevent_fd *fde;
59
60 bool (*release_ip_handler)(const char *ip_addr, void *private_data);
61 void *release_ip_priv;
62 };
63
64 static uint32_t ctdbd_next_reqid(struct ctdbd_connection *conn)
65 {
66 conn->reqid += 1;
67 if (conn->reqid == 0) {
68 conn->reqid += 1;
69 }
70 return conn->reqid;
71 }
72
73 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
74 uint32_t vnn, uint32_t opcode,
75 uint64_t srvid, uint32_t flags, TDB_DATA data,
76 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
77 int *cstatus);
78
79 /*
80 * exit on fatal communications errors with the ctdbd daemon
81 */
82 static void cluster_fatal(const char *why)
83 {
84 DEBUG(0,("cluster fatal event: %s - exiting immediately\n", why));
85 /* we don't use smb_panic() as we don't want to delay to write
86 a core file. We need to release this process id immediately
87 so that someone else can take over without getting sharing
88 violations */
89 _exit(1);
90 }
91
92 /*
93 *
94 */
95 static void ctdb_packet_dump(struct ctdb_req_header *hdr)
96 {
97 if (DEBUGLEVEL < 11) {
98 return;
99 }
100 DEBUGADD(11, ("len=%d, magic=%x, vers=%d, gen=%d, op=%d, reqid=%d\n",
101 (int)hdr->length, (int)hdr->ctdb_magic,
102 (int)hdr->ctdb_version, (int)hdr->generation,
103 (int)hdr->operation, (int)hdr->reqid));
104 }
105
106 /*
107 * Register a srvid with ctdbd
108 */
109 NTSTATUS register_with_ctdbd(struct ctdbd_connection *conn, uint64_t srvid)
110 {
111
112 int cstatus;
113 return ctdbd_control(conn, CTDB_CURRENT_NODE,
114 CTDB_CONTROL_REGISTER_SRVID, srvid, 0,
115 tdb_null, NULL, NULL, &cstatus);
116 }
117
118 /*
119 * get our vnn from the cluster
120 */
121 static NTSTATUS get_cluster_vnn(struct ctdbd_connection *conn, uint32_t *vnn)
122 {
123 int32_t cstatus=-1;
124 NTSTATUS status;
125 status = ctdbd_control(conn,
126 CTDB_CURRENT_NODE, CTDB_CONTROL_GET_PNN, 0, 0,
127 tdb_null, NULL, NULL, &cstatus);
128 if (!NT_STATUS_IS_OK(status)) {
129 DEBUG(1, ("ctdbd_control failed: %s\n", nt_errstr(status)));
130 return status;
131 }
132 *vnn = (uint32_t)cstatus;
133 return status;
134 }
135
136 /*
137 * Are we active (i.e. not banned or stopped?)
138 */
139 static bool ctdbd_working(struct ctdbd_connection *conn, uint32_t vnn)
140 {
141 int32_t cstatus=-1;
142 NTSTATUS status;
143 TDB_DATA outdata;
144 struct ctdb_node_map *m;
145 uint32_t failure_flags;
146 bool ret = false;
147 int i;
148
149 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
150 CTDB_CONTROL_GET_NODEMAP, 0, 0,
151 tdb_null, talloc_tos(), &outdata, &cstatus);
152 if (!NT_STATUS_IS_OK(status)) {
153 DEBUG(1, ("ctdbd_control failed: %s\n", nt_errstr(status)));
154 return false;
155 }
156 if ((cstatus != 0) || (outdata.dptr == NULL)) {
157 DEBUG(2, ("Received invalid ctdb data\n"));
158 return false;
159 }
160
161 m = (struct ctdb_node_map *)outdata.dptr;
162
163 for (i=0; i<m->num; i++) {
164 if (vnn == m->nodes[i].pnn) {
165 break;
166 }
167 }
168
169 if (i == m->num) {
170 DEBUG(2, ("Did not find ourselves (node %d) in nodemap\n",
171 (int)vnn));
172 goto fail;
173 }
174
175 failure_flags = NODE_FLAGS_BANNED | NODE_FLAGS_DISCONNECTED
176 | NODE_FLAGS_PERMANENTLY_DISABLED | NODE_FLAGS_STOPPED;
177
178 if ((m->nodes[i].flags & failure_flags) != 0) {
179 DEBUG(2, ("Node has status %x, not active\n",
180 (int)m->nodes[i].flags));
181 goto fail;
182 }
183
184 ret = true;
185 fail:
186 TALLOC_FREE(outdata.dptr);
187 return ret;
188 }
189
190 uint32_t ctdbd_vnn(const struct ctdbd_connection *conn)
191 {
192 return conn->our_vnn;
193 }
194
195 const char *lp_ctdbd_socket(void)
196 {
197 const char *ret;
198
199 ret = lp__ctdbd_socket();
200 if (ret != NULL && strlen(ret) > 0) {
201 return ret;
202 }
203
204 return CTDB_PATH;
205 }
206
207 /*
208 * Get us a ctdb connection
209 */
210
211 static NTSTATUS ctdbd_connect(TALLOC_CTX *mem_ctx,
212 struct ctdb_packet_context **presult)
213 {
214 struct ctdb_packet_context *result;
215 const char *sockname = lp_ctdbd_socket();
216 struct sockaddr_un addr = { 0, };
217 int fd;
218 socklen_t salen;
219
220 fd = socket(AF_UNIX, SOCK_STREAM, 0);
221 if (fd == -1) {
222 DEBUG(3, ("Could not create socket: %s\n", strerror(errno)));
223 return map_nt_error_from_unix(errno);
224 }
225
226 addr.sun_family = AF_UNIX;
227 snprintf(addr.sun_path, sizeof(addr.sun_path), "%s", sockname);
228
229 salen = sizeof(struct sockaddr_un);
230 if (connect(fd, (struct sockaddr *)(void *)&addr, salen) == -1) {
231 DEBUG(1, ("connect(%s) failed: %s\n", sockname,
232 strerror(errno)));
233 close(fd);
234 return map_nt_error_from_unix(errno);
235 }
236
237 if (!(result = ctdb_packet_init(mem_ctx, fd))) {
238 close(fd);
239 return NT_STATUS_NO_MEMORY;
240 }
241
242 *presult = result;
243 return NT_STATUS_OK;
244 }
245
246 /*
247 * Do we have a complete ctdb packet in the queue?
248 */
249
250 static bool ctdb_req_complete(const uint8_t *buf, size_t available,
251 size_t *length,
252 void *private_data)
253 {
254 uint32_t msglen;
255
256 if (available < sizeof(msglen)) {
257 return False;
258 }
259
260 msglen = *((const uint32_t *)buf);
261
262 DEBUG(11, ("msglen = %d\n", msglen));
263
264 if (msglen < sizeof(struct ctdb_req_header)) {
265 DEBUG(0, ("Got invalid msglen: %d, expected at least %d for "
266 "the req_header\n", (int)msglen,
267 (int)sizeof(struct ctdb_req_header)));
268 cluster_fatal("ctdbd protocol error\n");
269 }
270
271 if (available < msglen) {
272 return false;
273 }
274
275 *length = msglen;
276 return true;
277 }
278
279 /*
280 * State necessary to defer an incoming message while we are waiting for a
281 * ctdb reply.
282 */
283
284 struct deferred_msg_state {
285 struct messaging_context *msg_ctx;
286 struct messaging_rec *rec;
287 };
288
289 /*
290 * Timed event handler for the deferred message
291 */
292
293 static void deferred_message_dispatch(struct tevent_context *event_ctx,
294 struct tevent_timer *te,
295 struct timeval now,
296 void *private_data)
297 {
298 struct deferred_msg_state *state = talloc_get_type_abort(
299 private_data, struct deferred_msg_state);
300
301 messaging_dispatch_rec(state->msg_ctx, state->rec);
302 TALLOC_FREE(state);
303 TALLOC_FREE(te);
304 }
305
306 struct req_pull_state {
307 TALLOC_CTX *mem_ctx;
308 DATA_BLOB req;
309 };
310
311 /*
312 * Pull a ctdb request out of the incoming ctdb_packet queue
313 */
314
315 static NTSTATUS ctdb_req_pull(uint8_t *buf, size_t length,
316 void *private_data)
317 {
318 struct req_pull_state *state = (struct req_pull_state *)private_data;
319
320 state->req.data = talloc_move(state->mem_ctx, &buf);
321 state->req.length = length;
322 return NT_STATUS_OK;
323 }
324
325 /*
326 * Fetch a messaging_rec from an incoming ctdb style message
327 */
328
329 static struct messaging_rec *ctdb_pull_messaging_rec(TALLOC_CTX *mem_ctx,
330 size_t overall_length,
331 struct ctdb_req_message *msg)
332 {
333 struct messaging_rec *result;
334 DATA_BLOB blob;
335 enum ndr_err_code ndr_err;
336
337 if ((overall_length < offsetof(struct ctdb_req_message, data))
338 || (overall_length
339 < offsetof(struct ctdb_req_message, data) + msg->datalen)) {
340
341 cluster_fatal("got invalid msg length");
342 }
343
344 if (!(result = talloc(mem_ctx, struct messaging_rec))) {
345 DEBUG(0, ("talloc failed\n"));
346 return NULL;
347 }
348
349 blob = data_blob_const(msg->data, msg->datalen);
350
351 ndr_err = ndr_pull_struct_blob(
352 &blob, result, result,
353 (ndr_pull_flags_fn_t)ndr_pull_messaging_rec);
354
355 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
356 DEBUG(0, ("ndr_pull_struct_blob failed: %s\n",
357 ndr_errstr(ndr_err)));
358 TALLOC_FREE(result);
359 return NULL;
360 }
361
362 if (DEBUGLEVEL >= 11) {
363 DEBUG(11, ("ctdb_pull_messaging_rec:\n"));
364 NDR_PRINT_DEBUG(messaging_rec, result);
365 }
366
367 return result;
368 }
369
370 static NTSTATUS ctdb_packet_fd_read_sync(struct ctdb_packet_context *ctx)
371 {
372 int timeout = lp_ctdb_timeout();
373
374 if (timeout == 0) {
375 timeout = -1;
376 }
377 return ctdb_packet_fd_read_sync_timeout(ctx, timeout);
378 }
379
380 /*
381 * Read a full ctdbd request. If we have a messaging context, defer incoming
382 * messages that might come in between.
383 */
384
385 static NTSTATUS ctdb_read_req(struct ctdbd_connection *conn, uint32_t reqid,
386 TALLOC_CTX *mem_ctx, void *result)
387 {
388 struct ctdb_req_header *hdr;
389 struct req_pull_state state;
390 NTSTATUS status;
391
392 next_pkt:
393 ZERO_STRUCT(state);
394 state.mem_ctx = mem_ctx;
395
396 while (!ctdb_packet_handler(conn->pkt, ctdb_req_complete,
397 ctdb_req_pull, &state, &status)) {
398 /*
399 * Not enough data
400 */
401 status = ctdb_packet_fd_read_sync(conn->pkt);
402
403 if (NT_STATUS_EQUAL(status, NT_STATUS_NETWORK_BUSY)) {
404 /* EAGAIN */
405 continue;
406 } else if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
407 /* EAGAIN */
408 continue;
409 }
410
411 if (!NT_STATUS_IS_OK(status)) {
412 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
413 cluster_fatal("ctdbd died\n");
414 }
415 }
416
417 if (!NT_STATUS_IS_OK(status)) {
418 DEBUG(0, ("Could not read ctdb_packet: %s\n", nt_errstr(status)));
419 cluster_fatal("ctdbd died\n");
420 }
421
422 hdr = (struct ctdb_req_header *)state.req.data;
423
424 DEBUG(11, ("Received ctdb packet\n"));
425 ctdb_packet_dump(hdr);
426
427 if (hdr->operation == CTDB_REQ_MESSAGE) {
428 struct tevent_timer *evt;
429 struct deferred_msg_state *msg_state;
430 struct ctdb_req_message *msg = (struct ctdb_req_message *)hdr;
431
432 if (conn->msg_ctx == NULL) {
433 DEBUG(1, ("Got a message without having a msg ctx, "
434 "dropping msg %llu\n",
435 (long long unsigned)msg->srvid));
436 goto next_pkt;
437 }
438
439 if ((conn->release_ip_handler != NULL)
440 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
441 bool ret;
442
443 /* must be dispatched immediately */
444 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
445 ret = conn->release_ip_handler((const char *)msg->data,
446 conn->release_ip_priv);
447 if (ret) {
448 /*
449 * We need to release the ip,
450 * so return an error to the upper layers.
451 *
452 * We make sure we don't trigger this again.
453 */
454 conn->release_ip_handler = NULL;
455 conn->release_ip_priv = NULL;
456 return NT_STATUS_ADDRESS_CLOSED;
457 }
458 TALLOC_FREE(hdr);
459 goto next_pkt;
460 }
461
462 if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
463 || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)) {
464
465 DEBUG(1, ("ctdb_read_req: Got %s message\n",
466 (msg->srvid == CTDB_SRVID_RECONFIGURE)
467 ? "cluster reconfigure" : "SAMBA_NOTIFY"));
468
469 messaging_send(conn->msg_ctx,
470 messaging_server_id(conn->msg_ctx),
471 MSG_SMB_BRL_VALIDATE, &data_blob_null);
472 TALLOC_FREE(hdr);
473 goto next_pkt;
474 }
475
476 msg_state = talloc(NULL, struct deferred_msg_state);
477 if (msg_state == NULL) {
478 DEBUG(0, ("talloc failed\n"));
479 TALLOC_FREE(hdr);
480 goto next_pkt;
481 }
482
483 if (!(msg_state->rec = ctdb_pull_messaging_rec(
484 msg_state, state.req.length, msg))) {
485 DEBUG(0, ("ctdbd_pull_messaging_rec failed\n"));
486 TALLOC_FREE(msg_state);
487 TALLOC_FREE(hdr);
488 goto next_pkt;
489 }
490
491 TALLOC_FREE(hdr);
492
493 msg_state->msg_ctx = conn->msg_ctx;
494
495 /*
496 * We're waiting for a call reply, but an async message has
497 * crossed. Defer dispatching to the toplevel event loop.
498 */
499 evt = tevent_add_timer(messaging_tevent_context(conn->msg_ctx),
500 messaging_tevent_context(conn->msg_ctx),
501 timeval_zero(),
502 deferred_message_dispatch,
503 msg_state);
504 if (evt == NULL) {
505 DEBUG(0, ("event_add_timed failed\n"));
506 TALLOC_FREE(msg_state);
507 TALLOC_FREE(hdr);
508 goto next_pkt;
509 }
510
511 goto next_pkt;
512 }
513
514 if ((reqid != 0) && (hdr->reqid != reqid)) {
515 /* we got the wrong reply */
516 DEBUG(0,("Discarding mismatched ctdb reqid %u should have "
517 "been %u\n", hdr->reqid, reqid));
518 TALLOC_FREE(hdr);
519 goto next_pkt;
520 }
521
522 *((void **)result) = talloc_move(mem_ctx, &hdr);
523
524 return NT_STATUS_OK;
525 }
526
527 /*
528 * Get us a ctdbd connection
529 */
530
531 static NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
532 struct ctdbd_connection **pconn)
533 {
534 struct ctdbd_connection *conn;
535 NTSTATUS status;
536
537 if (!(conn = talloc_zero(mem_ctx, struct ctdbd_connection))) {
538 DEBUG(0, ("talloc failed\n"));
539 return NT_STATUS_NO_MEMORY;
540 }
541
542 status = ctdbd_connect(conn, &conn->pkt);
543
544 if (!NT_STATUS_IS_OK(status)) {
545 DEBUG(10, ("ctdbd_connect failed: %s\n", nt_errstr(status)));
546 goto fail;
547 }
548
549 status = get_cluster_vnn(conn, &conn->our_vnn);
550
551 if (!NT_STATUS_IS_OK(status)) {
552 DEBUG(10, ("get_cluster_vnn failed: %s\n", nt_errstr(status)));
553 goto fail;
554 }
555
556 if (!ctdbd_working(conn, conn->our_vnn)) {
557 DEBUG(2, ("Node is not working, can not connect\n"));
558 status = NT_STATUS_INTERNAL_DB_ERROR;
559 goto fail;
560 }
561
562 generate_random_buffer((unsigned char *)&conn->rand_srvid,
563 sizeof(conn->rand_srvid));
564
565 status = register_with_ctdbd(conn, conn->rand_srvid);
566
567 if (!NT_STATUS_IS_OK(status)) {
568 DEBUG(5, ("Could not register random srvid: %s\n",
569 nt_errstr(status)));
570 goto fail;
571 }
572
573 *pconn = conn;
574 return NT_STATUS_OK;
575
576 fail:
577 TALLOC_FREE(conn);
578 return status;
579 }
580
581 /*
582 * Get us a ctdbd connection and register us as a process
583 */
584
585 NTSTATUS ctdbd_messaging_connection(TALLOC_CTX *mem_ctx,
586 struct ctdbd_connection **pconn)
587 {
588 struct ctdbd_connection *conn;
589 NTSTATUS status;
590
591 status = ctdbd_init_connection(mem_ctx, &conn);
592
593 if (!NT_STATUS_IS_OK(status)) {
594 return status;
595 }
596
597 status = register_with_ctdbd(conn, (uint64_t)getpid());
598 if (!NT_STATUS_IS_OK(status)) {
599 goto fail;
600 }
601
602 status = register_with_ctdbd(conn, MSG_SRVID_SAMBA);
603 if (!NT_STATUS_IS_OK(status)) {
604 goto fail;
605 }
606
607 status = register_with_ctdbd(conn, CTDB_SRVID_SAMBA_NOTIFY);
608 if (!NT_STATUS_IS_OK(status)) {
609 goto fail;
610 }
611
612 *pconn = conn;
613 return NT_STATUS_OK;
614
615 fail:
616 TALLOC_FREE(conn);
617 return status;
618 }
619
620 struct messaging_context *ctdb_conn_msg_ctx(struct ctdbd_connection *conn)
621 {
622 return conn->msg_ctx;
623 }
624
625 int ctdbd_conn_get_fd(struct ctdbd_connection *conn)
626 {
627 return ctdb_packet_get_fd(conn->pkt);
628 }
629
630 /*
631 * Packet handler to receive and handle a ctdb message
632 */
633 static NTSTATUS ctdb_handle_message(uint8_t *buf, size_t length,
634 void *private_data)
635 {
636 struct ctdbd_connection *conn = talloc_get_type_abort(
637 private_data, struct ctdbd_connection);
638 struct ctdb_req_message *msg;
639 struct messaging_rec *msg_rec;
640
641 msg = (struct ctdb_req_message *)buf;
642
643 if (msg->hdr.operation != CTDB_REQ_MESSAGE) {
644 DEBUG(0, ("Received async msg of type %u, discarding\n",
645 msg->hdr.operation));
646 TALLOC_FREE(buf);
647 return NT_STATUS_INVALID_PARAMETER;
648 }
649
650 if ((conn->release_ip_handler != NULL)
651 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
652 bool ret;
653
654 /* must be dispatched immediately */
655 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
656 ret = conn->release_ip_handler((const char *)msg->data,
657 conn->release_ip_priv);
658 if (ret) {
659 /*
660 * We need to release the ip.
661 *
662 * We make sure we don't trigger this again.
663 */
664 conn->release_ip_handler = NULL;
665 conn->release_ip_priv = NULL;
666 }
667 TALLOC_FREE(buf);
668 return NT_STATUS_OK;
669 }
670
671 SMB_ASSERT(conn->msg_ctx != NULL);
672
673 if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
674 || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)){
675 DEBUG(0,("Got cluster reconfigure message\n"));
676 /*
677 * when the cluster is reconfigured or someone of the
678 * family has passed away (SAMBA_NOTIFY), we need to
679 * clean the brl database
680 */
681 messaging_send(conn->msg_ctx,
682 messaging_server_id(conn->msg_ctx),
683 MSG_SMB_BRL_VALIDATE, &data_blob_null);
684
685 TALLOC_FREE(buf);
686 return NT_STATUS_OK;
687 }
688
689 /* only messages to our pid or the broadcast are valid here */
690 if (msg->srvid != getpid() && msg->srvid != MSG_SRVID_SAMBA) {
691 DEBUG(0,("Got unexpected message with srvid=%llu\n",
692 (unsigned long long)msg->srvid));
693 TALLOC_FREE(buf);
694 return NT_STATUS_OK;
695 }
696
697 if (!(msg_rec = ctdb_pull_messaging_rec(NULL, length, msg))) {
698 DEBUG(10, ("ctdb_pull_messaging_rec failed\n"));
699 TALLOC_FREE(buf);
700 return NT_STATUS_NO_MEMORY;
701 }
702
703 messaging_dispatch_rec(conn->msg_ctx, msg_rec);
704
705 TALLOC_FREE(msg_rec);
706 TALLOC_FREE(buf);
707 return NT_STATUS_OK;
708 }
709
710 /*
711 * The ctdbd socket is readable asynchronuously
712 */
713
714 static void ctdbd_socket_handler(struct tevent_context *event_ctx,
715 struct tevent_fd *event,
716 uint16 flags,
717 void *private_data)
718 {
719 struct ctdbd_connection *conn = talloc_get_type_abort(
720 private_data, struct ctdbd_connection);
721
722 NTSTATUS status;
723
724 status = ctdb_packet_fd_read(conn->pkt);
725
726 if (!NT_STATUS_IS_OK(status)) {
727 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
728 cluster_fatal("ctdbd died\n");
729 }
730
731 while (ctdb_packet_handler(conn->pkt, ctdb_req_complete,
732 ctdb_handle_message, conn, &status)) {
733 if (!NT_STATUS_IS_OK(status)) {
734 DEBUG(10, ("could not handle incoming message: %s\n",
735 nt_errstr(status)));
736 }
737 }
738 }
739
740 /*
741 * Prepare a ctdbd connection to receive messages
742 */
743
744 NTSTATUS ctdbd_register_msg_ctx(struct ctdbd_connection *conn,
745 struct messaging_context *msg_ctx)
746 {
747 SMB_ASSERT(conn->msg_ctx == NULL);
748 SMB_ASSERT(conn->fde == NULL);
749
750 if (!(conn->fde = tevent_add_fd(messaging_tevent_context(msg_ctx),
751 conn,
752 ctdb_packet_get_fd(conn->pkt),
753 TEVENT_FD_READ,
754 ctdbd_socket_handler,
755 conn))) {
756 DEBUG(0, ("event_add_fd failed\n"));
757 return NT_STATUS_NO_MEMORY;
758 }
759
760 conn->msg_ctx = msg_ctx;
761
762 return NT_STATUS_OK;
763 }
764
765 /*
766 * Send a messaging message across a ctdbd
767 */
768
769 NTSTATUS ctdbd_messaging_send(struct ctdbd_connection *conn,
770 uint32_t dst_vnn, uint64_t dst_srvid,
771 struct messaging_rec *msg)
772 {
773 DATA_BLOB blob;
774 NTSTATUS status;
775 enum ndr_err_code ndr_err;
776
777 ndr_err = ndr_push_struct_blob(
778 &blob, talloc_tos(), msg,
779 (ndr_push_flags_fn_t)ndr_push_messaging_rec);
780
781 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
782 DEBUG(0, ("ndr_push_struct_blob failed: %s\n",
783 ndr_errstr(ndr_err)));
784 return ndr_map_error2ntstatus(ndr_err);
785 }
786
787 status = ctdbd_messaging_send_blob(conn, dst_vnn, dst_srvid,
788 blob.data, blob.length);
789 TALLOC_FREE(blob.data);
790 return status;
791 }
792
793 NTSTATUS ctdbd_messaging_send_blob(struct ctdbd_connection *conn,
794 uint32_t dst_vnn, uint64_t dst_srvid,
795 const uint8_t *buf, size_t buflen)
796 {
797 struct ctdb_req_message r;
798 NTSTATUS status;
799
800 r.hdr.length = offsetof(struct ctdb_req_message, data) + buflen;
801 r.hdr.ctdb_magic = CTDB_MAGIC;
802 r.hdr.ctdb_version = CTDB_VERSION;
803 r.hdr.generation = 1;
804 r.hdr.operation = CTDB_REQ_MESSAGE;
805 r.hdr.destnode = dst_vnn;
806 r.hdr.srcnode = conn->our_vnn;
807 r.hdr.reqid = 0;
808 r.srvid = dst_srvid;
809 r.datalen = buflen;
810
811 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
812 ctdb_packet_dump(&r.hdr);
813
814 status = ctdb_packet_send(
815 conn->pkt, 2,
816 data_blob_const(&r, offsetof(struct ctdb_req_message, data)),
817 data_blob_const(buf, buflen));
818
819 if (!NT_STATUS_IS_OK(status)) {
820 DEBUG(0, ("ctdb_packet_send failed: %s\n", nt_errstr(status)));
821 return status;
822 }
823
824 status = ctdb_packet_flush(conn->pkt);
825 if (!NT_STATUS_IS_OK(status)) {
826 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
827 cluster_fatal("cluster dispatch daemon msg write error\n");
828 }
829 return NT_STATUS_OK;
830 }
831
832 /*
833 * send/recv a generic ctdb control message
834 */
835 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
836 uint32_t vnn, uint32_t opcode,
837 uint64_t srvid, uint32_t flags,
838 TDB_DATA data,
839 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
840 int *cstatus)
841 {
842 struct ctdb_req_control req;
843 struct ctdb_reply_control *reply = NULL;
844 struct ctdbd_connection *new_conn = NULL;
845 NTSTATUS status;
846
847 if (conn == NULL) {
848 status = ctdbd_init_connection(NULL, &new_conn);
849
850 if (!NT_STATUS_IS_OK(status)) {
851 DEBUG(10, ("Could not init temp connection: %s\n",
852 nt_errstr(status)));
853 goto fail;
854 }
855
856 conn = new_conn;
857 }
858
859 ZERO_STRUCT(req);
860 req.hdr.length = offsetof(struct ctdb_req_control, data) + data.dsize;
861 req.hdr.ctdb_magic = CTDB_MAGIC;
862 req.hdr.ctdb_version = CTDB_VERSION;
863 req.hdr.operation = CTDB_REQ_CONTROL;
864 req.hdr.reqid = ctdbd_next_reqid(conn);
865 req.hdr.destnode = vnn;
866 req.opcode = opcode;
867 req.srvid = srvid;
868 req.datalen = data.dsize;
869 req.flags = flags;
870
871 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
872 ctdb_packet_dump(&req.hdr);
873
874 status = ctdb_packet_send(
875 conn->pkt, 2,
876 data_blob_const(&req, offsetof(struct ctdb_req_control, data)),
877 data_blob_const(data.dptr, data.dsize));
878
879 if (!NT_STATUS_IS_OK(status)) {
880 DEBUG(3, ("ctdb_packet_send failed: %s\n", nt_errstr(status)));
881 goto fail;
882 }
883
884 status = ctdb_packet_flush(conn->pkt);
885
886 if (!NT_STATUS_IS_OK(status)) {
887 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
888 cluster_fatal("cluster dispatch daemon control write error\n");
889 }
890
891 if (flags & CTDB_CTRL_FLAG_NOREPLY) {
892 TALLOC_FREE(new_conn);
893 if (cstatus) {
894 *cstatus = 0;
895 }
896 return NT_STATUS_OK;
897 }
898
899 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
900
901 if (!NT_STATUS_IS_OK(status)) {
902 DEBUG(10, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
903 goto fail;
904 }
905
906 if (reply->hdr.operation != CTDB_REPLY_CONTROL) {
907 DEBUG(0, ("received invalid reply\n"));
908 goto fail;
909 }
910
911 if (outdata) {
912 if (!(outdata->dptr = (uint8 *)talloc_memdup(
913 mem_ctx, reply->data, reply->datalen))) {
914 TALLOC_FREE(reply);
915 return NT_STATUS_NO_MEMORY;
916 }
917 outdata->dsize = reply->datalen;
918 }
919 if (cstatus) {
920 (*cstatus) = reply->status;
921 }
922
923 status = NT_STATUS_OK;
924
925 fail:
926 TALLOC_FREE(new_conn);
927 TALLOC_FREE(reply);
928 return status;
929 }
930
931 /*
932 * see if a remote process exists
933 */
934 bool ctdbd_process_exists(struct ctdbd_connection *conn, uint32_t vnn, pid_t pid)
935 {
936 struct server_id id;
937 bool result;
938
939 id.pid = pid;
940 id.vnn = vnn;
941
942 if (!ctdb_processes_exist(conn, &id, 1, &result)) {
943 DEBUG(10, ("ctdb_processes_exist failed\n"));
944 return false;
945 }
946 return result;
947 }
948
949 bool ctdb_processes_exist(struct ctdbd_connection *conn,
950 const struct server_id *pids, int num_pids,
951 bool *results)
952 {
953 TALLOC_CTX *frame = talloc_stackframe();
954 int i, num_received;
955 NTSTATUS status;
956 uint32_t *reqids;
957 bool result = false;
958
959 reqids = talloc_array(talloc_tos(), uint32_t, num_pids);
960 if (reqids == NULL) {
961 goto fail;
962 }
963
964 for (i=0; i<num_pids; i++) {
965 struct ctdb_req_control req;
966 pid_t pid;
967
968 results[i] = false;
969 reqids[i] = ctdbd_next_reqid(conn);
970
971 ZERO_STRUCT(req);
972
973 /*
974 * pids[i].pid is uint64_t, scale down to pid_t which
975 * is the wire protocol towards ctdb.
976 */
977 pid = pids[i].pid;
978
979 DEBUG(10, ("Requesting PID %d/%d, reqid=%d\n",
980 (int)pids[i].vnn, (int)pid,
981 (int)reqids[i]));
982
983 req.hdr.length = offsetof(struct ctdb_req_control, data);
984 req.hdr.length += sizeof(pid);
985 req.hdr.ctdb_magic = CTDB_MAGIC;
986 req.hdr.ctdb_version = CTDB_VERSION;
987 req.hdr.operation = CTDB_REQ_CONTROL;
988 req.hdr.reqid = reqids[i];
989 req.hdr.destnode = pids[i].vnn;
990 req.opcode = CTDB_CONTROL_PROCESS_EXISTS;
991 req.srvid = 0;
992 req.datalen = sizeof(pid);
993 req.flags = 0;
994
995 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
996 ctdb_packet_dump(&req.hdr);
997
998 status = ctdb_packet_send(
999 conn->pkt, 2,
1000 data_blob_const(
1001 &req, offsetof(struct ctdb_req_control, data)),
1002 data_blob_const(&pid, sizeof(pid)));
1003 if (!NT_STATUS_IS_OK(status)) {
1004 DEBUG(10, ("ctdb_packet_send failed: %s\n",
1005 nt_errstr(status)));
1006 goto fail;
1007 }
1008 }
1009
1010 status = ctdb_packet_flush(conn->pkt);
1011 if (!NT_STATUS_IS_OK(status)) {
1012 DEBUG(10, ("ctdb_packet_flush failed: %s\n",
1013 nt_errstr(status)));
1014 goto fail;
1015 }
1016
1017 num_received = 0;
1018
1019 while (num_received < num_pids) {
1020 struct ctdb_reply_control *reply = NULL;
1021 uint32_t reqid;
1022
1023 status = ctdb_read_req(conn, 0, talloc_tos(), (void *)&reply);
1024 if (!NT_STATUS_IS_OK(status)) {
1025 DEBUG(10, ("ctdb_read_req failed: %s\n",
1026 nt_errstr(status)));
1027 goto fail;
1028 }
1029
1030 if (reply->hdr.operation != CTDB_REPLY_CONTROL) {
1031 DEBUG(10, ("Received invalid reply\n"));
1032 goto fail;
1033 }
1034
1035 reqid = reply->hdr.reqid;
1036
1037 DEBUG(10, ("Received reqid %d\n", (int)reqid));
1038
1039 for (i=0; i<num_pids; i++) {
1040 if (reqid == reqids[i]) {
1041 break;
1042 }
1043 }
1044 if (i == num_pids) {
1045 DEBUG(10, ("Received unknown record number %u\n",
1046 (unsigned)reqid));
1047 goto fail;
1048 }
1049 results[i] = ((reply->status) == 0);
1050 TALLOC_FREE(reply);
1051 num_received += 1;
1052 }
1053
1054 result = true;
1055 fail:
1056 TALLOC_FREE(frame);
1057 return result;
1058 }
1059
1060 struct ctdb_vnn_list {
1061 uint32_t vnn;
1062 uint32_t reqid;
1063 unsigned num_srvids;
1064 unsigned num_filled;
1065 uint64_t *srvids;
1066 unsigned *pid_indexes;
1067 };
1068
1069 /*
1070 * Get a list of all vnns mentioned in a list of
1071 * server_ids. vnn_indexes tells where in the vnns array we have to
1072 * place the pids.
1073 */
1074 static bool ctdb_collect_vnns(TALLOC_CTX *mem_ctx,
1075 const struct server_id *pids, unsigned num_pids,
1076 struct ctdb_vnn_list **pvnns,
1077 unsigned *pnum_vnns)
1078 {
1079 struct ctdb_vnn_list *vnns = NULL;
1080 unsigned *vnn_indexes = NULL;
1081 unsigned i, num_vnns = 0;
1082
1083 vnn_indexes = talloc_array(mem_ctx, unsigned, num_pids);
1084 if (vnn_indexes == NULL) {
1085 DEBUG(1, ("talloc_array failed\n"));
1086 goto fail;
1087 }
1088
1089 for (i=0; i<num_pids; i++) {
1090 unsigned j;
1091 uint32_t vnn = pids[i].vnn;
1092
1093 for (j=0; j<num_vnns; j++) {
1094 if (vnn == vnns[j].vnn) {
1095 break;
1096 }
1097 }
1098 vnn_indexes[i] = j;
1099
1100 if (j < num_vnns) {
1101 /*
1102 * Already in the array
1103 */
1104 vnns[j].num_srvids += 1;
1105 continue;
1106 }
1107 vnns = talloc_realloc(mem_ctx, vnns, struct ctdb_vnn_list,
1108 num_vnns+1);
1109 if (vnns == NULL) {
1110 DEBUG(1, ("talloc_realloc failed\n"));
1111 goto fail;
1112 }
1113 vnns[num_vnns].vnn = vnn;
1114 vnns[num_vnns].num_srvids = 1;
1115 vnns[num_vnns].num_filled = 0;
1116 num_vnns += 1;
1117 }
1118 for (i=0; i<num_vnns; i++) {
1119 struct ctdb_vnn_list *vnn = &vnns[i];
1120
1121 vnn->srvids = talloc_array(vnns, uint64_t, vnn->num_srvids);
1122 if (vnn->srvids == NULL) {
1123 DEBUG(1, ("talloc_array failed\n"));
1124 goto fail;
1125 }
1126 vnn->pid_indexes = talloc_array(vnns, unsigned,
1127 vnn->num_srvids);
1128 if (vnn->pid_indexes == NULL) {
1129 DEBUG(1, ("talloc_array failed\n"));
1130 goto fail;
1131 }
1132 }
1133 for (i=0; i<num_pids; i++) {
1134 struct ctdb_vnn_list *vnn = &vnns[vnn_indexes[i]];
1135 vnn->srvids[vnn->num_filled] = pids[i].unique_id;
1136 vnn->pid_indexes[vnn->num_filled] = i;
1137 vnn->num_filled += 1;
1138 }
1139
1140 TALLOC_FREE(vnn_indexes);
1141 *pvnns = vnns;
1142 *pnum_vnns = num_vnns;
1143 return true;
1144 fail:
1145 TALLOC_FREE(vnns);
1146 TALLOC_FREE(vnn_indexes);
1147 return false;
1148 }
1149
1150 bool ctdb_serverids_exist_supported(struct ctdbd_connection *conn)
1151 {
1152 #ifndef HAVE_CTDB_CONTROL_CHECK_SRVIDS_DECL
1153 return false;
1154 #else /* HAVE_CTDB_CONTROL_CHECK_SRVIDS_DECL */
1155 return true;
1156 #endif /* HAVE_CTDB_CONTROL_CHECK_SRVIDS_DECL */
1157 }
1158
1159 bool ctdb_serverids_exist(struct ctdbd_connection *conn,
1160 const struct server_id *pids, unsigned num_pids,
1161 bool *results)
1162 {
1163 #ifndef HAVE_CTDB_CONTROL_CHECK_SRVIDS_DECL
1164 return false;
1165 #else /* HAVE_CTDB_CONTROL_CHECK_SRVIDS_DECL */
1166 unsigned i, num_received;
1167 NTSTATUS status;
1168 struct ctdb_vnn_list *vnns = NULL;
1169 unsigned num_vnns;
1170
1171 if (!ctdb_collect_vnns(talloc_tos(), pids, num_pids,
1172 &vnns, &num_vnns)) {
1173 DEBUG(1, ("ctdb_collect_vnns failed\n"));
1174 goto fail;
1175 }
1176
1177 for (i=0; i<num_vnns; i++) {
1178 struct ctdb_vnn_list *vnn = &vnns[i];
1179 struct ctdb_req_control req;
1180
1181 vnn->reqid = ctdbd_next_reqid(conn);
1182
1183 ZERO_STRUCT(req);
1184
1185 DEBUG(10, ("Requesting VNN %d, reqid=%d, num_srvids=%u\n",
1186 (int)vnn->vnn, (int)vnn->reqid, vnn->num_srvids));
1187
1188 req.hdr.length = offsetof(struct ctdb_req_control, data);
1189 req.hdr.ctdb_magic = CTDB_MAGIC;
1190 req.hdr.ctdb_version = CTDB_VERSION;
1191 req.hdr.operation = CTDB_REQ_CONTROL;
1192 req.hdr.reqid = vnn->reqid;
1193 req.hdr.destnode = vnn->vnn;
1194 req.opcode = CTDB_CONTROL_CHECK_SRVIDS;
1195 req.srvid = 0;
1196 req.datalen = sizeof(uint64_t) * vnn->num_srvids;
1197 req.hdr.length += req.datalen;
1198 req.flags = 0;
1199
1200 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
1201 ctdb_packet_dump(&req.hdr);
1202
1203 status = ctdb_packet_send(
1204 conn->pkt, 2,
1205 data_blob_const(
1206 &req, offsetof(struct ctdb_req_control,
1207 data)),
1208 data_blob_const(vnn->srvids, req.datalen));
1209 if (!NT_STATUS_IS_OK(status)) {
1210 DEBUG(1, ("ctdb_packet_send failed: %s\n",
1211 nt_errstr(status)));
1212 goto fail;
1213 }
1214 }
1215
1216 status = ctdb_packet_flush(conn->pkt);
1217 if (!NT_STATUS_IS_OK(status)) {
1218 DEBUG(1, ("ctdb_packet_flush failed: %s\n",
1219 nt_errstr(status)));
1220 goto fail;
1221 }
1222
1223 num_received = 0;
1224
1225 while (num_received < num_vnns) {
1226 struct ctdb_reply_control *reply = NULL;
1227 struct ctdb_vnn_list *vnn;
1228 uint32_t reqid;
1229 uint8_t *reply_data;
1230
1231 status = ctdb_read_req(conn, 0, talloc_tos(), (void *)&reply);
1232 if (!NT_STATUS_IS_OK(status)) {
1233 DEBUG(1, ("ctdb_read_req failed: %s\n",
1234 nt_errstr(status)));
1235 goto fail;
1236 }
1237
1238 if (reply->hdr.operation != CTDB_REPLY_CONTROL) {
1239 DEBUG(1, ("Received invalid reply %u\n",
1240 (unsigned)reply->hdr.operation));
1241 goto fail;
1242 }
1243
1244 reqid = reply->hdr.reqid;
1245
1246 DEBUG(10, ("Received reqid %d\n", (int)reqid));
1247
1248 for (i=0; i<num_vnns; i++) {
1249 if (reqid == vnns[i].reqid) {
1250 break;
1251 }
1252 }
1253 if (i == num_vnns) {
1254 DEBUG(1, ("Received unknown reqid number %u\n",
1255 (unsigned)reqid));
1256 goto fail;
1257 }
1258
1259 DEBUG(10, ("Found index %u\n", i));
1260
1261 vnn = &vnns[i];
1262
1263 DEBUG(10, ("Received vnn %u, vnn->num_srvids %u, datalen %u\n",
1264 (unsigned)vnn->vnn, vnn->num_srvids,
1265 (unsigned)reply->datalen));
1266
1267 if (reply->datalen >= ((vnn->num_srvids+7)/8)) {
1268 /*
1269 * Got a real reply
1270 */
1271 reply_data = reply->data;
1272 } else {
1273 /*
1274 * Got an error reply
1275 */
1276 DEBUG(5, ("Received short reply len %d, status %u, "
1277 "errorlen %u\n",
1278 (unsigned)reply->datalen,
1279 (unsigned)reply->status,
1280 (unsigned)reply->errorlen));
1281 dump_data(5, reply->data, reply->errorlen);
1282
1283 /*
1284 * This will trigger everything set to false
1285 */
1286 reply_data = NULL;
1287 }
1288
1289 for (i=0; i<vnn->num_srvids; i++) {
1290 int idx = vnn->pid_indexes[i];
1291
1292 if (pids[i].unique_id ==
1293 SERVERID_UNIQUE_ID_NOT_TO_VERIFY) {
1294 results[idx] = true;
1295 continue;
1296 }
1297 results[idx] =
1298 (reply_data != NULL) &&
1299 ((reply_data[i/8] & (1<<(i%8))) != 0);
1300 }
1301
1302 TALLOC_FREE(reply);
1303 num_received += 1;
1304 }
1305
1306 TALLOC_FREE(vnns);
1307 return true;
1308 fail:
1309 cluster_fatal("serverids_exist failed");
1310 return false;
1311 #endif /* HAVE_CTDB_CONTROL_CHECK_SRVIDS_DECL */
1312 }
1313
1314 /*
1315 * Get a db path
1316 */
1317 char *ctdbd_dbpath(struct ctdbd_connection *conn,
1318 TALLOC_CTX *mem_ctx, uint32_t db_id)
1319 {
1320 NTSTATUS status;
1321 TDB_DATA data;
1322 int32_t cstatus;
1323
1324 data.dptr = (uint8_t*)&db_id;
1325 data.dsize = sizeof(db_id);
1326
1327 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1328 CTDB_CONTROL_GETDBPATH, 0, 0, data,
1329 mem_ctx, &data, &cstatus);
1330 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
1331 DEBUG(0,(__location__ " ctdb_control for getdbpath failed\n"));
1332 return NULL;
1333 }
1334
1335 return (char *)data.dptr;
1336 }
1337
1338 /*
1339 * attach to a ctdb database
1340 */
1341 NTSTATUS ctdbd_db_attach(struct ctdbd_connection *conn,
1342 const char *name, uint32_t *db_id, int tdb_flags)
1343 {
1344 NTSTATUS status;
1345 TDB_DATA data;
1346 int32_t cstatus;
1347 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
1348
1349 data = string_term_tdb_data(name);
1350
1351 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1352 persistent
1353 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
1354 : CTDB_CONTROL_DB_ATTACH,
1355 tdb_flags, 0, data, NULL, &data, &cstatus);
1356 if (!NT_STATUS_IS_OK(status)) {
1357 DEBUG(0, (__location__ " ctdb_control for db_attach "
1358 "failed: %s\n", nt_errstr(status)));
1359 return status;
1360 }
1361
1362 if (cstatus != 0 || data.dsize != sizeof(uint32_t)) {
1363 DEBUG(0,(__location__ " ctdb_control for db_attach failed\n"));
1364 return NT_STATUS_INTERNAL_ERROR;
1365 }
1366
1367 *db_id = *(uint32_t *)data.dptr;
1368 talloc_free(data.dptr);
1369
1370 if (!(tdb_flags & TDB_SEQNUM)) {
1371 return NT_STATUS_OK;
1372 }
1373
1374 data.dptr = (uint8_t *)db_id;
1375 data.dsize = sizeof(*db_id);
1376
1377 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1378 CTDB_CONTROL_ENABLE_SEQNUM, 0, 0, data,
1379 NULL, NULL, &cstatus);
1380 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
1381 DEBUG(0,(__location__ " ctdb_control for enable seqnum "
1382 "failed\n"));
1383 return NT_STATUS_IS_OK(status) ? NT_STATUS_INTERNAL_ERROR :
1384 status;
1385 }
1386
1387 return NT_STATUS_OK;
1388 }
1389
1390 /*
1391 * force the migration of a record to this node
1392 */
1393 NTSTATUS ctdbd_migrate(struct ctdbd_connection *conn, uint32_t db_id,
1394 TDB_DATA key)
1395 {
1396 struct ctdb_req_call req;
1397 struct ctdb_reply_call *reply;
1398 NTSTATUS status;
1399
1400 ZERO_STRUCT(req);
1401
1402 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
1403 req.hdr.ctdb_magic = CTDB_MAGIC;
1404 req.hdr.ctdb_version = CTDB_VERSION;
1405 req.hdr.operation = CTDB_REQ_CALL;
1406 req.hdr.reqid = ctdbd_next_reqid(conn);
1407 req.flags = CTDB_IMMEDIATE_MIGRATION;
1408 req.callid = CTDB_NULL_FUNC;
1409 req.db_id = db_id;
1410 req.keylen = key.dsize;
1411
1412 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
1413 ctdb_packet_dump(&req.hdr);
1414
1415 status = ctdb_packet_send(
1416 conn->pkt, 2,
1417 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
1418 data_blob_const(key.dptr, key.dsize));
1419
1420 if (!NT_STATUS_IS_OK(status)) {
1421 DEBUG(3, ("ctdb_packet_send failed: %s\n", nt_errstr(status)));
1422 return status;
1423 }
1424
1425 status = ctdb_packet_flush(conn->pkt);
1426
1427 if (!NT_STATUS_IS_OK(status)) {
1428 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
1429 cluster_fatal("cluster dispatch daemon control write error\n");
1430 }
1431
1432 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
1433
1434 if (!NT_STATUS_IS_OK(status)) {
1435 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
1436 goto fail;
1437 }
1438
1439 if (reply->hdr.operation != CTDB_REPLY_CALL) {
1440 DEBUG(0, ("received invalid reply\n"));
1441 status = NT_STATUS_INTERNAL_ERROR;
1442 goto fail;
1443 }
1444
1445 status = NT_STATUS_OK;
1446 fail:
1447
1448 TALLOC_FREE(reply);
1449 return status;
1450 }
1451
1452 /*
1453 * Fetch a record and parse it
1454 */
1455 NTSTATUS ctdbd_parse(struct ctdbd_connection *conn, uint32_t db_id,
1456 TDB_DATA key, bool local_copy,
1457 void (*parser)(TDB_DATA key, TDB_DATA data,
1458 void *private_data),
1459 void *private_data)
1460 {
1461 struct ctdb_req_call req;
1462 struct ctdb_reply_call *reply;
1463 NTSTATUS status;
1464 uint32_t flags;
1465
1466 #ifdef HAVE_CTDB_WANT_READONLY_DECL
1467 flags = local_copy ? CTDB_WANT_READONLY : 0;
1468 #else
1469 flags = 0;
1470 #endif
1471
1472 ZERO_STRUCT(req);
1473
1474 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
1475 req.hdr.ctdb_magic = CTDB_MAGIC;
1476 req.hdr.ctdb_version = CTDB_VERSION;
1477 req.hdr.operation = CTDB_REQ_CALL;
1478 req.hdr.reqid = ctdbd_next_reqid(conn);
1479 req.flags = flags;
1480 req.callid = CTDB_FETCH_FUNC;
1481 req.db_id = db_id;
1482 req.keylen = key.dsize;
1483
1484 status = ctdb_packet_send(
1485 conn->pkt, 2,
1486 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
1487 data_blob_const(key.dptr, key.dsize));
1488
1489 if (!NT_STATUS_IS_OK(status)) {
1490 DEBUG(3, ("ctdb_packet_send failed: %s\n", nt_errstr(status)));
1491 return status;
1492 }
1493
1494 status = ctdb_packet_flush(conn->pkt);
1495
1496 if (!NT_STATUS_IS_OK(status)) {
1497 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
1498 cluster_fatal("cluster dispatch daemon control write error\n");
1499 }
1500
1501 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
1502
1503 if (!NT_STATUS_IS_OK(status)) {
1504 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
1505 goto fail;
1506 }
1507
1508 if (reply->hdr.operation != CTDB_REPLY_CALL) {
1509 DEBUG(0, ("received invalid reply\n"));
1510 status = NT_STATUS_INTERNAL_ERROR;
1511 goto fail;
1512 }
1513
1514 if (reply->datalen == 0) {
1515 /*
1516 * Treat an empty record as non-existing
1517 */
1518 status = NT_STATUS_NOT_FOUND;
1519 goto fail;
1520 }
1521
1522 parser(key, make_tdb_data(&reply->data[0], reply->datalen),
1523 private_data);
1524
1525 status = NT_STATUS_OK;
1526 fail:
1527 TALLOC_FREE(reply);
1528 return status;
1529 }
1530
1531 struct ctdbd_traverse_state {
1532 void (*fn)(TDB_DATA key, TDB_DATA data, void *private_data);
1533 void *private_data;
1534 };
1535
1536 /*
1537 * Handle a traverse record coming in on the ctdbd connection
1538 */
1539
1540 static NTSTATUS ctdb_traverse_handler(uint8_t *buf, size_t length,
1541 void *private_data)
1542 {
1543 struct ctdbd_traverse_state *state =
1544 (struct ctdbd_traverse_state *)private_data;
1545
1546 struct ctdb_req_message *m;
1547 struct ctdb_rec_data *d;
1548 TDB_DATA key, data;
1549
1550 m = (struct ctdb_req_message *)buf;
1551
1552 if (length < sizeof(*m) || m->hdr.length != length) {
1553 DEBUG(0, ("Got invalid message of length %d\n", (int)length));
1554 TALLOC_FREE(buf);
1555 return NT_STATUS_UNEXPECTED_IO_ERROR;
1556 }
1557
1558 d = (struct ctdb_rec_data *)&m->data[0];
1559 if (m->datalen < sizeof(uint32_t) || m->datalen != d->length) {
1560 DEBUG(0, ("Got invalid traverse data of length %d\n",
1561 (int)m->datalen));
1562 TALLOC_FREE(buf);
1563 return NT_STATUS_UNEXPECTED_IO_ERROR;
1564 }
1565
1566 key.dsize = d->keylen;
1567 key.dptr = &d->data[0];
1568 data.dsize = d->datalen;
1569 data.dptr = &d->data[d->keylen];
1570
1571 if (key.dsize == 0 && data.dsize == 0) {
1572 /* end of traverse */
1573 return NT_STATUS_END_OF_FILE;
1574 }
1575
1576 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1577 DEBUG(0, ("Got invalid ltdb header length %d\n",
1578 (int)data.dsize));
1579 TALLOC_FREE(buf);
1580 return NT_STATUS_UNEXPECTED_IO_ERROR;
1581 }
1582 data.dsize -= sizeof(struct ctdb_ltdb_header);
1583 data.dptr += sizeof(struct ctdb_ltdb_header);
1584
1585 if (state->fn) {
1586 state->fn(key, data, state->private_data);
1587 }
1588
1589 TALLOC_FREE(buf);
1590 return NT_STATUS_OK;
1591 }
1592
1593 /*
1594 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1595 connection to ctdbd to avoid the hairy recursive and async problems with
1596 everything in-line.
1597 */
1598
1599 NTSTATUS ctdbd_traverse(uint32_t db_id,
1600 void (*fn)(TDB_DATA key, TDB_DATA data,
1601 void *private_data),
1602 void *private_data)
1603 {
1604 struct ctdbd_connection *conn;
1605 NTSTATUS status;
1606
1607 TDB_DATA data;
1608 struct ctdb_traverse_start t;
1609 int cstatus;
1610 struct ctdbd_traverse_state state;
1611
1612 become_root();
1613 status = ctdbd_init_connection(NULL, &conn);
1614 unbecome_root();
1615 if (!NT_STATUS_IS_OK(status)) {
1616 DEBUG(0, ("ctdbd_init_connection failed: %s\n",
1617 nt_errstr(status)));
1618 return status;
1619 }
1620
1621 t.db_id = db_id;
1622 t.srvid = conn->rand_srvid;
1623 t.reqid = ctdbd_next_reqid(conn);
1624
1625 data.dptr = (uint8_t *)&t;
1626 data.dsize = sizeof(t);
1627
1628 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1629 CTDB_CONTROL_TRAVERSE_START, conn->rand_srvid, 0,
1630 data, NULL, NULL, &cstatus);
1631
1632 if (!NT_STATUS_IS_OK(status) || (cstatus != 0)) {
1633
1634 DEBUG(0,("ctdbd_control failed: %s, %d\n", nt_errstr(status),
1635 cstatus));
1636
1637 if (NT_STATUS_IS_OK(status)) {
1638 /*
1639 * We need a mapping here
1640 */
1641 status = NT_STATUS_UNSUCCESSFUL;
1642 }
1643 goto done;
1644 }
1645
1646 state.fn = fn;
1647 state.private_data = private_data;
1648
1649 while (True) {
1650
1651 status = NT_STATUS_OK;
1652
1653 if (ctdb_packet_handler(conn->pkt, ctdb_req_complete,
1654 ctdb_traverse_handler, &state, &status)) {
1655
1656 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1657 status = NT_STATUS_OK;
1658 break;
1659 }
1660
1661 /*
1662 * There might be more in the queue
1663 */
1664 continue;
1665 }
1666
1667 if (!NT_STATUS_IS_OK(status)) {
1668 break;
1669 }
1670
1671 status = ctdb_packet_fd_read_sync(conn->pkt);
1672
1673 if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
1674 /*
1675 * There might be more in the queue
1676 */
1677 continue;
1678 }
1679
1680 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1681 status = NT_STATUS_OK;
1682 break;
1683 }
1684
1685 if (!NT_STATUS_IS_OK(status)) {
1686 DEBUG(0, ("ctdb_packet_fd_read_sync failed: %s\n", nt_errstr(status)));
1687 cluster_fatal("ctdbd died\n");
1688 }
1689 }
1690
1691 done:
1692 TALLOC_FREE(conn);
1693 return status;
1694 }
1695
1696 /*
1697 This is used to canonicalize a ctdb_sock_addr structure.
1698 */
1699 static void smbd_ctdb_canonicalize_ip(const struct sockaddr_storage *in,
1700 struct sockaddr_storage *out)
1701 {
1702 memcpy(out, in, sizeof (*out));
1703
1704 #ifdef HAVE_IPV6
1705 if (in->ss_family == AF_INET6) {
1706 const char prefix[12] = { 0,0,0,0,0,0,0,0,0,0,0xff,0xff };
1707 const struct sockaddr_in6 *in6 =
1708 (const struct sockaddr_in6 *)in;
1709 struct sockaddr_in *out4 = (struct sockaddr_in *)out;
1710 if (memcmp(&in6->sin6_addr, prefix, 12) == 0) {
1711 memset(out, 0, sizeof(*out));
1712 #ifdef HAVE_SOCK_SIN_LEN
1713 out4->sin_len = sizeof(*out);
1714 #endif
1715 out4->sin_family = AF_INET;
1716 out4->sin_port = in6->sin6_port;
1717 memcpy(&out4->sin_addr, &in6->sin6_addr.s6_addr[12], 4);
1718 }
1719 }
1720 #endif
1721 }
1722
1723 /*
1724 * Register us as a server for a particular tcp connection
1725 */
1726
1727 NTSTATUS ctdbd_register_ips(struct ctdbd_connection *conn,
1728 const struct sockaddr_storage *_server,
1729 const struct sockaddr_storage *_client,
1730 bool (*release_ip_handler)(const char *ip_addr,
1731 void *private_data),
1732 void *private_data)
1733 {
1734 /*
1735 * we still use ctdb_control_tcp for ipv4
1736 * because we want to work against older ctdb
1737 * versions at runtime
1738 */
1739 struct ctdb_control_tcp p4;
1740 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1741 struct ctdb_control_tcp_addr p;
1742 #endif
1743 TDB_DATA data;
1744 NTSTATUS status;
1745 struct sockaddr_storage client;
1746 struct sockaddr_storage server;
1747
1748 /*
1749 * Only one connection so far
1750 */
1751 SMB_ASSERT(conn->release_ip_handler == NULL);
1752
1753 smbd_ctdb_canonicalize_ip(_client, &client);
1754 smbd_ctdb_canonicalize_ip(_server, &server);
1755
1756 switch (client.ss_family) {
1757 case AF_INET:
1758 memcpy(&p4.dest, &server, sizeof(p4.dest));
1759 memcpy(&p4.src, &client, sizeof(p4.src));
1760 data.dptr = (uint8_t *)&p4;
1761 data.dsize = sizeof(p4);
1762 break;
1763 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1764 case AF_INET6:
1765 memcpy(&p.dest.ip6, &server, sizeof(p.dest.ip6));
1766 memcpy(&p.src.ip6, &client, sizeof(p.src.ip6));
1767 data.dptr = (uint8_t *)&p;
1768 data.dsize = sizeof(p);
1769 break;
1770 #endif
1771 default:
1772 return NT_STATUS_INTERNAL_ERROR;
1773 }
1774
1775 conn->release_ip_handler = release_ip_handler;
1776 conn->release_ip_priv = private_data;
1777
1778 /*
1779 * We want to be told about IP releases
1780 */
1781
1782 status = register_with_ctdbd(conn, CTDB_SRVID_RELEASE_IP);
1783 if (!NT_STATUS_IS_OK(status)) {
1784 return status;
1785 }
1786
1787 /*
1788 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1789 * can send an extra ack to trigger a reset for our client, so it
1790 * immediately reconnects
1791 */
1792 return ctdbd_control(conn, CTDB_CURRENT_NODE,
1793 CTDB_CONTROL_TCP_CLIENT, 0,
1794 CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL, NULL);
1795 }
1796
1797 /*
1798 * We want to handle reconfigure events
1799 */
1800 NTSTATUS ctdbd_register_reconfigure(struct ctdbd_connection *conn)
1801 {
1802 return register_with_ctdbd(conn, CTDB_SRVID_RECONFIGURE);
1803 }
1804
1805 /*
1806 call a control on the local node
1807 */
1808 NTSTATUS ctdbd_control_local(struct ctdbd_connection *conn, uint32_t opcode,
1809 uint64_t srvid, uint32_t flags, TDB_DATA data,
1810 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
1811 int *cstatus)
1812 {
1813 return ctdbd_control(conn, CTDB_CURRENT_NODE, opcode, srvid, flags, data, mem_ctx, outdata, cstatus);
1814 }
1815
1816 NTSTATUS ctdb_watch_us(struct ctdbd_connection *conn)
1817 {
1818 struct ctdb_client_notify_register reg_data;
1819 size_t struct_len;
1820 NTSTATUS status;
1821 int cstatus;
1822
1823 reg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1824 reg_data.len = 1;
1825 reg_data.notify_data[0] = 0;
1826
1827 struct_len = offsetof(struct ctdb_client_notify_register,
1828 notify_data) + reg_data.len;
1829
1830 status = ctdbd_control_local(
1831 conn, CTDB_CONTROL_REGISTER_NOTIFY, conn->rand_srvid, 0,
1832 make_tdb_data((uint8_t *)&reg_data, struct_len),
1833 NULL, NULL, &cstatus);
1834 if (!NT_STATUS_IS_OK(status)) {
1835 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1836 nt_errstr(status)));
1837 }
1838 return status;
1839 }
1840
1841 NTSTATUS ctdb_unwatch(struct ctdbd_connection *conn)
1842 {
1843 struct ctdb_client_notify_deregister dereg_data;
1844 NTSTATUS status;
1845 int cstatus;
1846
1847 dereg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1848
1849 status = ctdbd_control_local(
1850 conn, CTDB_CONTROL_DEREGISTER_NOTIFY, conn->rand_srvid, 0,
1851 make_tdb_data((uint8_t *)&dereg_data, sizeof(dereg_data)),
1852 NULL, NULL, &cstatus);
1853 if (!NT_STATUS_IS_OK(status)) {
1854 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1855 nt_errstr(status)));
1856 }
1857 return status;
1858 }
1859
1860 NTSTATUS ctdbd_probe(void)
1861 {
1862 /*
1863 * Do a very early check if ctdbd is around to avoid an abort and core
1864 * later
1865 */
1866 struct ctdbd_connection *conn = NULL;
1867 NTSTATUS status;
1868
1869 status = ctdbd_messaging_connection(talloc_tos(), &conn);
1870
1871 /*
1872 * We only care if we can connect.
1873 */
1874 TALLOC_FREE(conn);
1875
1876 return status;
1877 }
Anubhav's Work In Progress branches