?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
lib: Make ctdbd_register_msg_ctx return 0/errno
[Samba.git] / source3 / lib / ctdbd_conn.c
blobdeb3e4e3b469e122dc186dc49af4ecebe8cf3142
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 #include "system/select.h"
26 #include "lib/sys_rw_data.h"
27 #include "lib/util/iov_buf.h"
28
29 #include "messages.h"
30
31 /* paths to these include files come from --with-ctdb= in configure */
32
33 #include "ctdb.h"
34 #include "ctdb_private.h"
35
36 struct ctdbd_srvid_cb {
37 uint64_t srvid;
38 int (*cb)(uint32_t src_vnn, uint32_t dst_vnn,
39 uint64_t dst_srvid,
40 const uint8_t *msg, size_t msglen,
41 void *private_data);
42 void *private_data;
43 };
44
45 struct ctdbd_connection {
46 const char *sockname; /* Needed in ctdbd_traverse */
47 struct messaging_context *msg_ctx;
48 uint32_t reqid;
49 uint32_t our_vnn;
50 uint64_t rand_srvid;
51 struct ctdbd_srvid_cb *callbacks;
52 int fd;
53 struct tevent_fd *fde;
54 int timeout;
55 };
56
57 static uint32_t ctdbd_next_reqid(struct ctdbd_connection *conn)
58 {
59 conn->reqid += 1;
60 if (conn->reqid == 0) {
61 conn->reqid += 1;
62 }
63 return conn->reqid;
64 }
65
66 static int ctdbd_control(struct ctdbd_connection *conn,
67 uint32_t vnn, uint32_t opcode,
68 uint64_t srvid, uint32_t flags,
69 TDB_DATA data,
70 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
71 int *cstatus);
72
73 /*
74 * exit on fatal communications errors with the ctdbd daemon
75 */
76 static void cluster_fatal(const char *why)
77 {
78 DEBUG(0,("cluster fatal event: %s - exiting immediately\n", why));
79 /* we don't use smb_panic() as we don't want to delay to write
80 a core file. We need to release this process id immediately
81 so that someone else can take over without getting sharing
82 violations */
83 _exit(1);
84 }
85
86 /*
87 *
88 */
89 static void ctdb_packet_dump(struct ctdb_req_header *hdr)
90 {
91 if (DEBUGLEVEL < 11) {
92 return;
93 }
94 DEBUGADD(11, ("len=%d, magic=%x, vers=%d, gen=%d, op=%d, reqid=%d\n",
95 (int)hdr->length, (int)hdr->ctdb_magic,
96 (int)hdr->ctdb_version, (int)hdr->generation,
97 (int)hdr->operation, (int)hdr->reqid));
98 }
99
100 /*
101 * Register a srvid with ctdbd
102 */
103 int register_with_ctdbd(struct ctdbd_connection *conn, uint64_t srvid,
104 int (*cb)(uint32_t src_vnn, uint32_t dst_vnn,
105 uint64_t dst_srvid,
106 const uint8_t *msg, size_t msglen,
107 void *private_data),
108 void *private_data)
109 {
110
111 int ret, cstatus;
112 size_t num_callbacks;
113 struct ctdbd_srvid_cb *tmp;
114
115 ret = ctdbd_control(conn, CTDB_CURRENT_NODE,
116 CTDB_CONTROL_REGISTER_SRVID, srvid, 0,
117 tdb_null, NULL, NULL, &cstatus);
118 if (ret != 0) {
119 return ret;
120 }
121
122 num_callbacks = talloc_array_length(conn->callbacks);
123
124 tmp = talloc_realloc(conn, conn->callbacks, struct ctdbd_srvid_cb,
125 num_callbacks + 1);
126 if (tmp == NULL) {
127 return ENOMEM;
128 }
129 conn->callbacks = tmp;
130
131 conn->callbacks[num_callbacks] = (struct ctdbd_srvid_cb) {
132 .srvid = srvid, .cb = cb, .private_data = private_data
133 };
134
135 return 0;
136 }
137
138 static int ctdbd_msg_call_back(struct ctdbd_connection *conn,
139 struct ctdb_req_message *msg)
140 {
141 size_t msg_len;
142 size_t i, num_callbacks;
143
144 msg_len = msg->hdr.length;
145 if (msg_len < offsetof(struct ctdb_req_message, data)) {
146 DEBUG(10, ("%s: len %u too small\n", __func__,
147 (unsigned)msg_len));
148 return 0;
149 }
150 msg_len -= offsetof(struct ctdb_req_message, data);
151
152 if (msg_len < msg->datalen) {
153 DEBUG(10, ("%s: msg_len=%u < msg->datalen=%u\n", __func__,
154 (unsigned)msg_len, (unsigned)msg->datalen));
155 return 0;
156 }
157
158 num_callbacks = talloc_array_length(conn->callbacks);
159
160 for (i=0; i<num_callbacks; i++) {
161 struct ctdbd_srvid_cb *cb = &conn->callbacks[i];
162
163 if ((cb->srvid == msg->srvid) && (cb->cb != NULL)) {
164 int ret;
165
166 ret = cb->cb(msg->hdr.srcnode, msg->hdr.destnode,
167 msg->srvid, msg->data, msg->datalen,
168 cb->private_data);
169 if (ret != 0) {
170 return ret;
171 }
172 }
173 }
174 return 0;
175 }
176
177 /*
178 * get our vnn from the cluster
179 */
180 static int get_cluster_vnn(struct ctdbd_connection *conn, uint32_t *vnn)
181 {
182 int32_t cstatus=-1;
183 int ret;
184 ret = ctdbd_control(conn,
185 CTDB_CURRENT_NODE, CTDB_CONTROL_GET_PNN, 0, 0,
186 tdb_null, NULL, NULL, &cstatus);
187 if (ret != 0) {
188 DEBUG(1, ("ctdbd_control failed: %s\n", strerror(ret)));
189 return ret;
190 }
191 *vnn = (uint32_t)cstatus;
192 return ret;
193 }
194
195 /*
196 * Are we active (i.e. not banned or stopped?)
197 */
198 static bool ctdbd_working(struct ctdbd_connection *conn, uint32_t vnn)
199 {
200 int32_t cstatus=-1;
201 TDB_DATA outdata;
202 struct ctdb_node_map *m;
203 uint32_t failure_flags;
204 bool ok = false;
205 int i, ret;
206
207 ret = ctdbd_control(conn, CTDB_CURRENT_NODE,
208 CTDB_CONTROL_GET_NODEMAP, 0, 0,
209 tdb_null, talloc_tos(), &outdata, &cstatus);
210 if (ret != 0) {
211 DEBUG(1, ("ctdbd_control failed: %s\n", strerror(ret)));
212 return false;
213 }
214 if ((cstatus != 0) || (outdata.dptr == NULL)) {
215 DEBUG(2, ("Received invalid ctdb data\n"));
216 return false;
217 }
218
219 m = (struct ctdb_node_map *)outdata.dptr;
220
221 for (i=0; i<m->num; i++) {
222 if (vnn == m->nodes[i].pnn) {
223 break;
224 }
225 }
226
227 if (i == m->num) {
228 DEBUG(2, ("Did not find ourselves (node %d) in nodemap\n",
229 (int)vnn));
230 goto fail;
231 }
232
233 failure_flags = NODE_FLAGS_BANNED | NODE_FLAGS_DISCONNECTED
234 | NODE_FLAGS_PERMANENTLY_DISABLED | NODE_FLAGS_STOPPED;
235
236 if ((m->nodes[i].flags & failure_flags) != 0) {
237 DEBUG(2, ("Node has status %x, not active\n",
238 (int)m->nodes[i].flags));
239 goto fail;
240 }
241
242 ok = true;
243 fail:
244 TALLOC_FREE(outdata.dptr);
245 return ok;
246 }
247
248 uint32_t ctdbd_vnn(const struct ctdbd_connection *conn)
249 {
250 return conn->our_vnn;
251 }
252
253 /*
254 * Get us a ctdb connection
255 */
256
257 static int ctdbd_connect(const char *sockname, int *pfd)
258 {
259 struct sockaddr_un addr = { 0, };
260 int fd;
261 socklen_t salen;
262 size_t namelen;
263
264 fd = socket(AF_UNIX, SOCK_STREAM, 0);
265 if (fd == -1) {
266 int err = errno;
267 DEBUG(3, ("Could not create socket: %s\n", strerror(err)));
268 return err;
269 }
270
271 addr.sun_family = AF_UNIX;
272
273 namelen = strlcpy(addr.sun_path, sockname, sizeof(addr.sun_path));
274 if (namelen >= sizeof(addr.sun_path)) {
275 DEBUG(3, ("%s: Socket name too long: %s\n", __func__,
276 sockname));
277 close(fd);
278 return ENAMETOOLONG;
279 }
280
281 salen = sizeof(struct sockaddr_un);
282
283 if (connect(fd, (struct sockaddr *)(void *)&addr, salen) == -1) {
284 int err = errno;
285 DEBUG(1, ("connect(%s) failed: %s\n", sockname,
286 strerror(err)));
287 close(fd);
288 return err;
289 }
290
291 *pfd = fd;
292 return 0;
293 }
294
295 static int ctdb_read_packet(int fd, int timeout, TALLOC_CTX *mem_ctx,
296 struct ctdb_req_header **result)
297 {
298 struct ctdb_req_header *req;
299 int ret, revents;
300 uint32_t msglen;
301 ssize_t nread;
302
303 if (timeout != -1) {
304 ret = poll_one_fd(fd, POLLIN, timeout, &revents);
305 if (ret == -1) {
306 return errno;
307 }
308 if (ret == 0) {
309 return ETIMEDOUT;
310 }
311 if (ret != 1) {
312 return EIO;
313 }
314 }
315
316 nread = read_data(fd, &msglen, sizeof(msglen));
317 if (nread == -1) {
318 return errno;
319 }
320 if (nread == 0) {
321 return EIO;
322 }
323
324 if (msglen < sizeof(struct ctdb_req_header)) {
325 return EIO;
326 }
327
328 req = talloc_size(mem_ctx, msglen);
329 if (req == NULL) {
330 return ENOMEM;
331 }
332 talloc_set_name_const(req, "struct ctdb_req_header");
333
334 req->length = msglen;
335
336 nread = read_data(fd, ((char *)req) + sizeof(msglen),
337 msglen - sizeof(msglen));
338 if (nread == -1) {
339 TALLOC_FREE(req);
340 return errno;
341 }
342 if (nread == 0) {
343 TALLOC_FREE(req);
344 return EIO;
345 }
346
347 *result = req;
348 return 0;
349 }
350
351 /*
352 * Read a full ctdbd request. If we have a messaging context, defer incoming
353 * messages that might come in between.
354 */
355
356 static int ctdb_read_req(struct ctdbd_connection *conn, uint32_t reqid,
357 TALLOC_CTX *mem_ctx, struct ctdb_req_header **result)
358 {
359 struct ctdb_req_header *hdr;
360 int ret;
361
362 next_pkt:
363
364 ret = ctdb_read_packet(conn->fd, conn->timeout, mem_ctx, &hdr);
365 if (ret != 0) {
366 DEBUG(0, ("ctdb_read_packet failed: %s\n", strerror(ret)));
367 cluster_fatal("ctdbd died\n");
368 }
369
370 DEBUG(11, ("Received ctdb packet\n"));
371 ctdb_packet_dump(hdr);
372
373 if (hdr->operation == CTDB_REQ_MESSAGE) {
374 struct ctdb_req_message *msg = (struct ctdb_req_message *)hdr;
375
376 if (conn->msg_ctx == NULL) {
377 DEBUG(1, ("Got a message without having a msg ctx, "
378 "dropping msg %llu\n",
379 (long long unsigned)msg->srvid));
380 TALLOC_FREE(hdr);
381 goto next_pkt;
382 }
383
384 ret = ctdbd_msg_call_back(conn, msg);
385 if (ret != 0) {
386 TALLOC_FREE(hdr);
387 return ret;
388 }
389
390 TALLOC_FREE(hdr);
391 goto next_pkt;
392 }
393
394 if ((reqid != 0) && (hdr->reqid != reqid)) {
395 /* we got the wrong reply */
396 DEBUG(0,("Discarding mismatched ctdb reqid %u should have "
397 "been %u\n", hdr->reqid, reqid));
398 TALLOC_FREE(hdr);
399 goto next_pkt;
400 }
401
402 *result = talloc_move(mem_ctx, &hdr);
403
404 return 0;
405 }
406
407 static int ctdbd_connection_destructor(struct ctdbd_connection *c)
408 {
409 TALLOC_FREE(c->fde);
410 if (c->fd != -1) {
411 close(c->fd);
412 c->fd = -1;
413 }
414 return 0;
415 }
416 /*
417 * Get us a ctdbd connection
418 */
419
420 static NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
421 const char *sockname, int timeout,
422 struct ctdbd_connection **pconn)
423 {
424 struct ctdbd_connection *conn;
425 int ret;
426 NTSTATUS status;
427
428 if (!(conn = talloc_zero(mem_ctx, struct ctdbd_connection))) {
429 DEBUG(0, ("talloc failed\n"));
430 return NT_STATUS_NO_MEMORY;
431 }
432
433 conn->sockname = talloc_strdup(conn, sockname);
434 if (conn->sockname == NULL) {
435 DBG_ERR("%s: talloc failed\n", __func__);
436 status = NT_STATUS_NO_MEMORY;
437 goto fail;
438 }
439
440 conn->timeout = timeout;
441
442 if (conn->timeout == 0) {
443 conn->timeout = -1;
444 }
445
446 ret = ctdbd_connect(conn->sockname, &conn->fd);
447 if (ret != 0) {
448 status = map_nt_error_from_unix(ret);
449 DEBUG(1, ("ctdbd_connect failed: %s\n", strerror(ret)));
450 goto fail;
451 }
452 talloc_set_destructor(conn, ctdbd_connection_destructor);
453
454 ret = get_cluster_vnn(conn, &conn->our_vnn);
455
456 if (ret != 0) {
457 DEBUG(10, ("get_cluster_vnn failed: %s\n", strerror(ret)));
458 status = map_nt_error_from_unix(ret);
459 goto fail;
460 }
461
462 if (!ctdbd_working(conn, conn->our_vnn)) {
463 DEBUG(2, ("Node is not working, can not connect\n"));
464 status = NT_STATUS_INTERNAL_DB_ERROR;
465 goto fail;
466 }
467
468 generate_random_buffer((unsigned char *)&conn->rand_srvid,
469 sizeof(conn->rand_srvid));
470
471 ret = register_with_ctdbd(conn, conn->rand_srvid, NULL, NULL);
472
473 if (ret != 0) {
474 DEBUG(5, ("Could not register random srvid: %s\n",
475 strerror(ret)));
476 status = map_nt_error_from_unix(ret);
477 goto fail;
478 }
479
480 *pconn = conn;
481 return NT_STATUS_OK;
482
483 fail:
484 TALLOC_FREE(conn);
485 return status;
486 }
487
488 /*
489 * Get us a ctdbd connection and register us as a process
490 */
491
492 NTSTATUS ctdbd_messaging_connection(TALLOC_CTX *mem_ctx,
493 const char *sockname, int timeout,
494 struct ctdbd_connection **pconn)
495 {
496 struct ctdbd_connection *conn;
497 NTSTATUS status;
498 int ret;
499
500 status = ctdbd_init_connection(mem_ctx, sockname, timeout, &conn);
501
502 if (!NT_STATUS_IS_OK(status)) {
503 return status;
504 }
505
506 ret = register_with_ctdbd(conn, MSG_SRVID_SAMBA, NULL, NULL);
507 if (ret != 0) {
508 status = map_nt_error_from_unix(ret);
509 goto fail;
510 }
511
512 *pconn = conn;
513 return NT_STATUS_OK;
514
515 fail:
516 TALLOC_FREE(conn);
517 return status;
518 }
519
520 struct messaging_context *ctdb_conn_msg_ctx(struct ctdbd_connection *conn)
521 {
522 return conn->msg_ctx;
523 }
524
525 int ctdbd_conn_get_fd(struct ctdbd_connection *conn)
526 {
527 return conn->fd;
528 }
529
530 /*
531 * Packet handler to receive and handle a ctdb message
532 */
533 static int ctdb_handle_message(struct ctdbd_connection *conn,
534 struct ctdb_req_header *hdr)
535 {
536 struct ctdb_req_message *msg;
537
538 if (hdr->operation != CTDB_REQ_MESSAGE) {
539 DEBUG(0, ("Received async msg of type %u, discarding\n",
540 hdr->operation));
541 return EINVAL;
542 }
543
544 msg = (struct ctdb_req_message *)hdr;
545
546 ctdbd_msg_call_back(conn, msg);
547
548 return 0;
549 }
550
551 /*
552 * The ctdbd socket is readable asynchronuously
553 */
554
555 static void ctdbd_socket_handler(struct tevent_context *event_ctx,
556 struct tevent_fd *event,
557 uint16_t flags,
558 void *private_data)
559 {
560 struct ctdbd_connection *conn = talloc_get_type_abort(
561 private_data, struct ctdbd_connection);
562 struct ctdb_req_header *hdr = NULL;
563 int ret;
564
565 ret = ctdb_read_packet(conn->fd, conn->timeout, talloc_tos(), &hdr);
566 if (ret != 0) {
567 DEBUG(0, ("ctdb_read_packet failed: %s\n", strerror(ret)));
568 cluster_fatal("ctdbd died\n");
569 }
570
571 ret = ctdb_handle_message(conn, hdr);
572
573 TALLOC_FREE(hdr);
574
575 if (ret != 0) {
576 DEBUG(10, ("could not handle incoming message: %s\n",
577 strerror(ret)));
578 }
579 }
580
581 /*
582 * Prepare a ctdbd connection to receive messages
583 */
584
585 int ctdbd_register_msg_ctx(struct ctdbd_connection *conn,
586 struct messaging_context *msg_ctx)
587 {
588 SMB_ASSERT(conn->msg_ctx == NULL);
589 SMB_ASSERT(conn->fde == NULL);
590
591 if (!(conn->fde = tevent_add_fd(messaging_tevent_context(msg_ctx),
592 conn,
593 conn->fd,
594 TEVENT_FD_READ,
595 ctdbd_socket_handler,
596 conn))) {
597 DEBUG(0, ("event_add_fd failed\n"));
598 return ENOMEM;
599 }
600
601 conn->msg_ctx = msg_ctx;
602
603 return 0;
604 }
605
606 NTSTATUS ctdbd_messaging_send_iov(struct ctdbd_connection *conn,
607 uint32_t dst_vnn, uint64_t dst_srvid,
608 const struct iovec *iov, int iovlen)
609 {
610 struct ctdb_req_message r;
611 struct iovec iov2[iovlen+1];
612 size_t buflen = iov_buflen(iov, iovlen);
613 ssize_t nwritten;
614
615 r.hdr.length = offsetof(struct ctdb_req_message, data) + buflen;
616 r.hdr.ctdb_magic = CTDB_MAGIC;
617 r.hdr.ctdb_version = CTDB_PROTOCOL;
618 r.hdr.generation = 1;
619 r.hdr.operation = CTDB_REQ_MESSAGE;
620 r.hdr.destnode = dst_vnn;
621 r.hdr.srcnode = conn->our_vnn;
622 r.hdr.reqid = 0;
623 r.srvid = dst_srvid;
624 r.datalen = buflen;
625
626 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
627 ctdb_packet_dump(&r.hdr);
628
629 iov2[0].iov_base = &r;
630 iov2[0].iov_len = offsetof(struct ctdb_req_message, data);
631 memcpy(&iov2[1], iov, iovlen * sizeof(struct iovec));
632
633 nwritten = write_data_iov(conn->fd, iov2, iovlen+1);
634 if (nwritten == -1) {
635 DEBUG(3, ("write_data_iov failed: %s\n", strerror(errno)));
636 cluster_fatal("cluster dispatch daemon msg write error\n");
637 }
638
639 return NT_STATUS_OK;
640 }
641
642 /*
643 * send/recv a generic ctdb control message
644 */
645 static int ctdbd_control(struct ctdbd_connection *conn,
646 uint32_t vnn, uint32_t opcode,
647 uint64_t srvid, uint32_t flags,
648 TDB_DATA data,
649 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
650 int *cstatus)
651 {
652 struct ctdb_req_control req;
653 struct ctdb_req_header *hdr;
654 struct ctdb_reply_control *reply = NULL;
655 struct iovec iov[2];
656 ssize_t nwritten;
657 int ret;
658
659 ZERO_STRUCT(req);
660 req.hdr.length = offsetof(struct ctdb_req_control, data) + data.dsize;
661 req.hdr.ctdb_magic = CTDB_MAGIC;
662 req.hdr.ctdb_version = CTDB_PROTOCOL;
663 req.hdr.operation = CTDB_REQ_CONTROL;
664 req.hdr.reqid = ctdbd_next_reqid(conn);
665 req.hdr.destnode = vnn;
666 req.opcode = opcode;
667 req.srvid = srvid;
668 req.datalen = data.dsize;
669 req.flags = flags;
670
671 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
672 ctdb_packet_dump(&req.hdr);
673
674 iov[0].iov_base = &req;
675 iov[0].iov_len = offsetof(struct ctdb_req_control, data);
676 iov[1].iov_base = data.dptr;
677 iov[1].iov_len = data.dsize;
678
679 nwritten = write_data_iov(conn->fd, iov, ARRAY_SIZE(iov));
680 if (nwritten == -1) {
681 DEBUG(3, ("write_data_iov failed: %s\n", strerror(errno)));
682 cluster_fatal("cluster dispatch daemon msg write error\n");
683 }
684
685 if (flags & CTDB_CTRL_FLAG_NOREPLY) {
686 if (cstatus) {
687 *cstatus = 0;
688 }
689 return 0;
690 }
691
692 ret = ctdb_read_req(conn, req.hdr.reqid, NULL, &hdr);
693 if (ret != 0) {
694 DEBUG(10, ("ctdb_read_req failed: %s\n", strerror(ret)));
695 return ret;
696 }
697
698 if (hdr->operation != CTDB_REPLY_CONTROL) {
699 DEBUG(0, ("received invalid reply\n"));
700 TALLOC_FREE(hdr);
701 return EIO;
702 }
703 reply = (struct ctdb_reply_control *)hdr;
704
705 if (outdata) {
706 if (!(outdata->dptr = (uint8_t *)talloc_memdup(
707 mem_ctx, reply->data, reply->datalen))) {
708 TALLOC_FREE(reply);
709 return ENOMEM;
710 }
711 outdata->dsize = reply->datalen;
712 }
713 if (cstatus) {
714 (*cstatus) = reply->status;
715 }
716
717 TALLOC_FREE(reply);
718 return ret;
719 }
720
721 /*
722 * see if a remote process exists
723 */
724 bool ctdbd_process_exists(struct ctdbd_connection *conn, uint32_t vnn, pid_t pid)
725 {
726 struct server_id id;
727 bool result;
728
729 id.pid = pid;
730 id.vnn = vnn;
731
732 if (!ctdb_processes_exist(conn, &id, 1, &result)) {
733 DEBUG(10, ("ctdb_processes_exist failed\n"));
734 return false;
735 }
736 return result;
737 }
738
739 bool ctdb_processes_exist(struct ctdbd_connection *conn,
740 const struct server_id *pids, int num_pids,
741 bool *results)
742 {
743 TALLOC_CTX *frame = talloc_stackframe();
744 int i, num_received;
745 uint32_t *reqids;
746 bool result = false;
747
748 reqids = talloc_array(talloc_tos(), uint32_t, num_pids);
749 if (reqids == NULL) {
750 goto fail;
751 }
752
753 for (i=0; i<num_pids; i++) {
754 struct ctdb_req_control req;
755 pid_t pid;
756 struct iovec iov[2];
757 ssize_t nwritten;
758
759 results[i] = false;
760 reqids[i] = ctdbd_next_reqid(conn);
761
762 ZERO_STRUCT(req);
763
764 /*
765 * pids[i].pid is uint64_t, scale down to pid_t which
766 * is the wire protocol towards ctdb.
767 */
768 pid = pids[i].pid;
769
770 DEBUG(10, ("Requesting PID %d/%d, reqid=%d\n",
771 (int)pids[i].vnn, (int)pid,
772 (int)reqids[i]));
773
774 req.hdr.length = offsetof(struct ctdb_req_control, data);
775 req.hdr.length += sizeof(pid);
776 req.hdr.ctdb_magic = CTDB_MAGIC;
777 req.hdr.ctdb_version = CTDB_PROTOCOL;
778 req.hdr.operation = CTDB_REQ_CONTROL;
779 req.hdr.reqid = reqids[i];
780 req.hdr.destnode = pids[i].vnn;
781 req.opcode = CTDB_CONTROL_PROCESS_EXISTS;
782 req.srvid = 0;
783 req.datalen = sizeof(pid);
784 req.flags = 0;
785
786 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
787 ctdb_packet_dump(&req.hdr);
788
789 iov[0].iov_base = &req;
790 iov[0].iov_len = offsetof(struct ctdb_req_control, data);
791 iov[1].iov_base = &pid;
792 iov[1].iov_len = sizeof(pid);
793
794 nwritten = write_data_iov(conn->fd, iov, ARRAY_SIZE(iov));
795 if (nwritten == -1) {
796 DEBUG(10, ("write_data_iov failed: %s\n",
797 strerror(errno)));
798 goto fail;
799 }
800 }
801
802 num_received = 0;
803
804 while (num_received < num_pids) {
805 struct ctdb_req_header *hdr;
806 struct ctdb_reply_control *reply;
807 uint32_t reqid;
808 int ret;
809
810 ret = ctdb_read_req(conn, 0, talloc_tos(), &hdr);
811 if (ret != 0) {
812 DEBUG(10, ("ctdb_read_req failed: %s\n",
813 strerror(ret)));
814 goto fail;
815 }
816
817 if (hdr->operation != CTDB_REPLY_CONTROL) {
818 DEBUG(10, ("Received invalid reply\n"));
819 goto fail;
820 }
821 reply = (struct ctdb_reply_control *)hdr;
822
823 reqid = reply->hdr.reqid;
824
825 DEBUG(10, ("Received reqid %d\n", (int)reqid));
826
827 for (i=0; i<num_pids; i++) {
828 if (reqid == reqids[i]) {
829 break;
830 }
831 }
832 if (i == num_pids) {
833 DEBUG(10, ("Received unknown record number %u\n",
834 (unsigned)reqid));
835 goto fail;
836 }
837 results[i] = ((reply->status) == 0);
838 TALLOC_FREE(reply);
839 num_received += 1;
840 }
841
842 result = true;
843 fail:
844 TALLOC_FREE(frame);
845 return result;
846 }
847
848 /*
849 * Get a db path
850 */
851 char *ctdbd_dbpath(struct ctdbd_connection *conn,
852 TALLOC_CTX *mem_ctx, uint32_t db_id)
853 {
854 int ret;
855 TDB_DATA data;
856 TDB_DATA rdata = {0};
857 int32_t cstatus = 0;
858
859 data.dptr = (uint8_t*)&db_id;
860 data.dsize = sizeof(db_id);
861
862 ret = ctdbd_control(conn, CTDB_CURRENT_NODE,
863 CTDB_CONTROL_GETDBPATH, 0, 0, data,
864 mem_ctx, &rdata, &cstatus);
865 if ((ret != 0) || cstatus != 0) {
866 DEBUG(0, (__location__ " ctdb_control for getdbpath failed: %s\n",
867 strerror(ret)));
868 return NULL;
869 }
870
871 return (char *)rdata.dptr;
872 }
873
874 /*
875 * attach to a ctdb database
876 */
877 NTSTATUS ctdbd_db_attach(struct ctdbd_connection *conn,
878 const char *name, uint32_t *db_id, int tdb_flags)
879 {
880 int ret;
881 TDB_DATA data;
882 int32_t cstatus;
883 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
884
885 data = string_term_tdb_data(name);
886
887 ret = ctdbd_control(conn, CTDB_CURRENT_NODE,
888 persistent
889 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
890 : CTDB_CONTROL_DB_ATTACH,
891 tdb_flags, 0, data, NULL, &data, &cstatus);
892 if (ret != 0) {
893 DEBUG(0, (__location__ " ctdb_control for db_attach "
894 "failed: %s\n", strerror(ret)));
895 return map_nt_error_from_unix(ret);
896 }
897
898 if (cstatus != 0 || data.dsize != sizeof(uint32_t)) {
899 DEBUG(0,(__location__ " ctdb_control for db_attach failed\n"));
900 return NT_STATUS_INTERNAL_ERROR;
901 }
902
903 *db_id = *(uint32_t *)data.dptr;
904 talloc_free(data.dptr);
905
906 if (!(tdb_flags & TDB_SEQNUM)) {
907 return NT_STATUS_OK;
908 }
909
910 data.dptr = (uint8_t *)db_id;
911 data.dsize = sizeof(*db_id);
912
913 ret = ctdbd_control(conn, CTDB_CURRENT_NODE,
914 CTDB_CONTROL_ENABLE_SEQNUM, 0, 0, data,
915 NULL, NULL, &cstatus);
916 if ((ret != 0) || cstatus != 0) {
917 DEBUG(0, (__location__ " ctdb_control for enable seqnum "
918 "failed: %s\n", strerror(ret)));
919 return (ret == 0) ? NT_STATUS_INTERNAL_ERROR :
920 map_nt_error_from_unix(ret);
921 }
922
923 return NT_STATUS_OK;
924 }
925
926 /*
927 * force the migration of a record to this node
928 */
929 NTSTATUS ctdbd_migrate(struct ctdbd_connection *conn, uint32_t db_id,
930 TDB_DATA key)
931 {
932 struct ctdb_req_call req;
933 struct ctdb_req_header *hdr;
934 struct iovec iov[2];
935 ssize_t nwritten;
936 NTSTATUS status;
937 int ret;
938
939 ZERO_STRUCT(req);
940
941 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
942 req.hdr.ctdb_magic = CTDB_MAGIC;
943 req.hdr.ctdb_version = CTDB_PROTOCOL;
944 req.hdr.operation = CTDB_REQ_CALL;
945 req.hdr.reqid = ctdbd_next_reqid(conn);
946 req.flags = CTDB_IMMEDIATE_MIGRATION;
947 req.callid = CTDB_NULL_FUNC;
948 req.db_id = db_id;
949 req.keylen = key.dsize;
950
951 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
952 ctdb_packet_dump(&req.hdr);
953
954 iov[0].iov_base = &req;
955 iov[0].iov_len = offsetof(struct ctdb_req_call, data);
956 iov[1].iov_base = key.dptr;
957 iov[1].iov_len = key.dsize;
958
959 nwritten = write_data_iov(conn->fd, iov, ARRAY_SIZE(iov));
960 if (nwritten == -1) {
961 DEBUG(3, ("write_data_iov failed: %s\n", strerror(errno)));
962 cluster_fatal("cluster dispatch daemon msg write error\n");
963 }
964
965 ret = ctdb_read_req(conn, req.hdr.reqid, NULL, &hdr);
966 if (ret != 0) {
967 DEBUG(10, ("ctdb_read_req failed: %s\n", strerror(ret)));
968 status = map_nt_error_from_unix(ret);
969 goto fail;
970 }
971
972 if (hdr->operation != CTDB_REPLY_CALL) {
973 DEBUG(0, ("received invalid reply\n"));
974 status = NT_STATUS_INTERNAL_ERROR;
975 goto fail;
976 }
977
978 status = NT_STATUS_OK;
979 fail:
980
981 TALLOC_FREE(hdr);
982 return status;
983 }
984
985 /*
986 * Fetch a record and parse it
987 */
988 NTSTATUS ctdbd_parse(struct ctdbd_connection *conn, uint32_t db_id,
989 TDB_DATA key, bool local_copy,
990 void (*parser)(TDB_DATA key, TDB_DATA data,
991 void *private_data),
992 void *private_data)
993 {
994 struct ctdb_req_call req;
995 struct ctdb_req_header *hdr = NULL;
996 struct ctdb_reply_call *reply;
997 struct iovec iov[2];
998 ssize_t nwritten;
999 NTSTATUS status;
1000 uint32_t flags;
1001 int ret;
1002
1003 flags = local_copy ? CTDB_WANT_READONLY : 0;
1004
1005 ZERO_STRUCT(req);
1006
1007 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
1008 req.hdr.ctdb_magic = CTDB_MAGIC;
1009 req.hdr.ctdb_version = CTDB_PROTOCOL;
1010 req.hdr.operation = CTDB_REQ_CALL;
1011 req.hdr.reqid = ctdbd_next_reqid(conn);
1012 req.flags = flags;
1013 req.callid = CTDB_FETCH_FUNC;
1014 req.db_id = db_id;
1015 req.keylen = key.dsize;
1016
1017 iov[0].iov_base = &req;
1018 iov[0].iov_len = offsetof(struct ctdb_req_call, data);
1019 iov[1].iov_base = key.dptr;
1020 iov[1].iov_len = key.dsize;
1021
1022 nwritten = write_data_iov(conn->fd, iov, ARRAY_SIZE(iov));
1023 if (nwritten == -1) {
1024 DEBUG(3, ("write_data_iov failed: %s\n", strerror(errno)));
1025 cluster_fatal("cluster dispatch daemon msg write error\n");
1026 }
1027
1028 ret = ctdb_read_req(conn, req.hdr.reqid, NULL, &hdr);
1029 if (ret != 0) {
1030 DEBUG(10, ("ctdb_read_req failed: %s\n", strerror(ret)));
1031 status = map_nt_error_from_unix(ret);
1032 goto fail;
1033 }
1034
1035 if ((hdr == NULL) || (hdr->operation != CTDB_REPLY_CALL)) {
1036 DEBUG(0, ("received invalid reply\n"));
1037 status = NT_STATUS_INTERNAL_ERROR;
1038 goto fail;
1039 }
1040 reply = (struct ctdb_reply_call *)hdr;
1041
1042 if (reply->datalen == 0) {
1043 /*
1044 * Treat an empty record as non-existing
1045 */
1046 status = NT_STATUS_NOT_FOUND;
1047 goto fail;
1048 }
1049
1050 parser(key, make_tdb_data(&reply->data[0], reply->datalen),
1051 private_data);
1052
1053 status = NT_STATUS_OK;
1054 fail:
1055 TALLOC_FREE(hdr);
1056 return status;
1057 }
1058
1059 /*
1060 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1061 connection to ctdbd to avoid the hairy recursive and async problems with
1062 everything in-line.
1063 */
1064
1065 NTSTATUS ctdbd_traverse(struct ctdbd_connection *master, uint32_t db_id,
1066 void (*fn)(TDB_DATA key, TDB_DATA data,
1067 void *private_data),
1068 void *private_data)
1069 {
1070 struct ctdbd_connection *conn;
1071 NTSTATUS status;
1072 int ret;
1073 TDB_DATA key, data;
1074 struct ctdb_traverse_start t;
1075 int cstatus;
1076
1077 become_root();
1078 status = ctdbd_init_connection(NULL, master->sockname, master->timeout,
1079 &conn);
1080 unbecome_root();
1081 if (!NT_STATUS_IS_OK(status)) {
1082 DEBUG(0, ("ctdbd_init_connection failed: %s\n",
1083 nt_errstr(status)));
1084 return status;
1085 }
1086
1087 t.db_id = db_id;
1088 t.srvid = conn->rand_srvid;
1089 t.reqid = ctdbd_next_reqid(conn);
1090
1091 data.dptr = (uint8_t *)&t;
1092 data.dsize = sizeof(t);
1093
1094 ret = ctdbd_control(conn, CTDB_CURRENT_NODE,
1095 CTDB_CONTROL_TRAVERSE_START, conn->rand_srvid,
1096 0, data, NULL, NULL, &cstatus);
1097
1098 if ((ret != 0) || (cstatus != 0)) {
1099 status = map_nt_error_from_unix(ret);
1100
1101 DEBUG(0,("ctdbd_control failed: %s, %d\n", strerror(ret),
1102 cstatus));
1103
1104 if (NT_STATUS_IS_OK(status)) {
1105 /*
1106 * We need a mapping here
1107 */
1108 status = NT_STATUS_UNSUCCESSFUL;
1109 }
1110 TALLOC_FREE(conn);
1111 return status;
1112 }
1113
1114 while (True) {
1115 struct ctdb_req_header *hdr = NULL;
1116 struct ctdb_req_message *m;
1117 struct ctdb_rec_data *d;
1118
1119 ret = ctdb_read_packet(conn->fd, conn->timeout, conn, &hdr);
1120 if (ret != 0) {
1121 DEBUG(0, ("ctdb_read_packet failed: %s\n",
1122 strerror(ret)));
1123 cluster_fatal("ctdbd died\n");
1124 }
1125
1126 if (hdr->operation != CTDB_REQ_MESSAGE) {
1127 DEBUG(0, ("Got operation %u, expected a message\n",
1128 (unsigned)hdr->operation));
1129 TALLOC_FREE(conn);
1130 return NT_STATUS_UNEXPECTED_IO_ERROR;
1131 }
1132
1133 m = (struct ctdb_req_message *)hdr;
1134 d = (struct ctdb_rec_data *)&m->data[0];
1135 if (m->datalen < sizeof(uint32_t) || m->datalen != d->length) {
1136 DEBUG(0, ("Got invalid traverse data of length %d\n",
1137 (int)m->datalen));
1138 TALLOC_FREE(conn);
1139 return NT_STATUS_UNEXPECTED_IO_ERROR;
1140 }
1141
1142 key.dsize = d->keylen;
1143 key.dptr = &d->data[0];
1144 data.dsize = d->datalen;
1145 data.dptr = &d->data[d->keylen];
1146
1147 if (key.dsize == 0 && data.dsize == 0) {
1148 /* end of traverse */
1149 TALLOC_FREE(conn);
1150 return NT_STATUS_OK;
1151 }
1152
1153 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1154 DEBUG(0, ("Got invalid ltdb header length %d\n",
1155 (int)data.dsize));
1156 TALLOC_FREE(conn);
1157 return NT_STATUS_UNEXPECTED_IO_ERROR;
1158 }
1159 data.dsize -= sizeof(struct ctdb_ltdb_header);
1160 data.dptr += sizeof(struct ctdb_ltdb_header);
1161
1162 if (fn != NULL) {
1163 fn(key, data, private_data);
1164 }
1165 }
1166 return NT_STATUS_OK;
1167 }
1168
1169 /*
1170 This is used to canonicalize a ctdb_sock_addr structure.
1171 */
1172 static void smbd_ctdb_canonicalize_ip(const struct sockaddr_storage *in,
1173 struct sockaddr_storage *out)
1174 {
1175 memcpy(out, in, sizeof (*out));
1176
1177 #ifdef HAVE_IPV6
1178 if (in->ss_family == AF_INET6) {
1179 const char prefix[12] = { 0,0,0,0,0,0,0,0,0,0,0xff,0xff };
1180 const struct sockaddr_in6 *in6 =
1181 (const struct sockaddr_in6 *)in;
1182 struct sockaddr_in *out4 = (struct sockaddr_in *)out;
1183 if (memcmp(&in6->sin6_addr, prefix, 12) == 0) {
1184 memset(out, 0, sizeof(*out));
1185 #ifdef HAVE_SOCK_SIN_LEN
1186 out4->sin_len = sizeof(*out);
1187 #endif
1188 out4->sin_family = AF_INET;
1189 out4->sin_port = in6->sin6_port;
1190 memcpy(&out4->sin_addr, &in6->sin6_addr.s6_addr[12], 4);
1191 }
1192 }
1193 #endif
1194 }
1195
1196 /*
1197 * Register us as a server for a particular tcp connection
1198 */
1199
1200 NTSTATUS ctdbd_register_ips(struct ctdbd_connection *conn,
1201 const struct sockaddr_storage *_server,
1202 const struct sockaddr_storage *_client,
1203 int (*cb)(uint32_t src_vnn, uint32_t dst_vnn,
1204 uint64_t dst_srvid,
1205 const uint8_t *msg, size_t msglen,
1206 void *private_data),
1207 void *private_data)
1208 {
1209 struct ctdb_control_tcp_addr p;
1210 TDB_DATA data = { .dptr = (uint8_t *)&p, .dsize = sizeof(p) };
1211 int ret;
1212 struct sockaddr_storage client;
1213 struct sockaddr_storage server;
1214
1215 /*
1216 * Only one connection so far
1217 */
1218
1219 smbd_ctdb_canonicalize_ip(_client, &client);
1220 smbd_ctdb_canonicalize_ip(_server, &server);
1221
1222 switch (client.ss_family) {
1223 case AF_INET:
1224 memcpy(&p.dest.ip, &server, sizeof(p.dest.ip));
1225 memcpy(&p.src.ip, &client, sizeof(p.src.ip));
1226 break;
1227 case AF_INET6:
1228 memcpy(&p.dest.ip6, &server, sizeof(p.dest.ip6));
1229 memcpy(&p.src.ip6, &client, sizeof(p.src.ip6));
1230 break;
1231 default:
1232 return NT_STATUS_INTERNAL_ERROR;
1233 }
1234
1235 /*
1236 * We want to be told about IP releases
1237 */
1238
1239 ret = register_with_ctdbd(conn, CTDB_SRVID_RELEASE_IP,
1240 cb, private_data);
1241 if (ret != 0) {
1242 return map_nt_error_from_unix(ret);
1243 }
1244
1245 /*
1246 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1247 * can send an extra ack to trigger a reset for our client, so it
1248 * immediately reconnects
1249 */
1250 ret = ctdbd_control(conn, CTDB_CURRENT_NODE,
1251 CTDB_CONTROL_TCP_CLIENT, 0,
1252 CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL,
1253 NULL);
1254 if (ret != 0) {
1255 return map_nt_error_from_unix(ret);
1256 }
1257 return NT_STATUS_OK;
1258 }
1259
1260 /*
1261 call a control on the local node
1262 */
1263 NTSTATUS ctdbd_control_local(struct ctdbd_connection *conn, uint32_t opcode,
1264 uint64_t srvid, uint32_t flags, TDB_DATA data,
1265 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
1266 int *cstatus)
1267 {
1268 int ret;
1269
1270 ret = ctdbd_control(conn, CTDB_CURRENT_NODE, opcode, srvid, flags, data,
1271 mem_ctx, outdata, cstatus);
1272 if (ret != 0) {
1273 return map_nt_error_from_unix(ret);
1274 }
1275 return NT_STATUS_OK;
1276 }
1277
1278 NTSTATUS ctdb_watch_us(struct ctdbd_connection *conn)
1279 {
1280 struct ctdb_client_notify_register reg_data;
1281 size_t struct_len;
1282 NTSTATUS status;
1283 int cstatus;
1284
1285 reg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1286 reg_data.len = 1;
1287 reg_data.notify_data[0] = 0;
1288
1289 struct_len = offsetof(struct ctdb_client_notify_register,
1290 notify_data) + reg_data.len;
1291
1292 status = ctdbd_control_local(
1293 conn, CTDB_CONTROL_REGISTER_NOTIFY, conn->rand_srvid, 0,
1294 make_tdb_data((uint8_t *)&reg_data, struct_len),
1295 NULL, NULL, &cstatus);
1296 if (!NT_STATUS_IS_OK(status)) {
1297 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1298 nt_errstr(status)));
1299 }
1300 return status;
1301 }
1302
1303 NTSTATUS ctdb_unwatch(struct ctdbd_connection *conn)
1304 {
1305 struct ctdb_client_notify_deregister dereg_data;
1306 NTSTATUS status;
1307 int cstatus;
1308
1309 dereg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1310
1311 status = ctdbd_control_local(
1312 conn, CTDB_CONTROL_DEREGISTER_NOTIFY, conn->rand_srvid, 0,
1313 make_tdb_data((uint8_t *)&dereg_data, sizeof(dereg_data)),
1314 NULL, NULL, &cstatus);
1315 if (!NT_STATUS_IS_OK(status)) {
1316 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1317 nt_errstr(status)));
1318 }
1319 return status;
1320 }
1321
1322 NTSTATUS ctdbd_probe(const char *sockname, int timeout)
1323 {
1324 /*
1325 * Do a very early check if ctdbd is around to avoid an abort and core
1326 * later
1327 */
1328 struct ctdbd_connection *conn = NULL;
1329 NTSTATUS status;
1330
1331 status = ctdbd_messaging_connection(talloc_tos(), sockname, timeout,
1332 &conn);
1333
1334 /*
1335 * We only care if we can connect.
1336 */
1337 TALLOC_FREE(conn);
1338
1339 return status;
1340 }
Samba GIT mirror