Fix a race condition in winbind leading to a crash
[samba.git] / source / winbindd / winbindd_dual.c
1 /* 
2    Unix SMB/CIFS implementation.
3
4    Winbind child daemons
5
6    Copyright (C) Andrew Tridgell 2002
7    Copyright (C) Volker Lendecke 2004,2005
8    
9    This program is free software; you can redistribute it and/or modify
10    it under the terms of the GNU General Public License as published by
11    the Free Software Foundation; either version 3 of the License, or
12    (at your option) any later version.
13    
14    This program is distributed in the hope that it will be useful,
15    but WITHOUT ANY WARRANTY; without even the implied warranty of
16    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17    GNU General Public License for more details.
18    
19    You should have received a copy of the GNU General Public License
20    along with this program.  If not, see <http://www.gnu.org/licenses/>.
21 */
22
23 /*
24  * We fork a child per domain to be able to act non-blocking in the main
25  * winbind daemon. A domain controller thousands of miles away being being
26  * slow replying with a 10.000 user list should not hold up netlogon calls
27  * that can be handled locally.
28  */
29
30 #include "includes.h"
31 #include "winbindd.h"
32
33 #undef DBGC_CLASS
34 #define DBGC_CLASS DBGC_WINBIND
35
36 extern bool override_logfile;
37 extern struct winbindd_methods cache_methods;
38
39 /* Read some data from a client connection */
40
41 static void child_read_request(struct winbindd_cli_state *state)
42 {
43         NTSTATUS status;
44
45         /* Read data */
46
47         status = read_data(state->sock, (char *)&state->request,
48                            sizeof(state->request));
49
50         if (!NT_STATUS_IS_OK(status)) {
51                 DEBUG(3, ("child_read_request: read_data failed: %s\n",
52                           nt_errstr(status)));
53                 state->finished = True;
54                 return;
55         }
56
57         if (state->request.extra_len == 0) {
58                 state->request.extra_data.data = NULL;
59                 return;
60         }
61
62         DEBUG(10, ("Need to read %d extra bytes\n", (int)state->request.extra_len));
63
64         state->request.extra_data.data =
65                 SMB_MALLOC_ARRAY(char, state->request.extra_len + 1);
66
67         if (state->request.extra_data.data == NULL) {
68                 DEBUG(0, ("malloc failed\n"));
69                 state->finished = True;
70                 return;
71         }
72
73         /* Ensure null termination */
74         state->request.extra_data.data[state->request.extra_len] = '\0';
75
76         status= read_data(state->sock, state->request.extra_data.data,
77                           state->request.extra_len);
78
79         if (!NT_STATUS_IS_OK(status)) {
80                 DEBUG(0, ("Could not read extra data: %s\n",
81                           nt_errstr(status)));
82                 state->finished = True;
83                 return;
84         }
85 }
86
87 /*
88  * Machinery for async requests sent to children. You set up a
89  * winbindd_request, select a child to query, and issue a async_request
90  * call. When the request is completed, the callback function you specified is
91  * called back with the private pointer you gave to async_request.
92  */
93
94 struct winbindd_async_request {
95         struct winbindd_async_request *next, *prev;
96         TALLOC_CTX *mem_ctx;
97         struct winbindd_child *child;
98         struct winbindd_request *request;
99         struct winbindd_response *response;
100         void (*continuation)(void *private_data, bool success);
101         struct timed_event *reply_timeout_event;
102         pid_t child_pid; /* pid of the child we're waiting on. Used to detect
103                             a restart of the child (child->pid != child_pid). */
104         void *private_data;
105 };
106
107 static void async_request_fail(struct winbindd_async_request *state);
108 static void async_main_request_sent(void *private_data, bool success);
109 static void async_request_sent(void *private_data, bool success);
110 static void async_reply_recv(void *private_data, bool success);
111 static void schedule_async_request(struct winbindd_child *child);
112
113 void async_request(TALLOC_CTX *mem_ctx, struct winbindd_child *child,
114                    struct winbindd_request *request,
115                    struct winbindd_response *response,
116                    void (*continuation)(void *private_data, bool success),
117                    void *private_data)
118 {
119         struct winbindd_async_request *state;
120
121         SMB_ASSERT(continuation != NULL);
122
123         state = TALLOC_P(mem_ctx, struct winbindd_async_request);
124
125         if (state == NULL) {
126                 DEBUG(0, ("talloc failed\n"));
127                 continuation(private_data, False);
128                 return;
129         }
130
131         state->mem_ctx = mem_ctx;
132         state->child = child;
133         state->reply_timeout_event = NULL;
134         state->request = request;
135         state->response = response;
136         state->continuation = continuation;
137         state->private_data = private_data;
138
139         DLIST_ADD_END(child->requests, state, struct winbindd_async_request *);
140
141         schedule_async_request(child);
142
143         return;
144 }
145
146 static void async_main_request_sent(void *private_data, bool success)
147 {
148         struct winbindd_async_request *state =
149                 talloc_get_type_abort(private_data, struct winbindd_async_request);
150
151         if (!success) {
152                 DEBUG(5, ("Could not send async request\n"));
153                 async_request_fail(state);
154                 return;
155         }
156
157         if (state->request->extra_len == 0) {
158                 async_request_sent(private_data, True);
159                 return;
160         }
161
162         setup_async_write(&state->child->event, state->request->extra_data.data,
163                           state->request->extra_len,
164                           async_request_sent, state);
165 }
166
167 /****************************************************************
168  Handler triggered if the child winbindd doesn't respond within
169  a given timeout.
170 ****************************************************************/
171
172 static void async_request_timeout_handler(struct event_context *ctx,
173                                         struct timed_event *te,
174                                         const struct timeval *now,
175                                         void *private_data)
176 {
177         struct winbindd_async_request *state =
178                 talloc_get_type_abort(private_data, struct winbindd_async_request);
179
180         DEBUG(0,("async_request_timeout_handler: child pid %u is not responding. "
181                 "Closing connection to it.\n",
182                 state->child_pid ));
183
184         /* Deal with the reply - set to error. */
185         async_reply_recv(private_data, False);
186 }
187
188 /**************************************************************
189  Common function called on both async send and recv fail.
190  Cleans up the child and schedules the next request.
191 **************************************************************/
192
193 static void async_request_fail(struct winbindd_async_request *state)
194 {
195         DLIST_REMOVE(state->child->requests, state);
196
197         TALLOC_FREE(state->reply_timeout_event);
198
199         SMB_ASSERT(state->child_pid != (pid_t)0);
200
201         /* If not already reaped, send kill signal to child. */
202         if (state->child->pid == state->child_pid) {
203                 kill(state->child_pid, SIGTERM);
204
205                 /* 
206                  * Close the socket to the child.
207                  */
208                 winbind_child_died(state->child_pid);
209         }
210
211         state->response->length = sizeof(struct winbindd_response);
212         state->response->result = WINBINDD_ERROR;
213         state->continuation(state->private_data, False);
214 }
215
216 static void async_request_sent(void *private_data_data, bool success)
217 {
218         struct winbindd_async_request *state =
219                 talloc_get_type_abort(private_data_data, struct winbindd_async_request);
220
221         if (!success) {
222                 DEBUG(5, ("Could not send async request to child pid %u\n",
223                         (unsigned int)state->child_pid ));
224                 async_request_fail(state);
225                 return;
226         }
227
228         /* Request successfully sent to the child, setup the wait for reply */
229
230         setup_async_read(&state->child->event,
231                          &state->response->result,
232                          sizeof(state->response->result),
233                          async_reply_recv, state);
234
235         /* 
236          * Set up a timeout of 300 seconds for the response.
237          * If we don't get it close the child socket and
238          * report failure.
239          */
240
241         state->reply_timeout_event = event_add_timed(winbind_event_context(),
242                                                         NULL,
243                                                         timeval_current_ofs(300,0),
244                                                         "async_request_timeout",
245                                                         async_request_timeout_handler,
246                                                         state);
247         if (!state->reply_timeout_event) {
248                 smb_panic("async_request_sent: failed to add timeout handler.\n");
249         }
250 }
251
252 static void async_reply_recv(void *private_data, bool success)
253 {
254         struct winbindd_async_request *state =
255                 talloc_get_type_abort(private_data, struct winbindd_async_request);
256         struct winbindd_child *child = state->child;
257
258         TALLOC_FREE(state->reply_timeout_event);
259
260         state->response->length = sizeof(struct winbindd_response);
261
262         if (!success) {
263                 DEBUG(5, ("Could not receive async reply from child pid %u\n",
264                         (unsigned int)state->child_pid ));
265
266                 cache_cleanup_response(state->child_pid);
267                 async_request_fail(state);
268                 return;
269         }
270
271         SMB_ASSERT(cache_retrieve_response(state->child_pid,
272                                            state->response));
273
274         cache_cleanup_response(state->child_pid);
275         
276         DLIST_REMOVE(child->requests, state);
277
278         schedule_async_request(child);
279
280         state->continuation(state->private_data, True);
281 }
282
283 static bool fork_domain_child(struct winbindd_child *child);
284
285 static void schedule_async_request(struct winbindd_child *child)
286 {
287         struct winbindd_async_request *request = child->requests;
288
289         if (request == NULL) {
290                 return;
291         }
292
293         if (child->event.flags != 0) {
294                 return;         /* Busy */
295         }
296
297         if ((child->pid == 0) && (!fork_domain_child(child))) {
298                 /* Cancel all outstanding requests */
299
300                 while (request != NULL) {
301                         /* request might be free'd in the continuation */
302                         struct winbindd_async_request *next = request->next;
303                         request->continuation(request->private_data, False);
304                         request = next;
305                 }
306                 return;
307         }
308
309         /* Now we know who we're sending to - remember the pid. */
310         request->child_pid = child->pid;
311
312         setup_async_write(&child->event, request->request,
313                           sizeof(*request->request),
314                           async_main_request_sent, request);
315
316         return;
317 }
318
319 struct domain_request_state {
320         TALLOC_CTX *mem_ctx;
321         struct winbindd_domain *domain;
322         struct winbindd_request *request;
323         struct winbindd_response *response;
324         void (*continuation)(void *private_data_data, bool success);
325         void *private_data_data;
326 };
327
328 static void domain_init_recv(void *private_data_data, bool success);
329
330 void async_domain_request(TALLOC_CTX *mem_ctx,
331                           struct winbindd_domain *domain,
332                           struct winbindd_request *request,
333                           struct winbindd_response *response,
334                           void (*continuation)(void *private_data_data, bool success),
335                           void *private_data_data)
336 {
337         struct domain_request_state *state;
338
339         if (domain->initialized) {
340                 async_request(mem_ctx, &domain->child, request, response,
341                               continuation, private_data_data);
342                 return;
343         }
344
345         state = TALLOC_P(mem_ctx, struct domain_request_state);
346         if (state == NULL) {
347                 DEBUG(0, ("talloc failed\n"));
348                 continuation(private_data_data, False);
349                 return;
350         }
351
352         state->mem_ctx = mem_ctx;
353         state->domain = domain;
354         state->request = request;
355         state->response = response;
356         state->continuation = continuation;
357         state->private_data_data = private_data_data;
358
359         init_child_connection(domain, domain_init_recv, state);
360 }
361
362 static void domain_init_recv(void *private_data_data, bool success)
363 {
364         struct domain_request_state *state =
365                 talloc_get_type_abort(private_data_data, struct domain_request_state);
366
367         if (!success) {
368                 DEBUG(5, ("Domain init returned an error\n"));
369                 state->continuation(state->private_data_data, False);
370                 return;
371         }
372
373         async_request(state->mem_ctx, &state->domain->child,
374                       state->request, state->response,
375                       state->continuation, state->private_data_data);
376 }
377
378 static void recvfrom_child(void *private_data_data, bool success)
379 {
380         struct winbindd_cli_state *state =
381                 talloc_get_type_abort(private_data_data, struct winbindd_cli_state);
382         enum winbindd_result result = state->response.result;
383
384         /* This is an optimization: The child has written directly to the
385          * response buffer. The request itself is still in pending state,
386          * state that in the result code. */
387
388         state->response.result = WINBINDD_PENDING;
389
390         if ((!success) || (result != WINBINDD_OK)) {
391                 request_error(state);
392                 return;
393         }
394
395         request_ok(state);
396 }
397
398 void sendto_child(struct winbindd_cli_state *state,
399                   struct winbindd_child *child)
400 {
401         async_request(state->mem_ctx, child, &state->request,
402                       &state->response, recvfrom_child, state);
403 }
404
405 void sendto_domain(struct winbindd_cli_state *state,
406                    struct winbindd_domain *domain)
407 {
408         async_domain_request(state->mem_ctx, domain,
409                              &state->request, &state->response,
410                              recvfrom_child, state);
411 }
412
413 static void child_process_request(struct winbindd_child *child,
414                                   struct winbindd_cli_state *state)
415 {
416         struct winbindd_domain *domain = child->domain;
417         const struct winbindd_child_dispatch_table *table = child->table;
418
419         /* Free response data - we may be interrupted and receive another
420            command before being able to send this data off. */
421
422         state->response.result = WINBINDD_ERROR;
423         state->response.length = sizeof(struct winbindd_response);
424
425         /* as all requests in the child are sync, we can use talloc_tos() */
426         state->mem_ctx = talloc_tos();
427
428         /* Process command */
429
430         for (; table->name; table++) {
431                 if (state->request.cmd == table->struct_cmd) {
432                         DEBUG(10,("child_process_request: request fn %s\n",
433                                   table->name));
434                         state->response.result = table->struct_fn(domain, state);
435                         return;
436                 }
437         }
438
439         DEBUG(1 ,("child_process_request: unknown request fn number %d\n",
440                   (int)state->request.cmd));
441         state->response.result = WINBINDD_ERROR;
442 }
443
444 void setup_child(struct winbindd_child *child,
445                  const struct winbindd_child_dispatch_table *table,
446                  const char *logprefix,
447                  const char *logname)
448 {
449         if (logprefix && logname) {
450                 if (asprintf(&child->logfilename, "%s/%s-%s",
451                              get_dyn_LOGFILEBASE(), logprefix, logname) < 0) {
452                         smb_panic("Internal error: asprintf failed");
453                 }
454         } else {
455                 smb_panic("Internal error: logprefix == NULL && "
456                           "logname == NULL");
457         }
458
459         child->domain = NULL;
460         child->table = table;
461 }
462
463 struct winbindd_child *children = NULL;
464
465 void winbind_child_died(pid_t pid)
466 {
467         struct winbindd_child *child;
468
469         for (child = children; child != NULL; child = child->next) {
470                 if (child->pid == pid) {
471                         break;
472                 }
473         }
474
475         if (child == NULL) {
476                 DEBUG(5, ("Already reaped child %u died\n", (unsigned int)pid));
477                 return;
478         }
479
480         /* This will be re-added in fork_domain_child() */
481
482         DLIST_REMOVE(children, child);
483         
484         remove_fd_event(&child->event);
485         close(child->event.fd);
486         child->event.fd = 0;
487         child->event.flags = 0;
488         child->pid = 0;
489
490         schedule_async_request(child);
491 }
492
493 /* Ensure any negative cache entries with the netbios or realm names are removed. */
494
495 void winbindd_flush_negative_conn_cache(struct winbindd_domain *domain)
496 {
497         flush_negative_conn_cache_for_domain(domain->name);
498         if (*domain->alt_name) {
499                 flush_negative_conn_cache_for_domain(domain->alt_name);
500         }
501 }
502
503 /* 
504  * Parent winbindd process sets its own debug level first and then
505  * sends a message to all the winbindd children to adjust their debug
506  * level to that of parents.
507  */
508
509 void winbind_msg_debug(struct messaging_context *msg_ctx,
510                          void *private_data,
511                          uint32_t msg_type,
512                          struct server_id server_id,
513                          DATA_BLOB *data)
514 {
515         struct winbindd_child *child;
516
517         DEBUG(10,("winbind_msg_debug: got debug message.\n"));
518         
519         debug_message(msg_ctx, private_data, MSG_DEBUG, server_id, data);
520
521         for (child = children; child != NULL; child = child->next) {
522
523                 DEBUG(10,("winbind_msg_debug: sending message to pid %u.\n",
524                         (unsigned int)child->pid));
525
526                 messaging_send_buf(msg_ctx, pid_to_procid(child->pid),
527                            MSG_DEBUG,
528                            data->data,
529                            strlen((char *) data->data) + 1);
530         }
531 }
532
533 /* Set our domains as offline and forward the offline message to our children. */
534
535 void winbind_msg_offline(struct messaging_context *msg_ctx,
536                          void *private_data,
537                          uint32_t msg_type,
538                          struct server_id server_id,
539                          DATA_BLOB *data)
540 {
541         struct winbindd_child *child;
542         struct winbindd_domain *domain;
543
544         DEBUG(10,("winbind_msg_offline: got offline message.\n"));
545
546         if (!lp_winbind_offline_logon()) {
547                 DEBUG(10,("winbind_msg_offline: rejecting offline message.\n"));
548                 return;
549         }
550
551         /* Set our global state as offline. */
552         if (!set_global_winbindd_state_offline()) {
553                 DEBUG(10,("winbind_msg_offline: offline request failed.\n"));
554                 return;
555         }
556
557         /* Set all our domains as offline. */
558         for (domain = domain_list(); domain; domain = domain->next) {
559                 if (domain->internal) {
560                         continue;
561                 }
562                 DEBUG(5,("winbind_msg_offline: marking %s offline.\n", domain->name));
563                 set_domain_offline(domain);
564         }
565
566         for (child = children; child != NULL; child = child->next) {
567                 /* Don't send message to internal childs.  We've already
568                    done so above. */
569                 if (!child->domain || winbindd_internal_child(child)) {
570                         continue;
571                 }
572
573                 /* Or internal domains (this should not be possible....) */
574                 if (child->domain->internal) {
575                         continue;
576                 }
577
578                 /* Each winbindd child should only process requests for one domain - make sure
579                    we only set it online / offline for that domain. */
580
581                 DEBUG(10,("winbind_msg_offline: sending message to pid %u for domain %s.\n",
582                         (unsigned int)child->pid, domain->name ));
583
584                 messaging_send_buf(msg_ctx, pid_to_procid(child->pid),
585                                    MSG_WINBIND_OFFLINE,
586                                    (uint8 *)child->domain->name,
587                                    strlen(child->domain->name)+1);
588         }
589 }
590
591 /* Set our domains as online and forward the online message to our children. */
592
593 void winbind_msg_online(struct messaging_context *msg_ctx,
594                         void *private_data,
595                         uint32_t msg_type,
596                         struct server_id server_id,
597                         DATA_BLOB *data)
598 {
599         struct winbindd_child *child;
600         struct winbindd_domain *domain;
601
602         DEBUG(10,("winbind_msg_online: got online message.\n"));
603
604         if (!lp_winbind_offline_logon()) {
605                 DEBUG(10,("winbind_msg_online: rejecting online message.\n"));
606                 return;
607         }
608
609         /* Set our global state as online. */
610         set_global_winbindd_state_online();
611
612         smb_nscd_flush_user_cache();
613         smb_nscd_flush_group_cache();
614
615         /* Set all our domains as online. */
616         for (domain = domain_list(); domain; domain = domain->next) {
617                 if (domain->internal) {
618                         continue;
619                 }
620                 DEBUG(5,("winbind_msg_online: requesting %s to go online.\n", domain->name));
621
622                 winbindd_flush_negative_conn_cache(domain);
623                 set_domain_online_request(domain);
624
625                 /* Send an online message to the idmap child when our
626                    primary domain comes back online */
627
628                 if ( domain->primary ) {
629                         struct winbindd_child *idmap = idmap_child();
630                         
631                         if ( idmap->pid != 0 ) {
632                                 messaging_send_buf(msg_ctx,
633                                                    pid_to_procid(idmap->pid), 
634                                                    MSG_WINBIND_ONLINE,
635                                                    (uint8 *)domain->name,
636                                                    strlen(domain->name)+1);
637                         }
638                         
639                 }
640         }
641
642         for (child = children; child != NULL; child = child->next) {
643                 /* Don't send message to internal childs. */
644                 if (!child->domain || winbindd_internal_child(child)) {
645                         continue;
646                 }
647
648                 /* Or internal domains (this should not be possible....) */
649                 if (child->domain->internal) {
650                         continue;
651                 }
652
653                 /* Each winbindd child should only process requests for one domain - make sure
654                    we only set it online / offline for that domain. */
655
656                 DEBUG(10,("winbind_msg_online: sending message to pid %u for domain %s.\n",
657                         (unsigned int)child->pid, child->domain->name ));
658
659                 messaging_send_buf(msg_ctx, pid_to_procid(child->pid),
660                                    MSG_WINBIND_ONLINE,
661                                    (uint8 *)child->domain->name,
662                                    strlen(child->domain->name)+1);
663         }
664 }
665
666 /* Forward the online/offline messages to our children. */
667 void winbind_msg_onlinestatus(struct messaging_context *msg_ctx,
668                               void *private_data,
669                               uint32_t msg_type,
670                               struct server_id server_id,
671                               DATA_BLOB *data)
672 {
673         struct winbindd_child *child;
674
675         DEBUG(10,("winbind_msg_onlinestatus: got onlinestatus message.\n"));
676
677         for (child = children; child != NULL; child = child->next) {
678                 if (child->domain && child->domain->primary) {
679                         DEBUG(10,("winbind_msg_onlinestatus: "
680                                   "sending message to pid %u of primary domain.\n",
681                                   (unsigned int)child->pid));
682                         messaging_send_buf(msg_ctx, pid_to_procid(child->pid), 
683                                            MSG_WINBIND_ONLINESTATUS,
684                                            (uint8 *)data->data,
685                                            data->length);
686                         break;
687                 }
688         }
689 }
690
691 void winbind_msg_dump_event_list(struct messaging_context *msg_ctx,
692                                  void *private_data,
693                                  uint32_t msg_type,
694                                  struct server_id server_id,
695                                  DATA_BLOB *data)
696 {
697         struct winbindd_child *child;
698
699         DEBUG(10,("winbind_msg_dump_event_list received\n"));
700
701         dump_event_list(winbind_event_context());
702
703         for (child = children; child != NULL; child = child->next) {
704
705                 DEBUG(10,("winbind_msg_dump_event_list: sending message to pid %u\n",
706                         (unsigned int)child->pid));
707
708                 messaging_send_buf(msg_ctx, pid_to_procid(child->pid),
709                                    MSG_DUMP_EVENT_LIST,
710                                    NULL, 0);
711         }
712
713 }
714
715 void winbind_msg_dump_domain_list(struct messaging_context *msg_ctx,
716                                   void *private_data,
717                                   uint32_t msg_type,
718                                   struct server_id server_id,
719                                   DATA_BLOB *data)
720 {
721         TALLOC_CTX *mem_ctx;
722         const char *message = NULL;
723         struct server_id *sender = NULL;
724         const char *domain = NULL;
725         char *s = NULL;
726         NTSTATUS status;
727         struct winbindd_domain *dom = NULL;
728
729         DEBUG(5,("winbind_msg_dump_domain_list received.\n"));
730
731         if (!data || !data->data) {
732                 return;
733         }
734
735         if (data->length < sizeof(struct server_id)) {
736                 return;
737         }
738
739         mem_ctx = talloc_init("winbind_msg_dump_domain_list");
740         if (!mem_ctx) {
741                 return;
742         }
743
744         sender = (struct server_id *)data->data;
745         if (data->length > sizeof(struct server_id)) {
746                 domain = (const char *)data->data+sizeof(struct server_id);
747         }
748
749         if (domain) {
750
751                 DEBUG(5,("winbind_msg_dump_domain_list for domain: %s\n",
752                         domain));
753
754                 message = NDR_PRINT_STRUCT_STRING(mem_ctx, winbindd_domain,
755                                                   find_domain_from_name_noinit(domain));
756                 if (!message) {
757                         talloc_destroy(mem_ctx);
758                         return;
759                 }
760
761                 messaging_send_buf(msg_ctx, *sender,
762                                    MSG_WINBIND_DUMP_DOMAIN_LIST,
763                                    (uint8_t *)message, strlen(message) + 1);
764
765                 talloc_destroy(mem_ctx);
766
767                 return;
768         }
769
770         DEBUG(5,("winbind_msg_dump_domain_list all domains\n"));
771
772         for (dom = domain_list(); dom; dom=dom->next) {
773                 message = NDR_PRINT_STRUCT_STRING(mem_ctx, winbindd_domain, dom);
774                 if (!message) {
775                         talloc_destroy(mem_ctx);
776                         return;
777                 }
778
779                 s = talloc_asprintf_append(s, "%s\n", message);
780                 if (!s) {
781                         talloc_destroy(mem_ctx);
782                         return;
783                 }
784         }
785
786         status = messaging_send_buf(msg_ctx, *sender,
787                                     MSG_WINBIND_DUMP_DOMAIN_LIST,
788                                     (uint8_t *)s, strlen(s) + 1);
789         if (!NT_STATUS_IS_OK(status)) {
790                 DEBUG(0,("failed to send message: %s\n",
791                 nt_errstr(status)));
792         }
793
794         talloc_destroy(mem_ctx);
795 }
796
797 static void account_lockout_policy_handler(struct event_context *ctx,
798                                            struct timed_event *te,
799                                            const struct timeval *now,
800                                            void *private_data)
801 {
802         struct winbindd_child *child =
803                 (struct winbindd_child *)private_data;
804         TALLOC_CTX *mem_ctx = NULL;
805         struct winbindd_methods *methods;
806         struct samr_DomInfo12 lockout_policy;
807         NTSTATUS result;
808
809         DEBUG(10,("account_lockout_policy_handler called\n"));
810
811         TALLOC_FREE(child->lockout_policy_event);
812
813         if ( !winbindd_can_contact_domain( child->domain ) ) {
814                 DEBUG(10,("account_lockout_policy_handler: Removing myself since I "
815                           "do not have an incoming trust to domain %s\n", 
816                           child->domain->name));
817
818                 return;         
819         }
820
821         methods = child->domain->methods;
822
823         mem_ctx = talloc_init("account_lockout_policy_handler ctx");
824         if (!mem_ctx) {
825                 result = NT_STATUS_NO_MEMORY;
826         } else {
827                 result = methods->lockout_policy(child->domain, mem_ctx, &lockout_policy);
828         }
829         TALLOC_FREE(mem_ctx);
830
831         if (!NT_STATUS_IS_OK(result)) {
832                 DEBUG(10,("account_lockout_policy_handler: lockout_policy failed error %s\n",
833                          nt_errstr(result)));
834         }
835
836         child->lockout_policy_event = event_add_timed(winbind_event_context(), NULL,
837                                                       timeval_current_ofs(3600, 0),
838                                                       "account_lockout_policy_handler",
839                                                       account_lockout_policy_handler,
840                                                       child);
841 }
842
843 /* Deal with a request to go offline. */
844
845 static void child_msg_offline(struct messaging_context *msg,
846                               void *private_data,
847                               uint32_t msg_type,
848                               struct server_id server_id,
849                               DATA_BLOB *data)
850 {
851         struct winbindd_domain *domain;
852         const char *domainname = (const char *)data->data;
853
854         if (data->data == NULL || data->length == 0) {
855                 return;
856         }
857
858         DEBUG(5,("child_msg_offline received for domain %s.\n", domainname));
859
860         if (!lp_winbind_offline_logon()) {
861                 DEBUG(10,("child_msg_offline: rejecting offline message.\n"));
862                 return;
863         }
864
865         /* Mark the requested domain offline. */
866
867         for (domain = domain_list(); domain; domain = domain->next) {
868                 if (domain->internal) {
869                         continue;
870                 }
871                 if (strequal(domain->name, domainname)) {
872                         DEBUG(5,("child_msg_offline: marking %s offline.\n", domain->name));
873                         set_domain_offline(domain);
874                 }
875         }
876 }
877
878 /* Deal with a request to go online. */
879
880 static void child_msg_online(struct messaging_context *msg,
881                              void *private_data,
882                              uint32_t msg_type,
883                              struct server_id server_id,
884                              DATA_BLOB *data)
885 {
886         struct winbindd_domain *domain;
887         const char *domainname = (const char *)data->data;
888
889         if (data->data == NULL || data->length == 0) {
890                 return;
891         }
892
893         DEBUG(5,("child_msg_online received for domain %s.\n", domainname));
894
895         if (!lp_winbind_offline_logon()) {
896                 DEBUG(10,("child_msg_online: rejecting online message.\n"));
897                 return;
898         }
899
900         /* Set our global state as online. */
901         set_global_winbindd_state_online();
902
903         /* Try and mark everything online - delete any negative cache entries
904            to force a reconnect now. */
905
906         for (domain = domain_list(); domain; domain = domain->next) {
907                 if (domain->internal) {
908                         continue;
909                 }
910                 if (strequal(domain->name, domainname)) {
911                         DEBUG(5,("child_msg_online: requesting %s to go online.\n", domain->name));
912                         winbindd_flush_negative_conn_cache(domain);
913                         set_domain_online_request(domain);
914                 }
915         }
916 }
917
918 static const char *collect_onlinestatus(TALLOC_CTX *mem_ctx)
919 {
920         struct winbindd_domain *domain;
921         char *buf = NULL;
922
923         if ((buf = talloc_asprintf(mem_ctx, "global:%s ", 
924                                    get_global_winbindd_state_offline() ? 
925                                    "Offline":"Online")) == NULL) {
926                 return NULL;
927         }
928
929         for (domain = domain_list(); domain; domain = domain->next) {
930                 if ((buf = talloc_asprintf_append_buffer(buf, "%s:%s ", 
931                                                   domain->name, 
932                                                   domain->online ?
933                                                   "Online":"Offline")) == NULL) {
934                         return NULL;
935                 }
936         }
937
938         buf = talloc_asprintf_append_buffer(buf, "\n");
939
940         DEBUG(5,("collect_onlinestatus: %s", buf));
941
942         return buf;
943 }
944
945 static void child_msg_onlinestatus(struct messaging_context *msg_ctx,
946                                    void *private_data,
947                                    uint32_t msg_type,
948                                    struct server_id server_id,
949                                    DATA_BLOB *data)
950 {
951         TALLOC_CTX *mem_ctx;
952         const char *message;
953         struct server_id *sender;
954         
955         DEBUG(5,("winbind_msg_onlinestatus received.\n"));
956
957         if (!data->data) {
958                 return;
959         }
960
961         sender = (struct server_id *)data->data;
962
963         mem_ctx = talloc_init("winbind_msg_onlinestatus");
964         if (mem_ctx == NULL) {
965                 return;
966         }
967         
968         message = collect_onlinestatus(mem_ctx);
969         if (message == NULL) {
970                 talloc_destroy(mem_ctx);
971                 return;
972         }
973
974         messaging_send_buf(msg_ctx, *sender, MSG_WINBIND_ONLINESTATUS, 
975                            (uint8 *)message, strlen(message) + 1);
976
977         talloc_destroy(mem_ctx);
978 }
979
980 static void child_msg_dump_event_list(struct messaging_context *msg,
981                                       void *private_data,
982                                       uint32_t msg_type,
983                                       struct server_id server_id,
984                                       DATA_BLOB *data)
985 {
986         DEBUG(5,("child_msg_dump_event_list received\n"));
987
988         dump_event_list(winbind_event_context());
989 }
990
991
992 static bool fork_domain_child(struct winbindd_child *child)
993 {
994         int fdpair[2];
995         struct winbindd_cli_state state;
996         struct winbindd_domain *domain;
997         struct winbindd_domain *primary_domain = NULL;
998
999         if (child->domain) {
1000                 DEBUG(10, ("fork_domain_child called for domain '%s'\n",
1001                            child->domain->name));
1002         } else {
1003                 DEBUG(10, ("fork_domain_child called without domain.\n"));
1004         }
1005
1006         if (socketpair(AF_UNIX, SOCK_STREAM, 0, fdpair) != 0) {
1007                 DEBUG(0, ("Could not open child pipe: %s\n",
1008                           strerror(errno)));
1009                 return False;
1010         }
1011
1012         ZERO_STRUCT(state);
1013         state.pid = sys_getpid();
1014
1015         child->pid = sys_fork();
1016
1017         if (child->pid == -1) {
1018                 DEBUG(0, ("Could not fork: %s\n", strerror(errno)));
1019                 return False;
1020         }
1021
1022         if (child->pid != 0) {
1023                 /* Parent */
1024                 close(fdpair[0]);
1025                 child->next = child->prev = NULL;
1026                 DLIST_ADD(children, child);
1027                 child->event.fd = fdpair[1];
1028                 child->event.flags = 0;
1029                 child->requests = NULL;
1030                 add_fd_event(&child->event);
1031                 return True;
1032         }
1033
1034         /* Child */
1035
1036         /* Stop zombies in children */
1037         CatchChild();
1038
1039         state.sock = fdpair[0];
1040         close(fdpair[1]);
1041
1042         if (!reinit_after_fork(winbind_messaging_context(), true)) {
1043                 DEBUG(0,("reinit_after_fork() failed\n"));
1044                 _exit(0);
1045         }
1046
1047         close_conns_after_fork();
1048
1049         if (!override_logfile) {
1050                 lp_set_logfile(child->logfilename);
1051                 reopen_logs();
1052         }
1053
1054         /*
1055          * For clustering, we need to re-init our ctdbd connection after the
1056          * fork
1057          */
1058         if (!NT_STATUS_IS_OK(messaging_reinit(winbind_messaging_context())))
1059                 exit(1);
1060
1061         /* Don't handle the same messages as our parent. */
1062         messaging_deregister(winbind_messaging_context(),
1063                              MSG_SMB_CONF_UPDATED, NULL);
1064         messaging_deregister(winbind_messaging_context(),
1065                              MSG_SHUTDOWN, NULL);
1066         messaging_deregister(winbind_messaging_context(),
1067                              MSG_WINBIND_OFFLINE, NULL);
1068         messaging_deregister(winbind_messaging_context(),
1069                              MSG_WINBIND_ONLINE, NULL);
1070         messaging_deregister(winbind_messaging_context(),
1071                              MSG_WINBIND_ONLINESTATUS, NULL);
1072         messaging_deregister(winbind_messaging_context(),
1073                              MSG_DUMP_EVENT_LIST, NULL);
1074         messaging_deregister(winbind_messaging_context(),
1075                              MSG_WINBIND_DUMP_DOMAIN_LIST, NULL);
1076         messaging_deregister(winbind_messaging_context(),
1077                              MSG_DEBUG, NULL);
1078
1079         /* Handle online/offline messages. */
1080         messaging_register(winbind_messaging_context(), NULL,
1081                            MSG_WINBIND_OFFLINE, child_msg_offline);
1082         messaging_register(winbind_messaging_context(), NULL,
1083                            MSG_WINBIND_ONLINE, child_msg_online);
1084         messaging_register(winbind_messaging_context(), NULL,
1085                            MSG_WINBIND_ONLINESTATUS, child_msg_onlinestatus);
1086         messaging_register(winbind_messaging_context(), NULL,
1087                            MSG_DUMP_EVENT_LIST, child_msg_dump_event_list);
1088         messaging_register(winbind_messaging_context(), NULL,
1089                            MSG_DEBUG, debug_message);
1090
1091         if ( child->domain ) {
1092                 child->domain->startup = True;
1093                 child->domain->startup_time = time(NULL);
1094         }
1095
1096         /* Ensure we have no pending check_online events other
1097            than one for this domain or the primary domain. */
1098
1099         for (domain = domain_list(); domain; domain = domain->next) {
1100                 if (domain->primary) {
1101                         primary_domain = domain;
1102                 }
1103                 if ((domain != child->domain) && !domain->primary) {
1104                         TALLOC_FREE(domain->check_online_event);
1105                 }
1106         }
1107
1108         /* Ensure we're not handling an event inherited from
1109            our parent. */
1110
1111         cancel_named_event(winbind_event_context(),
1112                            "krb5_ticket_refresh_handler");
1113
1114         /* We might be in the idmap child...*/
1115         if (child->domain && !(child->domain->internal) &&
1116             lp_winbind_offline_logon()) {
1117
1118                 set_domain_online_request(child->domain);
1119
1120                 if (primary_domain != child->domain) {
1121                         /* We need to talk to the primary
1122                          * domain as well as the trusted
1123                          * domain inside a trusted domain
1124                          * child.
1125                          * See the code in :
1126                          * set_dc_type_and_flags_trustinfo()
1127                          * for details.
1128                          */
1129                         set_domain_online_request(primary_domain);
1130                 }
1131
1132                 child->lockout_policy_event = event_add_timed(
1133                         winbind_event_context(), NULL, timeval_zero(),
1134                         "account_lockout_policy_handler",
1135                         account_lockout_policy_handler,
1136                         child);
1137         }
1138
1139         while (1) {
1140
1141                 int ret;
1142                 fd_set read_fds;
1143                 struct timeval t;
1144                 struct timeval *tp;
1145                 struct timeval now;
1146                 TALLOC_CTX *frame = talloc_stackframe();
1147
1148                 /* check for signals */
1149                 winbind_check_sigterm(false);
1150                 winbind_check_sighup(override_logfile ? NULL :
1151                                 child->logfilename);
1152
1153                 run_events(winbind_event_context(), 0, NULL, NULL);
1154
1155                 GetTimeOfDay(&now);
1156
1157                 if (child->domain && child->domain->startup &&
1158                                 (now.tv_sec > child->domain->startup_time + 30)) {
1159                         /* No longer in "startup" mode. */
1160                         DEBUG(10,("fork_domain_child: domain %s no longer in 'startup' mode.\n",
1161                                 child->domain->name ));
1162                         child->domain->startup = False;
1163                 }
1164
1165                 tp = get_timed_events_timeout(winbind_event_context(), &t);
1166                 if (tp) {
1167                         DEBUG(11,("select will use timeout of %u.%u seconds\n",
1168                                 (unsigned int)tp->tv_sec, (unsigned int)tp->tv_usec ));
1169                 }
1170
1171                 /* Handle messages */
1172
1173                 message_dispatch(winbind_messaging_context());
1174
1175                 FD_ZERO(&read_fds);
1176                 FD_SET(state.sock, &read_fds);
1177
1178                 ret = sys_select(state.sock + 1, &read_fds, NULL, NULL, tp);
1179
1180                 if (ret == 0) {
1181                         DEBUG(11,("nothing is ready yet, continue\n"));
1182                         TALLOC_FREE(frame);
1183                         continue;
1184                 }
1185
1186                 if (ret == -1 && errno == EINTR) {
1187                         /* We got a signal - continue. */
1188                         TALLOC_FREE(frame);
1189                         continue;
1190                 }
1191
1192                 if (ret == -1 && errno != EINTR) {
1193                         DEBUG(0,("select error occured\n"));
1194                         TALLOC_FREE(frame);
1195                         perror("select");
1196                         return False;
1197                 }
1198
1199                 /* fetch a request from the main daemon */
1200                 child_read_request(&state);
1201
1202                 if (state.finished) {
1203                         /* we lost contact with our parent */
1204                         exit(0);
1205                 }
1206
1207                 DEBUG(4,("child daemon request %d\n", (int)state.request.cmd));
1208
1209                 ZERO_STRUCT(state.response);
1210                 state.request.null_term = '\0';
1211                 child_process_request(child, &state);
1212
1213                 SAFE_FREE(state.request.extra_data.data);
1214
1215                 cache_store_response(sys_getpid(), &state.response);
1216
1217                 SAFE_FREE(state.response.extra_data.data);
1218
1219                 /* We just send the result code back, the result
1220                  * structure needs to be fetched via the
1221                  * winbindd_cache. Hmm. That needs fixing... */
1222
1223                 if (write_data(state.sock,
1224                                (const char *)&state.response.result,
1225                                sizeof(state.response.result)) !=
1226                     sizeof(state.response.result)) {
1227                         DEBUG(0, ("Could not write result\n"));
1228                         exit(1);
1229                 }
1230                 TALLOC_FREE(frame);
1231         }
1232 }