Add winbindd_reinit_after_fork(), cleaning out all possible events
[jra/samba/.git] / source3 / 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         DEBUG(10, ("Sending request to child pid %d (domain=%s)\n",
124                    (int)child->pid,
125                    (child->domain != NULL) ? child->domain->name : "''"));
126
127         state = TALLOC_P(mem_ctx, struct winbindd_async_request);
128
129         if (state == NULL) {
130                 DEBUG(0, ("talloc failed\n"));
131                 continuation(private_data, False);
132                 return;
133         }
134
135         state->mem_ctx = mem_ctx;
136         state->child = child;
137         state->reply_timeout_event = NULL;
138         state->request = request;
139         state->response = response;
140         state->continuation = continuation;
141         state->private_data = private_data;
142
143         DLIST_ADD_END(child->requests, state, struct winbindd_async_request *);
144
145         schedule_async_request(child);
146
147         return;
148 }
149
150 static void async_main_request_sent(void *private_data, bool success)
151 {
152         struct winbindd_async_request *state =
153                 talloc_get_type_abort(private_data, struct winbindd_async_request);
154
155         if (!success) {
156                 DEBUG(5, ("Could not send async request\n"));
157                 async_request_fail(state);
158                 return;
159         }
160
161         if (state->request->extra_len == 0) {
162                 async_request_sent(private_data, True);
163                 return;
164         }
165
166         setup_async_write(&state->child->event, state->request->extra_data.data,
167                           state->request->extra_len,
168                           async_request_sent, state);
169 }
170
171 /****************************************************************
172  Handler triggered if the child winbindd doesn't respond within
173  a given timeout.
174 ****************************************************************/
175
176 static void async_request_timeout_handler(struct event_context *ctx,
177                                         struct timed_event *te,
178                                         struct timeval now,
179                                         void *private_data)
180 {
181         struct winbindd_async_request *state =
182                 talloc_get_type_abort(private_data, struct winbindd_async_request);
183
184         DEBUG(0,("async_request_timeout_handler: child pid %u is not responding. "
185                 "Closing connection to it.\n",
186                 state->child_pid ));
187
188         /* Deal with the reply - set to error. */
189         async_reply_recv(private_data, False);
190 }
191
192 /**************************************************************
193  Common function called on both async send and recv fail.
194  Cleans up the child and schedules the next request.
195 **************************************************************/
196
197 static void async_request_fail(struct winbindd_async_request *state)
198 {
199         DLIST_REMOVE(state->child->requests, state);
200
201         TALLOC_FREE(state->reply_timeout_event);
202
203         /* If child exists and is not already reaped,
204            send kill signal to child. */
205
206         if ((state->child->pid != (pid_t)0) &&
207                         (state->child->pid != (pid_t)-1) &&
208                         (state->child->pid == state->child_pid)) {
209                 kill(state->child_pid, SIGTERM);
210
211                 /* 
212                  * Close the socket to the child.
213                  */
214                 winbind_child_died(state->child_pid);
215         }
216
217         state->response->length = sizeof(struct winbindd_response);
218         state->response->result = WINBINDD_ERROR;
219         state->continuation(state->private_data, False);
220 }
221
222 static void async_request_sent(void *private_data_data, bool success)
223 {
224         struct winbindd_async_request *state =
225                 talloc_get_type_abort(private_data_data, struct winbindd_async_request);
226
227         if (!success) {
228                 DEBUG(5, ("Could not send async request to child pid %u\n",
229                         (unsigned int)state->child_pid ));
230                 async_request_fail(state);
231                 return;
232         }
233
234         /* Request successfully sent to the child, setup the wait for reply */
235
236         setup_async_read(&state->child->event,
237                          &state->response->result,
238                          sizeof(state->response->result),
239                          async_reply_recv, state);
240
241         /* 
242          * Set up a timeout of 300 seconds for the response.
243          * If we don't get it close the child socket and
244          * report failure.
245          */
246
247         state->reply_timeout_event = event_add_timed(winbind_event_context(),
248                                                         NULL,
249                                                         timeval_current_ofs(300,0),
250                                                         async_request_timeout_handler,
251                                                         state);
252         if (!state->reply_timeout_event) {
253                 smb_panic("async_request_sent: failed to add timeout handler.\n");
254         }
255 }
256
257 static void async_reply_recv(void *private_data, bool success)
258 {
259         struct winbindd_async_request *state =
260                 talloc_get_type_abort(private_data, struct winbindd_async_request);
261         struct winbindd_child *child = state->child;
262
263         TALLOC_FREE(state->reply_timeout_event);
264
265         state->response->length = sizeof(struct winbindd_response);
266
267         if (!success) {
268                 DEBUG(5, ("Could not receive async reply from child pid %u\n",
269                         (unsigned int)state->child_pid ));
270
271                 cache_cleanup_response(state->child_pid);
272                 async_request_fail(state);
273                 return;
274         }
275
276         SMB_ASSERT(cache_retrieve_response(state->child_pid,
277                                            state->response));
278
279         cache_cleanup_response(state->child_pid);
280         
281         DLIST_REMOVE(child->requests, state);
282
283         schedule_async_request(child);
284
285         state->continuation(state->private_data, True);
286 }
287
288 static bool fork_domain_child(struct winbindd_child *child);
289
290 static void schedule_async_request(struct winbindd_child *child)
291 {
292         struct winbindd_async_request *request = child->requests;
293
294         if (request == NULL) {
295                 return;
296         }
297
298         if (child->event.flags != 0) {
299                 return;         /* Busy */
300         }
301
302         /*
303          * This may be a reschedule, so we might
304          * have an existing timeout event pending on
305          * the first entry in the child->requests list
306          * (we only send one request at a time).
307          * Ensure we free it before we reschedule.
308          * Bug #5814, from hargagan <shargagan@novell.com>.
309          * JRA.
310          */
311
312         TALLOC_FREE(request->reply_timeout_event);
313
314         if ((child->pid == 0) && (!fork_domain_child(child))) {
315                 /* fork_domain_child failed.
316                    Cancel all outstanding requests */
317
318                 while (request != NULL) {
319                         /* request might be free'd in the continuation */
320                         struct winbindd_async_request *next = request->next;
321
322                         async_request_fail(request);
323                         request = next;
324                 }
325                 return;
326         }
327
328         /* Now we know who we're sending to - remember the pid. */
329         request->child_pid = child->pid;
330
331         setup_async_write(&child->event, request->request,
332                           sizeof(*request->request),
333                           async_main_request_sent, request);
334
335         return;
336 }
337
338 struct domain_request_state {
339         TALLOC_CTX *mem_ctx;
340         struct winbindd_domain *domain;
341         struct winbindd_request *request;
342         struct winbindd_response *response;
343         void (*continuation)(void *private_data_data, bool success);
344         void *private_data_data;
345 };
346
347 static void domain_init_recv(void *private_data_data, bool success);
348
349 void async_domain_request(TALLOC_CTX *mem_ctx,
350                           struct winbindd_domain *domain,
351                           struct winbindd_request *request,
352                           struct winbindd_response *response,
353                           void (*continuation)(void *private_data_data, bool success),
354                           void *private_data_data)
355 {
356         struct domain_request_state *state;
357
358         if (domain->initialized) {
359                 async_request(mem_ctx, &domain->child, request, response,
360                               continuation, private_data_data);
361                 return;
362         }
363
364         state = TALLOC_P(mem_ctx, struct domain_request_state);
365         if (state == NULL) {
366                 DEBUG(0, ("talloc failed\n"));
367                 continuation(private_data_data, False);
368                 return;
369         }
370
371         state->mem_ctx = mem_ctx;
372         state->domain = domain;
373         state->request = request;
374         state->response = response;
375         state->continuation = continuation;
376         state->private_data_data = private_data_data;
377
378         init_child_connection(domain, domain_init_recv, state);
379 }
380
381 static void domain_init_recv(void *private_data_data, bool success)
382 {
383         struct domain_request_state *state =
384                 talloc_get_type_abort(private_data_data, struct domain_request_state);
385
386         if (!success) {
387                 DEBUG(5, ("Domain init returned an error\n"));
388                 state->continuation(state->private_data_data, False);
389                 return;
390         }
391
392         async_request(state->mem_ctx, &state->domain->child,
393                       state->request, state->response,
394                       state->continuation, state->private_data_data);
395 }
396
397 static void recvfrom_child(void *private_data_data, bool success)
398 {
399         struct winbindd_cli_state *state =
400                 talloc_get_type_abort(private_data_data, struct winbindd_cli_state);
401         enum winbindd_result result = state->response.result;
402
403         /* This is an optimization: The child has written directly to the
404          * response buffer. The request itself is still in pending state,
405          * state that in the result code. */
406
407         state->response.result = WINBINDD_PENDING;
408
409         if ((!success) || (result != WINBINDD_OK)) {
410                 request_error(state);
411                 return;
412         }
413
414         request_ok(state);
415 }
416
417 void sendto_child(struct winbindd_cli_state *state,
418                   struct winbindd_child *child)
419 {
420         async_request(state->mem_ctx, child, &state->request,
421                       &state->response, recvfrom_child, state);
422 }
423
424 void sendto_domain(struct winbindd_cli_state *state,
425                    struct winbindd_domain *domain)
426 {
427         async_domain_request(state->mem_ctx, domain,
428                              &state->request, &state->response,
429                              recvfrom_child, state);
430 }
431
432 static void child_process_request(struct winbindd_child *child,
433                                   struct winbindd_cli_state *state)
434 {
435         struct winbindd_domain *domain = child->domain;
436         const struct winbindd_child_dispatch_table *table = child->table;
437
438         /* Free response data - we may be interrupted and receive another
439            command before being able to send this data off. */
440
441         state->response.result = WINBINDD_ERROR;
442         state->response.length = sizeof(struct winbindd_response);
443
444         /* as all requests in the child are sync, we can use talloc_tos() */
445         state->mem_ctx = talloc_tos();
446
447         /* Process command */
448
449         for (; table->name; table++) {
450                 if (state->request.cmd == table->struct_cmd) {
451                         DEBUG(10,("child_process_request: request fn %s\n",
452                                   table->name));
453                         state->response.result = table->struct_fn(domain, state);
454                         return;
455                 }
456         }
457
458         DEBUG(1 ,("child_process_request: unknown request fn number %d\n",
459                   (int)state->request.cmd));
460         state->response.result = WINBINDD_ERROR;
461 }
462
463 void setup_child(struct winbindd_child *child,
464                  const struct winbindd_child_dispatch_table *table,
465                  const char *logprefix,
466                  const char *logname)
467 {
468         if (logprefix && logname) {
469                 if (asprintf(&child->logfilename, "%s/%s-%s",
470                              get_dyn_LOGFILEBASE(), logprefix, logname) < 0) {
471                         smb_panic("Internal error: asprintf failed");
472                 }
473         } else {
474                 smb_panic("Internal error: logprefix == NULL && "
475                           "logname == NULL");
476         }
477
478         child->domain = NULL;
479         child->table = table;
480 }
481
482 struct winbindd_child *children = NULL;
483
484 void winbind_child_died(pid_t pid)
485 {
486         struct winbindd_child *child;
487
488         for (child = children; child != NULL; child = child->next) {
489                 if (child->pid == pid) {
490                         break;
491                 }
492         }
493
494         if (child == NULL) {
495                 DEBUG(5, ("Already reaped child %u died\n", (unsigned int)pid));
496                 return;
497         }
498
499         /* This will be re-added in fork_domain_child() */
500
501         DLIST_REMOVE(children, child);
502         
503         remove_fd_event(&child->event);
504         close(child->event.fd);
505         child->event.fd = 0;
506         child->event.flags = 0;
507         child->pid = 0;
508
509         if (child->requests) {
510                 /*
511                  * schedule_async_request() will also
512                  * clear this event but the call is
513                  * idempotent so it doesn't hurt to
514                  * cover all possible future code
515                  * paths. JRA.
516                  */
517                 TALLOC_FREE(child->requests->reply_timeout_event);
518         }
519
520         schedule_async_request(child);
521 }
522
523 /* Ensure any negative cache entries with the netbios or realm names are removed. */
524
525 void winbindd_flush_negative_conn_cache(struct winbindd_domain *domain)
526 {
527         flush_negative_conn_cache_for_domain(domain->name);
528         if (*domain->alt_name) {
529                 flush_negative_conn_cache_for_domain(domain->alt_name);
530         }
531 }
532
533 /* 
534  * Parent winbindd process sets its own debug level first and then
535  * sends a message to all the winbindd children to adjust their debug
536  * level to that of parents.
537  */
538
539 void winbind_msg_debug(struct messaging_context *msg_ctx,
540                          void *private_data,
541                          uint32_t msg_type,
542                          struct server_id server_id,
543                          DATA_BLOB *data)
544 {
545         struct winbindd_child *child;
546
547         DEBUG(10,("winbind_msg_debug: got debug message.\n"));
548         
549         debug_message(msg_ctx, private_data, MSG_DEBUG, server_id, data);
550
551         for (child = children; child != NULL; child = child->next) {
552
553                 DEBUG(10,("winbind_msg_debug: sending message to pid %u.\n",
554                         (unsigned int)child->pid));
555
556                 messaging_send_buf(msg_ctx, pid_to_procid(child->pid),
557                            MSG_DEBUG,
558                            data->data,
559                            strlen((char *) data->data) + 1);
560         }
561 }
562
563 /* Set our domains as offline and forward the offline message to our children. */
564
565 void winbind_msg_offline(struct messaging_context *msg_ctx,
566                          void *private_data,
567                          uint32_t msg_type,
568                          struct server_id server_id,
569                          DATA_BLOB *data)
570 {
571         struct winbindd_child *child;
572         struct winbindd_domain *domain;
573
574         DEBUG(10,("winbind_msg_offline: got offline message.\n"));
575
576         if (!lp_winbind_offline_logon()) {
577                 DEBUG(10,("winbind_msg_offline: rejecting offline message.\n"));
578                 return;
579         }
580
581         /* Set our global state as offline. */
582         if (!set_global_winbindd_state_offline()) {
583                 DEBUG(10,("winbind_msg_offline: offline request failed.\n"));
584                 return;
585         }
586
587         /* Set all our domains as offline. */
588         for (domain = domain_list(); domain; domain = domain->next) {
589                 if (domain->internal) {
590                         continue;
591                 }
592                 DEBUG(5,("winbind_msg_offline: marking %s offline.\n", domain->name));
593                 set_domain_offline(domain);
594         }
595
596         for (child = children; child != NULL; child = child->next) {
597                 /* Don't send message to internal childs.  We've already
598                    done so above. */
599                 if (!child->domain || winbindd_internal_child(child)) {
600                         continue;
601                 }
602
603                 /* Or internal domains (this should not be possible....) */
604                 if (child->domain->internal) {
605                         continue;
606                 }
607
608                 /* Each winbindd child should only process requests for one domain - make sure
609                    we only set it online / offline for that domain. */
610
611                 DEBUG(10,("winbind_msg_offline: sending message to pid %u for domain %s.\n",
612                         (unsigned int)child->pid, domain->name ));
613
614                 messaging_send_buf(msg_ctx, pid_to_procid(child->pid),
615                                    MSG_WINBIND_OFFLINE,
616                                    (uint8 *)child->domain->name,
617                                    strlen(child->domain->name)+1);
618         }
619 }
620
621 /* Set our domains as online and forward the online message to our children. */
622
623 void winbind_msg_online(struct messaging_context *msg_ctx,
624                         void *private_data,
625                         uint32_t msg_type,
626                         struct server_id server_id,
627                         DATA_BLOB *data)
628 {
629         struct winbindd_child *child;
630         struct winbindd_domain *domain;
631
632         DEBUG(10,("winbind_msg_online: got online message.\n"));
633
634         if (!lp_winbind_offline_logon()) {
635                 DEBUG(10,("winbind_msg_online: rejecting online message.\n"));
636                 return;
637         }
638
639         /* Set our global state as online. */
640         set_global_winbindd_state_online();
641
642         smb_nscd_flush_user_cache();
643         smb_nscd_flush_group_cache();
644
645         /* Set all our domains as online. */
646         for (domain = domain_list(); domain; domain = domain->next) {
647                 if (domain->internal) {
648                         continue;
649                 }
650                 DEBUG(5,("winbind_msg_online: requesting %s to go online.\n", domain->name));
651
652                 winbindd_flush_negative_conn_cache(domain);
653                 set_domain_online_request(domain);
654
655                 /* Send an online message to the idmap child when our
656                    primary domain comes back online */
657
658                 if ( domain->primary ) {
659                         struct winbindd_child *idmap = idmap_child();
660                         
661                         if ( idmap->pid != 0 ) {
662                                 messaging_send_buf(msg_ctx,
663                                                    pid_to_procid(idmap->pid), 
664                                                    MSG_WINBIND_ONLINE,
665                                                    (uint8 *)domain->name,
666                                                    strlen(domain->name)+1);
667                         }
668                         
669                 }
670         }
671
672         for (child = children; child != NULL; child = child->next) {
673                 /* Don't send message to internal childs. */
674                 if (!child->domain || winbindd_internal_child(child)) {
675                         continue;
676                 }
677
678                 /* Or internal domains (this should not be possible....) */
679                 if (child->domain->internal) {
680                         continue;
681                 }
682
683                 /* Each winbindd child should only process requests for one domain - make sure
684                    we only set it online / offline for that domain. */
685
686                 DEBUG(10,("winbind_msg_online: sending message to pid %u for domain %s.\n",
687                         (unsigned int)child->pid, child->domain->name ));
688
689                 messaging_send_buf(msg_ctx, pid_to_procid(child->pid),
690                                    MSG_WINBIND_ONLINE,
691                                    (uint8 *)child->domain->name,
692                                    strlen(child->domain->name)+1);
693         }
694 }
695
696 /* Forward the online/offline messages to our children. */
697 void winbind_msg_onlinestatus(struct messaging_context *msg_ctx,
698                               void *private_data,
699                               uint32_t msg_type,
700                               struct server_id server_id,
701                               DATA_BLOB *data)
702 {
703         struct winbindd_child *child;
704
705         DEBUG(10,("winbind_msg_onlinestatus: got onlinestatus message.\n"));
706
707         for (child = children; child != NULL; child = child->next) {
708                 if (child->domain && child->domain->primary) {
709                         DEBUG(10,("winbind_msg_onlinestatus: "
710                                   "sending message to pid %u of primary domain.\n",
711                                   (unsigned int)child->pid));
712                         messaging_send_buf(msg_ctx, pid_to_procid(child->pid), 
713                                            MSG_WINBIND_ONLINESTATUS,
714                                            (uint8 *)data->data,
715                                            data->length);
716                         break;
717                 }
718         }
719 }
720
721 void winbind_msg_dump_event_list(struct messaging_context *msg_ctx,
722                                  void *private_data,
723                                  uint32_t msg_type,
724                                  struct server_id server_id,
725                                  DATA_BLOB *data)
726 {
727         struct winbindd_child *child;
728
729         DEBUG(10,("winbind_msg_dump_event_list received\n"));
730
731         dump_event_list(winbind_event_context());
732
733         for (child = children; child != NULL; child = child->next) {
734
735                 DEBUG(10,("winbind_msg_dump_event_list: sending message to pid %u\n",
736                         (unsigned int)child->pid));
737
738                 messaging_send_buf(msg_ctx, pid_to_procid(child->pid),
739                                    MSG_DUMP_EVENT_LIST,
740                                    NULL, 0);
741         }
742
743 }
744
745 void winbind_msg_dump_domain_list(struct messaging_context *msg_ctx,
746                                   void *private_data,
747                                   uint32_t msg_type,
748                                   struct server_id server_id,
749                                   DATA_BLOB *data)
750 {
751         TALLOC_CTX *mem_ctx;
752         const char *message = NULL;
753         struct server_id *sender = NULL;
754         const char *domain = NULL;
755         char *s = NULL;
756         NTSTATUS status;
757         struct winbindd_domain *dom = NULL;
758
759         DEBUG(5,("winbind_msg_dump_domain_list received.\n"));
760
761         if (!data || !data->data) {
762                 return;
763         }
764
765         if (data->length < sizeof(struct server_id)) {
766                 return;
767         }
768
769         mem_ctx = talloc_init("winbind_msg_dump_domain_list");
770         if (!mem_ctx) {
771                 return;
772         }
773
774         sender = (struct server_id *)data->data;
775         if (data->length > sizeof(struct server_id)) {
776                 domain = (const char *)data->data+sizeof(struct server_id);
777         }
778
779         if (domain) {
780
781                 DEBUG(5,("winbind_msg_dump_domain_list for domain: %s\n",
782                         domain));
783
784                 message = NDR_PRINT_STRUCT_STRING(mem_ctx, winbindd_domain,
785                                                   find_domain_from_name_noinit(domain));
786                 if (!message) {
787                         talloc_destroy(mem_ctx);
788                         return;
789                 }
790
791                 messaging_send_buf(msg_ctx, *sender,
792                                    MSG_WINBIND_DUMP_DOMAIN_LIST,
793                                    (uint8_t *)message, strlen(message) + 1);
794
795                 talloc_destroy(mem_ctx);
796
797                 return;
798         }
799
800         DEBUG(5,("winbind_msg_dump_domain_list all domains\n"));
801
802         for (dom = domain_list(); dom; dom=dom->next) {
803                 message = NDR_PRINT_STRUCT_STRING(mem_ctx, winbindd_domain, dom);
804                 if (!message) {
805                         talloc_destroy(mem_ctx);
806                         return;
807                 }
808
809                 s = talloc_asprintf_append(s, "%s\n", message);
810                 if (!s) {
811                         talloc_destroy(mem_ctx);
812                         return;
813                 }
814         }
815
816         status = messaging_send_buf(msg_ctx, *sender,
817                                     MSG_WINBIND_DUMP_DOMAIN_LIST,
818                                     (uint8_t *)s, strlen(s) + 1);
819         if (!NT_STATUS_IS_OK(status)) {
820                 DEBUG(0,("failed to send message: %s\n",
821                 nt_errstr(status)));
822         }
823
824         talloc_destroy(mem_ctx);
825 }
826
827 static void account_lockout_policy_handler(struct event_context *ctx,
828                                            struct timed_event *te,
829                                            struct timeval now,
830                                            void *private_data)
831 {
832         struct winbindd_child *child =
833                 (struct winbindd_child *)private_data;
834         TALLOC_CTX *mem_ctx = NULL;
835         struct winbindd_methods *methods;
836         struct samr_DomInfo12 lockout_policy;
837         NTSTATUS result;
838
839         DEBUG(10,("account_lockout_policy_handler called\n"));
840
841         TALLOC_FREE(child->lockout_policy_event);
842
843         if ( !winbindd_can_contact_domain( child->domain ) ) {
844                 DEBUG(10,("account_lockout_policy_handler: Removing myself since I "
845                           "do not have an incoming trust to domain %s\n", 
846                           child->domain->name));
847
848                 return;         
849         }
850
851         methods = child->domain->methods;
852
853         mem_ctx = talloc_init("account_lockout_policy_handler ctx");
854         if (!mem_ctx) {
855                 result = NT_STATUS_NO_MEMORY;
856         } else {
857                 result = methods->lockout_policy(child->domain, mem_ctx, &lockout_policy);
858         }
859         TALLOC_FREE(mem_ctx);
860
861         if (!NT_STATUS_IS_OK(result)) {
862                 DEBUG(10,("account_lockout_policy_handler: lockout_policy failed error %s\n",
863                          nt_errstr(result)));
864         }
865
866         child->lockout_policy_event = event_add_timed(winbind_event_context(), NULL,
867                                                       timeval_current_ofs(3600, 0),
868                                                       account_lockout_policy_handler,
869                                                       child);
870 }
871
872 static time_t get_machine_password_timeout(void)
873 {
874         /* until we have gpo support use lp setting */
875         return lp_machine_password_timeout();
876 }
877
878 static bool calculate_next_machine_pwd_change(const char *domain,
879                                               struct timeval *t)
880 {
881         time_t pass_last_set_time;
882         time_t timeout;
883         time_t next_change;
884         char *pw;
885
886         pw = secrets_fetch_machine_password(domain,
887                                             &pass_last_set_time,
888                                             NULL);
889
890         if (pw == NULL) {
891                 DEBUG(0,("cannot fetch own machine password ????"));
892                 return false;
893         }
894
895         SAFE_FREE(pw);
896
897         timeout = get_machine_password_timeout();
898         if (timeout == 0) {
899                 DEBUG(10,("machine password never expires\n"));
900                 return false;
901         }
902
903         if (time(NULL) < (pass_last_set_time + timeout)) {
904                 next_change = pass_last_set_time + timeout;
905                 DEBUG(10,("machine password still valid until: %s\n",
906                         http_timestring(talloc_tos(), next_change)));
907                 *t = timeval_set(next_change, 0);
908                 return true;
909         }
910
911         DEBUG(10,("machine password expired, needs immediate change\n"));
912
913         *t = timeval_zero();
914
915         return true;
916 }
917
918 static void machine_password_change_handler(struct event_context *ctx,
919                                             struct timed_event *te,
920                                             struct timeval now,
921                                             void *private_data)
922 {
923         struct winbindd_child *child =
924                 (struct winbindd_child *)private_data;
925         struct rpc_pipe_client *netlogon_pipe = NULL;
926         TALLOC_CTX *frame;
927         NTSTATUS result;
928         struct timeval next_change;
929
930         DEBUG(10,("machine_password_change_handler called\n"));
931
932         TALLOC_FREE(child->machine_password_change_event);
933
934         if (!calculate_next_machine_pwd_change(child->domain->name,
935                                                &next_change)) {
936                 return;
937         }
938
939         if (!winbindd_can_contact_domain(child->domain)) {
940                 DEBUG(10,("machine_password_change_handler: Removing myself since I "
941                           "do not have an incoming trust to domain %s\n",
942                           child->domain->name));
943                 return;
944         }
945
946         result = cm_connect_netlogon(child->domain, &netlogon_pipe);
947         if (!NT_STATUS_IS_OK(result)) {
948                 DEBUG(10,("machine_password_change_handler: "
949                         "failed to connect netlogon pipe: %s\n",
950                          nt_errstr(result)));
951                 return;
952         }
953
954         frame = talloc_stackframe();
955
956         result = trust_pw_find_change_and_store_it(netlogon_pipe,
957                                                    frame,
958                                                    child->domain->name);
959         TALLOC_FREE(frame);
960
961         if (!NT_STATUS_IS_OK(result)) {
962                 DEBUG(10,("machine_password_change_handler: "
963                         "failed to change machine password: %s\n",
964                          nt_errstr(result)));
965         } else {
966                 DEBUG(10,("machine_password_change_handler: "
967                         "successfully changed machine password\n"));
968         }
969
970         child->machine_password_change_event = event_add_timed(winbind_event_context(), NULL,
971                                                               next_change,
972                                                               machine_password_change_handler,
973                                                               child);
974 }
975
976 /* Deal with a request to go offline. */
977
978 static void child_msg_offline(struct messaging_context *msg,
979                               void *private_data,
980                               uint32_t msg_type,
981                               struct server_id server_id,
982                               DATA_BLOB *data)
983 {
984         struct winbindd_domain *domain;
985         struct winbindd_domain *primary_domain = NULL;
986         const char *domainname = (const char *)data->data;
987
988         if (data->data == NULL || data->length == 0) {
989                 return;
990         }
991
992         DEBUG(5,("child_msg_offline received for domain %s.\n", domainname));
993
994         if (!lp_winbind_offline_logon()) {
995                 DEBUG(10,("child_msg_offline: rejecting offline message.\n"));
996                 return;
997         }
998
999         primary_domain = find_our_domain();
1000
1001         /* Mark the requested domain offline. */
1002
1003         for (domain = domain_list(); domain; domain = domain->next) {
1004                 if (domain->internal) {
1005                         continue;
1006                 }
1007                 if (strequal(domain->name, domainname)) {
1008                         DEBUG(5,("child_msg_offline: marking %s offline.\n", domain->name));
1009                         set_domain_offline(domain);
1010                         /* we are in the trusted domain, set the primary domain 
1011                          * offline too */
1012                         if (domain != primary_domain) {
1013                                 set_domain_offline(primary_domain);
1014                         }
1015                 }
1016         }
1017 }
1018
1019 /* Deal with a request to go online. */
1020
1021 static void child_msg_online(struct messaging_context *msg,
1022                              void *private_data,
1023                              uint32_t msg_type,
1024                              struct server_id server_id,
1025                              DATA_BLOB *data)
1026 {
1027         struct winbindd_domain *domain;
1028         struct winbindd_domain *primary_domain = NULL;
1029         const char *domainname = (const char *)data->data;
1030
1031         if (data->data == NULL || data->length == 0) {
1032                 return;
1033         }
1034
1035         DEBUG(5,("child_msg_online received for domain %s.\n", domainname));
1036
1037         if (!lp_winbind_offline_logon()) {
1038                 DEBUG(10,("child_msg_online: rejecting online message.\n"));
1039                 return;
1040         }
1041
1042         primary_domain = find_our_domain();
1043
1044         /* Set our global state as online. */
1045         set_global_winbindd_state_online();
1046
1047         /* Try and mark everything online - delete any negative cache entries
1048            to force a reconnect now. */
1049
1050         for (domain = domain_list(); domain; domain = domain->next) {
1051                 if (domain->internal) {
1052                         continue;
1053                 }
1054                 if (strequal(domain->name, domainname)) {
1055                         DEBUG(5,("child_msg_online: requesting %s to go online.\n", domain->name));
1056                         winbindd_flush_negative_conn_cache(domain);
1057                         set_domain_online_request(domain);
1058
1059                         /* we can be in trusted domain, which will contact primary domain
1060                          * we have to bring primary domain online in trusted domain process
1061                          * see, winbindd_dual_pam_auth() --> winbindd_dual_pam_auth_samlogon()
1062                          * --> contact_domain = find_our_domain()
1063                          * */
1064                         if (domain != primary_domain) {
1065                                 winbindd_flush_negative_conn_cache(primary_domain);
1066                                 set_domain_online_request(primary_domain);
1067                         }
1068                 }
1069         }
1070 }
1071
1072 static const char *collect_onlinestatus(TALLOC_CTX *mem_ctx)
1073 {
1074         struct winbindd_domain *domain;
1075         char *buf = NULL;
1076
1077         if ((buf = talloc_asprintf(mem_ctx, "global:%s ", 
1078                                    get_global_winbindd_state_offline() ? 
1079                                    "Offline":"Online")) == NULL) {
1080                 return NULL;
1081         }
1082
1083         for (domain = domain_list(); domain; domain = domain->next) {
1084                 if ((buf = talloc_asprintf_append_buffer(buf, "%s:%s ", 
1085                                                   domain->name, 
1086                                                   domain->online ?
1087                                                   "Online":"Offline")) == NULL) {
1088                         return NULL;
1089                 }
1090         }
1091
1092         buf = talloc_asprintf_append_buffer(buf, "\n");
1093
1094         DEBUG(5,("collect_onlinestatus: %s", buf));
1095
1096         return buf;
1097 }
1098
1099 static void child_msg_onlinestatus(struct messaging_context *msg_ctx,
1100                                    void *private_data,
1101                                    uint32_t msg_type,
1102                                    struct server_id server_id,
1103                                    DATA_BLOB *data)
1104 {
1105         TALLOC_CTX *mem_ctx;
1106         const char *message;
1107         struct server_id *sender;
1108         
1109         DEBUG(5,("winbind_msg_onlinestatus received.\n"));
1110
1111         if (!data->data) {
1112                 return;
1113         }
1114
1115         sender = (struct server_id *)data->data;
1116
1117         mem_ctx = talloc_init("winbind_msg_onlinestatus");
1118         if (mem_ctx == NULL) {
1119                 return;
1120         }
1121         
1122         message = collect_onlinestatus(mem_ctx);
1123         if (message == NULL) {
1124                 talloc_destroy(mem_ctx);
1125                 return;
1126         }
1127
1128         messaging_send_buf(msg_ctx, *sender, MSG_WINBIND_ONLINESTATUS, 
1129                            (uint8 *)message, strlen(message) + 1);
1130
1131         talloc_destroy(mem_ctx);
1132 }
1133
1134 static void child_msg_dump_event_list(struct messaging_context *msg,
1135                                       void *private_data,
1136                                       uint32_t msg_type,
1137                                       struct server_id server_id,
1138                                       DATA_BLOB *data)
1139 {
1140         DEBUG(5,("child_msg_dump_event_list received\n"));
1141
1142         dump_event_list(winbind_event_context());
1143 }
1144
1145 static void winbindd_reinit_after_fork(struct winbindd_child *child)
1146 {
1147         struct winbindd_domain *domain;
1148         struct winbindd_child *cl;
1149
1150         if (!reinit_after_fork(winbind_messaging_context(),
1151                                winbind_event_context(), true)) {
1152                 DEBUG(0,("reinit_after_fork() failed\n"));
1153                 _exit(0);
1154         }
1155
1156         close_conns_after_fork();
1157
1158         if (!override_logfile) {
1159                 lp_set_logfile(child->logfilename);
1160                 reopen_logs();
1161         }
1162
1163         /* Don't handle the same messages as our parent. */
1164         messaging_deregister(winbind_messaging_context(),
1165                              MSG_SMB_CONF_UPDATED, NULL);
1166         messaging_deregister(winbind_messaging_context(),
1167                              MSG_SHUTDOWN, NULL);
1168         messaging_deregister(winbind_messaging_context(),
1169                              MSG_WINBIND_OFFLINE, NULL);
1170         messaging_deregister(winbind_messaging_context(),
1171                              MSG_WINBIND_ONLINE, NULL);
1172         messaging_deregister(winbind_messaging_context(),
1173                              MSG_WINBIND_ONLINESTATUS, NULL);
1174         messaging_deregister(winbind_messaging_context(),
1175                              MSG_DUMP_EVENT_LIST, NULL);
1176         messaging_deregister(winbind_messaging_context(),
1177                              MSG_WINBIND_DUMP_DOMAIN_LIST, NULL);
1178         messaging_deregister(winbind_messaging_context(),
1179                              MSG_DEBUG, NULL);
1180
1181         /* Handle online/offline messages. */
1182         messaging_register(winbind_messaging_context(), NULL,
1183                            MSG_WINBIND_OFFLINE, child_msg_offline);
1184         messaging_register(winbind_messaging_context(), NULL,
1185                            MSG_WINBIND_ONLINE, child_msg_online);
1186         messaging_register(winbind_messaging_context(), NULL,
1187                            MSG_WINBIND_ONLINESTATUS, child_msg_onlinestatus);
1188         messaging_register(winbind_messaging_context(), NULL,
1189                            MSG_DUMP_EVENT_LIST, child_msg_dump_event_list);
1190         messaging_register(winbind_messaging_context(), NULL,
1191                            MSG_DEBUG, debug_message);
1192
1193         /* We have destroyed all events in the winbindd_event_context
1194          * in reinit_after_fork(), so clean out all possible pending
1195          * event pointers. */
1196
1197         /* Deal with check_online_events. */
1198
1199         for (domain = domain_list(); domain; domain = domain->next) {
1200                 TALLOC_FREE(domain->check_online_event);
1201         }
1202
1203         /* Ensure we're not handling a credential cache event inherited
1204          * from our parent. */
1205
1206         ccache_remove_all_after_fork();
1207
1208         /* Destroy all possible events in child list. */
1209         for (cl = children; cl != NULL; cl = cl->next) {
1210                 struct winbindd_async_request *request;
1211
1212                 for (request = cl->requests; request; request = request->next) {
1213                         TALLOC_FREE(request->reply_timeout_event);
1214                 }
1215                 TALLOC_FREE(cl->lockout_policy_event);
1216                 TALLOC_FREE(cl->machine_password_change_event);
1217         }
1218 }
1219
1220 static bool fork_domain_child(struct winbindd_child *child)
1221 {
1222         int fdpair[2];
1223         struct winbindd_cli_state state;
1224         struct winbindd_domain *primary_domain = NULL;
1225
1226         if (child->domain) {
1227                 DEBUG(10, ("fork_domain_child called for domain '%s'\n",
1228                            child->domain->name));
1229         } else {
1230                 DEBUG(10, ("fork_domain_child called without domain.\n"));
1231         }
1232
1233         if (socketpair(AF_UNIX, SOCK_STREAM, 0, fdpair) != 0) {
1234                 DEBUG(0, ("Could not open child pipe: %s\n",
1235                           strerror(errno)));
1236                 return False;
1237         }
1238
1239         ZERO_STRUCT(state);
1240         state.pid = sys_getpid();
1241
1242         child->pid = sys_fork();
1243
1244         if (child->pid == -1) {
1245                 DEBUG(0, ("Could not fork: %s\n", strerror(errno)));
1246                 return False;
1247         }
1248
1249         if (child->pid != 0) {
1250                 /* Parent */
1251                 close(fdpair[0]);
1252                 child->next = child->prev = NULL;
1253                 DLIST_ADD(children, child);
1254                 child->event.fd = fdpair[1];
1255                 child->event.flags = 0;
1256                 child->requests = NULL;
1257                 add_fd_event(&child->event);
1258                 return True;
1259         }
1260
1261         /* Child */
1262
1263         DEBUG(10, ("Child process %d\n", (int)sys_getpid()));
1264
1265         /* Stop zombies in children */
1266         CatchChild();
1267
1268         state.sock = fdpair[0];
1269         close(fdpair[1]);
1270
1271         winbindd_reinit_after_fork(child);
1272
1273         primary_domain = find_our_domain();
1274
1275         if (primary_domain == NULL) {
1276                 smb_panic("no primary domain found");
1277         }
1278
1279         /* It doesn't matter if we allow cache login,
1280          * try to bring domain online after fork. */
1281         if ( child->domain ) {
1282                 child->domain->startup = True;
1283                 child->domain->startup_time = time(NULL);
1284                 /* we can be in primary domain or in trusted domain
1285                  * If we are in trusted domain, set the primary domain
1286                  * in start-up mode */
1287                 if (!(child->domain->internal)) {
1288                         set_domain_online_request(child->domain);
1289                         if (!(child->domain->primary)) {
1290                                 primary_domain->startup = True;
1291                                 primary_domain->startup_time = time(NULL);
1292                                 set_domain_online_request(primary_domain);
1293                         }
1294                 }
1295         }
1296
1297         /* We might be in the idmap child...*/
1298         if (child->domain && !(child->domain->internal) &&
1299             lp_winbind_offline_logon()) {
1300
1301                 set_domain_online_request(child->domain);
1302
1303                 if (primary_domain && (primary_domain != child->domain)) {
1304                         /* We need to talk to the primary
1305                          * domain as well as the trusted
1306                          * domain inside a trusted domain
1307                          * child.
1308                          * See the code in :
1309                          * set_dc_type_and_flags_trustinfo()
1310                          * for details.
1311                          */
1312                         set_domain_online_request(primary_domain);
1313                 }
1314
1315                 child->lockout_policy_event = event_add_timed(
1316                         winbind_event_context(), NULL, timeval_zero(),
1317                         account_lockout_policy_handler,
1318                         child);
1319         }
1320
1321         if (child->domain && child->domain->primary &&
1322             !lp_use_kerberos_keytab() &&
1323             lp_server_role() == ROLE_DOMAIN_MEMBER) {
1324
1325                 struct timeval next_change;
1326
1327                 if (calculate_next_machine_pwd_change(child->domain->name,
1328                                                        &next_change)) {
1329                         child->machine_password_change_event = event_add_timed(
1330                                 winbind_event_context(), NULL, next_change,
1331                                 machine_password_change_handler,
1332                                 child);
1333                 }
1334         }
1335
1336         while (1) {
1337
1338                 int ret;
1339                 fd_set read_fds;
1340                 struct timeval t;
1341                 struct timeval *tp;
1342                 struct timeval now;
1343                 TALLOC_CTX *frame = talloc_stackframe();
1344
1345                 /* check for signals */
1346                 winbind_check_sigterm(false);
1347                 winbind_check_sighup(override_logfile ? NULL :
1348                                 child->logfilename);
1349
1350                 run_events(winbind_event_context(), 0, NULL, NULL);
1351
1352                 GetTimeOfDay(&now);
1353
1354                 if (child->domain && child->domain->startup &&
1355                                 (now.tv_sec > child->domain->startup_time + 30)) {
1356                         /* No longer in "startup" mode. */
1357                         DEBUG(10,("fork_domain_child: domain %s no longer in 'startup' mode.\n",
1358                                 child->domain->name ));
1359                         child->domain->startup = False;
1360                 }
1361
1362                 tp = get_timed_events_timeout(winbind_event_context(), &t);
1363                 if (tp) {
1364                         DEBUG(11,("select will use timeout of %u.%u seconds\n",
1365                                 (unsigned int)tp->tv_sec, (unsigned int)tp->tv_usec ));
1366                 }
1367
1368                 /* Handle messages */
1369
1370                 message_dispatch(winbind_messaging_context());
1371
1372                 FD_ZERO(&read_fds);
1373                 FD_SET(state.sock, &read_fds);
1374
1375                 ret = sys_select(state.sock + 1, &read_fds, NULL, NULL, tp);
1376
1377                 if (ret == 0) {
1378                         DEBUG(11,("nothing is ready yet, continue\n"));
1379                         TALLOC_FREE(frame);
1380                         continue;
1381                 }
1382
1383                 if (ret == -1 && errno == EINTR) {
1384                         /* We got a signal - continue. */
1385                         TALLOC_FREE(frame);
1386                         continue;
1387                 }
1388
1389                 if (ret == -1 && errno != EINTR) {
1390                         DEBUG(0,("select error occured\n"));
1391                         TALLOC_FREE(frame);
1392                         perror("select");
1393                         return False;
1394                 }
1395
1396                 /* fetch a request from the main daemon */
1397                 child_read_request(&state);
1398
1399                 if (state.finished) {
1400                         /* we lost contact with our parent */
1401                         exit(0);
1402                 }
1403
1404                 DEBUG(4,("child daemon request %d\n", (int)state.request.cmd));
1405
1406                 ZERO_STRUCT(state.response);
1407                 state.request.null_term = '\0';
1408                 child_process_request(child, &state);
1409
1410                 SAFE_FREE(state.request.extra_data.data);
1411
1412                 cache_store_response(sys_getpid(), &state.response);
1413
1414                 SAFE_FREE(state.response.extra_data.data);
1415
1416                 /* We just send the result code back, the result
1417                  * structure needs to be fetched via the
1418                  * winbindd_cache. Hmm. That needs fixing... */
1419
1420                 if (write_data(state.sock,
1421                                (const char *)&state.response.result,
1422                                sizeof(state.response.result)) !=
1423                     sizeof(state.response.result)) {
1424                         DEBUG(0, ("Could not write result\n"));
1425                         exit(1);
1426                 }
1427                 TALLOC_FREE(frame);
1428         }
1429 }