Fix winbindd crash in an unusual failure mode. Bug #5737. Based on original patch...
[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         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         /* If child exists and is not already reaped,
200            send kill signal to child. */
201
202         if ((state->child->pid != (pid_t)0) &&
203                         (state->child->pid != (pid_t)-1) &&
204                         (state->child->pid == state->child_pid)) {
205                 kill(state->child_pid, SIGTERM);
206
207                 /* 
208                  * Close the socket to the child.
209                  */
210                 winbind_child_died(state->child_pid);
211         }
212
213         state->response->length = sizeof(struct winbindd_response);
214         state->response->result = WINBINDD_ERROR;
215         state->continuation(state->private_data, False);
216 }
217
218 static void async_request_sent(void *private_data_data, bool success)
219 {
220         struct winbindd_async_request *state =
221                 talloc_get_type_abort(private_data_data, struct winbindd_async_request);
222
223         if (!success) {
224                 DEBUG(5, ("Could not send async request to child pid %u\n",
225                         (unsigned int)state->child_pid ));
226                 async_request_fail(state);
227                 return;
228         }
229
230         /* Request successfully sent to the child, setup the wait for reply */
231
232         setup_async_read(&state->child->event,
233                          &state->response->result,
234                          sizeof(state->response->result),
235                          async_reply_recv, state);
236
237         /* 
238          * Set up a timeout of 300 seconds for the response.
239          * If we don't get it close the child socket and
240          * report failure.
241          */
242
243         state->reply_timeout_event = event_add_timed(winbind_event_context(),
244                                                         NULL,
245                                                         timeval_current_ofs(300,0),
246                                                         "async_request_timeout",
247                                                         async_request_timeout_handler,
248                                                         state);
249         if (!state->reply_timeout_event) {
250                 smb_panic("async_request_sent: failed to add timeout handler.\n");
251         }
252 }
253
254 static void async_reply_recv(void *private_data, bool success)
255 {
256         struct winbindd_async_request *state =
257                 talloc_get_type_abort(private_data, struct winbindd_async_request);
258         struct winbindd_child *child = state->child;
259
260         TALLOC_FREE(state->reply_timeout_event);
261
262         state->response->length = sizeof(struct winbindd_response);
263
264         if (!success) {
265                 DEBUG(5, ("Could not receive async reply from child pid %u\n",
266                         (unsigned int)state->child_pid ));
267
268                 cache_cleanup_response(state->child_pid);
269                 async_request_fail(state);
270                 return;
271         }
272
273         SMB_ASSERT(cache_retrieve_response(state->child_pid,
274                                            state->response));
275
276         cache_cleanup_response(state->child_pid);
277         
278         DLIST_REMOVE(child->requests, state);
279
280         schedule_async_request(child);
281
282         state->continuation(state->private_data, True);
283 }
284
285 static bool fork_domain_child(struct winbindd_child *child);
286
287 static void schedule_async_request(struct winbindd_child *child)
288 {
289         struct winbindd_async_request *request = child->requests;
290
291         if (request == NULL) {
292                 return;
293         }
294
295         if (child->event.flags != 0) {
296                 return;         /* Busy */
297         }
298
299         if ((child->pid == 0) && (!fork_domain_child(child))) {
300                 /* fork_domain_child failed.
301                    Cancel all outstanding requests */
302
303                 while (request != NULL) {
304                         /* request might be free'd in the continuation */
305                         struct winbindd_async_request *next = request->next;
306
307                         async_request_fail(request);
308                         request = next;
309                 }
310                 return;
311         }
312
313         /* Now we know who we're sending to - remember the pid. */
314         request->child_pid = child->pid;
315
316         setup_async_write(&child->event, request->request,
317                           sizeof(*request->request),
318                           async_main_request_sent, request);
319
320         return;
321 }
322
323 struct domain_request_state {
324         TALLOC_CTX *mem_ctx;
325         struct winbindd_domain *domain;
326         struct winbindd_request *request;
327         struct winbindd_response *response;
328         void (*continuation)(void *private_data_data, bool success);
329         void *private_data_data;
330 };
331
332 static void domain_init_recv(void *private_data_data, bool success);
333
334 void async_domain_request(TALLOC_CTX *mem_ctx,
335                           struct winbindd_domain *domain,
336                           struct winbindd_request *request,
337                           struct winbindd_response *response,
338                           void (*continuation)(void *private_data_data, bool success),
339                           void *private_data_data)
340 {
341         struct domain_request_state *state;
342
343         if (domain->initialized) {
344                 async_request(mem_ctx, &domain->child, request, response,
345                               continuation, private_data_data);
346                 return;
347         }
348
349         state = TALLOC_P(mem_ctx, struct domain_request_state);
350         if (state == NULL) {
351                 DEBUG(0, ("talloc failed\n"));
352                 continuation(private_data_data, False);
353                 return;
354         }
355
356         state->mem_ctx = mem_ctx;
357         state->domain = domain;
358         state->request = request;
359         state->response = response;
360         state->continuation = continuation;
361         state->private_data_data = private_data_data;
362
363         init_child_connection(domain, domain_init_recv, state);
364 }
365
366 static void domain_init_recv(void *private_data_data, bool success)
367 {
368         struct domain_request_state *state =
369                 talloc_get_type_abort(private_data_data, struct domain_request_state);
370
371         if (!success) {
372                 DEBUG(5, ("Domain init returned an error\n"));
373                 state->continuation(state->private_data_data, False);
374                 return;
375         }
376
377         async_request(state->mem_ctx, &state->domain->child,
378                       state->request, state->response,
379                       state->continuation, state->private_data_data);
380 }
381
382 static void recvfrom_child(void *private_data_data, bool success)
383 {
384         struct winbindd_cli_state *state =
385                 talloc_get_type_abort(private_data_data, struct winbindd_cli_state);
386         enum winbindd_result result = state->response.result;
387
388         /* This is an optimization: The child has written directly to the
389          * response buffer. The request itself is still in pending state,
390          * state that in the result code. */
391
392         state->response.result = WINBINDD_PENDING;
393
394         if ((!success) || (result != WINBINDD_OK)) {
395                 request_error(state);
396                 return;
397         }
398
399         request_ok(state);
400 }
401
402 void sendto_child(struct winbindd_cli_state *state,
403                   struct winbindd_child *child)
404 {
405         async_request(state->mem_ctx, child, &state->request,
406                       &state->response, recvfrom_child, state);
407 }
408
409 void sendto_domain(struct winbindd_cli_state *state,
410                    struct winbindd_domain *domain)
411 {
412         async_domain_request(state->mem_ctx, domain,
413                              &state->request, &state->response,
414                              recvfrom_child, state);
415 }
416
417 static void child_process_request(struct winbindd_child *child,
418                                   struct winbindd_cli_state *state)
419 {
420         struct winbindd_domain *domain = child->domain;
421         const struct winbindd_child_dispatch_table *table = child->table;
422
423         /* Free response data - we may be interrupted and receive another
424            command before being able to send this data off. */
425
426         state->response.result = WINBINDD_ERROR;
427         state->response.length = sizeof(struct winbindd_response);
428
429         /* as all requests in the child are sync, we can use talloc_tos() */
430         state->mem_ctx = talloc_tos();
431
432         /* Process command */
433
434         for (; table->name; table++) {
435                 if (state->request.cmd == table->struct_cmd) {
436                         DEBUG(10,("child_process_request: request fn %s\n",
437                                   table->name));
438                         state->response.result = table->struct_fn(domain, state);
439                         return;
440                 }
441         }
442
443         DEBUG(1 ,("child_process_request: unknown request fn number %d\n",
444                   (int)state->request.cmd));
445         state->response.result = WINBINDD_ERROR;
446 }
447
448 void setup_child(struct winbindd_child *child,
449                  const struct winbindd_child_dispatch_table *table,
450                  const char *logprefix,
451                  const char *logname)
452 {
453         if (logprefix && logname) {
454                 if (asprintf(&child->logfilename, "%s/%s-%s",
455                              get_dyn_LOGFILEBASE(), logprefix, logname) < 0) {
456                         smb_panic("Internal error: asprintf failed");
457                 }
458         } else {
459                 smb_panic("Internal error: logprefix == NULL && "
460                           "logname == NULL");
461         }
462
463         child->domain = NULL;
464         child->table = table;
465 }
466
467 struct winbindd_child *children = NULL;
468
469 void winbind_child_died(pid_t pid)
470 {
471         struct winbindd_child *child;
472
473         for (child = children; child != NULL; child = child->next) {
474                 if (child->pid == pid) {
475                         break;
476                 }
477         }
478
479         if (child == NULL) {
480                 DEBUG(5, ("Already reaped child %u died\n", (unsigned int)pid));
481                 return;
482         }
483
484         /* This will be re-added in fork_domain_child() */
485
486         DLIST_REMOVE(children, child);
487         
488         remove_fd_event(&child->event);
489         close(child->event.fd);
490         child->event.fd = 0;
491         child->event.flags = 0;
492         child->pid = 0;
493
494         schedule_async_request(child);
495 }
496
497 /* Ensure any negative cache entries with the netbios or realm names are removed. */
498
499 void winbindd_flush_negative_conn_cache(struct winbindd_domain *domain)
500 {
501         flush_negative_conn_cache_for_domain(domain->name);
502         if (*domain->alt_name) {
503                 flush_negative_conn_cache_for_domain(domain->alt_name);
504         }
505 }
506
507 /* 
508  * Parent winbindd process sets its own debug level first and then
509  * sends a message to all the winbindd children to adjust their debug
510  * level to that of parents.
511  */
512
513 void winbind_msg_debug(struct messaging_context *msg_ctx,
514                          void *private_data,
515                          uint32_t msg_type,
516                          struct server_id server_id,
517                          DATA_BLOB *data)
518 {
519         struct winbindd_child *child;
520
521         DEBUG(10,("winbind_msg_debug: got debug message.\n"));
522         
523         debug_message(msg_ctx, private_data, MSG_DEBUG, server_id, data);
524
525         for (child = children; child != NULL; child = child->next) {
526
527                 DEBUG(10,("winbind_msg_debug: sending message to pid %u.\n",
528                         (unsigned int)child->pid));
529
530                 messaging_send_buf(msg_ctx, pid_to_procid(child->pid),
531                            MSG_DEBUG,
532                            data->data,
533                            strlen((char *) data->data) + 1);
534         }
535 }
536
537 /* Set our domains as offline and forward the offline message to our children. */
538
539 void winbind_msg_offline(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         struct winbindd_domain *domain;
547
548         DEBUG(10,("winbind_msg_offline: got offline message.\n"));
549
550         if (!lp_winbind_offline_logon()) {
551                 DEBUG(10,("winbind_msg_offline: rejecting offline message.\n"));
552                 return;
553         }
554
555         /* Set our global state as offline. */
556         if (!set_global_winbindd_state_offline()) {
557                 DEBUG(10,("winbind_msg_offline: offline request failed.\n"));
558                 return;
559         }
560
561         /* Set all our domains as offline. */
562         for (domain = domain_list(); domain; domain = domain->next) {
563                 if (domain->internal) {
564                         continue;
565                 }
566                 DEBUG(5,("winbind_msg_offline: marking %s offline.\n", domain->name));
567                 set_domain_offline(domain);
568         }
569
570         for (child = children; child != NULL; child = child->next) {
571                 /* Don't send message to internal childs.  We've already
572                    done so above. */
573                 if (!child->domain || winbindd_internal_child(child)) {
574                         continue;
575                 }
576
577                 /* Or internal domains (this should not be possible....) */
578                 if (child->domain->internal) {
579                         continue;
580                 }
581
582                 /* Each winbindd child should only process requests for one domain - make sure
583                    we only set it online / offline for that domain. */
584
585                 DEBUG(10,("winbind_msg_offline: sending message to pid %u for domain %s.\n",
586                         (unsigned int)child->pid, domain->name ));
587
588                 messaging_send_buf(msg_ctx, pid_to_procid(child->pid),
589                                    MSG_WINBIND_OFFLINE,
590                                    (uint8 *)child->domain->name,
591                                    strlen(child->domain->name)+1);
592         }
593 }
594
595 /* Set our domains as online and forward the online message to our children. */
596
597 void winbind_msg_online(struct messaging_context *msg_ctx,
598                         void *private_data,
599                         uint32_t msg_type,
600                         struct server_id server_id,
601                         DATA_BLOB *data)
602 {
603         struct winbindd_child *child;
604         struct winbindd_domain *domain;
605
606         DEBUG(10,("winbind_msg_online: got online message.\n"));
607
608         if (!lp_winbind_offline_logon()) {
609                 DEBUG(10,("winbind_msg_online: rejecting online message.\n"));
610                 return;
611         }
612
613         /* Set our global state as online. */
614         set_global_winbindd_state_online();
615
616         smb_nscd_flush_user_cache();
617         smb_nscd_flush_group_cache();
618
619         /* Set all our domains as online. */
620         for (domain = domain_list(); domain; domain = domain->next) {
621                 if (domain->internal) {
622                         continue;
623                 }
624                 DEBUG(5,("winbind_msg_online: requesting %s to go online.\n", domain->name));
625
626                 winbindd_flush_negative_conn_cache(domain);
627                 set_domain_online_request(domain);
628
629                 /* Send an online message to the idmap child when our
630                    primary domain comes back online */
631
632                 if ( domain->primary ) {
633                         struct winbindd_child *idmap = idmap_child();
634                         
635                         if ( idmap->pid != 0 ) {
636                                 messaging_send_buf(msg_ctx,
637                                                    pid_to_procid(idmap->pid), 
638                                                    MSG_WINBIND_ONLINE,
639                                                    (uint8 *)domain->name,
640                                                    strlen(domain->name)+1);
641                         }
642                         
643                 }
644         }
645
646         for (child = children; child != NULL; child = child->next) {
647                 /* Don't send message to internal childs. */
648                 if (!child->domain || winbindd_internal_child(child)) {
649                         continue;
650                 }
651
652                 /* Or internal domains (this should not be possible....) */
653                 if (child->domain->internal) {
654                         continue;
655                 }
656
657                 /* Each winbindd child should only process requests for one domain - make sure
658                    we only set it online / offline for that domain. */
659
660                 DEBUG(10,("winbind_msg_online: sending message to pid %u for domain %s.\n",
661                         (unsigned int)child->pid, child->domain->name ));
662
663                 messaging_send_buf(msg_ctx, pid_to_procid(child->pid),
664                                    MSG_WINBIND_ONLINE,
665                                    (uint8 *)child->domain->name,
666                                    strlen(child->domain->name)+1);
667         }
668 }
669
670 /* Forward the online/offline messages to our children. */
671 void winbind_msg_onlinestatus(struct messaging_context *msg_ctx,
672                               void *private_data,
673                               uint32_t msg_type,
674                               struct server_id server_id,
675                               DATA_BLOB *data)
676 {
677         struct winbindd_child *child;
678
679         DEBUG(10,("winbind_msg_onlinestatus: got onlinestatus message.\n"));
680
681         for (child = children; child != NULL; child = child->next) {
682                 if (child->domain && child->domain->primary) {
683                         DEBUG(10,("winbind_msg_onlinestatus: "
684                                   "sending message to pid %u of primary domain.\n",
685                                   (unsigned int)child->pid));
686                         messaging_send_buf(msg_ctx, pid_to_procid(child->pid), 
687                                            MSG_WINBIND_ONLINESTATUS,
688                                            (uint8 *)data->data,
689                                            data->length);
690                         break;
691                 }
692         }
693 }
694
695 void winbind_msg_dump_event_list(struct messaging_context *msg_ctx,
696                                  void *private_data,
697                                  uint32_t msg_type,
698                                  struct server_id server_id,
699                                  DATA_BLOB *data)
700 {
701         struct winbindd_child *child;
702
703         DEBUG(10,("winbind_msg_dump_event_list received\n"));
704
705         dump_event_list(winbind_event_context());
706
707         for (child = children; child != NULL; child = child->next) {
708
709                 DEBUG(10,("winbind_msg_dump_event_list: sending message to pid %u\n",
710                         (unsigned int)child->pid));
711
712                 messaging_send_buf(msg_ctx, pid_to_procid(child->pid),
713                                    MSG_DUMP_EVENT_LIST,
714                                    NULL, 0);
715         }
716
717 }
718
719 void winbind_msg_dump_domain_list(struct messaging_context *msg_ctx,
720                                   void *private_data,
721                                   uint32_t msg_type,
722                                   struct server_id server_id,
723                                   DATA_BLOB *data)
724 {
725         TALLOC_CTX *mem_ctx;
726         const char *message = NULL;
727         struct server_id *sender = NULL;
728         const char *domain = NULL;
729         char *s = NULL;
730         NTSTATUS status;
731         struct winbindd_domain *dom = NULL;
732
733         DEBUG(5,("winbind_msg_dump_domain_list received.\n"));
734
735         if (!data || !data->data) {
736                 return;
737         }
738
739         if (data->length < sizeof(struct server_id)) {
740                 return;
741         }
742
743         mem_ctx = talloc_init("winbind_msg_dump_domain_list");
744         if (!mem_ctx) {
745                 return;
746         }
747
748         sender = (struct server_id *)data->data;
749         if (data->length > sizeof(struct server_id)) {
750                 domain = (const char *)data->data+sizeof(struct server_id);
751         }
752
753         if (domain) {
754
755                 DEBUG(5,("winbind_msg_dump_domain_list for domain: %s\n",
756                         domain));
757
758                 message = NDR_PRINT_STRUCT_STRING(mem_ctx, winbindd_domain,
759                                                   find_domain_from_name_noinit(domain));
760                 if (!message) {
761                         talloc_destroy(mem_ctx);
762                         return;
763                 }
764
765                 messaging_send_buf(msg_ctx, *sender,
766                                    MSG_WINBIND_DUMP_DOMAIN_LIST,
767                                    (uint8_t *)message, strlen(message) + 1);
768
769                 talloc_destroy(mem_ctx);
770
771                 return;
772         }
773
774         DEBUG(5,("winbind_msg_dump_domain_list all domains\n"));
775
776         for (dom = domain_list(); dom; dom=dom->next) {
777                 message = NDR_PRINT_STRUCT_STRING(mem_ctx, winbindd_domain, dom);
778                 if (!message) {
779                         talloc_destroy(mem_ctx);
780                         return;
781                 }
782
783                 s = talloc_asprintf_append(s, "%s\n", message);
784                 if (!s) {
785                         talloc_destroy(mem_ctx);
786                         return;
787                 }
788         }
789
790         status = messaging_send_buf(msg_ctx, *sender,
791                                     MSG_WINBIND_DUMP_DOMAIN_LIST,
792                                     (uint8_t *)s, strlen(s) + 1);
793         if (!NT_STATUS_IS_OK(status)) {
794                 DEBUG(0,("failed to send message: %s\n",
795                 nt_errstr(status)));
796         }
797
798         talloc_destroy(mem_ctx);
799 }
800
801 static void account_lockout_policy_handler(struct event_context *ctx,
802                                            struct timed_event *te,
803                                            const struct timeval *now,
804                                            void *private_data)
805 {
806         struct winbindd_child *child =
807                 (struct winbindd_child *)private_data;
808         TALLOC_CTX *mem_ctx = NULL;
809         struct winbindd_methods *methods;
810         struct samr_DomInfo12 lockout_policy;
811         NTSTATUS result;
812
813         DEBUG(10,("account_lockout_policy_handler called\n"));
814
815         TALLOC_FREE(child->lockout_policy_event);
816
817         if ( !winbindd_can_contact_domain( child->domain ) ) {
818                 DEBUG(10,("account_lockout_policy_handler: Removing myself since I "
819                           "do not have an incoming trust to domain %s\n", 
820                           child->domain->name));
821
822                 return;         
823         }
824
825         methods = child->domain->methods;
826
827         mem_ctx = talloc_init("account_lockout_policy_handler ctx");
828         if (!mem_ctx) {
829                 result = NT_STATUS_NO_MEMORY;
830         } else {
831                 result = methods->lockout_policy(child->domain, mem_ctx, &lockout_policy);
832         }
833         TALLOC_FREE(mem_ctx);
834
835         if (!NT_STATUS_IS_OK(result)) {
836                 DEBUG(10,("account_lockout_policy_handler: lockout_policy failed error %s\n",
837                          nt_errstr(result)));
838         }
839
840         child->lockout_policy_event = event_add_timed(winbind_event_context(), NULL,
841                                                       timeval_current_ofs(3600, 0),
842                                                       "account_lockout_policy_handler",
843                                                       account_lockout_policy_handler,
844                                                       child);
845 }
846
847 static time_t get_machine_password_timeout(void)
848 {
849         /* until we have gpo support use lp setting */
850         return lp_machine_password_timeout();
851 }
852
853 static bool calculate_next_machine_pwd_change(const char *domain,
854                                               struct timeval *t)
855 {
856         time_t pass_last_set_time;
857         time_t timeout;
858         time_t next_change;
859         char *pw;
860
861         pw = secrets_fetch_machine_password(domain,
862                                             &pass_last_set_time,
863                                             NULL);
864
865         if (pw == NULL) {
866                 DEBUG(0,("cannot fetch own machine password ????"));
867                 return false;
868         }
869
870         SAFE_FREE(pw);
871
872         timeout = get_machine_password_timeout();
873         if (timeout == 0) {
874                 DEBUG(10,("machine password never expires\n"));
875                 return false;
876         }
877
878         if (time(NULL) < (pass_last_set_time + timeout)) {
879                 next_change = pass_last_set_time + timeout;
880                 DEBUG(10,("machine password still valid until: %s\n",
881                         http_timestring(next_change)));
882                 *t = timeval_set(next_change, 0);
883                 return true;
884         }
885
886         DEBUG(10,("machine password expired, needs immediate change\n"));
887
888         *t = timeval_zero();
889
890         return true;
891 }
892
893 static void machine_password_change_handler(struct event_context *ctx,
894                                             struct timed_event *te,
895                                             const struct timeval *now,
896                                             void *private_data)
897 {
898         struct winbindd_child *child =
899                 (struct winbindd_child *)private_data;
900         struct rpc_pipe_client *netlogon_pipe = NULL;
901         TALLOC_CTX *frame;
902         NTSTATUS result;
903         struct timeval next_change;
904
905         DEBUG(10,("machine_password_change_handler called\n"));
906
907         TALLOC_FREE(child->machine_password_change_event);
908
909         if (!calculate_next_machine_pwd_change(child->domain->name,
910                                                &next_change)) {
911                 return;
912         }
913
914         if (!winbindd_can_contact_domain(child->domain)) {
915                 DEBUG(10,("machine_password_change_handler: Removing myself since I "
916                           "do not have an incoming trust to domain %s\n",
917                           child->domain->name));
918                 return;
919         }
920
921         result = cm_connect_netlogon(child->domain, &netlogon_pipe);
922         if (!NT_STATUS_IS_OK(result)) {
923                 DEBUG(10,("machine_password_change_handler: "
924                         "failed to connect netlogon pipe: %s\n",
925                          nt_errstr(result)));
926                 return;
927         }
928
929         frame = talloc_stackframe();
930
931         result = trust_pw_find_change_and_store_it(netlogon_pipe,
932                                                    frame,
933                                                    child->domain->name);
934         TALLOC_FREE(frame);
935
936         if (!NT_STATUS_IS_OK(result)) {
937                 DEBUG(10,("machine_password_change_handler: "
938                         "failed to change machine password: %s\n",
939                          nt_errstr(result)));
940         } else {
941                 DEBUG(10,("machine_password_change_handler: "
942                         "successfully changed machine password\n"));
943         }
944
945         child->machine_password_change_event = event_add_timed(winbind_event_context(), NULL,
946                                                               next_change,
947                                                               "machine_password_change_handler",
948                                                               machine_password_change_handler,
949                                                               child);
950 }
951
952 /* Deal with a request to go offline. */
953
954 static void child_msg_offline(struct messaging_context *msg,
955                               void *private_data,
956                               uint32_t msg_type,
957                               struct server_id server_id,
958                               DATA_BLOB *data)
959 {
960         struct winbindd_domain *domain;
961         const char *domainname = (const char *)data->data;
962
963         if (data->data == NULL || data->length == 0) {
964                 return;
965         }
966
967         DEBUG(5,("child_msg_offline received for domain %s.\n", domainname));
968
969         if (!lp_winbind_offline_logon()) {
970                 DEBUG(10,("child_msg_offline: rejecting offline message.\n"));
971                 return;
972         }
973
974         /* Mark the requested domain offline. */
975
976         for (domain = domain_list(); domain; domain = domain->next) {
977                 if (domain->internal) {
978                         continue;
979                 }
980                 if (strequal(domain->name, domainname)) {
981                         DEBUG(5,("child_msg_offline: marking %s offline.\n", domain->name));
982                         set_domain_offline(domain);
983                 }
984         }
985 }
986
987 /* Deal with a request to go online. */
988
989 static void child_msg_online(struct messaging_context *msg,
990                              void *private_data,
991                              uint32_t msg_type,
992                              struct server_id server_id,
993                              DATA_BLOB *data)
994 {
995         struct winbindd_domain *domain;
996         const char *domainname = (const char *)data->data;
997
998         if (data->data == NULL || data->length == 0) {
999                 return;
1000         }
1001
1002         DEBUG(5,("child_msg_online received for domain %s.\n", domainname));
1003
1004         if (!lp_winbind_offline_logon()) {
1005                 DEBUG(10,("child_msg_online: rejecting online message.\n"));
1006                 return;
1007         }
1008
1009         /* Set our global state as online. */
1010         set_global_winbindd_state_online();
1011
1012         /* Try and mark everything online - delete any negative cache entries
1013            to force a reconnect now. */
1014
1015         for (domain = domain_list(); domain; domain = domain->next) {
1016                 if (domain->internal) {
1017                         continue;
1018                 }
1019                 if (strequal(domain->name, domainname)) {
1020                         DEBUG(5,("child_msg_online: requesting %s to go online.\n", domain->name));
1021                         winbindd_flush_negative_conn_cache(domain);
1022                         set_domain_online_request(domain);
1023                 }
1024         }
1025 }
1026
1027 static const char *collect_onlinestatus(TALLOC_CTX *mem_ctx)
1028 {
1029         struct winbindd_domain *domain;
1030         char *buf = NULL;
1031
1032         if ((buf = talloc_asprintf(mem_ctx, "global:%s ", 
1033                                    get_global_winbindd_state_offline() ? 
1034                                    "Offline":"Online")) == NULL) {
1035                 return NULL;
1036         }
1037
1038         for (domain = domain_list(); domain; domain = domain->next) {
1039                 if ((buf = talloc_asprintf_append_buffer(buf, "%s:%s ", 
1040                                                   domain->name, 
1041                                                   domain->online ?
1042                                                   "Online":"Offline")) == NULL) {
1043                         return NULL;
1044                 }
1045         }
1046
1047         buf = talloc_asprintf_append_buffer(buf, "\n");
1048
1049         DEBUG(5,("collect_onlinestatus: %s", buf));
1050
1051         return buf;
1052 }
1053
1054 static void child_msg_onlinestatus(struct messaging_context *msg_ctx,
1055                                    void *private_data,
1056                                    uint32_t msg_type,
1057                                    struct server_id server_id,
1058                                    DATA_BLOB *data)
1059 {
1060         TALLOC_CTX *mem_ctx;
1061         const char *message;
1062         struct server_id *sender;
1063         
1064         DEBUG(5,("winbind_msg_onlinestatus received.\n"));
1065
1066         if (!data->data) {
1067                 return;
1068         }
1069
1070         sender = (struct server_id *)data->data;
1071
1072         mem_ctx = talloc_init("winbind_msg_onlinestatus");
1073         if (mem_ctx == NULL) {
1074                 return;
1075         }
1076         
1077         message = collect_onlinestatus(mem_ctx);
1078         if (message == NULL) {
1079                 talloc_destroy(mem_ctx);
1080                 return;
1081         }
1082
1083         messaging_send_buf(msg_ctx, *sender, MSG_WINBIND_ONLINESTATUS, 
1084                            (uint8 *)message, strlen(message) + 1);
1085
1086         talloc_destroy(mem_ctx);
1087 }
1088
1089 static void child_msg_dump_event_list(struct messaging_context *msg,
1090                                       void *private_data,
1091                                       uint32_t msg_type,
1092                                       struct server_id server_id,
1093                                       DATA_BLOB *data)
1094 {
1095         DEBUG(5,("child_msg_dump_event_list received\n"));
1096
1097         dump_event_list(winbind_event_context());
1098 }
1099
1100
1101 static bool fork_domain_child(struct winbindd_child *child)
1102 {
1103         int fdpair[2];
1104         struct winbindd_cli_state state;
1105         struct winbindd_domain *domain;
1106         struct winbindd_domain *primary_domain = NULL;
1107
1108         if (child->domain) {
1109                 DEBUG(10, ("fork_domain_child called for domain '%s'\n",
1110                            child->domain->name));
1111         } else {
1112                 DEBUG(10, ("fork_domain_child called without domain.\n"));
1113         }
1114
1115         if (socketpair(AF_UNIX, SOCK_STREAM, 0, fdpair) != 0) {
1116                 DEBUG(0, ("Could not open child pipe: %s\n",
1117                           strerror(errno)));
1118                 return False;
1119         }
1120
1121         ZERO_STRUCT(state);
1122         state.pid = sys_getpid();
1123
1124         child->pid = sys_fork();
1125
1126         if (child->pid == -1) {
1127                 DEBUG(0, ("Could not fork: %s\n", strerror(errno)));
1128                 return False;
1129         }
1130
1131         if (child->pid != 0) {
1132                 /* Parent */
1133                 close(fdpair[0]);
1134                 child->next = child->prev = NULL;
1135                 DLIST_ADD(children, child);
1136                 child->event.fd = fdpair[1];
1137                 child->event.flags = 0;
1138                 child->requests = NULL;
1139                 add_fd_event(&child->event);
1140                 return True;
1141         }
1142
1143         /* Child */
1144
1145         DEBUG(10, ("Child process %d\n", (int)sys_getpid()));
1146
1147         /* Stop zombies in children */
1148         CatchChild();
1149
1150         state.sock = fdpair[0];
1151         close(fdpair[1]);
1152
1153         if (!reinit_after_fork(winbind_messaging_context(), true)) {
1154                 DEBUG(0,("reinit_after_fork() failed\n"));
1155                 _exit(0);
1156         }
1157
1158         close_conns_after_fork();
1159
1160         if (!override_logfile) {
1161                 lp_set_logfile(child->logfilename);
1162                 reopen_logs();
1163         }
1164
1165         /*
1166          * For clustering, we need to re-init our ctdbd connection after the
1167          * fork
1168          */
1169         if (!NT_STATUS_IS_OK(messaging_reinit(winbind_messaging_context())))
1170                 exit(1);
1171
1172         /* Don't handle the same messages as our parent. */
1173         messaging_deregister(winbind_messaging_context(),
1174                              MSG_SMB_CONF_UPDATED, NULL);
1175         messaging_deregister(winbind_messaging_context(),
1176                              MSG_SHUTDOWN, NULL);
1177         messaging_deregister(winbind_messaging_context(),
1178                              MSG_WINBIND_OFFLINE, NULL);
1179         messaging_deregister(winbind_messaging_context(),
1180                              MSG_WINBIND_ONLINE, NULL);
1181         messaging_deregister(winbind_messaging_context(),
1182                              MSG_WINBIND_ONLINESTATUS, NULL);
1183         messaging_deregister(winbind_messaging_context(),
1184                              MSG_DUMP_EVENT_LIST, NULL);
1185         messaging_deregister(winbind_messaging_context(),
1186                              MSG_WINBIND_DUMP_DOMAIN_LIST, NULL);
1187         messaging_deregister(winbind_messaging_context(),
1188                              MSG_DEBUG, NULL);
1189
1190         /* Handle online/offline messages. */
1191         messaging_register(winbind_messaging_context(), NULL,
1192                            MSG_WINBIND_OFFLINE, child_msg_offline);
1193         messaging_register(winbind_messaging_context(), NULL,
1194                            MSG_WINBIND_ONLINE, child_msg_online);
1195         messaging_register(winbind_messaging_context(), NULL,
1196                            MSG_WINBIND_ONLINESTATUS, child_msg_onlinestatus);
1197         messaging_register(winbind_messaging_context(), NULL,
1198                            MSG_DUMP_EVENT_LIST, child_msg_dump_event_list);
1199         messaging_register(winbind_messaging_context(), NULL,
1200                            MSG_DEBUG, debug_message);
1201
1202         if ( child->domain ) {
1203                 child->domain->startup = True;
1204                 child->domain->startup_time = time(NULL);
1205         }
1206
1207         /* Ensure we have no pending check_online events other
1208            than one for this domain or the primary domain. */
1209
1210         for (domain = domain_list(); domain; domain = domain->next) {
1211                 if (domain->primary) {
1212                         primary_domain = domain;
1213                 }
1214                 if ((domain != child->domain) && !domain->primary) {
1215                         TALLOC_FREE(domain->check_online_event);
1216                 }
1217         }
1218
1219         if (primary_domain == NULL) {
1220                 smb_panic("no primary domain found");
1221         }
1222
1223         /* Ensure we're not handling an event inherited from
1224            our parent. */
1225
1226         cancel_named_event(winbind_event_context(),
1227                            "krb5_ticket_refresh_handler");
1228
1229         /* We might be in the idmap child...*/
1230         if (child->domain && !(child->domain->internal) &&
1231             lp_winbind_offline_logon()) {
1232
1233                 set_domain_online_request(child->domain);
1234
1235                 if (primary_domain != child->domain) {
1236                         /* We need to talk to the primary
1237                          * domain as well as the trusted
1238                          * domain inside a trusted domain
1239                          * child.
1240                          * See the code in :
1241                          * set_dc_type_and_flags_trustinfo()
1242                          * for details.
1243                          */
1244                         set_domain_online_request(primary_domain);
1245                 }
1246
1247                 child->lockout_policy_event = event_add_timed(
1248                         winbind_event_context(), NULL, timeval_zero(),
1249                         "account_lockout_policy_handler",
1250                         account_lockout_policy_handler,
1251                         child);
1252         }
1253
1254         if (child->domain && child->domain->primary &&
1255             lp_server_role() == ROLE_DOMAIN_MEMBER) {
1256
1257                 struct timeval next_change;
1258
1259                 if (calculate_next_machine_pwd_change(child->domain->name,
1260                                                        &next_change)) {
1261                         child->machine_password_change_event = event_add_timed(
1262                                 winbind_event_context(), NULL, next_change,
1263                                 "machine_password_change_handler",
1264                                 machine_password_change_handler,
1265                                 child);
1266                 }
1267         }
1268
1269         while (1) {
1270
1271                 int ret;
1272                 fd_set read_fds;
1273                 struct timeval t;
1274                 struct timeval *tp;
1275                 struct timeval now;
1276                 TALLOC_CTX *frame = talloc_stackframe();
1277
1278                 /* check for signals */
1279                 winbind_check_sigterm(false);
1280                 winbind_check_sighup(override_logfile ? NULL :
1281                                 child->logfilename);
1282
1283                 run_events(winbind_event_context(), 0, NULL, NULL);
1284
1285                 GetTimeOfDay(&now);
1286
1287                 if (child->domain && child->domain->startup &&
1288                                 (now.tv_sec > child->domain->startup_time + 30)) {
1289                         /* No longer in "startup" mode. */
1290                         DEBUG(10,("fork_domain_child: domain %s no longer in 'startup' mode.\n",
1291                                 child->domain->name ));
1292                         child->domain->startup = False;
1293                 }
1294
1295                 tp = get_timed_events_timeout(winbind_event_context(), &t);
1296                 if (tp) {
1297                         DEBUG(11,("select will use timeout of %u.%u seconds\n",
1298                                 (unsigned int)tp->tv_sec, (unsigned int)tp->tv_usec ));
1299                 }
1300
1301                 /* Handle messages */
1302
1303                 message_dispatch(winbind_messaging_context());
1304
1305                 FD_ZERO(&read_fds);
1306                 FD_SET(state.sock, &read_fds);
1307
1308                 ret = sys_select(state.sock + 1, &read_fds, NULL, NULL, tp);
1309
1310                 if (ret == 0) {
1311                         DEBUG(11,("nothing is ready yet, continue\n"));
1312                         TALLOC_FREE(frame);
1313                         continue;
1314                 }
1315
1316                 if (ret == -1 && errno == EINTR) {
1317                         /* We got a signal - continue. */
1318                         TALLOC_FREE(frame);
1319                         continue;
1320                 }
1321
1322                 if (ret == -1 && errno != EINTR) {
1323                         DEBUG(0,("select error occured\n"));
1324                         TALLOC_FREE(frame);
1325                         perror("select");
1326                         return False;
1327                 }
1328
1329                 /* fetch a request from the main daemon */
1330                 child_read_request(&state);
1331
1332                 if (state.finished) {
1333                         /* we lost contact with our parent */
1334                         exit(0);
1335                 }
1336
1337                 DEBUG(4,("child daemon request %d\n", (int)state.request.cmd));
1338
1339                 ZERO_STRUCT(state.response);
1340                 state.request.null_term = '\0';
1341                 child_process_request(child, &state);
1342
1343                 SAFE_FREE(state.request.extra_data.data);
1344
1345                 cache_store_response(sys_getpid(), &state.response);
1346
1347                 SAFE_FREE(state.response.extra_data.data);
1348
1349                 /* We just send the result code back, the result
1350                  * structure needs to be fetched via the
1351                  * winbindd_cache. Hmm. That needs fixing... */
1352
1353                 if (write_data(state.sock,
1354                                (const char *)&state.response.result,
1355                                sizeof(state.response.result)) !=
1356                     sizeof(state.response.result)) {
1357                         DEBUG(0, ("Could not write result\n"));
1358                         exit(1);
1359                 }
1360                 TALLOC_FREE(frame);
1361         }
1362 }