smbd: remove process shortname arg from reinit_after_fork()
[bbaumbach/samba-autobuild/.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 #include "rpc_client/rpc_client.h"
33 #include "nsswitch/wb_reqtrans.h"
34 #include "secrets.h"
35 #include "../lib/util/select.h"
36 #include "winbindd_traceid.h"
37 #include "../libcli/security/security.h"
38 #include "system/select.h"
39 #include "messages.h"
40 #include "../lib/util/tevent_unix.h"
41 #include "lib/param/loadparm.h"
42 #include "lib/util/sys_rw.h"
43 #include "lib/util/sys_rw_data.h"
44 #include "passdb.h"
45 #include "lib/util/string_wrappers.h"
46 #include "lib/global_contexts.h"
47 #include "idmap.h"
48 #include "libcli/auth/netlogon_creds_cli.h"
49 #include "../lib/util/pidfile.h"
50 #include "librpc/gen_ndr/ndr_winbind_c.h"
51 #include "lib/util/util_process.h"
52
53 #undef DBGC_CLASS
54 #define DBGC_CLASS DBGC_WINBIND
55
56 static void forall_domain_children(bool (*fn)(struct winbindd_child *c,
57                                               void *private_data),
58                                    void *private_data)
59 {
60         struct winbindd_domain *d;
61
62         for (d = domain_list(); d != NULL; d = d->next) {
63                 int i;
64
65                 for (i = 0; i < lp_winbind_max_domain_connections(); i++) {
66                         struct winbindd_child *c = &d->children[i];
67                         bool ok;
68
69                         if (c->pid == 0) {
70                                 continue;
71                         }
72
73                         ok = fn(c, private_data);
74                         if (!ok) {
75                                 return;
76                         }
77                 }
78         }
79 }
80
81 static void forall_children(bool (*fn)(struct winbindd_child *c,
82                                        void *private_data),
83                             void *private_data)
84 {
85         struct winbindd_child *c;
86         bool ok;
87
88         c = idmap_child();
89         if (c->pid != 0) {
90                 ok = fn(c, private_data);
91                 if (!ok) {
92                         return;
93                 }
94         }
95
96         c = locator_child();
97         if (c->pid != 0) {
98                 ok = fn(c, private_data);
99                 if (!ok) {
100                         return;
101                 }
102         }
103
104         forall_domain_children(fn, private_data);
105 }
106
107 /* Read some data from a client connection */
108
109 static NTSTATUS child_read_request(int sock, struct winbindd_request *wreq)
110 {
111         NTSTATUS status;
112
113         status = read_data_ntstatus(sock, (char *)wreq, sizeof(*wreq));
114         if (!NT_STATUS_IS_OK(status)) {
115                 DEBUG(3, ("child_read_request: read_data failed: %s\n",
116                           nt_errstr(status)));
117                 return status;
118         }
119
120         if (wreq->extra_len == 0) {
121                 wreq->extra_data.data = NULL;
122                 return NT_STATUS_OK;
123         }
124
125         DEBUG(10, ("Need to read %d extra bytes\n", (int)wreq->extra_len));
126
127         wreq->extra_data.data = SMB_MALLOC_ARRAY(char, wreq->extra_len + 1);
128         if (wreq->extra_data.data == NULL) {
129                 DEBUG(0, ("malloc failed\n"));
130                 return NT_STATUS_NO_MEMORY;
131         }
132
133         /* Ensure null termination */
134         wreq->extra_data.data[wreq->extra_len] = '\0';
135
136         status = read_data_ntstatus(sock, wreq->extra_data.data,
137                                     wreq->extra_len);
138         if (!NT_STATUS_IS_OK(status)) {
139                 DEBUG(0, ("Could not read extra data: %s\n",
140                           nt_errstr(status)));
141         }
142         return status;
143 }
144
145 static NTSTATUS child_write_response(int sock, struct winbindd_response *wrsp)
146 {
147         struct iovec iov[2];
148         int iov_count;
149
150         iov[0].iov_base = (void *)wrsp;
151         iov[0].iov_len = sizeof(struct winbindd_response);
152         iov_count = 1;
153
154         if (wrsp->length > sizeof(struct winbindd_response)) {
155                 iov[1].iov_base = (void *)wrsp->extra_data.data;
156                 iov[1].iov_len = wrsp->length-iov[0].iov_len;
157                 iov_count = 2;
158         }
159
160         DEBUG(10, ("Writing %d bytes to parent\n", (int)wrsp->length));
161
162         if (write_data_iov(sock, iov, iov_count) != wrsp->length) {
163                 DEBUG(0, ("Could not write result\n"));
164                 return NT_STATUS_INVALID_HANDLE;
165         }
166
167         return NT_STATUS_OK;
168 }
169
170 /*
171  * Do winbind child async request. This is not simply wb_simple_trans. We have
172  * to do the queueing ourselves because while a request is queued, the child
173  * might have crashed, and we have to re-fork it in the _trigger function.
174  */
175
176 struct wb_child_request_state {
177         struct tevent_context *ev;
178         struct tevent_req *queue_subreq;
179         struct tevent_req *subreq;
180         struct winbindd_child *child;
181         struct winbindd_request *request;
182         struct winbindd_response *response;
183 };
184
185 static bool fork_domain_child(struct winbindd_child *child);
186
187 static void wb_child_request_waited(struct tevent_req *subreq);
188 static void wb_child_request_done(struct tevent_req *subreq);
189 static void wb_child_request_orphaned(struct tevent_req *subreq);
190
191 static void wb_child_request_cleanup(struct tevent_req *req,
192                                      enum tevent_req_state req_state);
193
194 struct tevent_req *wb_child_request_send(TALLOC_CTX *mem_ctx,
195                                          struct tevent_context *ev,
196                                          struct winbindd_child *child,
197                                          struct winbindd_request *request)
198 {
199         struct tevent_req *req;
200         struct wb_child_request_state *state;
201         struct tevent_req *subreq;
202
203         req = tevent_req_create(mem_ctx, &state,
204                                 struct wb_child_request_state);
205         if (req == NULL) {
206                 return NULL;
207         }
208
209         state->ev = ev;
210         state->child = child;
211
212         /*
213          * We have to make a copy of "request", because our caller
214          * might drop us via talloc_free().
215          *
216          * The talloc_move() magic in wb_child_request_cleanup() keeps
217          * all the requests, but if we are sitting deep within
218          * writev_send() down to the client, we have given it the
219          * pointer to "request". As our caller lost interest, it will
220          * just free "request", while writev_send still references it.
221          */
222
223         state->request = talloc_memdup(state, request, sizeof(*request));
224         if (tevent_req_nomem(state->request, req)) {
225                 return tevent_req_post(req, ev);
226         }
227
228         state->request->traceid = debug_traceid_get();
229
230         if (request->extra_data.data != NULL) {
231                 state->request->extra_data.data = talloc_memdup(
232                         state->request,
233                         request->extra_data.data,
234                         request->extra_len);
235                 if (tevent_req_nomem(state->request->extra_data.data, req)) {
236                         return tevent_req_post(req, ev);
237                 }
238         }
239
240         subreq = tevent_queue_wait_send(state, ev, child->queue);
241         if (tevent_req_nomem(subreq, req)) {
242                 return tevent_req_post(req, ev);
243         }
244         tevent_req_set_callback(subreq, wb_child_request_waited, req);
245         state->queue_subreq = subreq;
246
247         tevent_req_set_cleanup_fn(req, wb_child_request_cleanup);
248
249         return req;
250 }
251
252 static void wb_child_request_waited(struct tevent_req *subreq)
253 {
254         struct tevent_req *req = tevent_req_callback_data(
255                 subreq, struct tevent_req);
256         struct wb_child_request_state *state = tevent_req_data(
257                 req, struct wb_child_request_state);
258         bool ok;
259
260         ok = tevent_queue_wait_recv(subreq);
261         if (!ok) {
262                 tevent_req_oom(req);
263                 return;
264         }
265         /*
266          * We need to keep state->queue_subreq
267          * in order to block the queue.
268          */
269         subreq = NULL;
270
271         if ((state->child->sock == -1) && (!fork_domain_child(state->child))) {
272                 tevent_req_error(req, errno);
273                 return;
274         }
275
276         tevent_fd_set_flags(state->child->monitor_fde, 0);
277
278         subreq = wb_simple_trans_send(state, global_event_context(), NULL,
279                                       state->child->sock, state->request);
280         if (tevent_req_nomem(subreq, req)) {
281                 return;
282         }
283
284         state->subreq = subreq;
285         tevent_req_set_callback(subreq, wb_child_request_done, req);
286         tevent_req_set_endtime(req, state->ev, timeval_current_ofs(300, 0));
287 }
288
289 static void wb_child_request_done(struct tevent_req *subreq)
290 {
291         struct tevent_req *req = tevent_req_callback_data(
292                 subreq, struct tevent_req);
293         struct wb_child_request_state *state = tevent_req_data(
294                 req, struct wb_child_request_state);
295         int ret, err;
296
297         ret = wb_simple_trans_recv(subreq, state, &state->response, &err);
298         /* Freeing the subrequest is deferred until the cleanup function,
299          * which has to know whether a subrequest exists, and consequently
300          * decide whether to shut down the pipe to the child process.
301          */
302         if (ret == -1) {
303                 tevent_req_error(req, err);
304                 return;
305         }
306         tevent_req_done(req);
307 }
308
309 static void wb_child_request_orphaned(struct tevent_req *subreq)
310 {
311         struct winbindd_child *child =
312                 (struct winbindd_child *)tevent_req_callback_data_void(subreq);
313
314         DBG_WARNING("cleanup orphaned subreq[%p]\n", subreq);
315         TALLOC_FREE(subreq);
316
317         if (child->domain != NULL) {
318                 /*
319                  * If the child is attached to a domain,
320                  * we need to make sure the domain queue
321                  * can move forward, after the orphaned
322                  * request is done.
323                  */
324                 tevent_queue_start(child->domain->queue);
325         }
326 }
327
328 int wb_child_request_recv(struct tevent_req *req, TALLOC_CTX *mem_ctx,
329                           struct winbindd_response **presponse, int *err)
330 {
331         struct wb_child_request_state *state = tevent_req_data(
332                 req, struct wb_child_request_state);
333
334         if (tevent_req_is_unix_error(req, err)) {
335                 return -1;
336         }
337         *presponse = talloc_move(mem_ctx, &state->response);
338         return 0;
339 }
340
341 static void wb_child_request_cleanup(struct tevent_req *req,
342                                      enum tevent_req_state req_state)
343 {
344         struct wb_child_request_state *state =
345             tevent_req_data(req, struct wb_child_request_state);
346
347         if (state->subreq == NULL) {
348                 /* nothing to cleanup */
349                 return;
350         }
351
352         if (req_state == TEVENT_REQ_RECEIVED) {
353                 struct tevent_req *subreq = NULL;
354
355                 /*
356                  * Our caller gave up, but we need to keep
357                  * the low level request (wb_simple_trans)
358                  * in order to maintain the parent child protocol.
359                  *
360                  * We also need to keep the child queue blocked
361                  * until we got the response from the child.
362                  */
363
364                 subreq = talloc_move(state->child->queue, &state->subreq);
365                 talloc_move(subreq, &state->queue_subreq);
366                 talloc_move(subreq, &state->request);
367                 tevent_req_set_callback(subreq,
368                                         wb_child_request_orphaned,
369                                         state->child);
370
371                 DBG_WARNING("keep orphaned subreq[%p]\n", subreq);
372                 return;
373         }
374
375         TALLOC_FREE(state->subreq);
376         TALLOC_FREE(state->queue_subreq);
377
378         tevent_fd_set_flags(state->child->monitor_fde, TEVENT_FD_READ);
379
380         if (state->child->domain != NULL) {
381                 /*
382                  * If the child is attached to a domain,
383                  * we need to make sure the domain queue
384                  * can move forward, after the request
385                  * is done.
386                  */
387                 tevent_queue_start(state->child->domain->queue);
388         }
389
390         if (req_state == TEVENT_REQ_DONE) {
391                 /* transmitted request and got response */
392                 return;
393         }
394
395         /*
396          * Failed to transmit and receive response, or request
397          * cancelled while being serviced.
398          * The basic parent/child communication broke, close
399          * our socket
400          */
401         TALLOC_FREE(state->child->monitor_fde);
402         close(state->child->sock);
403         state->child->sock = -1;
404 }
405
406 static void child_socket_readable(struct tevent_context *ev,
407                                   struct tevent_fd *fde,
408                                   uint16_t flags,
409                                   void *private_data)
410 {
411         struct winbindd_child *child = private_data;
412
413         if ((flags & TEVENT_FD_READ) == 0) {
414                 return;
415         }
416
417         TALLOC_FREE(child->monitor_fde);
418
419         /*
420          * We're only active when there is no outstanding child
421          * request. Arriving here means the child closed its socket,
422          * it died. Do the same here.
423          */
424
425         SMB_ASSERT(child->sock != -1);
426
427         close(child->sock);
428         child->sock = -1;
429 }
430
431 static struct winbindd_child *choose_domain_child(struct winbindd_domain *domain)
432 {
433         struct winbindd_child *shortest = &domain->children[0];
434         struct winbindd_child *current;
435         int i;
436
437         for (i=0; i<lp_winbind_max_domain_connections(); i++) {
438                 size_t shortest_len, current_len;
439
440                 current = &domain->children[i];
441                 current_len = tevent_queue_length(current->queue);
442
443                 if (current_len == 0) {
444                         /* idle child */
445                         return current;
446                 }
447
448                 shortest_len = tevent_queue_length(shortest->queue);
449
450                 if (current_len < shortest_len) {
451                         shortest = current;
452                 }
453         }
454
455         return shortest;
456 }
457
458 struct dcerpc_binding_handle *dom_child_handle(struct winbindd_domain *domain)
459 {
460         return domain->binding_handle;
461 }
462
463 struct wb_domain_request_state {
464         struct tevent_context *ev;
465         struct tevent_queue_entry *queue_entry;
466         struct winbindd_domain *domain;
467         struct winbindd_child *child;
468         struct winbindd_request *request;
469         struct winbindd_request *init_req;
470         struct winbindd_response *response;
471         struct tevent_req *pending_subreq;
472         struct wbint_InitConnection r;
473 };
474
475 static void wb_domain_request_cleanup(struct tevent_req *req,
476                                       enum tevent_req_state req_state)
477 {
478         struct wb_domain_request_state *state = tevent_req_data(
479                 req, struct wb_domain_request_state);
480
481         /*
482          * If we're completely done or got a failure.
483          * we should remove ourself from the domain queue,
484          * after removing the child subreq from the child queue
485          * and give the next one in the queue the chance
486          * to check for an idle child.
487          */
488         TALLOC_FREE(state->pending_subreq);
489         TALLOC_FREE(state->queue_entry);
490         tevent_queue_start(state->domain->queue);
491 }
492
493 static void wb_domain_request_trigger(struct tevent_req *req,
494                                       void *private_data);
495 static void wb_domain_request_gotdc(struct tevent_req *subreq);
496 static void wb_domain_request_initialized(struct tevent_req *subreq);
497 static void wb_domain_request_done(struct tevent_req *subreq);
498
499 struct tevent_req *wb_domain_request_send(TALLOC_CTX *mem_ctx,
500                                           struct tevent_context *ev,
501                                           struct winbindd_domain *domain,
502                                           struct winbindd_request *request)
503 {
504         struct tevent_req *req;
505         struct wb_domain_request_state *state;
506
507         req = tevent_req_create(mem_ctx, &state,
508                                 struct wb_domain_request_state);
509         if (req == NULL) {
510                 return NULL;
511         }
512
513         state->domain = domain;
514         state->ev = ev;
515         state->request = request;
516
517         tevent_req_set_cleanup_fn(req, wb_domain_request_cleanup);
518
519         state->queue_entry = tevent_queue_add_entry(
520                         domain->queue, state->ev, req,
521                         wb_domain_request_trigger, NULL);
522         if (tevent_req_nomem(state->queue_entry, req)) {
523                 return tevent_req_post(req, ev);
524         }
525
526         return req;
527 }
528
529 static void wb_domain_request_trigger(struct tevent_req *req,
530                                       void *private_data)
531 {
532         struct wb_domain_request_state *state = tevent_req_data(
533                 req, struct wb_domain_request_state);
534         struct winbindd_domain *domain = state->domain;
535         struct tevent_req *subreq = NULL;
536         size_t shortest_queue_length;
537
538         state->child = choose_domain_child(domain);
539         shortest_queue_length = tevent_queue_length(state->child->queue);
540         if (shortest_queue_length > 0) {
541                 /*
542                  * All children are busy, we need to stop
543                  * the queue and untrigger our own queue
544                  * entry. Once a pending request
545                  * is done it calls tevent_queue_start
546                  * and we get retriggered.
547                  */
548                 state->child = NULL;
549                 tevent_queue_stop(state->domain->queue);
550                 tevent_queue_entry_untrigger(state->queue_entry);
551                 return;
552         }
553
554         if (domain->initialized) {
555                 subreq = wb_child_request_send(state, state->ev, state->child,
556                                                state->request);
557                 if (tevent_req_nomem(subreq, req)) {
558                         return;
559                 }
560                 tevent_req_set_callback(subreq, wb_domain_request_done, req);
561                 state->pending_subreq = subreq;
562
563                 /*
564                  * Once the domain is initialized and
565                  * once we placed our real request into the child queue,
566                  * we can remove ourself from the domain queue
567                  * and give the next one in the queue the chance
568                  * to check for an idle child.
569                  */
570                 TALLOC_FREE(state->queue_entry);
571                 return;
572         }
573
574         state->init_req = talloc_zero(state, struct winbindd_request);
575         if (tevent_req_nomem(state->init_req, req)) {
576                 return;
577         }
578
579         if (IS_DC || domain->primary || domain->internal) {
580                 /* The primary domain has to find the DC name itself */
581                 state->r.in.dcname = talloc_strdup(state, "");
582                 if (tevent_req_nomem(state->r.in.dcname, req)) {
583                         return;
584                 }
585
586                 subreq = dcerpc_wbint_InitConnection_r_send(state,
587                                                 state->ev,
588                                                 state->child->binding_handle,
589                                                 &state->r);
590                 if (tevent_req_nomem(subreq, req)) {
591                         return;
592                 }
593                 tevent_req_set_callback(subreq, wb_domain_request_initialized,
594                                         req);
595                 state->pending_subreq = subreq;
596                 return;
597         }
598
599         /*
600          * This is *not* the primary domain,
601          * let's ask our DC about a DC name.
602          *
603          * We prefer getting a dns name in dc_unc,
604          * which is indicated by DS_RETURN_DNS_NAME.
605          * For NT4 domains we still get the netbios name.
606          */
607         subreq = wb_dsgetdcname_send(state, state->ev,
608                                      state->domain->name,
609                                      NULL, /* domain_guid */
610                                      NULL, /* site_name */
611                                      DS_RETURN_DNS_NAME); /* flags */
612         if (tevent_req_nomem(subreq, req)) {
613                 return;
614         }
615         tevent_req_set_callback(subreq, wb_domain_request_gotdc, req);
616         state->pending_subreq = subreq;
617         return;
618 }
619
620 static void wb_domain_request_gotdc(struct tevent_req *subreq)
621 {
622         struct tevent_req *req = tevent_req_callback_data(
623                 subreq, struct tevent_req);
624         struct wb_domain_request_state *state = tevent_req_data(
625                 req, struct wb_domain_request_state);
626         struct netr_DsRGetDCNameInfo *dcinfo = NULL;
627         NTSTATUS status;
628         const char *dcname = NULL;
629
630         state->pending_subreq = NULL;
631
632         status = wb_dsgetdcname_recv(subreq, state, &dcinfo);
633         TALLOC_FREE(subreq);
634         if (tevent_req_nterror(req, status)) {
635                 return;
636         }
637         dcname = dcinfo->dc_unc;
638         while (dcname != NULL && *dcname == '\\') {
639                 dcname++;
640         }
641
642         state->r.in.dcname = talloc_strdup(state, dcname);
643         if (tevent_req_nomem(state->r.in.dcname, req)) {
644                 return;
645         }
646
647         TALLOC_FREE(dcinfo);
648
649         subreq = dcerpc_wbint_InitConnection_r_send(state,
650                                                 state->ev,
651                                                 state->child->binding_handle,
652                                                 &state->r);
653         if (tevent_req_nomem(subreq, req)) {
654                 return;
655         }
656         tevent_req_set_callback(subreq, wb_domain_request_initialized, req);
657         state->pending_subreq = subreq;
658 }
659
660 static void wb_domain_request_initialized(struct tevent_req *subreq)
661 {
662         struct tevent_req *req = tevent_req_callback_data(
663                 subreq, struct tevent_req);
664         struct wb_domain_request_state *state = tevent_req_data(
665                 req, struct wb_domain_request_state);
666         NTSTATUS status;
667
668         state->pending_subreq = NULL;
669
670         status = dcerpc_wbint_InitConnection_r_recv(subreq, state);
671         TALLOC_FREE(subreq);
672         if (NT_STATUS_IS_ERR(status)) {
673                 tevent_req_error(req, map_errno_from_nt_status(status));
674                 return;
675         }
676
677         status = state->r.out.result;
678         if (NT_STATUS_IS_ERR(status)) {
679                 tevent_req_error(req, map_errno_from_nt_status(status));
680                 return;
681         }
682
683         state->domain->sid = *state->r.out.sid;
684
685         talloc_free(state->domain->name);
686         state->domain->name = talloc_strdup(state->domain, *state->r.out.name);
687         if (state->domain->name == NULL) {
688                 tevent_req_error(req, ENOMEM);
689                 return;
690         }
691
692         if (*state->r.out.alt_name != NULL &&
693             strlen(*state->r.out.alt_name) > 0) {
694                 talloc_free(state->domain->alt_name);
695
696                 state->domain->alt_name = talloc_strdup(state->domain,
697                                                         *state->r.out.alt_name);
698                 if (state->domain->alt_name == NULL) {
699                         tevent_req_error(req, ENOMEM);
700                         return;
701                 }
702         }
703
704         state->domain->native_mode =
705                         (*state->r.out.flags & WB_DOMINFO_DOMAIN_NATIVE);
706         state->domain->active_directory =
707                         (*state->r.out.flags & WB_DOMINFO_DOMAIN_AD);
708         state->domain->initialized = true;
709
710         subreq = wb_child_request_send(state, state->ev, state->child,
711                                        state->request);
712         if (tevent_req_nomem(subreq, req)) {
713                 return;
714         }
715         tevent_req_set_callback(subreq, wb_domain_request_done, req);
716         state->pending_subreq = subreq;
717
718         /*
719          * Once the domain is initialized and
720          * once we placed our real request into the child queue,
721          * we can remove ourself from the domain queue
722          * and give the next one in the queue the chance
723          * to check for an idle child.
724          */
725         TALLOC_FREE(state->queue_entry);
726 }
727
728 static void wb_domain_request_done(struct tevent_req *subreq)
729 {
730         struct tevent_req *req = tevent_req_callback_data(
731                 subreq, struct tevent_req);
732         struct wb_domain_request_state *state = tevent_req_data(
733                 req, struct wb_domain_request_state);
734         int ret, err;
735
736         state->pending_subreq = NULL;
737
738         ret = wb_child_request_recv(subreq, talloc_tos(), &state->response,
739                                     &err);
740         TALLOC_FREE(subreq);
741         if (ret == -1) {
742                 tevent_req_error(req, err);
743                 return;
744         }
745         tevent_req_done(req);
746 }
747
748 int wb_domain_request_recv(struct tevent_req *req, TALLOC_CTX *mem_ctx,
749                            struct winbindd_response **presponse, int *err)
750 {
751         struct wb_domain_request_state *state = tevent_req_data(
752                 req, struct wb_domain_request_state);
753
754         if (tevent_req_is_unix_error(req, err)) {
755                 return -1;
756         }
757         *presponse = talloc_move(mem_ctx, &state->response);
758         return 0;
759 }
760
761 static void child_process_request(struct winbindd_child *child,
762                                   struct winbindd_cli_state *state)
763 {
764         struct winbindd_domain *domain = child->domain;
765
766         /* Free response data - we may be interrupted and receive another
767            command before being able to send this data off. */
768
769         state->response->result = WINBINDD_ERROR;
770         state->response->length = sizeof(struct winbindd_response);
771
772         /* as all requests in the child are sync, we can use talloc_tos() */
773         state->mem_ctx = talloc_tos();
774
775         /* Process command */
776         state->response->result = winbindd_dual_ndrcmd(domain, state);
777 }
778
779 void setup_child(struct winbindd_domain *domain, struct winbindd_child *child,
780                  const char *logprefix,
781                  const char *logname)
782 {
783         const struct loadparm_substitution *lp_sub =
784                 loadparm_s3_global_substitution();
785
786         if (logprefix && logname) {
787                 char *logbase = NULL;
788
789                 if (*lp_logfile(talloc_tos(), lp_sub)) {
790                         char *end = NULL;
791
792                         if (asprintf(&logbase, "%s", lp_logfile(talloc_tos(), lp_sub)) < 0) {
793                                 smb_panic("Internal error: asprintf failed");
794                         }
795
796                         if ((end = strrchr_m(logbase, '/'))) {
797                                 *end = '\0';
798                         }
799                 } else {
800                         if (asprintf(&logbase, "%s", get_dyn_LOGFILEBASE()) < 0) {
801                                 smb_panic("Internal error: asprintf failed");
802                         }
803                 }
804
805                 if (asprintf(&child->logfilename, "%s/%s-%s",
806                              logbase, logprefix, logname) < 0) {
807                         SAFE_FREE(logbase);
808                         smb_panic("Internal error: asprintf failed");
809                 }
810
811                 SAFE_FREE(logbase);
812         } else {
813                 smb_panic("Internal error: logprefix == NULL && "
814                           "logname == NULL");
815         }
816
817         child->pid = 0;
818         child->sock = -1;
819         child->domain = domain;
820         child->queue = tevent_queue_create(NULL, "winbind_child");
821         SMB_ASSERT(child->queue != NULL);
822
823         child->binding_handle = wbint_binding_handle(NULL, NULL, child);
824         SMB_ASSERT(child->binding_handle != NULL);
825 }
826
827 struct winbind_child_died_state {
828         pid_t pid;
829         struct winbindd_child *child;
830 };
831
832 static bool winbind_child_died_fn(struct winbindd_child *child,
833                                   void *private_data)
834 {
835         struct winbind_child_died_state *state = private_data;
836
837         if (child->pid == state->pid) {
838                 state->child = child;
839                 return false;
840         }
841         return true;
842 }
843
844 void winbind_child_died(pid_t pid)
845 {
846         struct winbind_child_died_state state = { .pid = pid };
847
848         forall_children(winbind_child_died_fn, &state);
849
850         if (state.child == NULL) {
851                 DEBUG(5, ("Already reaped child %u died\n", (unsigned int)pid));
852                 return;
853         }
854
855         state.child->pid = 0;
856 }
857
858 /* Ensure any negative cache entries with the netbios or realm names are removed. */
859
860 void winbindd_flush_negative_conn_cache(struct winbindd_domain *domain)
861 {
862         flush_negative_conn_cache_for_domain(domain->name);
863         if (domain->alt_name != NULL) {
864                 flush_negative_conn_cache_for_domain(domain->alt_name);
865         }
866 }
867
868 /*
869  * Parent winbindd process sets its own debug level first and then
870  * sends a message to all the winbindd children to adjust their debug
871  * level to that of parents.
872  */
873
874 struct winbind_msg_relay_state {
875         struct messaging_context *msg_ctx;
876         uint32_t msg_type;
877         DATA_BLOB *data;
878 };
879
880 static bool winbind_msg_relay_fn(struct winbindd_child *child,
881                                  void *private_data)
882 {
883         struct winbind_msg_relay_state *state = private_data;
884
885         DBG_DEBUG("sending message to pid %u.\n",
886                   (unsigned int)child->pid);
887
888         messaging_send(state->msg_ctx, pid_to_procid(child->pid),
889                        state->msg_type, state->data);
890         return true;
891 }
892
893 void winbind_msg_debug(struct messaging_context *msg_ctx,
894                          void *private_data,
895                          uint32_t msg_type,
896                          struct server_id server_id,
897                          DATA_BLOB *data)
898 {
899         struct winbind_msg_relay_state state = {
900                 .msg_ctx = msg_ctx, .msg_type = msg_type, .data = data
901         };
902
903         DEBUG(10,("winbind_msg_debug: got debug message.\n"));
904
905         debug_message(msg_ctx, private_data, MSG_DEBUG, server_id, data);
906
907         forall_children(winbind_msg_relay_fn, &state);
908 }
909
910 void winbind_disconnect_dc_parent(struct messaging_context *msg_ctx,
911                                   void *private_data,
912                                   uint32_t msg_type,
913                                   struct server_id server_id,
914                                   DATA_BLOB *data)
915 {
916         struct winbind_msg_relay_state state = {
917                 .msg_ctx = msg_ctx, .msg_type = msg_type, .data = data
918         };
919
920         DBG_DEBUG("Got disconnect_dc message\n");
921
922         forall_children(winbind_msg_relay_fn, &state);
923 }
924
925 static void winbindd_msg_reload_services_child(struct messaging_context *msg,
926                                                void *private_data,
927                                                uint32_t msg_type,
928                                                struct server_id server_id,
929                                                DATA_BLOB *data)
930 {
931         DBG_DEBUG("Got reload-config message\n");
932         winbindd_reload_services_file((const char *)private_data);
933 }
934
935 /* React on 'smbcontrol winbindd reload-config' in the same way as on SIGHUP*/
936 void winbindd_msg_reload_services_parent(struct messaging_context *msg,
937                                          void *private_data,
938                                          uint32_t msg_type,
939                                          struct server_id server_id,
940                                          DATA_BLOB *data)
941 {
942         struct winbind_msg_relay_state state = {
943                 .msg_ctx = msg,
944                 .msg_type = msg_type,
945                 .data = data,
946         };
947
948         DBG_DEBUG("Got reload-config message\n");
949
950         /* Flush various caches */
951         winbindd_flush_caches();
952
953         winbindd_reload_services_file((const char *)private_data);
954
955         forall_children(winbind_msg_relay_fn, &state);
956 }
957
958 /* Set our domains as offline and forward the offline message to our children. */
959
960 struct winbind_msg_on_offline_state {
961         struct messaging_context *msg_ctx;
962         uint32_t msg_type;
963 };
964
965 static bool winbind_msg_on_offline_fn(struct winbindd_child *child,
966                                       void *private_data)
967 {
968         struct winbind_msg_on_offline_state *state = private_data;
969
970         if (child->domain->internal) {
971                 return true;
972         }
973
974         /*
975          * Each winbindd child should only process requests for one
976          * domain - make sure we only set it online / offline for that
977          * domain.
978          */
979         DBG_DEBUG("sending message to pid %u for domain %s.\n",
980                   (unsigned int)child->pid, child->domain->name);
981
982         messaging_send_buf(state->msg_ctx,
983                            pid_to_procid(child->pid),
984                            state->msg_type,
985                            (const uint8_t *)child->domain->name,
986                            strlen(child->domain->name)+1);
987
988         return true;
989 }
990
991 void winbind_msg_offline(struct messaging_context *msg_ctx,
992                          void *private_data,
993                          uint32_t msg_type,
994                          struct server_id server_id,
995                          DATA_BLOB *data)
996 {
997         struct winbind_msg_on_offline_state state = {
998                 .msg_ctx = msg_ctx,
999                 .msg_type = MSG_WINBIND_OFFLINE,
1000         };
1001         struct winbindd_domain *domain;
1002
1003         DEBUG(10,("winbind_msg_offline: got offline message.\n"));
1004
1005         if (!lp_winbind_offline_logon()) {
1006                 DEBUG(10,("winbind_msg_offline: rejecting offline message.\n"));
1007                 return;
1008         }
1009
1010         /* Set our global state as offline. */
1011         if (!set_global_winbindd_state_offline()) {
1012                 DEBUG(10,("winbind_msg_offline: offline request failed.\n"));
1013                 return;
1014         }
1015
1016         /* Set all our domains as offline. */
1017         for (domain = domain_list(); domain; domain = domain->next) {
1018                 if (domain->internal) {
1019                         continue;
1020                 }
1021                 DEBUG(5,("winbind_msg_offline: marking %s offline.\n", domain->name));
1022                 domain->online = false;
1023         }
1024
1025         forall_domain_children(winbind_msg_on_offline_fn, &state);
1026 }
1027
1028 /* Set our domains as online and forward the online message to our children. */
1029
1030 void winbind_msg_online(struct messaging_context *msg_ctx,
1031                         void *private_data,
1032                         uint32_t msg_type,
1033                         struct server_id server_id,
1034                         DATA_BLOB *data)
1035 {
1036         struct winbind_msg_on_offline_state state = {
1037                 .msg_ctx = msg_ctx,
1038                 .msg_type = MSG_WINBIND_ONLINE,
1039         };
1040
1041         DEBUG(10,("winbind_msg_online: got online message.\n"));
1042
1043         if (!lp_winbind_offline_logon()) {
1044                 DEBUG(10,("winbind_msg_online: rejecting online message.\n"));
1045                 return;
1046         }
1047
1048         /* Set our global state as online. */
1049         set_global_winbindd_state_online();
1050
1051         smb_nscd_flush_user_cache();
1052         smb_nscd_flush_group_cache();
1053
1054         /* Tell all our child domains to go online online. */
1055         forall_domain_children(winbind_msg_on_offline_fn, &state);
1056 }
1057
1058 static const char *collect_onlinestatus(TALLOC_CTX *mem_ctx)
1059 {
1060         struct winbindd_domain *domain;
1061         char *buf = NULL;
1062
1063         if ((buf = talloc_asprintf(mem_ctx, "global:%s ",
1064                                    get_global_winbindd_state_offline() ?
1065                                    "Offline":"Online")) == NULL) {
1066                 return NULL;
1067         }
1068
1069         for (domain = domain_list(); domain; domain = domain->next) {
1070                 if ((buf = talloc_asprintf_append_buffer(buf, "%s:%s ",
1071                                                   domain->name,
1072                                                   domain->online ?
1073                                                   "Online":"Offline")) == NULL) {
1074                         return NULL;
1075                 }
1076         }
1077
1078         buf = talloc_asprintf_append_buffer(buf, "\n");
1079
1080         DEBUG(5,("collect_onlinestatus: %s", buf));
1081
1082         return buf;
1083 }
1084
1085 void winbind_msg_onlinestatus(struct messaging_context *msg_ctx,
1086                               void *private_data,
1087                               uint32_t msg_type,
1088                               struct server_id server_id,
1089                               DATA_BLOB *data)
1090 {
1091         TALLOC_CTX *mem_ctx;
1092         const char *message;
1093
1094         DEBUG(5,("winbind_msg_onlinestatus received.\n"));
1095
1096         mem_ctx = talloc_init("winbind_msg_onlinestatus");
1097         if (mem_ctx == NULL) {
1098                 return;
1099         }
1100
1101         message = collect_onlinestatus(mem_ctx);
1102         if (message == NULL) {
1103                 talloc_destroy(mem_ctx);
1104                 return;
1105         }
1106
1107         messaging_send_buf(msg_ctx, server_id, MSG_WINBIND_ONLINESTATUS,
1108                            (const uint8_t *)message, strlen(message) + 1);
1109
1110         talloc_destroy(mem_ctx);
1111 }
1112
1113 void winbind_msg_dump_domain_list(struct messaging_context *msg_ctx,
1114                                   void *private_data,
1115                                   uint32_t msg_type,
1116                                   struct server_id server_id,
1117                                   DATA_BLOB *data)
1118 {
1119         TALLOC_CTX *mem_ctx;
1120         const char *message = NULL;
1121         const char *domain = NULL;
1122         char *s = NULL;
1123         NTSTATUS status;
1124         struct winbindd_domain *dom = NULL;
1125
1126         DEBUG(5,("winbind_msg_dump_domain_list received.\n"));
1127
1128         mem_ctx = talloc_init("winbind_msg_dump_domain_list");
1129         if (!mem_ctx) {
1130                 return;
1131         }
1132
1133         if (data->length > 0) {
1134                 domain = (const char *)data->data;
1135         }
1136
1137         if (domain) {
1138
1139                 DEBUG(5,("winbind_msg_dump_domain_list for domain: %s\n",
1140                         domain));
1141
1142                 message = NDR_PRINT_STRUCT_STRING(mem_ctx, winbindd_domain,
1143                                                   find_domain_from_name_noinit(domain));
1144                 if (!message) {
1145                         talloc_destroy(mem_ctx);
1146                         return;
1147                 }
1148
1149                 messaging_send_buf(msg_ctx, server_id,
1150                                    MSG_WINBIND_DUMP_DOMAIN_LIST,
1151                                    (const uint8_t *)message, strlen(message) + 1);
1152
1153                 talloc_destroy(mem_ctx);
1154
1155                 return;
1156         }
1157
1158         DEBUG(5,("winbind_msg_dump_domain_list all domains\n"));
1159
1160         for (dom = domain_list(); dom; dom=dom->next) {
1161                 message = NDR_PRINT_STRUCT_STRING(mem_ctx, winbindd_domain, dom);
1162                 if (!message) {
1163                         talloc_destroy(mem_ctx);
1164                         return;
1165                 }
1166
1167                 s = talloc_asprintf_append(s, "%s\n", message);
1168                 if (!s) {
1169                         talloc_destroy(mem_ctx);
1170                         return;
1171                 }
1172         }
1173
1174         status = messaging_send_buf(msg_ctx, server_id,
1175                                     MSG_WINBIND_DUMP_DOMAIN_LIST,
1176                                     (uint8_t *)s, strlen(s) + 1);
1177         if (!NT_STATUS_IS_OK(status)) {
1178                 DEBUG(0,("failed to send message: %s\n",
1179                 nt_errstr(status)));
1180         }
1181
1182         talloc_destroy(mem_ctx);
1183 }
1184
1185 static void account_lockout_policy_handler(struct tevent_context *ctx,
1186                                            struct tevent_timer *te,
1187                                            struct timeval now,
1188                                            void *private_data)
1189 {
1190         struct winbindd_child *child =
1191                 (struct winbindd_child *)private_data;
1192         TALLOC_CTX *mem_ctx = NULL;
1193         struct samr_DomInfo12 lockout_policy;
1194         NTSTATUS result;
1195
1196         DEBUG(10,("account_lockout_policy_handler called\n"));
1197
1198         TALLOC_FREE(child->lockout_policy_event);
1199
1200         if ( !winbindd_can_contact_domain( child->domain ) ) {
1201                 DEBUG(10,("account_lockout_policy_handler: Removing myself since I "
1202                           "do not have an incoming trust to domain %s\n",
1203                           child->domain->name));
1204
1205                 return;
1206         }
1207
1208         mem_ctx = talloc_init("account_lockout_policy_handler ctx");
1209         if (!mem_ctx) {
1210                 result = NT_STATUS_NO_MEMORY;
1211         } else {
1212                 result = wb_cache_lockout_policy(child->domain, mem_ctx,
1213                                                  &lockout_policy);
1214         }
1215         TALLOC_FREE(mem_ctx);
1216
1217         if (!NT_STATUS_IS_OK(result)) {
1218                 DEBUG(10,("account_lockout_policy_handler: lockout_policy failed error %s\n",
1219                          nt_errstr(result)));
1220         }
1221
1222         child->lockout_policy_event = tevent_add_timer(global_event_context(), NULL,
1223                                                       timeval_current_ofs(3600, 0),
1224                                                       account_lockout_policy_handler,
1225                                                       child);
1226 }
1227
1228 static time_t get_machine_password_timeout(void)
1229 {
1230         /* until we have gpo support use lp setting */
1231         return lp_machine_password_timeout();
1232 }
1233
1234 static bool calculate_next_machine_pwd_change(const char *domain,
1235                                               struct timeval *t)
1236 {
1237         time_t pass_last_set_time;
1238         time_t timeout;
1239         time_t next_change;
1240         struct timeval tv;
1241         char *pw;
1242
1243         pw = secrets_fetch_machine_password(domain,
1244                                             &pass_last_set_time,
1245                                             NULL);
1246
1247         if (pw == NULL) {
1248                 DEBUG(0,("cannot fetch own machine password ????"));
1249                 return false;
1250         }
1251
1252         SAFE_FREE(pw);
1253
1254         timeout = get_machine_password_timeout();
1255         if (timeout == 0) {
1256                 DEBUG(10,("machine password never expires\n"));
1257                 return false;
1258         }
1259
1260         tv.tv_sec = pass_last_set_time;
1261         DEBUG(10, ("password last changed %s\n",
1262                    timeval_string(talloc_tos(), &tv, false)));
1263         tv.tv_sec += timeout;
1264         DEBUGADD(10, ("password valid until %s\n",
1265                       timeval_string(talloc_tos(), &tv, false)));
1266
1267         if (time(NULL) < (pass_last_set_time + timeout)) {
1268                 next_change = pass_last_set_time + timeout;
1269                 DEBUG(10,("machine password still valid until: %s\n",
1270                         http_timestring(talloc_tos(), next_change)));
1271                 *t = timeval_set(next_change, 0);
1272
1273                 if (lp_clustering()) {
1274                         uint8_t randbuf;
1275                         /*
1276                          * When having a cluster, we have several
1277                          * winbinds racing for the password change. In
1278                          * the machine_password_change_handler()
1279                          * function we check if someone else was
1280                          * faster when the event triggers. We add a
1281                          * 255-second random delay here, so that we
1282                          * don't run to change the password at the
1283                          * exact same moment.
1284                          */
1285                         generate_random_buffer(&randbuf, sizeof(randbuf));
1286                         DEBUG(10, ("adding %d seconds randomness\n",
1287                                    (int)randbuf));
1288                         t->tv_sec += randbuf;
1289                 }
1290                 return true;
1291         }
1292
1293         DEBUG(10,("machine password expired, needs immediate change\n"));
1294
1295         *t = timeval_zero();
1296
1297         return true;
1298 }
1299
1300 static void machine_password_change_handler(struct tevent_context *ctx,
1301                                             struct tevent_timer *te,
1302                                             struct timeval now,
1303                                             void *private_data)
1304 {
1305         struct messaging_context *msg_ctx = global_messaging_context();
1306         struct winbindd_child *child =
1307                 (struct winbindd_child *)private_data;
1308         struct rpc_pipe_client *netlogon_pipe = NULL;
1309         struct netlogon_creds_cli_context *netlogon_creds_ctx = NULL;
1310         NTSTATUS result;
1311         struct timeval next_change;
1312
1313         DEBUG(10,("machine_password_change_handler called\n"));
1314
1315         TALLOC_FREE(child->machine_password_change_event);
1316
1317         if (!calculate_next_machine_pwd_change(child->domain->name,
1318                                                &next_change)) {
1319                 DEBUG(10, ("calculate_next_machine_pwd_change failed\n"));
1320                 return;
1321         }
1322
1323         DEBUG(10, ("calculate_next_machine_pwd_change returned %s\n",
1324                    timeval_string(talloc_tos(), &next_change, false)));
1325
1326         if (!timeval_expired(&next_change)) {
1327                 DEBUG(10, ("Someone else has already changed the pw\n"));
1328                 goto done;
1329         }
1330
1331         if (!winbindd_can_contact_domain(child->domain)) {
1332                 DEBUG(10,("machine_password_change_handler: Removing myself since I "
1333                           "do not have an incoming trust to domain %s\n",
1334                           child->domain->name));
1335                 return;
1336         }
1337
1338         result = cm_connect_netlogon_secure(child->domain,
1339                                             &netlogon_pipe,
1340                                             &netlogon_creds_ctx);
1341         if (!NT_STATUS_IS_OK(result)) {
1342                 DEBUG(10,("machine_password_change_handler: "
1343                         "failed to connect netlogon pipe: %s\n",
1344                          nt_errstr(result)));
1345                 return;
1346         }
1347
1348         result = trust_pw_change(netlogon_creds_ctx,
1349                                  msg_ctx,
1350                                  netlogon_pipe->binding_handle,
1351                                  child->domain->name,
1352                                  child->domain->dcname,
1353                                  false); /* force */
1354
1355         DEBUG(10, ("machine_password_change_handler: "
1356                    "trust_pw_change returned %s\n",
1357                    nt_errstr(result)));
1358
1359         if (NT_STATUS_EQUAL(result, NT_STATUS_ACCESS_DENIED) ) {
1360                 DEBUG(3,("machine_password_change_handler: password set returned "
1361                          "ACCESS_DENIED.  Maybe the trust account "
1362                          "password was changed and we didn't know it. "
1363                          "Killing connections to domain %s\n",
1364                          child->domain->name));
1365                 invalidate_cm_connection(child->domain);
1366         }
1367
1368         if (!calculate_next_machine_pwd_change(child->domain->name,
1369                                                &next_change)) {
1370                 DEBUG(10, ("calculate_next_machine_pwd_change failed\n"));
1371                 return;
1372         }
1373
1374         DEBUG(10, ("calculate_next_machine_pwd_change returned %s\n",
1375                    timeval_string(talloc_tos(), &next_change, false)));
1376
1377         if (!NT_STATUS_IS_OK(result)) {
1378                 struct timeval tmp;
1379                 /*
1380                  * In case of failure, give the DC a minute to recover
1381                  */
1382                 tmp = timeval_current_ofs(60, 0);
1383                 next_change = timeval_max(&next_change, &tmp);
1384         }
1385
1386 done:
1387         child->machine_password_change_event = tevent_add_timer(global_event_context(), NULL,
1388                                                               next_change,
1389                                                               machine_password_change_handler,
1390                                                               child);
1391 }
1392
1393 /* Deal with a request to go offline. */
1394
1395 static void child_msg_offline(struct messaging_context *msg,
1396                               void *private_data,
1397                               uint32_t msg_type,
1398                               struct server_id server_id,
1399                               DATA_BLOB *data)
1400 {
1401         struct winbindd_domain *domain;
1402         struct winbindd_domain *primary_domain = NULL;
1403         const char *domainname = (const char *)data->data;
1404
1405         if (data->data == NULL || data->length == 0) {
1406                 return;
1407         }
1408
1409         DEBUG(5,("child_msg_offline received for domain %s.\n", domainname));
1410
1411         if (!lp_winbind_offline_logon()) {
1412                 DEBUG(10,("child_msg_offline: rejecting offline message.\n"));
1413                 return;
1414         }
1415
1416         primary_domain = find_our_domain();
1417
1418         /* Mark the requested domain offline. */
1419
1420         for (domain = domain_list(); domain; domain = domain->next) {
1421                 if (domain->internal) {
1422                         continue;
1423                 }
1424                 if (strequal(domain->name, domainname)) {
1425                         DEBUG(5,("child_msg_offline: marking %s offline.\n", domain->name));
1426                         set_domain_offline(domain);
1427                         /* we are in the trusted domain, set the primary domain
1428                          * offline too */
1429                         if (domain != primary_domain) {
1430                                 set_domain_offline(primary_domain);
1431                         }
1432                 }
1433         }
1434 }
1435
1436 /* Deal with a request to go online. */
1437
1438 static void child_msg_online(struct messaging_context *msg,
1439                              void *private_data,
1440                              uint32_t msg_type,
1441                              struct server_id server_id,
1442                              DATA_BLOB *data)
1443 {
1444         struct winbindd_domain *domain;
1445         struct winbindd_domain *primary_domain = NULL;
1446         const char *domainname = (const char *)data->data;
1447
1448         if (data->data == NULL || data->length == 0) {
1449                 return;
1450         }
1451
1452         DEBUG(5,("child_msg_online received for domain %s.\n", domainname));
1453
1454         if (!lp_winbind_offline_logon()) {
1455                 DEBUG(10,("child_msg_online: rejecting online message.\n"));
1456                 return;
1457         }
1458
1459         primary_domain = find_our_domain();
1460
1461         /* Set our global state as online. */
1462         set_global_winbindd_state_online();
1463
1464         /* Try and mark everything online - delete any negative cache entries
1465            to force a reconnect now. */
1466
1467         for (domain = domain_list(); domain; domain = domain->next) {
1468                 if (domain->internal) {
1469                         continue;
1470                 }
1471                 if (strequal(domain->name, domainname)) {
1472                         DEBUG(5,("child_msg_online: requesting %s to go online.\n", domain->name));
1473                         winbindd_flush_negative_conn_cache(domain);
1474                         set_domain_online_request(domain);
1475
1476                         /* we can be in trusted domain, which will contact primary domain
1477                          * we have to bring primary domain online in trusted domain process
1478                          * see, winbindd_dual_pam_auth() --> winbindd_dual_pam_auth_samlogon()
1479                          * --> contact_domain = find_our_domain()
1480                          * */
1481                         if (domain != primary_domain) {
1482                                 winbindd_flush_negative_conn_cache(primary_domain);
1483                                 set_domain_online_request(primary_domain);
1484                         }
1485                 }
1486         }
1487 }
1488
1489 struct winbindd_reinit_after_fork_state {
1490         const struct winbindd_child *myself;
1491 };
1492
1493 static bool winbindd_reinit_after_fork_fn(struct winbindd_child *child,
1494                                           void *private_data)
1495 {
1496         struct winbindd_reinit_after_fork_state *state = private_data;
1497
1498         if (child == state->myself) {
1499                 return true;
1500         }
1501
1502         /* Destroy all possible events in child list. */
1503         TALLOC_FREE(child->lockout_policy_event);
1504         TALLOC_FREE(child->machine_password_change_event);
1505
1506         /*
1507          * Children should never be able to send each other messages,
1508          * all messages must go through the parent.
1509          */
1510         child->pid = (pid_t)0;
1511
1512         /*
1513          * Close service sockets to all other children
1514          */
1515         if (child->sock != -1) {
1516                 close(child->sock);
1517                 child->sock = -1;
1518         }
1519
1520         return true;
1521 }
1522
1523 NTSTATUS winbindd_reinit_after_fork(const struct winbindd_child *myself,
1524                                     const char *logfilename)
1525 {
1526         struct winbindd_reinit_after_fork_state state = { .myself = myself };
1527         struct winbindd_domain *domain;
1528         NTSTATUS status;
1529
1530         status = reinit_after_fork(
1531                 global_messaging_context(),
1532                 global_event_context(),
1533                 true);
1534         if (!NT_STATUS_IS_OK(status)) {
1535                 DEBUG(0,("reinit_after_fork() failed\n"));
1536                 return status;
1537         }
1538         initialize_password_db(true, global_event_context());
1539
1540         close_conns_after_fork();
1541
1542         if (logfilename != NULL) {
1543                 lp_set_logfile(logfilename);
1544                 reopen_logs();
1545         }
1546
1547         if (!winbindd_setup_sig_term_handler(false)) {
1548                 return NT_STATUS_NO_MEMORY;
1549         }
1550
1551         if (!winbindd_setup_sig_hup_handler(logfilename)) {
1552                 return NT_STATUS_NO_MEMORY;
1553         }
1554
1555         /* Stop zombies in children */
1556         CatchChild();
1557
1558         /* Don't handle the same messages as our parent. */
1559         messaging_deregister(global_messaging_context(),
1560                              MSG_SMB_CONF_UPDATED, NULL);
1561         messaging_deregister(global_messaging_context(),
1562                              MSG_SHUTDOWN, NULL);
1563         messaging_deregister(global_messaging_context(),
1564                              MSG_WINBIND_OFFLINE, NULL);
1565         messaging_deregister(global_messaging_context(),
1566                              MSG_WINBIND_ONLINE, NULL);
1567         messaging_deregister(global_messaging_context(),
1568                              MSG_WINBIND_ONLINESTATUS, NULL);
1569         messaging_deregister(global_messaging_context(),
1570                              MSG_WINBIND_DUMP_DOMAIN_LIST, NULL);
1571         messaging_deregister(global_messaging_context(),
1572                              MSG_DEBUG, NULL);
1573
1574         messaging_deregister(global_messaging_context(),
1575                              MSG_WINBIND_DOMAIN_OFFLINE, NULL);
1576         messaging_deregister(global_messaging_context(),
1577                              MSG_WINBIND_DOMAIN_ONLINE, NULL);
1578
1579         /* We have destroyed all events in the winbindd_event_context
1580          * in reinit_after_fork(), so clean out all possible pending
1581          * event pointers. */
1582
1583         /* Deal with check_online_events. */
1584
1585         for (domain = domain_list(); domain; domain = domain->next) {
1586                 TALLOC_FREE(domain->check_online_event);
1587         }
1588
1589         /* Ensure we're not handling a credential cache event inherited
1590          * from our parent. */
1591
1592         ccache_remove_all_after_fork();
1593
1594         forall_children(winbindd_reinit_after_fork_fn, &state);
1595
1596         return NT_STATUS_OK;
1597 }
1598
1599 /*
1600  * In a child there will be only one domain, reference that here.
1601  */
1602 static struct winbindd_domain *child_domain;
1603
1604 struct winbindd_domain *wb_child_domain(void)
1605 {
1606         return child_domain;
1607 }
1608
1609 struct child_handler_state {
1610         struct winbindd_child *child;
1611         struct winbindd_cli_state cli;
1612 };
1613
1614 static void child_handler(struct tevent_context *ev, struct tevent_fd *fde,
1615                           uint16_t flags, void *private_data)
1616 {
1617         struct child_handler_state *state =
1618                 (struct child_handler_state *)private_data;
1619         NTSTATUS status;
1620         uint64_t parent_traceid;
1621
1622         /* fetch a request from the main daemon */
1623         status = child_read_request(state->cli.sock, state->cli.request);
1624
1625         if (!NT_STATUS_IS_OK(status)) {
1626                 /* we lost contact with our parent */
1627                 _exit(0);
1628         }
1629
1630         /* read traceid from request */
1631         parent_traceid = state->cli.request->traceid;
1632         debug_traceid_set(parent_traceid);
1633
1634         DEBUG(4,("child daemon request %d\n",
1635                  (int)state->cli.request->cmd));
1636
1637         ZERO_STRUCTP(state->cli.response);
1638         state->cli.request->null_term = '\0';
1639         state->cli.mem_ctx = talloc_tos();
1640         child_process_request(state->child, &state->cli);
1641
1642         DEBUG(4, ("Finished processing child request %d\n",
1643                   (int)state->cli.request->cmd));
1644
1645         SAFE_FREE(state->cli.request->extra_data.data);
1646
1647         status = child_write_response(state->cli.sock, state->cli.response);
1648         if (!NT_STATUS_IS_OK(status)) {
1649                 exit(1);
1650         }
1651 }
1652
1653 static bool fork_domain_child(struct winbindd_child *child)
1654 {
1655         int fdpair[2];
1656         struct child_handler_state state;
1657         struct winbindd_request request;
1658         struct winbindd_response response;
1659         struct winbindd_domain *primary_domain = NULL;
1660         NTSTATUS status;
1661         ssize_t nwritten;
1662         struct tevent_fd *fde;
1663
1664         if (child->domain) {
1665                 DEBUG(10, ("fork_domain_child called for domain '%s'\n",
1666                            child->domain->name));
1667         } else {
1668                 DEBUG(10, ("fork_domain_child called without domain.\n"));
1669         }
1670
1671         if (socketpair(AF_UNIX, SOCK_STREAM, 0, fdpair) != 0) {
1672                 DEBUG(0, ("Could not open child pipe: %s\n",
1673                           strerror(errno)));
1674                 return False;
1675         }
1676
1677         ZERO_STRUCT(state);
1678         state.child = child;
1679         state.cli.pid = getpid();
1680         state.cli.request = &request;
1681         state.cli.response = &response;
1682
1683         child->pid = fork();
1684
1685         if (child->pid == -1) {
1686                 DEBUG(0, ("Could not fork: %s\n", strerror(errno)));
1687                 close(fdpair[0]);
1688                 close(fdpair[1]);
1689                 return False;
1690         }
1691
1692         if (child->pid != 0) {
1693                 /* Parent */
1694                 ssize_t nread;
1695                 int rc;
1696
1697                 close(fdpair[0]);
1698
1699                 nread = sys_read(fdpair[1], &status, sizeof(status));
1700                 if (nread != sizeof(status)) {
1701                         DEBUG(1, ("fork_domain_child: Could not read child status: "
1702                                   "nread=%d, error=%s\n", (int)nread,
1703                                   strerror(errno)));
1704                         close(fdpair[1]);
1705                         return false;
1706                 }
1707                 if (!NT_STATUS_IS_OK(status)) {
1708                         DEBUG(1, ("fork_domain_child: Child status is %s\n",
1709                                   nt_errstr(status)));
1710                         close(fdpair[1]);
1711                         return false;
1712                 }
1713
1714                 child->monitor_fde = tevent_add_fd(global_event_context(),
1715                                                    global_event_context(),
1716                                                    fdpair[1],
1717                                                    TEVENT_FD_READ,
1718                                                    child_socket_readable,
1719                                                    child);
1720                 if (child->monitor_fde == NULL) {
1721                         DBG_WARNING("tevent_add_fd failed\n");
1722                         close(fdpair[1]);
1723                         return false;
1724                 }
1725
1726                 rc = set_blocking(fdpair[1], false);
1727                 if (rc < 0) {
1728                         close(fdpair[1]);
1729                         return false;
1730                 }
1731
1732                 child->sock = fdpair[1];
1733
1734                 return true;
1735         }
1736
1737         /* Child */
1738         child_domain = child->domain;
1739
1740         DEBUG(10, ("Child process %d\n", (int)getpid()));
1741
1742         state.cli.sock = fdpair[0];
1743         close(fdpair[1]);
1744
1745         status = winbindd_reinit_after_fork(child, child->logfilename);
1746
1747         /* setup callbacks again, one of them is removed in reinit_after_fork */
1748         if (lp_winbind_debug_traceid()) {
1749                 winbind_debug_traceid_setup(global_event_context());
1750         }
1751
1752         nwritten = sys_write(state.cli.sock, &status, sizeof(status));
1753         if (nwritten != sizeof(status)) {
1754                 DEBUG(1, ("fork_domain_child: Could not write status: "
1755                           "nwritten=%d, error=%s\n", (int)nwritten,
1756                           strerror(errno)));
1757                 _exit(0);
1758         }
1759         if (!NT_STATUS_IS_OK(status)) {
1760                 DEBUG(1, ("winbindd_reinit_after_fork failed: %s\n",
1761                           nt_errstr(status)));
1762                 _exit(0);
1763         }
1764
1765         if (child_domain != NULL) {
1766                 process_set_title("wb[%s]", "domain child [%s]", child_domain->name);
1767         } else if (is_idmap_child(child)) {
1768                 process_set_title("wb-idmap", "idmap child");
1769         }
1770
1771         /* Handle online/offline messages. */
1772         messaging_register(global_messaging_context(), NULL,
1773                            MSG_WINBIND_OFFLINE, child_msg_offline);
1774         messaging_register(global_messaging_context(), NULL,
1775                            MSG_WINBIND_ONLINE, child_msg_online);
1776         messaging_register(global_messaging_context(), NULL,
1777                            MSG_DEBUG, debug_message);
1778         messaging_register(global_messaging_context(), NULL,
1779                            MSG_WINBIND_IP_DROPPED,
1780                            winbind_msg_ip_dropped);
1781         messaging_register(global_messaging_context(), NULL,
1782                            MSG_WINBIND_DISCONNECT_DC,
1783                            winbind_msg_disconnect_dc);
1784         messaging_register(
1785                 global_messaging_context(),
1786                 child->logfilename,
1787                 MSG_SMB_CONF_UPDATED,
1788                 winbindd_msg_reload_services_child);
1789
1790         primary_domain = find_our_domain();
1791
1792         if (primary_domain == NULL) {
1793                 smb_panic("no primary domain found");
1794         }
1795
1796         /* It doesn't matter if we allow cache login,
1797          * try to bring domain online after fork. */
1798         if ( child->domain ) {
1799                 child->domain->startup = True;
1800                 child->domain->startup_time = time_mono(NULL);
1801                 /* we can be in primary domain or in trusted domain
1802                  * If we are in trusted domain, set the primary domain
1803                  * in start-up mode */
1804                 if (!(child->domain->internal)) {
1805                         set_domain_online_request(child->domain);
1806                         if (!(child->domain->primary)) {
1807                                 primary_domain->startup = True;
1808                                 primary_domain->startup_time = time_mono(NULL);
1809                                 set_domain_online_request(primary_domain);
1810                         }
1811                 }
1812         }
1813
1814         /*
1815          * We are in idmap child, bring primary domain online.
1816          */
1817         if (is_idmap_child(child)) {
1818                 set_domain_online_request(primary_domain);
1819         }
1820
1821         /* We might be in the idmap child...*/
1822         if (child->domain && !(child->domain->internal) &&
1823             lp_winbind_offline_logon()) {
1824
1825                 set_domain_online_request(child->domain);
1826
1827                 if (primary_domain && (primary_domain != child->domain)) {
1828                         /* We need to talk to the primary
1829                          * domain as well as the trusted
1830                          * domain inside a trusted domain
1831                          * child.
1832                          * See the code in :
1833                          * set_dc_type_and_flags_trustinfo()
1834                          * for details.
1835                          */
1836                         set_domain_online_request(primary_domain);
1837                 }
1838
1839                 child->lockout_policy_event = tevent_add_timer(
1840                         global_event_context(), NULL, timeval_zero(),
1841                         account_lockout_policy_handler,
1842                         child);
1843         }
1844
1845         if (child->domain && child->domain->primary &&
1846             !USE_KERBEROS_KEYTAB &&
1847             lp_server_role() == ROLE_DOMAIN_MEMBER) {
1848
1849                 struct timeval next_change;
1850
1851                 if (calculate_next_machine_pwd_change(child->domain->name,
1852                                                        &next_change)) {
1853                         child->machine_password_change_event = tevent_add_timer(
1854                                 global_event_context(), NULL, next_change,
1855                                 machine_password_change_handler,
1856                                 child);
1857                 }
1858         }
1859
1860         fde = tevent_add_fd(global_event_context(), NULL, state.cli.sock,
1861                             TEVENT_FD_READ, child_handler, &state);
1862         if (fde == NULL) {
1863                 DEBUG(1, ("tevent_add_fd failed\n"));
1864                 _exit(1);
1865         }
1866
1867         while (1) {
1868
1869                 int ret;
1870                 TALLOC_CTX *frame = talloc_stackframe();
1871
1872                 ret = tevent_loop_once(global_event_context());
1873                 if (ret != 0) {
1874                         DEBUG(1, ("tevent_loop_once failed: %s\n",
1875                                   strerror(errno)));
1876                         _exit(1);
1877                 }
1878
1879                 if (child->domain && child->domain->startup &&
1880                                 (time_mono(NULL) > child->domain->startup_time + 30)) {
1881                         /* No longer in "startup" mode. */
1882                         DEBUG(10,("fork_domain_child: domain %s no longer in 'startup' mode.\n",
1883                                 child->domain->name ));
1884                         child->domain->startup = False;
1885                 }
1886
1887                 TALLOC_FREE(frame);
1888         }
1889 }
1890
1891 void winbind_msg_ip_dropped_parent(struct messaging_context *msg_ctx,
1892                                    void *private_data,
1893                                    uint32_t msg_type,
1894                                    struct server_id server_id,
1895                                    DATA_BLOB *data)
1896 {
1897         struct winbind_msg_relay_state state = {
1898                 .msg_ctx = msg_ctx,
1899                 .msg_type = msg_type,
1900                 .data = data,
1901         };
1902
1903         winbind_msg_ip_dropped(msg_ctx, private_data, msg_type,
1904                                server_id, data);
1905
1906         forall_children(winbind_msg_relay_fn, &state);
1907 }
1908
1909 void winbindd_terminate(bool is_parent)
1910 {
1911         if (is_parent) {
1912                 /* When parent goes away we should
1913                  * remove the socket file. Not so
1914                  * when children terminate.
1915                  */
1916                 char *path = NULL;
1917
1918                 if (asprintf(&path, "%s/%s",
1919                         lp_winbindd_socket_directory(), WINBINDD_SOCKET_NAME) > 0) {
1920                         unlink(path);
1921                         SAFE_FREE(path);
1922                 }
1923         }
1924
1925         idmap_close();
1926
1927         netlogon_creds_cli_close_global_db();
1928
1929 #if 0
1930         if (interactive) {
1931                 TALLOC_CTX *mem_ctx = talloc_init("end_description");
1932                 char *description = talloc_describe_all(mem_ctx);
1933
1934                 DEBUG(3, ("tallocs left:\n%s\n", description));
1935                 talloc_destroy(mem_ctx);
1936         }
1937 #endif
1938
1939         if (is_parent) {
1940                 pidfile_unlink(lp_pid_directory(), "winbindd");
1941         }
1942
1943         exit(0);
1944 }
1945
1946 static void winbindd_sig_term_handler(struct tevent_context *ev,
1947                                       struct tevent_signal *se,
1948                                       int signum,
1949                                       int count,
1950                                       void *siginfo,
1951                                       void *private_data)
1952 {
1953         bool *p = talloc_get_type_abort(private_data, bool);
1954         bool is_parent = *p;
1955
1956         TALLOC_FREE(p);
1957
1958         DEBUG(0,("Got sig[%d] terminate (is_parent=%d)\n",
1959                  signum, is_parent));
1960         winbindd_terminate(is_parent);
1961 }
1962
1963 bool winbindd_setup_sig_term_handler(bool parent)
1964 {
1965         struct tevent_signal *se;
1966         bool *is_parent;
1967
1968         is_parent = talloc(global_event_context(), bool);
1969         if (!is_parent) {
1970                 return false;
1971         }
1972
1973         *is_parent = parent;
1974
1975         se = tevent_add_signal(global_event_context(),
1976                                is_parent,
1977                                SIGTERM, 0,
1978                                winbindd_sig_term_handler,
1979                                is_parent);
1980         if (!se) {
1981                 DEBUG(0,("failed to setup SIGTERM handler"));
1982                 talloc_free(is_parent);
1983                 return false;
1984         }
1985
1986         se = tevent_add_signal(global_event_context(),
1987                                is_parent,
1988                                SIGINT, 0,
1989                                winbindd_sig_term_handler,
1990                                is_parent);
1991         if (!se) {
1992                 DEBUG(0,("failed to setup SIGINT handler"));
1993                 talloc_free(is_parent);
1994                 return false;
1995         }
1996
1997         se = tevent_add_signal(global_event_context(),
1998                                is_parent,
1999                                SIGQUIT, 0,
2000                                winbindd_sig_term_handler,
2001                                is_parent);
2002         if (!se) {
2003                 DEBUG(0,("failed to setup SIGINT handler"));
2004                 talloc_free(is_parent);
2005                 return false;
2006         }
2007
2008         return true;
2009 }
2010
2011 static void flush_caches_noinit(void)
2012 {
2013         /*
2014          * We need to invalidate cached user list entries on a SIGHUP
2015          * otherwise cached access denied errors due to restrict anonymous
2016          * hang around until the sequence number changes.
2017          * NB
2018          * Skip uninitialized domains when flush cache.
2019          * If domain is not initialized, it means it is never
2020          * used or never become online. look, wcache_invalidate_cache()
2021          * -> get_cache() -> init_dc_connection(). It causes a lot of traffic
2022          * for unused domains and large traffic for primay domain's DC if there
2023          * are many domains..
2024          */
2025
2026         if (!wcache_invalidate_cache_noinit()) {
2027                 DEBUG(0, ("invalidating the cache failed; revalidate the cache\n"));
2028                 if (!winbindd_cache_validate_and_initialize()) {
2029                         exit(1);
2030                 }
2031         }
2032 }
2033
2034 static void winbindd_sig_hup_handler(struct tevent_context *ev,
2035                                      struct tevent_signal *se,
2036                                      int signum,
2037                                      int count,
2038                                      void *siginfo,
2039                                      void *private_data)
2040 {
2041         const char *file = (const char *)private_data;
2042
2043         DEBUG(1,("Reloading services after SIGHUP\n"));
2044         flush_caches_noinit();
2045         winbindd_reload_services_file(file);
2046 }
2047
2048 bool winbindd_setup_sig_hup_handler(const char *lfile)
2049 {
2050         struct tevent_signal *se;
2051         char *file = NULL;
2052
2053         if (lfile) {
2054                 file = talloc_strdup(global_event_context(),
2055                                      lfile);
2056                 if (!file) {
2057                         return false;
2058                 }
2059         }
2060
2061         se = tevent_add_signal(global_event_context(),
2062                                global_event_context(),
2063                                SIGHUP, 0,
2064                                winbindd_sig_hup_handler,
2065                                file);
2066         if (!se) {
2067                 return false;
2068         }
2069
2070         return true;
2071 }