ctdb-daemon: Simplify code a bit
[obnox/samba/samba-obnox.git] / ctdb / server / ctdb_ltdb_server.c
1 /* 
2    ctdb ltdb code - server side
3
4    Copyright (C) Andrew Tridgell  2007
5
6    This program is free software; you can redistribute it and/or modify
7    it under the terms of the GNU General Public License as published by
8    the Free Software Foundation; either version 3 of the License, or
9    (at your option) any later version.
10    
11    This program is distributed in the hope that it will be useful,
12    but WITHOUT ANY WARRANTY; without even the implied warranty of
13    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14    GNU General Public License for more details.
15    
16    You should have received a copy of the GNU General Public License
17    along with this program; if not, see <http://www.gnu.org/licenses/>.
18 */
19
20 #include "includes.h"
21 #include "tdb.h"
22 #include "system/network.h"
23 #include "system/filesys.h"
24 #include "system/dir.h"
25 #include "system/time.h"
26 #include "../include/ctdb_private.h"
27 #include "../common/rb_tree.h"
28 #include "db_wrap.h"
29 #include "lib/util/dlinklist.h"
30 #include <ctype.h>
31
32 #define PERSISTENT_HEALTH_TDB "persistent_health.tdb"
33
34 /**
35  * write a record to a normal database
36  *
37  * This is the server-variant of the ctdb_ltdb_store function.
38  * It contains logic to determine whether a record should be
39  * stored or deleted. It also sends SCHEDULE_FOR_DELETION
40  * controls to the local ctdb daemon if apporpriate.
41  */
42 static int ctdb_ltdb_store_server(struct ctdb_db_context *ctdb_db,
43                                   TDB_DATA key,
44                                   struct ctdb_ltdb_header *header,
45                                   TDB_DATA data)
46 {
47         struct ctdb_context *ctdb = ctdb_db->ctdb;
48         TDB_DATA rec;
49         int ret;
50         bool seqnum_suppressed = false;
51         bool keep = false;
52         bool schedule_for_deletion = false;
53         bool remove_from_delete_queue = false;
54         uint32_t lmaster;
55
56         if (ctdb->flags & CTDB_FLAG_TORTURE) {
57                 struct ctdb_ltdb_header *h2;
58                 rec = tdb_fetch(ctdb_db->ltdb->tdb, key);
59                 h2 = (struct ctdb_ltdb_header *)rec.dptr;
60                 if (rec.dptr && rec.dsize >= sizeof(h2) && h2->rsn > header->rsn) {
61                         DEBUG(DEBUG_CRIT,("RSN regression! %llu %llu\n",
62                                  (unsigned long long)h2->rsn, (unsigned long long)header->rsn));
63                 }
64                 if (rec.dptr) free(rec.dptr);
65         }
66
67         if (ctdb->vnn_map == NULL) {
68                 /*
69                  * Called from a client: always store the record
70                  * Also don't call ctdb_lmaster since it uses the vnn_map!
71                  */
72                 keep = true;
73                 goto store;
74         }
75
76         lmaster = ctdb_lmaster(ctdb_db->ctdb, &key);
77
78         /*
79          * If we migrate an empty record off to another node
80          * and the record has not been migrated with data,
81          * delete the record instead of storing the empty record.
82          */
83         if (data.dsize != 0) {
84                 keep = true;
85         } else if (header->flags & CTDB_REC_RO_FLAGS) {
86                 keep = true;
87         } else if (ctdb_db->persistent) {
88                 keep = true;
89         } else if (header->flags & CTDB_REC_FLAG_AUTOMATIC) {
90                 /*
91                  * The record is not created by the client but
92                  * automatically by the ctdb_ltdb_fetch logic that
93                  * creates a record with an initial header in the
94                  * ltdb before trying to migrate the record from
95                  * the current lmaster. Keep it instead of trying
96                  * to delete the non-existing record...
97                  */
98                 keep = true;
99                 schedule_for_deletion = true;
100         } else if (header->flags & CTDB_REC_FLAG_MIGRATED_WITH_DATA) {
101                 keep = true;
102         } else if (ctdb_db->ctdb->pnn == lmaster) {
103                 /*
104                  * If we are lmaster, then we usually keep the record.
105                  * But if we retrieve the dmaster role by a VACUUM_MIGRATE
106                  * and the record is empty and has never been migrated
107                  * with data, then we should delete it instead of storing it.
108                  * This is part of the vacuuming process.
109                  *
110                  * The reason that we usually need to store even empty records
111                  * on the lmaster is that a client operating directly on the
112                  * lmaster (== dmaster) expects the local copy of the record to
113                  * exist after successful ctdb migrate call. If the record does
114                  * not exist, the client goes into a migrate loop and eventually
115                  * fails. So storing the empty record makes sure that we do not
116                  * need to change the client code.
117                  */
118                 if (!(header->flags & CTDB_REC_FLAG_VACUUM_MIGRATED)) {
119                         keep = true;
120                 } else if (ctdb_db->ctdb->pnn != header->dmaster) {
121                         keep = true;
122                 }
123         } else if (ctdb_db->ctdb->pnn == header->dmaster) {
124                 keep = true;
125         }
126
127         if (keep) {
128                 if (!ctdb_db->persistent &&
129                     (ctdb_db->ctdb->pnn == header->dmaster) &&
130                     !(header->flags & CTDB_REC_RO_FLAGS))
131                 {
132                         header->rsn++;
133
134                         if (data.dsize == 0) {
135                                 schedule_for_deletion = true;
136                         }
137                 }
138                 remove_from_delete_queue = !schedule_for_deletion;
139         }
140
141 store:
142         /*
143          * The VACUUM_MIGRATED flag is only set temporarily for
144          * the above logic when the record was retrieved by a
145          * VACUUM_MIGRATE call and should not be stored in the
146          * database.
147          *
148          * The VACUUM_MIGRATE call is triggered by a vacuum fetch,
149          * and there are two cases in which the corresponding record
150          * is stored in the local database:
151          * 1. The record has been migrated with data in the past
152          *    (the MIGRATED_WITH_DATA record flag is set).
153          * 2. The record has been filled with data again since it
154          *    had been submitted in the VACUUM_FETCH message to the
155          *    lmaster.
156          * For such records it is important to not store the
157          * VACUUM_MIGRATED flag in the database.
158          */
159         header->flags &= ~CTDB_REC_FLAG_VACUUM_MIGRATED;
160
161         /*
162          * Similarly, clear the AUTOMATIC flag which should not enter
163          * the local database copy since this would require client
164          * modifications to clear the flag when the client stores
165          * the record.
166          */
167         header->flags &= ~CTDB_REC_FLAG_AUTOMATIC;
168
169         rec.dsize = sizeof(*header) + data.dsize;
170         rec.dptr = talloc_size(ctdb, rec.dsize);
171         CTDB_NO_MEMORY(ctdb, rec.dptr);
172
173         memcpy(rec.dptr, header, sizeof(*header));
174         memcpy(rec.dptr + sizeof(*header), data.dptr, data.dsize);
175
176         /* Databases with seqnum updates enabled only get their seqnum
177            changes when/if we modify the data */
178         if (ctdb_db->seqnum_update != NULL) {
179                 TDB_DATA old;
180                 old = tdb_fetch(ctdb_db->ltdb->tdb, key);
181
182                 if ( (old.dsize == rec.dsize)
183                 && !memcmp(old.dptr+sizeof(struct ctdb_ltdb_header),
184                           rec.dptr+sizeof(struct ctdb_ltdb_header),
185                           rec.dsize-sizeof(struct ctdb_ltdb_header)) ) {
186                         tdb_remove_flags(ctdb_db->ltdb->tdb, TDB_SEQNUM);
187                         seqnum_suppressed = true;
188                 }
189                 if (old.dptr) free(old.dptr);
190         }
191
192         DEBUG(DEBUG_DEBUG, (__location__ " db[%s]: %s record: hash[0x%08x]\n",
193                             ctdb_db->db_name,
194                             keep?"storing":"deleting",
195                             ctdb_hash(&key)));
196
197         if (keep) {
198                 ret = tdb_store(ctdb_db->ltdb->tdb, key, rec, TDB_REPLACE);
199         } else {
200                 ret = tdb_delete(ctdb_db->ltdb->tdb, key);
201         }
202
203         if (ret != 0) {
204                 int lvl = DEBUG_ERR;
205
206                 if (keep == false &&
207                     tdb_error(ctdb_db->ltdb->tdb) == TDB_ERR_NOEXIST)
208                 {
209                         lvl = DEBUG_DEBUG;
210                 }
211
212                 DEBUG(lvl, (__location__ " db[%s]: Failed to %s record: "
213                             "%d - %s\n",
214                             ctdb_db->db_name,
215                             keep?"store":"delete", ret,
216                             tdb_errorstr(ctdb_db->ltdb->tdb)));
217
218                 schedule_for_deletion = false;
219                 remove_from_delete_queue = false;
220         }
221         if (seqnum_suppressed) {
222                 tdb_add_flags(ctdb_db->ltdb->tdb, TDB_SEQNUM);
223         }
224
225         talloc_free(rec.dptr);
226
227         if (schedule_for_deletion) {
228                 int ret2;
229                 ret2 = ctdb_local_schedule_for_deletion(ctdb_db, header, key);
230                 if (ret2 != 0) {
231                         DEBUG(DEBUG_ERR, (__location__ " ctdb_local_schedule_for_deletion failed.\n"));
232                 }
233         }
234
235         if (remove_from_delete_queue) {
236                 ctdb_local_remove_from_delete_queue(ctdb_db, header, key);
237         }
238
239         return ret;
240 }
241
242 struct lock_fetch_state {
243         struct ctdb_context *ctdb;
244         void (*recv_pkt)(void *, struct ctdb_req_header *);
245         void *recv_context;
246         struct ctdb_req_header *hdr;
247         uint32_t generation;
248         bool ignore_generation;
249 };
250
251 /*
252   called when we should retry the operation
253  */
254 static void lock_fetch_callback(void *p, bool locked)
255 {
256         struct lock_fetch_state *state = talloc_get_type(p, struct lock_fetch_state);
257         if (!state->ignore_generation &&
258             state->generation != state->ctdb->vnn_map->generation) {
259                 DEBUG(DEBUG_NOTICE,("Discarding previous generation lockwait packet\n"));
260                 talloc_free(state->hdr);
261                 return;
262         }
263         state->recv_pkt(state->recv_context, state->hdr);
264         DEBUG(DEBUG_INFO,(__location__ " PACKET REQUEUED\n"));
265 }
266
267
268 /*
269   do a non-blocking ltdb_lock, deferring this ctdb request until we
270   have the chainlock
271
272   It does the following:
273
274    1) tries to get the chainlock. If it succeeds, then it returns 0
275
276    2) if it fails to get a chainlock immediately then it sets up a
277    non-blocking chainlock via ctdb_lock_record, and when it gets the
278    chainlock it re-submits this ctdb request to the main packet
279    receive function.
280
281    This effectively queues all ctdb requests that cannot be
282    immediately satisfied until it can get the lock. This means that
283    the main ctdb daemon will not block waiting for a chainlock held by
284    a client
285
286    There are 3 possible return values:
287
288        0:    means that it got the lock immediately.
289       -1:    means that it failed to get the lock, and won't retry
290       -2:    means that it failed to get the lock immediately, but will retry
291  */
292 int ctdb_ltdb_lock_requeue(struct ctdb_db_context *ctdb_db, 
293                            TDB_DATA key, struct ctdb_req_header *hdr,
294                            void (*recv_pkt)(void *, struct ctdb_req_header *),
295                            void *recv_context, bool ignore_generation)
296 {
297         int ret;
298         struct tdb_context *tdb = ctdb_db->ltdb->tdb;
299         struct lock_request *lreq;
300         struct lock_fetch_state *state;
301         
302         ret = tdb_chainlock_nonblock(tdb, key);
303
304         if (ret != 0 &&
305             !(errno == EACCES || errno == EAGAIN || errno == EDEADLK)) {
306                 /* a hard failure - don't try again */
307                 return -1;
308         }
309
310         /* when torturing, ensure we test the contended path */
311         if ((ctdb_db->ctdb->flags & CTDB_FLAG_TORTURE) &&
312             random() % 5 == 0) {
313                 ret = -1;
314                 tdb_chainunlock(tdb, key);
315         }
316
317         /* first the non-contended path */
318         if (ret == 0) {
319                 return 0;
320         }
321
322         state = talloc(hdr, struct lock_fetch_state);
323         state->ctdb = ctdb_db->ctdb;
324         state->hdr = hdr;
325         state->recv_pkt = recv_pkt;
326         state->recv_context = recv_context;
327         state->generation = ctdb_db->ctdb->vnn_map->generation;
328         state->ignore_generation = ignore_generation;
329
330         /* now the contended path */
331         lreq = ctdb_lock_record(ctdb_db, key, true, lock_fetch_callback, state);
332         if (lreq == NULL) {
333                 return -1;
334         }
335
336         /* we need to move the packet off the temporary context in ctdb_input_pkt(),
337            so it won't be freed yet */
338         talloc_steal(state, hdr);
339
340         /* now tell the caller than we will retry asynchronously */
341         return -2;
342 }
343
344 /*
345   a varient of ctdb_ltdb_lock_requeue that also fetches the record
346  */
347 int ctdb_ltdb_lock_fetch_requeue(struct ctdb_db_context *ctdb_db, 
348                                  TDB_DATA key, struct ctdb_ltdb_header *header, 
349                                  struct ctdb_req_header *hdr, TDB_DATA *data,
350                                  void (*recv_pkt)(void *, struct ctdb_req_header *),
351                                  void *recv_context, bool ignore_generation)
352 {
353         int ret;
354
355         ret = ctdb_ltdb_lock_requeue(ctdb_db, key, hdr, recv_pkt, 
356                                      recv_context, ignore_generation);
357         if (ret == 0) {
358                 ret = ctdb_ltdb_fetch(ctdb_db, key, header, hdr, data);
359                 if (ret != 0) {
360                         int uret;
361                         uret = ctdb_ltdb_unlock(ctdb_db, key);
362                         if (uret != 0) {
363                                 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", uret));
364                         }
365                 }
366         }
367         return ret;
368 }
369
370
371 /*
372   paraoid check to see if the db is empty
373  */
374 static void ctdb_check_db_empty(struct ctdb_db_context *ctdb_db)
375 {
376         struct tdb_context *tdb = ctdb_db->ltdb->tdb;
377         int count = tdb_traverse_read(tdb, NULL, NULL);
378         if (count != 0) {
379                 DEBUG(DEBUG_ALERT,(__location__ " tdb '%s' not empty on attach! aborting\n",
380                          ctdb_db->db_path));
381                 ctdb_fatal(ctdb_db->ctdb, "database not empty on attach");
382         }
383 }
384
385 int ctdb_load_persistent_health(struct ctdb_context *ctdb,
386                                 struct ctdb_db_context *ctdb_db)
387 {
388         struct tdb_context *tdb = ctdb->db_persistent_health->tdb;
389         char *old;
390         char *reason = NULL;
391         TDB_DATA key;
392         TDB_DATA val;
393
394         key.dptr = discard_const_p(uint8_t, ctdb_db->db_name);
395         key.dsize = strlen(ctdb_db->db_name);
396
397         old = ctdb_db->unhealthy_reason;
398         ctdb_db->unhealthy_reason = NULL;
399
400         val = tdb_fetch(tdb, key);
401         if (val.dsize > 0) {
402                 reason = talloc_strndup(ctdb_db,
403                                         (const char *)val.dptr,
404                                         val.dsize);
405                 if (reason == NULL) {
406                         DEBUG(DEBUG_ALERT,(__location__ " talloc_strndup(%d) failed\n",
407                                            (int)val.dsize));
408                         ctdb_db->unhealthy_reason = old;
409                         free(val.dptr);
410                         return -1;
411                 }
412         }
413
414         if (val.dptr) {
415                 free(val.dptr);
416         }
417
418         talloc_free(old);
419         ctdb_db->unhealthy_reason = reason;
420         return 0;
421 }
422
423 int ctdb_update_persistent_health(struct ctdb_context *ctdb,
424                                   struct ctdb_db_context *ctdb_db,
425                                   const char *given_reason,/* NULL means healthy */
426                                   int num_healthy_nodes)
427 {
428         struct tdb_context *tdb = ctdb->db_persistent_health->tdb;
429         int ret;
430         TDB_DATA key;
431         TDB_DATA val;
432         char *new_reason = NULL;
433         char *old_reason = NULL;
434
435         ret = tdb_transaction_start(tdb);
436         if (ret != 0) {
437                 DEBUG(DEBUG_ALERT,(__location__ " tdb_transaction_start('%s') failed: %d - %s\n",
438                                    tdb_name(tdb), ret, tdb_errorstr(tdb)));
439                 return -1;
440         }
441
442         ret = ctdb_load_persistent_health(ctdb, ctdb_db);
443         if (ret != 0) {
444                 DEBUG(DEBUG_ALERT,(__location__ " ctdb_load_persistent_health('%s') failed: %d\n",
445                                    ctdb_db->db_name, ret));
446                 return -1;
447         }
448         old_reason = ctdb_db->unhealthy_reason;
449
450         key.dptr = discard_const_p(uint8_t, ctdb_db->db_name);
451         key.dsize = strlen(ctdb_db->db_name);
452
453         if (given_reason) {
454                 new_reason = talloc_strdup(ctdb_db, given_reason);
455                 if (new_reason == NULL) {
456                         DEBUG(DEBUG_ALERT,(__location__ " talloc_strdup(%s) failed\n",
457                                           given_reason));
458                         return -1;
459                 }
460         } else if (old_reason && num_healthy_nodes == 0) {
461                 /*
462                  * If the reason indicates ok, but there where no healthy nodes
463                  * available, that it means, we have not recovered valid content
464                  * of the db. So if there's an old reason, prefix it with
465                  * "NO-HEALTHY-NODES - "
466                  */
467                 const char *prefix;
468
469 #define _TMP_PREFIX "NO-HEALTHY-NODES - "
470                 ret = strncmp(_TMP_PREFIX, old_reason, strlen(_TMP_PREFIX));
471                 if (ret != 0) {
472                         prefix = _TMP_PREFIX;
473                 } else {
474                         prefix = "";
475                 }
476                 new_reason = talloc_asprintf(ctdb_db, "%s%s",
477                                          prefix, old_reason);
478                 if (new_reason == NULL) {
479                         DEBUG(DEBUG_ALERT,(__location__ " talloc_asprintf(%s%s) failed\n",
480                                           prefix, old_reason));
481                         return -1;
482                 }
483 #undef _TMP_PREFIX
484         }
485
486         if (new_reason) {
487                 val.dptr = discard_const_p(uint8_t, new_reason);
488                 val.dsize = strlen(new_reason);
489
490                 ret = tdb_store(tdb, key, val, TDB_REPLACE);
491                 if (ret != 0) {
492                         tdb_transaction_cancel(tdb);
493                         DEBUG(DEBUG_ALERT,(__location__ " tdb_store('%s', %s, %s) failed: %d - %s\n",
494                                            tdb_name(tdb), ctdb_db->db_name, new_reason,
495                                            ret, tdb_errorstr(tdb)));
496                         talloc_free(new_reason);
497                         return -1;
498                 }
499                 DEBUG(DEBUG_ALERT,("Updated db health for db(%s) to: %s\n",
500                                    ctdb_db->db_name, new_reason));
501         } else if (old_reason) {
502                 ret = tdb_delete(tdb, key);
503                 if (ret != 0) {
504                         tdb_transaction_cancel(tdb);
505                         DEBUG(DEBUG_ALERT,(__location__ " tdb_delete('%s', %s) failed: %d - %s\n",
506                                            tdb_name(tdb), ctdb_db->db_name,
507                                            ret, tdb_errorstr(tdb)));
508                         talloc_free(new_reason);
509                         return -1;
510                 }
511                 DEBUG(DEBUG_NOTICE,("Updated db health for db(%s): OK\n",
512                                    ctdb_db->db_name));
513         }
514
515         ret = tdb_transaction_commit(tdb);
516         if (ret != TDB_SUCCESS) {
517                 DEBUG(DEBUG_ALERT,(__location__ " tdb_transaction_commit('%s') failed: %d - %s\n",
518                                    tdb_name(tdb), ret, tdb_errorstr(tdb)));
519                 talloc_free(new_reason);
520                 return -1;
521         }
522
523         talloc_free(old_reason);
524         ctdb_db->unhealthy_reason = new_reason;
525
526         return 0;
527 }
528
529 static int ctdb_backup_corrupted_tdb(struct ctdb_context *ctdb,
530                                      struct ctdb_db_context *ctdb_db)
531 {
532         time_t now = time(NULL);
533         char *new_path;
534         char *new_reason;
535         int ret;
536         struct tm *tm;
537
538         tm = gmtime(&now);
539
540         /* formatted like: foo.tdb.0.corrupted.20091204160825.0Z */
541         new_path = talloc_asprintf(ctdb_db, "%s.corrupted."
542                                    "%04u%02u%02u%02u%02u%02u.0Z",
543                                    ctdb_db->db_path,
544                                    tm->tm_year+1900, tm->tm_mon+1,
545                                    tm->tm_mday, tm->tm_hour, tm->tm_min,
546                                    tm->tm_sec);
547         if (new_path == NULL) {
548                 DEBUG(DEBUG_CRIT,(__location__ " talloc_asprintf() failed\n"));
549                 return -1;
550         }
551
552         new_reason = talloc_asprintf(ctdb_db,
553                                      "ERROR - Backup of corrupted TDB in '%s'",
554                                      new_path);
555         if (new_reason == NULL) {
556                 DEBUG(DEBUG_CRIT,(__location__ " talloc_asprintf() failed\n"));
557                 return -1;
558         }
559         ret = ctdb_update_persistent_health(ctdb, ctdb_db, new_reason, 0);
560         talloc_free(new_reason);
561         if (ret != 0) {
562                 DEBUG(DEBUG_CRIT,(__location__
563                                  ": ctdb_backup_corrupted_tdb(%s) not implemented yet\n",
564                                  ctdb_db->db_path));
565                 return -1;
566         }
567
568         ret = rename(ctdb_db->db_path, new_path);
569         if (ret != 0) {
570                 DEBUG(DEBUG_CRIT,(__location__
571                                   ": ctdb_backup_corrupted_tdb(%s) rename to %s failed: %d - %s\n",
572                                   ctdb_db->db_path, new_path,
573                                   errno, strerror(errno)));
574                 talloc_free(new_path);
575                 return -1;
576         }
577
578         DEBUG(DEBUG_CRIT,(__location__
579                          ": ctdb_backup_corrupted_tdb(%s) renamed to %s\n",
580                          ctdb_db->db_path, new_path));
581         talloc_free(new_path);
582         return 0;
583 }
584
585 int ctdb_recheck_persistent_health(struct ctdb_context *ctdb)
586 {
587         struct ctdb_db_context *ctdb_db;
588         int ret;
589         int ok = 0;
590         int fail = 0;
591
592         for (ctdb_db = ctdb->db_list; ctdb_db; ctdb_db = ctdb_db->next) {
593                 if (!ctdb_db->persistent) {
594                         continue;
595                 }
596
597                 ret = ctdb_load_persistent_health(ctdb, ctdb_db);
598                 if (ret != 0) {
599                         DEBUG(DEBUG_ALERT,(__location__
600                                            " load persistent health for '%s' failed\n",
601                                            ctdb_db->db_path));
602                         return -1;
603                 }
604
605                 if (ctdb_db->unhealthy_reason == NULL) {
606                         ok++;
607                         DEBUG(DEBUG_INFO,(__location__
608                                    " persistent db '%s' healthy\n",
609                                    ctdb_db->db_path));
610                         continue;
611                 }
612
613                 fail++;
614                 DEBUG(DEBUG_ALERT,(__location__
615                                    " persistent db '%s' unhealthy: %s\n",
616                                    ctdb_db->db_path,
617                                    ctdb_db->unhealthy_reason));
618         }
619         DEBUG((fail!=0)?DEBUG_ALERT:DEBUG_NOTICE,
620               ("ctdb_recheck_presistent_health: OK[%d] FAIL[%d]\n",
621                ok, fail));
622
623         if (fail != 0) {
624                 return -1;
625         }
626
627         return 0;
628 }
629
630
631 /*
632   mark a database - as healthy
633  */
634 int32_t ctdb_control_db_set_healthy(struct ctdb_context *ctdb, TDB_DATA indata)
635 {
636         uint32_t db_id = *(uint32_t *)indata.dptr;
637         struct ctdb_db_context *ctdb_db;
638         int ret;
639         bool may_recover = false;
640
641         ctdb_db = find_ctdb_db(ctdb, db_id);
642         if (!ctdb_db) {
643                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%x\n", db_id));
644                 return -1;
645         }
646
647         if (ctdb_db->unhealthy_reason) {
648                 may_recover = true;
649         }
650
651         ret = ctdb_update_persistent_health(ctdb, ctdb_db, NULL, 1);
652         if (ret != 0) {
653                 DEBUG(DEBUG_ERR,(__location__
654                                  " ctdb_update_persistent_health(%s) failed\n",
655                                  ctdb_db->db_name));
656                 return -1;
657         }
658
659         if (may_recover && ctdb->runstate == CTDB_RUNSTATE_STARTUP) {
660                 DEBUG(DEBUG_ERR, (__location__ " db %s become healthy  - force recovery for startup\n",
661                                   ctdb_db->db_name));
662                 ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
663         }
664
665         return 0;
666 }
667
668 int32_t ctdb_control_db_get_health(struct ctdb_context *ctdb,
669                                    TDB_DATA indata,
670                                    TDB_DATA *outdata)
671 {
672         uint32_t db_id = *(uint32_t *)indata.dptr;
673         struct ctdb_db_context *ctdb_db;
674         int ret;
675
676         ctdb_db = find_ctdb_db(ctdb, db_id);
677         if (!ctdb_db) {
678                 DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%x\n", db_id));
679                 return -1;
680         }
681
682         ret = ctdb_load_persistent_health(ctdb, ctdb_db);
683         if (ret != 0) {
684                 DEBUG(DEBUG_ERR,(__location__
685                                  " ctdb_load_persistent_health(%s) failed\n",
686                                  ctdb_db->db_name));
687                 return -1;
688         }
689
690         *outdata = tdb_null;
691         if (ctdb_db->unhealthy_reason) {
692                 outdata->dptr = (uint8_t *)ctdb_db->unhealthy_reason;
693                 outdata->dsize = strlen(ctdb_db->unhealthy_reason)+1;
694         }
695
696         return 0;
697 }
698
699
700 int ctdb_set_db_readonly(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db)
701 {
702         char *ropath;
703
704         if (ctdb_db->readonly) {
705                 return 0;
706         }
707
708         if (ctdb_db->persistent) {
709                 DEBUG(DEBUG_ERR,("Persistent databases do not support readonly property\n"));
710                 return -1;
711         }
712
713         ropath = talloc_asprintf(ctdb_db, "%s.RO", ctdb_db->db_path);
714         if (ropath == NULL) {
715                 DEBUG(DEBUG_CRIT,("Failed to asprintf the tracking database\n"));
716                 return -1;
717         }
718         ctdb_db->rottdb = tdb_open(ropath, 
719                               ctdb->tunable.database_hash_size, 
720                               TDB_NOLOCK|TDB_CLEAR_IF_FIRST|TDB_NOSYNC,
721                               O_CREAT|O_RDWR, 0);
722         if (ctdb_db->rottdb == NULL) {
723                 DEBUG(DEBUG_CRIT,("Failed to open/create the tracking database '%s'\n", ropath));
724                 talloc_free(ropath);
725                 return -1;
726         }
727
728         DEBUG(DEBUG_NOTICE,("OPENED tracking database : '%s'\n", ropath));
729
730         ctdb_db->readonly = true;
731
732         DEBUG(DEBUG_NOTICE, ("Readonly property set on DB %s\n", ctdb_db->db_name));
733
734         talloc_free(ropath);
735         return 0;
736 }
737
738 /*
739   attach to a database, handling both persistent and non-persistent databases
740   return 0 on success, -1 on failure
741  */
742 static int ctdb_local_attach(struct ctdb_context *ctdb, const char *db_name,
743                              bool persistent, const char *unhealthy_reason,
744                              bool jenkinshash)
745 {
746         struct ctdb_db_context *ctdb_db, *tmp_db;
747         int ret;
748         struct TDB_DATA key;
749         unsigned tdb_flags;
750         int mode = 0600;
751         int remaining_tries = 0;
752
753         ctdb_db = talloc_zero(ctdb, struct ctdb_db_context);
754         CTDB_NO_MEMORY(ctdb, ctdb_db);
755
756         ctdb_db->priority = 1;
757         ctdb_db->ctdb = ctdb;
758         ctdb_db->db_name = talloc_strdup(ctdb_db, db_name);
759         CTDB_NO_MEMORY(ctdb, ctdb_db->db_name);
760
761         key.dsize = strlen(db_name)+1;
762         key.dptr  = discard_const(db_name);
763         ctdb_db->db_id = ctdb_hash(&key);
764         ctdb_db->persistent = persistent;
765
766         if (!ctdb_db->persistent) {
767                 ctdb_db->delete_queue = trbt_create(ctdb_db, 0);
768                 if (ctdb_db->delete_queue == NULL) {
769                         CTDB_NO_MEMORY(ctdb, ctdb_db->delete_queue);
770                 }
771
772                 ctdb_db->ctdb_ltdb_store_fn = ctdb_ltdb_store_server;
773         }
774
775         /* check for hash collisions */
776         for (tmp_db=ctdb->db_list;tmp_db;tmp_db=tmp_db->next) {
777                 if (tmp_db->db_id == ctdb_db->db_id) {
778                         DEBUG(DEBUG_CRIT,("db_id 0x%x hash collision. name1='%s' name2='%s'\n",
779                                  tmp_db->db_id, db_name, tmp_db->db_name));
780                         talloc_free(ctdb_db);
781                         return -1;
782                 }
783         }
784
785         if (persistent) {
786                 if (unhealthy_reason) {
787                         ret = ctdb_update_persistent_health(ctdb, ctdb_db,
788                                                             unhealthy_reason, 0);
789                         if (ret != 0) {
790                                 DEBUG(DEBUG_ALERT,(__location__ " ctdb_update_persistent_health('%s','%s') failed: %d\n",
791                                                    ctdb_db->db_name, unhealthy_reason, ret));
792                                 talloc_free(ctdb_db);
793                                 return -1;
794                         }
795                 }
796
797                 if (ctdb->max_persistent_check_errors > 0) {
798                         remaining_tries = 1;
799                 }
800                 if (ctdb->runstate == CTDB_RUNSTATE_RUNNING) {
801                         remaining_tries = 0;
802                 }
803
804                 ret = ctdb_load_persistent_health(ctdb, ctdb_db);
805                 if (ret != 0) {
806                         DEBUG(DEBUG_ALERT,(__location__ " ctdb_load_persistent_health('%s') failed: %d\n",
807                                    ctdb_db->db_name, ret));
808                         talloc_free(ctdb_db);
809                         return -1;
810                 }
811         }
812
813         if (ctdb_db->unhealthy_reason && remaining_tries == 0) {
814                 DEBUG(DEBUG_ALERT,(__location__ "ERROR: tdb %s is marked as unhealthy: %s\n",
815                                    ctdb_db->db_name, ctdb_db->unhealthy_reason));
816                 talloc_free(ctdb_db);
817                 return -1;
818         }
819
820         if (ctdb_db->unhealthy_reason) {
821                 /* this is just a warning, but we want that in the log file! */
822                 DEBUG(DEBUG_ALERT,(__location__ "Warning: tdb %s is marked as unhealthy: %s\n",
823                                    ctdb_db->db_name, ctdb_db->unhealthy_reason));
824         }
825
826         /* open the database */
827         ctdb_db->db_path = talloc_asprintf(ctdb_db, "%s/%s.%u", 
828                                            persistent?ctdb->db_directory_persistent:ctdb->db_directory, 
829                                            db_name, ctdb->pnn);
830
831         tdb_flags = persistent? TDB_DEFAULT : TDB_CLEAR_IF_FIRST | TDB_NOSYNC;
832         if (ctdb->valgrinding) {
833                 tdb_flags |= TDB_NOMMAP;
834         }
835         tdb_flags |= TDB_DISALLOW_NESTING;
836         if (jenkinshash) {
837                 tdb_flags |= TDB_INCOMPATIBLE_HASH;
838         }
839
840 again:
841         ctdb_db->ltdb = tdb_wrap_open(ctdb_db, ctdb_db->db_path,
842                                       ctdb->tunable.database_hash_size, 
843                                       tdb_flags, 
844                                       O_CREAT|O_RDWR, mode);
845         if (ctdb_db->ltdb == NULL) {
846                 struct stat st;
847                 int saved_errno = errno;
848
849                 if (!persistent) {
850                         DEBUG(DEBUG_CRIT,("Failed to open tdb '%s': %d - %s\n",
851                                           ctdb_db->db_path,
852                                           saved_errno,
853                                           strerror(saved_errno)));
854                         talloc_free(ctdb_db);
855                         return -1;
856                 }
857
858                 if (remaining_tries == 0) {
859                         DEBUG(DEBUG_CRIT,(__location__
860                                           "Failed to open persistent tdb '%s': %d - %s\n",
861                                           ctdb_db->db_path,
862                                           saved_errno,
863                                           strerror(saved_errno)));
864                         talloc_free(ctdb_db);
865                         return -1;
866                 }
867
868                 ret = stat(ctdb_db->db_path, &st);
869                 if (ret != 0) {
870                         DEBUG(DEBUG_CRIT,(__location__
871                                           "Failed to open persistent tdb '%s': %d - %s\n",
872                                           ctdb_db->db_path,
873                                           saved_errno,
874                                           strerror(saved_errno)));
875                         talloc_free(ctdb_db);
876                         return -1;
877                 }
878
879                 ret = ctdb_backup_corrupted_tdb(ctdb, ctdb_db);
880                 if (ret != 0) {
881                         DEBUG(DEBUG_CRIT,(__location__
882                                           "Failed to open persistent tdb '%s': %d - %s\n",
883                                           ctdb_db->db_path,
884                                           saved_errno,
885                                           strerror(saved_errno)));
886                         talloc_free(ctdb_db);
887                         return -1;
888                 }
889
890                 remaining_tries--;
891                 mode = st.st_mode;
892                 goto again;
893         }
894
895         if (!persistent) {
896                 ctdb_check_db_empty(ctdb_db);
897         } else {
898                 ret = tdb_check(ctdb_db->ltdb->tdb, NULL, NULL);
899                 if (ret != 0) {
900                         int fd;
901                         struct stat st;
902
903                         DEBUG(DEBUG_CRIT,("tdb_check(%s) failed: %d - %s\n",
904                                           ctdb_db->db_path, ret,
905                                           tdb_errorstr(ctdb_db->ltdb->tdb)));
906                         if (remaining_tries == 0) {
907                                 talloc_free(ctdb_db);
908                                 return -1;
909                         }
910
911                         fd = tdb_fd(ctdb_db->ltdb->tdb);
912                         ret = fstat(fd, &st);
913                         if (ret != 0) {
914                                 DEBUG(DEBUG_CRIT,(__location__
915                                                   "Failed to fstat() persistent tdb '%s': %d - %s\n",
916                                                   ctdb_db->db_path,
917                                                   errno,
918                                                   strerror(errno)));
919                                 talloc_free(ctdb_db);
920                                 return -1;
921                         }
922
923                         /* close the TDB */
924                         talloc_free(ctdb_db->ltdb);
925                         ctdb_db->ltdb = NULL;
926
927                         ret = ctdb_backup_corrupted_tdb(ctdb, ctdb_db);
928                         if (ret != 0) {
929                                 DEBUG(DEBUG_CRIT,("Failed to backup corrupted tdb '%s'\n",
930                                                   ctdb_db->db_path));
931                                 talloc_free(ctdb_db);
932                                 return -1;
933                         }
934
935                         remaining_tries--;
936                         mode = st.st_mode;
937                         goto again;
938                 }
939         }
940
941         /* set up a rb tree we can use to track which records we have a 
942            fetch-lock in-flight for so we can defer any additional calls
943            for the same record.
944          */
945         ctdb_db->deferred_fetch = trbt_create(ctdb_db, 0);
946         if (ctdb_db->deferred_fetch == NULL) {
947                 DEBUG(DEBUG_ERR,("Failed to create deferred fetch rb tree for ctdb database\n"));
948                 talloc_free(ctdb_db);
949                 return -1;
950         }
951
952         DLIST_ADD(ctdb->db_list, ctdb_db);
953
954         /* setting this can help some high churn databases */
955         tdb_set_max_dead(ctdb_db->ltdb->tdb, ctdb->tunable.database_max_dead);
956
957         /* 
958            all databases support the "null" function. we need this in
959            order to do forced migration of records
960         */
961         ret = ctdb_daemon_set_call(ctdb, ctdb_db->db_id, ctdb_null_func, CTDB_NULL_FUNC);
962         if (ret != 0) {
963                 DEBUG(DEBUG_CRIT,("Failed to setup null function for '%s'\n", ctdb_db->db_name));
964                 talloc_free(ctdb_db);
965                 return -1;
966         }
967
968         /* 
969            all databases support the "fetch" function. we need this
970            for efficient Samba3 ctdb fetch
971         */
972         ret = ctdb_daemon_set_call(ctdb, ctdb_db->db_id, ctdb_fetch_func, CTDB_FETCH_FUNC);
973         if (ret != 0) {
974                 DEBUG(DEBUG_CRIT,("Failed to setup fetch function for '%s'\n", ctdb_db->db_name));
975                 talloc_free(ctdb_db);
976                 return -1;
977         }
978
979         /* 
980            all databases support the "fetch_with_header" function. we need this
981            for efficient readonly record fetches
982         */
983         ret = ctdb_daemon_set_call(ctdb, ctdb_db->db_id, ctdb_fetch_with_header_func, CTDB_FETCH_WITH_HEADER_FUNC);
984         if (ret != 0) {
985                 DEBUG(DEBUG_CRIT,("Failed to setup fetch function for '%s'\n", ctdb_db->db_name));
986                 talloc_free(ctdb_db);
987                 return -1;
988         }
989
990         ret = ctdb_vacuum_init(ctdb_db);
991         if (ret != 0) {
992                 DEBUG(DEBUG_CRIT,("Failed to setup vacuuming for "
993                                   "database '%s'\n", ctdb_db->db_name));
994                 talloc_free(ctdb_db);
995                 return -1;
996         }
997
998
999         DEBUG(DEBUG_NOTICE,("Attached to database '%s' with flags 0x%x\n",
1000                             ctdb_db->db_path, tdb_flags));
1001
1002         /* success */
1003         return 0;
1004 }
1005
1006
1007 struct ctdb_deferred_attach_context {
1008         struct ctdb_deferred_attach_context *next, *prev;
1009         struct ctdb_context *ctdb;
1010         struct ctdb_req_control *c;
1011 };
1012
1013
1014 static int ctdb_deferred_attach_destructor(struct ctdb_deferred_attach_context *da_ctx)
1015 {
1016         DLIST_REMOVE(da_ctx->ctdb->deferred_attach, da_ctx);
1017
1018         return 0;
1019 }
1020
1021 static void ctdb_deferred_attach_timeout(struct event_context *ev, struct timed_event *te, struct timeval t, void *private_data)
1022 {
1023         struct ctdb_deferred_attach_context *da_ctx = talloc_get_type(private_data, struct ctdb_deferred_attach_context);
1024         struct ctdb_context *ctdb = da_ctx->ctdb;
1025
1026         ctdb_request_control_reply(ctdb, da_ctx->c, NULL, -1, NULL);
1027         talloc_free(da_ctx);
1028 }
1029
1030 static void ctdb_deferred_attach_callback(struct event_context *ev, struct timed_event *te, struct timeval t, void *private_data)
1031 {
1032         struct ctdb_deferred_attach_context *da_ctx = talloc_get_type(private_data, struct ctdb_deferred_attach_context);
1033         struct ctdb_context *ctdb = da_ctx->ctdb;
1034
1035         /* This talloc-steals the packet ->c */
1036         ctdb_input_pkt(ctdb, (struct ctdb_req_header *)da_ctx->c);
1037         talloc_free(da_ctx);
1038 }
1039
1040 int ctdb_process_deferred_attach(struct ctdb_context *ctdb)
1041 {
1042         struct ctdb_deferred_attach_context *da_ctx;
1043
1044         /* call it from the main event loop as soon as the current event 
1045            finishes.
1046          */
1047         while ((da_ctx = ctdb->deferred_attach) != NULL) {
1048                 DLIST_REMOVE(ctdb->deferred_attach, da_ctx);
1049                 event_add_timed(ctdb->ev, da_ctx, timeval_current_ofs(1,0), ctdb_deferred_attach_callback, da_ctx);
1050         }
1051
1052         return 0;
1053 }
1054
1055 /*
1056   a client has asked to attach a new database
1057  */
1058 int32_t ctdb_control_db_attach(struct ctdb_context *ctdb, TDB_DATA indata,
1059                                TDB_DATA *outdata, uint64_t tdb_flags, 
1060                                bool persistent, uint32_t client_id,
1061                                struct ctdb_req_control *c,
1062                                bool *async_reply)
1063 {
1064         const char *db_name = (const char *)indata.dptr;
1065         struct ctdb_db_context *db;
1066         struct ctdb_node *node = ctdb->nodes[ctdb->pnn];
1067         struct ctdb_client *client = NULL;
1068         bool with_jenkinshash;
1069
1070         if (ctdb->tunable.allow_client_db_attach == 0) {
1071                 DEBUG(DEBUG_ERR, ("DB Attach to database %s denied by tunable "
1072                                   "AllowClientDBAccess == 0\n", db_name));
1073                 return -1;
1074         }
1075
1076         /* dont allow any local clients to attach while we are in recovery mode
1077          * except for the recovery daemon.
1078          * allow all attach from the network since these are always from remote
1079          * recovery daemons.
1080          */
1081         if (client_id != 0) {
1082                 client = ctdb_reqid_find(ctdb, client_id, struct ctdb_client);
1083         }
1084         if (client != NULL) {
1085                 /* If the node is inactive it is not part of the cluster
1086                    and we should not allow clients to attach to any
1087                    databases
1088                 */
1089                 if (node->flags & NODE_FLAGS_INACTIVE) {
1090                         DEBUG(DEBUG_ERR,("DB Attach to database %s refused since node is inactive (flags=0x%x)\n", db_name, node->flags));
1091                         return -1;
1092                 }
1093
1094                 if (ctdb->recovery_mode == CTDB_RECOVERY_ACTIVE &&
1095                     client->pid != ctdb->recoverd_pid &&
1096                     ctdb->runstate < CTDB_RUNSTATE_RUNNING) {
1097                         struct ctdb_deferred_attach_context *da_ctx = talloc(client, struct ctdb_deferred_attach_context);
1098
1099                         if (da_ctx == NULL) {
1100                                 DEBUG(DEBUG_ERR,("DB Attach to database %s deferral for client with pid:%d failed due to OOM.\n", db_name, client->pid));
1101                                 return -1;
1102                         }
1103
1104                         da_ctx->ctdb = ctdb;
1105                         da_ctx->c = talloc_steal(da_ctx, c);
1106                         talloc_set_destructor(da_ctx, ctdb_deferred_attach_destructor);
1107                         DLIST_ADD(ctdb->deferred_attach, da_ctx);
1108
1109                         event_add_timed(ctdb->ev, da_ctx, timeval_current_ofs(ctdb->tunable.deferred_attach_timeout, 0), ctdb_deferred_attach_timeout, da_ctx);
1110
1111                         DEBUG(DEBUG_ERR,("DB Attach to database %s deferred for client with pid:%d since node is in recovery mode.\n", db_name, client->pid));
1112                         *async_reply = true;
1113                         return 0;
1114                 }
1115         }
1116
1117         /* the client can optionally pass additional tdb flags, but we
1118            only allow a subset of those on the database in ctdb. Note
1119            that tdb_flags is passed in via the (otherwise unused)
1120            srvid to the attach control */
1121         tdb_flags &= (TDB_NOSYNC|TDB_INCOMPATIBLE_HASH);
1122
1123         /* see if we already have this name */
1124         db = ctdb_db_handle(ctdb, db_name);
1125         if (db) {
1126                 if (db->persistent != persistent) {
1127                         DEBUG(DEBUG_ERR, ("ERROR: DB Attach %spersistent to %spersistent "
1128                                           "database %s\n", persistent ? "" : "non-",
1129                                           db-> persistent ? "" : "non-", db_name));
1130                         return -1;
1131                 }
1132                 outdata->dptr  = (uint8_t *)&db->db_id;
1133                 outdata->dsize = sizeof(db->db_id);
1134                 tdb_add_flags(db->ltdb->tdb, tdb_flags);
1135                 return 0;
1136         }
1137
1138         with_jenkinshash = (tdb_flags & TDB_INCOMPATIBLE_HASH) ? true : false;
1139
1140         if (ctdb_local_attach(ctdb, db_name, persistent, NULL,
1141                               with_jenkinshash) != 0) {
1142                 return -1;
1143         }
1144
1145         db = ctdb_db_handle(ctdb, db_name);
1146         if (!db) {
1147                 DEBUG(DEBUG_ERR,("Failed to find db handle for name '%s'\n", db_name));
1148                 return -1;
1149         }
1150
1151         /* remember the flags the client has specified */
1152         tdb_add_flags(db->ltdb->tdb, tdb_flags);
1153
1154         outdata->dptr  = (uint8_t *)&db->db_id;
1155         outdata->dsize = sizeof(db->db_id);
1156
1157         /* Try to ensure it's locked in mem */
1158         lockdown_memory(ctdb->valgrinding);
1159
1160         /* tell all the other nodes about this database */
1161         ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_ALL, tdb_flags,
1162                                  persistent?CTDB_CONTROL_DB_ATTACH_PERSISTENT:
1163                                                 CTDB_CONTROL_DB_ATTACH,
1164                                  0, CTDB_CTRL_FLAG_NOREPLY,
1165                                  indata, NULL, NULL);
1166
1167         /* success */
1168         return 0;
1169 }
1170
1171 /*
1172  * a client has asked to detach from a database
1173  */
1174 int32_t ctdb_control_db_detach(struct ctdb_context *ctdb, TDB_DATA indata,
1175                                uint32_t client_id)
1176 {
1177         uint32_t db_id;
1178         struct ctdb_db_context *ctdb_db;
1179         struct ctdb_client *client = NULL;
1180
1181         db_id = *(uint32_t *)indata.dptr;
1182         ctdb_db = find_ctdb_db(ctdb, db_id);
1183         if (ctdb_db == NULL) {
1184                 DEBUG(DEBUG_ERR, ("Invalid dbid 0x%08x in DB detach\n",
1185                                   db_id));
1186                 return -1;
1187         }
1188
1189         if (ctdb->tunable.allow_client_db_attach == 1) {
1190                 DEBUG(DEBUG_ERR, ("DB detach from database %s denied. "
1191                                   "Clients are allowed access to databases "
1192                                   "(AllowClientDBAccess == 1)\n",
1193                                   ctdb_db->db_name));
1194                 return -1;
1195         }
1196
1197         if (ctdb_db->persistent) {
1198                 DEBUG(DEBUG_ERR, ("DB detach from persistent database %s "
1199                                   "denied\n", ctdb_db->db_name));
1200                 return -1;
1201         }
1202
1203         /* Cannot detach from database when in recovery */
1204         if (ctdb->recovery_mode == CTDB_RECOVERY_ACTIVE) {
1205                 DEBUG(DEBUG_ERR, ("DB detach denied while in recovery\n"));
1206                 return -1;
1207         }
1208
1209         /* If a control comes from a client, then broadcast it to all nodes.
1210          * Do the actual detach only if the control comes from other daemons.
1211          */
1212         if (client_id != 0) {
1213                 client = ctdb_reqid_find(ctdb, client_id, struct ctdb_client);
1214                 if (client != NULL) {
1215                         /* forward the control to all the nodes */
1216                         ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_ALL, 0,
1217                                                  CTDB_CONTROL_DB_DETACH, 0,
1218                                                  CTDB_CTRL_FLAG_NOREPLY,
1219                                                  indata, NULL, NULL);
1220                         return 0;
1221                 }
1222                 DEBUG(DEBUG_ERR, ("Client has gone away. Failing DB detach "
1223                                   "for database '%s'\n", ctdb_db->db_name));
1224                 return -1;
1225         }
1226
1227         /* Detach database from recoverd */
1228         if (ctdb_daemon_send_message(ctdb, ctdb->pnn,
1229                                      CTDB_SRVID_DETACH_DATABASE,
1230                                      indata) != 0) {
1231                 DEBUG(DEBUG_ERR, ("Unable to detach DB from recoverd\n"));
1232                 return -1;
1233         }
1234
1235         /* Disable vacuuming and drop all vacuuming data */
1236         talloc_free(ctdb_db->vacuum_handle);
1237         talloc_free(ctdb_db->delete_queue);
1238
1239         /* Terminate any deferred fetch */
1240         talloc_free(ctdb_db->deferred_fetch);
1241
1242         /* Terminate any traverses */
1243         while (ctdb_db->traverse) {
1244                 talloc_free(ctdb_db->traverse);
1245         }
1246
1247         /* Terminate any revokes */
1248         while (ctdb_db->revokechild_active) {
1249                 talloc_free(ctdb_db->revokechild_active);
1250         }
1251
1252         /* Free readonly tracking database */
1253         if (ctdb_db->readonly) {
1254                 talloc_free(ctdb_db->rottdb);
1255         }
1256
1257         DLIST_REMOVE(ctdb->db_list, ctdb_db);
1258
1259         DEBUG(DEBUG_NOTICE, ("Detached from database '%s'\n",
1260                              ctdb_db->db_name));
1261         talloc_free(ctdb_db);
1262
1263         return 0;
1264 }
1265
1266 /*
1267   attach to all existing persistent databases
1268  */
1269 static int ctdb_attach_persistent(struct ctdb_context *ctdb,
1270                                   const char *unhealthy_reason)
1271 {
1272         DIR *d;
1273         struct dirent *de;
1274
1275         /* open the persistent db directory and scan it for files */
1276         d = opendir(ctdb->db_directory_persistent);
1277         if (d == NULL) {
1278                 return 0;
1279         }
1280
1281         while ((de=readdir(d))) {
1282                 char *p, *s, *q;
1283                 size_t len = strlen(de->d_name);
1284                 uint32_t node;
1285                 int invalid_name = 0;
1286                 
1287                 s = talloc_strdup(ctdb, de->d_name);
1288                 if (s == NULL) {
1289                         closedir(d);
1290                         CTDB_NO_MEMORY(ctdb, s);
1291                 }
1292
1293                 /* only accept names ending in .tdb */
1294                 p = strstr(s, ".tdb.");
1295                 if (len < 7 || p == NULL) {
1296                         talloc_free(s);
1297                         continue;
1298                 }
1299
1300                 /* only accept names ending with .tdb. and any number of digits */
1301                 q = p+5;
1302                 while (*q != 0 && invalid_name == 0) {
1303                         if (!isdigit(*q++)) {
1304                                 invalid_name = 1;
1305                         }
1306                 }
1307                 if (invalid_name == 1 || sscanf(p+5, "%u", &node) != 1 || node != ctdb->pnn) {
1308                         DEBUG(DEBUG_ERR,("Ignoring persistent database '%s'\n", de->d_name));
1309                         talloc_free(s);
1310                         continue;
1311                 }
1312                 p[4] = 0;
1313
1314                 if (ctdb_local_attach(ctdb, s, true, unhealthy_reason, false) != 0) {
1315                         DEBUG(DEBUG_ERR,("Failed to attach to persistent database '%s'\n", de->d_name));
1316                         closedir(d);
1317                         talloc_free(s);
1318                         return -1;
1319                 }
1320
1321                 DEBUG(DEBUG_INFO,("Attached to persistent database %s\n", s));
1322
1323                 talloc_free(s);
1324         }
1325         closedir(d);
1326         return 0;
1327 }
1328
1329 int ctdb_attach_databases(struct ctdb_context *ctdb)
1330 {
1331         int ret;
1332         char *persistent_health_path = NULL;
1333         char *unhealthy_reason = NULL;
1334         bool first_try = true;
1335
1336         persistent_health_path = talloc_asprintf(ctdb, "%s/%s.%u",
1337                                                  ctdb->db_directory_state,
1338                                                  PERSISTENT_HEALTH_TDB,
1339                                                  ctdb->pnn);
1340         if (persistent_health_path == NULL) {
1341                 DEBUG(DEBUG_CRIT,(__location__ " talloc_asprintf() failed\n"));
1342                 return -1;
1343         }
1344
1345 again:
1346
1347         ctdb->db_persistent_health = tdb_wrap_open(ctdb, persistent_health_path,
1348                                                    0, TDB_DISALLOW_NESTING,
1349                                                    O_CREAT | O_RDWR, 0600);
1350         if (ctdb->db_persistent_health == NULL) {
1351                 struct tdb_wrap *tdb;
1352
1353                 if (!first_try) {
1354                         DEBUG(DEBUG_CRIT,("Failed to open tdb '%s': %d - %s\n",
1355                                           persistent_health_path,
1356                                           errno,
1357                                           strerror(errno)));
1358                         talloc_free(persistent_health_path);
1359                         talloc_free(unhealthy_reason);
1360                         return -1;
1361                 }
1362                 first_try = false;
1363
1364                 unhealthy_reason = talloc_asprintf(ctdb, "WARNING - '%s' %s - %s",
1365                                                    persistent_health_path,
1366                                                    "was cleared after a failure",
1367                                                    "manual verification needed");
1368                 if (unhealthy_reason == NULL) {
1369                         DEBUG(DEBUG_CRIT,(__location__ " talloc_asprintf() failed\n"));
1370                         talloc_free(persistent_health_path);
1371                         return -1;
1372                 }
1373
1374                 DEBUG(DEBUG_CRIT,("Failed to open tdb '%s' - retrying after CLEAR_IF_FIRST\n",
1375                                   persistent_health_path));
1376                 tdb = tdb_wrap_open(ctdb, persistent_health_path,
1377                                     0, TDB_CLEAR_IF_FIRST | TDB_DISALLOW_NESTING,
1378                                     O_CREAT | O_RDWR, 0600);
1379                 if (tdb) {
1380                         DEBUG(DEBUG_CRIT,("Failed to open tdb '%s' - with CLEAR_IF_FIRST: %d - %s\n",
1381                                           persistent_health_path,
1382                                           errno,
1383                                           strerror(errno)));
1384                         talloc_free(persistent_health_path);
1385                         talloc_free(unhealthy_reason);
1386                         return -1;
1387                 }
1388
1389                 talloc_free(tdb);
1390                 goto again;
1391         }
1392         ret = tdb_check(ctdb->db_persistent_health->tdb, NULL, NULL);
1393         if (ret != 0) {
1394                 struct tdb_wrap *tdb;
1395
1396                 talloc_free(ctdb->db_persistent_health);
1397                 ctdb->db_persistent_health = NULL;
1398
1399                 if (!first_try) {
1400                         DEBUG(DEBUG_CRIT,("tdb_check('%s') failed\n",
1401                                           persistent_health_path));
1402                         talloc_free(persistent_health_path);
1403                         talloc_free(unhealthy_reason);
1404                         return -1;
1405                 }
1406                 first_try = false;
1407
1408                 unhealthy_reason = talloc_asprintf(ctdb, "WARNING - '%s' %s - %s",
1409                                                    persistent_health_path,
1410                                                    "was cleared after a failure",
1411                                                    "manual verification needed");
1412                 if (unhealthy_reason == NULL) {
1413                         DEBUG(DEBUG_CRIT,(__location__ " talloc_asprintf() failed\n"));
1414                         talloc_free(persistent_health_path);
1415                         return -1;
1416                 }
1417
1418                 DEBUG(DEBUG_CRIT,("tdb_check('%s') failed - retrying after CLEAR_IF_FIRST\n",
1419                                   persistent_health_path));
1420                 tdb = tdb_wrap_open(ctdb, persistent_health_path,
1421                                     0, TDB_CLEAR_IF_FIRST | TDB_DISALLOW_NESTING,
1422                                     O_CREAT | O_RDWR, 0600);
1423                 if (tdb) {
1424                         DEBUG(DEBUG_CRIT,("Failed to open tdb '%s' - with CLEAR_IF_FIRST: %d - %s\n",
1425                                           persistent_health_path,
1426                                           errno,
1427                                           strerror(errno)));
1428                         talloc_free(persistent_health_path);
1429                         talloc_free(unhealthy_reason);
1430                         return -1;
1431                 }
1432
1433                 talloc_free(tdb);
1434                 goto again;
1435         }
1436         talloc_free(persistent_health_path);
1437
1438         ret = ctdb_attach_persistent(ctdb, unhealthy_reason);
1439         talloc_free(unhealthy_reason);
1440         if (ret != 0) {
1441                 return ret;
1442         }
1443
1444         return 0;
1445 }
1446
1447 /*
1448   called when a broadcast seqnum update comes in
1449  */
1450 int32_t ctdb_ltdb_update_seqnum(struct ctdb_context *ctdb, uint32_t db_id, uint32_t srcnode)
1451 {
1452         struct ctdb_db_context *ctdb_db;
1453         if (srcnode == ctdb->pnn) {
1454                 /* don't update ourselves! */
1455                 return 0;
1456         }
1457
1458         ctdb_db = find_ctdb_db(ctdb, db_id);
1459         if (!ctdb_db) {
1460                 DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_ltdb_update_seqnum\n", db_id));
1461                 return -1;
1462         }
1463
1464         if (ctdb_db->unhealthy_reason) {
1465                 DEBUG(DEBUG_ERR,("db(%s) unhealty in ctdb_ltdb_update_seqnum: %s\n",
1466                                  ctdb_db->db_name, ctdb_db->unhealthy_reason));
1467                 return -1;
1468         }
1469
1470         tdb_increment_seqnum_nonblock(ctdb_db->ltdb->tdb);
1471         ctdb_db->seqnum = tdb_get_seqnum(ctdb_db->ltdb->tdb);
1472         return 0;
1473 }
1474
1475 /*
1476   timer to check for seqnum changes in a ltdb and propogate them
1477  */
1478 static void ctdb_ltdb_seqnum_check(struct event_context *ev, struct timed_event *te, 
1479                                    struct timeval t, void *p)
1480 {
1481         struct ctdb_db_context *ctdb_db = talloc_get_type(p, struct ctdb_db_context);
1482         struct ctdb_context *ctdb = ctdb_db->ctdb;
1483         uint32_t new_seqnum = tdb_get_seqnum(ctdb_db->ltdb->tdb);
1484         if (new_seqnum != ctdb_db->seqnum) {
1485                 /* something has changed - propogate it */
1486                 TDB_DATA data;
1487                 data.dptr = (uint8_t *)&ctdb_db->db_id;
1488                 data.dsize = sizeof(uint32_t);
1489                 ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_VNNMAP, 0,
1490                                          CTDB_CONTROL_UPDATE_SEQNUM, 0, CTDB_CTRL_FLAG_NOREPLY,
1491                                          data, NULL, NULL);             
1492         }
1493         ctdb_db->seqnum = new_seqnum;
1494
1495         /* setup a new timer */
1496         ctdb_db->seqnum_update =
1497                 event_add_timed(ctdb->ev, ctdb_db, 
1498                                 timeval_current_ofs(ctdb->tunable.seqnum_interval/1000, (ctdb->tunable.seqnum_interval%1000)*1000),
1499                                 ctdb_ltdb_seqnum_check, ctdb_db);
1500 }
1501
1502 /*
1503   enable seqnum handling on this db
1504  */
1505 int32_t ctdb_ltdb_enable_seqnum(struct ctdb_context *ctdb, uint32_t db_id)
1506 {
1507         struct ctdb_db_context *ctdb_db;
1508         ctdb_db = find_ctdb_db(ctdb, db_id);
1509         if (!ctdb_db) {
1510                 DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_ltdb_enable_seqnum\n", db_id));
1511                 return -1;
1512         }
1513
1514         if (ctdb_db->seqnum_update == NULL) {
1515                 ctdb_db->seqnum_update =
1516                         event_add_timed(ctdb->ev, ctdb_db, 
1517                                         timeval_current_ofs(ctdb->tunable.seqnum_interval/1000, (ctdb->tunable.seqnum_interval%1000)*1000),
1518                                         ctdb_ltdb_seqnum_check, ctdb_db);
1519         }
1520
1521         tdb_enable_seqnum(ctdb_db->ltdb->tdb);
1522         ctdb_db->seqnum = tdb_get_seqnum(ctdb_db->ltdb->tdb);
1523         return 0;
1524 }
1525
1526 int32_t ctdb_control_set_db_priority(struct ctdb_context *ctdb, TDB_DATA indata,
1527                                      uint32_t client_id)
1528 {
1529         struct ctdb_db_priority *db_prio = (struct ctdb_db_priority *)indata.dptr;
1530         struct ctdb_db_context *ctdb_db;
1531
1532         ctdb_db = find_ctdb_db(ctdb, db_prio->db_id);
1533         if (!ctdb_db) {
1534                 if (!(ctdb->nodes[ctdb->pnn]->flags & NODE_FLAGS_INACTIVE)) {
1535                         DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_set_db_priority\n",
1536                                          db_prio->db_id));
1537                 }
1538                 return 0;
1539         }
1540
1541         if ((db_prio->priority<1) || (db_prio->priority>NUM_DB_PRIORITIES)) {
1542                 DEBUG(DEBUG_ERR,("Trying to set invalid priority : %u\n", db_prio->priority));
1543                 return 0;
1544         }
1545
1546         ctdb_db->priority = db_prio->priority;
1547         DEBUG(DEBUG_INFO,("Setting DB priority to %u for db 0x%08x\n", db_prio->priority, db_prio->db_id));
1548
1549         if (client_id != 0) {
1550                 /* Broadcast the update to the rest of the cluster */
1551                 ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_ALL, 0,
1552                                          CTDB_CONTROL_SET_DB_PRIORITY, 0,
1553                                          CTDB_CTRL_FLAG_NOREPLY, indata,
1554                                          NULL, NULL);
1555         }
1556         return 0;
1557 }
1558
1559
1560 int ctdb_set_db_sticky(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db)
1561 {
1562         if (ctdb_db->sticky) {
1563                 return 0;
1564         }
1565
1566         if (ctdb_db->persistent) {
1567                 DEBUG(DEBUG_ERR,("Trying to set persistent database with sticky property\n"));
1568                 return -1;
1569         }
1570
1571         ctdb_db->sticky_records = trbt_create(ctdb_db, 0);
1572
1573         ctdb_db->sticky = true;
1574
1575         DEBUG(DEBUG_NOTICE,("set db sticky %s\n", ctdb_db->db_name));
1576
1577         return 0;
1578 }
1579
1580 int32_t ctdb_control_get_db_statistics(struct ctdb_context *ctdb,
1581                                 uint32_t db_id,
1582                                 TDB_DATA *outdata)
1583 {
1584         struct ctdb_db_context *ctdb_db;
1585         struct ctdb_db_statistics *stats;
1586         int i;
1587         int len;
1588         char *ptr;
1589
1590         ctdb_db = find_ctdb_db(ctdb, db_id);
1591         if (!ctdb_db) {
1592                 DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in get_db_statistics\n", db_id));
1593                 return -1;
1594         }
1595
1596         len = offsetof(struct ctdb_db_statistics, hot_keys_wire);
1597         for (i = 0; i < MAX_HOT_KEYS; i++) {
1598                 len += ctdb_db->statistics.hot_keys[i].key.dsize;
1599         }
1600
1601         stats = talloc_size(outdata, len);
1602         if (stats == NULL) {
1603                 DEBUG(DEBUG_ERR,("Failed to allocate db statistics structure\n"));
1604                 return -1;
1605         }
1606
1607         *stats = ctdb_db->statistics;
1608
1609         stats->num_hot_keys = MAX_HOT_KEYS;
1610
1611         ptr = &stats->hot_keys_wire[0];
1612         for (i = 0; i < MAX_HOT_KEYS; i++) {
1613                 memcpy(ptr, ctdb_db->statistics.hot_keys[i].key.dptr,
1614                        ctdb_db->statistics.hot_keys[i].key.dsize);
1615                 ptr += ctdb_db->statistics.hot_keys[i].key.dsize;
1616         }
1617
1618         outdata->dptr  = (uint8_t *)stats;
1619         outdata->dsize = len;
1620
1621         return 0;
1622 }