1 /* cache.c - routines to maintain an in-core cache of entries */
3 /* This work is part of OpenLDAP Software <http://www.openldap.org/>.
5 * Copyright 2000-2008 The OpenLDAP Foundation.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted only as authorized by the OpenLDAP
12 * A copy of this license is available in the file LICENSE in the
13 * top-level directory of the distribution or, alternatively, at
14 * <http://www.OpenLDAP.org/license.html>.
22 #include <ac/string.h>
23 #include <ac/socket.h>
32 #define bdb_cache_lru_purge hdb_cache_lru_purge
34 static void bdb_cache_lru_purge( struct bdb_info *bdb );
36 static int bdb_cache_delete_internal(Cache *cache, EntryInfo *e, int decr);
40 static void bdb_lru_print(Cache *cache);
44 /* For concurrency experiments only! */
46 #define ldap_pvt_thread_rdwr_wlock(a) 0
47 #define ldap_pvt_thread_rdwr_wunlock(a) 0
48 #define ldap_pvt_thread_rdwr_rlock(a) 0
49 #define ldap_pvt_thread_rdwr_runlock(a) 0
53 #define ldap_pvt_thread_mutex_trylock(a) 0
57 bdb_cache_entryinfo_new( Cache *cache )
61 if ( cache->c_eifree ) {
62 ldap_pvt_thread_mutex_lock( &cache->c_eifree_mutex );
63 if ( cache->c_eifree ) {
65 cache->c_eifree = ei->bei_lrunext;
68 ldap_pvt_thread_mutex_unlock( &cache->c_eifree_mutex );
71 ei = ch_calloc(1, sizeof(EntryInfo));
72 ldap_pvt_thread_mutex_init( &ei->bei_kids_mutex );
75 ei->bei_state = CACHE_ENTRY_REFERENCED;
81 bdb_cache_entryinfo_free( Cache *cache, EntryInfo *ei )
83 free( ei->bei_nrdn.bv_val );
84 ei->bei_nrdn.bv_val = NULL;
86 free( ei->bei_rdn.bv_val );
87 ei->bei_rdn.bv_val = NULL;
92 ei->bei_parent = NULL;
94 ei->bei_lruprev = NULL;
96 ldap_pvt_thread_mutex_lock( &cache->c_eifree_mutex );
97 ei->bei_lrunext = cache->c_eifree;
99 ldap_pvt_thread_mutex_unlock( &cache->c_eifree_mutex );
102 #define LRU_DEL( c, e ) do { \
103 if ( e == (c)->c_lruhead ) (c)->c_lruhead = e->bei_lruprev; \
104 if ( e == (c)->c_lrutail ) (c)->c_lrutail = e->bei_lruprev; \
105 e->bei_lrunext->bei_lruprev = e->bei_lruprev; \
106 e->bei_lruprev->bei_lrunext = e->bei_lrunext; \
107 e->bei_lruprev = NULL; \
110 /* Note - we now use a Second-Chance / Clock algorithm instead of
111 * Least-Recently-Used. This tremendously improves concurrency
112 * because we no longer need to manipulate the lists every time an
113 * entry is touched. We only need to lock the lists when adding
114 * or deleting an entry. It's now a circular doubly-linked list.
115 * We always append to the tail, but the head traverses the circle
116 * during a purge operation.
119 bdb_cache_lru_link( struct bdb_info *bdb, EntryInfo *ei )
122 /* Already linked, ignore */
123 if ( ei->bei_lruprev )
126 /* Insert into circular LRU list */
127 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_lru_mutex );
129 ei->bei_lruprev = bdb->bi_cache.c_lrutail;
130 if ( bdb->bi_cache.c_lrutail ) {
131 ei->bei_lrunext = bdb->bi_cache.c_lrutail->bei_lrunext;
132 bdb->bi_cache.c_lrutail->bei_lrunext = ei;
133 if ( ei->bei_lrunext )
134 ei->bei_lrunext->bei_lruprev = ei;
136 ei->bei_lrunext = ei->bei_lruprev = ei;
137 bdb->bi_cache.c_lruhead = ei;
139 bdb->bi_cache.c_lrutail = ei;
140 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_lru_mutex );
147 /* #define NO_DB_LOCK 1 */
148 /* Note: The BerkeleyDB locks are much slower than regular
149 * mutexes or rdwr locks. But the BDB implementation has the
150 * advantage of using a fixed size lock table, instead of
151 * allocating a lock object per entry in the DB. That's a
152 * key benefit for scaling. It also frees us from worrying
153 * about undetectable deadlocks between BDB activity and our
154 * own cache activity. It's still worth exploring faster
155 * alternatives though.
158 /* Atomically release and reacquire a lock */
160 bdb_cache_entry_db_relock(
161 struct bdb_info *bdb,
175 if ( !lock ) return 0;
177 lockobj.data = &ei->bei_id;
178 lockobj.size = sizeof(ei->bei_id) + 1;
180 list[0].op = DB_LOCK_PUT;
181 list[0].lock = *lock;
182 list[1].op = DB_LOCK_GET;
183 list[1].lock = *lock;
184 list[1].mode = rw ? DB_LOCK_WRITE : DB_LOCK_READ;
185 list[1].obj = &lockobj;
186 rc = bdb->bi_dbenv->lock_vec(bdb->bi_dbenv, BDB_LOCKID(locker), tryOnly ? DB_LOCK_NOWAIT : 0,
189 if (rc && !tryOnly) {
190 Debug( LDAP_DEBUG_TRACE,
191 "bdb_cache_entry_db_relock: entry %ld, rw %d, rc %d\n",
192 ei->bei_id, rw, rc );
194 *lock = list[1].lock;
201 bdb_cache_entry_db_lock( struct bdb_info *bdb, BDB_LOCKER locker, EntryInfo *ei,
202 int rw, int tryOnly, DB_LOCK *lock )
211 if ( !lock ) return 0;
214 db_rw = DB_LOCK_WRITE;
216 db_rw = DB_LOCK_READ;
218 lockobj.data = &ei->bei_id;
219 lockobj.size = sizeof(ei->bei_id) + 1;
221 rc = LOCK_GET(bdb->bi_dbenv, BDB_LOCKID(locker), tryOnly ? DB_LOCK_NOWAIT : 0,
222 &lockobj, db_rw, lock);
223 if (rc && !tryOnly) {
224 Debug( LDAP_DEBUG_TRACE,
225 "bdb_cache_entry_db_lock: entry %ld, rw %d, rc %d\n",
226 ei->bei_id, rw, rc );
229 #endif /* NO_DB_LOCK */
233 bdb_cache_entry_db_unlock ( struct bdb_info *bdb, DB_LOCK *lock )
240 if ( !lock || lock->mode == DB_LOCK_NG ) return 0;
242 rc = LOCK_PUT ( bdb->bi_dbenv, lock );
248 bdb_cache_return_entry_rw( struct bdb_info *bdb, Entry *e,
249 int rw, DB_LOCK *lock )
255 bdb_cache_entry_db_unlock( bdb, lock );
257 bdb_cache_entryinfo_lock( ei );
258 if ( ei->bei_state & CACHE_ENTRY_NOT_CACHED ) {
260 ei->bei_state ^= CACHE_ENTRY_NOT_CACHED;
263 bdb_cache_entryinfo_unlock( ei );
267 bdb_entry_return( e );
272 bdb_cache_entryinfo_destroy( EntryInfo *e )
274 ldap_pvt_thread_mutex_destroy( &e->bei_kids_mutex );
275 free( e->bei_nrdn.bv_val );
277 free( e->bei_rdn.bv_val );
283 /* Do a length-ordered sort on normalized RDNs */
285 bdb_rdn_cmp( const void *v_e1, const void *v_e2 )
287 const EntryInfo *e1 = v_e1, *e2 = v_e2;
288 int rc = e1->bei_nrdn.bv_len - e2->bei_nrdn.bv_len;
290 rc = strncmp( e1->bei_nrdn.bv_val, e2->bei_nrdn.bv_val,
291 e1->bei_nrdn.bv_len );
297 bdb_id_cmp( const void *v_e1, const void *v_e2 )
299 const EntryInfo *e1 = v_e1, *e2 = v_e2;
300 return e1->bei_id - e2->bei_id;
304 bdb_id_dup_err( void *v1, void *v2 )
307 e2->bei_lrunext = v1;
311 /* Create an entryinfo in the cache. Caller must release the locks later.
314 bdb_entryinfo_add_internal(
315 struct bdb_info *bdb,
319 EntryInfo *ei2 = NULL;
323 ei2 = bdb_cache_entryinfo_new( &bdb->bi_cache );
325 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
326 bdb_cache_entryinfo_lock( ei->bei_parent );
328 ei2->bei_id = ei->bei_id;
329 ei2->bei_parent = ei->bei_parent;
331 ei2->bei_rdn = ei->bei_rdn;
333 #ifdef SLAP_ZONE_ALLOC
337 /* Add to cache ID tree */
338 if (avl_insert( &bdb->bi_cache.c_idtree, ei2, bdb_id_cmp,
340 EntryInfo *eix = ei2->bei_lrunext;
341 bdb_cache_entryinfo_free( &bdb->bi_cache, ei2 );
344 /* It got freed above because its value was
347 ei->bei_rdn.bv_val = NULL;
352 bdb->bi_cache.c_eiused++;
353 ber_dupbv( &ei2->bei_nrdn, &ei->bei_nrdn );
355 /* This is a new leaf node. But if parent had no kids, then it was
356 * a leaf and we would be decrementing that. So, only increment if
357 * the parent already has kids.
359 if ( ei->bei_parent->bei_kids || !ei->bei_parent->bei_id )
360 bdb->bi_cache.c_leaves++;
361 rc = avl_insert( &ei->bei_parent->bei_kids, ei2, bdb_rdn_cmp,
364 /* This should never happen; entry cache is corrupt */
365 bdb->bi_dbenv->log_flush( bdb->bi_dbenv, NULL );
369 ei->bei_parent->bei_ckids++;
377 /* Find the EntryInfo for the requested DN. If the DN cannot be found, return
378 * the info for its closest ancestor. *res should be NULL to process a
379 * complete DN starting from the tree root. Otherwise *res must be the
380 * immediate parent of the requested DN, and only the RDN will be searched.
381 * The EntryInfo is locked upon return and must be unlocked by the caller.
390 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
391 EntryInfo ei, *eip, *ei2;
395 /* this function is always called with normalized DN */
397 /* we're doing a onelevel search for an RDN */
398 ei.bei_nrdn.bv_val = ndn->bv_val;
399 ei.bei_nrdn.bv_len = dn_rdnlen( op->o_bd, ndn );
402 /* we're searching a full DN from the root */
403 ptr = ndn->bv_val + ndn->bv_len - op->o_bd->be_nsuffix[0].bv_len;
404 ei.bei_nrdn.bv_val = ptr;
405 ei.bei_nrdn.bv_len = op->o_bd->be_nsuffix[0].bv_len;
406 /* Skip to next rdn if suffix is empty */
407 if ( ei.bei_nrdn.bv_len == 0 ) {
408 for (ptr = ei.bei_nrdn.bv_val - 2; ptr > ndn->bv_val
409 && !DN_SEPARATOR(*ptr); ptr--) /* empty */;
410 if ( ptr >= ndn->bv_val ) {
411 if (DN_SEPARATOR(*ptr)) ptr++;
412 ei.bei_nrdn.bv_len = ei.bei_nrdn.bv_val - ptr;
413 ei.bei_nrdn.bv_val = ptr;
416 eip = &bdb->bi_cache.c_dntree;
419 for ( bdb_cache_entryinfo_lock( eip ); eip; ) {
420 eip->bei_state |= CACHE_ENTRY_REFERENCED;
422 ei2 = (EntryInfo *)avl_find( eip->bei_kids, &ei, bdb_rdn_cmp );
425 int len = ei.bei_nrdn.bv_len;
427 if ( BER_BVISEMPTY( ndn )) {
432 ei.bei_nrdn.bv_len = ndn->bv_len -
433 (ei.bei_nrdn.bv_val - ndn->bv_val);
434 bdb_cache_entryinfo_unlock( eip );
436 BDB_LOG_PRINTF( bdb->bi_dbenv, NULL, "slapd Reading %s",
437 ei.bei_nrdn.bv_val );
439 lock.mode = DB_LOCK_NG;
440 rc = bdb_dn2id( op, &ei.bei_nrdn, &ei, locker, &lock );
442 bdb_cache_entryinfo_lock( eip );
443 bdb_cache_entry_db_unlock( bdb, &lock );
448 BDB_LOG_PRINTF( bdb->bi_dbenv, NULL, "slapd Read got %s(%d)",
449 ei.bei_nrdn.bv_val, ei.bei_id );
451 /* DN exists but needs to be added to cache */
452 ei.bei_nrdn.bv_len = len;
453 rc = bdb_entryinfo_add_internal( bdb, &ei, &ei2 );
454 /* add_internal left eip and c_rwlock locked */
455 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
456 bdb_cache_entry_db_unlock( bdb, &lock );
461 } else if ( ei2->bei_state & CACHE_ENTRY_DELETED ) {
462 /* In the midst of deleting? Give it a chance to
465 bdb_cache_entryinfo_unlock( eip );
466 ldap_pvt_thread_yield();
467 bdb_cache_entryinfo_lock( eip );
471 bdb_cache_entryinfo_unlock( eip );
472 bdb_cache_entryinfo_lock( ei2 );
476 /* Advance to next lower RDN */
477 for (ptr = ei.bei_nrdn.bv_val - 2; ptr > ndn->bv_val
478 && !DN_SEPARATOR(*ptr); ptr--) /* empty */;
479 if ( ptr >= ndn->bv_val ) {
480 if (DN_SEPARATOR(*ptr)) ptr++;
481 ei.bei_nrdn.bv_len = ei.bei_nrdn.bv_val - ptr - 1;
482 ei.bei_nrdn.bv_val = ptr;
484 if ( ptr < ndn->bv_val ) {
494 /* Walk up the tree from a child node, looking for an ID that's already
495 * been linked into the cache.
498 hdb_cache_find_parent(
504 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
505 EntryInfo ei, eip, *ei2 = NULL, *ein = NULL, *eir = NULL;
513 rc = hdb_dn2id_parent( op, locker, &ei, &eip.bei_id );
516 /* Save the previous node, if any */
519 /* Create a new node for the current ID */
520 ein = bdb_cache_entryinfo_new( &bdb->bi_cache );
521 ein->bei_id = ei.bei_id;
522 ein->bei_kids = ei.bei_kids;
523 ein->bei_nrdn = ei.bei_nrdn;
524 ein->bei_rdn = ei.bei_rdn;
525 ein->bei_ckids = ei.bei_ckids;
526 #ifdef SLAP_ZONE_ALLOC
531 /* This node is not fully connected yet */
532 ein->bei_state |= CACHE_ENTRY_NOT_LINKED;
534 /* Insert this node into the ID tree */
535 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
536 if ( avl_insert( &bdb->bi_cache.c_idtree, (caddr_t)ein,
537 bdb_id_cmp, bdb_id_dup_err ) ) {
538 EntryInfo *eix = ein->bei_lrunext;
540 /* Someone else created this node just before us.
541 * Free our new copy and use the existing one.
543 bdb_cache_entryinfo_free( &bdb->bi_cache, ein );
546 /* Link in any kids we've already processed */
548 bdb_cache_entryinfo_lock( ein );
549 avl_insert( &ein->bei_kids, (caddr_t)ei2,
550 bdb_rdn_cmp, avl_dup_error );
552 bdb_cache_entryinfo_unlock( ein );
556 /* If this is the first time, save this node
557 * to be returned later.
559 if ( eir == NULL ) eir = ein;
561 /* If there was a previous node, link it to this one */
562 if ( ei2 ) ei2->bei_parent = ein;
564 /* Look for this node's parent */
566 ei2 = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
567 (caddr_t) &eip, bdb_id_cmp );
569 ei2 = &bdb->bi_cache.c_dntree;
571 bdb->bi_cache.c_eiused++;
572 if ( ei2 && ( ei2->bei_kids || !ei2->bei_id ))
573 bdb->bi_cache.c_leaves++;
574 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
576 /* Got the parent, link in and we're done. */
578 bdb_cache_entryinfo_lock( eir );
579 bdb_cache_entryinfo_lock( ei2 );
580 ein->bei_parent = ei2;
582 avl_insert( &ei2->bei_kids, (caddr_t)ein, bdb_rdn_cmp,
586 /* Reset all the state info */
587 for (ein = eir; ein != ei2; ein=ein->bei_parent)
588 ein->bei_state &= ~CACHE_ENTRY_NOT_LINKED;
590 bdb_cache_entryinfo_unlock( ei2 );
596 ei.bei_id = eip.bei_id;
598 avl_insert( &ei.bei_kids, (caddr_t)ein, bdb_rdn_cmp,
604 /* Used by hdb_dn2idl when loading the EntryInfo for all the children
608 struct bdb_info *bdb,
615 /* See if we already have this one */
616 bdb_cache_entryinfo_lock( ei->bei_parent );
617 ei2 = (EntryInfo *)avl_find( ei->bei_parent->bei_kids, ei, bdb_rdn_cmp );
618 bdb_cache_entryinfo_unlock( ei->bei_parent );
621 /* Not found, add it */
624 /* bei_rdn was not malloc'd before, do it now */
625 ber_dupbv( &bv, &ei->bei_rdn );
628 rc = bdb_entryinfo_add_internal( bdb, ei, res );
629 bdb_cache_entryinfo_unlock( ei->bei_parent );
630 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
632 /* Found, return it */
640 /* This is best-effort only. If all entries in the cache are
641 * busy, they will all be kept. This is unlikely to happen
642 * unless the cache is very much smaller than the working set.
645 bdb_cache_lru_purge( struct bdb_info *bdb )
647 DB_LOCK lock, *lockp;
648 EntryInfo *elru, *elnext = NULL;
649 int count, islocked, eimax;
651 /* Wait for the mutex; we're the only one trying to purge. */
652 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_lru_mutex );
654 if ( bdb->bi_cache.c_cursize <= bdb->bi_cache.c_maxsize ) {
655 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_lru_mutex );
656 bdb->bi_cache.c_purging = 0;
660 if ( bdb->bi_cache.c_locker ) {
668 /* maximum number of EntryInfo leaves to cache. In slapcat
669 * we always free all leaf nodes.
671 if ( slapMode & SLAP_TOOL_READONLY )
674 eimax = bdb->bi_cache.c_eimax;
676 /* Look for an unused entry to remove */
677 for ( elru = bdb->bi_cache.c_lruhead; elru; elru = elnext ) {
678 elnext = elru->bei_lrunext;
680 if ( bdb_cache_entryinfo_trylock( elru ))
683 /* This flag implements the clock replacement behavior */
684 if ( elru->bei_state & ( CACHE_ENTRY_REFERENCED )) {
685 elru->bei_state &= ~CACHE_ENTRY_REFERENCED;
686 bdb_cache_entryinfo_unlock( elru );
690 /* If this node is in the process of linking into the cache,
691 * or this node is being deleted, skip it.
693 if (( elru->bei_state & ( CACHE_ENTRY_NOT_LINKED |
694 CACHE_ENTRY_DELETED | CACHE_ENTRY_LOADING )) ||
695 elru->bei_finders > 0 ) {
696 bdb_cache_entryinfo_unlock( elru );
700 /* entryinfo is locked */
703 /* If we can successfully writelock it, then
704 * the object is idle.
706 if ( bdb_cache_entry_db_lock( bdb,
707 bdb->bi_cache.c_locker, elru, 1, 1, lockp ) == 0 ) {
709 /* Free entry for this node if it's present */
711 elru->bei_e->e_private = NULL;
712 #ifdef SLAP_ZONE_ALLOC
713 bdb_entry_return( bdb, elru->bei_e, elru->bei_zseq );
715 bdb_entry_return( elru->bei_e );
720 bdb_cache_entry_db_unlock( bdb, lockp );
723 * If it is a leaf node, and we're over the limit, free it.
725 if ( elru->bei_kids ) {
726 /* Drop from list, we ignore it... */
727 LRU_DEL( &bdb->bi_cache, elru );
728 } else if ( bdb->bi_cache.c_leaves > eimax ) {
729 /* Too many leaf nodes, free this one */
730 bdb_cache_delete_internal( &bdb->bi_cache, elru, 0 );
731 bdb_cache_delete_cleanup( &bdb->bi_cache, elru );
733 } /* Leave on list until we need to free it */
737 bdb_cache_entryinfo_unlock( elru );
739 if ( count >= bdb->bi_cache.c_minfree ) {
740 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_count_mutex );
741 bdb->bi_cache.c_cursize -= count;
742 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_count_mutex );
746 if ( elnext == bdb->bi_cache.c_lruhead )
750 bdb->bi_cache.c_lruhead = elnext;
751 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_lru_mutex );
752 bdb->bi_cache.c_purging = 0;
757 struct bdb_info *bdb,
760 EntryInfo ei = { 0 },
765 ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
766 ei2 = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
767 (caddr_t) &ei, bdb_id_cmp );
768 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
773 * cache_find_id - find an entry in the cache, given id.
774 * The entry is locked for Read upon return. Call with flag ID_LOCKED if
775 * the supplied *eip was already locked.
788 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
790 int rc = 0, load = 0;
791 EntryInfo ei = { 0 };
795 #ifdef SLAP_ZONE_ALLOC
796 slap_zh_rlock(bdb->bi_cache.c_zctx);
798 /* If we weren't given any info, see if we have it already cached */
800 again: ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
801 *eip = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
802 (caddr_t) &ei, bdb_id_cmp );
804 /* If the lock attempt fails, the info is in use */
805 if ( bdb_cache_entryinfo_trylock( *eip )) {
806 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
807 /* If this node is being deleted, treat
808 * as if the delete has already finished
810 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
813 /* otherwise, wait for the info to free up */
814 ldap_pvt_thread_yield();
817 /* If this info isn't hooked up to its parent yet,
818 * unlock and wait for it to be fully initialized
820 if ( (*eip)->bei_state & CACHE_ENTRY_NOT_LINKED ) {
821 bdb_cache_entryinfo_unlock( *eip );
822 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
823 ldap_pvt_thread_yield();
828 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
831 /* See if the ID exists in the database; add it to the cache if so */
834 rc = bdb_id2entry( op->o_bd, tid, locker, id, &ep );
836 rc = bdb_cache_find_ndn( op, locker,
838 if ( *eip ) flag |= ID_LOCKED;
840 ep->e_private = NULL;
841 #ifdef SLAP_ZONE_ALLOC
842 bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
844 bdb_entry_return( ep );
850 rc = hdb_cache_find_parent(op, locker, id, eip );
851 if ( rc == 0 ) flag |= ID_LOCKED;
855 /* Ok, we found the info, do we have the entry? */
857 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
860 (*eip)->bei_finders++;
861 (*eip)->bei_state |= CACHE_ENTRY_REFERENCED;
862 /* Make sure only one thread tries to load the entry */
864 #ifdef SLAP_ZONE_ALLOC
865 if ((*eip)->bei_e && !slap_zn_validate(
866 bdb->bi_cache.c_zctx, (*eip)->bei_e, (*eip)->bei_zseq)) {
867 (*eip)->bei_e = NULL;
868 (*eip)->bei_zseq = 0;
871 if ( !(*eip)->bei_e && !((*eip)->bei_state & CACHE_ENTRY_LOADING)) {
873 (*eip)->bei_state |= CACHE_ENTRY_LOADING;
876 /* If the entry was loaded before but uncached, and we need
877 * it again, clear the uncached state
879 if ( (*eip)->bei_state & CACHE_ENTRY_NOT_CACHED ) {
880 (*eip)->bei_state ^= CACHE_ENTRY_NOT_CACHED;
881 if ( flag & ID_NOCACHE )
885 if ( flag & ID_LOCKED ) {
886 bdb_cache_entryinfo_unlock( *eip );
889 rc = bdb_cache_entry_db_lock( bdb, locker, *eip, load, 0, lock );
890 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
892 bdb_cache_entry_db_unlock( bdb, lock );
893 } else if ( rc == 0 ) {
896 rc = bdb_id2entry( op->o_bd, tid, locker, id, &ep );
899 ep->e_private = *eip;
904 #ifdef SLAP_ZONE_ALLOC
905 (*eip)->bei_zseq = *((ber_len_t *)ep - 2);
908 bdb_cache_lru_link( bdb, *eip );
909 if ( flag & ID_NOCACHE ) {
910 bdb_cache_entryinfo_lock( *eip );
911 (*eip)->bei_state |= CACHE_ENTRY_NOT_CACHED;
912 bdb_cache_entryinfo_unlock( *eip );
916 /* If we succeeded, downgrade back to a readlock. */
917 rc = bdb_cache_entry_db_relock( bdb, locker,
920 /* Otherwise, release the lock. */
921 bdb_cache_entry_db_unlock( bdb, lock );
923 } else if ( !(*eip)->bei_e ) {
924 /* Some other thread is trying to load the entry,
925 * wait for it to finish.
927 bdb_cache_entry_db_unlock( bdb, lock );
928 bdb_cache_entryinfo_lock( *eip );
933 /* Check for subtree renames
935 rc = bdb_fix_dn( (*eip)->bei_e, 1 );
937 bdb_cache_entry_db_relock( bdb,
938 locker, *eip, 1, 0, lock );
939 /* check again in case other modifier did it already */
940 if ( bdb_fix_dn( (*eip)->bei_e, 1 ) )
941 rc = bdb_fix_dn( (*eip)->bei_e, 2 );
942 bdb_cache_entry_db_relock( bdb,
943 locker, *eip, 0, 0, lock );
947 bdb_cache_entryinfo_lock( *eip );
948 (*eip)->bei_finders--;
950 (*eip)->bei_state ^= CACHE_ENTRY_LOADING;
951 bdb_cache_entryinfo_unlock( *eip );
955 if ( flag & ID_LOCKED ) {
956 bdb_cache_entryinfo_unlock( *eip );
959 ep->e_private = NULL;
960 #ifdef SLAP_ZONE_ALLOC
961 bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
963 bdb_entry_return( ep );
970 if ( !( flag & ID_NOCACHE )) {
971 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_count_mutex );
972 bdb->bi_cache.c_cursize++;
973 if ( bdb->bi_cache.c_cursize > bdb->bi_cache.c_maxsize &&
974 !bdb->bi_cache.c_purging ) {
976 bdb->bi_cache.c_purging = 1;
978 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_count_mutex );
982 bdb_cache_lru_purge( bdb );
985 #ifdef SLAP_ZONE_ALLOC
986 if (rc == 0 && (*eip)->bei_e) {
987 slap_zn_rlock(bdb->bi_cache.c_zctx, (*eip)->bei_e);
989 slap_zh_runlock(bdb->bi_cache.c_zctx);
1002 if ( BEI(e)->bei_kids ) {
1005 if ( BEI(e)->bei_state & CACHE_ENTRY_NO_KIDS ) {
1008 rc = bdb_dn2id_children( op, txn, e );
1009 if ( rc == DB_NOTFOUND ) {
1010 BEI(e)->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
1015 /* Update the cache after a successful database Add. */
1018 struct bdb_info *bdb,
1021 struct berval *nrdn,
1028 struct berval rdn = e->e_name;
1031 ei.bei_id = e->e_id;
1032 ei.bei_parent = eip;
1033 ei.bei_nrdn = *nrdn;
1036 /* Lock this entry so that bdb_add can run to completion.
1037 * It can only fail if BDB has run out of lock resources.
1039 rc = bdb_cache_entry_db_lock( bdb, locker, &ei, 0, 0, lock );
1041 bdb_cache_entryinfo_unlock( eip );
1046 if ( nrdn->bv_len != e->e_nname.bv_len ) {
1047 char *ptr = ber_bvchr( &rdn, ',' );
1048 assert( ptr != NULL );
1049 rdn.bv_len = ptr - rdn.bv_val;
1051 ber_dupbv( &ei.bei_rdn, &rdn );
1052 if ( eip->bei_dkids ) eip->bei_dkids++;
1055 rc = bdb_entryinfo_add_internal( bdb, &ei, &new );
1056 /* bdb_csn_commit can cause this when adding the database root entry */
1058 new->bei_e->e_private = NULL;
1059 #ifdef SLAP_ZONE_ALLOC
1060 bdb_entry_return( bdb, new->bei_e, new->bei_zseq );
1062 bdb_entry_return( new->bei_e );
1067 new->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
1068 eip->bei_state &= ~CACHE_ENTRY_NO_KIDS;
1069 if (eip->bei_parent) {
1070 eip->bei_parent->bei_state &= ~CACHE_ENTRY_NO_GRANDKIDS;
1072 bdb_cache_entryinfo_unlock( eip );
1074 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
1075 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_count_mutex );
1076 ++bdb->bi_cache.c_cursize;
1077 if ( bdb->bi_cache.c_cursize > bdb->bi_cache.c_maxsize &&
1078 !bdb->bi_cache.c_purging ) {
1080 bdb->bi_cache.c_purging = 1;
1082 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_count_mutex );
1084 bdb_cache_lru_link( bdb, new );
1087 bdb_cache_lru_purge( bdb );
1094 struct bdb_info *bdb,
1096 Attribute *newAttrs,
1100 EntryInfo *ei = BEI(e);
1102 /* Get write lock on data */
1103 rc = bdb_cache_entry_db_relock( bdb, locker, ei, 1, 0, lock );
1105 /* If we've done repeated mods on a cached entry, then e_attrs
1106 * is no longer contiguous with the entry, and must be freed.
1109 if ( (void *)e->e_attrs != (void *)(e+1) ) {
1110 attrs_free( e->e_attrs );
1112 e->e_attrs = newAttrs;
1118 * Change the rdn in the entryinfo. Also move to a new parent if needed.
1122 struct bdb_info *bdb,
1124 struct berval *nrdn,
1130 EntryInfo *ei = BEI(e), *pei;
1136 /* Get write lock on data */
1137 rc = bdb_cache_entry_db_relock( bdb, locker, ei, 1, 0, lock );
1138 if ( rc ) return rc;
1140 /* If we've done repeated mods on a cached entry, then e_attrs
1141 * is no longer contiguous with the entry, and must be freed.
1143 if ( (void *)e->e_attrs != (void *)(e+1) ) {
1144 attrs_free( e->e_attrs );
1146 e->e_attrs = new->e_attrs;
1147 if( e->e_nname.bv_val < e->e_bv.bv_val ||
1148 e->e_nname.bv_val > e->e_bv.bv_val + e->e_bv.bv_len )
1150 ch_free(e->e_name.bv_val);
1151 ch_free(e->e_nname.bv_val);
1153 e->e_name = new->e_name;
1154 e->e_nname = new->e_nname;
1156 /* Lock the parent's kids AVL tree */
1157 pei = ei->bei_parent;
1158 bdb_cache_entryinfo_lock( pei );
1159 avl_delete( &pei->bei_kids, (caddr_t) ei, bdb_rdn_cmp );
1160 free( ei->bei_nrdn.bv_val );
1161 ber_dupbv( &ei->bei_nrdn, nrdn );
1164 free( ei->bei_rdn.bv_val );
1167 if ( nrdn->bv_len != e->e_nname.bv_len ) {
1168 char *ptr = ber_bvchr(&rdn, ',');
1169 assert( ptr != NULL );
1170 rdn.bv_len = ptr - rdn.bv_val;
1172 ber_dupbv( &ei->bei_rdn, &rdn );
1174 /* If new parent, decrement kid counts */
1177 if ( pei->bei_dkids ) {
1179 if ( pei->bei_dkids < 2 )
1180 pei->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
1186 ein = ei->bei_parent;
1188 ei->bei_parent = ein;
1189 bdb_cache_entryinfo_unlock( pei );
1190 bdb_cache_entryinfo_lock( ein );
1192 /* new parent now has kids */
1193 if ( ein->bei_state & CACHE_ENTRY_NO_KIDS )
1194 ein->bei_state ^= CACHE_ENTRY_NO_KIDS;
1195 /* grandparent has grandkids */
1196 if ( ein->bei_parent )
1197 ein->bei_parent->bei_state &= ~CACHE_ENTRY_NO_GRANDKIDS;
1199 /* parent might now have grandkids */
1200 if ( ein->bei_state & CACHE_ENTRY_NO_GRANDKIDS &&
1201 !(ei->bei_state & CACHE_ENTRY_NO_KIDS))
1202 ein->bei_state ^= CACHE_ENTRY_NO_GRANDKIDS;
1205 if ( ein->bei_dkids ) ein->bei_dkids++;
1210 /* Record the generation number of this change */
1211 ldap_pvt_thread_mutex_lock( &bdb->bi_modrdns_mutex );
1213 ei->bei_modrdns = bdb->bi_modrdns;
1214 ldap_pvt_thread_mutex_unlock( &bdb->bi_modrdns_mutex );
1217 avl_insert( &ein->bei_kids, ei, bdb_rdn_cmp, avl_dup_error );
1218 bdb_cache_entryinfo_unlock( ein );
1222 * cache_delete - delete the entry e from the cache.
1224 * returns: 0 e was deleted ok
1225 * 1 e was not in the cache
1226 * -1 something bad happened
1230 struct bdb_info *bdb,
1235 EntryInfo *ei = BEI(e);
1238 assert( e->e_private != NULL );
1240 /* Set this early, warn off any queriers */
1241 ei->bei_state |= CACHE_ENTRY_DELETED;
1243 /* Lock the entry's info */
1244 bdb_cache_entryinfo_lock( ei );
1246 /* Get write lock on the data */
1247 rc = bdb_cache_entry_db_relock( bdb, locker, ei, 1, 0, lock );
1249 /* couldn't lock, undo and give up */
1250 ei->bei_state ^= CACHE_ENTRY_DELETED;
1251 bdb_cache_entryinfo_unlock( ei );
1255 Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_delete( %ld )\n",
1259 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_lru_mutex );
1261 rc = bdb_cache_delete_internal( &bdb->bi_cache, e->e_private, 1 );
1263 /* free lru mutex */
1264 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_lru_mutex );
1266 /* Leave entry info locked */
1272 bdb_cache_delete_cleanup(
1277 ei->bei_e->e_private = NULL;
1278 #ifdef SLAP_ZONE_ALLOC
1279 bdb_entry_return( ei->bei_bdb, ei->bei_e, ei->bei_zseq );
1281 bdb_entry_return( ei->bei_e );
1286 bdb_cache_entryinfo_free( cache, ei );
1287 bdb_cache_entryinfo_unlock( ei );
1291 bdb_cache_delete_internal(
1296 int rc = 0; /* return code */
1299 /* Lock the parent's kids tree */
1300 bdb_cache_entryinfo_lock( e->bei_parent );
1303 e->bei_parent->bei_ckids--;
1304 if ( decr && e->bei_parent->bei_dkids ) e->bei_parent->bei_dkids--;
1307 if ( avl_delete( &e->bei_parent->bei_kids, (caddr_t) e, bdb_rdn_cmp )
1312 if ( e->bei_parent->bei_kids )
1315 bdb_cache_entryinfo_unlock( e->bei_parent );
1317 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
1319 if ( avl_delete( &cache->c_idtree, (caddr_t) e, bdb_id_cmp )) {
1326 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
1330 LRU_DEL( cache, e );
1333 ldap_pvt_thread_mutex_lock( &cache->c_count_mutex );
1335 ldap_pvt_thread_mutex_unlock( &cache->c_count_mutex );
1343 bdb_entryinfo_release( void *data )
1345 EntryInfo *ei = (EntryInfo *)data;
1346 if ( ei->bei_kids ) {
1347 avl_free( ei->bei_kids, NULL );
1350 ei->bei_e->e_private = NULL;
1351 #ifdef SLAP_ZONE_ALLOC
1352 bdb_entry_return( ei->bei_bdb, ei->bei_e, ei->bei_zseq );
1354 bdb_entry_return( ei->bei_e );
1357 bdb_cache_entryinfo_destroy( ei );
1361 bdb_cache_release_all( Cache *cache )
1363 /* set cache write lock */
1364 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
1366 ldap_pvt_thread_mutex_lock( &cache->c_lru_mutex );
1368 Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_release_all\n", 0, 0, 0 );
1370 avl_free( cache->c_dntree.bei_kids, NULL );
1371 avl_free( cache->c_idtree, bdb_entryinfo_release );
1372 for (;cache->c_eifree;cache->c_eifree = cache->c_lruhead) {
1373 cache->c_lruhead = cache->c_eifree->bei_lrunext;
1374 bdb_cache_entryinfo_destroy(cache->c_eifree);
1376 cache->c_cursize = 0;
1377 cache->c_eiused = 0;
1378 cache->c_leaves = 0;
1379 cache->c_idtree = NULL;
1380 cache->c_lruhead = NULL;
1381 cache->c_lrutail = NULL;
1382 cache->c_dntree.bei_kids = NULL;
1384 /* free lru mutex */
1385 ldap_pvt_thread_mutex_unlock( &cache->c_lru_mutex );
1386 /* free cache write lock */
1387 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
1393 bdb_lru_print( Cache *cache )
1397 fprintf( stderr, "LRU circle head: %p\n", (void *) cache->c_lruhead );
1398 fprintf( stderr, "LRU circle (tail forward):\n" );
1399 for ( e = cache->c_lrutail; ; ) {
1400 fprintf( stderr, "\t%p, %p id %ld rdn \"%s\"\n",
1401 (void *) e, (void *) e->bei_e, e->bei_id, e->bei_nrdn.bv_val );
1403 if ( e == cache->c_lrutail )
1406 fprintf( stderr, "LRU circle (tail backward):\n" );
1407 for ( e = cache->c_lrutail; ; ) {
1408 fprintf( stderr, "\t%p, %p id %ld rdn \"%s\"\n",
1409 (void *) e, (void *) e->bei_e, e->bei_id, e->bei_nrdn.bv_val );
1411 if ( e == cache->c_lrutail )
1418 #ifdef BDB_REUSE_LOCKERS
1420 bdb_locker_id_free( void *key, void *data )
1426 #if DB_VERSION_FULL >= 0x04060012
1427 BDB_LOCKER lptr = data;
1430 lockid = (long)data;
1432 rc = XLOCK_ID_FREE( env, lockid );
1433 if ( rc == EINVAL ) {
1435 Debug( LDAP_DEBUG_ANY,
1436 "bdb_locker_id_free: %lu err %s(%d)\n",
1437 (unsigned long) lockid, db_strerror(rc), rc );
1438 /* release all locks held by this locker. */
1439 lr.op = DB_LOCK_PUT_ALL;
1441 env->lock_vec( env, lockid, 0, &lr, 1, NULL );
1442 XLOCK_ID_FREE( env, lockid );
1446 /* free up any keys used by the main thread */
1448 bdb_locker_flush( DB_ENV *env )
1451 void *ctx = ldap_pvt_thread_pool_context();
1453 if ( !ldap_pvt_thread_pool_getkey( ctx, env, &data, NULL ) ) {
1454 ldap_pvt_thread_pool_setkey( ctx, env, NULL, 0, NULL, NULL );
1455 bdb_locker_id_free( env, data );
1460 bdb_locker_id( Operation *op, DB_ENV *env, BDB_LOCKER *locker )
1467 if ( !env || !locker ) return -1;
1469 /* If no op was provided, try to find the ctx anyway... */
1471 ctx = op->o_threadctx;
1473 ctx = ldap_pvt_thread_pool_context();
1476 /* Shouldn't happen unless we're single-threaded */
1482 if ( ldap_pvt_thread_pool_getkey( ctx, env, &data, NULL ) ) {
1483 for ( i=0, rc=1; rc != 0 && i<4; i++ ) {
1484 rc = XLOCK_ID( env, &lockid );
1485 if (rc) ldap_pvt_thread_yield();
1490 #if DB_VERSION_FULL >= 0x04060012
1492 __lock_getlocker( env->lk_handle, lockid, 0, &lptr );
1496 data = (void *)((long)lockid);
1498 if ( ( rc = ldap_pvt_thread_pool_setkey( ctx, env,
1499 data, bdb_locker_id_free, NULL, NULL ) ) ) {
1500 XLOCK_ID_FREE( env, lockid );
1501 Debug( LDAP_DEBUG_ANY, "bdb_locker_id: err %s(%d)\n",
1502 db_strerror(rc), rc, 0 );
1507 lockid = (long)data;
1509 #if DB_VERSION_FULL >= 0x04060012
1516 #endif /* BDB_REUSE_LOCKERS */