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 2000-2006 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);
39 static void bdb_lru_print(Cache *cache);
43 /* For concurrency experiments only! */
45 #define ldap_pvt_thread_rdwr_wlock(a) 0
46 #define ldap_pvt_thread_rdwr_wunlock(a) 0
47 #define ldap_pvt_thread_rdwr_rlock(a) 0
48 #define ldap_pvt_thread_rdwr_runlock(a) 0
52 #define ldap_pvt_thread_mutex_trylock(a) 0
56 bdb_cache_entryinfo_new( Cache *cache )
60 if ( cache->c_eifree ) {
61 ldap_pvt_thread_mutex_lock( &cache->c_eifree_mutex );
62 if ( cache->c_eifree ) {
64 cache->c_eifree = ei->bei_lrunext;
66 ldap_pvt_thread_mutex_unlock( &cache->c_eifree_mutex );
69 ei = ch_calloc(1, sizeof(EntryInfo));
70 ldap_pvt_thread_mutex_init( &ei->bei_kids_mutex );
73 ei->bei_state = CACHE_ENTRY_REFERENCED;
78 /* Note - we now use a Second-Chance / Clock algorithm instead of
79 * Least-Recently-Used. This tremendously improves concurrency
80 * because we no longer need to manipulate the lists every time an
81 * entry is touched. We only need to lock the lists when adding
82 * or deleting an entry. It's now a circular doubly-linked list.
83 * We always append to the tail, but the head traverses the circle
84 * during a purge operation.
87 bdb_cache_lru_link( Cache *cache, EntryInfo *ei )
89 /* Insert into circular LRU list */
90 ldap_pvt_thread_mutex_lock( &cache->lru_tail_mutex );
91 ei->bei_lruprev = cache->c_lrutail;
92 if ( cache->c_lrutail ) {
93 ei->bei_lrunext = cache->c_lrutail->bei_lrunext;
94 cache->c_lrutail->bei_lrunext = ei;
95 if ( ei->bei_lrunext )
96 ei->bei_lrunext->bei_lruprev = ei;
98 ei->bei_lrunext = ei->bei_lruprev = ei;
99 cache->c_lruhead = ei;
101 cache->c_lrutail = ei;
102 ldap_pvt_thread_mutex_unlock( &cache->lru_tail_mutex );
109 /* #define NO_DB_LOCK 1 */
110 /* Note: The BerkeleyDB locks are much slower than regular
111 * mutexes or rdwr locks. But the BDB implementation has the
112 * advantage of using a fixed size lock table, instead of
113 * allocating a lock object per entry in the DB. That's a
114 * key benefit for scaling. It also frees us from worrying
115 * about undetectable deadlocks between BDB activity and our
116 * own cache activity. It's still worth exploring faster
117 * alternatives though.
120 /* Atomically release and reacquire a lock */
122 bdb_cache_entry_db_relock(
123 struct bdb_info *bdb,
137 if ( !lock ) return 0;
139 lockobj.data = &ei->bei_id;
140 lockobj.size = sizeof(ei->bei_id) + 1;
142 list[0].op = DB_LOCK_PUT;
143 list[0].lock = *lock;
144 list[1].op = DB_LOCK_GET;
145 list[1].lock = *lock;
146 list[1].mode = rw ? DB_LOCK_WRITE : DB_LOCK_READ;
147 list[1].obj = &lockobj;
148 rc = bdb->bi_dbenv->lock_vec(bdb->bi_dbenv, locker, tryOnly ? DB_LOCK_NOWAIT : 0,
151 if (rc && !tryOnly) {
152 Debug( LDAP_DEBUG_TRACE,
153 "bdb_cache_entry_db_relock: entry %ld, rw %d, rc %d\n",
154 ei->bei_id, rw, rc );
156 *lock = list[1].lock;
163 bdb_cache_entry_db_lock( struct bdb_info *bdb, u_int32_t locker, EntryInfo *ei,
164 int rw, int tryOnly, DB_LOCK *lock )
173 if ( !lock ) return 0;
176 db_rw = DB_LOCK_WRITE;
178 db_rw = DB_LOCK_READ;
180 lockobj.data = &ei->bei_id;
181 lockobj.size = sizeof(ei->bei_id) + 1;
183 rc = LOCK_GET(bdb->bi_dbenv, locker, tryOnly ? DB_LOCK_NOWAIT : 0,
184 &lockobj, db_rw, lock);
185 if (rc && !tryOnly) {
186 Debug( LDAP_DEBUG_TRACE,
187 "bdb_cache_entry_db_lock: entry %ld, rw %d, rc %d\n",
188 ei->bei_id, rw, rc );
191 #endif /* NO_DB_LOCK */
195 bdb_cache_entry_db_unlock ( struct bdb_info *bdb, DB_LOCK *lock )
202 if ( !lock || lock->mode == DB_LOCK_NG ) return 0;
204 rc = LOCK_PUT ( bdb->bi_dbenv, lock );
210 bdb_cache_entryinfo_destroy( EntryInfo *e )
212 ldap_pvt_thread_mutex_destroy( &e->bei_kids_mutex );
213 free( e->bei_nrdn.bv_val );
215 free( e->bei_rdn.bv_val );
221 /* Do a length-ordered sort on normalized RDNs */
223 bdb_rdn_cmp( const void *v_e1, const void *v_e2 )
225 const EntryInfo *e1 = v_e1, *e2 = v_e2;
226 int rc = e1->bei_nrdn.bv_len - e2->bei_nrdn.bv_len;
228 rc = strncmp( e1->bei_nrdn.bv_val, e2->bei_nrdn.bv_val,
229 e1->bei_nrdn.bv_len );
235 bdb_id_cmp( const void *v_e1, const void *v_e2 )
237 const EntryInfo *e1 = v_e1, *e2 = v_e2;
238 return e1->bei_id - e2->bei_id;
241 /* Create an entryinfo in the cache. Caller must release the locks later.
244 bdb_entryinfo_add_internal(
245 struct bdb_info *bdb,
249 EntryInfo *ei2 = NULL;
253 ei2 = bdb_cache_entryinfo_new( &bdb->bi_cache );
255 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
256 bdb_cache_entryinfo_lock( ei->bei_parent );
258 ei2->bei_id = ei->bei_id;
259 ei2->bei_parent = ei->bei_parent;
261 ei2->bei_rdn = ei->bei_rdn;
263 #ifdef SLAP_ZONE_ALLOC
267 /* Add to cache ID tree */
268 if (avl_insert( &bdb->bi_cache.c_idtree, ei2, bdb_id_cmp, avl_dup_error )) {
270 eix = avl_find( bdb->bi_cache.c_idtree, ei2, bdb_id_cmp );
271 bdb_cache_entryinfo_destroy( ei2 );
274 /* It got freed above because its value was
277 ei->bei_rdn.bv_val = NULL;
280 bdb->bi_cache.c_eiused++;
281 ber_dupbv( &ei2->bei_nrdn, &ei->bei_nrdn );
283 /* This is a new leaf node. But if parent had no kids, then it was
284 * a leaf and we would be decrementing that. So, only increment if
285 * the parent already has kids.
287 if ( ei->bei_parent->bei_kids || !ei->bei_parent->bei_id )
288 bdb->bi_cache.c_leaves++;
289 avl_insert( &ei->bei_parent->bei_kids, ei2, bdb_rdn_cmp,
292 ei->bei_parent->bei_ckids++;
295 bdb_cache_lru_link( &bdb->bi_cache, ei2 );
301 /* Find the EntryInfo for the requested DN. If the DN cannot be found, return
302 * the info for its closest ancestor. *res should be NULL to process a
303 * complete DN starting from the tree root. Otherwise *res must be the
304 * immediate parent of the requested DN, and only the RDN will be searched.
305 * The EntryInfo is locked upon return and must be unlocked by the caller.
314 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
315 EntryInfo ei, *eip, *ei2;
319 /* this function is always called with normalized DN */
321 /* we're doing a onelevel search for an RDN */
322 ei.bei_nrdn.bv_val = ndn->bv_val;
323 ei.bei_nrdn.bv_len = dn_rdnlen( op->o_bd, ndn );
326 /* we're searching a full DN from the root */
327 ptr = ndn->bv_val + ndn->bv_len - op->o_bd->be_nsuffix[0].bv_len;
328 ei.bei_nrdn.bv_val = ptr;
329 ei.bei_nrdn.bv_len = op->o_bd->be_nsuffix[0].bv_len;
330 /* Skip to next rdn if suffix is empty */
331 if ( ei.bei_nrdn.bv_len == 0 ) {
332 for (ptr = ei.bei_nrdn.bv_val - 2; ptr > ndn->bv_val
333 && !DN_SEPARATOR(*ptr); ptr--) /* empty */;
334 if ( ptr >= ndn->bv_val ) {
335 if (DN_SEPARATOR(*ptr)) ptr++;
336 ei.bei_nrdn.bv_len = ei.bei_nrdn.bv_val - ptr;
337 ei.bei_nrdn.bv_val = ptr;
340 eip = &bdb->bi_cache.c_dntree;
343 for ( bdb_cache_entryinfo_lock( eip ); eip; ) {
344 eip->bei_state |= CACHE_ENTRY_REFERENCED;
346 ei2 = (EntryInfo *)avl_find( eip->bei_kids, &ei, bdb_rdn_cmp );
348 int len = ei.bei_nrdn.bv_len;
350 if ( BER_BVISEMPTY( ndn )) {
355 ei.bei_nrdn.bv_len = ndn->bv_len -
356 (ei.bei_nrdn.bv_val - ndn->bv_val);
357 bdb_cache_entryinfo_unlock( eip );
359 rc = bdb_dn2id( op, txn, &ei.bei_nrdn, &ei );
361 bdb_cache_entryinfo_lock( eip );
366 /* DN exists but needs to be added to cache */
367 ei.bei_nrdn.bv_len = len;
368 rc = bdb_entryinfo_add_internal( bdb, &ei, &ei2 );
369 /* add_internal left eip and c_rwlock locked */
370 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
375 } else if ( ei2->bei_state & CACHE_ENTRY_DELETED ) {
376 /* In the midst of deleting? Give it a chance to
379 bdb_cache_entryinfo_unlock( eip );
380 ldap_pvt_thread_yield();
381 bdb_cache_entryinfo_lock( eip );
385 bdb_cache_entryinfo_unlock( eip );
386 bdb_cache_entryinfo_lock( ei2 );
390 /* Advance to next lower RDN */
391 for (ptr = ei.bei_nrdn.bv_val - 2; ptr > ndn->bv_val
392 && !DN_SEPARATOR(*ptr); ptr--) /* empty */;
393 if ( ptr >= ndn->bv_val ) {
394 if (DN_SEPARATOR(*ptr)) ptr++;
395 ei.bei_nrdn.bv_len = ei.bei_nrdn.bv_val - ptr - 1;
396 ei.bei_nrdn.bv_val = ptr;
398 if ( ptr < ndn->bv_val ) {
408 /* Walk up the tree from a child node, looking for an ID that's already
409 * been linked into the cache.
412 hdb_cache_find_parent(
419 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
420 EntryInfo ei, eip, *ei2 = NULL, *ein = NULL, *eir = NULL;
428 rc = hdb_dn2id_parent( op, txn, locker, &ei, &eip.bei_id );
431 /* Save the previous node, if any */
434 /* Create a new node for the current ID */
435 ein = bdb_cache_entryinfo_new( &bdb->bi_cache );
436 ein->bei_id = ei.bei_id;
437 ein->bei_kids = ei.bei_kids;
438 ein->bei_nrdn = ei.bei_nrdn;
439 ein->bei_rdn = ei.bei_rdn;
440 ein->bei_ckids = ei.bei_ckids;
441 #ifdef SLAP_ZONE_ALLOC
446 /* This node is not fully connected yet */
447 ein->bei_state |= CACHE_ENTRY_NOT_LINKED;
449 /* Insert this node into the ID tree */
450 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
451 if ( avl_insert( &bdb->bi_cache.c_idtree, (caddr_t)ein,
452 bdb_id_cmp, avl_dup_error ) ) {
454 /* Someone else created this node just before us.
455 * Free our new copy and use the existing one.
457 bdb_cache_entryinfo_destroy( ein );
458 ein = (EntryInfo *)avl_find( bdb->bi_cache.c_idtree,
459 (caddr_t) &ei, bdb_id_cmp );
461 /* Link in any kids we've already processed */
463 bdb_cache_entryinfo_lock( ein );
464 avl_insert( &ein->bei_kids, (caddr_t)ei2,
465 bdb_rdn_cmp, avl_dup_error );
467 bdb_cache_entryinfo_unlock( ein );
470 bdb_cache_lru_link( &bdb->bi_cache, ein );
473 /* If this is the first time, save this node
474 * to be returned later.
476 if ( eir == NULL ) eir = ein;
478 /* If there was a previous node, link it to this one */
479 if ( ei2 ) ei2->bei_parent = ein;
481 /* Look for this node's parent */
483 ei2 = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
484 (caddr_t) &eip, bdb_id_cmp );
486 ei2 = &bdb->bi_cache.c_dntree;
488 bdb->bi_cache.c_eiused++;
489 if ( ei2 && ( ei2->bei_kids || !ei2->bei_id ))
490 bdb->bi_cache.c_leaves++;
491 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
493 /* Got the parent, link in and we're done. */
495 bdb_cache_entryinfo_lock( ei2 );
496 ein->bei_parent = ei2;
498 avl_insert( &ei2->bei_kids, (caddr_t)ein, bdb_rdn_cmp,
502 /* Reset all the state info */
503 for (ein = eir; ein != ei2; ein=ein->bei_parent)
504 ein->bei_state &= ~CACHE_ENTRY_NOT_LINKED;
506 bdb_cache_entryinfo_unlock( ei2 );
507 bdb_cache_entryinfo_lock( eir );
513 ei.bei_id = eip.bei_id;
515 avl_insert( &ei.bei_kids, (caddr_t)ein, bdb_rdn_cmp,
521 /* Used by hdb_dn2idl when loading the EntryInfo for all the children
525 struct bdb_info *bdb,
532 /* See if we already have this one */
533 bdb_cache_entryinfo_lock( ei->bei_parent );
534 ei2 = (EntryInfo *)avl_find( ei->bei_parent->bei_kids, ei, bdb_rdn_cmp );
535 bdb_cache_entryinfo_unlock( ei->bei_parent );
538 /* Not found, add it */
541 /* bei_rdn was not malloc'd before, do it now */
542 ber_dupbv( &bv, &ei->bei_rdn );
545 rc = bdb_entryinfo_add_internal( bdb, ei, res );
546 bdb_cache_entryinfo_unlock( ei->bei_parent );
547 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
549 /* Found, return it */
558 bdb_cache_lru_purge( struct bdb_info *bdb )
560 DB_LOCK lock, *lockp;
561 EntryInfo *elru, *elnext;
564 /* Don't bother if we can't get the lock */
565 if ( ldap_pvt_thread_mutex_trylock( &bdb->bi_cache.lru_head_mutex ) )
568 if ( bdb->bi_cache.c_cursize <= bdb->bi_cache.c_maxsize ) {
569 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_head_mutex );
573 if ( bdb->bi_cache.c_locker ) {
580 /* Look for an unused entry to remove */
581 for (elru = bdb->bi_cache.c_lruhead; elru; elru = elnext ) {
582 elnext = elru->bei_lrunext;
584 if ( ldap_pvt_thread_mutex_trylock( &elru->bei_kids_mutex ))
587 /* This flag implements the clock replacement behavior */
588 if ( elru->bei_state & ( CACHE_ENTRY_REFERENCED )) {
589 elru->bei_state &= ~CACHE_ENTRY_REFERENCED;
590 bdb_cache_entryinfo_unlock( elru );
594 /* If this node is in the process of linking into the cache,
595 * or this node is being deleted, skip it.
597 if ( elru->bei_state & ( CACHE_ENTRY_NOT_LINKED |
598 CACHE_ENTRY_DELETED | CACHE_ENTRY_LOADING )) {
599 bdb_cache_entryinfo_unlock( elru );
603 /* entryinfo is locked */
606 /* If we can successfully writelock it, then
607 * the object is idle.
609 if ( bdb_cache_entry_db_lock( bdb,
610 bdb->bi_cache.c_locker, elru, 1, 1, lockp ) == 0 ) {
612 /* Free entry for this node if it's present */
614 elru->bei_e->e_private = NULL;
615 #ifdef SLAP_ZONE_ALLOC
616 bdb_entry_return( bdb, elru->bei_e, elru->bei_zseq );
618 bdb_entry_return( elru->bei_e );
623 bdb_cache_entry_db_unlock( bdb, lockp );
625 /* ITS#4010 if we're in slapcat, and this node is a leaf
628 * FIXME: we need to do this for slapd as well, (which is
629 * why we compute bi_cache.c_leaves now) but at the moment
630 * we can't because it causes unresolvable deadlocks.
632 if ( slapMode & SLAP_TOOL_READONLY ) {
633 if ( !elru->bei_kids ) {
634 bdb_cache_delete_internal( &bdb->bi_cache, elru, 0 );
635 bdb_cache_delete_cleanup( &bdb->bi_cache, elru );
638 /* Leave node on LRU list for a future pass */
643 bdb_cache_entryinfo_unlock( elru );
645 if ( count >= bdb->bi_cache.c_minfree ) {
646 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_count_mutex );
647 bdb->bi_cache.c_cursize -= count;
648 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_count_mutex );
653 bdb->bi_cache.c_lruhead = elru;
654 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_head_mutex );
659 struct bdb_info *bdb,
662 EntryInfo ei = { 0 },
667 ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
668 ei2 = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
669 (caddr_t) &ei, bdb_id_cmp );
670 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
675 * cache_find_id - find an entry in the cache, given id.
676 * The entry is locked for Read upon return. Call with islocked TRUE if
677 * the supplied *eip was already locked.
690 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
692 int rc = 0, load = 0;
693 EntryInfo ei = { 0 };
697 #ifdef SLAP_ZONE_ALLOC
698 slap_zh_rlock(bdb->bi_cache.c_zctx);
700 /* If we weren't given any info, see if we have it already cached */
702 again: ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
703 *eip = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
704 (caddr_t) &ei, bdb_id_cmp );
706 /* If the lock attempt fails, the info is in use */
707 if ( ldap_pvt_thread_mutex_trylock(
708 &(*eip)->bei_kids_mutex )) {
709 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
710 /* If this node is being deleted, treat
711 * as if the delete has already finished
713 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
716 /* otherwise, wait for the info to free up */
717 ldap_pvt_thread_yield();
720 /* If this info isn't hooked up to its parent yet,
721 * unlock and wait for it to be fully initialized
723 if ( (*eip)->bei_state & CACHE_ENTRY_NOT_LINKED ) {
724 bdb_cache_entryinfo_unlock( *eip );
725 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
726 ldap_pvt_thread_yield();
731 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
734 /* See if the ID exists in the database; add it to the cache if so */
737 rc = bdb_id2entry( op->o_bd, tid, locker, id, &ep );
739 rc = bdb_cache_find_ndn( op, tid,
741 if ( *eip ) islocked = 1;
743 ep->e_private = NULL;
744 #ifdef SLAP_ZONE_ALLOC
745 bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
747 bdb_entry_return( ep );
753 rc = hdb_cache_find_parent(op, tid, locker, id, eip );
754 if ( rc == 0 ) islocked = 1;
758 /* Ok, we found the info, do we have the entry? */
760 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
763 /* Make sure only one thread tries to load the entry */
765 #ifdef SLAP_ZONE_ALLOC
766 if ((*eip)->bei_e && !slap_zn_validate(
767 bdb->bi_cache.c_zctx, (*eip)->bei_e, (*eip)->bei_zseq)) {
768 (*eip)->bei_e = NULL;
769 (*eip)->bei_zseq = 0;
772 if ( !(*eip)->bei_e && !((*eip)->bei_state & CACHE_ENTRY_LOADING)) {
774 (*eip)->bei_state |= CACHE_ENTRY_LOADING;
777 bdb_cache_entryinfo_unlock( *eip );
780 rc = bdb_cache_entry_db_lock( bdb, locker, *eip, 0, 0, lock );
781 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
783 bdb_cache_entry_db_unlock( bdb, lock );
784 } else if ( rc == 0 ) {
786 /* Give up original read lock, obtain write lock
789 rc = bdb_cache_entry_db_relock( bdb, locker,
792 if ( rc == 0 && !ep) {
793 rc = bdb_id2entry( op->o_bd, tid, locker, id, &ep );
796 ep->e_private = *eip;
801 #ifdef SLAP_ZONE_ALLOC
802 (*eip)->bei_zseq = *((ber_len_t *)ep - 2);
806 bdb_cache_entryinfo_lock( *eip );
807 (*eip)->bei_state ^= CACHE_ENTRY_LOADING;
808 bdb_cache_entryinfo_unlock( *eip );
810 /* If we succeeded, downgrade back to a readlock. */
811 rc = bdb_cache_entry_db_relock( bdb, locker,
814 /* Otherwise, release the lock. */
815 bdb_cache_entry_db_unlock( bdb, lock );
817 } else if ( !(*eip)->bei_e ) {
818 /* Some other thread is trying to load the entry,
819 * wait for it to finish.
821 bdb_cache_entry_db_unlock( bdb, lock );
822 bdb_cache_entryinfo_lock( *eip );
827 /* Check for subtree renames
829 rc = bdb_fix_dn( (*eip)->bei_e, 1 );
831 bdb_cache_entry_db_relock( bdb,
832 locker, *eip, 1, 0, lock );
833 /* check again in case other modifier did it already */
834 if ( bdb_fix_dn( (*eip)->bei_e, 1 ) )
835 rc = bdb_fix_dn( (*eip)->bei_e, 2 );
836 bdb_cache_entry_db_relock( bdb,
837 locker, *eip, 0, 0, lock );
846 bdb_cache_entryinfo_unlock( *eip );
849 ep->e_private = NULL;
850 #ifdef SLAP_ZONE_ALLOC
851 bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
853 bdb_entry_return( ep );
860 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_count_mutex );
861 bdb->bi_cache.c_cursize++;
862 if ( bdb->bi_cache.c_cursize > bdb->bi_cache.c_maxsize )
864 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_count_mutex );
867 bdb_cache_lru_purge( bdb );
870 #ifdef SLAP_ZONE_ALLOC
871 if (rc == 0 && (*eip)->bei_e) {
872 slap_zn_rlock(bdb->bi_cache.c_zctx, (*eip)->bei_e);
874 slap_zh_runlock(bdb->bi_cache.c_zctx);
887 if ( BEI(e)->bei_kids ) {
890 if ( BEI(e)->bei_state & CACHE_ENTRY_NO_KIDS ) {
893 rc = bdb_dn2id_children( op, txn, e );
894 if ( rc == DB_NOTFOUND ) {
895 BEI(e)->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
900 /* Update the cache after a successful database Add. */
903 struct bdb_info *bdb,
913 struct berval rdn = e->e_name;
921 /* Lock this entry so that bdb_add can run to completion.
922 * It can only fail if BDB has run out of lock resources.
924 rc = bdb_cache_entry_db_lock( bdb, locker, &ei, 0, 0, lock );
926 bdb_cache_entryinfo_unlock( eip );
931 if ( nrdn->bv_len != e->e_nname.bv_len ) {
932 char *ptr = ber_bvchr( &rdn, ',' );
933 assert( ptr != NULL );
934 rdn.bv_len = ptr - rdn.bv_val;
936 ber_dupbv( &ei.bei_rdn, &rdn );
937 if ( eip->bei_dkids ) eip->bei_dkids++;
940 rc = bdb_entryinfo_add_internal( bdb, &ei, &new );
941 /* bdb_csn_commit can cause this when adding the database root entry */
943 new->bei_e->e_private = NULL;
944 #ifdef SLAP_ZONE_ALLOC
945 bdb_entry_return( bdb, new->bei_e, new->bei_zseq );
947 bdb_entry_return( new->bei_e );
952 new->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
953 eip->bei_state &= ~CACHE_ENTRY_NO_KIDS;
954 if (eip->bei_parent) {
955 eip->bei_parent->bei_state &= ~CACHE_ENTRY_NO_GRANDKIDS;
957 bdb_cache_entryinfo_unlock( eip );
959 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
960 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_count_mutex );
961 ++bdb->bi_cache.c_cursize;
962 if ( bdb->bi_cache.c_cursize > bdb->bi_cache.c_maxsize )
964 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_count_mutex );
967 bdb_cache_lru_purge( bdb );
974 struct bdb_info *bdb,
980 EntryInfo *ei = BEI(e);
982 /* Get write lock on data */
983 rc = bdb_cache_entry_db_relock( bdb, locker, ei, 1, 0, lock );
985 /* If we've done repeated mods on a cached entry, then e_attrs
986 * is no longer contiguous with the entry, and must be freed.
989 if ( (void *)e->e_attrs != (void *)(e+1) ) {
990 attrs_free( e->e_attrs );
992 e->e_attrs = newAttrs;
998 * Change the rdn in the entryinfo. Also move to a new parent if needed.
1002 struct bdb_info *bdb,
1004 struct berval *nrdn,
1010 EntryInfo *ei = BEI(e), *pei;
1016 /* Get write lock on data */
1017 rc = bdb_cache_entry_db_relock( bdb, locker, ei, 1, 0, lock );
1018 if ( rc ) return rc;
1020 /* If we've done repeated mods on a cached entry, then e_attrs
1021 * is no longer contiguous with the entry, and must be freed.
1023 if ( (void *)e->e_attrs != (void *)(e+1) ) {
1024 attrs_free( e->e_attrs );
1026 e->e_attrs = new->e_attrs;
1027 if( e->e_nname.bv_val < e->e_bv.bv_val ||
1028 e->e_nname.bv_val > e->e_bv.bv_val + e->e_bv.bv_len )
1030 ch_free(e->e_name.bv_val);
1031 ch_free(e->e_nname.bv_val);
1033 e->e_name = new->e_name;
1034 e->e_nname = new->e_nname;
1036 /* Lock the parent's kids AVL tree */
1037 pei = ei->bei_parent;
1038 bdb_cache_entryinfo_lock( pei );
1039 avl_delete( &pei->bei_kids, (caddr_t) ei, bdb_rdn_cmp );
1040 free( ei->bei_nrdn.bv_val );
1041 ber_dupbv( &ei->bei_nrdn, nrdn );
1043 if ( !pei->bei_kids )
1044 pei->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
1047 free( ei->bei_rdn.bv_val );
1050 if ( nrdn->bv_len != e->e_nname.bv_len ) {
1051 char *ptr = ber_bvchr(&rdn, ',');
1052 assert( ptr != NULL );
1053 rdn.bv_len = ptr - rdn.bv_val;
1055 ber_dupbv( &ei->bei_rdn, &rdn );
1057 if ( pei->bei_dkids ) pei->bei_dkids--;
1061 ein = ei->bei_parent;
1063 ei->bei_parent = ein;
1064 bdb_cache_entryinfo_unlock( pei );
1065 bdb_cache_entryinfo_lock( ein );
1067 /* parent now has kids */
1068 if ( ein->bei_state & CACHE_ENTRY_NO_KIDS )
1069 ein->bei_state ^= CACHE_ENTRY_NO_KIDS;
1072 /* parent might now have grandkids */
1073 if ( ein->bei_state & CACHE_ENTRY_NO_GRANDKIDS &&
1074 !(ei->bei_state & (CACHE_ENTRY_NO_KIDS)))
1075 ein->bei_state ^= CACHE_ENTRY_NO_GRANDKIDS;
1078 /* Record the generation number of this change */
1079 ldap_pvt_thread_mutex_lock( &bdb->bi_modrdns_mutex );
1081 ei->bei_modrdns = bdb->bi_modrdns;
1082 ldap_pvt_thread_mutex_unlock( &bdb->bi_modrdns_mutex );
1085 if ( ein->bei_dkids ) ein->bei_dkids++;
1087 avl_insert( &ein->bei_kids, ei, bdb_rdn_cmp, avl_dup_error );
1088 bdb_cache_entryinfo_unlock( ein );
1092 * cache_delete - delete the entry e from the cache.
1094 * returns: 0 e was deleted ok
1095 * 1 e was not in the cache
1096 * -1 something bad happened
1100 struct bdb_info *bdb,
1105 EntryInfo *ei = BEI(e);
1108 assert( e->e_private != NULL );
1110 /* Set this early, warn off any queriers */
1111 ei->bei_state |= CACHE_ENTRY_DELETED;
1113 /* Lock the entry's info */
1114 bdb_cache_entryinfo_lock( ei );
1116 /* Get write lock on the data */
1117 rc = bdb_cache_entry_db_relock( bdb, locker, ei, 1, 0, lock );
1119 /* couldn't lock, undo and give up */
1120 ei->bei_state ^= CACHE_ENTRY_DELETED;
1121 bdb_cache_entryinfo_unlock( ei );
1125 Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_delete( %ld )\n",
1129 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_head_mutex );
1131 rc = bdb_cache_delete_internal( &bdb->bi_cache, e->e_private, 1 );
1133 /* free lru mutex */
1134 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_head_mutex );
1136 /* Leave entry info locked */
1142 bdb_cache_delete_cleanup(
1147 ei->bei_e->e_private = NULL;
1148 #ifdef SLAP_ZONE_ALLOC
1149 bdb_entry_return( ei->bei_bdb, ei->bei_e, ei->bei_zseq );
1151 bdb_entry_return( ei->bei_e );
1156 free( ei->bei_nrdn.bv_val );
1157 ei->bei_nrdn.bv_val = NULL;
1159 free( ei->bei_rdn.bv_val );
1160 ei->bei_rdn.bv_val = NULL;
1161 ei->bei_modrdns = 0;
1165 ei->bei_parent = NULL;
1166 ei->bei_kids = NULL;
1167 ei->bei_lruprev = NULL;
1169 ldap_pvt_thread_mutex_lock( &cache->c_eifree_mutex );
1170 ei->bei_lrunext = cache->c_eifree;
1171 cache->c_eifree = ei;
1172 ldap_pvt_thread_mutex_unlock( &cache->c_eifree_mutex );
1173 bdb_cache_entryinfo_unlock( ei );
1177 bdb_cache_delete_internal(
1182 int rc = 0; /* return code */
1185 /* Lock the parent's kids tree */
1186 bdb_cache_entryinfo_lock( e->bei_parent );
1189 e->bei_parent->bei_ckids--;
1190 if ( decr && e->bei_parent->bei_dkids ) e->bei_parent->bei_dkids--;
1193 if ( avl_delete( &e->bei_parent->bei_kids, (caddr_t) e, bdb_rdn_cmp )
1198 if ( e->bei_parent->bei_kids )
1201 bdb_cache_entryinfo_unlock( e->bei_parent );
1203 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
1205 if ( avl_delete( &cache->c_idtree, (caddr_t) e, bdb_id_cmp )) {
1212 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
1216 if ( e == cache->c_lruhead ) cache->c_lruhead = e->bei_lrunext;
1217 if ( e == cache->c_lrutail ) {
1218 ldap_pvt_thread_mutex_lock( &cache->lru_tail_mutex );
1219 if ( e == cache->c_lrutail ) cache->c_lrutail = e->bei_lruprev;
1220 ldap_pvt_thread_mutex_unlock( &cache->lru_tail_mutex );
1223 if ( e->bei_lrunext ) e->bei_lrunext->bei_lruprev = e->bei_lruprev;
1224 if ( e->bei_lruprev ) e->bei_lruprev->bei_lrunext = e->bei_lrunext;
1227 ldap_pvt_thread_mutex_lock( &cache->c_count_mutex );
1229 ldap_pvt_thread_mutex_unlock( &cache->c_count_mutex );
1237 bdb_entryinfo_release( void *data )
1239 EntryInfo *ei = (EntryInfo *)data;
1240 if ( ei->bei_kids ) {
1241 avl_free( ei->bei_kids, NULL );
1244 ei->bei_e->e_private = NULL;
1245 #ifdef SLAP_ZONE_ALLOC
1246 bdb_entry_return( ei->bei_bdb, ei->bei_e, ei->bei_zseq );
1248 bdb_entry_return( ei->bei_e );
1251 bdb_cache_entryinfo_destroy( ei );
1255 bdb_cache_release_all( Cache *cache )
1257 /* set cache write lock */
1258 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
1260 ldap_pvt_thread_mutex_lock( &cache->lru_tail_mutex );
1262 Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_release_all\n", 0, 0, 0 );
1264 avl_free( cache->c_dntree.bei_kids, NULL );
1265 avl_free( cache->c_idtree, bdb_entryinfo_release );
1266 for (;cache->c_eifree;cache->c_eifree = cache->c_lruhead) {
1267 cache->c_lruhead = cache->c_eifree->bei_lrunext;
1268 bdb_cache_entryinfo_destroy(cache->c_eifree);
1270 cache->c_cursize = 0;
1271 cache->c_eiused = 0;
1272 cache->c_leaves = 0;
1273 cache->c_idtree = NULL;
1274 cache->c_lruhead = NULL;
1275 cache->c_lrutail = NULL;
1276 cache->c_dntree.bei_kids = NULL;
1278 /* free lru mutex */
1279 ldap_pvt_thread_mutex_unlock( &cache->lru_tail_mutex );
1280 /* free cache write lock */
1281 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
1287 bdb_lru_print( Cache *cache )
1291 fprintf( stderr, "LRU queue (head to tail):\n" );
1292 for ( e = cache->c_lruhead; e != NULL; e = e->bei_lrunext ) {
1293 fprintf( stderr, "\trdn \"%20s\" id %ld\n",
1294 e->bei_nrdn.bv_val, e->bei_id );
1296 fprintf( stderr, "LRU queue (tail to head):\n" );
1297 for ( e = cache->c_lrutail; e != NULL; e = e->bei_lruprev ) {
1298 fprintf( stderr, "\trdn \"%20s\" id %ld\n",
1299 e->bei_nrdn.bv_val, e->bei_id );
1305 #ifdef BDB_REUSE_LOCKERS
1307 bdb_locker_id_free( void *key, void *data )
1310 u_int32_t lockid = (long)data;
1313 rc = XLOCK_ID_FREE( env, lockid );
1314 if ( rc == EINVAL ) {
1316 Debug( LDAP_DEBUG_ANY,
1317 "bdb_locker_id_free: %lu err %s(%d)\n",
1318 (unsigned long) lockid, db_strerror(rc), rc );
1319 /* release all locks held by this locker. */
1320 lr.op = DB_LOCK_PUT_ALL;
1322 env->lock_vec( env, lockid, 0, &lr, 1, NULL );
1323 XLOCK_ID_FREE( env, lockid );
1328 bdb_locker_id( Operation *op, DB_ENV *env, u_int32_t *locker )
1335 if ( !env || !locker ) return -1;
1337 /* If no op was provided, try to find the ctx anyway... */
1339 ctx = op->o_threadctx;
1341 ctx = ldap_pvt_thread_pool_context();
1344 /* Shouldn't happen unless we're single-threaded */
1350 if ( ldap_pvt_thread_pool_getkey( ctx, env, &data, NULL ) ) {
1351 for ( i=0, rc=1; rc != 0 && i<4; i++ ) {
1352 rc = XLOCK_ID( env, &lockid );
1353 if (rc) ldap_pvt_thread_yield();
1358 data = (void *)((long)lockid);
1359 if ( ( rc = ldap_pvt_thread_pool_setkey( ctx, env,
1360 data, bdb_locker_id_free ) ) ) {
1361 XLOCK_ID_FREE( env, lockid );
1362 Debug( LDAP_DEBUG_ANY, "bdb_locker_id: err %s(%d)\n",
1363 db_strerror(rc), rc, 0 );
1368 lockid = (long)data;
1373 #endif /* BDB_REUSE_LOCKERS */