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 2000-2007 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_add hdb_cache_lru_add
34 static void bdb_cache_lru_add( struct bdb_info *bdb, EntryInfo *ei );
36 static int bdb_cache_delete_internal(Cache *cache, EntryInfo *e, int decr);
39 static void bdb_lru_print(Cache *cache);
44 bdb_cache_entryinfo_new( Cache *cache )
48 if ( cache->c_eifree ) {
49 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
50 if ( cache->c_eifree ) {
52 cache->c_eifree = ei->bei_lrunext;
54 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
57 ei->bei_lrunext = NULL;
60 ei = ch_calloc(1, sizeof(struct bdb_entry_info));
61 ldap_pvt_thread_mutex_init( &ei->bei_kids_mutex );
67 /* Atomically release and reacquire a lock */
69 bdb_cache_entry_db_relock(
84 if ( !lock ) return 0;
86 lockobj.data = &ei->bei_id;
87 lockobj.size = sizeof(ei->bei_id) + 1;
89 list[0].op = DB_LOCK_PUT;
91 list[1].op = DB_LOCK_GET;
93 list[1].mode = rw ? DB_LOCK_WRITE : DB_LOCK_READ;
94 list[1].obj = &lockobj;
95 rc = env->lock_vec(env, locker, tryOnly ? DB_LOCK_NOWAIT : 0,
99 Debug( LDAP_DEBUG_TRACE,
100 "bdb_cache_entry_db_relock: entry %ld, rw %d, rc %d\n",
101 ei->bei_id, rw, rc );
103 *lock = list[1].lock;
110 bdb_cache_entry_db_lock( DB_ENV *env, u_int32_t locker, EntryInfo *ei,
111 int rw, int tryOnly, DB_LOCK *lock )
120 if ( !lock ) return 0;
123 db_rw = DB_LOCK_WRITE;
125 db_rw = DB_LOCK_READ;
127 lockobj.data = &ei->bei_id;
128 lockobj.size = sizeof(ei->bei_id) + 1;
130 rc = LOCK_GET(env, locker, tryOnly ? DB_LOCK_NOWAIT : 0,
131 &lockobj, db_rw, lock);
132 if (rc && !tryOnly) {
133 Debug( LDAP_DEBUG_TRACE,
134 "bdb_cache_entry_db_lock: entry %ld, rw %d, rc %d\n",
135 ei->bei_id, rw, rc );
138 #endif /* NO_THREADS */
142 bdb_cache_entry_db_unlock ( DB_ENV *env, DB_LOCK *lock )
149 if ( !lock || lock->mode == DB_LOCK_NG ) return 0;
151 rc = LOCK_PUT ( env, lock );
157 bdb_cache_entryinfo_destroy( EntryInfo *e )
159 ldap_pvt_thread_mutex_destroy( &e->bei_kids_mutex );
160 free( e->bei_nrdn.bv_val );
162 free( e->bei_rdn.bv_val );
168 #define LRU_DELETE( cache, ei ) do { \
169 if ( (ei)->bei_lruprev != NULL ) { \
170 (ei)->bei_lruprev->bei_lrunext = (ei)->bei_lrunext; \
172 (cache)->c_lruhead = (ei)->bei_lrunext; \
174 if ( (ei)->bei_lrunext != NULL ) { \
175 (ei)->bei_lrunext->bei_lruprev = (ei)->bei_lruprev; \
177 (cache)->c_lrutail = (ei)->bei_lruprev; \
179 (ei)->bei_lrunext = (ei)->bei_lruprev = NULL; \
182 #define LRU_ADD( cache, ei ) do { \
183 (ei)->bei_lrunext = (cache)->c_lruhead; \
184 if ( (ei)->bei_lrunext != NULL ) { \
185 (ei)->bei_lrunext->bei_lruprev = (ei); \
187 (cache)->c_lruhead = (ei); \
188 (ei)->bei_lruprev = NULL; \
189 if ( !ldap_pvt_thread_mutex_trylock( &(cache)->lru_tail_mutex )) { \
190 if ( (cache)->c_lrutail == NULL ) \
191 (cache)->c_lrutail = (ei); \
192 ldap_pvt_thread_mutex_unlock( &(cache)->lru_tail_mutex ); \
196 /* Do a length-ordered sort on normalized RDNs */
198 bdb_rdn_cmp( const void *v_e1, const void *v_e2 )
200 const EntryInfo *e1 = v_e1, *e2 = v_e2;
201 int rc = e1->bei_nrdn.bv_len - e2->bei_nrdn.bv_len;
203 rc = strncmp( e1->bei_nrdn.bv_val, e2->bei_nrdn.bv_val,
204 e1->bei_nrdn.bv_len );
210 bdb_id_cmp( const void *v_e1, const void *v_e2 )
212 const EntryInfo *e1 = v_e1, *e2 = v_e2;
213 return e1->bei_id - e2->bei_id;
216 /* Create an entryinfo in the cache. Caller must release the locks later.
219 bdb_entryinfo_add_internal(
220 struct bdb_info *bdb,
224 EntryInfo *ei2 = NULL;
228 ei2 = bdb_cache_entryinfo_new( &bdb->bi_cache );
230 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
231 bdb_cache_entryinfo_lock( ei->bei_parent );
233 ei2->bei_id = ei->bei_id;
234 ei2->bei_parent = ei->bei_parent;
236 ei2->bei_rdn = ei->bei_rdn;
238 #ifdef SLAP_ZONE_ALLOC
242 /* Add to cache ID tree */
243 if (avl_insert( &bdb->bi_cache.c_idtree, ei2, bdb_id_cmp, avl_dup_error )) {
245 eix = avl_find( bdb->bi_cache.c_idtree, ei2, bdb_id_cmp );
246 bdb_cache_entryinfo_destroy( ei2 );
249 /* It got freed above because its value was
252 ei->bei_rdn.bv_val = NULL;
257 bdb->bi_cache.c_eiused++;
258 ber_dupbv( &ei2->bei_nrdn, &ei->bei_nrdn );
260 /* This is a new leaf node. But if parent had no kids, then it was
261 * a leaf and we would be decrementing that. So, only increment if
262 * the parent already has kids.
264 if ( ei->bei_parent->bei_kids || !ei->bei_parent->bei_id )
265 bdb->bi_cache.c_leaves++;
266 rc = avl_insert( &ei->bei_parent->bei_kids, ei2, bdb_rdn_cmp,
269 /* This should never happen; entry cache is corrupt */
270 bdb->bi_dbenv->log_flush( bdb->bi_dbenv, NULL );
274 ei->bei_parent->bei_ckids++;
282 /* Find the EntryInfo for the requested DN. If the DN cannot be found, return
283 * the info for its closest ancestor. *res should be NULL to process a
284 * complete DN starting from the tree root. Otherwise *res must be the
285 * immediate parent of the requested DN, and only the RDN will be searched.
286 * The EntryInfo is locked upon return and must be unlocked by the caller.
295 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
296 EntryInfo ei, *eip, *ei2;
300 /* this function is always called with normalized DN */
302 /* we're doing a onelevel search for an RDN */
303 ei.bei_nrdn.bv_val = ndn->bv_val;
304 ei.bei_nrdn.bv_len = dn_rdnlen( op->o_bd, ndn );
307 /* we're searching a full DN from the root */
308 ptr = ndn->bv_val + ndn->bv_len - op->o_bd->be_nsuffix[0].bv_len;
309 ei.bei_nrdn.bv_val = ptr;
310 ei.bei_nrdn.bv_len = op->o_bd->be_nsuffix[0].bv_len;
311 /* Skip to next rdn if suffix is empty */
312 if ( ei.bei_nrdn.bv_len == 0 ) {
313 for (ptr = ei.bei_nrdn.bv_val - 2; ptr > ndn->bv_val
314 && !DN_SEPARATOR(*ptr); ptr--) /* empty */;
315 if ( ptr >= ndn->bv_val ) {
316 if (DN_SEPARATOR(*ptr)) ptr++;
317 ei.bei_nrdn.bv_len = ei.bei_nrdn.bv_val - ptr;
318 ei.bei_nrdn.bv_val = ptr;
321 eip = &bdb->bi_cache.c_dntree;
324 for ( bdb_cache_entryinfo_lock( eip ); eip; ) {
326 ei2 = (EntryInfo *)avl_find( eip->bei_kids, &ei, bdb_rdn_cmp );
329 int len = ei.bei_nrdn.bv_len;
331 if ( BER_BVISEMPTY( ndn )) {
336 ei.bei_nrdn.bv_len = ndn->bv_len -
337 (ei.bei_nrdn.bv_val - ndn->bv_val);
338 bdb_cache_entryinfo_unlock( eip );
340 lock.mode = DB_LOCK_NG;
341 rc = bdb_dn2id( op, &ei.bei_nrdn, &ei, locker, &lock );
343 bdb_cache_entryinfo_lock( eip );
344 bdb_cache_entry_db_unlock( bdb->bi_dbenv, &lock );
349 /* DN exists but needs to be added to cache */
350 ei.bei_nrdn.bv_len = len;
351 rc = bdb_entryinfo_add_internal( bdb, &ei, &ei2 );
352 /* add_internal left eip and c_rwlock locked */
353 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
354 bdb_cache_entry_db_unlock( bdb->bi_dbenv, &lock );
359 } else if ( ei2->bei_state & CACHE_ENTRY_DELETED ) {
360 /* In the midst of deleting? Give it a chance to
363 bdb_cache_entryinfo_unlock( eip );
364 ldap_pvt_thread_yield();
365 bdb_cache_entryinfo_lock( eip );
369 bdb_cache_entryinfo_unlock( eip );
370 bdb_cache_entryinfo_lock( ei2 );
374 /* Advance to next lower RDN */
375 for (ptr = ei.bei_nrdn.bv_val - 2; ptr > ndn->bv_val
376 && !DN_SEPARATOR(*ptr); ptr--) /* empty */;
377 if ( ptr >= ndn->bv_val ) {
378 if (DN_SEPARATOR(*ptr)) ptr++;
379 ei.bei_nrdn.bv_len = ei.bei_nrdn.bv_val - ptr - 1;
380 ei.bei_nrdn.bv_val = ptr;
382 if ( ptr < ndn->bv_val ) {
392 /* Walk up the tree from a child node, looking for an ID that's already
393 * been linked into the cache.
396 hdb_cache_find_parent(
402 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
403 EntryInfo ei, eip, *ei2 = NULL, *ein = NULL, *eir = NULL;
412 rc = hdb_dn2id_parent( op, locker, &ei, &eip.bei_id );
415 /* Save the previous node, if any */
418 /* Create a new node for the current ID */
419 ein = bdb_cache_entryinfo_new( &bdb->bi_cache );
420 ein->bei_id = ei.bei_id;
421 ein->bei_kids = ei.bei_kids;
422 ein->bei_nrdn = ei.bei_nrdn;
423 ein->bei_rdn = ei.bei_rdn;
424 ein->bei_ckids = ei.bei_ckids;
425 #ifdef SLAP_ZONE_ALLOC
430 /* This node is not fully connected yet */
431 ein->bei_state = CACHE_ENTRY_NOT_LINKED;
433 /* Insert this node into the ID tree */
434 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
435 if ( avl_insert( &bdb->bi_cache.c_idtree, (caddr_t)ein,
436 bdb_id_cmp, avl_dup_error ) ) {
438 /* Someone else created this node just before us.
439 * Free our new copy and use the existing one.
441 bdb_cache_entryinfo_destroy( ein );
442 ein = (EntryInfo *)avl_find( bdb->bi_cache.c_idtree,
443 (caddr_t) &ei, bdb_id_cmp );
445 /* Link in any kids we've already processed */
447 bdb_cache_entryinfo_lock( ein );
448 avl_insert( &ein->bei_kids, (caddr_t)ei2,
449 bdb_rdn_cmp, avl_dup_error );
451 bdb_cache_entryinfo_unlock( ein );
457 /* If this is the first time, save this node
458 * to be returned later.
460 if ( eir == NULL ) eir = ein;
462 /* If there was a previous node, link it to this one */
463 if ( ei2 ) ei2->bei_parent = ein;
465 /* Look for this node's parent */
467 ei2 = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
468 (caddr_t) &eip, bdb_id_cmp );
470 ei2 = &bdb->bi_cache.c_dntree;
472 bdb->bi_cache.c_eiused++;
473 if ( ei2 && ( ei2->bei_kids || !ei2->bei_id ))
474 bdb->bi_cache.c_leaves++;
475 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
478 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_head_mutex );
479 bdb_cache_lru_add( bdb, ein );
483 /* Got the parent, link in and we're done. */
485 bdb_cache_entryinfo_lock( ei2 );
486 ein->bei_parent = ei2;
487 avl_insert( &ei2->bei_kids, (caddr_t)ein, bdb_rdn_cmp,
490 bdb_cache_entryinfo_unlock( ei2 );
491 bdb_cache_entryinfo_lock( eir );
493 /* Reset all the state info */
494 for (ein = eir; ein != ei2; ein=ein->bei_parent)
495 ein->bei_state &= ~CACHE_ENTRY_NOT_LINKED;
500 ei.bei_id = eip.bei_id;
502 avl_insert( &ei.bei_kids, (caddr_t)ein, bdb_rdn_cmp,
508 /* Used by hdb_dn2idl when loading the EntryInfo for all the children
512 struct bdb_info *bdb,
519 /* See if we already have this one */
520 bdb_cache_entryinfo_lock( ei->bei_parent );
521 ei2 = (EntryInfo *)avl_find( ei->bei_parent->bei_kids, ei, bdb_rdn_cmp );
522 bdb_cache_entryinfo_unlock( ei->bei_parent );
525 /* Not found, add it */
528 /* bei_rdn was not malloc'd before, do it now */
529 ber_dupbv( &bv, &ei->bei_rdn );
532 rc = bdb_entryinfo_add_internal( bdb, ei, res );
533 bdb_cache_entryinfo_unlock( ei->bei_parent );
534 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
536 /* Found, return it */
544 /* caller must have lru_head_mutex locked. mutex
545 * will be unlocked on return.
549 struct bdb_info *bdb,
552 DB_LOCK lock, *lockp;
553 EntryInfo *elru, *elprev;
556 LRU_ADD( &bdb->bi_cache, ei );
557 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_head_mutex );
559 /* See if we're above the cache size limit */
560 if ( bdb->bi_cache.c_cursize <= bdb->bi_cache.c_maxsize )
563 if ( bdb->bi_cache.c_locker ) {
569 /* Don't bother if we can't get the lock */
570 if ( ldap_pvt_thread_mutex_trylock( &bdb->bi_cache.lru_tail_mutex ) )
573 /* Look for an unused entry to remove */
574 for (elru = bdb->bi_cache.c_lrutail; elru; elru = elprev ) {
575 elprev = elru->bei_lruprev;
577 /* If we can successfully writelock it, then
578 * the object is idle.
580 if ( bdb_cache_entry_db_lock( bdb->bi_dbenv,
581 bdb->bi_cache.c_locker, elru, 1, 1, lockp ) == 0 ) {
584 /* If this node is in the process of linking into the cache,
585 * or this node is being deleted, skip it.
587 if ( elru->bei_state &
588 ( CACHE_ENTRY_NOT_LINKED | CACHE_ENTRY_DELETED )) {
589 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lockp );
592 /* Free entry for this node if it's present */
594 elru->bei_e->e_private = NULL;
595 #ifdef SLAP_ZONE_ALLOC
596 bdb_entry_return( bdb, elru->bei_e, elru->bei_zseq );
598 bdb_entry_return( elru->bei_e );
603 /* ITS#4010 if we're in slapcat, and this node is a leaf
606 * FIXME: we need to do this for slapd as well, (which is
607 * why we compute bi_cache.c_leaves now) but at the moment
608 * we can't because it causes unresolvable deadlocks.
610 if ( slapMode & SLAP_TOOL_READONLY ) {
611 if ( !elru->bei_kids ) {
612 /* This does LRU_DELETE for us */
613 bdb_cache_delete_internal( &bdb->bi_cache, elru, 0 );
614 bdb_cache_delete_cleanup( &bdb->bi_cache, elru );
616 /* Leave node on LRU list for a future pass */
618 LRU_DELETE( &bdb->bi_cache, elru );
620 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lockp );
622 if ( count >= bdb->bi_cache.c_minfree ) {
623 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
624 bdb->bi_cache.c_cursize -= count;
625 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
631 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_tail_mutex );
636 struct bdb_info *bdb,
639 EntryInfo ei = { 0 },
644 ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
645 ei2 = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
646 (caddr_t) &ei, bdb_id_cmp );
647 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
652 * cache_find_id - find an entry in the cache, given id.
653 * The entry is locked for Read upon return. Call with islocked TRUE if
654 * the supplied *eip was already locked.
667 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
669 int rc = 0, load = 0;
670 EntryInfo ei = { 0 };
674 #ifdef SLAP_ZONE_ALLOC
675 slap_zh_rlock(bdb->bi_cache.c_zctx);
677 /* If we weren't given any info, see if we have it already cached */
679 again: ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
680 *eip = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
681 (caddr_t) &ei, bdb_id_cmp );
683 /* If the lock attempt fails, the info is in use */
684 if ( ldap_pvt_thread_mutex_trylock(
685 &(*eip)->bei_kids_mutex )) {
686 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
687 /* If this node is being deleted, treat
688 * as if the delete has already finished
690 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
693 /* otherwise, wait for the info to free up */
694 ldap_pvt_thread_yield();
697 /* If this info isn't hooked up to its parent yet,
698 * unlock and wait for it to be fully initialized
700 if ( (*eip)->bei_state & CACHE_ENTRY_NOT_LINKED ) {
701 bdb_cache_entryinfo_unlock( *eip );
702 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
703 ldap_pvt_thread_yield();
708 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
711 /* See if the ID exists in the database; add it to the cache if so */
714 rc = bdb_id2entry( op->o_bd, tid, locker, id, &ep );
716 rc = bdb_cache_find_ndn( op, locker,
718 if ( *eip ) islocked = 1;
720 ep->e_private = NULL;
721 #ifdef SLAP_ZONE_ALLOC
722 bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
724 bdb_entry_return( ep );
730 rc = hdb_cache_find_parent(op, locker, id, eip );
731 if ( rc == 0 ) islocked = 1;
735 /* Ok, we found the info, do we have the entry? */
737 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
740 /* Make sure only one thread tries to load the entry */
742 #ifdef SLAP_ZONE_ALLOC
743 if ((*eip)->bei_e && !slap_zn_validate(
744 bdb->bi_cache.c_zctx, (*eip)->bei_e, (*eip)->bei_zseq)) {
745 (*eip)->bei_e = NULL;
746 (*eip)->bei_zseq = 0;
749 if ( !(*eip)->bei_e && !((*eip)->bei_state & CACHE_ENTRY_LOADING)) {
751 (*eip)->bei_state |= CACHE_ENTRY_LOADING;
754 bdb_cache_entryinfo_unlock( *eip );
757 rc = bdb_cache_entry_db_lock( bdb->bi_dbenv, locker, *eip, 0, 0, lock );
758 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
760 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
761 } else if ( rc == 0 ) {
763 /* Give up original read lock, obtain write lock
766 rc = bdb_cache_entry_db_relock( bdb->bi_dbenv, locker,
769 if ( rc == 0 && !ep) {
770 rc = bdb_id2entry( op->o_bd, tid, locker, id, &ep );
773 ep->e_private = *eip;
778 #ifdef SLAP_ZONE_ALLOC
779 (*eip)->bei_zseq = *((ber_len_t *)ep - 2);
783 bdb_cache_entryinfo_lock( *eip );
784 (*eip)->bei_state ^= CACHE_ENTRY_LOADING;
785 bdb_cache_entryinfo_unlock( *eip );
787 /* If we succeeded, downgrade back to a readlock. */
788 rc = bdb_cache_entry_db_relock( bdb->bi_dbenv, locker,
791 /* Otherwise, release the lock. */
792 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
794 } else if ( !(*eip)->bei_e ) {
795 /* Some other thread is trying to load the entry,
796 * give it a chance to finish.
798 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
799 ldap_pvt_thread_yield();
800 bdb_cache_entryinfo_lock( *eip );
805 /* Check for subtree renames
807 rc = bdb_fix_dn( (*eip)->bei_e, 1 );
809 bdb_cache_entry_db_relock( bdb->bi_dbenv,
810 locker, *eip, 1, 0, lock );
811 /* check again in case other modifier did it already */
812 if ( bdb_fix_dn( (*eip)->bei_e, 1 ) )
813 rc = bdb_fix_dn( (*eip)->bei_e, 2 );
814 bdb_cache_entry_db_relock( bdb->bi_dbenv,
815 locker, *eip, 0, 0, lock );
824 bdb_cache_entryinfo_unlock( *eip );
827 ep->e_private = NULL;
828 #ifdef SLAP_ZONE_ALLOC
829 bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
831 bdb_entry_return( ep );
837 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
838 bdb->bi_cache.c_cursize++;
839 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
842 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_head_mutex );
844 /* If the LRU list has only one entry and this is it, it
845 * doesn't need to be added again.
847 if ( bdb->bi_cache.c_lruhead == bdb->bi_cache.c_lrutail &&
848 bdb->bi_cache.c_lruhead == *eip ) {
849 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_head_mutex );
851 /* if entry is on LRU list, remove from old spot */
852 if ( (*eip)->bei_lrunext || (*eip)->bei_lruprev ) {
853 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_tail_mutex );
854 LRU_DELETE( &bdb->bi_cache, *eip );
855 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_tail_mutex );
857 /* lru_head_mutex is unlocked for us */
858 bdb_cache_lru_add( bdb, *eip );
862 #ifdef SLAP_ZONE_ALLOC
863 if (rc == 0 && (*eip)->bei_e) {
864 slap_zn_rlock(bdb->bi_cache.c_zctx, (*eip)->bei_e);
866 slap_zh_runlock(bdb->bi_cache.c_zctx);
879 if ( BEI(e)->bei_kids ) {
882 if ( BEI(e)->bei_state & CACHE_ENTRY_NO_KIDS ) {
885 rc = bdb_dn2id_children( op, txn, e );
886 if ( rc == DB_NOTFOUND ) {
887 BEI(e)->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
892 /* Update the cache after a successful database Add. */
895 struct bdb_info *bdb,
905 struct berval rdn = e->e_name;
913 /* Lock this entry so that bdb_add can run to completion.
914 * It can only fail if BDB has run out of lock resources.
916 rc = bdb_cache_entry_db_lock( bdb->bi_dbenv, locker, &ei, 1, 0, &lock );
918 bdb_cache_entryinfo_unlock( eip );
923 if ( nrdn->bv_len != e->e_nname.bv_len ) {
924 char *ptr = ber_bvchr( &rdn, ',' );
925 assert( ptr != NULL );
926 rdn.bv_len = ptr - rdn.bv_val;
928 ber_dupbv( &ei.bei_rdn, &rdn );
929 if ( eip->bei_dkids ) eip->bei_dkids++;
932 rc = bdb_entryinfo_add_internal( bdb, &ei, &new );
933 /* bdb_csn_commit can cause this when adding the database root entry */
935 new->bei_e->e_private = NULL;
936 #ifdef SLAP_ZONE_ALLOC
937 bdb_entry_return( bdb, new->bei_e, new->bei_zseq );
939 bdb_entry_return( new->bei_e );
944 new->bei_state = CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
945 eip->bei_state &= ~CACHE_ENTRY_NO_KIDS;
946 if (eip->bei_parent) {
947 eip->bei_parent->bei_state &= ~CACHE_ENTRY_NO_GRANDKIDS;
949 bdb_cache_entryinfo_unlock( eip );
951 ++bdb->bi_cache.c_cursize;
952 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
955 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_head_mutex );
957 /* lru_head_mutex is unlocked for us */
958 bdb_cache_lru_add( bdb, new );
971 EntryInfo *ei = BEI(e);
973 /* Get write lock on data */
974 rc = bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock );
976 /* If we've done repeated mods on a cached entry, then e_attrs
977 * is no longer contiguous with the entry, and must be freed.
980 if ( (void *)e->e_attrs != (void *)(e+1) ) {
981 attrs_free( e->e_attrs );
983 e->e_attrs = newAttrs;
989 * Change the rdn in the entryinfo. Also move to a new parent if needed.
993 struct bdb_info *bdb,
1001 EntryInfo *ei = BEI(e), *pei;
1007 /* Get write lock on data */
1008 rc = bdb_cache_entry_db_relock( bdb->bi_dbenv, locker, ei, 1, 0, lock );
1009 if ( rc ) return rc;
1011 /* If we've done repeated mods on a cached entry, then e_attrs
1012 * is no longer contiguous with the entry, and must be freed.
1014 if ( (void *)e->e_attrs != (void *)(e+1) ) {
1015 attrs_free( e->e_attrs );
1017 e->e_attrs = new->e_attrs;
1018 if( e->e_nname.bv_val < e->e_bv.bv_val ||
1019 e->e_nname.bv_val > e->e_bv.bv_val + e->e_bv.bv_len )
1021 ch_free(e->e_name.bv_val);
1022 ch_free(e->e_nname.bv_val);
1024 e->e_name = new->e_name;
1025 e->e_nname = new->e_nname;
1027 /* Lock the parent's kids AVL tree */
1028 pei = ei->bei_parent;
1029 bdb_cache_entryinfo_lock( pei );
1030 avl_delete( &pei->bei_kids, (caddr_t) ei, bdb_rdn_cmp );
1031 free( ei->bei_nrdn.bv_val );
1032 ber_dupbv( &ei->bei_nrdn, nrdn );
1034 if ( !pei->bei_kids )
1035 pei->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
1038 free( ei->bei_rdn.bv_val );
1041 if ( nrdn->bv_len != e->e_nname.bv_len ) {
1042 char *ptr = ber_bvchr(&rdn, ',');
1043 assert( ptr != NULL );
1044 rdn.bv_len = ptr - rdn.bv_val;
1046 ber_dupbv( &ei->bei_rdn, &rdn );
1048 if ( pei->bei_dkids ) pei->bei_dkids--;
1052 ein = ei->bei_parent;
1054 ei->bei_parent = ein;
1055 bdb_cache_entryinfo_unlock( pei );
1056 bdb_cache_entryinfo_lock( ein );
1058 /* parent now has kids */
1059 if ( ein->bei_state & CACHE_ENTRY_NO_KIDS )
1060 ein->bei_state ^= CACHE_ENTRY_NO_KIDS;
1062 /* parent might now have grandkids */
1063 if ( ein->bei_state & CACHE_ENTRY_NO_GRANDKIDS &&
1064 !(ei->bei_state & (CACHE_ENTRY_NO_KIDS)))
1065 ein->bei_state ^= CACHE_ENTRY_NO_GRANDKIDS;
1068 /* Record the generation number of this change */
1069 ldap_pvt_thread_mutex_lock( &bdb->bi_modrdns_mutex );
1071 ei->bei_modrdns = bdb->bi_modrdns;
1072 ldap_pvt_thread_mutex_unlock( &bdb->bi_modrdns_mutex );
1075 if ( ein->bei_dkids ) ein->bei_dkids++;
1077 avl_insert( &ein->bei_kids, ei, bdb_rdn_cmp, avl_dup_error );
1078 bdb_cache_entryinfo_unlock( ein );
1082 * cache_delete - delete the entry e from the cache.
1084 * returns: 0 e was deleted ok
1085 * 1 e was not in the cache
1086 * -1 something bad happened
1096 EntryInfo *ei = BEI(e);
1099 assert( e->e_private != NULL );
1101 /* Set this early, warn off any queriers */
1102 ei->bei_state |= CACHE_ENTRY_DELETED;
1104 /* Lock the entry's info */
1105 bdb_cache_entryinfo_lock( ei );
1107 /* Get write lock on the data */
1108 rc = bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock );
1110 /* couldn't lock, undo and give up */
1111 ei->bei_state ^= CACHE_ENTRY_DELETED;
1112 bdb_cache_entryinfo_unlock( ei );
1116 Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_delete( %ld )\n",
1120 ldap_pvt_thread_mutex_lock( &cache->lru_tail_mutex );
1122 /* set cache write lock */
1123 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
1125 rc = bdb_cache_delete_internal( cache, e->e_private, 1 );
1127 /* free cache write lock */
1128 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
1130 /* free lru mutex */
1131 ldap_pvt_thread_mutex_unlock( &cache->lru_tail_mutex );
1133 /* Leave entry info locked */
1139 bdb_cache_delete_cleanup(
1144 ei->bei_e->e_private = NULL;
1145 #ifdef SLAP_ZONE_ALLOC
1146 bdb_entry_return( ei->bei_bdb, ei->bei_e, ei->bei_zseq );
1148 bdb_entry_return( ei->bei_e );
1153 free( ei->bei_nrdn.bv_val );
1154 ei->bei_nrdn.bv_val = NULL;
1156 free( ei->bei_rdn.bv_val );
1157 ei->bei_rdn.bv_val = NULL;
1158 ei->bei_modrdns = 0;
1162 ei->bei_parent = NULL;
1163 ei->bei_kids = NULL;
1164 ei->bei_lruprev = NULL;
1166 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
1167 ei->bei_lrunext = cache->c_eifree;
1168 cache->c_eifree = ei;
1169 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
1170 bdb_cache_entryinfo_unlock( ei );
1174 bdb_cache_delete_internal(
1179 int rc = 0; /* return code */
1181 /* Lock the parent's kids tree */
1182 bdb_cache_entryinfo_lock( e->bei_parent );
1185 e->bei_parent->bei_ckids--;
1186 if ( decr && e->bei_parent->bei_dkids ) e->bei_parent->bei_dkids--;
1189 if ( avl_delete( &e->bei_parent->bei_kids, (caddr_t) e, bdb_rdn_cmp )
1194 if ( e->bei_parent->bei_kids )
1198 if ( avl_delete( &cache->c_idtree, (caddr_t) e, bdb_id_cmp ) == NULL ) {
1206 LRU_DELETE( cache, e );
1207 if ( e->bei_e ) cache->c_cursize--;
1210 bdb_cache_entryinfo_unlock( e->bei_parent );
1216 bdb_entryinfo_release( void *data )
1218 EntryInfo *ei = (EntryInfo *)data;
1219 if ( ei->bei_kids ) {
1220 avl_free( ei->bei_kids, NULL );
1223 ei->bei_e->e_private = NULL;
1224 #ifdef SLAP_ZONE_ALLOC
1225 bdb_entry_return( ei->bei_bdb, ei->bei_e, ei->bei_zseq );
1227 bdb_entry_return( ei->bei_e );
1230 bdb_cache_entryinfo_destroy( ei );
1234 bdb_cache_release_all( Cache *cache )
1236 /* set cache write lock */
1237 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
1239 ldap_pvt_thread_mutex_lock( &cache->lru_tail_mutex );
1241 Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_release_all\n", 0, 0, 0 );
1243 avl_free( cache->c_dntree.bei_kids, NULL );
1244 avl_free( cache->c_idtree, bdb_entryinfo_release );
1245 for (;cache->c_eifree;cache->c_eifree = cache->c_lruhead) {
1246 cache->c_lruhead = cache->c_eifree->bei_lrunext;
1247 bdb_cache_entryinfo_destroy(cache->c_eifree);
1249 cache->c_cursize = 0;
1250 cache->c_eiused = 0;
1251 cache->c_leaves = 0;
1252 cache->c_idtree = NULL;
1253 cache->c_lruhead = NULL;
1254 cache->c_lrutail = NULL;
1255 cache->c_dntree.bei_kids = NULL;
1257 /* free lru mutex */
1258 ldap_pvt_thread_mutex_unlock( &cache->lru_tail_mutex );
1259 /* free cache write lock */
1260 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
1266 bdb_lru_print( Cache *cache )
1270 fprintf( stderr, "LRU queue (head to tail):\n" );
1271 for ( e = cache->c_lruhead; e != NULL; e = e->bei_lrunext ) {
1272 fprintf( stderr, "\trdn \"%20s\" id %ld\n",
1273 e->bei_nrdn.bv_val, e->bei_id );
1275 fprintf( stderr, "LRU queue (tail to head):\n" );
1276 for ( e = cache->c_lrutail; e != NULL; e = e->bei_lruprev ) {
1277 fprintf( stderr, "\trdn \"%20s\" id %ld\n",
1278 e->bei_nrdn.bv_val, e->bei_id );
1284 #ifdef BDB_REUSE_LOCKERS
1286 bdb_locker_id_free( void *key, void *data )
1289 u_int32_t lockid = (long)data;
1292 rc = XLOCK_ID_FREE( env, lockid );
1293 if ( rc == EINVAL ) {
1295 Debug( LDAP_DEBUG_ANY,
1296 "bdb_locker_id_free: %lu err %s(%d)\n",
1297 (unsigned long) lockid, db_strerror(rc), rc );
1298 /* release all locks held by this locker. */
1299 lr.op = DB_LOCK_PUT_ALL;
1301 env->lock_vec( env, lockid, 0, &lr, 1, NULL );
1302 XLOCK_ID_FREE( env, lockid );
1306 /* free up any keys used by the main thread */
1308 bdb_locker_flush( DB_ENV *env )
1311 void *ctx = ldap_pvt_thread_pool_context();
1313 if ( !ldap_pvt_thread_pool_getkey( ctx, env, &data, NULL ) ) {
1314 ldap_pvt_thread_pool_setkey( ctx, env, NULL, NULL );
1315 bdb_locker_id_free( env, data );
1320 bdb_locker_id( Operation *op, DB_ENV *env, u_int32_t *locker )
1327 if ( !env || !locker ) return -1;
1329 /* If no op was provided, try to find the ctx anyway... */
1331 ctx = op->o_threadctx;
1333 ctx = ldap_pvt_thread_pool_context();
1336 /* Shouldn't happen unless we're single-threaded */
1342 if ( ldap_pvt_thread_pool_getkey( ctx, env, &data, NULL ) ) {
1343 for ( i=0, rc=1; rc != 0 && i<4; i++ ) {
1344 rc = XLOCK_ID( env, &lockid );
1345 if (rc) ldap_pvt_thread_yield();
1350 data = (void *)((long)lockid);
1351 if ( ( rc = ldap_pvt_thread_pool_setkey( ctx, env,
1352 data, bdb_locker_id_free ) ) ) {
1353 XLOCK_ID_FREE( env, lockid );
1354 Debug( LDAP_DEBUG_ANY, "bdb_locker_id: err %s(%d)\n",
1355 db_strerror(rc), rc, 0 );
1360 lockid = (long)data;
1365 #endif /* BDB_REUSE_LOCKERS */
1368 bdb_cache_delete_entry(
1369 struct bdb_info *bdb,
1374 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
1375 if ( bdb_cache_entry_db_lock( bdb->bi_dbenv, bdb->bi_cache.c_locker, ei, 1, 1, lock ) == 0 )
1377 if ( ei->bei_e && !(ei->bei_state & CACHE_ENTRY_NOT_LINKED )) {
1378 LRU_DELETE( &bdb->bi_cache, ei );
1379 ei->bei_e->e_private = NULL;
1380 #ifdef SLAP_ZONE_ALLOC
1381 bdb_entry_return( bdb, ei->bei_e, ei->bei_zseq );
1383 bdb_entry_return( ei->bei_e );
1386 --bdb->bi_cache.c_cursize;
1388 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
1390 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );