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 2000-2005 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>
30 #define bdb_cache_lru_add hdb_cache_lru_add
33 static int bdb_cache_delete_internal(Cache *cache, EntryInfo *e, int decr);
35 static void bdb_lru_print(Cache *cache);
38 static int bdb_txn_get( Operation *op, DB_ENV *env, DB_TXN **txn, int reset );
41 #if DB_VERSION_FULL == 0x04020034
42 #define READ_TXN_FLAG ReadFlag
43 static int ReadFlag = DB_TXN_NOT_DURABLE;
45 #define READ_TXN_FLAG 0
49 bdb_cache_entryinfo_new( Cache *cache )
53 if ( cache->c_eifree ) {
54 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
55 if ( cache->c_eifree ) {
57 cache->c_eifree = ei->bei_lrunext;
59 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
62 ei->bei_lrunext = NULL;
65 ei = ch_calloc(1, sizeof(struct bdb_entry_info));
66 ldap_pvt_thread_mutex_init( &ei->bei_kids_mutex );
72 /* Atomically release and reacquire a lock */
74 bdb_cache_entry_db_relock(
89 if ( !lock ) return 0;
91 lockobj.data = &ei->bei_id;
92 lockobj.size = sizeof(ei->bei_id) + 1;
94 list[0].op = DB_LOCK_PUT;
96 list[1].op = DB_LOCK_GET;
98 list[1].mode = rw ? DB_LOCK_WRITE : DB_LOCK_READ;
99 list[1].obj = &lockobj;
100 rc = env->lock_vec(env, locker, tryOnly ? DB_LOCK_NOWAIT : 0,
103 if (rc && !tryOnly) {
104 Debug( LDAP_DEBUG_TRACE,
105 "bdb_cache_entry_db_relock: entry %ld, rw %d, rc %d\n",
106 ei->bei_id, rw, rc );
108 *lock = list[1].lock;
115 bdb_cache_entry_db_lock( DB_ENV *env, u_int32_t locker, EntryInfo *ei,
116 int rw, int tryOnly, DB_LOCK *lock )
125 if ( !lock ) return 0;
128 db_rw = DB_LOCK_WRITE;
130 db_rw = DB_LOCK_READ;
132 lockobj.data = &ei->bei_id;
133 lockobj.size = sizeof(ei->bei_id) + 1;
135 rc = LOCK_GET(env, locker, tryOnly ? DB_LOCK_NOWAIT : 0,
136 &lockobj, db_rw, lock);
137 if (rc && !tryOnly) {
138 Debug( LDAP_DEBUG_TRACE,
139 "bdb_cache_entry_db_lock: entry %ld, rw %d, rc %d\n",
140 ei->bei_id, rw, rc );
143 #endif /* NO_THREADS */
147 bdb_cache_entry_db_unlock ( DB_ENV *env, DB_LOCK *lock )
154 if ( !lock ) return 0;
156 rc = LOCK_PUT ( env, lock );
162 bdb_cache_entryinfo_destroy( EntryInfo *e )
164 ldap_pvt_thread_mutex_destroy( &e->bei_kids_mutex );
165 free( e->bei_nrdn.bv_val );
167 free( e->bei_rdn.bv_val );
173 #define LRU_DELETE( cache, ei ) do { \
174 if ( (ei)->bei_lruprev != NULL ) { \
175 (ei)->bei_lruprev->bei_lrunext = (ei)->bei_lrunext; \
177 (cache)->c_lruhead = (ei)->bei_lrunext; \
179 if ( (ei)->bei_lrunext != NULL ) { \
180 (ei)->bei_lrunext->bei_lruprev = (ei)->bei_lruprev; \
182 (cache)->c_lrutail = (ei)->bei_lruprev; \
184 (ei)->bei_lrunext = (ei)->bei_lruprev = NULL; \
187 #define LRU_ADD( cache, ei ) do { \
188 (ei)->bei_lrunext = (cache)->c_lruhead; \
189 if ( (ei)->bei_lrunext != NULL ) { \
190 (ei)->bei_lrunext->bei_lruprev = (ei); \
192 (cache)->c_lruhead = (ei); \
193 (ei)->bei_lruprev = NULL; \
194 if ( (cache)->c_lrutail == NULL ) { \
195 (cache)->c_lrutail = (ei); \
199 /* Do a length-ordered sort on normalized RDNs */
201 bdb_rdn_cmp( const void *v_e1, const void *v_e2 )
203 const EntryInfo *e1 = v_e1, *e2 = v_e2;
204 int rc = e1->bei_nrdn.bv_len - e2->bei_nrdn.bv_len;
206 rc = strncmp( e1->bei_nrdn.bv_val, e2->bei_nrdn.bv_val,
207 e1->bei_nrdn.bv_len );
213 bdb_id_cmp( const void *v_e1, const void *v_e2 )
215 const EntryInfo *e1 = v_e1, *e2 = v_e2;
216 return e1->bei_id - e2->bei_id;
219 /* Create an entryinfo in the cache. Caller must release the locks later.
222 bdb_entryinfo_add_internal(
223 struct bdb_info *bdb,
227 EntryInfo *ei2 = NULL;
231 ei2 = bdb_cache_entryinfo_new( &bdb->bi_cache );
233 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
234 bdb_cache_entryinfo_lock( ei->bei_parent );
236 ei2->bei_id = ei->bei_id;
237 ei2->bei_parent = ei->bei_parent;
239 ei2->bei_rdn = ei->bei_rdn;
241 #ifdef SLAP_ZONE_ALLOC
245 /* Add to cache ID tree */
246 if (avl_insert( &bdb->bi_cache.c_idtree, ei2, bdb_id_cmp, avl_dup_error )) {
248 eix = avl_find( bdb->bi_cache.c_idtree, ei2, bdb_id_cmp );
249 bdb_cache_entryinfo_destroy( ei2 );
252 /* It got freed above because its value was
255 ei->bei_rdn.bv_val = NULL;
258 bdb->bi_cache.c_eiused++;
259 ber_dupbv( &ei2->bei_nrdn, &ei->bei_nrdn );
261 /* This is a new leaf node. But if parent had no kids, then it was
262 * a leaf and we would be decrementing that. So, only increment if
263 * the parent already has kids.
265 if ( ei->bei_parent->bei_kids || !ei->bei_parent->bei_id )
266 bdb->bi_cache.c_leaves++;
267 avl_insert( &ei->bei_parent->bei_kids, ei2, bdb_rdn_cmp,
270 ei->bei_parent->bei_ckids++;
278 /* Find the EntryInfo for the requested DN. If the DN cannot be found, return
279 * the info for its closest ancestor. *res should be NULL to process a
280 * complete DN starting from the tree root. Otherwise *res must be the
281 * immediate parent of the requested DN, and only the RDN will be searched.
282 * The EntryInfo is locked upon return and must be unlocked by the caller.
291 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
292 EntryInfo ei, *eip, *ei2;
296 /* this function is always called with normalized DN */
298 /* we're doing a onelevel search for an RDN */
299 ei.bei_nrdn.bv_val = ndn->bv_val;
300 ei.bei_nrdn.bv_len = dn_rdnlen( op->o_bd, ndn );
303 /* we're searching a full DN from the root */
304 ptr = ndn->bv_val + ndn->bv_len - op->o_bd->be_nsuffix[0].bv_len;
305 ei.bei_nrdn.bv_val = ptr;
306 ei.bei_nrdn.bv_len = op->o_bd->be_nsuffix[0].bv_len;
307 /* Skip to next rdn if suffix is empty */
308 if ( ei.bei_nrdn.bv_len == 0 ) {
309 for (ptr = ei.bei_nrdn.bv_val - 2; ptr > ndn->bv_val
310 && !DN_SEPARATOR(*ptr); ptr--) /* empty */;
311 if ( ptr >= ndn->bv_val ) {
312 if (DN_SEPARATOR(*ptr)) ptr++;
313 ei.bei_nrdn.bv_len = ei.bei_nrdn.bv_val - ptr;
314 ei.bei_nrdn.bv_val = ptr;
317 eip = &bdb->bi_cache.c_dntree;
320 for ( bdb_cache_entryinfo_lock( eip ); eip; ) {
322 ei2 = (EntryInfo *)avl_find( eip->bei_kids, &ei, bdb_rdn_cmp );
324 int len = ei.bei_nrdn.bv_len;
326 if ( BER_BVISEMPTY( ndn )) {
331 ei.bei_nrdn.bv_len = ndn->bv_len -
332 (ei.bei_nrdn.bv_val - ndn->bv_val);
333 bdb_cache_entryinfo_unlock( eip );
335 rc = bdb_dn2id( op, txn, &ei.bei_nrdn, &ei );
337 bdb_cache_entryinfo_lock( eip );
342 /* DN exists but needs to be added to cache */
343 ei.bei_nrdn.bv_len = len;
344 rc = bdb_entryinfo_add_internal( bdb, &ei, &ei2 );
345 /* add_internal left eip and c_rwlock locked */
346 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
351 } else if ( ei2->bei_state & CACHE_ENTRY_DELETED ) {
352 /* In the midst of deleting? Give it a chance to
355 bdb_cache_entryinfo_unlock( eip );
356 ldap_pvt_thread_yield();
357 bdb_cache_entryinfo_lock( eip );
361 bdb_cache_entryinfo_unlock( eip );
362 bdb_cache_entryinfo_lock( ei2 );
366 /* Advance to next lower RDN */
367 for (ptr = ei.bei_nrdn.bv_val - 2; ptr > ndn->bv_val
368 && !DN_SEPARATOR(*ptr); ptr--) /* empty */;
369 if ( ptr >= ndn->bv_val ) {
370 if (DN_SEPARATOR(*ptr)) ptr++;
371 ei.bei_nrdn.bv_len = ei.bei_nrdn.bv_val - ptr - 1;
372 ei.bei_nrdn.bv_val = ptr;
374 if ( ptr < ndn->bv_val ) {
384 /* Walk up the tree from a child node, looking for an ID that's already
385 * been linked into the cache.
388 hdb_cache_find_parent(
394 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
395 EntryInfo ei, eip, *ei2 = NULL, *ein = NULL, *eir = NULL;
404 rc = hdb_dn2id_parent( op, txn, &ei, &eip.bei_id );
407 /* Save the previous node, if any */
410 /* Create a new node for the current ID */
411 ein = bdb_cache_entryinfo_new( &bdb->bi_cache );
412 ein->bei_id = ei.bei_id;
413 ein->bei_kids = ei.bei_kids;
414 ein->bei_nrdn = ei.bei_nrdn;
415 ein->bei_rdn = ei.bei_rdn;
416 ein->bei_ckids = ei.bei_ckids;
417 #ifdef SLAP_ZONE_ALLOC
422 /* This node is not fully connected yet */
423 ein->bei_state = CACHE_ENTRY_NOT_LINKED;
425 /* Insert this node into the ID tree */
426 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
427 if ( avl_insert( &bdb->bi_cache.c_idtree, (caddr_t)ein,
428 bdb_id_cmp, avl_dup_error ) ) {
430 /* Someone else created this node just before us.
431 * Free our new copy and use the existing one.
433 bdb_cache_entryinfo_destroy( ein );
434 ein = (EntryInfo *)avl_find( bdb->bi_cache.c_idtree,
435 (caddr_t) &ei, bdb_id_cmp );
437 /* Link in any kids we've already processed */
439 bdb_cache_entryinfo_lock( ein );
440 avl_insert( &ein->bei_kids, (caddr_t)ei2,
441 bdb_rdn_cmp, avl_dup_error );
443 bdb_cache_entryinfo_unlock( ein );
449 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_mutex );
450 LRU_ADD( &bdb->bi_cache, ein );
451 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_mutex );
455 /* If this is the first time, save this node
456 * to be returned later.
458 if ( eir == NULL ) eir = ein;
460 /* If there was a previous node, link it to this one */
461 if ( ei2 ) ei2->bei_parent = ein;
463 /* Look for this node's parent */
465 ei2 = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
466 (caddr_t) &eip, bdb_id_cmp );
468 ei2 = &bdb->bi_cache.c_dntree;
470 bdb->bi_cache.c_eiused++;
471 if ( ei2 && ( ei2->bei_kids || !ei2->bei_id ))
472 bdb->bi_cache.c_leaves++;
473 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
475 /* Got the parent, link in and we're done. */
477 bdb_cache_entryinfo_lock( ei2 );
478 ein->bei_parent = ei2;
479 avl_insert( &ei2->bei_kids, (caddr_t)ein, bdb_rdn_cmp,
482 bdb_cache_entryinfo_unlock( ei2 );
483 bdb_cache_entryinfo_lock( eir );
485 /* Reset all the state info */
486 for (ein = eir; ein != ei2; ein=ein->bei_parent)
487 ein->bei_state &= ~CACHE_ENTRY_NOT_LINKED;
492 ei.bei_id = eip.bei_id;
494 avl_insert( &ei.bei_kids, (caddr_t)ein, bdb_rdn_cmp,
500 /* Used by hdb_dn2idl when loading the EntryInfo for all the children
504 struct bdb_info *bdb,
511 /* See if we already have this one */
512 bdb_cache_entryinfo_lock( ei->bei_parent );
513 ei2 = (EntryInfo *)avl_find( ei->bei_parent->bei_kids, ei, bdb_rdn_cmp );
514 bdb_cache_entryinfo_unlock( ei->bei_parent );
517 /* Not found, add it */
520 /* bei_rdn was not malloc'd before, do it now */
521 ber_dupbv( &bv, &ei->bei_rdn );
524 rc = bdb_entryinfo_add_internal( bdb, ei, res );
525 bdb_cache_entryinfo_unlock( ei->bei_parent );
526 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
528 /* Found, return it */
536 /* caller must have lru_mutex locked. mutex
537 * will be unlocked on return.
541 struct bdb_info *bdb,
545 DB_LOCK lock, *lockp;
553 /* See if we're above the cache size limit */
554 if ( bdb->bi_cache.c_cursize > bdb->bi_cache.c_maxsize ) {
555 EntryInfo *elru, *elprev;
558 /* Look for an unused entry to remove */
559 for (elru = bdb->bi_cache.c_lrutail; elru; elru = elprev, i++ ) {
560 elprev = elru->bei_lruprev;
562 /* Too many probes, not enough idle, give up */
566 /* If we can successfully writelock it, then
567 * the object is idle.
569 if ( bdb_cache_entry_db_lock( bdb->bi_dbenv,
570 bdb->bi_cache.c_locker, elru, 1, 1, lockp ) == 0 ) {
572 int stop = 0, decr = 0;
574 /* If this node is in the process of linking into the cache,
575 * or this node is being deleted, skip it.
577 if ( elru->bei_state &
578 ( CACHE_ENTRY_NOT_LINKED | CACHE_ENTRY_DELETED )) {
579 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lockp );
582 /* Free entry for this node if it's present */
584 elru->bei_e->e_private = NULL;
585 #ifdef SLAP_ZONE_ALLOC
586 bdb_entry_return( bdb, elru->bei_e, elru->bei_zseq );
588 bdb_entry_return( elru->bei_e );
593 /* ITS#4010 if we're in slapcat, and this node is a leaf
596 * FIXME: we need to do this for slapd as well, (which is
597 * why we compute bi_cache.c_leaves now) but at the moment
598 * we can't because it causes unresolvable deadlocks.
600 if ( slapMode & SLAP_TOOL_READONLY ) {
601 if ( !elru->bei_kids ) {
602 /* This does LRU_DELETE for us */
603 bdb_cache_delete_internal( &bdb->bi_cache, elru, 0 );
604 bdb_cache_delete_cleanup( &bdb->bi_cache, elru );
606 /* Leave node on LRU list for a future pass */
608 LRU_DELETE( &bdb->bi_cache, elru );
610 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lockp );
612 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
614 --bdb->bi_cache.c_cursize;
615 if (bdb->bi_cache.c_cursize <= bdb->bi_cache.c_maxsize)
617 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
622 LRU_ADD( &bdb->bi_cache, ei );
623 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_mutex );
628 struct bdb_info *bdb,
631 EntryInfo ei = { 0 },
636 ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
637 ei2 = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
638 (caddr_t) &ei, bdb_id_cmp );
639 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
644 * cache_find_id - find an entry in the cache, given id.
645 * The entry is locked for Read upon return. Call with islocked TRUE if
646 * the supplied *eip was already locked.
659 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
661 int rc = 0, load = 0;
662 EntryInfo ei = { 0 };
666 #ifdef SLAP_ZONE_ALLOC
667 slap_zh_rlock(bdb->bi_cache.c_zctx);
669 /* If we weren't given any info, see if we have it already cached */
671 again: ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
672 *eip = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
673 (caddr_t) &ei, bdb_id_cmp );
675 /* If the lock attempt fails, the info is in use */
676 if ( ldap_pvt_thread_mutex_trylock(
677 &(*eip)->bei_kids_mutex )) {
678 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
679 /* If this node is being deleted, treat
680 * as if the delete has already finished
682 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
685 /* otherwise, wait for the info to free up */
686 ldap_pvt_thread_yield();
689 /* If this info isn't hooked up to its parent yet,
690 * unlock and wait for it to be fully initialized
692 if ( (*eip)->bei_state & CACHE_ENTRY_NOT_LINKED ) {
693 bdb_cache_entryinfo_unlock( *eip );
694 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
695 ldap_pvt_thread_yield();
700 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
703 /* See if the ID exists in the database; add it to the cache if so */
706 rc = bdb_id2entry( op->o_bd, tid, id, &ep );
708 rc = bdb_cache_find_ndn( op, tid,
710 if ( *eip ) islocked = 1;
712 #ifdef SLAP_ZONE_ALLOC
713 bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
715 bdb_entry_return( ep );
721 rc = hdb_cache_find_parent(op, tid, id, eip );
722 if ( rc == 0 && *eip ) islocked = 1;
726 /* Ok, we found the info, do we have the entry? */
727 if ( *eip && rc == 0 ) {
728 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
731 /* Make sure only one thread tries to load the entry */
733 #ifdef SLAP_ZONE_ALLOC
734 if ((*eip)->bei_e && !slap_zn_validate(
735 bdb->bi_cache.c_zctx, (*eip)->bei_e, (*eip)->bei_zseq)) {
736 (*eip)->bei_e = NULL;
737 (*eip)->bei_zseq = 0;
740 if ( !(*eip)->bei_e && !((*eip)->bei_state & CACHE_ENTRY_LOADING)) {
742 (*eip)->bei_state |= CACHE_ENTRY_LOADING;
745 bdb_cache_entryinfo_unlock( *eip );
748 rc = bdb_cache_entry_db_lock( bdb->bi_dbenv, locker, *eip, 0, 0, lock );
749 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
751 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
752 } else if ( rc == 0 ) {
755 u_int32_t locker2 = locker;
757 /* We don't wrap entire read operations in txn's, but
758 * we need our cache entry lock and any DB page locks
759 * to be associated, in order for deadlock detection
760 * to work properly. So if we need to read from the DB,
761 * we use a long-lived per-thread txn for this step.
764 rc = bdb_txn_get( op, bdb->bi_dbenv, <id, 0 );
766 locker2 = TXN_ID( ltid );
770 /* Give up original read lock, obtain write lock with
771 * (possibly) new locker ID.
774 rc = bdb_cache_entry_db_relock( bdb->bi_dbenv, locker2,
777 if ( rc == 0 && !ep) {
778 rc = bdb_id2entry( op->o_bd, ltid, id, &ep );
781 ep->e_private = *eip;
786 #ifdef SLAP_ZONE_ALLOC
787 (*eip)->bei_zseq = *((ber_len_t *)ep - 2);
791 (*eip)->bei_state ^= CACHE_ENTRY_LOADING;
793 /* If we succeeded, downgrade back to a readlock. */
794 rc = bdb_cache_entry_db_relock( bdb->bi_dbenv, locker,
797 /* Otherwise, release the lock. */
798 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
800 if ( locker2 != locker ) {
801 /* If we're using the per-thread txn, release all
802 * of its page locks now.
805 list.op = DB_LOCK_PUT_ALL;
807 bdb->bi_dbenv->lock_vec( bdb->bi_dbenv, locker2,
809 /* If this txn was deadlocked, we must abort it
810 * and invalidate this per-thread txn.
812 if ( rc == DB_LOCK_DEADLOCK ) {
813 bdb_txn_get( op, bdb->bi_dbenv, <id, 1 );
816 } else if ( !(*eip)->bei_e ) {
817 /* Some other thread is trying to load the entry,
818 * give it a chance to finish.
820 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
821 ldap_pvt_thread_yield();
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->bi_dbenv,
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->bi_dbenv,
837 locker, *eip, 0, 0, lock );
846 bdb_cache_entryinfo_unlock( *eip );
849 #ifdef SLAP_ZONE_ALLOC
850 bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
852 bdb_entry_return( ep );
858 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
859 bdb->bi_cache.c_cursize++;
860 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
863 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_mutex );
865 /* If the LRU list has only one entry and this is it, it
866 * doesn't need to be added again.
868 if ( bdb->bi_cache.c_lruhead == bdb->bi_cache.c_lrutail &&
869 bdb->bi_cache.c_lruhead == *eip ) {
870 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_mutex );
873 /* if entry is on LRU list, remove from old spot */
874 if ( (*eip)->bei_lrunext || (*eip)->bei_lruprev ) {
875 LRU_DELETE( &bdb->bi_cache, *eip );
877 /* lru_mutex is unlocked for us */
878 bdb_cache_lru_add( bdb, locker, *eip );
882 #ifdef SLAP_ZONE_ALLOC
883 if (rc == 0 && (*eip)->bei_e) {
884 slap_zn_rlock(bdb->bi_cache.c_zctx, (*eip)->bei_e);
886 slap_zh_runlock(bdb->bi_cache.c_zctx);
899 if ( BEI(e)->bei_kids ) {
902 if ( BEI(e)->bei_state & CACHE_ENTRY_NO_KIDS ) {
905 rc = bdb_dn2id_children( op, txn, e );
906 if ( rc == DB_NOTFOUND ) {
907 BEI(e)->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
912 /* Update the cache after a successful database Add. */
915 struct bdb_info *bdb,
925 struct berval rdn = e->e_name;
933 /* Lock this entry so that bdb_add can run to completion.
934 * It can only fail if BDB has run out of lock resources.
936 rc = bdb_cache_entry_db_lock( bdb->bi_dbenv, locker, &ei, 1, 0, &lock );
938 bdb_cache_entryinfo_unlock( eip );
943 if ( nrdn->bv_len != e->e_nname.bv_len ) {
944 char *ptr = strchr( rdn.bv_val, ',' );
945 rdn.bv_len = ptr - rdn.bv_val;
947 ber_dupbv( &ei.bei_rdn, &rdn );
948 if ( eip->bei_dkids ) eip->bei_dkids++;
951 rc = bdb_entryinfo_add_internal( bdb, &ei, &new );
952 /* bdb_csn_commit can cause this when adding the database root entry */
954 new->bei_e->e_private = NULL;
955 #ifdef SLAP_ZONE_ALLOC
956 bdb_entry_return( bdb, new->bei_e, new->bei_zseq );
958 bdb_entry_return( new->bei_e );
963 new->bei_state = CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
964 eip->bei_state &= ~CACHE_ENTRY_NO_KIDS;
965 if (eip->bei_parent) {
966 eip->bei_parent->bei_state &= ~CACHE_ENTRY_NO_GRANDKIDS;
968 bdb_cache_entryinfo_unlock( eip );
970 ++bdb->bi_cache.c_cursize;
971 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
974 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_mutex );
976 /* lru_mutex is unlocked for us */
977 bdb_cache_lru_add( bdb, locker, new );
990 EntryInfo *ei = BEI(e);
992 /* Get write lock on data */
993 rc = bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock );
995 /* If we've done repeated mods on a cached entry, then e_attrs
996 * is no longer contiguous with the entry, and must be freed.
999 if ( (void *)e->e_attrs != (void *)(e+1) ) {
1000 attrs_free( e->e_attrs );
1002 e->e_attrs = newAttrs;
1008 * Change the rdn in the entryinfo. Also move to a new parent if needed.
1012 struct bdb_info *bdb,
1014 struct berval *nrdn,
1020 EntryInfo *ei = BEI(e), *pei;
1026 /* Get write lock on data */
1027 rc = bdb_cache_entry_db_relock( bdb->bi_dbenv, locker, ei, 1, 0, lock );
1028 if ( rc ) return rc;
1030 /* If we've done repeated mods on a cached entry, then e_attrs
1031 * is no longer contiguous with the entry, and must be freed.
1033 if ( (void *)e->e_attrs != (void *)(e+1) ) {
1034 attrs_free( e->e_attrs );
1036 e->e_attrs = new->e_attrs;
1037 if( e->e_nname.bv_val < e->e_bv.bv_val ||
1038 e->e_nname.bv_val > e->e_bv.bv_val + e->e_bv.bv_len )
1040 ch_free(e->e_name.bv_val);
1041 ch_free(e->e_nname.bv_val);
1043 e->e_name = new->e_name;
1044 e->e_nname = new->e_nname;
1046 /* Lock the parent's kids AVL tree */
1047 pei = ei->bei_parent;
1048 bdb_cache_entryinfo_lock( pei );
1049 avl_delete( &pei->bei_kids, (caddr_t) ei, bdb_rdn_cmp );
1050 free( ei->bei_nrdn.bv_val );
1051 ber_dupbv( &ei->bei_nrdn, nrdn );
1053 free( ei->bei_rdn.bv_val );
1056 if ( nrdn->bv_len != e->e_nname.bv_len ) {
1057 char *ptr = strchr(rdn.bv_val, ',');
1058 rdn.bv_len = ptr - rdn.bv_val;
1060 ber_dupbv( &ei->bei_rdn, &rdn );
1064 ein = ei->bei_parent;
1066 ei->bei_parent = ein;
1067 bdb_cache_entryinfo_unlock( pei );
1068 bdb_cache_entryinfo_lock( ein );
1072 /* Record the generation number of this change */
1073 ldap_pvt_thread_mutex_lock( &bdb->bi_modrdns_mutex );
1075 ei->bei_modrdns = bdb->bi_modrdns;
1076 ldap_pvt_thread_mutex_unlock( &bdb->bi_modrdns_mutex );
1079 avl_insert( &ein->bei_kids, ei, bdb_rdn_cmp, avl_dup_error );
1080 bdb_cache_entryinfo_unlock( ein );
1084 * cache_delete - delete the entry e from the cache.
1086 * returns: 0 e was deleted ok
1087 * 1 e was not in the cache
1088 * -1 something bad happened
1098 EntryInfo *ei = BEI(e);
1101 assert( e->e_private != NULL );
1103 /* Set this early, warn off any queriers */
1104 ei->bei_state |= CACHE_ENTRY_DELETED;
1106 /* Lock the entry's info */
1107 bdb_cache_entryinfo_lock( ei );
1109 /* Get write lock on the data */
1110 rc = bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock );
1112 /* couldn't lock, undo and give up */
1113 ei->bei_state ^= CACHE_ENTRY_DELETED;
1114 bdb_cache_entryinfo_unlock( ei );
1118 Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_delete( %ld )\n",
1122 ldap_pvt_thread_mutex_lock( &cache->lru_mutex );
1124 /* set cache write lock */
1125 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
1127 rc = bdb_cache_delete_internal( cache, e->e_private, 1 );
1129 /* free cache write lock */
1130 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
1132 /* free lru mutex */
1133 ldap_pvt_thread_mutex_unlock( &cache->lru_mutex );
1135 /* Leave entry info locked */
1141 bdb_cache_delete_cleanup(
1146 ei->bei_e->e_private = NULL;
1147 #ifdef SLAP_ZONE_ALLOC
1148 bdb_entry_return( ei->bei_bdb, ei->bei_e, ei->bei_zseq );
1150 bdb_entry_return( ei->bei_e );
1155 free( ei->bei_nrdn.bv_val );
1156 ei->bei_nrdn.bv_val = NULL;
1158 free( ei->bei_rdn.bv_val );
1159 ei->bei_rdn.bv_val = NULL;
1160 ei->bei_modrdns = 0;
1164 ei->bei_parent = NULL;
1165 ei->bei_kids = NULL;
1166 ei->bei_lruprev = NULL;
1168 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
1169 ei->bei_lrunext = cache->c_eifree;
1170 cache->c_eifree = ei;
1171 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
1172 bdb_cache_entryinfo_unlock( ei );
1176 bdb_cache_delete_internal(
1181 int rc = 0; /* return code */
1183 /* Lock the parent's kids tree */
1184 bdb_cache_entryinfo_lock( e->bei_parent );
1187 e->bei_parent->bei_ckids--;
1188 if ( decr && e->bei_parent->bei_dkids ) e->bei_parent->bei_dkids--;
1191 if ( avl_delete( &e->bei_parent->bei_kids, (caddr_t) e, bdb_rdn_cmp )
1196 if ( e->bei_parent->bei_kids )
1200 if ( avl_delete( &cache->c_idtree, (caddr_t) e, bdb_id_cmp ) == NULL ) {
1208 LRU_DELETE( cache, e );
1209 if ( e->bei_e ) cache->c_cursize--;
1212 bdb_cache_entryinfo_unlock( e->bei_parent );
1218 bdb_entryinfo_release( void *data )
1220 EntryInfo *ei = (EntryInfo *)data;
1221 if ( ei->bei_kids ) {
1222 avl_free( ei->bei_kids, NULL );
1225 ei->bei_e->e_private = NULL;
1226 #ifdef SLAP_ZONE_ALLOC
1227 bdb_entry_return( ei->bei_bdb, ei->bei_e, ei->bei_zseq );
1229 bdb_entry_return( ei->bei_e );
1232 bdb_cache_entryinfo_destroy( ei );
1236 bdb_cache_release_all( Cache *cache )
1238 /* set cache write lock */
1239 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
1241 ldap_pvt_thread_mutex_lock( &cache->lru_mutex );
1243 Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_release_all\n", 0, 0, 0 );
1245 avl_free( cache->c_dntree.bei_kids, NULL );
1246 avl_free( cache->c_idtree, bdb_entryinfo_release );
1247 for (;cache->c_eifree;cache->c_eifree = cache->c_lruhead) {
1248 cache->c_lruhead = cache->c_eifree->bei_lrunext;
1249 bdb_cache_entryinfo_destroy(cache->c_eifree);
1251 cache->c_cursize = 0;
1252 cache->c_eiused = 0;
1253 cache->c_leaves = 0;
1254 cache->c_idtree = NULL;
1255 cache->c_lruhead = NULL;
1256 cache->c_lrutail = NULL;
1257 cache->c_dntree.bei_kids = NULL;
1259 /* free lru mutex */
1260 ldap_pvt_thread_mutex_unlock( &cache->lru_mutex );
1261 /* free cache write lock */
1262 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
1267 bdb_lru_print( Cache *cache )
1271 fprintf( stderr, "LRU queue (head to tail):\n" );
1272 for ( e = cache->c_lruhead; e != NULL; e = e->bei_lrunext ) {
1273 fprintf( stderr, "\trdn \"%20s\" id %ld\n",
1274 e->bei_nrdn.bv_val, e->bei_id );
1276 fprintf( stderr, "LRU queue (tail to head):\n" );
1277 for ( e = cache->c_lrutail; e != NULL; e = e->bei_lruprev ) {
1278 fprintf( stderr, "\trdn \"%20s\" id %ld\n",
1279 e->bei_nrdn.bv_val, e->bei_id );
1285 bdb_txn_free( void *key, void *data )
1291 /* Obtain a long-lived transaction for the current thread.
1292 * If reset == 1, remove the current transaction. */
1294 bdb_txn_get( Operation *op, DB_ENV *env, DB_TXN **txn, int reset )
1297 void *ctx, *data = NULL;
1299 if ( slapMode & SLAP_TOOL_MODE ) {
1304 /* If no op was provided, try to find the ctx anyway... */
1306 ctx = op->o_threadctx;
1308 ctx = ldap_pvt_thread_pool_context();
1311 /* Shouldn't happen unless we're single-threaded */
1319 return ldap_pvt_thread_pool_setkey( ctx, ((char *)env)+1, NULL, NULL );
1322 if ( ldap_pvt_thread_pool_getkey( ctx, ((char *)env)+1, &data, NULL ) ||
1324 for ( i=0, rc=1; rc != 0 && i<4; i++ ) {
1325 rc = TXN_BEGIN( env, NULL, txn, READ_TXN_FLAG );
1326 #if DB_VERSION_FULL == 0x04020034
1327 if ( rc == EINVAL && READ_TXN_FLAG ) {
1329 Debug( LDAP_DEBUG_ANY,
1330 "bdb_txn_get: BerkeleyDB 4.2.52 library needs TXN patch!\n",
1336 if (rc) ldap_pvt_thread_yield();
1341 if ( ( rc = ldap_pvt_thread_pool_setkey( ctx, ((char *)env)+1,
1342 *txn, bdb_txn_free ) ) ) {
1344 Debug( LDAP_DEBUG_ANY, "bdb_txn_get: err %s(%d)\n",
1345 db_strerror(rc), rc, 0 );
1355 #ifdef BDB_REUSE_LOCKERS
1357 bdb_locker_id_free( void *key, void *data )
1360 u_int32_t lockid = (long)data;
1363 rc = XLOCK_ID_FREE( env, lockid );
1364 if ( rc == EINVAL ) {
1366 Debug( LDAP_DEBUG_ANY,
1367 "bdb_locker_id_free: %lu err %s(%d)\n",
1368 (unsigned long) lockid, db_strerror(rc), rc );
1369 /* release all locks held by this locker. */
1370 lr.op = DB_LOCK_PUT_ALL;
1372 env->lock_vec( env, lockid, 0, &lr, 1, NULL );
1373 XLOCK_ID_FREE( env, lockid );
1378 bdb_locker_id( Operation *op, DB_ENV *env, u_int32_t *locker )
1385 if ( !env || !locker ) return -1;
1387 /* If no op was provided, try to find the ctx anyway... */
1389 ctx = op->o_threadctx;
1391 ctx = ldap_pvt_thread_pool_context();
1394 /* Shouldn't happen unless we're single-threaded */
1400 if ( ldap_pvt_thread_pool_getkey( ctx, env, &data, NULL ) ) {
1401 for ( i=0, rc=1; rc != 0 && i<4; i++ ) {
1402 rc = XLOCK_ID( env, &lockid );
1403 if (rc) ldap_pvt_thread_yield();
1408 data = (void *)((long)lockid);
1409 if ( ( rc = ldap_pvt_thread_pool_setkey( ctx, env,
1410 data, bdb_locker_id_free ) ) ) {
1411 XLOCK_ID_FREE( env, lockid );
1412 Debug( LDAP_DEBUG_ANY, "bdb_locker_id: err %s(%d)\n",
1413 db_strerror(rc), rc, 0 );
1418 lockid = (long)data;
1423 #endif /* BDB_REUSE_LOCKERS */
1426 bdb_cache_delete_entry(
1427 struct bdb_info *bdb,
1432 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
1433 if ( bdb_cache_entry_db_lock( bdb->bi_dbenv, bdb->bi_cache.c_locker, ei, 1, 1, lock ) == 0 )
1435 if ( ei->bei_e && !(ei->bei_state & CACHE_ENTRY_NOT_LINKED )) {
1436 LRU_DELETE( &bdb->bi_cache, ei );
1437 ei->bei_e->e_private = NULL;
1438 #ifdef SLAP_ZONE_ALLOC
1439 bdb_entry_return( bdb, ei->bei_e, ei->bei_zseq );
1441 bdb_entry_return( ei->bei_e );
1444 --bdb->bi_cache.c_cursize;
1446 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
1448 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );