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_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 #if DB_VERSION_FULL >= 0x04060012
159 #define BDB_LOCKID(locker) locker->id
161 #define BDB_LOCKID(locker) locker
164 /* Atomically release and reacquire a lock */
166 bdb_cache_entry_db_relock(
167 struct bdb_info *bdb,
181 if ( !lock ) return 0;
183 lockobj.data = &ei->bei_id;
184 lockobj.size = sizeof(ei->bei_id) + 1;
186 list[0].op = DB_LOCK_PUT;
187 list[0].lock = *lock;
188 list[1].op = DB_LOCK_GET;
189 list[1].lock = *lock;
190 list[1].mode = rw ? DB_LOCK_WRITE : DB_LOCK_READ;
191 list[1].obj = &lockobj;
192 rc = bdb->bi_dbenv->lock_vec(bdb->bi_dbenv, BDB_LOCKID(locker), tryOnly ? DB_LOCK_NOWAIT : 0,
195 if (rc && !tryOnly) {
196 Debug( LDAP_DEBUG_TRACE,
197 "bdb_cache_entry_db_relock: entry %ld, rw %d, rc %d\n",
198 ei->bei_id, rw, rc );
200 *lock = list[1].lock;
207 bdb_cache_entry_db_lock( struct bdb_info *bdb, BDB_LOCKER locker, EntryInfo *ei,
208 int rw, int tryOnly, DB_LOCK *lock )
217 if ( !lock ) return 0;
220 db_rw = DB_LOCK_WRITE;
222 db_rw = DB_LOCK_READ;
224 lockobj.data = &ei->bei_id;
225 lockobj.size = sizeof(ei->bei_id) + 1;
227 rc = LOCK_GET(bdb->bi_dbenv, BDB_LOCKID(locker), tryOnly ? DB_LOCK_NOWAIT : 0,
228 &lockobj, db_rw, lock);
229 if (rc && !tryOnly) {
230 Debug( LDAP_DEBUG_TRACE,
231 "bdb_cache_entry_db_lock: entry %ld, rw %d, rc %d\n",
232 ei->bei_id, rw, rc );
235 #endif /* NO_DB_LOCK */
239 bdb_cache_entry_db_unlock ( struct bdb_info *bdb, DB_LOCK *lock )
246 if ( !lock || lock->mode == DB_LOCK_NG ) return 0;
248 rc = LOCK_PUT ( bdb->bi_dbenv, lock );
254 bdb_cache_return_entry_rw( struct bdb_info *bdb, Entry *e,
255 int rw, DB_LOCK *lock )
260 bdb_cache_entry_db_unlock( bdb, lock );
262 bdb_cache_entryinfo_lock( ei );
263 if ( ei->bei_state & CACHE_ENTRY_NOT_CACHED ) {
265 ei->bei_state ^= CACHE_ENTRY_NOT_CACHED;
268 bdb_cache_entryinfo_unlock( ei );
271 bdb_entry_return( e );
276 bdb_cache_entryinfo_destroy( EntryInfo *e )
278 ldap_pvt_thread_mutex_destroy( &e->bei_kids_mutex );
279 free( e->bei_nrdn.bv_val );
281 free( e->bei_rdn.bv_val );
287 /* Do a length-ordered sort on normalized RDNs */
289 bdb_rdn_cmp( const void *v_e1, const void *v_e2 )
291 const EntryInfo *e1 = v_e1, *e2 = v_e2;
292 int rc = e1->bei_nrdn.bv_len - e2->bei_nrdn.bv_len;
294 rc = strncmp( e1->bei_nrdn.bv_val, e2->bei_nrdn.bv_val,
295 e1->bei_nrdn.bv_len );
301 bdb_id_cmp( const void *v_e1, const void *v_e2 )
303 const EntryInfo *e1 = v_e1, *e2 = v_e2;
304 return e1->bei_id - e2->bei_id;
308 bdb_id_dup_err( void *v1, void *v2 )
311 e2->bei_lrunext = v1;
315 /* Create an entryinfo in the cache. Caller must release the locks later.
318 bdb_entryinfo_add_internal(
319 struct bdb_info *bdb,
323 EntryInfo *ei2 = NULL;
327 ei2 = bdb_cache_entryinfo_new( &bdb->bi_cache );
329 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
330 bdb_cache_entryinfo_lock( ei->bei_parent );
332 ei2->bei_id = ei->bei_id;
333 ei2->bei_parent = ei->bei_parent;
335 ei2->bei_rdn = ei->bei_rdn;
337 #ifdef SLAP_ZONE_ALLOC
341 /* Add to cache ID tree */
342 if (avl_insert( &bdb->bi_cache.c_idtree, ei2, bdb_id_cmp,
344 EntryInfo *eix = ei2->bei_lrunext;
345 bdb_cache_entryinfo_free( &bdb->bi_cache, ei2 );
348 /* It got freed above because its value was
351 ei->bei_rdn.bv_val = NULL;
356 bdb->bi_cache.c_eiused++;
357 ber_dupbv( &ei2->bei_nrdn, &ei->bei_nrdn );
359 /* This is a new leaf node. But if parent had no kids, then it was
360 * a leaf and we would be decrementing that. So, only increment if
361 * the parent already has kids.
363 if ( ei->bei_parent->bei_kids || !ei->bei_parent->bei_id )
364 bdb->bi_cache.c_leaves++;
365 rc = avl_insert( &ei->bei_parent->bei_kids, ei2, bdb_rdn_cmp,
367 #if defined(LDAP_DEBUG) && defined(LDAP_DEVEL)
369 bdb->bi_dbenv->log_flush( bdb->bi_dbenv, NULL );
374 ei->bei_parent->bei_ckids++;
382 /* Find the EntryInfo for the requested DN. If the DN cannot be found, return
383 * the info for its closest ancestor. *res should be NULL to process a
384 * complete DN starting from the tree root. Otherwise *res must be the
385 * immediate parent of the requested DN, and only the RDN will be searched.
386 * The EntryInfo is locked upon return and must be unlocked by the caller.
395 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
396 EntryInfo ei, *eip, *ei2;
400 /* this function is always called with normalized DN */
402 /* we're doing a onelevel search for an RDN */
403 ei.bei_nrdn.bv_val = ndn->bv_val;
404 ei.bei_nrdn.bv_len = dn_rdnlen( op->o_bd, ndn );
407 /* we're searching a full DN from the root */
408 ptr = ndn->bv_val + ndn->bv_len - op->o_bd->be_nsuffix[0].bv_len;
409 ei.bei_nrdn.bv_val = ptr;
410 ei.bei_nrdn.bv_len = op->o_bd->be_nsuffix[0].bv_len;
411 /* Skip to next rdn if suffix is empty */
412 if ( ei.bei_nrdn.bv_len == 0 ) {
413 for (ptr = ei.bei_nrdn.bv_val - 2; ptr > ndn->bv_val
414 && !DN_SEPARATOR(*ptr); ptr--) /* empty */;
415 if ( ptr >= ndn->bv_val ) {
416 if (DN_SEPARATOR(*ptr)) ptr++;
417 ei.bei_nrdn.bv_len = ei.bei_nrdn.bv_val - ptr;
418 ei.bei_nrdn.bv_val = ptr;
421 eip = &bdb->bi_cache.c_dntree;
424 for ( bdb_cache_entryinfo_lock( eip ); eip; ) {
425 eip->bei_state |= CACHE_ENTRY_REFERENCED;
427 ei2 = (EntryInfo *)avl_find( eip->bei_kids, &ei, bdb_rdn_cmp );
429 int len = ei.bei_nrdn.bv_len;
431 if ( BER_BVISEMPTY( ndn )) {
436 ei.bei_nrdn.bv_len = ndn->bv_len -
437 (ei.bei_nrdn.bv_val - ndn->bv_val);
438 bdb_cache_entryinfo_unlock( eip );
440 #if defined(LDAP_DEBUG) && defined(LDAP_DEVEL)
441 bdb->bi_dbenv->log_printf( bdb->bi_dbenv, NULL, "Reading %s",
442 ei.bei_nrdn.bv_val );
445 rc = bdb_dn2id( op, locker, &ei.bei_nrdn, &ei );
447 bdb_cache_entryinfo_lock( eip );
452 #if defined(LDAP_DEBUG) && defined(LDAP_DEVEL)
453 bdb->bi_dbenv->log_printf( bdb->bi_dbenv, NULL, "Read got %s(%d)",
454 ei.bei_nrdn.bv_val, ei.bei_id );
457 /* DN exists but needs to be added to cache */
458 ei.bei_nrdn.bv_len = len;
459 rc = bdb_entryinfo_add_internal( bdb, &ei, &ei2 );
460 Debug( LDAP_DEBUG_TRACE, "add_internal: \"%s\" %d\n",
461 ei.bei_nrdn.bv_val, ei.bei_id, 0 );
462 /* add_internal left eip and c_rwlock locked */
463 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
468 } else if ( ei2->bei_state & CACHE_ENTRY_DELETED ) {
469 /* In the midst of deleting? Give it a chance to
472 bdb_cache_entryinfo_unlock( eip );
473 ldap_pvt_thread_yield();
474 bdb_cache_entryinfo_lock( eip );
478 bdb_cache_entryinfo_unlock( eip );
479 bdb_cache_entryinfo_lock( ei2 );
483 /* Advance to next lower RDN */
484 for (ptr = ei.bei_nrdn.bv_val - 2; ptr > ndn->bv_val
485 && !DN_SEPARATOR(*ptr); ptr--) /* empty */;
486 if ( ptr >= ndn->bv_val ) {
487 if (DN_SEPARATOR(*ptr)) ptr++;
488 ei.bei_nrdn.bv_len = ei.bei_nrdn.bv_val - ptr - 1;
489 ei.bei_nrdn.bv_val = ptr;
491 if ( ptr < ndn->bv_val ) {
501 /* Walk up the tree from a child node, looking for an ID that's already
502 * been linked into the cache.
505 hdb_cache_find_parent(
511 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
512 EntryInfo ei, eip, *ei2 = NULL, *ein = NULL, *eir = NULL;
520 rc = hdb_dn2id_parent( op, locker, &ei, &eip.bei_id );
523 /* Save the previous node, if any */
526 /* Create a new node for the current ID */
527 ein = bdb_cache_entryinfo_new( &bdb->bi_cache );
528 ein->bei_id = ei.bei_id;
529 ein->bei_kids = ei.bei_kids;
530 ein->bei_nrdn = ei.bei_nrdn;
531 ein->bei_rdn = ei.bei_rdn;
532 ein->bei_ckids = ei.bei_ckids;
533 #ifdef SLAP_ZONE_ALLOC
538 /* This node is not fully connected yet */
539 ein->bei_state |= CACHE_ENTRY_NOT_LINKED;
541 /* Insert this node into the ID tree */
542 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
543 if ( avl_insert( &bdb->bi_cache.c_idtree, (caddr_t)ein,
544 bdb_id_cmp, bdb_id_dup_err ) ) {
545 EntryInfo *eix = ein->bei_lrunext;
547 /* Someone else created this node just before us.
548 * Free our new copy and use the existing one.
550 bdb_cache_entryinfo_free( &bdb->bi_cache, ein );
553 /* Link in any kids we've already processed */
555 bdb_cache_entryinfo_lock( ein );
556 avl_insert( &ein->bei_kids, (caddr_t)ei2,
557 bdb_rdn_cmp, avl_dup_error );
559 bdb_cache_entryinfo_unlock( ein );
563 /* If this is the first time, save this node
564 * to be returned later.
566 if ( eir == NULL ) eir = ein;
568 /* If there was a previous node, link it to this one */
569 if ( ei2 ) ei2->bei_parent = ein;
571 /* Look for this node's parent */
573 ei2 = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
574 (caddr_t) &eip, bdb_id_cmp );
576 ei2 = &bdb->bi_cache.c_dntree;
578 bdb->bi_cache.c_eiused++;
579 if ( ei2 && ( ei2->bei_kids || !ei2->bei_id ))
580 bdb->bi_cache.c_leaves++;
581 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
583 /* Got the parent, link in and we're done. */
585 bdb_cache_entryinfo_lock( eir );
586 bdb_cache_entryinfo_lock( ei2 );
587 ein->bei_parent = ei2;
589 avl_insert( &ei2->bei_kids, (caddr_t)ein, bdb_rdn_cmp,
593 /* Reset all the state info */
594 for (ein = eir; ein != ei2; ein=ein->bei_parent)
595 ein->bei_state &= ~CACHE_ENTRY_NOT_LINKED;
597 bdb_cache_entryinfo_unlock( ei2 );
603 ei.bei_id = eip.bei_id;
605 avl_insert( &ei.bei_kids, (caddr_t)ein, bdb_rdn_cmp,
611 /* Used by hdb_dn2idl when loading the EntryInfo for all the children
615 struct bdb_info *bdb,
622 /* See if we already have this one */
623 bdb_cache_entryinfo_lock( ei->bei_parent );
624 ei2 = (EntryInfo *)avl_find( ei->bei_parent->bei_kids, ei, bdb_rdn_cmp );
625 bdb_cache_entryinfo_unlock( ei->bei_parent );
628 /* Not found, add it */
631 /* bei_rdn was not malloc'd before, do it now */
632 ber_dupbv( &bv, &ei->bei_rdn );
635 rc = bdb_entryinfo_add_internal( bdb, ei, res );
636 bdb_cache_entryinfo_unlock( ei->bei_parent );
637 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
639 /* Found, return it */
647 /* This is best-effort only. If all entries in the cache are
648 * busy, they will all be kept. This is unlikely to happen
649 * unless the cache is very much smaller than the working set.
652 bdb_cache_lru_purge( struct bdb_info *bdb )
654 DB_LOCK lock, *lockp;
655 EntryInfo *elru, *elnext = NULL;
656 int count, islocked, eimax;
658 /* Wait for the mutex; we're the only one trying to purge. */
659 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_lru_mutex );
661 if ( bdb->bi_cache.c_cursize <= bdb->bi_cache.c_maxsize ) {
662 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_lru_mutex );
663 bdb->bi_cache.c_purging = 0;
667 if ( bdb->bi_cache.c_locker ) {
675 /* maximum number of EntryInfo leaves to cache. In slapcat
676 * we always free all leaf nodes.
678 if ( slapMode & SLAP_TOOL_READONLY )
681 eimax = bdb->bi_cache.c_eimax;
683 /* Look for an unused entry to remove */
684 for ( elru = bdb->bi_cache.c_lruhead; elru; elru = elnext ) {
685 elnext = elru->bei_lrunext;
687 if ( bdb_cache_entryinfo_trylock( elru ))
690 /* This flag implements the clock replacement behavior */
691 if ( elru->bei_state & ( CACHE_ENTRY_REFERENCED )) {
692 elru->bei_state &= ~CACHE_ENTRY_REFERENCED;
693 bdb_cache_entryinfo_unlock( elru );
697 /* If this node is in the process of linking into the cache,
698 * or this node is being deleted, skip it.
700 if (( elru->bei_state & ( CACHE_ENTRY_NOT_LINKED |
701 CACHE_ENTRY_DELETED | CACHE_ENTRY_LOADING )) ||
702 elru->bei_finders > 0 ) {
703 bdb_cache_entryinfo_unlock( elru );
707 /* entryinfo is locked */
710 /* If we can successfully writelock it, then
711 * the object is idle.
713 if ( bdb_cache_entry_db_lock( bdb,
714 bdb->bi_cache.c_locker, elru, 1, 1, lockp ) == 0 ) {
716 /* Free entry for this node if it's present */
718 elru->bei_e->e_private = NULL;
719 #ifdef SLAP_ZONE_ALLOC
720 bdb_entry_return( bdb, elru->bei_e, elru->bei_zseq );
722 bdb_entry_return( elru->bei_e );
727 bdb_cache_entry_db_unlock( bdb, lockp );
730 * If it is a leaf node, and we're over the limit, free it.
732 if ( elru->bei_kids ) {
733 /* Drop from list, we ignore it... */
734 LRU_DEL( &bdb->bi_cache, elru );
735 } else if ( bdb->bi_cache.c_leaves > eimax ) {
736 /* Too many leaf nodes, free this one */
737 bdb_cache_delete_internal( &bdb->bi_cache, elru, 0 );
738 bdb_cache_delete_cleanup( &bdb->bi_cache, elru );
740 } /* Leave on list until we need to free it */
744 bdb_cache_entryinfo_unlock( elru );
746 if ( count >= bdb->bi_cache.c_minfree ) {
747 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_count_mutex );
748 bdb->bi_cache.c_cursize -= count;
749 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_count_mutex );
753 if ( elnext == bdb->bi_cache.c_lruhead )
757 bdb->bi_cache.c_lruhead = elnext;
758 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_lru_mutex );
759 bdb->bi_cache.c_purging = 0;
764 struct bdb_info *bdb,
767 EntryInfo ei = { 0 },
772 ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
773 ei2 = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
774 (caddr_t) &ei, bdb_id_cmp );
775 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
780 * cache_find_id - find an entry in the cache, given id.
781 * The entry is locked for Read upon return. Call with flag ID_LOCKED if
782 * the supplied *eip was already locked.
795 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
797 int rc = 0, load = 0;
798 EntryInfo ei = { 0 };
802 #ifdef SLAP_ZONE_ALLOC
803 slap_zh_rlock(bdb->bi_cache.c_zctx);
805 /* If we weren't given any info, see if we have it already cached */
807 again: ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
808 *eip = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
809 (caddr_t) &ei, bdb_id_cmp );
811 /* If the lock attempt fails, the info is in use */
812 if ( bdb_cache_entryinfo_trylock( *eip )) {
813 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
814 /* If this node is being deleted, treat
815 * as if the delete has already finished
817 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
820 /* otherwise, wait for the info to free up */
821 ldap_pvt_thread_yield();
824 /* If this info isn't hooked up to its parent yet,
825 * unlock and wait for it to be fully initialized
827 if ( (*eip)->bei_state & CACHE_ENTRY_NOT_LINKED ) {
828 bdb_cache_entryinfo_unlock( *eip );
829 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
830 ldap_pvt_thread_yield();
835 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
838 /* See if the ID exists in the database; add it to the cache if so */
841 rc = bdb_id2entry( op->o_bd, tid, locker, id, &ep );
843 rc = bdb_cache_find_ndn( op, locker,
845 if ( *eip ) flag |= ID_LOCKED;
847 ep->e_private = NULL;
848 #ifdef SLAP_ZONE_ALLOC
849 bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
851 bdb_entry_return( ep );
857 rc = hdb_cache_find_parent(op, locker, id, eip );
858 if ( rc == 0 ) flag |= ID_LOCKED;
862 /* Ok, we found the info, do we have the entry? */
864 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
867 (*eip)->bei_finders++;
868 (*eip)->bei_state |= CACHE_ENTRY_REFERENCED;
869 /* Make sure only one thread tries to load the entry */
871 #ifdef SLAP_ZONE_ALLOC
872 if ((*eip)->bei_e && !slap_zn_validate(
873 bdb->bi_cache.c_zctx, (*eip)->bei_e, (*eip)->bei_zseq)) {
874 (*eip)->bei_e = NULL;
875 (*eip)->bei_zseq = 0;
878 if ( !(*eip)->bei_e && !((*eip)->bei_state & CACHE_ENTRY_LOADING)) {
880 (*eip)->bei_state |= CACHE_ENTRY_LOADING;
883 /* If the entry was loaded before but uncached, and we need
884 * it again, clear the uncached state
886 if ( (*eip)->bei_state & CACHE_ENTRY_NOT_CACHED ) {
887 (*eip)->bei_state ^= CACHE_ENTRY_NOT_CACHED;
888 if ( flag & ID_NOCACHE )
892 if ( flag & ID_LOCKED ) {
893 bdb_cache_entryinfo_unlock( *eip );
896 rc = bdb_cache_entry_db_lock( bdb, locker, *eip, load, 0, lock );
897 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
899 bdb_cache_entry_db_unlock( bdb, lock );
900 } else if ( rc == 0 ) {
903 rc = bdb_id2entry( op->o_bd, tid, locker, id, &ep );
906 ep->e_private = *eip;
911 #ifdef SLAP_ZONE_ALLOC
912 (*eip)->bei_zseq = *((ber_len_t *)ep - 2);
915 bdb_cache_lru_link( bdb, *eip );
916 if ( flag & ID_NOCACHE ) {
917 bdb_cache_entryinfo_lock( *eip );
918 (*eip)->bei_state |= CACHE_ENTRY_NOT_CACHED;
919 bdb_cache_entryinfo_unlock( *eip );
923 /* If we succeeded, downgrade back to a readlock. */
924 rc = bdb_cache_entry_db_relock( bdb, locker,
927 /* Otherwise, release the lock. */
928 bdb_cache_entry_db_unlock( bdb, lock );
930 } else if ( !(*eip)->bei_e ) {
931 /* Some other thread is trying to load the entry,
932 * wait for it to finish.
934 bdb_cache_entry_db_unlock( bdb, lock );
935 bdb_cache_entryinfo_lock( *eip );
940 /* Check for subtree renames
942 rc = bdb_fix_dn( (*eip)->bei_e, 1 );
944 bdb_cache_entry_db_relock( bdb,
945 locker, *eip, 1, 0, lock );
946 /* check again in case other modifier did it already */
947 if ( bdb_fix_dn( (*eip)->bei_e, 1 ) )
948 rc = bdb_fix_dn( (*eip)->bei_e, 2 );
949 bdb_cache_entry_db_relock( bdb,
950 locker, *eip, 0, 0, lock );
954 bdb_cache_entryinfo_lock( *eip );
955 (*eip)->bei_finders--;
957 (*eip)->bei_state ^= CACHE_ENTRY_LOADING;
958 bdb_cache_entryinfo_unlock( *eip );
962 if ( flag & ID_LOCKED ) {
963 bdb_cache_entryinfo_unlock( *eip );
966 ep->e_private = NULL;
967 #ifdef SLAP_ZONE_ALLOC
968 bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
970 bdb_entry_return( ep );
977 if ( !( flag & ID_NOCACHE )) {
978 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_count_mutex );
979 bdb->bi_cache.c_cursize++;
980 if ( bdb->bi_cache.c_cursize > bdb->bi_cache.c_maxsize &&
981 !bdb->bi_cache.c_purging ) {
983 bdb->bi_cache.c_purging = 1;
985 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_count_mutex );
989 bdb_cache_lru_purge( bdb );
992 #ifdef SLAP_ZONE_ALLOC
993 if (rc == 0 && (*eip)->bei_e) {
994 slap_zn_rlock(bdb->bi_cache.c_zctx, (*eip)->bei_e);
996 slap_zh_runlock(bdb->bi_cache.c_zctx);
1009 if ( BEI(e)->bei_kids ) {
1012 if ( BEI(e)->bei_state & CACHE_ENTRY_NO_KIDS ) {
1015 rc = bdb_dn2id_children( op, txn, e );
1016 if ( rc == DB_NOTFOUND ) {
1017 BEI(e)->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
1022 /* Update the cache after a successful database Add. */
1025 struct bdb_info *bdb,
1028 struct berval *nrdn,
1035 struct berval rdn = e->e_name;
1038 ei.bei_id = e->e_id;
1039 ei.bei_parent = eip;
1040 ei.bei_nrdn = *nrdn;
1043 /* Lock this entry so that bdb_add can run to completion.
1044 * It can only fail if BDB has run out of lock resources.
1046 rc = bdb_cache_entry_db_lock( bdb, locker, &ei, 0, 0, lock );
1048 bdb_cache_entryinfo_unlock( eip );
1053 if ( nrdn->bv_len != e->e_nname.bv_len ) {
1054 char *ptr = ber_bvchr( &rdn, ',' );
1055 assert( ptr != NULL );
1056 rdn.bv_len = ptr - rdn.bv_val;
1058 ber_dupbv( &ei.bei_rdn, &rdn );
1059 if ( eip->bei_dkids ) eip->bei_dkids++;
1062 rc = bdb_entryinfo_add_internal( bdb, &ei, &new );
1063 /* bdb_csn_commit can cause this when adding the database root entry */
1065 new->bei_e->e_private = NULL;
1066 #ifdef SLAP_ZONE_ALLOC
1067 bdb_entry_return( bdb, new->bei_e, new->bei_zseq );
1069 bdb_entry_return( new->bei_e );
1074 new->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
1075 eip->bei_state &= ~CACHE_ENTRY_NO_KIDS;
1076 if (eip->bei_parent) {
1077 eip->bei_parent->bei_state &= ~CACHE_ENTRY_NO_GRANDKIDS;
1079 bdb_cache_entryinfo_unlock( eip );
1081 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
1082 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_count_mutex );
1083 ++bdb->bi_cache.c_cursize;
1084 if ( bdb->bi_cache.c_cursize > bdb->bi_cache.c_maxsize &&
1085 !bdb->bi_cache.c_purging ) {
1087 bdb->bi_cache.c_purging = 1;
1089 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_count_mutex );
1091 bdb_cache_lru_link( bdb, new );
1094 bdb_cache_lru_purge( bdb );
1101 struct bdb_info *bdb,
1103 Attribute *newAttrs,
1107 EntryInfo *ei = BEI(e);
1109 /* Get write lock on data */
1110 rc = bdb_cache_entry_db_relock( bdb, locker, ei, 1, 0, lock );
1112 /* If we've done repeated mods on a cached entry, then e_attrs
1113 * is no longer contiguous with the entry, and must be freed.
1116 if ( (void *)e->e_attrs != (void *)(e+1) ) {
1117 attrs_free( e->e_attrs );
1119 e->e_attrs = newAttrs;
1125 * Change the rdn in the entryinfo. Also move to a new parent if needed.
1129 struct bdb_info *bdb,
1131 struct berval *nrdn,
1137 EntryInfo *ei = BEI(e), *pei;
1143 /* Get write lock on data */
1144 rc = bdb_cache_entry_db_relock( bdb, locker, ei, 1, 0, lock );
1145 if ( rc ) return rc;
1147 /* If we've done repeated mods on a cached entry, then e_attrs
1148 * is no longer contiguous with the entry, and must be freed.
1150 if ( (void *)e->e_attrs != (void *)(e+1) ) {
1151 attrs_free( e->e_attrs );
1153 e->e_attrs = new->e_attrs;
1154 if( e->e_nname.bv_val < e->e_bv.bv_val ||
1155 e->e_nname.bv_val > e->e_bv.bv_val + e->e_bv.bv_len )
1157 ch_free(e->e_name.bv_val);
1158 ch_free(e->e_nname.bv_val);
1160 e->e_name = new->e_name;
1161 e->e_nname = new->e_nname;
1163 /* Lock the parent's kids AVL tree */
1164 pei = ei->bei_parent;
1165 bdb_cache_entryinfo_lock( pei );
1166 avl_delete( &pei->bei_kids, (caddr_t) ei, bdb_rdn_cmp );
1167 free( ei->bei_nrdn.bv_val );
1168 ber_dupbv( &ei->bei_nrdn, nrdn );
1171 free( ei->bei_rdn.bv_val );
1174 if ( nrdn->bv_len != e->e_nname.bv_len ) {
1175 char *ptr = ber_bvchr(&rdn, ',');
1176 assert( ptr != NULL );
1177 rdn.bv_len = ptr - rdn.bv_val;
1179 ber_dupbv( &ei->bei_rdn, &rdn );
1181 /* If new parent, decrement kid counts */
1184 if ( pei->bei_dkids ) {
1186 if ( pei->bei_dkids < 2 )
1187 pei->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
1193 ein = ei->bei_parent;
1195 ei->bei_parent = ein;
1196 bdb_cache_entryinfo_unlock( pei );
1197 bdb_cache_entryinfo_lock( ein );
1199 /* new parent now has kids */
1200 if ( ein->bei_state & CACHE_ENTRY_NO_KIDS )
1201 ein->bei_state ^= CACHE_ENTRY_NO_KIDS;
1202 /* grandparent has grandkids */
1203 if ( ein->bei_parent )
1204 ein->bei_parent->bei_state &= ~CACHE_ENTRY_NO_GRANDKIDS;
1206 /* parent might now have grandkids */
1207 if ( ein->bei_state & CACHE_ENTRY_NO_GRANDKIDS &&
1208 !(ei->bei_state & CACHE_ENTRY_NO_KIDS))
1209 ein->bei_state ^= CACHE_ENTRY_NO_GRANDKIDS;
1212 if ( ein->bei_dkids ) ein->bei_dkids++;
1217 /* Record the generation number of this change */
1218 ldap_pvt_thread_mutex_lock( &bdb->bi_modrdns_mutex );
1220 ei->bei_modrdns = bdb->bi_modrdns;
1221 ldap_pvt_thread_mutex_unlock( &bdb->bi_modrdns_mutex );
1224 avl_insert( &ein->bei_kids, ei, bdb_rdn_cmp, avl_dup_error );
1225 bdb_cache_entryinfo_unlock( ein );
1229 * cache_delete - delete the entry e from the cache.
1231 * returns: 0 e was deleted ok
1232 * 1 e was not in the cache
1233 * -1 something bad happened
1237 struct bdb_info *bdb,
1242 EntryInfo *ei = BEI(e);
1245 assert( e->e_private != NULL );
1247 /* Set this early, warn off any queriers */
1248 ei->bei_state |= CACHE_ENTRY_DELETED;
1250 /* Lock the entry's info */
1251 bdb_cache_entryinfo_lock( ei );
1253 /* Get write lock on the data */
1254 rc = bdb_cache_entry_db_relock( bdb, locker, ei, 1, 0, lock );
1256 /* couldn't lock, undo and give up */
1257 ei->bei_state ^= CACHE_ENTRY_DELETED;
1258 bdb_cache_entryinfo_unlock( ei );
1262 Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_delete( %ld )\n",
1266 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_lru_mutex );
1268 rc = bdb_cache_delete_internal( &bdb->bi_cache, e->e_private, 1 );
1270 /* free lru mutex */
1271 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_lru_mutex );
1273 /* Leave entry info locked */
1279 bdb_cache_delete_cleanup(
1284 ei->bei_e->e_private = NULL;
1285 #ifdef SLAP_ZONE_ALLOC
1286 bdb_entry_return( ei->bei_bdb, ei->bei_e, ei->bei_zseq );
1288 bdb_entry_return( ei->bei_e );
1293 bdb_cache_entryinfo_free( cache, ei );
1294 bdb_cache_entryinfo_unlock( ei );
1298 bdb_cache_delete_internal(
1303 int rc = 0; /* return code */
1306 /* Lock the parent's kids tree */
1307 bdb_cache_entryinfo_lock( e->bei_parent );
1310 e->bei_parent->bei_ckids--;
1311 if ( decr && e->bei_parent->bei_dkids ) e->bei_parent->bei_dkids--;
1314 if ( avl_delete( &e->bei_parent->bei_kids, (caddr_t) e, bdb_rdn_cmp )
1319 if ( e->bei_parent->bei_kids )
1322 bdb_cache_entryinfo_unlock( e->bei_parent );
1324 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
1326 if ( avl_delete( &cache->c_idtree, (caddr_t) e, bdb_id_cmp )) {
1333 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
1337 LRU_DEL( cache, e );
1340 ldap_pvt_thread_mutex_lock( &cache->c_count_mutex );
1342 ldap_pvt_thread_mutex_unlock( &cache->c_count_mutex );
1350 bdb_entryinfo_release( void *data )
1352 EntryInfo *ei = (EntryInfo *)data;
1353 if ( ei->bei_kids ) {
1354 avl_free( ei->bei_kids, NULL );
1357 ei->bei_e->e_private = NULL;
1358 #ifdef SLAP_ZONE_ALLOC
1359 bdb_entry_return( ei->bei_bdb, ei->bei_e, ei->bei_zseq );
1361 bdb_entry_return( ei->bei_e );
1364 bdb_cache_entryinfo_destroy( ei );
1368 bdb_cache_release_all( Cache *cache )
1370 /* set cache write lock */
1371 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
1373 ldap_pvt_thread_mutex_lock( &cache->c_lru_mutex );
1375 Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_release_all\n", 0, 0, 0 );
1377 avl_free( cache->c_dntree.bei_kids, NULL );
1378 avl_free( cache->c_idtree, bdb_entryinfo_release );
1379 for (;cache->c_eifree;cache->c_eifree = cache->c_lruhead) {
1380 cache->c_lruhead = cache->c_eifree->bei_lrunext;
1381 bdb_cache_entryinfo_destroy(cache->c_eifree);
1383 cache->c_cursize = 0;
1384 cache->c_eiused = 0;
1385 cache->c_leaves = 0;
1386 cache->c_idtree = NULL;
1387 cache->c_lruhead = NULL;
1388 cache->c_lrutail = NULL;
1389 cache->c_dntree.bei_kids = NULL;
1391 /* free lru mutex */
1392 ldap_pvt_thread_mutex_unlock( &cache->c_lru_mutex );
1393 /* free cache write lock */
1394 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
1400 bdb_lru_print( Cache *cache )
1404 fprintf( stderr, "LRU circle head: %p\n", (void *) cache->c_lruhead );
1405 fprintf( stderr, "LRU circle (tail forward):\n" );
1406 for ( e = cache->c_lrutail; ; ) {
1407 fprintf( stderr, "\t%p, %p id %ld rdn \"%s\"\n",
1408 (void *) e, (void *) e->bei_e, e->bei_id, e->bei_nrdn.bv_val );
1410 if ( e == cache->c_lrutail )
1413 fprintf( stderr, "LRU circle (tail backward):\n" );
1414 for ( e = cache->c_lrutail; ; ) {
1415 fprintf( stderr, "\t%p, %p id %ld rdn \"%s\"\n",
1416 (void *) e, (void *) e->bei_e, e->bei_id, e->bei_nrdn.bv_val );
1418 if ( e == cache->c_lrutail )
1425 #ifdef BDB_REUSE_LOCKERS
1427 bdb_locker_id_free( void *key, void *data )
1433 #if DB_VERSION_FULL >= 0x04060012
1434 BDB_LOCKER lptr = data;
1437 lockid = (long)data;
1439 rc = XLOCK_ID_FREE( env, lockid );
1440 if ( rc == EINVAL ) {
1442 Debug( LDAP_DEBUG_ANY,
1443 "bdb_locker_id_free: %lu err %s(%d)\n",
1444 (unsigned long) lockid, db_strerror(rc), rc );
1445 /* release all locks held by this locker. */
1446 lr.op = DB_LOCK_PUT_ALL;
1448 env->lock_vec( env, lockid, 0, &lr, 1, NULL );
1449 XLOCK_ID_FREE( env, lockid );
1453 /* free up any keys used by the main thread */
1455 bdb_locker_flush( DB_ENV *env )
1458 void *ctx = ldap_pvt_thread_pool_context();
1460 if ( !ldap_pvt_thread_pool_getkey( ctx, env, &data, NULL ) ) {
1461 ldap_pvt_thread_pool_setkey( ctx, env, NULL, NULL );
1462 bdb_locker_id_free( env, data );
1467 bdb_locker_id( Operation *op, DB_ENV *env, BDB_LOCKER *locker )
1474 if ( !env || !locker ) return -1;
1476 /* If no op was provided, try to find the ctx anyway... */
1478 ctx = op->o_threadctx;
1480 ctx = ldap_pvt_thread_pool_context();
1483 /* Shouldn't happen unless we're single-threaded */
1489 if ( ldap_pvt_thread_pool_getkey( ctx, env, &data, NULL ) ) {
1490 for ( i=0, rc=1; rc != 0 && i<4; i++ ) {
1491 rc = XLOCK_ID( env, &lockid );
1492 if (rc) ldap_pvt_thread_yield();
1497 #if DB_VERSION_FULL >= 0x04060012
1499 __lock_getlocker( env->lk_handle, lockid, 0, &lptr );
1503 data = (void *)((long)lockid);
1505 if ( ( rc = ldap_pvt_thread_pool_setkey( ctx, env,
1506 data, bdb_locker_id_free ) ) ) {
1507 XLOCK_ID_FREE( env, lockid );
1508 Debug( LDAP_DEBUG_ANY, "bdb_locker_id: err %s(%d)\n",
1509 db_strerror(rc), rc, 0 );
1514 lockid = (long)data;
1516 #if DB_VERSION_FULL >= 0x04060012
1523 #endif /* BDB_REUSE_LOCKERS */