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 );
260 avl_insert( &ei->bei_parent->bei_kids, ei2, bdb_rdn_cmp,
263 ei->bei_parent->bei_ckids++;
271 /* Find the EntryInfo for the requested DN. If the DN cannot be found, return
272 * the info for its closest ancestor. *res should be NULL to process a
273 * complete DN starting from the tree root. Otherwise *res must be the
274 * immediate parent of the requested DN, and only the RDN will be searched.
275 * The EntryInfo is locked upon return and must be unlocked by the caller.
284 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
285 EntryInfo ei, *eip, *ei2;
289 /* this function is always called with normalized DN */
291 /* we're doing a onelevel search for an RDN */
292 ei.bei_nrdn.bv_val = ndn->bv_val;
293 ei.bei_nrdn.bv_len = dn_rdnlen( op->o_bd, ndn );
296 /* we're searching a full DN from the root */
297 ptr = ndn->bv_val + ndn->bv_len - op->o_bd->be_nsuffix[0].bv_len;
298 ei.bei_nrdn.bv_val = ptr;
299 ei.bei_nrdn.bv_len = op->o_bd->be_nsuffix[0].bv_len;
300 /* Skip to next rdn if suffix is empty */
301 if ( ei.bei_nrdn.bv_len == 0 ) {
302 for (ptr = ei.bei_nrdn.bv_val - 2; ptr > ndn->bv_val
303 && !DN_SEPARATOR(*ptr); ptr--) /* empty */;
304 if ( ptr >= ndn->bv_val ) {
305 if (DN_SEPARATOR(*ptr)) ptr++;
306 ei.bei_nrdn.bv_len = ei.bei_nrdn.bv_val - ptr;
307 ei.bei_nrdn.bv_val = ptr;
310 eip = &bdb->bi_cache.c_dntree;
313 for ( bdb_cache_entryinfo_lock( eip ); eip; ) {
315 ei2 = (EntryInfo *)avl_find( eip->bei_kids, &ei, bdb_rdn_cmp );
317 int len = ei.bei_nrdn.bv_len;
319 if ( BER_BVISEMPTY( ndn )) {
324 ei.bei_nrdn.bv_len = ndn->bv_len -
325 (ei.bei_nrdn.bv_val - ndn->bv_val);
326 bdb_cache_entryinfo_unlock( eip );
328 rc = bdb_dn2id( op, txn, &ei.bei_nrdn, &ei );
330 bdb_cache_entryinfo_lock( eip );
335 /* DN exists but needs to be added to cache */
336 ei.bei_nrdn.bv_len = len;
337 rc = bdb_entryinfo_add_internal( bdb, &ei, &ei2 );
338 /* add_internal left eip and c_rwlock locked */
339 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
344 } else if ( ei2->bei_state & CACHE_ENTRY_DELETED ) {
345 /* In the midst of deleting? Give it a chance to
348 bdb_cache_entryinfo_unlock( eip );
349 ldap_pvt_thread_yield();
350 bdb_cache_entryinfo_lock( eip );
354 bdb_cache_entryinfo_unlock( eip );
355 bdb_cache_entryinfo_lock( ei2 );
359 /* Advance to next lower RDN */
360 for (ptr = ei.bei_nrdn.bv_val - 2; ptr > ndn->bv_val
361 && !DN_SEPARATOR(*ptr); ptr--) /* empty */;
362 if ( ptr >= ndn->bv_val ) {
363 if (DN_SEPARATOR(*ptr)) ptr++;
364 ei.bei_nrdn.bv_len = ei.bei_nrdn.bv_val - ptr - 1;
365 ei.bei_nrdn.bv_val = ptr;
367 if ( ptr < ndn->bv_val ) {
377 /* Walk up the tree from a child node, looking for an ID that's already
378 * been linked into the cache.
381 hdb_cache_find_parent(
387 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
388 EntryInfo ei, eip, *ei2 = NULL, *ein = NULL, *eir = NULL;
397 rc = hdb_dn2id_parent( op, txn, &ei, &eip.bei_id );
400 /* Save the previous node, if any */
403 /* Create a new node for the current ID */
404 ein = bdb_cache_entryinfo_new( &bdb->bi_cache );
405 ein->bei_id = ei.bei_id;
406 ein->bei_kids = ei.bei_kids;
407 ein->bei_nrdn = ei.bei_nrdn;
408 ein->bei_rdn = ei.bei_rdn;
409 ein->bei_ckids = ei.bei_ckids;
410 #ifdef SLAP_ZONE_ALLOC
415 /* This node is not fully connected yet */
416 ein->bei_state = CACHE_ENTRY_NOT_LINKED;
418 /* Insert this node into the ID tree */
419 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
420 if ( avl_insert( &bdb->bi_cache.c_idtree, (caddr_t)ein,
421 bdb_id_cmp, avl_dup_error ) ) {
423 /* Someone else created this node just before us.
424 * Free our new copy and use the existing one.
426 bdb_cache_entryinfo_destroy( ein );
427 ein = (EntryInfo *)avl_find( bdb->bi_cache.c_idtree,
428 (caddr_t) &ei, bdb_id_cmp );
430 /* Link in any kids we've already processed */
432 bdb_cache_entryinfo_lock( ein );
433 avl_insert( &ein->bei_kids, (caddr_t)ei2,
434 bdb_rdn_cmp, avl_dup_error );
436 bdb_cache_entryinfo_unlock( ein );
444 /* If this is the first time, save this node
445 * to be returned later.
447 if ( eir == NULL ) eir = ein;
449 /* If there was a previous node, link it to this one */
450 if ( ei2 ) ei2->bei_parent = ein;
452 /* Look for this node's parent */
454 ei2 = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
455 (caddr_t) &eip, bdb_id_cmp );
457 ei2 = &bdb->bi_cache.c_dntree;
459 bdb->bi_cache.c_eiused++;
460 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
462 /* Got the parent, link in and we're done. */
464 bdb_cache_entryinfo_lock( ei2 );
465 ein->bei_parent = ei2;
466 avl_insert( &ei2->bei_kids, (caddr_t)ein, bdb_rdn_cmp,
469 bdb_cache_entryinfo_unlock( ei2 );
470 bdb_cache_entryinfo_lock( eir );
472 /* Reset all the state info */
473 for (ein = eir; ein != ei2; ein=ein->bei_parent)
474 ein->bei_state &= ~CACHE_ENTRY_NOT_LINKED;
479 ei.bei_id = eip.bei_id;
481 avl_insert( &ei.bei_kids, (caddr_t)ein, bdb_rdn_cmp,
487 /* Used by hdb_dn2idl when loading the EntryInfo for all the children
491 struct bdb_info *bdb,
498 /* See if we already have this one */
499 bdb_cache_entryinfo_lock( ei->bei_parent );
500 ei2 = (EntryInfo *)avl_find( ei->bei_parent->bei_kids, ei, bdb_rdn_cmp );
501 bdb_cache_entryinfo_unlock( ei->bei_parent );
504 /* Not found, add it */
507 /* bei_rdn was not malloc'd before, do it now */
508 ber_dupbv( &bv, &ei->bei_rdn );
511 rc = bdb_entryinfo_add_internal( bdb, ei, res );
512 bdb_cache_entryinfo_unlock( ei->bei_parent );
513 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
515 /* Found, return it */
523 /* caller must have lru_mutex locked. mutex
524 * will be unlocked on return.
528 struct bdb_info *bdb,
532 DB_LOCK lock, *lockp;
540 /* See if we're above the cache size limit */
541 if ( bdb->bi_cache.c_cursize > bdb->bi_cache.c_maxsize ) {
542 EntryInfo *elru, *elprev;
545 /* Look for an unused entry to remove */
546 for (elru = bdb->bi_cache.c_lrutail; elru; elru = elprev, i++ ) {
547 elprev = elru->bei_lruprev;
549 /* Too many probes, not enough idle, give up */
553 /* If we can successfully writelock it, then
554 * the object is idle.
556 if ( bdb_cache_entry_db_lock( bdb->bi_dbenv,
557 bdb->bi_cache.c_locker, elru, 1, 1, lockp ) == 0 ) {
560 /* If there's no entry, or this node is in
561 * the process of linking into the cache,
562 * or this node is being deleted, skip it.
564 if ( !elru->bei_e || (elru->bei_state &
565 ( CACHE_ENTRY_NOT_LINKED | CACHE_ENTRY_DELETED ))) {
566 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lockp );
569 LRU_DELETE( &bdb->bi_cache, elru );
570 elru->bei_e->e_private = NULL;
571 #ifdef SLAP_ZONE_ALLOC
572 bdb_entry_return( bdb, elru->bei_e, elru->bei_zseq );
574 bdb_entry_return( elru->bei_e );
577 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
578 --bdb->bi_cache.c_cursize;
579 if (bdb->bi_cache.c_cursize <= bdb->bi_cache.c_maxsize)
581 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
582 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lockp );
587 LRU_ADD( &bdb->bi_cache, ei );
588 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_mutex );
593 struct bdb_info *bdb,
596 EntryInfo ei = { 0 },
601 ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
602 ei2 = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
603 (caddr_t) &ei, bdb_id_cmp );
604 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
609 * cache_find_id - find an entry in the cache, given id.
610 * The entry is locked for Read upon return. Call with islocked TRUE if
611 * the supplied *eip was already locked.
624 struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
626 int rc = 0, load = 0;
627 EntryInfo ei = { 0 };
631 #ifdef SLAP_ZONE_ALLOC
632 slap_zh_rlock(bdb->bi_cache.c_zctx);
634 /* If we weren't given any info, see if we have it already cached */
636 again: ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
637 *eip = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
638 (caddr_t) &ei, bdb_id_cmp );
640 /* If the lock attempt fails, the info is in use */
641 if ( ldap_pvt_thread_mutex_trylock(
642 &(*eip)->bei_kids_mutex )) {
643 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
644 /* If this node is being deleted, treat
645 * as if the delete has already finished
647 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
650 /* otherwise, wait for the info to free up */
651 ldap_pvt_thread_yield();
654 /* If this info isn't hooked up to its parent yet,
655 * unlock and wait for it to be fully initialized
657 if ( (*eip)->bei_state & CACHE_ENTRY_NOT_LINKED ) {
658 bdb_cache_entryinfo_unlock( *eip );
659 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
660 ldap_pvt_thread_yield();
665 ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
668 /* See if the ID exists in the database; add it to the cache if so */
671 rc = bdb_id2entry( op->o_bd, tid, id, &ep );
673 rc = bdb_cache_find_ndn( op, tid,
675 if ( *eip ) islocked = 1;
677 #ifdef SLAP_ZONE_ALLOC
678 bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
680 bdb_entry_return( ep );
686 rc = hdb_cache_find_parent(op, tid, id, eip );
687 if ( rc == 0 && *eip ) islocked = 1;
691 /* Ok, we found the info, do we have the entry? */
692 if ( *eip && rc == 0 ) {
693 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
696 /* Make sure only one thread tries to load the entry */
698 #ifdef SLAP_ZONE_ALLOC
699 if ((*eip)->bei_e && !slap_zn_validate(
700 bdb->bi_cache.c_zctx, (*eip)->bei_e, (*eip)->bei_zseq)) {
701 (*eip)->bei_e = NULL;
702 (*eip)->bei_zseq = 0;
705 if ( !(*eip)->bei_e && !((*eip)->bei_state & CACHE_ENTRY_LOADING)) {
707 (*eip)->bei_state |= CACHE_ENTRY_LOADING;
710 bdb_cache_entryinfo_unlock( *eip );
713 rc = bdb_cache_entry_db_lock( bdb->bi_dbenv, locker, *eip, 0, 0, lock );
714 if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
716 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
717 } else if ( rc == 0 ) {
720 u_int32_t locker2 = locker;
722 /* We don't wrap entire read operations in txn's, but
723 * we need our cache entry lock and any DB page locks
724 * to be associated, in order for deadlock detection
725 * to work properly. So if we need to read from the DB,
726 * we use a long-lived per-thread txn for this step.
729 rc = bdb_txn_get( op, bdb->bi_dbenv, <id, 0 );
731 locker2 = TXN_ID( ltid );
735 /* Give up original read lock, obtain write lock with
736 * (possibly) new locker ID.
739 rc = bdb_cache_entry_db_relock( bdb->bi_dbenv, locker2,
742 if ( rc == 0 && !ep) {
743 rc = bdb_id2entry( op->o_bd, ltid, id, &ep );
746 ep->e_private = *eip;
751 #ifdef SLAP_ZONE_ALLOC
752 (*eip)->bei_zseq = *((ber_len_t *)ep - 2);
756 (*eip)->bei_state ^= CACHE_ENTRY_LOADING;
758 /* If we succeeded, downgrade back to a readlock. */
759 rc = bdb_cache_entry_db_relock( bdb->bi_dbenv, locker,
762 /* Otherwise, release the lock. */
763 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
765 if ( locker2 != locker ) {
766 /* If we're using the per-thread txn, release all
767 * of its page locks now.
770 list.op = DB_LOCK_PUT_ALL;
772 bdb->bi_dbenv->lock_vec( bdb->bi_dbenv, locker2,
774 /* If this txn was deadlocked, we must abort it
775 * and invalidate this per-thread txn.
777 if ( rc == DB_LOCK_DEADLOCK ) {
778 bdb_txn_get( op, bdb->bi_dbenv, <id, 1 );
781 } else if ( !(*eip)->bei_e ) {
782 /* Some other thread is trying to load the entry,
783 * give it a chance to finish.
785 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
786 ldap_pvt_thread_yield();
787 bdb_cache_entryinfo_lock( *eip );
792 /* Check for subtree renames
794 rc = bdb_fix_dn( (*eip)->bei_e, 1 );
796 bdb_cache_entry_db_relock( bdb->bi_dbenv,
797 locker, *eip, 1, 0, lock );
798 /* check again in case other modifier did it already */
799 if ( bdb_fix_dn( (*eip)->bei_e, 1 ) )
800 rc = bdb_fix_dn( (*eip)->bei_e, 2 );
801 bdb_cache_entry_db_relock( bdb->bi_dbenv,
802 locker, *eip, 0, 0, lock );
811 bdb_cache_entryinfo_unlock( *eip );
814 #ifdef SLAP_ZONE_ALLOC
815 bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
817 bdb_entry_return( ep );
823 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
824 bdb->bi_cache.c_cursize++;
825 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
828 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_mutex );
830 /* If the LRU list has only one entry and this is it, it
831 * doesn't need to be added again.
833 if ( bdb->bi_cache.c_lruhead == bdb->bi_cache.c_lrutail &&
834 bdb->bi_cache.c_lruhead == *eip ) {
835 ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_mutex );
838 /* if entry is on LRU list, remove from old spot */
839 if ( (*eip)->bei_lrunext || (*eip)->bei_lruprev ) {
840 LRU_DELETE( &bdb->bi_cache, *eip );
842 /* lru_mutex is unlocked for us */
843 bdb_cache_lru_add( bdb, locker, *eip );
847 #ifdef SLAP_ZONE_ALLOC
848 if (rc == 0 && (*eip)->bei_e) {
849 slap_zn_rlock(bdb->bi_cache.c_zctx, (*eip)->bei_e);
851 slap_zh_runlock(bdb->bi_cache.c_zctx);
864 if ( BEI(e)->bei_kids ) {
867 if ( BEI(e)->bei_state & CACHE_ENTRY_NO_KIDS ) {
870 rc = bdb_dn2id_children( op, txn, e );
871 if ( rc == DB_NOTFOUND ) {
872 BEI(e)->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
877 /* Update the cache after a successful database Add. */
880 struct bdb_info *bdb,
887 struct berval rdn = e->e_name;
896 /* Lock this entry so that bdb_add can run to completion.
897 * It can only fail if BDB has run out of lock resources.
899 rc = bdb_cache_entry_db_lock( bdb->bi_dbenv, locker, &ei, 1, 0, &lock );
901 bdb_cache_entryinfo_unlock( eip );
906 if ( nrdn->bv_len != e->e_nname.bv_len ) {
907 char *ptr = strchr( rdn.bv_val, ',' );
908 rdn.bv_len = ptr - rdn.bv_val;
910 ber_dupbv( &ei.bei_rdn, &rdn );
911 if ( eip->bei_dkids ) eip->bei_dkids++;
914 rc = bdb_entryinfo_add_internal( bdb, &ei, &new );
915 /* bdb_csn_commit can cause this when adding the database root entry */
917 new->bei_e->e_private = NULL;
918 #ifdef SLAP_ZONE_ALLOC
919 bdb_entry_return( bdb, new->bei_e, new->bei_zseq );
921 bdb_entry_return( new->bei_e );
926 new->bei_state = CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
927 eip->bei_state &= ~CACHE_ENTRY_NO_KIDS;
928 if (eip->bei_parent) {
929 eip->bei_parent->bei_state &= ~CACHE_ENTRY_NO_GRANDKIDS;
931 bdb_cache_entryinfo_unlock( eip );
933 ++bdb->bi_cache.c_cursize;
934 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
937 ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_mutex );
939 /* lru_mutex is unlocked for us */
940 bdb_cache_lru_add( bdb, locker, new );
953 EntryInfo *ei = BEI(e);
955 /* Get write lock on data */
956 rc = bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock );
958 /* If we've done repeated mods on a cached entry, then e_attrs
959 * is no longer contiguous with the entry, and must be freed.
962 if ( (void *)e->e_attrs != (void *)(e+1) ) {
963 attrs_free( e->e_attrs );
965 e->e_attrs = newAttrs;
971 * Change the rdn in the entryinfo. Also move to a new parent if needed.
975 struct bdb_info *bdb,
983 EntryInfo *ei = BEI(e), *pei;
987 /* Get write lock on data */
988 rc = bdb_cache_entry_db_relock( bdb->bi_dbenv, locker, ei, 1, 0, lock );
991 /* If we've done repeated mods on a cached entry, then e_attrs
992 * is no longer contiguous with the entry, and must be freed.
994 if ( (void *)e->e_attrs != (void *)(e+1) ) {
995 attrs_free( e->e_attrs );
997 e->e_attrs = new->e_attrs;
998 if( e->e_nname.bv_val < e->e_bv.bv_val ||
999 e->e_nname.bv_val > e->e_bv.bv_val + e->e_bv.bv_len )
1001 ch_free(e->e_name.bv_val);
1002 ch_free(e->e_nname.bv_val);
1004 e->e_name = new->e_name;
1005 e->e_nname = new->e_nname;
1007 /* Lock the parent's kids AVL tree */
1008 pei = ei->bei_parent;
1009 bdb_cache_entryinfo_lock( pei );
1010 avl_delete( &pei->bei_kids, (caddr_t) ei, bdb_rdn_cmp );
1011 free( ei->bei_nrdn.bv_val );
1012 ber_dupbv( &ei->bei_nrdn, nrdn );
1014 free( ei->bei_rdn.bv_val );
1017 if ( nrdn->bv_len != e->e_nname.bv_len ) {
1018 char *ptr = strchr(rdn.bv_val, ',');
1019 rdn.bv_len = ptr - rdn.bv_val;
1021 ber_dupbv( &ei->bei_rdn, &rdn );
1025 ein = ei->bei_parent;
1027 ei->bei_parent = ein;
1028 bdb_cache_entryinfo_unlock( pei );
1029 bdb_cache_entryinfo_lock( ein );
1033 /* Record the generation number of this change */
1034 ldap_pvt_thread_mutex_lock( &bdb->bi_modrdns_mutex );
1036 ei->bei_modrdns = bdb->bi_modrdns;
1037 ldap_pvt_thread_mutex_unlock( &bdb->bi_modrdns_mutex );
1040 avl_insert( &ein->bei_kids, ei, bdb_rdn_cmp, avl_dup_error );
1041 bdb_cache_entryinfo_unlock( ein );
1045 * cache_delete - delete the entry e from the cache.
1047 * returns: 0 e was deleted ok
1048 * 1 e was not in the cache
1049 * -1 something bad happened
1059 EntryInfo *ei = BEI(e);
1062 assert( e->e_private );
1064 /* Set this early, warn off any queriers */
1065 ei->bei_state |= CACHE_ENTRY_DELETED;
1067 /* Lock the entry's info */
1068 bdb_cache_entryinfo_lock( ei );
1070 /* Get write lock on the data */
1071 rc = bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock );
1073 /* couldn't lock, undo and give up */
1074 ei->bei_state ^= CACHE_ENTRY_DELETED;
1075 bdb_cache_entryinfo_unlock( ei );
1079 Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_delete( %ld )\n",
1083 ldap_pvt_thread_mutex_lock( &cache->lru_mutex );
1084 rc = bdb_cache_delete_internal( cache, e->e_private, 1 );
1085 /* free lru mutex */
1086 ldap_pvt_thread_mutex_unlock( &cache->lru_mutex );
1088 /* Leave entry info locked */
1094 bdb_cache_delete_cleanup(
1099 ei->bei_e->e_private = NULL;
1100 #ifdef SLAP_ZONE_ALLOC
1101 bdb_entry_return( ei->bei_bdb, ei->bei_e, ei->bei_zseq );
1103 bdb_entry_return( ei->bei_e );
1108 free( ei->bei_nrdn.bv_val );
1109 ei->bei_nrdn.bv_val = NULL;
1111 free( ei->bei_rdn.bv_val );
1112 ei->bei_rdn.bv_val = NULL;
1113 ei->bei_modrdns = 0;
1117 ei->bei_parent = NULL;
1118 ei->bei_kids = NULL;
1119 ei->bei_lruprev = NULL;
1121 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
1122 ei->bei_lrunext = cache->c_eifree;
1123 cache->c_eifree = ei;
1124 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
1125 bdb_cache_entryinfo_unlock( ei );
1129 bdb_cache_delete_internal(
1134 int rc = 0; /* return code */
1136 /* set cache write lock */
1137 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
1139 /* Lock the parent's kids tree */
1140 bdb_cache_entryinfo_lock( e->bei_parent );
1143 e->bei_parent->bei_ckids--;
1144 if ( decr && e->bei_parent->bei_dkids ) e->bei_parent->bei_dkids--;
1147 if ( avl_delete( &e->bei_parent->bei_kids, (caddr_t) e, bdb_rdn_cmp )
1154 if ( avl_delete( &cache->c_idtree, (caddr_t) e, bdb_id_cmp ) == NULL ) {
1165 LRU_DELETE( cache, e );
1166 if ( e->bei_e ) cache->c_cursize--;
1168 /* free cache write lock */
1169 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
1170 bdb_cache_entryinfo_unlock( e->bei_parent );
1176 bdb_entryinfo_release( void *data )
1178 EntryInfo *ei = (EntryInfo *)data;
1179 if ( ei->bei_kids ) {
1180 avl_free( ei->bei_kids, NULL );
1183 ei->bei_e->e_private = NULL;
1184 #ifdef SLAP_ZONE_ALLOC
1185 bdb_entry_return( ei->bei_bdb, ei->bei_e, ei->bei_zseq );
1187 bdb_entry_return( ei->bei_e );
1190 bdb_cache_entryinfo_destroy( ei );
1194 bdb_cache_release_all( Cache *cache )
1196 /* set cache write lock */
1197 ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
1199 ldap_pvt_thread_mutex_lock( &cache->lru_mutex );
1201 Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_release_all\n", 0, 0, 0 );
1203 avl_free( cache->c_dntree.bei_kids, NULL );
1204 avl_free( cache->c_idtree, bdb_entryinfo_release );
1205 for (;cache->c_eifree;cache->c_eifree = cache->c_lruhead) {
1206 cache->c_lruhead = cache->c_eifree->bei_lrunext;
1207 bdb_cache_entryinfo_destroy(cache->c_eifree);
1209 cache->c_cursize = 0;
1210 cache->c_eiused = 0;
1211 cache->c_idtree = NULL;
1212 cache->c_lruhead = NULL;
1213 cache->c_lrutail = NULL;
1214 cache->c_dntree.bei_kids = NULL;
1216 /* free lru mutex */
1217 ldap_pvt_thread_mutex_unlock( &cache->lru_mutex );
1218 /* free cache write lock */
1219 ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
1224 bdb_lru_print( Cache *cache )
1228 fprintf( stderr, "LRU queue (head to tail):\n" );
1229 for ( e = cache->c_lruhead; e != NULL; e = e->bei_lrunext ) {
1230 fprintf( stderr, "\trdn \"%20s\" id %ld\n",
1231 e->bei_nrdn.bv_val, e->bei_id );
1233 fprintf( stderr, "LRU queue (tail to head):\n" );
1234 for ( e = cache->c_lrutail; e != NULL; e = e->bei_lruprev ) {
1235 fprintf( stderr, "\trdn \"%20s\" id %ld\n",
1236 e->bei_nrdn.bv_val, e->bei_id );
1242 bdb_txn_free( void *key, void *data )
1248 /* Obtain a long-lived transaction for the current thread.
1249 * If reset == 1, remove the current transaction. */
1251 bdb_txn_get( Operation *op, DB_ENV *env, DB_TXN **txn, int reset )
1254 void *ctx, *data = NULL;
1256 if ( slapMode & SLAP_TOOL_MODE ) {
1261 /* If no op was provided, try to find the ctx anyway... */
1263 ctx = op->o_threadctx;
1265 ctx = ldap_pvt_thread_pool_context();
1268 /* Shouldn't happen unless we're single-threaded */
1276 return ldap_pvt_thread_pool_setkey( ctx, ((char *)env)+1, NULL, NULL );
1279 if ( ldap_pvt_thread_pool_getkey( ctx, ((char *)env)+1, &data, NULL ) ||
1281 for ( i=0, rc=1; rc != 0 && i<4; i++ ) {
1282 rc = TXN_BEGIN( env, NULL, txn, READ_TXN_FLAG );
1283 #if DB_VERSION_FULL == 0x04020034
1284 if ( rc == EINVAL && READ_TXN_FLAG ) {
1286 Debug( LDAP_DEBUG_ANY,
1287 "bdb_txn_get: BerkeleyDB 4.2.52 library needs TXN patch!\n",
1293 if (rc) ldap_pvt_thread_yield();
1298 if ( ( rc = ldap_pvt_thread_pool_setkey( ctx, ((char *)env)+1,
1299 *txn, bdb_txn_free ) ) ) {
1301 Debug( LDAP_DEBUG_ANY, "bdb_txn_get: err %s(%d)\n",
1302 db_strerror(rc), rc, 0 );
1312 #ifdef BDB_REUSE_LOCKERS
1314 bdb_locker_id_free( void *key, void *data )
1317 int lockid = (int) data;
1320 rc = XLOCK_ID_FREE( env, lockid );
1321 if ( rc == EINVAL ) {
1323 Debug( LDAP_DEBUG_ANY,
1324 "bdb_locker_id_free: %d err %s(%d)\n",
1325 lockid, db_strerror(rc), rc );
1326 /* release all locks held by this locker. */
1327 lr.op = DB_LOCK_PUT_ALL;
1329 env->lock_vec( env, lockid, 0, &lr, 1, NULL );
1330 XLOCK_ID_FREE( env, lockid );
1335 bdb_locker_id( Operation *op, DB_ENV *env, int *locker )
1341 if ( !env || !locker ) return -1;
1343 /* If no op was provided, try to find the ctx anyway... */
1345 ctx = op->o_threadctx;
1347 ctx = ldap_pvt_thread_pool_context();
1350 /* Shouldn't happen unless we're single-threaded */
1356 if ( ldap_pvt_thread_pool_getkey( ctx, env, &data, NULL ) ) {
1357 for ( i=0, rc=1; rc != 0 && i<4; i++ ) {
1358 rc = XLOCK_ID( env, &lockid );
1359 if (rc) ldap_pvt_thread_yield();
1364 data = (void *)lockid;
1365 if ( ( rc = ldap_pvt_thread_pool_setkey( ctx, env,
1366 data, bdb_locker_id_free ) ) ) {
1367 XLOCK_ID_FREE( env, lockid );
1368 Debug( LDAP_DEBUG_ANY, "bdb_locker_id: err %s(%d)\n",
1369 db_strerror(rc), rc, 0 );
1382 bdb_cache_delete_entry(
1383 struct bdb_info *bdb,
1388 ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
1389 if ( bdb_cache_entry_db_lock( bdb->bi_dbenv, bdb->bi_cache.c_locker, ei, 1, 1, lock ) == 0 )
1391 if ( ei->bei_e && !(ei->bei_state & CACHE_ENTRY_NOT_LINKED )) {
1392 LRU_DELETE( &bdb->bi_cache, ei );
1393 ei->bei_e->e_private = NULL;
1394 #ifdef SLAP_ZONE_ALLOC
1395 bdb_entry_return( bdb, ei->bei_e, ei->bei_zseq );
1397 bdb_entry_return( ei->bei_e );
1400 --bdb->bi_cache.c_cursize;
1402 bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
1404 ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );