]> git.sur5r.net Git - openldap/blobdiff - servers/slapd/back-bdb/cache.c
Fix prev commit, cleanup fakeroot entry before freeing
[openldap] / servers / slapd / back-bdb / cache.c
index 86acd1316fb121c3786611f89add02aab3d16b1d..7f93b3bad1096373bf04c069f3da0cf9fd66a3a5 100644 (file)
@@ -2,7 +2,7 @@
 /* $OpenLDAP$ */
 /* This work is part of OpenLDAP Software <http://www.openldap.org/>.
  *
- * Copyright 2000-2006 The OpenLDAP Foundation.
+ * Copyright 2000-2007 The OpenLDAP Foundation.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -35,6 +35,7 @@ static void bdb_cache_lru_purge( struct bdb_info *bdb );
 
 static int     bdb_cache_delete_internal(Cache *cache, EntryInfo *e, int decr);
 #ifdef LDAP_DEBUG
+#define SLAPD_UNUSED
 #ifdef SLAPD_UNUSED
 static void    bdb_lru_print(Cache *cache);
 #endif
@@ -64,6 +65,7 @@ bdb_cache_entryinfo_new( Cache *cache )
                        cache->c_eifree = ei->bei_lrunext;
                }
                ldap_pvt_thread_mutex_unlock( &cache->c_eifree_mutex );
+               ei->bei_finders = 0;
        }
        if ( !ei ) {
                ei = ch_calloc(1, sizeof(EntryInfo));
@@ -75,6 +77,36 @@ bdb_cache_entryinfo_new( Cache *cache )
        return ei;
 }
 
+static void
+bdb_cache_entryinfo_free( Cache *cache, EntryInfo *ei )
+{
+       free( ei->bei_nrdn.bv_val );
+       ei->bei_nrdn.bv_val = NULL;
+#ifdef BDB_HIER
+       free( ei->bei_rdn.bv_val );
+       ei->bei_rdn.bv_val = NULL;
+       ei->bei_modrdns = 0;
+       ei->bei_ckids = 0;
+       ei->bei_dkids = 0;
+#endif
+       ei->bei_parent = NULL;
+       ei->bei_kids = NULL;
+       ei->bei_lruprev = NULL;
+
+       ldap_pvt_thread_mutex_lock( &cache->c_eifree_mutex );
+       ei->bei_lrunext = cache->c_eifree;
+       cache->c_eifree = ei;
+       ldap_pvt_thread_mutex_unlock( &cache->c_eifree_mutex );
+}
+
+#define LRU_DEL( c, e ) do { \
+       if ( e == (c)->c_lruhead ) (c)->c_lruhead = e->bei_lruprev; \
+       if ( e == (c)->c_lrutail ) (c)->c_lrutail = e->bei_lruprev; \
+       e->bei_lrunext->bei_lruprev = e->bei_lruprev; \
+       e->bei_lruprev->bei_lrunext = e->bei_lrunext; \
+       e->bei_lruprev = NULL; \
+} while ( 0 )
+
 /* Note - we now use a Second-Chance / Clock algorithm instead of
  * Least-Recently-Used. This tremendously improves concurrency
  * because we no longer need to manipulate the lists every time an
@@ -84,22 +116,28 @@ bdb_cache_entryinfo_new( Cache *cache )
  * during a purge operation.
  */
 static void
-bdb_cache_lru_link( Cache *cache, EntryInfo *ei )
+bdb_cache_lru_link( struct bdb_info *bdb, EntryInfo *ei )
 {
+
        /* Insert into circular LRU list */
-       ldap_pvt_thread_mutex_lock( &cache->lru_tail_mutex );
-       ei->bei_lruprev = cache->c_lrutail;
-       if ( cache->c_lrutail ) {
-               ei->bei_lrunext = cache->c_lrutail->bei_lrunext;
-               cache->c_lrutail->bei_lrunext = ei;
+       ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_lru_mutex );
+
+       /* Still linked, remove */
+       if ( ei->bei_lruprev ) {
+               LRU_DEL( &bdb->bi_cache, ei );
+       }
+       ei->bei_lruprev = bdb->bi_cache.c_lrutail;
+       if ( bdb->bi_cache.c_lrutail ) {
+               ei->bei_lrunext = bdb->bi_cache.c_lrutail->bei_lrunext;
+               bdb->bi_cache.c_lrutail->bei_lrunext = ei;
                if ( ei->bei_lrunext )
                        ei->bei_lrunext->bei_lruprev = ei;
        } else {
                ei->bei_lrunext = ei->bei_lruprev = ei;
-               cache->c_lruhead = ei;
+               bdb->bi_cache.c_lruhead = ei;
        }
-       cache->c_lrutail = ei;
-       ldap_pvt_thread_mutex_unlock( &cache->lru_tail_mutex );
+       bdb->bi_cache.c_lrutail = ei;
+       ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_lru_mutex );
 }
 
 #ifdef NO_THREADS
@@ -120,7 +158,7 @@ bdb_cache_lru_link( Cache *cache, EntryInfo *ei )
 /* Atomically release and reacquire a lock */
 int
 bdb_cache_entry_db_relock(
-       DB_ENV *env,
+       struct bdb_info *bdb,
        u_int32_t locker,
        EntryInfo *ei,
        int rw,
@@ -145,7 +183,7 @@ bdb_cache_entry_db_relock(
        list[1].lock = *lock;
        list[1].mode = rw ? DB_LOCK_WRITE : DB_LOCK_READ;
        list[1].obj = &lockobj;
-       rc = env->lock_vec(env, locker, tryOnly ? DB_LOCK_NOWAIT : 0,
+       rc = bdb->bi_dbenv->lock_vec(bdb->bi_dbenv, locker, tryOnly ? DB_LOCK_NOWAIT : 0,
                list, 2, NULL );
 
        if (rc && !tryOnly) {
@@ -160,7 +198,7 @@ bdb_cache_entry_db_relock(
 }
 
 static int
-bdb_cache_entry_db_lock( DB_ENV *env, u_int32_t locker, EntryInfo *ei,
+bdb_cache_entry_db_lock( struct bdb_info *bdb, u_int32_t locker, EntryInfo *ei,
        int rw, int tryOnly, DB_LOCK *lock )
 {
 #ifdef NO_DB_LOCK
@@ -180,7 +218,7 @@ bdb_cache_entry_db_lock( DB_ENV *env, u_int32_t locker, EntryInfo *ei,
        lockobj.data = &ei->bei_id;
        lockobj.size = sizeof(ei->bei_id) + 1;
 
-       rc = LOCK_GET(env, locker, tryOnly ? DB_LOCK_NOWAIT : 0,
+       rc = LOCK_GET(bdb->bi_dbenv, locker, tryOnly ? DB_LOCK_NOWAIT : 0,
                                        &lockobj, db_rw, lock);
        if (rc && !tryOnly) {
                Debug( LDAP_DEBUG_TRACE,
@@ -192,7 +230,7 @@ bdb_cache_entry_db_lock( DB_ENV *env, u_int32_t locker, EntryInfo *ei,
 }
 
 int
-bdb_cache_entry_db_unlock ( DB_ENV *env, DB_LOCK *lock )
+bdb_cache_entry_db_unlock ( struct bdb_info *bdb, DB_LOCK *lock )
 {
 #ifdef NO_DB_LOCK
        return 0;
@@ -201,7 +239,7 @@ bdb_cache_entry_db_unlock ( DB_ENV *env, DB_LOCK *lock )
 
        if ( !lock || lock->mode == DB_LOCK_NG ) return 0;
 
-       rc = LOCK_PUT ( env, lock );
+       rc = LOCK_PUT ( bdb->bi_dbenv, lock );
        return rc;
 #endif
 }
@@ -238,6 +276,14 @@ bdb_id_cmp( const void *v_e1, const void *v_e2 )
        return e1->bei_id - e2->bei_id;
 }
 
+static int
+bdb_id_dup_err( void *v1, void *v2 )
+{
+       EntryInfo *e2 = v2;
+       e2->bei_lrunext = v1;
+       return -1;
+}
+
 /* Create an entryinfo in the cache. Caller must release the locks later.
  */
 static int
@@ -265,10 +311,10 @@ bdb_entryinfo_add_internal(
 #endif
 
        /* Add to cache ID tree */
-       if (avl_insert( &bdb->bi_cache.c_idtree, ei2, bdb_id_cmp, avl_dup_error )) {
-               EntryInfo *eix;
-               eix = avl_find( bdb->bi_cache.c_idtree, ei2, bdb_id_cmp );
-               bdb_cache_entryinfo_destroy( ei2 );
+       if (avl_insert( &bdb->bi_cache.c_idtree, ei2, bdb_id_cmp,
+               bdb_id_dup_err )) {
+               EntryInfo *eix = ei2->bei_lrunext;
+               bdb_cache_entryinfo_free( &bdb->bi_cache, ei2 );
                ei2 = eix;
 #ifdef BDB_HIER
                /* It got freed above because its value was
@@ -292,7 +338,6 @@ bdb_entryinfo_add_internal(
                ei->bei_parent->bei_ckids++;
 #endif
        }
-       bdb_cache_lru_link( &bdb->bi_cache, ei2 );
 
        *res = ei2;
        return 0;
@@ -449,14 +494,14 @@ hdb_cache_find_parent(
                /* Insert this node into the ID tree */
                ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
                if ( avl_insert( &bdb->bi_cache.c_idtree, (caddr_t)ein,
-                       bdb_id_cmp, avl_dup_error ) ) {
+                       bdb_id_cmp, bdb_id_dup_err ) ) {
+                       EntryInfo *eix = ein->bei_lrunext;
 
                        /* Someone else created this node just before us.
                         * Free our new copy and use the existing one.
                         */
-                       bdb_cache_entryinfo_destroy( ein );
-                       ein = (EntryInfo *)avl_find( bdb->bi_cache.c_idtree,
-                               (caddr_t) &ei, bdb_id_cmp );
+                       bdb_cache_entryinfo_free( &bdb->bi_cache, ein );
+                       ein = eix;
                        
                        /* Link in any kids we've already processed */
                        if ( ei2 ) {
@@ -466,8 +511,6 @@ hdb_cache_find_parent(
                                ein->bei_ckids++;
                                bdb_cache_entryinfo_unlock( ein );
                        }
-               } else {
-                       bdb_cache_lru_link( &bdb->bi_cache, ein );
                }
 
                /* If this is the first time, save this node
@@ -495,13 +538,14 @@ hdb_cache_find_parent(
                        bdb_cache_entryinfo_lock( ei2 );
                        ein->bei_parent = ei2;
 
+                       avl_insert( &ei2->bei_kids, (caddr_t)ein, bdb_rdn_cmp,
+                               avl_dup_error);
+                       ei2->bei_ckids++;
+
                        /* Reset all the state info */
                        for (ein = eir; ein != ei2; ein=ein->bei_parent)
                                ein->bei_state &= ~CACHE_ENTRY_NOT_LINKED;
 
-                       avl_insert( &ei2->bei_kids, (caddr_t)ein, bdb_rdn_cmp,
-                               avl_dup_error);
-                       ei2->bei_ckids++;
                        bdb_cache_entryinfo_unlock( ei2 );
                        bdb_cache_entryinfo_lock( eir );
 
@@ -553,72 +597,100 @@ int hdb_cache_load(
 }
 #endif
 
+/* This is best-effort only. If all entries in the cache are
+ * busy, they will all be kept. This is unlikely to happen
+ * unless the cache is very much smaller than the working set.
+ */
 static void
 bdb_cache_lru_purge( struct bdb_info *bdb )
 {
-       EntryInfo *elru, *elnext;
-       int count, islocked;
+       DB_LOCK         lock, *lockp;
+       EntryInfo *elru, *elnext = NULL;
+       int count, islocked, eimax;
 
-       /* Don't bother if we can't get the lock */
-       if ( ldap_pvt_thread_mutex_trylock( &bdb->bi_cache.lru_head_mutex ) )
-               return;
+       /* Wait for the mutex; we're the only one trying to purge. */
+       ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_lru_mutex );
 
        if ( bdb->bi_cache.c_cursize <= bdb->bi_cache.c_maxsize ) {
-               ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_head_mutex );
+               ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_lru_mutex );
+               bdb->bi_cache.c_purging = 0;
                return;
        }
 
+       if ( bdb->bi_cache.c_locker ) {
+               lockp = &lock;
+       } else {
+               lockp = NULL;
+       }
+
        count = 0;
+
+       /* maximum number of EntryInfo leaves to cache. In slapcat
+        * we always free all leaf nodes.
+        */
+       if ( slapMode & SLAP_TOOL_READONLY )
+               eimax = 0;
+       else
+               eimax = bdb->bi_cache.c_maxsize * 4;
+
        /* Look for an unused entry to remove */
-       for (elru = bdb->bi_cache.c_lruhead; elru; elru = elnext ) {
+       for ( elru = bdb->bi_cache.c_lruhead; elru; elru = elnext ) {
                elnext = elru->bei_lrunext;
 
-               if ( ldap_pvt_thread_mutex_trylock( &elru->bei_kids_mutex ))
-                       continue;
+               if ( bdb_cache_entryinfo_trylock( elru ))
+                       goto bottom;
 
                /* This flag implements the clock replacement behavior */
                if ( elru->bei_state & ( CACHE_ENTRY_REFERENCED )) {
                        elru->bei_state &= ~CACHE_ENTRY_REFERENCED;
                        bdb_cache_entryinfo_unlock( elru );
-                       continue;
+                       goto bottom;
                }
 
                /* If this node is in the process of linking into the cache,
                 * or this node is being deleted, skip it.
                 */
-               if ( elru->bei_state & ( CACHE_ENTRY_NOT_LINKED |
-                       CACHE_ENTRY_DELETED | CACHE_ENTRY_LOADING )) {
+               if (( elru->bei_state & ( CACHE_ENTRY_NOT_LINKED |
+                       CACHE_ENTRY_DELETED | CACHE_ENTRY_LOADING )) ||
+                       elru->bei_finders > 0 ) {
                        bdb_cache_entryinfo_unlock( elru );
-                       continue;
+                       goto bottom;
                }
 
+               /* entryinfo is locked */
                islocked = 1;
 
-               /* Free entry for this node if it's present */
-               if ( elru->bei_e ) {
-                       elru->bei_e->e_private = NULL;
+               /* If we can successfully writelock it, then
+                * the object is idle.
+                */
+               if ( bdb_cache_entry_db_lock( bdb,
+                       bdb->bi_cache.c_locker, elru, 1, 1, lockp ) == 0 ) {
+
+                       /* Free entry for this node if it's present */
+                       if ( elru->bei_e ) {
+                               elru->bei_e->e_private = NULL;
 #ifdef SLAP_ZONE_ALLOC
-                       bdb_entry_return( bdb, elru->bei_e, elru->bei_zseq );
+                               bdb_entry_return( bdb, elru->bei_e, elru->bei_zseq );
 #else
-                       bdb_entry_return( elru->bei_e );
+                               bdb_entry_return( elru->bei_e );
 #endif
-                       elru->bei_e = NULL;
-                       count++;
-               }
-               /* ITS#4010 if we're in slapcat, and this node is a leaf
-                * node, free it.
-                *
-                * FIXME: we need to do this for slapd as well, (which is
-                * why we compute bi_cache.c_leaves now) but at the moment
-                * we can't because it causes unresolvable deadlocks. 
-                */
-               if ( slapMode & SLAP_TOOL_READONLY ) {
-                       if ( !elru->bei_kids ) {
+                               elru->bei_e = NULL;
+                               count++;
+                       }
+                       bdb_cache_entry_db_unlock( bdb, lockp );
+
+                       /* 
+                        * If it is a leaf node, and we're over the limit, free it.
+                        */
+                       if ( elru->bei_kids ) {
+                               /* Drop from list, we ignore it... */
+                               LRU_DEL( &bdb->bi_cache, elru );
+                       } else if ( bdb->bi_cache.c_leaves > eimax ) {
+                               /* Too many leaf nodes, free this one */
                                bdb_cache_delete_internal( &bdb->bi_cache, elru, 0 );
                                bdb_cache_delete_cleanup( &bdb->bi_cache, elru );
                                islocked = 0;
-                       }
-                       /* Leave node on LRU list for a future pass */
+                       }       /* Leave on list until we need to free it */
                }
 
                if ( islocked )
@@ -630,10 +702,14 @@ bdb_cache_lru_purge( struct bdb_info *bdb )
                        ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_count_mutex );
                        break;
                }
+bottom:
+               if ( elnext == bdb->bi_cache.c_lruhead )
+                       break;
        }
 
-       bdb->bi_cache.c_lruhead = elru;
-       ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_head_mutex );
+       bdb->bi_cache.c_lruhead = elnext;
+       ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_lru_mutex );
+       bdb->bi_cache.c_purging = 0;
 }
 
 EntryInfo *
@@ -686,8 +762,7 @@ again:      ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
                        (caddr_t) &ei, bdb_id_cmp );
                if ( *eip ) {
                        /* If the lock attempt fails, the info is in use */
-                       if ( ldap_pvt_thread_mutex_trylock(
-                                       &(*eip)->bei_kids_mutex )) {
+                       if ( bdb_cache_entryinfo_trylock( *eip )) {
                                ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
                                /* If this node is being deleted, treat
                                 * as if the delete has already finished
@@ -742,6 +817,8 @@ again:      ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
                if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
                        rc = DB_NOTFOUND;
                } else {
+                       (*eip)->bei_finders++;
+                       (*eip)->bei_state |= CACHE_ENTRY_REFERENCED;
                        /* Make sure only one thread tries to load the entry */
 load1:
 #ifdef SLAP_ZONE_ALLOC
@@ -755,23 +832,18 @@ load1:
                                load = 1;
                                (*eip)->bei_state |= CACHE_ENTRY_LOADING;
                        }
+
                        if ( islocked ) {
                                bdb_cache_entryinfo_unlock( *eip );
                                islocked = 0;
                        }
-                       rc = bdb_cache_entry_db_lock( bdb->bi_dbenv, locker, *eip, 0, 0, lock );
+                       rc = bdb_cache_entry_db_lock( bdb, locker, *eip, load, 0, lock );
                        if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
                                rc = DB_NOTFOUND;
-                               bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
+                               bdb_cache_entry_db_unlock( bdb, lock );
                        } else if ( rc == 0 ) {
                                if ( load ) {
-                                       /* Give up original read lock, obtain write lock
-                                        */
-                                   if ( rc == 0 ) {
-                                               rc = bdb_cache_entry_db_relock( bdb->bi_dbenv, locker,
-                                                       *eip, 1, 0, lock );
-                                       }
-                                       if ( rc == 0 && !ep) {
+                                       if ( !ep) {
                                                rc = bdb_id2entry( op->o_bd, tid, locker, id, &ep );
                                        }
                                        if ( rc == 0 ) {
@@ -784,24 +856,21 @@ load1:
                                                (*eip)->bei_zseq = *((ber_len_t *)ep - 2);
 #endif
                                                ep = NULL;
+                                               bdb_cache_lru_link( bdb, *eip );
                                        }
-                                       bdb_cache_entryinfo_lock( *eip );
-                                       (*eip)->bei_state ^= CACHE_ENTRY_LOADING;
-                                       bdb_cache_entryinfo_unlock( *eip );
                                        if ( rc == 0 ) {
                                                /* If we succeeded, downgrade back to a readlock. */
-                                               rc = bdb_cache_entry_db_relock( bdb->bi_dbenv, locker,
+                                               rc = bdb_cache_entry_db_relock( bdb, locker,
                                                        *eip, 0, 0, lock );
                                        } else {
                                                /* Otherwise, release the lock. */
-                                               bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
+                                               bdb_cache_entry_db_unlock( bdb, lock );
                                        }
                                } else if ( !(*eip)->bei_e ) {
                                        /* Some other thread is trying to load the entry,
-                                        * give it a chance to finish.
+                                        * wait for it to finish.
                                         */
-                                       bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
-                                       ldap_pvt_thread_yield();
+                                       bdb_cache_entry_db_unlock( bdb, lock );
                                        bdb_cache_entryinfo_lock( *eip );
                                        islocked = 1;
                                        goto load1;
@@ -811,17 +880,21 @@ load1:
                                         */
                                        rc = bdb_fix_dn( (*eip)->bei_e, 1 );
                                        if ( rc ) {
-                                               bdb_cache_entry_db_relock( bdb->bi_dbenv,
+                                               bdb_cache_entry_db_relock( bdb,
                                                        locker, *eip, 1, 0, lock );
                                                /* check again in case other modifier did it already */
                                                if ( bdb_fix_dn( (*eip)->bei_e, 1 ) )
                                                        rc = bdb_fix_dn( (*eip)->bei_e, 2 );
-                                               bdb_cache_entry_db_relock( bdb->bi_dbenv,
+                                               bdb_cache_entry_db_relock( bdb,
                                                        locker, *eip, 0, 0, lock );
                                        }
 #endif
                                }
-
+                               bdb_cache_entryinfo_lock( *eip );
+                               (*eip)->bei_finders--;
+                               if ( load )
+                                       (*eip)->bei_state ^= CACHE_ENTRY_LOADING;
+                               bdb_cache_entryinfo_unlock( *eip );
                        }
                }
        }
@@ -842,8 +915,11 @@ load1:
                if ( load ) {
                        ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_count_mutex );
                        bdb->bi_cache.c_cursize++;
-                       if ( bdb->bi_cache.c_cursize > bdb->bi_cache.c_maxsize )
+                       if ( bdb->bi_cache.c_cursize > bdb->bi_cache.c_maxsize &&
+                               !bdb->bi_cache.c_purging ) {
                                purge = 1;
+                               bdb->bi_cache.c_purging = 1;
+                       }
                        ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_count_mutex );
                }
                if ( purge )
@@ -904,7 +980,7 @@ bdb_cache_add(
        /* Lock this entry so that bdb_add can run to completion.
         * It can only fail if BDB has run out of lock resources.
         */
-       rc = bdb_cache_entry_db_lock( bdb->bi_dbenv, locker, &ei, 0, 0, lock );
+       rc = bdb_cache_entry_db_lock( bdb, locker, &ei, 0, 0, lock );
        if ( rc ) {
                bdb_cache_entryinfo_unlock( eip );
                return rc;
@@ -942,10 +1018,15 @@ bdb_cache_add(
        ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
        ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_count_mutex );
        ++bdb->bi_cache.c_cursize;
-       if ( bdb->bi_cache.c_cursize > bdb->bi_cache.c_maxsize )
+       if ( bdb->bi_cache.c_cursize > bdb->bi_cache.c_maxsize &&
+               !bdb->bi_cache.c_purging ) {
                purge = 1;
+               bdb->bi_cache.c_purging = 1;
+       }
        ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_count_mutex );
 
+       bdb_cache_lru_link( bdb, new );
+
        if ( purge )
                bdb_cache_lru_purge( bdb );
 
@@ -954,16 +1035,16 @@ bdb_cache_add(
 
 int
 bdb_cache_modify(
+       struct bdb_info *bdb,
        Entry *e,
        Attribute *newAttrs,
-       DB_ENV *env,
        u_int32_t locker,
        DB_LOCK *lock )
 {
        EntryInfo *ei = BEI(e);
        int rc;
        /* Get write lock on data */
-       rc = bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock );
+       rc = bdb_cache_entry_db_relock( bdb, locker, ei, 1, 0, lock );
 
        /* If we've done repeated mods on a cached entry, then e_attrs
         * is no longer contiguous with the entry, and must be freed.
@@ -997,7 +1078,7 @@ bdb_cache_modrdn(
 #endif
 
        /* Get write lock on data */
-       rc =  bdb_cache_entry_db_relock( bdb->bi_dbenv, locker, ei, 1, 0, lock );
+       rc =  bdb_cache_entry_db_relock( bdb, locker, ei, 1, 0, lock );
        if ( rc ) return rc;
 
        /* If we've done repeated mods on a cached entry, then e_attrs
@@ -1080,9 +1161,8 @@ bdb_cache_modrdn(
  */
 int
 bdb_cache_delete(
-    Cache      *cache,
+       struct bdb_info *bdb,
     Entry              *e,
-    DB_ENV     *env,
     u_int32_t  locker,
     DB_LOCK    *lock )
 {
@@ -1098,7 +1178,7 @@ bdb_cache_delete(
        bdb_cache_entryinfo_lock( ei );
 
        /* Get write lock on the data */
-       rc = bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock );
+       rc = bdb_cache_entry_db_relock( bdb, locker, ei, 1, 0, lock );
        if ( rc ) {
                /* couldn't lock, undo and give up */
                ei->bei_state ^= CACHE_ENTRY_DELETED;
@@ -1110,12 +1190,12 @@ bdb_cache_delete(
                e->e_id, 0, 0 );
 
        /* set lru mutex */
-       ldap_pvt_thread_mutex_lock( &cache->lru_head_mutex );
+       ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_lru_mutex );
 
-       rc = bdb_cache_delete_internal( cache, e->e_private, 1 );
+       rc = bdb_cache_delete_internal( &bdb->bi_cache, e->e_private, 1 );
 
        /* free lru mutex */
-       ldap_pvt_thread_mutex_unlock( &cache->lru_head_mutex );
+       ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_lru_mutex );
 
        /* Leave entry info locked */
 
@@ -1137,23 +1217,7 @@ bdb_cache_delete_cleanup(
                ei->bei_e = NULL;
        }
 
-       free( ei->bei_nrdn.bv_val );
-       ei->bei_nrdn.bv_val = NULL;
-#ifdef BDB_HIER
-       free( ei->bei_rdn.bv_val );
-       ei->bei_rdn.bv_val = NULL;
-       ei->bei_modrdns = 0;
-       ei->bei_ckids = 0;
-       ei->bei_dkids = 0;
-#endif
-       ei->bei_parent = NULL;
-       ei->bei_kids = NULL;
-       ei->bei_lruprev = NULL;
-
-       ldap_pvt_thread_mutex_lock( &cache->c_eifree_mutex );
-       ei->bei_lrunext = cache->c_eifree;
-       cache->c_eifree = ei;
-       ldap_pvt_thread_mutex_unlock( &cache->c_eifree_mutex );
+       bdb_cache_entryinfo_free( cache, ei );
        bdb_cache_entryinfo_unlock( ei );
 }
 
@@ -1197,15 +1261,7 @@ bdb_cache_delete_internal(
 
        if ( rc == 0 ){
                /* lru */
-               if ( e == cache->c_lruhead ) cache->c_lruhead = e->bei_lrunext;
-               if ( e == cache->c_lrutail ) {
-                       ldap_pvt_thread_mutex_lock( &cache->lru_tail_mutex );
-                       if ( e == cache->c_lrutail ) cache->c_lrutail = e->bei_lruprev;
-                       ldap_pvt_thread_mutex_unlock( &cache->lru_tail_mutex );
-               }
-
-               if ( e->bei_lrunext ) e->bei_lrunext->bei_lruprev = e->bei_lruprev;
-               if ( e->bei_lruprev ) e->bei_lruprev->bei_lrunext = e->bei_lrunext;
+               LRU_DEL( cache, e );
 
                if ( e->bei_e ) {
                        ldap_pvt_thread_mutex_lock( &cache->c_count_mutex );
@@ -1241,7 +1297,7 @@ bdb_cache_release_all( Cache *cache )
        /* set cache write lock */
        ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
        /* set lru mutex */
-       ldap_pvt_thread_mutex_lock( &cache->lru_tail_mutex );
+       ldap_pvt_thread_mutex_lock( &cache->c_lru_mutex );
 
        Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_release_all\n", 0, 0, 0 );
 
@@ -1260,7 +1316,7 @@ bdb_cache_release_all( Cache *cache )
        cache->c_dntree.bei_kids = NULL;
 
        /* free lru mutex */
-       ldap_pvt_thread_mutex_unlock( &cache->lru_tail_mutex );
+       ldap_pvt_thread_mutex_unlock( &cache->c_lru_mutex );
        /* free cache write lock */
        ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
 }
@@ -1272,15 +1328,22 @@ bdb_lru_print( Cache *cache )
 {
        EntryInfo       *e;
 
-       fprintf( stderr, "LRU queue (head to tail):\n" );
-       for ( e = cache->c_lruhead; e != NULL; e = e->bei_lrunext ) {
-               fprintf( stderr, "\trdn \"%20s\" id %ld\n",
-                       e->bei_nrdn.bv_val, e->bei_id );
+       fprintf( stderr, "LRU circle head: %p\n", cache->c_lruhead );
+       fprintf( stderr, "LRU circle (tail forward):\n" );
+       for ( e = cache->c_lrutail; ; ) {
+               fprintf( stderr, "\t%p, %p id %ld rdn \"%s\"\n",
+                       e, e->bei_e, e->bei_id, e->bei_nrdn.bv_val );
+               e = e->bei_lrunext;
+               if ( e == cache->c_lrutail )
+                       break;
        }
-       fprintf( stderr, "LRU queue (tail to head):\n" );
-       for ( e = cache->c_lrutail; e != NULL; e = e->bei_lruprev ) {
-               fprintf( stderr, "\trdn \"%20s\" id %ld\n",
-                       e->bei_nrdn.bv_val, e->bei_id );
+       fprintf( stderr, "LRU circle (tail backward):\n" );
+       for ( e = cache->c_lrutail; ; ) {
+               fprintf( stderr, "\t%p, %p id %ld rdn \"%s\"\n",
+                       e, e->bei_e, e->bei_id, e->bei_nrdn.bv_val );
+               e = e->bei_lruprev;
+               if ( e == cache->c_lrutail )
+                       break;
        }
 }
 #endif
@@ -1308,6 +1371,19 @@ bdb_locker_id_free( void *key, void *data )
        }
 }
 
+/* free up any keys used by the main thread */
+void
+bdb_locker_flush( DB_ENV *env )
+{
+       void *data;
+       void *ctx = ldap_pvt_thread_pool_context();
+
+       if ( !ldap_pvt_thread_pool_getkey( ctx, env, &data, NULL ) ) {
+               ldap_pvt_thread_pool_setkey( ctx, env, NULL, NULL );
+               bdb_locker_id_free( env, data );
+       }
+}
+
 int
 bdb_locker_id( Operation *op, DB_ENV *env, u_int32_t *locker )
 {