X-Git-Url: https://git.sur5r.net/?a=blobdiff_plain;f=servers%2Fslapd%2Fback-bdb%2Fcache.c;h=f4de1cb9eab933420f065f6139e6dc20f1e52340;hb=1adee08e8912c1f47c7b170fe62bebdd9797921f;hp=7064c576b393afafb991b123a34165a237fd9f34;hpb=e48f72c1b5a7ce571c7ced749aed473d20b32526;p=openldap

diff --git a/servers/slapd/back-bdb/cache.c b/servers/slapd/back-bdb/cache.c
index 7064c576b3..f4de1cb9ea 100644
--- a/servers/slapd/back-bdb/cache.c
+++ b/servers/slapd/back-bdb/cache.c
@@ -2,7 +2,7 @@
 /* $OpenLDAP$ */
 /* This work is part of OpenLDAP Software <http://www.openldap.org/>.
  *
- * Copyright 2000-2007 The OpenLDAP Foundation.
+ * Copyright 2000-2010 The OpenLDAP Foundation.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -31,12 +31,14 @@
 #ifdef BDB_HIER
 #define bdb_cache_lru_purge	hdb_cache_lru_purge
 #endif
-static void bdb_cache_lru_purge( struct bdb_info *bdb, uint32_t locker );
+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);
+static void	bdb_idtree_print(Cache *cache);
 #endif
 #endif
 
@@ -62,6 +64,8 @@ bdb_cache_entryinfo_new( Cache *cache )
 		if ( cache->c_eifree ) {
 			ei = cache->c_eifree;
 			cache->c_eifree = ei->bei_lrunext;
+			ei->bei_finders = 0;
+			ei->bei_lrunext = NULL;
 		}
 		ldap_pvt_thread_mutex_unlock( &cache->c_eifree_mutex );
 	}
@@ -75,6 +79,44 @@ bdb_cache_entryinfo_new( Cache *cache )
 	return ei;
 }
 
+static void
+bdb_cache_entryinfo_free( Cache *cache, EntryInfo *ei )
+{
+	free( ei->bei_nrdn.bv_val );
+	BER_BVZERO( &ei->bei_nrdn );
+#ifdef BDB_HIER
+	free( ei->bei_rdn.bv_val );
+	BER_BVZERO( &ei->bei_rdn );
+	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;
+
+#if 0
+	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 );
+#else
+	ch_free( ei );
+#endif
+}
+
+#define LRU_DEL( c, e ) do { \
+	if ( e == e->bei_lruprev ) { \
+		(c)->c_lruhead = (c)->c_lrutail = NULL; \
+	} else { \
+		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 +126,28 @@ 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 )
 {
+
+	/* Already linked, ignore */
+	if ( ei->bei_lruprev )
+		return;
+
 	/* 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 );
+
+	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
@@ -121,7 +169,7 @@ bdb_cache_lru_link( Cache *cache, EntryInfo *ei )
 int
 bdb_cache_entry_db_relock(
 	struct bdb_info *bdb,
-	u_int32_t locker,
+	DB_TXN *txn,
 	EntryInfo *ei,
 	int rw,
 	int tryOnly,
@@ -145,7 +193,7 @@ bdb_cache_entry_db_relock(
 	list[1].lock = *lock;
 	list[1].mode = rw ? DB_LOCK_WRITE : DB_LOCK_READ;
 	list[1].obj = &lockobj;
-	rc = bdb->bi_dbenv->lock_vec(bdb->bi_dbenv, locker, tryOnly ? DB_LOCK_NOWAIT : 0,
+	rc = bdb->bi_dbenv->lock_vec(bdb->bi_dbenv, TXN_ID(txn), tryOnly ? DB_LOCK_NOWAIT : 0,
 		list, 2, NULL );
 
 	if (rc && !tryOnly) {
@@ -160,7 +208,7 @@ bdb_cache_entry_db_relock(
 }
 
 static int
-bdb_cache_entry_db_lock( struct bdb_info *bdb, u_int32_t locker, EntryInfo *ei,
+bdb_cache_entry_db_lock( struct bdb_info *bdb, DB_TXN *txn, EntryInfo *ei,
 	int rw, int tryOnly, DB_LOCK *lock )
 {
 #ifdef NO_DB_LOCK
@@ -180,7 +228,7 @@ bdb_cache_entry_db_lock( struct bdb_info *bdb, u_int32_t locker, EntryInfo *ei,
 	lockobj.data = &ei->bei_id;
 	lockobj.size = sizeof(ei->bei_id) + 1;
 
-	rc = LOCK_GET(bdb->bi_dbenv, locker, tryOnly ? DB_LOCK_NOWAIT : 0,
+	rc = LOCK_GET(bdb->bi_dbenv, TXN_ID(txn), tryOnly ? DB_LOCK_NOWAIT : 0,
 					&lockobj, db_rw, lock);
 	if (rc && !tryOnly) {
 		Debug( LDAP_DEBUG_TRACE,
@@ -206,6 +254,41 @@ bdb_cache_entry_db_unlock ( struct bdb_info *bdb, DB_LOCK *lock )
 #endif
 }
 
+void
+bdb_cache_return_entry_rw( struct bdb_info *bdb, Entry *e,
+	int rw, DB_LOCK *lock )
+{
+	EntryInfo *ei;
+	int free = 0;
+
+	ei = e->e_private;
+	if ( ei &&
+		( ei->bei_state & CACHE_ENTRY_NOT_CACHED ) &&
+		( bdb_cache_entryinfo_trylock( ei ) == 0 )) {
+		if ( ei->bei_state & CACHE_ENTRY_NOT_CACHED ) {
+			/* Releasing the entry can only be done when
+			 * we know that nobody else is using it, i.e we
+			 * should have an entry_db writelock.  But the
+			 * flag is only set by the thread that loads the
+			 * entry, and only if no other threads has found
+			 * it while it was working.  All other threads
+			 * clear the flag, which mean that we should be
+			 * the only thread using the entry if the flag
+			 * is set here.
+			 */
+			ei->bei_e = NULL;
+			ei->bei_state ^= CACHE_ENTRY_NOT_CACHED;
+			free = 1;
+		}
+		bdb_cache_entryinfo_unlock( ei );
+	}
+	bdb_cache_entry_db_unlock( bdb, lock );
+	if ( free ) {
+		e->e_private = NULL;
+		bdb_entry_return( e );
+	}
+}
+
 static int
 bdb_cache_entryinfo_destroy( EntryInfo *e )
 {
@@ -238,6 +321,14 @@ 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
@@ -252,8 +343,8 @@ bdb_entryinfo_add_internal(
 
 	ei2 = bdb_cache_entryinfo_new( &bdb->bi_cache );
 
-	ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
 	bdb_cache_entryinfo_lock( ei->bei_parent );
+	ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
 
 	ei2->bei_id = ei->bei_id;
 	ei2->bei_parent = ei->bei_parent;
@@ -265,10 +356,10 @@ 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
@@ -277,6 +368,8 @@ bdb_entryinfo_add_internal(
 		ei->bei_rdn.bv_val = NULL;
 #endif
 	} else {
+		int rc;
+
 		bdb->bi_cache.c_eiused++;
 		ber_dupbv( &ei2->bei_nrdn, &ei->bei_nrdn );
 
@@ -286,13 +379,15 @@ bdb_entryinfo_add_internal(
 		 */
 		if ( ei->bei_parent->bei_kids || !ei->bei_parent->bei_id )
 			bdb->bi_cache.c_leaves++;
-		avl_insert( &ei->bei_parent->bei_kids, ei2, bdb_rdn_cmp,
+		rc = avl_insert( &ei->bei_parent->bei_kids, ei2, bdb_rdn_cmp,
 			avl_dup_error );
 #ifdef BDB_HIER
-		ei->bei_parent->bei_ckids++;
+		/* it's possible for hdb_cache_find_parent to beat us to it */
+		if ( !rc ) {
+			ei->bei_parent->bei_ckids++;
+		}
 #endif
 	}
-	bdb_cache_lru_link( &bdb->bi_cache, ei2 );
 
 	*res = ei2;
 	return 0;
@@ -345,6 +440,7 @@ bdb_cache_find_ndn(
 		ei.bei_parent = eip;
 		ei2 = (EntryInfo *)avl_find( eip->bei_kids, &ei, bdb_rdn_cmp );
 		if ( !ei2 ) {
+			DB_LOCK lock;
 			int len = ei.bei_nrdn.bv_len;
 				
 			if ( BER_BVISEMPTY( ndn )) {
@@ -354,20 +450,32 @@ bdb_cache_find_ndn(
 
 			ei.bei_nrdn.bv_len = ndn->bv_len -
 				(ei.bei_nrdn.bv_val - ndn->bv_val);
+			eip->bei_finders++;
 			bdb_cache_entryinfo_unlock( eip );
 
-			rc = bdb_dn2id( op, txn, &ei.bei_nrdn, &ei );
+			BDB_LOG_PRINTF( bdb->bi_dbenv, NULL, "slapd Reading %s",
+				ei.bei_nrdn.bv_val );
+
+			lock.mode = DB_LOCK_NG;
+			rc = bdb_dn2id( op, &ei.bei_nrdn, &ei, txn, &lock );
 			if (rc) {
 				bdb_cache_entryinfo_lock( eip );
+				eip->bei_finders--;
+				bdb_cache_entry_db_unlock( bdb, &lock );
 				*res = eip;
 				return rc;
 			}
 
+			BDB_LOG_PRINTF( bdb->bi_dbenv, NULL, "slapd Read got %s(%d)",
+				ei.bei_nrdn.bv_val, ei.bei_id );
+
 			/* DN exists but needs to be added to cache */
 			ei.bei_nrdn.bv_len = len;
 			rc = bdb_entryinfo_add_internal( bdb, &ei, &ei2 );
 			/* add_internal left eip and c_rwlock locked */
+			eip->bei_finders--;
 			ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
+			bdb_cache_entry_db_unlock( bdb, &lock );
 			if ( rc ) {
 				*res = eip;
 				return rc;
@@ -382,8 +490,8 @@ bdb_cache_find_ndn(
 			*res = eip;
 			return DB_NOTFOUND;
 		}
-		bdb_cache_entryinfo_unlock( eip );
 		bdb_cache_entryinfo_lock( ei2 );
+		bdb_cache_entryinfo_unlock( eip );
 
 		eip = ei2;
 
@@ -411,21 +519,20 @@ bdb_cache_find_ndn(
 int
 hdb_cache_find_parent(
 	Operation *op,
-	DB_TXN *txn,
-	u_int32_t	locker,
+	DB_TXN	*txn,
 	ID id,
 	EntryInfo **res )
 {
 	struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
 	EntryInfo ei, eip, *ei2 = NULL, *ein = NULL, *eir = NULL;
-	int rc;
+	int rc, add;
 
 	ei.bei_id = id;
 	ei.bei_kids = NULL;
 	ei.bei_ckids = 0;
 
 	for (;;) {
-		rc = hdb_dn2id_parent( op, txn, locker, &ei, &eip.bei_id );
+		rc = hdb_dn2id_parent( op, txn, &ei, &eip.bei_id );
 		if ( rc ) break;
 
 		/* Save the previous node, if any */
@@ -442,69 +549,92 @@ hdb_cache_find_parent(
 		ein->bei_bdb = bdb;
 #endif
 		ei.bei_ckids = 0;
+		add = 1;
 		
 		/* This node is not fully connected yet */
 		ein->bei_state |= CACHE_ENTRY_NOT_LINKED;
 
+		/* If this is the first time, save this node
+		 * to be returned later.
+		 */
+		if ( eir == NULL ) {
+			eir = ein;
+			ein->bei_finders++;
+		}
+
+again:
 		/* 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;
+
+			if ( bdb_cache_entryinfo_trylock( eix )) {
+				ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
+				ldap_pvt_thread_yield();
+				goto again;
+			}
+			ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
 
 			/* 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 );
-			
-			/* Link in any kids we've already processed */
-			if ( ei2 ) {
-				bdb_cache_entryinfo_lock( ein );
-				avl_insert( &ein->bei_kids, (caddr_t)ei2,
-					bdb_rdn_cmp, avl_dup_error );
-				ein->bei_ckids++;
-				bdb_cache_entryinfo_unlock( ein );
+			bdb_cache_entryinfo_free( &bdb->bi_cache, ein );
+
+			/* if it was the node we were looking for, just return it */
+			if ( eir == ein ) {
+				*res = eix;
+				rc = 0;
+				break;
 			}
-		} else {
-			bdb_cache_lru_link( &bdb->bi_cache, ein );
-		}
 
-		/* If this is the first time, save this node
-		 * to be returned later.
-		 */
-		if ( eir == NULL ) eir = ein;
+			ein = ei2;
+			ei2 = eix;
+			add = 0;
+
+			/* otherwise, link up what we have and return */
+			goto gotparent;
+		}
 
 		/* If there was a previous node, link it to this one */
 		if ( ei2 ) ei2->bei_parent = ein;
 
 		/* Look for this node's parent */
+par2:
 		if ( eip.bei_id ) {
 			ei2 = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
 					(caddr_t) &eip, bdb_id_cmp );
 		} else {
 			ei2 = &bdb->bi_cache.c_dntree;
 		}
-		bdb->bi_cache.c_eiused++;
+		if ( ei2 && bdb_cache_entryinfo_trylock( ei2 )) {
+			ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
+			ldap_pvt_thread_yield();
+			ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
+			goto par2;
+		}
+		if ( add )
+			bdb->bi_cache.c_eiused++;
 		if ( ei2 && ( ei2->bei_kids || !ei2->bei_id ))
-				bdb->bi_cache.c_leaves++;
+			bdb->bi_cache.c_leaves++;
 		ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
 
+gotparent:
 		/* Got the parent, link in and we're done. */
 		if ( ei2 ) {
-			bdb_cache_entryinfo_lock( ei2 );
+			bdb_cache_entryinfo_lock( eir );
 			ein->bei_parent = ei2;
 
-			avl_insert( &ei2->bei_kids, (caddr_t)ein, bdb_rdn_cmp,
-				avl_dup_error);
-			ei2->bei_ckids++;
+			if ( avl_insert( &ei2->bei_kids, (caddr_t)ein, bdb_rdn_cmp,
+				avl_dup_error) == 0 )
+				ei2->bei_ckids++;
 
 			/* Reset all the state info */
 			for (ein = eir; ein != ei2; ein=ein->bei_parent)
 				ein->bei_state &= ~CACHE_ENTRY_NOT_LINKED;
 
 			bdb_cache_entryinfo_unlock( ei2 );
-			bdb_cache_entryinfo_lock( eir );
+			eir->bei_finders--;
 
 			*res = eir;
 			break;
@@ -554,48 +684,92 @@ 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, uint32_t locker )
+bdb_cache_lru_purge( struct bdb_info *bdb )
 {
-	DB_LOCK		lock;
-	EntryInfo *elru, *elnext;
-	int count, islocked;
+	DB_LOCK		lock, *lockp;
+	EntryInfo *elru, *elnext = NULL;
+	int islocked;
+	ID eicount, ecount;
+	ID count, efree, eifree = 0;
+#ifdef LDAP_DEBUG
+	int iter;
+#endif
 
-	/* 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 );
+	if ( bdb->bi_cache.c_cursize > bdb->bi_cache.c_maxsize ) {
+		efree = bdb->bi_cache.c_cursize - bdb->bi_cache.c_maxsize;
+		efree += bdb->bi_cache.c_minfree;
+	} else {
+		efree = 0;
+	}
+
+	/* maximum number of EntryInfo leaves to cache. In slapcat
+	 * we always free all leaf nodes.
+	 */
+
+	if ( slapMode & SLAP_TOOL_READONLY ) {
+		eifree = bdb->bi_cache.c_leaves;
+	} else if ( bdb->bi_cache.c_eimax &&
+		bdb->bi_cache.c_leaves > bdb->bi_cache.c_eimax ) {
+		eifree = bdb->bi_cache.c_minfree * 10;
+		if ( eifree >= bdb->bi_cache.c_leaves )
+			eifree /= 2;
+	}
+
+	if ( !efree && !eifree ) {
+		ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_lru_mutex );
+		bdb->bi_cache.c_purging = 0;
 		return;
 	}
 
+	if ( bdb->bi_cache.c_txn ) {
+		lockp = &lock;
+	} else {
+		lockp = NULL;
+	}
+
 	count = 0;
+	eicount = 0;
+	ecount = 0;
+#ifdef LDAP_DEBUG
+	iter = 0;
+#endif
+
 	/* 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.
-		 *
-		 * Also, if this node has no entry attached, skip it, there's
-		 * nothing to purge anyway.
 		 */
 		if (( elru->bei_state & ( CACHE_ENTRY_NOT_LINKED |
-			CACHE_ENTRY_DELETED | CACHE_ENTRY_LOADING )) ||
-			!elru->bei_e ) {
+			CACHE_ENTRY_DELETED | CACHE_ENTRY_LOADING |
+			CACHE_ENTRY_ONELEVEL )) ||
+			elru->bei_finders > 0 ) {
 			bdb_cache_entryinfo_unlock( elru );
-			continue;
+			goto bottom;
+		}
+
+		if ( bdb_cache_entryinfo_trylock( elru->bei_parent )) {
+			bdb_cache_entryinfo_unlock( elru );
+			goto bottom;
 		}
 
 		/* entryinfo is locked */
@@ -604,73 +778,83 @@ bdb_cache_lru_purge( struct bdb_info *bdb, uint32_t locker )
 		/* If we can successfully writelock it, then
 		 * the object is idle.
 		 */
-		if ( bdb_cache_entry_db_lock( bdb, locker, elru, 1, 1, &lock ) == 0 ) {
+		if ( bdb_cache_entry_db_lock( bdb,
+			bdb->bi_cache.c_txn, elru, 1, 1, lockp ) == 0 ) {
 
 			/* Free entry for this node if it's present */
 			if ( elru->bei_e ) {
-				elru->bei_e->e_private = NULL;
+				ecount++;
+
+				/* the cache may have gone over the limit while we
+				 * weren't looking, so double check.
+				 */
+				if ( !efree && ecount > bdb->bi_cache.c_maxsize )
+					efree = bdb->bi_cache.c_minfree;
+
+				if ( count < efree ) {
+					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++;
+					elru->bei_e = NULL;
+					count++;
+				} else {
+					/* Keep this node cached, skip to next */
+					bdb_cache_entry_db_unlock( bdb, lockp );
+					goto next;
+				}
 			}
-			bdb_cache_entry_db_unlock( bdb, &lock );
+			bdb_cache_entry_db_unlock( bdb, lockp );
 
-			/* 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 it is a leaf node, and we're over the limit, free it.
 			 */
-			if ( slapMode & SLAP_TOOL_READONLY ) {
-				if ( !elru->bei_kids ) {
-					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 */
-			}
+			if ( elru->bei_kids ) {
+				/* Drop from list, we ignore it... */
+				LRU_DEL( &bdb->bi_cache, elru );
+			} else if ( eicount < eifree ) {
+				/* 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;
+				eicount++;
+			}	/* Leave on list until we need to free it */
 		}
 
-		if ( islocked )
+next:
+		if ( islocked ) {
 			bdb_cache_entryinfo_unlock( elru );
+			bdb_cache_entryinfo_unlock( elru->bei_parent );
+		}
 
-		if ( count >= bdb->bi_cache.c_minfree ) {
-			ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_count_mutex );
-			bdb->bi_cache.c_cursize -= count;
-			ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_count_mutex );
+		if ( count >= efree && eicount >= eifree )
 			break;
-		}
+bottom:
+		if ( elnext == bdb->bi_cache.c_lruhead )
+			break;
+#ifdef LDAP_DEBUG
+		iter++;
+#endif
 	}
 
+	if ( count || ecount > bdb->bi_cache.c_cursize ) {
+		ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_count_mutex );
+		/* HACK: we seem to be losing track, fix up now */
+		if ( ecount > bdb->bi_cache.c_cursize )
+			bdb->bi_cache.c_cursize = ecount;
+		bdb->bi_cache.c_cursize -= count;
+		ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_count_mutex );
+	}
 	bdb->bi_cache.c_lruhead = elnext;
-	ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_head_mutex );
-}
-
-EntryInfo *
-bdb_cache_find_info(
-	struct bdb_info *bdb,
-	ID id )
-{
-	EntryInfo	ei = { 0 },
-			*ei2;
-
-	ei.bei_id = id;
-
-	ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
-	ei2 = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
-					(caddr_t) &ei, bdb_id_cmp );
-	ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
-	return ei2;
+	ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_lru_mutex );
+	bdb->bi_cache.c_purging = 0;
 }
 
 /*
  * cache_find_id - find an entry in the cache, given id.
- * The entry is locked for Read upon return. Call with islocked TRUE if
+ * The entry is locked for Read upon return. Call with flag ID_LOCKED if
  * the supplied *eip was already locked.
  */
 
@@ -680,8 +864,7 @@ bdb_cache_find_id(
 	DB_TXN	*tid,
 	ID				id,
 	EntryInfo	**eip,
-	int		islocked,
-	u_int32_t	locker,
+	int		flag,
 	DB_LOCK		*lock )
 {
 	struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
@@ -701,13 +884,13 @@ 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 )) {
+				int del = (*eip)->bei_state & CACHE_ENTRY_DELETED;
 				ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
 				/* If this node is being deleted, treat
 				 * as if the delete has already finished
 				 */
-				if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
+				if ( del ) {
 					return DB_NOTFOUND;
 				}
 				/* otherwise, wait for the info to free up */
@@ -723,7 +906,7 @@ again:	ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
 				ldap_pvt_thread_yield();
 				goto again;
 			}
-			islocked = 1;
+			flag |= ID_LOCKED;
 		}
 		ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
 	}
@@ -731,11 +914,11 @@ again:	ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
 	/* See if the ID exists in the database; add it to the cache if so */
 	if ( !*eip ) {
 #ifndef BDB_HIER
-		rc = bdb_id2entry( op->o_bd, tid, locker, id, &ep );
+		rc = bdb_id2entry( op->o_bd, tid, id, &ep );
 		if ( rc == 0 ) {
 			rc = bdb_cache_find_ndn( op, tid,
 				&ep->e_nname, eip );
-			if ( *eip ) islocked = 1;
+			if ( *eip ) flag |= ID_LOCKED;
 			if ( rc ) {
 				ep->e_private = NULL;
 #ifdef SLAP_ZONE_ALLOC
@@ -747,16 +930,27 @@ again:	ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
 			}
 		}
 #else
-		rc = hdb_cache_find_parent(op, tid, locker, id, eip );
-		if ( rc == 0 ) islocked = 1;
+		rc = hdb_cache_find_parent(op, tid, id, eip );
+		if ( rc == 0 ) flag |= ID_LOCKED;
 #endif
 	}
 
 	/* Ok, we found the info, do we have the entry? */
 	if ( rc == 0 ) {
+		if ( !( flag & ID_LOCKED )) {
+			bdb_cache_entryinfo_lock( *eip );
+			flag |= ID_LOCKED;
+		}
+
 		if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
 			rc = DB_NOTFOUND;
 		} else {
+			(*eip)->bei_finders++;
+			(*eip)->bei_state |= CACHE_ENTRY_REFERENCED;
+			if ( flag & ID_NOENTRY ) {
+				bdb_cache_entryinfo_unlock( *eip );
+				return 0;
+			}
 			/* Make sure only one thread tries to load the entry */
 load1:
 #ifdef SLAP_ZONE_ALLOC
@@ -770,22 +964,42 @@ load1:
 				load = 1;
 				(*eip)->bei_state |= CACHE_ENTRY_LOADING;
 			}
-			if ( islocked ) {
+
+			if ( !load ) {
+				/* Clear the uncached state if we are not
+				 * loading it, i.e it is already cached or
+				 * another thread is currently loading it.
+				 */
+				if ( (*eip)->bei_state & CACHE_ENTRY_NOT_CACHED ) {
+					(*eip)->bei_state &= ~CACHE_ENTRY_NOT_CACHED;
+					ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_count_mutex );
+					++bdb->bi_cache.c_cursize;
+					ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_count_mutex );
+				}
+				flag &= ~ID_NOCACHE;
+			}
+
+			if ( flag & ID_LOCKED ) {
 				bdb_cache_entryinfo_unlock( *eip );
-				islocked = 0;
+				flag ^= ID_LOCKED;
 			}
-			rc = bdb_cache_entry_db_lock( bdb, locker, *eip, load, 0, lock );
+			rc = bdb_cache_entry_db_lock( bdb, tid, *eip, load, 0, lock );
 			if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
 				rc = DB_NOTFOUND;
 				bdb_cache_entry_db_unlock( bdb, lock );
+				bdb_cache_entryinfo_lock( *eip );
+				(*eip)->bei_finders--;
+				bdb_cache_entryinfo_unlock( *eip );
 			} else if ( rc == 0 ) {
 				if ( load ) {
 					if ( !ep) {
-						rc = bdb_id2entry( op->o_bd, tid, locker, id, &ep );
+						rc = bdb_id2entry( op->o_bd, tid, id, &ep );
 					}
 					if ( rc == 0 ) {
 						ep->e_private = *eip;
 #ifdef BDB_HIER
+						while ( (*eip)->bei_state & CACHE_ENTRY_NOT_LINKED )
+							ldap_pvt_thread_yield();
 						bdb_fix_dn( ep, 0 );
 #endif
 						(*eip)->bei_e = ep;
@@ -793,25 +1007,32 @@ load1:
 						(*eip)->bei_zseq = *((ber_len_t *)ep - 2);
 #endif
 						ep = NULL;
+						bdb_cache_lru_link( bdb, *eip );
+						if (( flag & ID_NOCACHE ) &&
+							( bdb_cache_entryinfo_trylock( *eip ) == 0 )) {
+							/* Set the cached state only if no other thread
+							 * found the info while we were loading the entry.
+							 */
+							if ( (*eip)->bei_finders == 1 )
+								(*eip)->bei_state |= CACHE_ENTRY_NOT_CACHED;
+							bdb_cache_entryinfo_unlock( *eip );
+						}
 					}
 					if ( rc == 0 ) {
 						/* If we succeeded, downgrade back to a readlock. */
-						rc = bdb_cache_entry_db_relock( bdb, locker,
+						rc = bdb_cache_entry_db_relock( bdb, tid,
 							*eip, 0, 0, lock );
 					} else {
 						/* Otherwise, release the lock. */
 						bdb_cache_entry_db_unlock( bdb, lock );
 					}
-					bdb_cache_entryinfo_lock( *eip );
-					(*eip)->bei_state ^= CACHE_ENTRY_LOADING;
-					bdb_cache_entryinfo_unlock( *eip );
 				} else if ( !(*eip)->bei_e ) {
 					/* Some other thread is trying to load the entry,
 					 * wait for it to finish.
 					 */
 					bdb_cache_entry_db_unlock( bdb, lock );
 					bdb_cache_entryinfo_lock( *eip );
-					islocked = 1;
+					flag |= ID_LOCKED;
 					goto load1;
 #ifdef BDB_HIER
 				} else {
@@ -820,20 +1041,24 @@ load1:
 					rc = bdb_fix_dn( (*eip)->bei_e, 1 );
 					if ( rc ) {
 						bdb_cache_entry_db_relock( bdb,
-							locker, *eip, 1, 0, lock );
+							tid, *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,
-							locker, *eip, 0, 0, lock );
+							tid, *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 );
 			}
 		}
 	}
-	if ( islocked ) {
+	if ( flag & ID_LOCKED ) {
 		bdb_cache_entryinfo_unlock( *eip );
 	}
 	if ( ep ) {
@@ -847,15 +1072,22 @@ load1:
 	if ( rc == 0 ) {
 		int purge = 0;
 
-		if ( load ) {
+		if (( load && !( flag & ID_NOCACHE )) || bdb->bi_cache.c_eimax ) {
 			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 ( load && !( flag & ID_NOCACHE )) {
+				bdb->bi_cache.c_cursize++;
+				if ( !bdb->bi_cache.c_purging && bdb->bi_cache.c_cursize > bdb->bi_cache.c_maxsize ) {
+					purge = 1;
+					bdb->bi_cache.c_purging = 1;
+				}
+			} else if ( !bdb->bi_cache.c_purging && bdb->bi_cache.c_eimax && bdb->bi_cache.c_leaves > bdb->bi_cache.c_eimax ) {
 				purge = 1;
+				bdb->bi_cache.c_purging = 1;
+			}
 			ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_count_mutex );
 		}
 		if ( purge )
-			bdb_cache_lru_purge( bdb, locker );
+			bdb_cache_lru_purge( bdb );
 	}
 
 #ifdef SLAP_ZONE_ALLOC
@@ -895,7 +1127,7 @@ bdb_cache_add(
 	EntryInfo *eip,
 	Entry *e,
 	struct berval *nrdn,
-	u_int32_t locker,
+	DB_TXN *txn,
 	DB_LOCK *lock )
 {
 	EntryInfo *new, ei;
@@ -912,7 +1144,7 @@ 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, locker, &ei, 0, 0, lock );
+	rc = bdb_cache_entry_db_lock( bdb, txn, &ei, 0, 0, lock );
 	if ( rc ) {
 		bdb_cache_entryinfo_unlock( eip );
 		return rc;
@@ -928,6 +1160,12 @@ bdb_cache_add(
 	if ( eip->bei_dkids ) eip->bei_dkids++;
 #endif
 
+	if (eip->bei_parent) {
+		bdb_cache_entryinfo_lock( eip->bei_parent );
+		eip->bei_parent->bei_state &= ~CACHE_ENTRY_NO_GRANDKIDS;
+		bdb_cache_entryinfo_unlock( eip->bei_parent );
+	}
+
 	rc = bdb_entryinfo_add_internal( bdb, &ei, &new );
 	/* bdb_csn_commit can cause this when adding the database root entry */
 	if ( new->bei_e ) {
@@ -942,20 +1180,22 @@ bdb_cache_add(
 	e->e_private = new;
 	new->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
 	eip->bei_state &= ~CACHE_ENTRY_NO_KIDS;
-	if (eip->bei_parent) {
-		eip->bei_parent->bei_state &= ~CACHE_ENTRY_NO_GRANDKIDS;
-	}
 	bdb_cache_entryinfo_unlock( eip );
 
 	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, locker );
+		bdb_cache_lru_purge( bdb );
 
 	return rc;
 }
@@ -965,13 +1205,13 @@ bdb_cache_modify(
 	struct bdb_info *bdb,
 	Entry *e,
 	Attribute *newAttrs,
-	u_int32_t locker,
+	DB_TXN *txn,
 	DB_LOCK *lock )
 {
 	EntryInfo *ei = BEI(e);
 	int rc;
 	/* Get write lock on data */
-	rc = bdb_cache_entry_db_relock( bdb, locker, ei, 1, 0, lock );
+	rc = bdb_cache_entry_db_relock( bdb, txn, 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.
@@ -995,7 +1235,7 @@ bdb_cache_modrdn(
 	struct berval *nrdn,
 	Entry *new,
 	EntryInfo *ein,
-	u_int32_t locker,
+	DB_TXN *txn,
 	DB_LOCK *lock )
 {
 	EntryInfo *ei = BEI(e), *pei;
@@ -1005,7 +1245,7 @@ bdb_cache_modrdn(
 #endif
 
 	/* Get write lock on data */
-	rc =  bdb_cache_entry_db_relock( bdb, locker, ei, 1, 0, lock );
+	rc =  bdb_cache_entry_db_relock( bdb, txn, ei, 1, 0, lock );
 	if ( rc ) return rc;
 
 	/* If we've done repeated mods on a cached entry, then e_attrs
@@ -1031,9 +1271,6 @@ bdb_cache_modrdn(
 	free( ei->bei_nrdn.bv_val );
 	ber_dupbv( &ei->bei_nrdn, nrdn );
 
-	if ( !pei->bei_kids )
-		pei->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
-
 #ifdef BDB_HIER
 	free( ei->bei_rdn.bv_val );
 
@@ -1044,8 +1281,16 @@ bdb_cache_modrdn(
 		rdn.bv_len = ptr - rdn.bv_val;
 	}
 	ber_dupbv( &ei->bei_rdn, &rdn );
-	pei->bei_ckids--;
-	if ( pei->bei_dkids ) pei->bei_dkids--;
+
+	/* If new parent, decrement kid counts */
+	if ( ein ) {
+		pei->bei_ckids--;
+		if ( pei->bei_dkids ) {
+			pei->bei_dkids--;
+			if ( pei->bei_dkids < 2 )
+				pei->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
+		}
+	}
 #endif
 
 	if (!ein) {
@@ -1054,27 +1299,32 @@ bdb_cache_modrdn(
 		ei->bei_parent = ein;
 		bdb_cache_entryinfo_unlock( pei );
 		bdb_cache_entryinfo_lock( ein );
-	}
-	/* parent now has kids */
-	if ( ein->bei_state & CACHE_ENTRY_NO_KIDS )
-		ein->bei_state ^= CACHE_ENTRY_NO_KIDS;
 
+		/* new parent now has kids */
+		if ( ein->bei_state & CACHE_ENTRY_NO_KIDS )
+			ein->bei_state ^= CACHE_ENTRY_NO_KIDS;
+		/* grandparent has grandkids */
+		if ( ein->bei_parent )
+			ein->bei_parent->bei_state &= ~CACHE_ENTRY_NO_GRANDKIDS;
 #ifdef BDB_HIER
-	/* parent might now have grandkids */
-	if ( ein->bei_state & CACHE_ENTRY_NO_GRANDKIDS &&
-		!(ei->bei_state & (CACHE_ENTRY_NO_KIDS)))
-		ein->bei_state ^= CACHE_ENTRY_NO_GRANDKIDS;
+		/* parent might now have grandkids */
+		if ( ein->bei_state & CACHE_ENTRY_NO_GRANDKIDS &&
+			!(ei->bei_state & CACHE_ENTRY_NO_KIDS))
+			ein->bei_state ^= CACHE_ENTRY_NO_GRANDKIDS;
 
-	{
-		/* Record the generation number of this change */
-		ldap_pvt_thread_mutex_lock( &bdb->bi_modrdns_mutex );
-		bdb->bi_modrdns++;
-		ei->bei_modrdns = bdb->bi_modrdns;
-		ldap_pvt_thread_mutex_unlock( &bdb->bi_modrdns_mutex );
+		ein->bei_ckids++;
+		if ( ein->bei_dkids ) ein->bei_dkids++;
+#endif
 	}
-	ein->bei_ckids++;
-	if ( ein->bei_dkids ) ein->bei_dkids++;
+
+#ifdef BDB_HIER
+	/* Record the generation number of this change */
+	ldap_pvt_thread_mutex_lock( &bdb->bi_modrdns_mutex );
+	bdb->bi_modrdns++;
+	ei->bei_modrdns = bdb->bi_modrdns;
+	ldap_pvt_thread_mutex_unlock( &bdb->bi_modrdns_mutex );
 #endif
+
 	avl_insert( &ein->bei_kids, ei, bdb_rdn_cmp, avl_dup_error );
 	bdb_cache_entryinfo_unlock( ein );
 	return rc;
@@ -1090,23 +1340,42 @@ int
 bdb_cache_delete(
 	struct bdb_info *bdb,
     Entry		*e,
-    u_int32_t	locker,
+    DB_TXN *txn,
     DB_LOCK	*lock )
 {
 	EntryInfo *ei = BEI(e);
-	int	rc;
+	int	rc, busy = 0;
 
 	assert( e->e_private != NULL );
 
+	/* Lock the entry's info */
+	bdb_cache_entryinfo_lock( ei );
+
 	/* Set this early, warn off any queriers */
 	ei->bei_state |= CACHE_ENTRY_DELETED;
 
-	/* Lock the entry's info */
-	bdb_cache_entryinfo_lock( ei );
+	if (( ei->bei_state & ( CACHE_ENTRY_NOT_LINKED |
+		CACHE_ENTRY_LOADING | CACHE_ENTRY_ONELEVEL )) ||
+		ei->bei_finders > 0 )
+		busy = 1;
+
+	bdb_cache_entryinfo_unlock( ei );
+
+	while ( busy ) {
+		ldap_pvt_thread_yield();
+		busy = 0;
+		bdb_cache_entryinfo_lock( ei );
+		if (( ei->bei_state & ( CACHE_ENTRY_NOT_LINKED |
+			CACHE_ENTRY_LOADING | CACHE_ENTRY_ONELEVEL )) ||
+			ei->bei_finders > 0 )
+			busy = 1;
+		bdb_cache_entryinfo_unlock( ei );
+	}
 
 	/* Get write lock on the data */
-	rc = bdb_cache_entry_db_relock( bdb, locker, ei, 1, 0, lock );
+	rc = bdb_cache_entry_db_relock( bdb, txn, ei, 1, 0, lock );
 	if ( rc ) {
+		bdb_cache_entryinfo_lock( ei );
 		/* couldn't lock, undo and give up */
 		ei->bei_state ^= CACHE_ENTRY_DELETED;
 		bdb_cache_entryinfo_unlock( ei );
@@ -1117,14 +1386,15 @@ bdb_cache_delete(
 		e->e_id, 0, 0 );
 
 	/* set lru mutex */
-	ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_head_mutex );
+	ldap_pvt_thread_mutex_lock( &bdb->bi_cache.c_lru_mutex );
 
+	bdb_cache_entryinfo_lock( ei->bei_parent );
+	bdb_cache_entryinfo_lock( ei );
 	rc = bdb_cache_delete_internal( &bdb->bi_cache, e->e_private, 1 );
+	bdb_cache_entryinfo_unlock( ei );
 
 	/* free lru mutex */
-	ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_head_mutex );
-
-	/* Leave entry info locked */
+	ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.c_lru_mutex );
 
 	return( rc );
 }
@@ -1134,6 +1404,11 @@ bdb_cache_delete_cleanup(
 	Cache *cache,
 	EntryInfo *ei )
 {
+	/* Enter with ei locked */
+
+	/* already freed? */
+	if ( !ei->bei_parent ) return;
+
 	if ( ei->bei_e ) {
 		ei->bei_e->e_private = NULL;
 #ifdef SLAP_ZONE_ALLOC
@@ -1144,24 +1419,8 @@ 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_unlock( ei );
+	bdb_cache_entryinfo_free( cache, ei );
 }
 
 static int
@@ -1173,8 +1432,11 @@ bdb_cache_delete_internal(
 	int rc = 0;	/* return code */
 	int decr_leaf = 0;
 
-	/* Lock the parent's kids tree */
-	bdb_cache_entryinfo_lock( e->bei_parent );
+	/* already freed? */
+	if ( !e->bei_parent ) {
+		assert(0);
+		return -1;
+	}
 
 #ifdef BDB_HIER
 	e->bei_parent->bei_ckids--;
@@ -1185,12 +1447,11 @@ bdb_cache_delete_internal(
 		== NULL )
 	{
 		rc = -1;
+		assert(0);
 	}
 	if ( e->bei_parent->bei_kids )
 		decr_leaf = 1;
 
-	bdb_cache_entryinfo_unlock( e->bei_parent );
-
 	ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
 	/* id tree */
 	if ( avl_delete( &cache->c_idtree, (caddr_t) e, bdb_id_cmp )) {
@@ -1199,20 +1460,14 @@ bdb_cache_delete_internal(
 			cache->c_leaves--;
 	} else {
 		rc = -1;
+		assert(0);
 	}
 	ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
+	bdb_cache_entryinfo_unlock( e->bei_parent );
 
 	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 );
@@ -1248,7 +1503,7 @@ 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 );
 
@@ -1267,7 +1522,7 @@ 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 );
 }
@@ -1279,11 +1534,11 @@ bdb_lru_print( Cache *cache )
 {
 	EntryInfo	*e;
 
-	fprintf( stderr, "LRU circle head: %p\n", cache->c_lruhead );
+	fprintf( stderr, "LRU circle head: %p\n", (void *) 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 );
+			(void *) e, (void *) e->bei_e, e->bei_id, e->bei_nrdn.bv_val );
 		e = e->bei_lrunext;
 		if ( e == cache->c_lrutail )
 			break;
@@ -1291,46 +1546,60 @@ bdb_lru_print( Cache *cache )
 	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 );
+			(void *) e, (void *) e->bei_e, e->bei_id, e->bei_nrdn.bv_val );
 		e = e->bei_lruprev;
 		if ( e == cache->c_lrutail )
 			break;
 	}
 }
+
+static int
+bdb_entryinfo_print(void *data, void *arg)
+{
+	EntryInfo *e = data;
+	fprintf( stderr, "\t%p, %p id %ld rdn \"%s\"\n",
+		(void *) e, (void *) e->bei_e, e->bei_id, e->bei_nrdn.bv_val );
+	return 0;
+}
+
+static void
+bdb_idtree_print(Cache *cache)
+{
+	avl_apply( cache->c_idtree, bdb_entryinfo_print, NULL, -1, AVL_INORDER );
+}
 #endif
 #endif
 
-#ifdef BDB_REUSE_LOCKERS
 static void
-bdb_locker_id_free( void *key, void *data )
+bdb_reader_free( void *key, void *data )
 {
-	DB_ENV *env = key;
-	u_int32_t lockid = (long)data;
-	int rc;
+	/* DB_ENV *env = key; */
+	DB_TXN *txn = data;
+
+	if ( txn ) TXN_ABORT( txn );
+}
+
+/* free up any keys used by the main thread */
+void
+bdb_reader_flush( DB_ENV *env )
+{
+	void *data;
+	void *ctx = ldap_pvt_thread_pool_context();
 
-	rc = XLOCK_ID_FREE( env, lockid );
-	if ( rc == EINVAL ) {
-		DB_LOCKREQ lr;
-		Debug( LDAP_DEBUG_ANY,
-			"bdb_locker_id_free: %lu err %s(%d)\n",
-			(unsigned long) lockid, db_strerror(rc), rc );
-		/* release all locks held by this locker. */
-		lr.op = DB_LOCK_PUT_ALL;
-		lr.obj = NULL;
-		env->lock_vec( env, lockid, 0, &lr, 1, NULL );
-		XLOCK_ID_FREE( env, lockid );
+	if ( !ldap_pvt_thread_pool_getkey( ctx, env, &data, NULL ) ) {
+		ldap_pvt_thread_pool_setkey( ctx, env, NULL, 0, NULL, NULL );
+		bdb_reader_free( env, data );
 	}
 }
 
 int
-bdb_locker_id( Operation *op, DB_ENV *env, u_int32_t *locker )
+bdb_reader_get( Operation *op, DB_ENV *env, DB_TXN **txn )
 {
 	int i, rc;
-	u_int32_t lockid;
 	void *data;
 	void *ctx;
 
-	if ( !env || !locker ) return -1;
+	if ( !env || !txn ) return -1;
 
 	/* If no op was provided, try to find the ctx anyway... */
 	if ( op ) {
@@ -1341,31 +1610,29 @@ bdb_locker_id( Operation *op, DB_ENV *env, u_int32_t *locker )
 
 	/* Shouldn't happen unless we're single-threaded */
 	if ( !ctx ) {
-		*locker = 0;
+		*txn = NULL;
 		return 0;
 	}
 
 	if ( ldap_pvt_thread_pool_getkey( ctx, env, &data, NULL ) ) {
 		for ( i=0, rc=1; rc != 0 && i<4; i++ ) {
-			rc = XLOCK_ID( env, &lockid );
+			rc = TXN_BEGIN( env, NULL, txn, DB_READ_COMMITTED );
 			if (rc) ldap_pvt_thread_yield();
 		}
 		if ( rc != 0) {
 			return rc;
 		}
-		data = (void *)((long)lockid);
+		data = *txn;
 		if ( ( rc = ldap_pvt_thread_pool_setkey( ctx, env,
-			data, bdb_locker_id_free ) ) ) {
-			XLOCK_ID_FREE( env, lockid );
-			Debug( LDAP_DEBUG_ANY, "bdb_locker_id: err %s(%d)\n",
+			data, bdb_reader_free, NULL, NULL ) ) ) {
+			TXN_ABORT( *txn );
+			Debug( LDAP_DEBUG_ANY, "bdb_reader_get: err %s(%d)\n",
 				db_strerror(rc), rc, 0 );
 
 			return rc;
 		}
 	} else {
-		lockid = (long)data;
+		*txn = data;
 	}
-	*locker = lockid;
 	return 0;
 }
-#endif /* BDB_REUSE_LOCKERS */