X-Git-Url: https://git.sur5r.net/?a=blobdiff_plain;f=servers%2Fslapd%2Fback-bdb%2Fcache.c;h=b470762ea91438b6f9042d601f68b8b7f2481755;hb=cfc8c67523837451fe2810755bf5e5cbf791ddac;hp=f68f89e4f01d4759183ca5646ac47ed7049ccb9a;hpb=f696f7ab55cbd526bc6e008aa3c89f41bf2c8fb9;p=openldap

diff --git a/servers/slapd/back-bdb/cache.c b/servers/slapd/back-bdb/cache.c
index f68f89e4f0..b470762ea9 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 2000-2004 The OpenLDAP Foundation.
+ * Copyright 2000-2008 The OpenLDAP Foundation.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -26,10 +26,19 @@
 
 #include "back-bdb.h"
 
-static int	bdb_cache_delete_internal(Cache *cache, EntryInfo *e);
+#include "ldap_rq.h"
+
+#ifdef BDB_HIER
+#define bdb_cache_lru_add	hdb_cache_lru_add
+#endif
+static void bdb_cache_lru_add( struct bdb_info *bdb, EntryInfo *ei );
+
+static int	bdb_cache_delete_internal(Cache *cache, EntryInfo *e, int decr);
 #ifdef LDAP_DEBUG
+#ifdef SLAPD_UNUSED
 static void	bdb_lru_print(Cache *cache);
 #endif
+#endif
 
 static EntryInfo *
 bdb_cache_entryinfo_new( Cache *cache )
@@ -86,16 +95,10 @@ bdb_cache_entry_db_relock(
 	rc = env->lock_vec(env, locker, tryOnly ? DB_LOCK_NOWAIT : 0,
 		list, 2, NULL );
 
-	if (rc) {
-#ifdef NEW_LOGGING
-		LDAP_LOG( CACHE, DETAIL1, 
-			"bdb_cache_entry_db_relock: entry %ld, rw %d, rc %d\n",
-			ei->bei_id, rw, rc );
-#else
+	if (rc && !tryOnly) {
 		Debug( LDAP_DEBUG_TRACE,
 			"bdb_cache_entry_db_relock: entry %ld, rw %d, rc %d\n",
 			ei->bei_id, rw, rc );
-#endif
 	} else {
 		*lock = list[1].lock;
 	}
@@ -126,16 +129,10 @@ bdb_cache_entry_db_lock( DB_ENV *env, u_int32_t locker, EntryInfo *ei,
 
 	rc = LOCK_GET(env, locker, tryOnly ? DB_LOCK_NOWAIT : 0,
 					&lockobj, db_rw, lock);
-	if (rc) {
-#ifdef NEW_LOGGING
-		LDAP_LOG( CACHE, DETAIL1, 
-			"bdb_cache_entry_db_lock: entry %ld, rw %d, rc %d\n",
-			ei->bei_id, rw, rc );
-#else
+	if (rc && !tryOnly) {
 		Debug( LDAP_DEBUG_TRACE,
 			"bdb_cache_entry_db_lock: entry %ld, rw %d, rc %d\n",
 			ei->bei_id, rw, rc );
-#endif
 	}
 	return rc;
 #endif /* NO_THREADS */
@@ -149,7 +146,7 @@ bdb_cache_entry_db_unlock ( DB_ENV *env, DB_LOCK *lock )
 #else
 	int rc;
 
-	if ( !lock ) return 0;
+	if ( !lock || lock->mode == DB_LOCK_NG ) return 0;
 
 	rc = LOCK_PUT ( env, lock );
 	return rc;
@@ -179,6 +176,7 @@ bdb_cache_entryinfo_destroy( EntryInfo *e )
 	} else { \
 		(cache)->c_lrutail = (ei)->bei_lruprev; \
 	} \
+	(ei)->bei_lrunext = (ei)->bei_lruprev = NULL; \
 } while(0)
 
 #define LRU_ADD( cache, ei ) do { \
@@ -188,8 +186,10 @@ bdb_cache_entryinfo_destroy( EntryInfo *e )
 	} \
 	(cache)->c_lruhead = (ei); \
 	(ei)->bei_lruprev = NULL; \
-	if ( (cache)->c_lrutail == NULL ) { \
-		(cache)->c_lrutail = (ei); \
+	if ( !ldap_pvt_thread_mutex_trylock( &(cache)->lru_tail_mutex )) { \
+		if ( (cache)->c_lrutail == NULL ) \
+			(cache)->c_lrutail = (ei); \
+		ldap_pvt_thread_mutex_unlock( &(cache)->lru_tail_mutex ); \
 	} \
 } while(0)
 
@@ -235,6 +235,9 @@ bdb_entryinfo_add_internal(
 #ifdef BDB_HIER
 	ei2->bei_rdn = ei->bei_rdn;
 #endif
+#ifdef SLAP_ZONE_ALLOC
+	ei2->bei_bdb = bdb;
+#endif
 
 	/* Add to cache ID tree */
 	if (avl_insert( &bdb->bi_cache.c_idtree, ei2, bdb_id_cmp, avl_dup_error )) {
@@ -249,9 +252,24 @@ 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 );
-		avl_insert( &ei->bei_parent->bei_kids, ei2, bdb_rdn_cmp,
+
+		/* This is a new leaf node. But if parent had no kids, then it was
+		 * a leaf and we would be decrementing that. So, only increment if
+		 * the parent already has kids.
+		 */
+		if ( ei->bei_parent->bei_kids || !ei->bei_parent->bei_id )
+			bdb->bi_cache.c_leaves++;
+		rc = avl_insert( &ei->bei_parent->bei_kids, ei2, bdb_rdn_cmp,
 			avl_dup_error );
+ 		if ( rc ) {
+ 			/* This should never happen; entry cache is corrupt */
+ 			bdb->bi_dbenv->log_flush( bdb->bi_dbenv, NULL );
+ 			assert( !rc );
+ 		}
 #ifdef BDB_HIER
 		ei->bei_parent->bei_ckids++;
 #endif
@@ -270,7 +288,7 @@ bdb_entryinfo_add_internal(
 int
 bdb_cache_find_ndn(
 	Operation	*op,
-	DB_TXN		*txn,
+	u_int32_t		locker,
 	struct berval	*ndn,
 	EntryInfo	**res )
 {
@@ -307,15 +325,23 @@ 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 )) {
+				*res = eip;
+				return LDAP_SUCCESS;
+			}
+
 			ei.bei_nrdn.bv_len = ndn->bv_len -
 				(ei.bei_nrdn.bv_val - ndn->bv_val);
 			bdb_cache_entryinfo_unlock( eip );
 
-			rc = bdb_dn2id( op, txn, &ei.bei_nrdn, &ei );
+			lock.mode = DB_LOCK_NG;
+			rc = bdb_dn2id( op, &ei.bei_nrdn, &ei, locker, &lock );
 			if (rc) {
 				bdb_cache_entryinfo_lock( eip );
+				bdb_cache_entry_db_unlock( bdb->bi_dbenv, &lock );
 				*res = eip;
 				return rc;
 			}
@@ -325,6 +351,7 @@ bdb_cache_find_ndn(
 			rc = bdb_entryinfo_add_internal( bdb, &ei, &ei2 );
 			/* add_internal left eip and c_rwlock locked */
 			ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
+			bdb_cache_entry_db_unlock( bdb->bi_dbenv, &lock );
 			if ( rc ) {
 				*res = eip;
 				return rc;
@@ -365,25 +392,24 @@ bdb_cache_find_ndn(
 /* Walk up the tree from a child node, looking for an ID that's already
  * been linked into the cache.
  */
-static int
+int
 hdb_cache_find_parent(
 	Operation *op,
-	DB_TXN *txn,
+	u_int32_t	locker,
 	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;
-	char ndn[SLAP_LDAPDN_MAXLEN];
-	ID parent;
 	int rc;
-	int addlru = 1;
+	int addlru = 0;
 
 	ei.bei_id = id;
 	ei.bei_kids = NULL;
+	ei.bei_ckids = 0;
 
 	for (;;) {
-		rc = hdb_dn2id_parent( op, txn, &ei, &eip.bei_id );
+		rc = hdb_dn2id_parent( op, locker, &ei, &eip.bei_id );
 		if ( rc ) break;
 
 		/* Save the previous node, if any */
@@ -395,6 +421,11 @@ hdb_cache_find_parent(
 		ein->bei_kids = ei.bei_kids;
 		ein->bei_nrdn = ei.bei_nrdn;
 		ein->bei_rdn = ei.bei_rdn;
+		ein->bei_ckids = ei.bei_ckids;
+#ifdef SLAP_ZONE_ALLOC
+		ein->bei_bdb = bdb;
+#endif
+		ei.bei_ckids = 0;
 		
 		/* This node is not fully connected yet */
 		ein->bei_state = CACHE_ENTRY_NOT_LINKED;
@@ -416,12 +447,11 @@ hdb_cache_find_parent(
 				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 );
 			}
+			addlru = 0;
 
-			if ( !eir ) {
-				addlru = 0;
-			}
 		}
 
 		/* If this is the first time, save this node
@@ -439,14 +469,24 @@ hdb_cache_find_parent(
 		} else {
 			ei2 = &bdb->bi_cache.c_dntree;
 		}
+		bdb->bi_cache.c_eiused++;
+		if ( ei2 && ( ei2->bei_kids || !ei2->bei_id ))
+				bdb->bi_cache.c_leaves++;
 		ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
 
+		if ( addlru ) {
+			ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_head_mutex );
+			bdb_cache_lru_add( bdb, ein );
+		}
+		addlru = 1;
+
 		/* Got the parent, link in and we're done. */
 		if ( ei2 ) {
 			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++;
 			bdb_cache_entryinfo_unlock( ei2 );
 			bdb_cache_entryinfo_lock( eir );
 
@@ -458,6 +498,7 @@ hdb_cache_find_parent(
 		}
 		ei.bei_kids = NULL;
 		ei.bei_id = eip.bei_id;
+		ei.bei_ckids = 1;
 		avl_insert( &ei.bei_kids, (caddr_t)ein, bdb_rdn_cmp,
 			avl_dup_error );
 	}
@@ -500,61 +541,94 @@ int hdb_cache_load(
 }
 #endif
 
-/* caller must have lru_mutex locked. mutex
+/* caller must have lru_head_mutex locked. mutex
  * will be unlocked on return.
  */
 static void
 bdb_cache_lru_add(
 	struct bdb_info *bdb,
-	u_int32_t	locker,
 	EntryInfo *ei )
 {
 	DB_LOCK		lock, *lockp;
+	EntryInfo *elru, *elprev;
+	int count = 0;
+
+	LRU_ADD( &bdb->bi_cache, ei );
+	ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_head_mutex );
 
-	if ( locker ) {
+	/* See if we're above the cache size limit */
+	if ( bdb->bi_cache.c_cursize <= bdb->bi_cache.c_maxsize )
+		return;
+
+	if ( bdb->bi_cache.c_locker ) {
 		lockp = &lock;
 	} else {
 		lockp = NULL;
 	}
 
-	/* See if we're above the cache size limit */
-	if ( bdb->bi_cache.c_cursize > bdb->bi_cache.c_maxsize ) {
-		EntryInfo *elru, *elprev;
-		int i = 0;
+	/* Don't bother if we can't get the lock */
+	if ( ldap_pvt_thread_mutex_trylock( &bdb->bi_cache.lru_tail_mutex ) )
+		return;
 
-		/* Look for an unused entry to remove */
-		for (elru = bdb->bi_cache.c_lrutail; elru; elru = elprev, i++ ) {
-			elprev = elru->bei_lruprev;
+	/* Look for an unused entry to remove */
+	for (elru = bdb->bi_cache.c_lrutail; elru; elru = elprev ) {
+		elprev = elru->bei_lruprev;
+
+		/* If we can successfully writelock it, then
+		 * the object is idle.
+		 */
+		if ( bdb_cache_entry_db_lock( bdb->bi_dbenv,
+				bdb->bi_cache.c_locker, elru, 1, 1, lockp ) == 0 ) {
 
-			/* Too many probes, not enough idle, give up */
-			if (i > 10) break;
 
-			/* If we can successfully writelock it, then
-			 * the object is idle.
+			/* If this node is in the process of linking into the cache,
+			 * or this node is being deleted, skip it.
 			 */
-			if ( bdb_cache_entry_db_lock( bdb->bi_dbenv, bdb->bi_cache.c_locker, elru, 1, 1,
-				lockp ) == 0 ) {
-				/* If there's no entry, or this node is in
-				 * the process of linking into the cache,
-				 * skip it.
-				 */
-				if ( !elru->bei_e || (elru->bei_state & CACHE_ENTRY_NOT_LINKED) ) {
-					bdb_cache_entry_db_unlock( bdb->bi_dbenv, lockp );
-					continue;
-				}
-				LRU_DELETE( &bdb->bi_cache, elru );
+			if ( elru->bei_state &
+				( CACHE_ENTRY_NOT_LINKED | CACHE_ENTRY_DELETED )) {
+				bdb_cache_entry_db_unlock( bdb->bi_dbenv, lockp );
+				continue;
+			}
+			/* 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 );
+#else
 				bdb_entry_return( elru->bei_e );
+#endif
 				elru->bei_e = NULL;
-				bdb_cache_entry_db_unlock( bdb->bi_dbenv, lockp );
-				--bdb->bi_cache.c_cursize;
-				if (bdb->bi_cache.c_cursize < bdb->bi_cache.c_maxsize)
-					break;
+				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 ) {
+					/* This does LRU_DELETE for us */
+					bdb_cache_delete_internal( &bdb->bi_cache, elru, 0 );
+					bdb_cache_delete_cleanup( &bdb->bi_cache, elru );
+				}
+				/* Leave node on LRU list for a future pass */
+			} else {
+				LRU_DELETE( &bdb->bi_cache, elru );
+			}
+			bdb_cache_entry_db_unlock( bdb->bi_dbenv, lockp );
+
+			if ( count >= bdb->bi_cache.c_minfree ) {
+				ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
+				bdb->bi_cache.c_cursize -= count;
+				ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
+				break;
 			}
 		}
 	}
-	LRU_ADD( &bdb->bi_cache, ei );
-	ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_mutex );
+
+	ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_tail_mutex );
 }
 
 EntryInfo *
@@ -562,7 +636,8 @@ bdb_cache_find_info(
 	struct bdb_info *bdb,
 	ID id )
 {
-	EntryInfo ei, *ei2;
+	EntryInfo	ei = { 0 },
+			*ei2;
 
 	ei.bei_id = id;
 
@@ -591,11 +666,14 @@ bdb_cache_find_id(
 {
 	struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
 	Entry	*ep = NULL;
-	int	rc = 0;
-	EntryInfo ei;
+	int	rc = 0, load = 0;
+	EntryInfo ei = { 0 };
 
 	ei.bei_id = id;
 
+#ifdef SLAP_ZONE_ALLOC
+	slap_zh_rlock(bdb->bi_cache.c_zctx);
+#endif
 	/* If we weren't given any info, see if we have it already cached */
 	if ( !*eip ) {
 again:	ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
@@ -633,60 +711,95 @@ 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, id, &ep );
+		rc = bdb_id2entry( op->o_bd, tid, locker, id, &ep );
 		if ( rc == 0 ) {
-			rc = bdb_cache_find_ndn( op, tid,
+			rc = bdb_cache_find_ndn( op, locker,
 				&ep->e_nname, eip );
 			if ( *eip ) islocked = 1;
 			if ( rc ) {
+				ep->e_private = NULL;
+#ifdef SLAP_ZONE_ALLOC
+				bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
+#else
 				bdb_entry_return( ep );
+#endif
 				ep = NULL;
 			}
 		}
 #else
-		rc = hdb_cache_find_parent(op, tid, id, eip );
-		if ( rc == 0 && *eip ) islocked = 1;
+		rc = hdb_cache_find_parent(op, locker, id, eip );
+		if ( rc == 0 ) islocked = 1;
 #endif
 	}
 
 	/* Ok, we found the info, do we have the entry? */
-	if ( *eip && rc == 0 ) {
+	if ( rc == 0 ) {
 		if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
 			rc = DB_NOTFOUND;
 		} else {
-			rc = bdb_cache_entry_db_lock( bdb->bi_dbenv, locker, *eip, 0, 0, lock );
-			/* entry is protected now, we don't need to hold the entryinfo */
+			/* Make sure only one thread tries to load the entry */
+load1:
+#ifdef SLAP_ZONE_ALLOC
+			if ((*eip)->bei_e && !slap_zn_validate(
+					bdb->bi_cache.c_zctx, (*eip)->bei_e, (*eip)->bei_zseq)) {
+				(*eip)->bei_e = NULL;
+				(*eip)->bei_zseq = 0;
+			}
+#endif
+			if ( !(*eip)->bei_e && !((*eip)->bei_state & CACHE_ENTRY_LOADING)) {
+				load = 1;
+				(*eip)->bei_state |= CACHE_ENTRY_LOADING;
+			}
 			if ( islocked ) {
 				bdb_cache_entryinfo_unlock( *eip );
 				islocked = 0;
 			}
-			if ( rc == 0 ) {
-				if ( !(*eip)->bei_e ) {
-					if (!ep) {
-						rc = bdb_id2entry( op->o_bd, tid, id, &ep );
+			rc = bdb_cache_entry_db_lock( bdb->bi_dbenv, locker, *eip, 0, 0, lock );
+			if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
+				rc = DB_NOTFOUND;
+				bdb_cache_entry_db_unlock( bdb->bi_dbenv, 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) {
+						rc = bdb_id2entry( op->o_bd, tid, locker, id, &ep );
 					}
 					if ( rc == 0 ) {
-						bdb_cache_entry_db_relock( bdb->bi_dbenv, locker,
-							*eip, 1, 0, lock );
-						/* Make sure no other modifier beat us to it */
-						if ( (*eip)->bei_e ) {
-							bdb_entry_return( ep );
-							ep = NULL;
+						ep->e_private = *eip;
 #ifdef BDB_HIER
-							/* Check for subtree renames */
-							rc = bdb_fix_dn( (*eip)->bei_e, 1 );
-							if ( rc ) rc = bdb_fix_dn( (*eip)->bei_e, 2 );
+						bdb_fix_dn( ep, 0 );
 #endif
-						} else {
-							ep->e_private = *eip;
-#ifdef BDB_HIER
-							bdb_fix_dn( ep, 0 );
+						(*eip)->bei_e = ep;
+#ifdef SLAP_ZONE_ALLOC
+						(*eip)->bei_zseq = *((ber_len_t *)ep - 2);
 #endif
-							(*eip)->bei_e = ep;
-						}
-						bdb_cache_entry_db_relock( bdb->bi_dbenv, locker,
+						ep = NULL;
+					}
+					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,
 							*eip, 0, 0, lock );
+					} else {
+						/* Otherwise, release the lock. */
+						bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
 					}
+				} else if ( !(*eip)->bei_e ) {
+					/* Some other thread is trying to load the entry,
+					 * give it a chance to finish.
+					 */
+					bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
+					ldap_pvt_thread_yield();
+					bdb_cache_entryinfo_lock( *eip );
+					islocked = 1;
+					goto load1;
 #ifdef BDB_HIER
 				} else {
 					/* Check for subtree renames
@@ -707,23 +820,51 @@ again:	ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
 			}
 		}
 	}
+	if ( islocked ) {
+		bdb_cache_entryinfo_unlock( *eip );
+	}
+	if ( ep ) {
+		ep->e_private = NULL;
+#ifdef SLAP_ZONE_ALLOC
+		bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
+#else
+		bdb_entry_return( ep );
+#endif
+	}
 	if ( rc == 0 ) {
-		/* set lru mutex */
-		ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_mutex );
-		/* if entry is on LRU list, remove from old spot */
-		if ( (*eip)->bei_lrunext || (*eip)->bei_lruprev ) {
-			LRU_DELETE( &bdb->bi_cache, *eip );
-		} else {
-		/* if entry is new, bump cache size */
+
+		if ( load ) {
+			ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
 			bdb->bi_cache.c_cursize++;
+			ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
+		}
+
+		ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_head_mutex );
+
+		/* If the LRU list has only one entry and this is it, it
+		 * doesn't need to be added again.
+		 */
+		if ( bdb->bi_cache.c_lruhead == bdb->bi_cache.c_lrutail &&
+			bdb->bi_cache.c_lruhead == *eip ) {
+			ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_head_mutex );
+		} else {
+			/* if entry is on LRU list, remove from old spot */
+			if ( (*eip)->bei_lrunext || (*eip)->bei_lruprev ) {
+				ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_tail_mutex );
+				LRU_DELETE( &bdb->bi_cache, *eip );
+				ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_tail_mutex );
+			}
+			/* lru_head_mutex is unlocked for us */
+			bdb_cache_lru_add( bdb, *eip );
 		}
-		/* lru_mutex is unlocked for us */
-		bdb_cache_lru_add( bdb, locker, *eip );
 	}
 
-	if ( islocked ) {
-		bdb_cache_entryinfo_unlock( *eip );
+#ifdef SLAP_ZONE_ALLOC
+	if (rc == 0 && (*eip)->bei_e) {
+		slap_zn_rlock(bdb->bi_cache.c_zctx, (*eip)->bei_e);
 	}
+	slap_zh_runlock(bdb->bi_cache.c_zctx);
+#endif
 	return rc;
 }
 
@@ -758,9 +899,11 @@ bdb_cache_add(
 	u_int32_t locker )
 {
 	EntryInfo *new, ei;
-	struct berval rdn = e->e_name;
 	DB_LOCK lock;
 	int rc;
+#ifdef BDB_HIER
+	struct berval rdn = e->e_name;
+#endif
 
 	ei.bei_id = e->e_id;
 	ei.bei_parent = eip;
@@ -778,7 +921,8 @@ bdb_cache_add(
 
 #ifdef BDB_HIER
 	if ( nrdn->bv_len != e->e_nname.bv_len ) {
-		char *ptr = strchr( rdn.bv_val, ',' );
+		char *ptr = ber_bvchr( &rdn, ',' );
+		assert( ptr != NULL );
 		rdn.bv_len = ptr - rdn.bv_val;
 	}
 	ber_dupbv( &ei.bei_rdn, &rdn );
@@ -789,7 +933,11 @@ bdb_cache_add(
 	/* bdb_csn_commit can cause this when adding the database root entry */
 	if ( new->bei_e ) {
 		new->bei_e->e_private = NULL;
+#ifdef SLAP_ZONE_ALLOC
+		bdb_entry_return( bdb, new->bei_e, new->bei_zseq );
+#else
 		bdb_entry_return( new->bei_e );
+#endif
 	}
 	new->bei_e = e;
 	e->e_private = new;
@@ -798,15 +946,17 @@ bdb_cache_add(
 	if (eip->bei_parent) {
 		eip->bei_parent->bei_state &= ~CACHE_ENTRY_NO_GRANDKIDS;
 	}
+	bdb_cache_entryinfo_unlock( eip );
 
-	/* set lru mutex */
-	ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_mutex );
 	++bdb->bi_cache.c_cursize;
-	/* lru_mutex is unlocked for us */
-	bdb_cache_lru_add( bdb, locker, new );
-
-	bdb_cache_entryinfo_unlock( eip );
 	ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
+
+	/* set lru mutex */
+	ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_head_mutex );
+
+	/* lru_head_mutex is unlocked for us */
+	bdb_cache_lru_add( bdb, new );
+
 	return rc;
 }
 
@@ -840,20 +990,22 @@ bdb_cache_modify(
  */
 int
 bdb_cache_modrdn(
+	struct bdb_info *bdb,
 	Entry *e,
 	struct berval *nrdn,
 	Entry *new,
 	EntryInfo *ein,
-	DB_ENV *env,
 	u_int32_t locker,
 	DB_LOCK *lock )
 {
 	EntryInfo *ei = BEI(e), *pei;
-	struct berval rdn;
 	int rc;
+#ifdef BDB_HIER
+	struct berval rdn;
+#endif
 
 	/* Get write lock on data */
-	rc =  bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock );
+	rc =  bdb_cache_entry_db_relock( bdb->bi_dbenv, locker, ei, 1, 0, lock );
 	if ( rc ) return rc;
 
 	/* If we've done repeated mods on a cached entry, then e_attrs
@@ -878,15 +1030,22 @@ bdb_cache_modrdn(
 	avl_delete( &pei->bei_kids, (caddr_t) ei, bdb_rdn_cmp );
 	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 );
 
 	rdn = e->e_name;
 	if ( nrdn->bv_len != e->e_nname.bv_len ) {
-		char *ptr = strchr(rdn.bv_val, ',');
+		char *ptr = ber_bvchr(&rdn, ',');
+		assert( ptr != NULL );
 		rdn.bv_len = ptr - rdn.bv_val;
 	}
 	ber_dupbv( &ei->bei_rdn, &rdn );
+	pei->bei_ckids--;
+	if ( pei->bei_dkids ) pei->bei_dkids--;
 #endif
 
 	if (!ein) {
@@ -896,15 +1055,24 @@ bdb_cache_modrdn(
 		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;
 #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;
+
 	{
-		int max = ei->bei_modrdns;
 		/* Record the generation number of this change */
-		for ( pei = ein; pei->bei_parent; pei = pei->bei_parent ) {
-			if ( pei->bei_modrdns > max ) max = pei->bei_modrdns;
-		}
-		ei->bei_modrdns = max + 1;
+		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
 	avl_insert( &ein->bei_kids, ei, bdb_rdn_cmp, avl_dup_error );
 	bdb_cache_entryinfo_unlock( ein );
@@ -928,7 +1096,7 @@ bdb_cache_delete(
 	EntryInfo *ei = BEI(e);
 	int	rc;
 
-	assert( e->e_private );
+	assert( e->e_private != NULL );
 
 	/* Set this early, warn off any queriers */
 	ei->bei_state |= CACHE_ENTRY_DELETED;
@@ -945,29 +1113,22 @@ bdb_cache_delete(
 		return rc;
 	}
 
-	/* set cache write lock */
-	ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
-
-	/* Lock the parent's kids tree */
-	bdb_cache_entryinfo_lock( ei->bei_parent );
-
-#ifdef NEW_LOGGING
-	LDAP_LOG( CACHE, ENTRY, 
-		"bdb_cache_delete: delete %ld.\n", e->e_id, 0, 0 );
-#else
 	Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_delete( %ld )\n",
 		e->e_id, 0, 0 );
-#endif
 
 	/* set lru mutex */
-	ldap_pvt_thread_mutex_lock( &cache->lru_mutex );
-	rc = bdb_cache_delete_internal( cache, e->e_private );
-	/* free lru mutex */
-	ldap_pvt_thread_mutex_unlock( &cache->lru_mutex );
+	ldap_pvt_thread_mutex_lock( &cache->lru_tail_mutex );
+
+	/* set cache write lock */
+	ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
+
+	rc = bdb_cache_delete_internal( cache, e->e_private, 1 );
 
 	/* free cache write lock */
 	ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
-	bdb_cache_entryinfo_unlock( ei->bei_parent );
+
+	/* free lru mutex */
+	ldap_pvt_thread_mutex_unlock( &cache->lru_tail_mutex );
 
 	/* Leave entry info locked */
 
@@ -977,13 +1138,17 @@ bdb_cache_delete(
 void
 bdb_cache_delete_cleanup(
 	Cache *cache,
-	Entry *e )
+	EntryInfo *ei )
 {
-	EntryInfo *ei = BEI(e);
-
-	ei->bei_e = NULL;
-	e->e_private = NULL;
-	bdb_entry_return( e );
+	if ( ei->bei_e ) {
+		ei->bei_e->e_private = NULL;
+#ifdef SLAP_ZONE_ALLOC
+		bdb_entry_return( ei->bei_bdb, ei->bei_e, ei->bei_zseq );
+#else
+		bdb_entry_return( ei->bei_e );
+#endif
+		ei->bei_e = NULL;
+	}
 
 	free( ei->bei_nrdn.bv_val );
 	ei->bei_nrdn.bv_val = NULL;
@@ -1008,13 +1173,17 @@ bdb_cache_delete_cleanup(
 static int
 bdb_cache_delete_internal(
     Cache	*cache,
-    EntryInfo		*e )
+    EntryInfo		*e,
+    int		decr )
 {
 	int rc = 0;	/* return code */
 
+	/* Lock the parent's kids tree */
+	bdb_cache_entryinfo_lock( e->bei_parent );
+
 #ifdef BDB_HIER
 	e->bei_parent->bei_ckids--;
-	if ( e->bei_parent->bei_dkids ) e->bei_parent->bei_dkids--;
+	if ( decr && e->bei_parent->bei_dkids ) e->bei_parent->bei_dkids--;
 #endif
 	/* dn tree */
 	if ( avl_delete( &e->bei_parent->bei_kids, (caddr_t) e, bdb_rdn_cmp )
@@ -1022,26 +1191,25 @@ bdb_cache_delete_internal(
 	{
 		rc = -1;
 	}
+	if ( e->bei_parent->bei_kids )
+		cache->c_leaves--;
 
 	/* id tree */
 	if ( avl_delete( &cache->c_idtree, (caddr_t) e, bdb_id_cmp ) == NULL ) {
 		rc = -1;
 	}
 
-	if (rc != 0) {
-		return rc;
-	}
+	if ( rc == 0 ){
+		cache->c_eiused--;
 
-	/* lru */
-	LRU_DELETE( cache, e );
-	cache->c_cursize--;
+		/* lru */
+		LRU_DELETE( cache, e );
+		if ( e->bei_e ) cache->c_cursize--;
+	}
 
-	/*
-	 * flag entry to be freed later by a call to cache_return_entry()
-	 */
-	e->bei_state |= CACHE_ENTRY_DELETED;
+	bdb_cache_entryinfo_unlock( e->bei_parent );
 
-	return( 0 );
+	return( rc );
 }
 
 static void
@@ -1053,7 +1221,11 @@ bdb_entryinfo_release( void *data )
 	}
 	if ( ei->bei_e ) {
 		ei->bei_e->e_private = NULL;
+#ifdef SLAP_ZONE_ALLOC
+		bdb_entry_return( ei->bei_bdb, ei->bei_e, ei->bei_zseq );
+#else
 		bdb_entry_return( ei->bei_e );
+#endif
 	}
 	bdb_cache_entryinfo_destroy( ei );
 }
@@ -1064,26 +1236,32 @@ 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_mutex );
+	ldap_pvt_thread_mutex_lock( &cache->lru_tail_mutex );
 
-#ifdef NEW_LOGGING
-	LDAP_LOG( CACHE, ENTRY, "bdb_cache_release_all: enter\n", 0, 0, 0 );
-#else
 	Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_release_all\n", 0, 0, 0 );
-#endif
 
 	avl_free( cache->c_dntree.bei_kids, NULL );
 	avl_free( cache->c_idtree, bdb_entryinfo_release );
+	for (;cache->c_eifree;cache->c_eifree = cache->c_lruhead) {
+		cache->c_lruhead = cache->c_eifree->bei_lrunext;
+		bdb_cache_entryinfo_destroy(cache->c_eifree);
+	}
+	cache->c_cursize = 0;
+	cache->c_eiused = 0;
+	cache->c_leaves = 0;
+	cache->c_idtree = NULL;
 	cache->c_lruhead = NULL;
 	cache->c_lrutail = NULL;
+	cache->c_dntree.bei_kids = NULL;
 
 	/* free lru mutex */
-	ldap_pvt_thread_mutex_unlock( &cache->lru_mutex );
+	ldap_pvt_thread_mutex_unlock( &cache->lru_tail_mutex );
 	/* free cache write lock */
 	ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
 }
 
 #ifdef LDAP_DEBUG
+#ifdef SLAPD_UNUSED
 static void
 bdb_lru_print( Cache *cache )
 {
@@ -1101,40 +1279,48 @@ bdb_lru_print( Cache *cache )
 	}
 }
 #endif
+#endif
 
 #ifdef BDB_REUSE_LOCKERS
 static void
 bdb_locker_id_free( void *key, void *data )
 {
 	DB_ENV *env = key;
-	int lockid = (int) data;
+	u_int32_t lockid = (long)data;
 	int rc;
 
 	rc = XLOCK_ID_FREE( env, lockid );
 	if ( rc == EINVAL ) {
 		DB_LOCKREQ lr;
-#ifdef NEW_LOGGING
-		LDAP_LOG( BACK_BDB, ERR,
-			"bdb_locker_id_free: %d err %s(%d)\n",
-			lockid, db_strerror(rc), rc );
-#else
 		Debug( LDAP_DEBUG_ANY,
-			"bdb_locker_id_free: %d err %s(%d)\n",
-			lockid, db_strerror(rc), rc );
-#endif
-		memset( &lr, 0, sizeof(lr) );
-
+			"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 );
 	}
 }
 
+/* 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, int *locker )
+bdb_locker_id( Operation *op, DB_ENV *env, u_int32_t *locker )
 {
-	int i, rc, lockid;
+	int i, rc;
+	u_int32_t lockid;
 	void *data;
 	void *ctx;
 
@@ -1161,27 +1347,22 @@ bdb_locker_id( Operation *op, DB_ENV *env, int *locker )
 		if ( rc != 0) {
 			return rc;
 		}
-		data = (void *)lockid;
+		data = (void *)((long)lockid);
 		if ( ( rc = ldap_pvt_thread_pool_setkey( ctx, env,
 			data, bdb_locker_id_free ) ) ) {
 			XLOCK_ID_FREE( env, lockid );
-#ifdef NEW_LOGGING
-			LDAP_LOG( BACK_BDB, ERR, "bdb_locker_id: err %s(%d)\n",
-				db_strerror(rc), rc, 0 );
-#else
 			Debug( LDAP_DEBUG_ANY, "bdb_locker_id: err %s(%d)\n",
 				db_strerror(rc), rc, 0 );
-#endif
 
 			return rc;
 		}
 	} else {
-		lockid = (int)data;
+		lockid = (long)data;
 	}
 	*locker = lockid;
 	return 0;
 }
-#endif
+#endif /* BDB_REUSE_LOCKERS */
 
 void
 bdb_cache_delete_entry(
@@ -1196,7 +1377,11 @@ bdb_cache_delete_entry(
 		if ( ei->bei_e && !(ei->bei_state & CACHE_ENTRY_NOT_LINKED )) {
 			LRU_DELETE( &bdb->bi_cache, ei );
 			ei->bei_e->e_private = NULL;
+#ifdef SLAP_ZONE_ALLOC
+			bdb_entry_return( bdb, ei->bei_e, ei->bei_zseq );
+#else
 			bdb_entry_return( ei->bei_e );
+#endif
 			ei->bei_e = NULL;
 			--bdb->bi_cache.c_cursize;
 		}