]> git.sur5r.net Git - openldap/blobdiff - servers/slapd/back-bdb/dn2id.c
ITS#5262 fixes from HEAD
[openldap] / servers / slapd / back-bdb / dn2id.c
index c76c8ea613288a9eac2353b30ca83e39d3d48e01..eb7fbdb3fa52b8843392884c0f944102c3739b39 100644 (file)
@@ -2,7 +2,7 @@
 /* $OpenLDAP$ */
 /* This work is part of OpenLDAP Software <http://www.openldap.org/>.
  *
- * Copyright 2000-2005 The OpenLDAP Foundation.
+ * Copyright 2000-2007 The OpenLDAP Foundation.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
 #include "idl.h"
 #include "lutil.h"
 
+#define bdb_dn2id_lock                                 BDB_SYMBOL(dn2id_lock)
+
+static int
+bdb_dn2id_lock( struct bdb_info *bdb, struct berval *dn,
+       int rw, u_int32_t locker, DB_LOCK *lock )
+{
+       int       rc;
+       DBT       lockobj;
+       int       db_rw;
+
+       if (rw)
+               db_rw = DB_LOCK_WRITE;
+       else
+               db_rw = DB_LOCK_READ;
+
+       lockobj.data = dn->bv_val;
+       lockobj.size = dn->bv_len;
+
+       rc = LOCK_GET(bdb->bi_dbenv, locker, DB_LOCK_NOWAIT,
+                                       &lockobj, db_rw, lock);
+       return rc;
+}
+
 #ifndef BDB_HIER
 int
 bdb_dn2id_add(
@@ -39,8 +62,8 @@ bdb_dn2id_add(
        char            *buf;
        struct berval   ptr, pdn;
 
-       Debug( LDAP_DEBUG_TRACE, "=> bdb_dn2id_add( \"%s\", 0x%08lx )\n",
-               e->e_ndn, (long) e->e_id, 0 );
+       Debug( LDAP_DEBUG_TRACE, "=> bdb_dn2id_add 0x%lx: \"%s\"\n",
+               e->e_id, e->e_ndn, 0 );
        assert( e->e_id != NOID );
 
        DBTzero( &key );
@@ -63,8 +86,8 @@ bdb_dn2id_add(
        /* store it -- don't override */
        rc = db->put( db, txn, &key, &data, DB_NOOVERWRITE );
        if( rc != 0 ) {
-               Debug( LDAP_DEBUG_ANY, "=> bdb_dn2id_add: put failed: %s %d\n",
-                       db_strerror(rc), rc, 0 );
+               Debug( LDAP_DEBUG_ANY, "=> bdb_dn2id_add 0x%lx: put failed: %s %d\n",
+                       e->e_id, db_strerror(rc), rc );
                goto done;
        }
 
@@ -76,8 +99,8 @@ bdb_dn2id_add(
                rc = db->put( db, txn, &key, &data, DB_NOOVERWRITE );
                if( rc != 0 ) {
                        Debug( LDAP_DEBUG_ANY,
-                       "=> bdb_dn2id_add: subtree (%s) put failed: %d\n",
-                       ptr.bv_val, rc, 0 );
+                       "=> bdb_dn2id_add 0x%lx: subtree (%s) put failed: %d\n",
+                       e->e_id, ptr.bv_val, rc );
                        goto done;
                }
                
@@ -97,8 +120,8 @@ bdb_dn2id_add(
 
                if( rc != 0 ) {
                        Debug( LDAP_DEBUG_ANY,
-                               "=> bdb_dn2id_add: parent (%s) insert failed: %d\n",
-                                       ptr.bv_val, rc, 0 );
+                               "=> bdb_dn2id_add 0x%lx: parent (%s) insert failed: %d\n",
+                                       e->e_id, ptr.bv_val, rc );
                        goto done;
                }
        }
@@ -115,8 +138,8 @@ bdb_dn2id_add(
 
                if( rc != 0 ) {
                        Debug( LDAP_DEBUG_ANY,
-                               "=> bdb_dn2id_add: subtree (%s) insert failed: %d\n",
-                                       ptr.bv_val, rc, 0 );
+                               "=> bdb_dn2id_add 0x%lx: subtree (%s) insert failed: %d\n",
+                                       e->e_id, ptr.bv_val, rc );
                        break;
                }
 #ifdef BDB_MULTIPLE_SUFFIXES
@@ -133,7 +156,7 @@ bdb_dn2id_add(
 
 done:
        op->o_tmpfree( buf, op->o_tmpmemctx );
-       Debug( LDAP_DEBUG_TRACE, "<= bdb_dn2id_add: %d\n", rc, 0, 0 );
+       Debug( LDAP_DEBUG_TRACE, "<= bdb_dn2id_add 0x%lx: %d\n", e->e_id, rc, 0 );
        return rc;
 }
 
@@ -146,13 +169,14 @@ bdb_dn2id_delete(
 {
        struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
        DB *db = bdb->bi_dn2id->bdi_db;
-       int             rc;
-       DBT             key;
        char            *buf;
+       DBT             key;
+       DB_LOCK lock;
        struct berval   pdn, ptr;
+       int             rc;
 
-       Debug( LDAP_DEBUG_TRACE, "=> bdb_dn2id_delete( \"%s\", 0x%08lx )\n",
-               e->e_ndn, e->e_id, 0 );
+       Debug( LDAP_DEBUG_TRACE, "=> bdb_dn2id_delete 0x%lx: \"%s\"\n",
+               e->e_id, e->e_ndn, 0 );
 
        DBTzero( &key );
        key.size = e->e_nname.bv_len + 2;
@@ -165,11 +189,15 @@ bdb_dn2id_delete(
        AC_MEMCPY( ptr.bv_val, e->e_nname.bv_val, e->e_nname.bv_len );
        ptr.bv_val[ptr.bv_len] = '\0';
 
+       /* We hold this lock until the TXN completes */
+       rc = bdb_dn2id_lock( bdb, &e->e_nname, 1, TXN_ID( txn ), &lock );
+       if ( rc ) goto done;
+
        /* delete it */
        rc = db->del( db, txn, &key, 0 );
        if( rc != 0 ) {
-               Debug( LDAP_DEBUG_ANY, "=> bdb_dn2id_delete: delete failed: %s %d\n",
-                       db_strerror(rc), rc, 0 );
+               Debug( LDAP_DEBUG_ANY, "=> bdb_dn2id_delete 0x%lx: delete failed: %s %d\n",
+                       e->e_id, db_strerror(rc), rc );
                goto done;
        }
 
@@ -178,11 +206,11 @@ bdb_dn2id_delete(
 #endif
        {
                buf[0] = DN_SUBTREE_PREFIX;
-               rc = db->del( db, txn, &key, 0 );
+               rc = bdb_idl_delete_key( op->o_bd, db, txn, &key, e->e_id );
                if( rc != 0 ) {
                        Debug( LDAP_DEBUG_ANY,
-                       "=> bdb_dn2id_delete: subtree (%s) delete failed: %d\n",
-                       ptr.bv_val, rc, 0 );
+                       "=> bdb_dn2id_delete 0x%lx: subtree (%s) delete failed: %d\n",
+                       e->e_id, ptr.bv_val, rc );
                        goto done;
                }
 
@@ -202,8 +230,8 @@ bdb_dn2id_delete(
 
                if( rc != 0 ) {
                        Debug( LDAP_DEBUG_ANY,
-                               "=> bdb_dn2id_delete: parent (%s) delete failed: %d\n",
-                               ptr.bv_val, rc, 0 );
+                               "=> bdb_dn2id_delete 0x%lx: parent (%s) delete failed: %d\n",
+                               e->e_id, ptr.bv_val, rc );
                        goto done;
                }
        }
@@ -219,8 +247,8 @@ bdb_dn2id_delete(
                rc = bdb_idl_delete_key( op->o_bd, db, txn, &key, e->e_id );
                if( rc != 0 ) {
                        Debug( LDAP_DEBUG_ANY,
-                               "=> bdb_dn2id_delete: subtree (%s) delete failed: %d\n",
-                               ptr.bv_val, rc, 0 );
+                               "=> bdb_dn2id_delete 0x%lx: subtree (%s) delete failed: %d\n",
+                               e->e_id, ptr.bv_val, rc );
                        goto done;
                }
 #ifdef BDB_MULTIPLE_SUFFIXES
@@ -237,24 +265,27 @@ bdb_dn2id_delete(
 
 done:
        op->o_tmpfree( buf, op->o_tmpmemctx );
-       Debug( LDAP_DEBUG_TRACE, "<= bdb_dn2id_delete %d\n", rc, 0, 0 );
+       Debug( LDAP_DEBUG_TRACE, "<= bdb_dn2id_delete 0x%lx: %d\n", e->e_id, rc, 0 );
        return rc;
 }
 
 int
 bdb_dn2id(
        Operation *op,
-       DB_TXN *txn,
        struct berval   *dn,
-       EntryInfo *ei )
+       EntryInfo *ei,
+       u_int32_t locker,
+       DB_LOCK *lock )
 {
        struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
        DB *db = bdb->bi_dn2id->bdi_db;
+       DBC     *cursor;
        int             rc;
        DBT             key, data;
        ID              nid;
 
        Debug( LDAP_DEBUG_TRACE, "=> bdb_dn2id(\"%s\")\n", dn->bv_val, 0, 0 );
+
        DBTzero( &key );
        key.size = dn->bv_len + 2;
        key.data = op->o_tmpalloc( key.size, op->o_tmpmemctx );
@@ -267,18 +298,31 @@ bdb_dn2id(
        data.ulen = sizeof(ID);
        data.flags = DB_DBT_USERMEM;
 
+       rc = db->cursor( db, NULL, &cursor, bdb->bi_db_opflags );
+       if ( rc ) goto leave;
+
+       rc = bdb_dn2id_lock( bdb, dn, 0, locker, lock );
+       if ( rc ) goto nolock;
+
+       if ( locker ) {
+               cursor->locker = locker;
+       }
+
        /* fetch it */
-       rc = db->get( db, txn, &key, &data, bdb->bi_db_opflags );
+       rc = cursor->c_get( cursor, &key, &data, DB_SET );
+
+nolock:
+       cursor->c_close( cursor );
+leave:
 
        if( rc != 0 ) {
                Debug( LDAP_DEBUG_TRACE, "<= bdb_dn2id: get failed: %s (%d)\n",
                        db_strerror( rc ), rc, 0 );
        } else {
                BDB_DISK2ID( &nid, &ei->bei_id );
-               Debug( LDAP_DEBUG_TRACE, "<= bdb_dn2id: got id=0x%08lx\n",
+               Debug( LDAP_DEBUG_TRACE, "<= bdb_dn2id: got id=0x%lx\n",
                        ei->bei_id, 0, 0 );
        }
-
        op->o_tmpfree( key.data, op->o_tmpmemctx );
        return rc;
 }
@@ -361,6 +405,7 @@ bdb_dn2idl(
        ((char *)key.data)[0] = prefix;
        AC_MEMCPY( &((char *)key.data)[1], e->e_nname.bv_val, key.size - 1 );
 
+       BDB_IDL_ZERO( ids );
        rc = bdb_idl_fetch_key( op->o_bd, db, NULL, &key, ids, NULL, 0 );
 
        if( rc != 0 ) {
@@ -380,10 +425,7 @@ bdb_dn2idl(
 }
 
 #else  /* BDB_HIER */
-/* Experimental management routines for a hierarchically structured database.
- *
- * Unsupported! Use at your own risk!
- * -- Howard Chu, Symas Corp. 2003.
+/* Management routines for a hierarchically structured database.
  *
  * Instead of a ldbm-style dn2id database, we use a hierarchical one. Each
  * entry in this database is a struct diskNode, keyed by entryID and with
@@ -401,11 +443,36 @@ bdb_dn2idl(
  */
 typedef struct diskNode {
        unsigned char nrdnlen[2];
-       unsigned char nrdn[1];
-       unsigned char rdn[1];
-       unsigned char entryID[sizeof(ID)];
+       char nrdn[1];
+       char rdn[1];                        /* variable placement */
+       unsigned char entryID[sizeof(ID)];  /* variable placement */
 } diskNode;
 
+/* Sort function for the sorted duplicate data items of a dn2id key.
+ * Sorts based on normalized RDN, in length order.
+ */
+int
+hdb_dup_compare(
+       DB *db, 
+       const DBT *usrkey,
+       const DBT *curkey
+)
+{
+       diskNode *un, *cn;
+       int rc, ul, cl;
+
+       un = (diskNode *)usrkey->data;
+       cn = (diskNode *)curkey->data;
+
+       /* data is not aligned, cannot compare directly */
+       rc = un->nrdnlen[0] - cn->nrdnlen[0];
+       if ( rc ) return rc;
+       rc = un->nrdnlen[1] - cn->nrdnlen[1];
+       if ( rc ) return rc;
+
+       return strcmp( un->nrdn, cn->nrdn );
+}
+
 /* This function constructs a full DN for a given entry.
  */
 int hdb_fix_dn(
@@ -417,6 +484,9 @@ int hdb_fix_dn(
        char *ptr, *nptr;
        int max = 0;
 
+       if ( !e->e_id )
+               return 0;
+
        /* count length of all DN components */
        for ( ei = BEI(e); ei && ei->bei_id; ei=ei->bei_parent ) {
                rlen += ei->bei_rdn.bv_len + 1;
@@ -478,6 +548,9 @@ hdb_dn2id_add(
        diskNode *d;
        char *ptr;
 
+       Debug( LDAP_DEBUG_TRACE, "=> hdb_dn2id_add 0x%lx: \"%s\"\n",
+               e->e_id, e->e_ndn, 0 );
+
        nrlen = dn_rdnlen( op->o_bd, &e->e_nname );
        if (nrlen) {
                rlen = dn_rdnlen( op->o_bd, &e->e_name );
@@ -497,16 +570,17 @@ hdb_dn2id_add(
 
        DBTzero(&key);
        DBTzero(&data);
-       key.data = &nid;
        key.size = sizeof(ID);
        key.flags = DB_DBT_USERMEM;
        BDB_ID2DISK( eip->bei_id, &nid );
 
+       key.data = &nid;
+
        /* Need to make dummy root node once. Subsequent attempts
         * will fail harmlessly.
         */
        if ( eip->bei_id == 0 ) {
-               diskNode dummy = {0};
+               diskNode dummy = {{0, 0}, "", "", ""};
                data.data = &dummy;
                data.size = sizeof(diskNode);
                data.flags = DB_DBT_USERMEM;
@@ -514,9 +588,6 @@ hdb_dn2id_add(
                db->put( db, txn, &key, &data, DB_NODUPDATA );
        }
 
-       if ( bdb->bi_idl_cache_size ) {
-               bdb_idl_cache_del( bdb, db, &key );
-       }
        data.data = d;
        data.size = sizeof(diskNode) + rlen + nrlen;
        data.flags = DB_DBT_USERMEM;
@@ -531,7 +602,25 @@ hdb_dn2id_add(
                rc = db->put( db, txn, &key, &data, DB_NODUPDATA );
        }
 
+       /* Update all parents' IDL cache entries */
+       if ( rc == 0 && bdb->bi_idl_cache_size ) {
+               ID tmp[2];
+               char *ptr = ((char *)&tmp[1])-1;
+               key.data = ptr;
+               key.size = sizeof(ID)+1;
+               tmp[1] = eip->bei_id;
+               *ptr = DN_ONE_PREFIX;
+               bdb_idl_cache_add_id( bdb, db, &key, e->e_id );
+               *ptr = DN_SUBTREE_PREFIX;
+               for (; eip && eip->bei_parent->bei_id; eip = eip->bei_parent) {
+                       tmp[1] = eip->bei_id;
+                       bdb_idl_cache_add_id( bdb, db, &key, e->e_id );
+               }
+       }
+
+leave:
        op->o_tmpfree( d, op->o_tmpmemctx );
+       Debug( LDAP_DEBUG_TRACE, "<= hdb_dn2id_add 0x%lx: %d\n", e->e_id, rc, 0 );
 
        return rc;
 }
@@ -548,13 +637,17 @@ hdb_dn2id_delete(
        DBT             key, data;
        DBC     *cursor;
        diskNode *d;
-       int rc, nrlen;
+       int rc;
        ID      nid;
+       unsigned char dlen[2];
+       DB_LOCK lock;
+
+       Debug( LDAP_DEBUG_TRACE, "=> hdb_dn2id_delete 0x%lx: \"%s\"\n",
+               e->e_id, e->e_ndn, 0 );
 
        DBTzero(&key);
        key.size = sizeof(ID);
        key.ulen = key.size;
-       key.data = &nid;
        key.flags = DB_DBT_USERMEM;
        BDB_ID2DISK( eip->bei_id, &nid );
 
@@ -564,22 +657,28 @@ hdb_dn2id_delete(
        data.dlen = data.size;
        data.flags = DB_DBT_USERMEM | DB_DBT_PARTIAL;
 
-       if ( bdb->bi_idl_cache_size ) {
-               bdb_idl_cache_del( bdb, db, &key );
-       }
-       rc = db->cursor( db, txn, &cursor, bdb->bi_db_opflags );
-       if ( rc ) return rc;
+       key.data = &nid;
 
        d = op->o_tmpalloc( data.size, op->o_tmpmemctx );
        d->nrdnlen[1] = BEI(e)->bei_nrdn.bv_len & 0xff;
        d->nrdnlen[0] = (BEI(e)->bei_nrdn.bv_len >> 8) | 0x80;
+       dlen[0] = d->nrdnlen[0];
+       dlen[1] = d->nrdnlen[1];
        strcpy( d->nrdn, BEI(e)->bei_nrdn.bv_val );
        data.data = d;
 
+       rc = db->cursor( db, txn, &cursor, bdb->bi_db_opflags );
+       if ( rc ) goto leave;
+
+       /* We hold this lock until the TXN completes */
+       rc = bdb_dn2id_lock( bdb, &e->e_nname, 1, TXN_ID( txn ), &lock );
+       if ( rc ) goto nolock;
+
        /* Delete our ID from the parent's list */
        rc = cursor->c_get( cursor, &key, &data, DB_GET_BOTH_RANGE );
        if ( rc == 0 ) {
-               if ( !strcmp( d->nrdn, BEI(e)->bei_nrdn.bv_val ))
+               if ( dlen[1] == d->nrdnlen[1] && dlen[0] == d->nrdnlen[0] &&
+                       !strcmp( d->nrdn, BEI(e)->bei_nrdn.bv_val ))
                        rc = cursor->c_del( cursor, 0 );
                else
                        rc = DB_NOTFOUND;
@@ -595,9 +694,28 @@ hdb_dn2id_delete(
                if ( rc == 0 )
                        rc = cursor->c_del( cursor, 0 );
        }
+
+nolock:
        cursor->c_close( cursor );
+leave:
        op->o_tmpfree( d, op->o_tmpmemctx );
 
+       /* Delete IDL cache entries */
+       if ( rc == 0 && bdb->bi_idl_cache_size ) {
+               ID tmp[2];
+               char *ptr = ((char *)&tmp[1])-1;
+               key.data = ptr;
+               key.size = sizeof(ID)+1;
+               tmp[1] = eip->bei_id;
+               *ptr = DN_ONE_PREFIX;
+               bdb_idl_cache_del_id( bdb, db, &key, e->e_id );
+               *ptr = DN_SUBTREE_PREFIX;
+               for (; eip && eip->bei_parent->bei_id; eip = eip->bei_parent) {
+                       tmp[1] = eip->bei_id;
+                       bdb_idl_cache_del_id( bdb, db, &key, e->e_id );
+               }
+       }
+       Debug( LDAP_DEBUG_TRACE, "<= hdb_dn2id_delete 0x%lx: %d\n", e->e_id, rc, 0 );
        return rc;
 }
 
@@ -605,9 +723,10 @@ hdb_dn2id_delete(
 int
 hdb_dn2id(
        Operation       *op,
-       DB_TXN *txn,
        struct berval   *in,
-       EntryInfo       *ei )
+       EntryInfo       *ei,
+       u_int32_t locker,
+       DB_LOCK *lock )
 {
        struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
        DB *db = bdb->bi_dn2id->bdi_db;
@@ -616,7 +735,10 @@ hdb_dn2id(
        int             rc = 0, nrlen;
        diskNode *d;
        char    *ptr;
-       ID idp;
+       unsigned char dlen[2];
+       ID idp, parentID;
+
+       Debug( LDAP_DEBUG_TRACE, "=> hdb_dn2id(\"%s\")\n", in->bv_val, 0, 0 );
 
        nrlen = dn_rdnlen( op->o_bd, in );
        if (!nrlen) nrlen = in->bv_len;
@@ -626,7 +748,8 @@ hdb_dn2id(
        key.data = &idp;
        key.ulen = sizeof(ID);
        key.flags = DB_DBT_USERMEM;
-       BDB_ID2DISK( ei->bei_parent->bei_id, &idp );
+       parentID = ( ei->bei_parent != NULL ) ? ei->bei_parent->bei_id : 0;
+       BDB_ID2DISK( parentID, &idp );
 
        DBTzero(&data);
        data.size = sizeof(diskNode) + nrlen - sizeof(ID) - 1;
@@ -634,27 +757,36 @@ hdb_dn2id(
        data.dlen = data.ulen;
        data.flags = DB_DBT_USERMEM | DB_DBT_PARTIAL;
 
-       rc = db->cursor( db, txn, &cursor, bdb->bi_db_opflags );
+       rc = db->cursor( db, NULL, &cursor, bdb->bi_db_opflags );
        if ( rc ) return rc;
+       if ( locker ) {
+               cursor->locker = locker;
+       }
 
        d = op->o_tmpalloc( data.size * 3, op->o_tmpmemctx );
        d->nrdnlen[1] = nrlen & 0xff;
        d->nrdnlen[0] = (nrlen >> 8) | 0x80;
+       dlen[0] = d->nrdnlen[0];
+       dlen[1] = d->nrdnlen[1];
        ptr = lutil_strncopy( d->nrdn, in->bv_val, nrlen );
        *ptr = '\0';
        data.data = d;
 
+       rc = bdb_dn2id_lock( bdb, in, 0, locker, lock );
+       if ( rc ) goto leave;
+
        rc = cursor->c_get( cursor, &key, &data, DB_GET_BOTH_RANGE );
-       if ( rc == 0 && strncmp( d->nrdn, in->bv_val, nrlen )) {
+       if ( rc == 0 && (dlen[1] != d->nrdnlen[1] || dlen[0] != d->nrdnlen[0] ||
+               strncmp( d->nrdn, in->bv_val, nrlen ))) {
                rc = DB_NOTFOUND;
        }
        if ( rc == 0 ) {
-               ptr = data.data + data.size - sizeof(ID);
+               ptr = (char *) data.data + data.size - sizeof(ID);
                BDB_DISK2ID( ptr, &ei->bei_id );
                ei->bei_rdn.bv_len = data.size - sizeof(diskNode) - nrlen;
                ptr = d->nrdn + nrlen + 1;
                ber_str2bv( ptr, ei->bei_rdn.bv_len, 1, &ei->bei_rdn );
-               if ( !ei->bei_parent->bei_dkids ) {
+               if ( ei->bei_parent != NULL && !ei->bei_parent->bei_dkids ) {
                        db_recno_t dkids;
                        /* How many children does the parent have? */
                        /* FIXME: do we need to lock the parent
@@ -664,8 +796,17 @@ hdb_dn2id(
                        ei->bei_parent->bei_dkids = dkids;
                }
        }
+
+leave:
        cursor->c_close( cursor );
        op->o_tmpfree( d, op->o_tmpmemctx );
+       if( rc != 0 ) {
+               Debug( LDAP_DEBUG_TRACE, "<= hdb_dn2id: get failed: %s (%d)\n",
+                       db_strerror( rc ), rc, 0 );
+       } else {
+               Debug( LDAP_DEBUG_TRACE, "<= hdb_dn2id: got id=0x%lx\n",
+                       ei->bei_id, 0, 0 );
+       }
 
        return rc;
 }
@@ -673,7 +814,7 @@ hdb_dn2id(
 int
 hdb_dn2id_parent(
        Operation *op,
-       DB_TXN *txn,
+       u_int32_t       locker,
        EntryInfo *ei,
        ID *idp )
 {
@@ -684,7 +825,6 @@ hdb_dn2id_parent(
        int             rc = 0;
        diskNode *d;
        char    *ptr;
-       unsigned char *pt2;
        ID      nid;
 
        DBTzero(&key);
@@ -697,8 +837,11 @@ hdb_dn2id_parent(
        DBTzero(&data);
        data.flags = DB_DBT_USERMEM;
 
-       rc = db->cursor( db, txn, &cursor, bdb->bi_db_opflags );
+       rc = db->cursor( db, NULL, &cursor, bdb->bi_db_opflags );
        if ( rc ) return rc;
+       if ( locker ) {
+               cursor->locker = locker;
+       }
 
        data.ulen = sizeof(diskNode) + (SLAP_LDAPDN_MAXLEN * 2);
        d = op->o_tmpalloc( data.ulen, op->o_tmpmemctx );
@@ -710,7 +853,7 @@ hdb_dn2id_parent(
                        rc = LDAP_OTHER;
                } else {
                        db_recno_t dkids;
-                       ptr = data.data + data.size - sizeof(ID);
+                       ptr = (char *) data.data + data.size - sizeof(ID);
                        BDB_DISK2ID( ptr, idp );
                        ei->bei_nrdn.bv_len = (d->nrdnlen[0] << 8) | d->nrdnlen[1];
                        ber_str2bv( d->nrdn, ei->bei_nrdn.bv_len, 1, &ei->bei_nrdn );
@@ -791,21 +934,22 @@ hdb_dn2id_children(
 
 struct dn2id_cookie {
        struct bdb_info *bdb;
-       DB *db;
-       int prefix;
-       int rc;
+       Operation *op;
        EntryInfo *ei;
-       ID id;
-       ID nid;
-       ID dbuf;
        ID *ids;
-       void *ptr;
-       ID tmp[BDB_IDL_DB_SIZE];
+       ID *tmp;
        ID *buf;
+       DB *db;
+       DBC *dbc;
        DBT key;
        DBT data;
-       DBC *dbc;
-       Operation *op;
+       ID dbuf;
+       ID id;
+       ID nid;
+       int rc;
+       int depth;
+       char need_sort;
+       char prefix;
 };
 
 static int
@@ -816,7 +960,7 @@ apply_func(
        EntryInfo *ei = data;
        ID *idl = arg;
 
-       bdb_idl_insert( idl, ei->bei_id );
+       bdb_idl_append_one( idl, ei->bei_id );
        return 0;
 }
 
@@ -827,23 +971,31 @@ hdb_dn2idl_internal(
 {
        BDB_IDL_ZERO( cx->tmp );
 
-       if ( !cx->ei ) {
-               cx->ei = bdb_cache_find_info( cx->bdb, cx->id );
-               if ( !cx->ei ) {
-                       cx->rc = DB_NOTFOUND;
-                       goto saveit;
-               }
-       }
-
        if ( cx->bdb->bi_idl_cache_size ) {
-               cx->key.data = &cx->id;
-               cx->rc = bdb_idl_cache_get(cx->bdb, cx->db, &cx->key, cx->tmp);
-               if ( cx->rc == DB_NOTFOUND ) {
-                       return cx->rc;
+               char *ptr = ((char *)&cx->id)-1;
+
+               cx->key.data = ptr;
+               cx->key.size = sizeof(ID)+1;
+               if ( cx->prefix == DN_SUBTREE_PREFIX ) {
+                       ID *ids = cx->depth ? cx->tmp : cx->ids;
+                       *ptr = cx->prefix;
+                       cx->rc = bdb_idl_cache_get(cx->bdb, cx->db, &cx->key, ids);
+                       if ( cx->rc == LDAP_SUCCESS ) {
+                               if ( cx->depth ) {
+                                       bdb_idl_append( cx->ids, cx->tmp );
+                                       cx->need_sort = 1;
+                               }
+                               return cx->rc;
+                       }
                }
+               *ptr = DN_ONE_PREFIX;
+               cx->rc = bdb_idl_cache_get(cx->bdb, cx->db, &cx->key, cx->tmp);
                if ( cx->rc == LDAP_SUCCESS ) {
                        goto gotit;
                }
+               if ( cx->rc == DB_NOTFOUND ) {
+                       return cx->rc;
+               }
        }
 
        bdb_cache_entryinfo_lock( cx->ei );
@@ -856,11 +1008,24 @@ hdb_dn2idl_internal(
                db_recno_t dkids = cx->ei->bei_dkids;
                ei.bei_parent = cx->ei;
 
+               /* Only one thread should load the cache */
+               while ( cx->ei->bei_state & CACHE_ENTRY_ONELEVEL ) {
+                       bdb_cache_entryinfo_unlock( cx->ei );
+                       ldap_pvt_thread_yield();
+                       bdb_cache_entryinfo_lock( cx->ei );
+                       if ( cx->ei->bei_ckids+1 == cx->ei->bei_dkids ) {
+                               goto synced;
+                       }
+               }
+
+               cx->ei->bei_state |= CACHE_ENTRY_ONELEVEL;
+
                bdb_cache_entryinfo_unlock( cx->ei );
 
                cx->rc = cx->db->cursor( cx->db, NULL, &cx->dbc,
                        cx->bdb->bi_db_opflags );
-               if ( cx->rc ) return cx->rc;
+               if ( cx->rc )
+                       goto done_one;
 
                cx->data.data = &cx->dbuf;
                cx->data.ulen = sizeof(ID);
@@ -869,11 +1034,11 @@ hdb_dn2idl_internal(
 
                /* The first item holds the parent ID. Ignore it. */
                cx->key.data = &cx->nid;
+               cx->key.size = sizeof(ID);
                cx->rc = cx->dbc->c_get( cx->dbc, &cx->key, &cx->data, DB_SET );
                if ( cx->rc ) {
                        cx->dbc->c_close( cx->dbc );
-                       if ( cx->rc == DB_NOTFOUND ) goto saveit;
-                       return cx->rc;
+                       goto done_one;
                }
 
                /* If the on-disk count is zero we've never checked it.
@@ -888,43 +1053,55 @@ hdb_dn2idl_internal(
                cx->data.ulen = BDB_IDL_UM_SIZE * sizeof(ID);
                cx->data.flags = DB_DBT_USERMEM;
 
-               /* Fetch the rest of the IDs in a loop... */
-               while ( (cx->rc = cx->dbc->c_get( cx->dbc, &cx->key, &cx->data,
-                       DB_MULTIPLE | DB_NEXT_DUP )) == 0 ) {
-                       u_int8_t *j;
-                       size_t len;
-                       DB_MULTIPLE_INIT( cx->ptr, &cx->data );
-                       while (cx->ptr) {
-                               DB_MULTIPLE_NEXT( cx->ptr, &cx->data, j, len );
-                               if (j) {
-                                       EntryInfo *ei2;
-                                       diskNode *d = (diskNode *)j;
-                                       short nrlen;
-
-                                       BDB_DISK2ID( j + len - sizeof(ID), &ei.bei_id );
-                                       nrlen = ((d->nrdnlen[0] ^ 0x80) << 8) | d->nrdnlen[1];
-                                       ei.bei_nrdn.bv_len = nrlen;
-                                       /* nrdn/rdn are set in-place.
-                                        * hdb_cache_load will copy them as needed
-                                        */
-                                       ei.bei_nrdn.bv_val = d->nrdn;
-                                       ei.bei_rdn.bv_len = len - sizeof(diskNode)
-                                               - ei.bei_nrdn.bv_len;
-                                       ei.bei_rdn.bv_val = d->nrdn + ei.bei_nrdn.bv_len + 1;
-                                       bdb_idl_insert( cx->tmp, ei.bei_id );
-                                       hdb_cache_load( cx->bdb, &ei, &ei2 );
+               if ( dkids > 1 ) {
+                       /* Fetch the rest of the IDs in a loop... */
+                       while ( (cx->rc = cx->dbc->c_get( cx->dbc, &cx->key, &cx->data,
+                               DB_MULTIPLE | DB_NEXT_DUP )) == 0 ) {
+                               u_int8_t *j;
+                               size_t len;
+                               void *ptr;
+                               DB_MULTIPLE_INIT( ptr, &cx->data );
+                               while (ptr) {
+                                       DB_MULTIPLE_NEXT( ptr, &cx->data, j, len );
+                                       if (j) {
+                                               EntryInfo *ei2;
+                                               diskNode *d = (diskNode *)j;
+                                               short nrlen;
+
+                                               BDB_DISK2ID( j + len - sizeof(ID), &ei.bei_id );
+                                               nrlen = ((d->nrdnlen[0] ^ 0x80) << 8) | d->nrdnlen[1];
+                                               ei.bei_nrdn.bv_len = nrlen;
+                                               /* nrdn/rdn are set in-place.
+                                                * hdb_cache_load will copy them as needed
+                                                */
+                                               ei.bei_nrdn.bv_val = d->nrdn;
+                                               ei.bei_rdn.bv_len = len - sizeof(diskNode)
+                                                       - ei.bei_nrdn.bv_len;
+                                               ei.bei_rdn.bv_val = d->nrdn + ei.bei_nrdn.bv_len + 1;
+                                               bdb_idl_append_one( cx->tmp, ei.bei_id );
+                                               hdb_cache_load( cx->bdb, &ei, &ei2 );
+                                       }
                                }
                        }
                }
+
                cx->rc = cx->dbc->c_close( cx->dbc );
+done_one:
+               bdb_cache_entryinfo_lock( cx->ei );
+               cx->ei->bei_state ^= CACHE_ENTRY_ONELEVEL;
+               bdb_cache_entryinfo_unlock( cx->ei );
+               if ( cx->rc )
+                       return cx->rc;
+
        } else {
                /* The in-memory cache is in sync with the on-disk data.
                 * do we have any kids?
                 */
+synced:
                cx->rc = 0;
                if ( cx->ei->bei_ckids > 0 ) {
-                       /* Walk the kids tree; order is irrelevant since bdb_idl_insert
-                        * will insert in sorted order.
+                       /* Walk the kids tree; order is irrelevant since bdb_idl_sort
+                        * will sort it later.
                         */
                        avl_apply( cx->ei->bei_kids, apply_func,
                                cx->tmp, -1, AVL_POSTORDER );
@@ -932,36 +1109,45 @@ hdb_dn2idl_internal(
                bdb_cache_entryinfo_unlock( cx->ei );
        }
 
-saveit:
-       if ( cx->bdb->bi_idl_cache_max_size ) {
-               cx->key.data = &cx->id;
+       if ( !BDB_IDL_IS_RANGE( cx->tmp ) && cx->tmp[0] > 3 )
+               bdb_idl_sort( cx->tmp, cx->buf );
+       if ( cx->bdb->bi_idl_cache_max_size && !BDB_IDL_IS_ZERO( cx->tmp )) {
+               char *ptr = ((char *)&cx->id)-1;
+               cx->key.data = ptr;
+               cx->key.size = sizeof(ID)+1;
+               *ptr = DN_ONE_PREFIX;
                bdb_idl_cache_put( cx->bdb, cx->db, &cx->key, cx->tmp, cx->rc );
        }
-       ;
+
 gotit:
        if ( !BDB_IDL_IS_ZERO( cx->tmp )) {
                if ( cx->prefix == DN_SUBTREE_PREFIX ) {
-                       if (cx->ei->bei_state & CACHE_ENTRY_NO_GRANDKIDS) {
-                               bdb_idl_union( cx->ids, cx->tmp );
-                       } else {
+                       bdb_idl_append( cx->ids, cx->tmp );
+                       cx->need_sort = 1;
+                       if ( !(cx->ei->bei_state & CACHE_ENTRY_NO_GRANDKIDS)) {
                                ID *save, idcurs;
                                EntryInfo *ei = cx->ei;
                                int nokids = 1;
                                save = cx->op->o_tmpalloc( BDB_IDL_SIZEOF( cx->tmp ),
                                        cx->op->o_tmpmemctx );
                                BDB_IDL_CPY( save, cx->tmp );
-                               bdb_idl_union( cx->ids, cx->tmp );
 
                                idcurs = 0;
+                               cx->depth++;
                                for ( cx->id = bdb_idl_first( save, &idcurs );
                                        cx->id != NOID;
                                        cx->id = bdb_idl_next( save, &idcurs )) {
+                                       cx->ei = bdb_cache_find_info( cx->bdb, cx->id );
+                                       if ( !cx->ei ||
+                                               ( cx->ei->bei_state & CACHE_ENTRY_NO_KIDS ))
+                                               continue;
+
                                        BDB_ID2DISK( cx->id, &cx->nid );
-                                       cx->ei = NULL;
                                        hdb_dn2idl_internal( cx );
                                        if ( !BDB_IDL_IS_ZERO( cx->tmp ))
                                                nokids = 0;
                                }
+                               cx->depth--;
                                cx->op->o_tmpfree( save, cx->op->o_tmpmemctx );
                                if ( nokids ) ei->bei_state |= CACHE_ENTRY_NO_GRANDKIDS;
                        }
@@ -1003,16 +1189,23 @@ hdb_dn2idl(
        cx.ei = e->e_id ? BEI(e) : &bdb->bi_cache.c_dntree;
        cx.bdb = bdb;
        cx.db = cx.bdb->bi_dn2id->bdi_db;
-       cx.prefix = op->ors_scope == LDAP_SCOPE_ONELEVEL
-               DN_ONE_PREFIX : DN_SUBTREE_PREFIX;
+       cx.prefix = (op->ors_scope == LDAP_SCOPE_ONELEVEL) ?
+               DN_ONE_PREFIX : DN_SUBTREE_PREFIX;
        cx.ids = ids;
-       cx.buf = stack;
+       cx.tmp = stack;
+       cx.buf = stack + BDB_IDL_UM_SIZE;
        cx.op = op;
+       cx.need_sort = 0;
+       cx.depth = 0;
 
-       BDB_IDL_ZERO( ids );
        if ( cx.prefix == DN_SUBTREE_PREFIX ) {
-               bdb_idl_insert( ids, cx.id );
+               ids[0] = 1;
+               ids[1] = cx.id;
+       } else {
+               BDB_IDL_ZERO( ids );
        }
+       if ( cx.ei->bei_state & CACHE_ENTRY_NO_KIDS )
+               return LDAP_SUCCESS;
 
        DBTzero(&cx.key);
        cx.key.ulen = sizeof(ID);
@@ -1021,6 +1214,21 @@ hdb_dn2idl(
 
        DBTzero(&cx.data);
 
-       return hdb_dn2idl_internal(&cx);
+       hdb_dn2idl_internal(&cx);
+       if ( cx.need_sort ) {
+               char *ptr = ((char *)&cx.id)-1;
+               if ( !BDB_IDL_IS_RANGE( cx.ids ) && cx.ids[0] > 3 ) 
+                       bdb_idl_sort( cx.ids, cx.tmp );
+               cx.key.data = ptr;
+               cx.key.size = sizeof(ID)+1;
+               *ptr = cx.prefix;
+               cx.id = e->e_id;
+               bdb_idl_cache_put( cx.bdb, cx.db, &cx.key, cx.ids, cx.rc );
+       }
+
+       if ( cx.rc == DB_NOTFOUND )
+               cx.rc = LDAP_SUCCESS;
+
+       return cx.rc;
 }
 #endif /* BDB_HIER */