More deadlock tweaks.

servers/slapd/back-bdb/cache.c

@@ -239,7 +239,11 @@ bdb_entryinfo_add_internal(
 			 */
 			if ( bdb_cache_entry_db_lock( env, locker, elru, 1, 1,
 				&lock ) == 0 ) {
-				if ( !elru->bei_e ) {
+				/* 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( env, &lock );
 					continue;
 				}
@@ -422,8 +426,10 @@ hdb_cache_find_parent(
 {
 	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;
 
 	ei.bei_id = id;
 	ei.bei_kids = NULL;
@@ -450,16 +456,24 @@ hdb_cache_find_parent(
 		if ( avl_insert( &bdb->bi_cache.c_idtree, (caddr_t)ein,
 			bdb_id_cmp, avl_dup_error ) ) {
 
-			/* Hm, can this really happen? */
+			/* 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 );
 				bdb_cache_entryinfo_unlock( ein );
 			}
+
+			if ( !eir ) {
+				addlru = 0;
+			}
 		}
 
 		/* If this is the first time, save this node
@@ -470,27 +484,28 @@ hdb_cache_find_parent(
 		/* If there was a previous node, link it to this one */
 		if ( ei2 ) ei2->bei_parent = ein;
 
+		/* Look for this node's parent */
 		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;
 		}
+		ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
 
+		/* Got the parent, link in and we're done. */
 		if ( ei2 ) {
-			ein->bei_parent = ei2;
 			bdb_cache_entryinfo_lock( ei2 );
+			ein->bei_parent = ei2;
 			avl_insert( &ei2->bei_kids, (caddr_t)ein, bdb_rdn_cmp,
 				avl_dup_error);
 			bdb_cache_entryinfo_unlock( ei2 );
-			*res = eir;
 			bdb_cache_entryinfo_lock( eir );
-		}
+
-		ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
+			/* Reset all the state info */
-		if ( ei2 ) {
-			/* Found a link. Reset all the state info */
 			for (ein = eir; ein != ei2; ein=ein->bei_parent)
 				ein->bei_state &= ~CACHE_ENTRY_NOT_LINKED;
+			*res = eir;
 			break;
 		}
 		ei.bei_kids = NULL;
@@ -532,12 +547,26 @@ again:		ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
 		*eip = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
 					(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 )) {
 				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 ) {
 					return DB_NOTFOUND;
 				}
+				/* otherwise, wait for the info to free up */
+				ldap_pvt_thread_yield();
+				goto again;
+			}
+			/* If this info isn't hooked up to its parent yet,
+			 * unlock and wait for it to be fully initialized
+			 */
+			if ( (*eip)->bei_state & CACHE_ENTRY_NOT_LINKED ) {
+				bdb_cache_entryinfo_unlock( *eip );
+				ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
 				ldap_pvt_thread_yield();
 				goto again;
 			}