Ticket 47978: Deadlock between two MODs on the same entry between entry cache and backend lock

Bug Description:
	During a modify, the modified entry gets into the entry cache and is locked.
	If after the be_txn_postop/txn_commit and before the modify returns
	the modified entry gets out of the entry cache, the entry is not unlocked.
	It can lead to hang as soon as an other write operation hit that unlocked entry.

	This is a side effect of fix:
		#47834 - Tombstone_to_glue: if parents are also converted to glue, the target entry's DN must be adjusted.

Fix Description:
	When the entry is locked, set a flag so that can later be unlocked
	independently of its presence in the entry cache

https://fedorahosted.org/389/ticket/47978

Reviewed by: Noriko Hosoi (Thanks Noriko)

Platforms tested: F22 (IPA CI test test_integration/test_vault.py, one failure out 2-4)

Flag Day: no

Doc impact: no

ldap/servers/slapd/back-ldbm/ldbm_modify.c

@@ -391,6 +391,7 @@ ldbm_back_modify( Slapi_PBlock *pb )
 	int mod_count = 0;
 	int not_an_error = 0;
 	int fixup_tombstone = 0;
+	int ec_locked = 0;
 
 	slapi_pblock_get( pb, SLAPI_BACKEND, &be);
 	slapi_pblock_get( pb, SLAPI_PLUGIN_PRIVATE, &li );
@@ -799,6 +800,7 @@ ldbm_back_modify( Slapi_PBlock *pb )
 	CACHE_RETURN( &inst->inst_cache, &e );
 	/* lock new entry in cache to prevent usage until we are complete */
 	cache_lock_entry( &inst->inst_cache, ec );
+	ec_locked = 1;
 	postentry = slapi_entry_dup( ec->ep_entry );
 	slapi_pblock_set( pb, SLAPI_ENTRY_POST_OP, postentry );
 
@@ -919,7 +921,7 @@ common_return:
 	slapi_mods_done(&smods);
 	
 	if (inst) {
-		if (cache_is_in_cache(&inst->inst_cache, ec)) {
+		if (ec_locked || cache_is_in_cache(&inst->inst_cache, ec)) {
 			cache_unlock_entry(&inst->inst_cache, ec);
 		} else if (e) {
 			/* if ec was not in cache, cache_replace was not done.