b2b_entities: avoid deadlock for certain scenarios

When a call was terminted with a BYE from the handler of the 200 OK
reply, the entities lock would remain locked, generating a deadlock.
This commit fixes this (and possible other) situations by generically
keeping a track when the lock is taken and unlock it before exiting.

Thanks to Vlad Patrascu for helping with debugging and solution for this

(cherry picked from commit 87661e1)

modules/b2b_entities/dlg.c

@@ -2658,6 +2658,7 @@ void b2b_tm_cback(struct cell *t, b2b_table htable, struct tmcb_params *ps)
 	int old_route_type;
 	bin_packet_t storage;
 	int b2b_ev = -1;
+	int lock_taken = 0;
 	struct b2b_context *ctx;
 	int b2b_cb_flags = 0;
 	unsigned int reqmask;
@@ -3394,26 +3395,28 @@ void b2b_tm_cback(struct cell *t, b2b_table htable, struct tmcb_params *ps)
 b2b_route:
 
 	if (B2BE_SERIALIZE_STORAGE()) {
+		lock_get(&htable[hash_index].lock);
+		lock_taken = 1;
 		if (dlg_state == B2B_CONFIRMED && prev_state == B2B_MODIFIED) {
-			lock_get(&htable[hash_index].lock);
 
 			if (dlg->state != B2B_TERMINATED) {
 				b2b_ev = B2B_EVENT_UPDATE;
 				b2b_run_cb(dlg, hash_index, etype, B2BCB_TRIGGER_EVENT, b2b_ev,
 					&storage, serialize_backend);
-			} else
+			} else {
 				b2b_ev = -1;
+			}
 		} else if (b2b_ev == B2B_EVENT_CREATE) {
-			lock_get(&htable[hash_index].lock);
 
 			if (dlg->state != B2B_TERMINATED) {
 				b2b_run_cb(dlg, hash_index, etype, B2BCB_TRIGGER_EVENT, b2b_ev,
 					&storage, serialize_backend);
 
 				if (b2be_db_mode == WRITE_THROUGH)
 					b2be_db_insert(dlg, etype);
-			} else
+			} else {
 				b2b_ev = -1;
+			}
 		}
 	}
 
@@ -3437,8 +3440,9 @@ void b2b_tm_cback(struct cell *t, b2b_table htable, struct tmcb_params *ps)
 		if (b2be_db_update(dlg, etype) < 0)
 			LM_ERR("Failed to update in database\n");
 		lock_release(&htable[hash_index].lock);
-	} else if (b2b_ev != -1)
+	} else if (lock_taken) {
 		lock_release(&htable[hash_index].lock);
+	}
 
 	if (b2be_cluster) {
 		if (b2b_ev == B2B_EVENT_UPDATE)