Broker bridge 2.0.10: Corruption of pollfds on Windows

[NiclasLindgren]

net__socket_close is called which deletes/closes the socket from db.contexts_by_sock, but doesn't call mux__delete(mosq) before trying another connect, when connect is called the socket fd is changed in the context, but the pollfd_index points to the old spot (which also isn't cleared)

This happens in bridge_check which calls mosquitto__check_keepalive which closes the socket, which triggers a call to rc = bridge__connect(context); in the same loop

I am not sure what the proper solution is, because it doesn't seem like you are supposed to call mux__delete from net__socket_close, rather that maybe bridge_check_pending should detect the socket as stale and call mux__delete

however adding mux__delete here (in net_mosq::net__socket_close) works

		HASH_FIND(hh_sock, db.contexts_by_sock, &mosq->sock, sizeof(mosq->sock), mosq_found);
		if(mosq_found){
			HASH_DELETE(hh_sock, db.contexts_by_sock, mosq_found);
			mux__delete(mosq);

This also seems to solve the issue where the broker is "stalled" if there is a bridge that isn't reachable (#2172 )

[NiclasLindgren]

Another way to make it work would be

static void bridge_check_pending(struct mosquitto *context)
{
int err;
socklen_t len;

if (context->sock == INVALID_SOCKET && context->pollfd_index != -1)
{
	mux__delete(context);
}

But it feels like it is dangerous because alot of code calls net__socket_close which doesn't immediately removes the socket from the poll array

[ralight]

Thanks, you're right it needs some careful consideration of where it should happen. I'll take a look.

[NiclasLindgren]

Cool, I was wrong that it solves #2172, that problem is much more serious, it seems when you have more than 1 bridge configured only one of them will ever work, all the other (even if reachable) will never connect up, I will investigate. I suspect this must all work on Linux but I am on Windows