When dns.ipv6 disabled ipv6 will fail

[spongebob888]

If dns.ipv6 is disabled, even if global ipv6 is enabled, the ipv6 traffic will fail in direct outbound. This is caused by following code:

clash-rs/clash-lib/src/proxy/direct/mod.rs

Lines 67 to 71 in 61736d2

	let remote_ip = resolver
	.resolve(sess.destination.host().as_str(), false)
	.map_err(map_io_error)
	.await?
	.ok_or_else(|| std::io::Error::other("no dns result"))?;

It will regard ipv6 address as a domain and try to resolve it and failed.

Don't know this is intended or a bug.

[mrushbersuz]

Hi, I've been looking into this issue. From my understanding, when dns.ipv6 is disabled, the resolver may treat IPv6 addresses as domain names and attempt DNS resolution, which fails because IPv6 resolution is disabled. This occurs in direct outbound where resolver.resolve is called with the destination host string.

I think the fix is to check if the host is already a valid IPv6 (or IPv4) address before calling resolver.resolve. If it's an IP address, we can parse it directly and skip DNS resolution. This would align with typical behavior where IP addresses don't require DNS lookup.

I'd like to work on a PR to add this IP address detection in the connect_stream and connect_datagram methods (and possibly other places). Let me know if this approach makes sense.

[spongebob888]

Makes sense to me

[ibigbug]

I think we somehow do it already https://github.com/Watfaq/clash-rs/blob/master/clash-lib/src/app/dns/resolver/enhanced.rs

Maybe improve the IP parse logic in both resolve functions and returns IP/error accordingly.

[spongebob888]

I didn't get it. So in the case globle ipv6=true while dns.ipv6=false. Should we return a error if received an IPv6 address

[ibigbug]

Actually nvm.

After thinking about it, I agree that when global v6 enabled and dns.v6 disabled, we should still process v6 traffic if we can, without going through the resolver.

So yeah i think the above suggested solution should work best.