Can't resolve IP address with square brakets

[digitwolf]

Problem

Sending to destination like https://[::1]/ fails with error invalid dnsname.

Error is thrown here https://github.com/rustls/hyper-rustls/blob/main/src/connector.rs#L95

Analysis

hyper-rustls resolves the Hostname as hyper::Uri::host
This makes hostname = "[::1]" which is HTTP-specific form of writing IPv6 address.

Because IPv6 addresses contain colons, they cannot be directly used in URLs because the colons would conflict with both the protocol declaration (http:// or https://) and port numbers. Therefore, when a literal IPv6 address is used, it is encased in a bracket, like http://[2001:db8::1]:80
https://en.wikiversity.org/wiki/IPv6

Root-Cause

hyper-rustls is resolving the host name using rustls::ServerName::try_from(hostname) which does not expect square brackets in IP because internally it uses

Proposed solution

Since square brakets are coming from hyper and not part standard IP notation I propose that we remove any square brakets inside of the connector before calling the rustls::ServerName::try_from(hostname)

Will post a PR in bit

[djc]

Are you setting an actual DNS server name separately? Because rustls doesn't currently support IP addresses in certificates it won't otherwise be able to establish a session anyway.

[digitwolf]

I am not using a DNS, but I am using custom certs and all custom rustls cert verifiers. I have checked locally and this works.

[digitwolf]

This is the only change (in external codebase) that is needed to make requests to IP with rustls work #182

[BiagioFesta]

Because rustls doesn't currently support IP addresses in certificates

Shouldn't rustls/rustls#1032 mitigate this?

I see this current main branch depends on rustls 0.20.1 (so probably that PR is not in the code)

With that patch on rustls and a custom cert verifiers (thus, dangerous config), it should be feasible if I am not wrong.

[digitwolf]

Hi BiagioFesta!
I tried your idea locally. I've upgraded hyper-rustls dependency of rustls to 0.20.7 however on it's own that does not address it.
I think that the main issue still exists - square brackets are not accounted for in rustls.
But I do like the idea of upgrading the dependency. I'll update the PR to include the version bump.

[digitwolf]

Guys, thank you for pulling in the PR.
Does anyone have an idea when I can expect it to be released?

I have to make a call today to decide if I am pulling in this fix from crates.io or if I need to embed a copy of hyper-rustls in my software release.

[djc]

I've submitted #183.