gateway: migrate subdomain and dnslink code

[lidel]

Logic that parses Host header and translates it to content path is still in Kubo.

We need to migrate it as well, because the gateway binary for Saturn MUST also support subdomain (for dweb.link) and DNSLink gateways (for things like https://en.wikipedia-on-ipfs.org).

Ref.

• kubo
  • https://github.com/ipfs/kubo/blob/f54b2bcf6061219f7f481e8660e6707ae3bc3c38/cmd/ipfs/daemon.go#L841-L842
  • https://github.com/ipfs/kubo/blob/4e8271b1021f7d31e7b201640e6214a521f328e7/core/corehttp/hostname.go#L42-L43
• https://github.com/ipfs/specs/blob/main/http-gateways/SUBDOMAIN_GATEWAY.md
• https://github.com/ipfs/specs/blob/main/http-gateways/DNSLINK_GATEWAY.md

Other ToDos:

• Update CAR Gateway example to support subdomains on localhost

[lidel]

@hacdias this is lower priority than PoC examples (#141 and #142), but something we will need for the production binary for dweb.link.

Not sure how tricky this is, but these additional types of gateways are essentially wrappers executed before passing request to the path one.

I think this is a good patter/model to follow in our library.
See mux and childMux in this file.

[hacdias]

(Oops, moving on...)

I think this is a good patter/model to follow in our library. See mux and childMux in this file.

@lidel Any reason to do so instead of doing the common middleware pattern we see in Go?

func Middleware(http.Handler) http.Handler

*http.Mux is a http.Handler but the opposite is not necessarily true. That is how I implemented it right now, but I can change if there's any reason to use a different pattern that what is common in the Go-land.

[lidel]

@hacdias don't remember, possible it is this way only for legacy reasons (subdomains were bolted-on years after path gateway was an ossified thing, did nto want to change too much because we had virtually no tests for anything gateway related back then).

If you can switch Kubo to middleware pattern and have subdomain sharness pass, then we can clean this up 👍