Docs: clarify how to use proxy-headers with unix sockets

[PeterJCLaw]

It would be great if it could be documented that when using a unix socket the client address which ProxyHeadersMiddleware gets will be None and that you therefore need to pass * as the trusted hosts to get the proxy headers to work.

Even better would be for the proxy headers to "just work" when using a unix socket, in the same way that ProxyHeadersMiddleware trusts localhost by default, though I suspect that may be more involved.

[euri10]

would be a good addition yes, PR welcome @PeterJCLaw :)

[PeterJCLaw]

So I had a quick look into the internals here and I can't see an easy way to be confident in the middleware that the connection is a unix socket.

I think that ideally we'd get something in scope['client'] which would tell us that, however the ASGI spec defines that value to be either a 2-tuple (host, port) or None which kinda rules that out.
While uvicorn does pass None there for a unix socket, it seems to also be the case that various error conditions in uvicorn.protocols.utils.get_remote_addr could cause it to be None for other reasons. Certainly as this is somewhat security sensitive I think I'd want to have actively determined that we're looking at a unix socket, rather than merely not a happy-path network socket.

scope['server'] might hold more promise, though uvicorn.protocols.utils.get_local_addr also seems to have some soft fallback logic which I'm unsure of.

Is there something I'm missing here which might make this possible, or is documenting the behaviour the only solution? (I'm happy to look at doing that anyway, just would be nice for it to work automatically)