fileserver: Redirect within the original URL

[diamondburned]

This commit changes the file_server directive to redirect using the original request's URL instead of the possibly trimmed URL. This should make file_server work with handle_path.

This fix is taken from mholt's comment in https://caddy.community/t/file-servers-on-different-paths-not-working/11698/11.

---

For testing purposes, below Caddyfile should reproduce the before and afters:

{
	debug
	admin off
}

http://127.0.0.1:28475 {
	root * /home/diamond

	handle_path /browse* {
		file_server browse
	}

	respond "Hi"
}

When accessing http://127.0.0.1:28475/browse, the redirection would lead to /; after the changes, it would lead to /browse/.

[CLAassistant]

All committers have signed the CLA.

[francislavoie]

Thanks for this! I'd been meaning to look into this at some point. Glad someone got to it before me 😁

FWIW, I realized this was a problem in https://caddy.community/t/route-directive-file-server-not-picking-up-index-html/12562 as well.

[francislavoie]

I'm also worried about the situation where the original URL already has a / but the rewritten URL doesn't (can happen when the rewrite directive is used, instead of it being inside of a handle_path).

I think the original request path needs to be considered as well to not do a redirect when it doesn't make sense.

[diamondburned]

I think the original request path needs to be considered as well to not do a redirect when it doesn't make sense.

Do you mean both the original path and the new path, or just the original path? I've changed the code to check the original path instead; I think it makes more sense that way as well.

[francislavoie]

Do you mean both the original path and the new path, or just the original path? I've changed the code to check the original path instead; I think it makes more sense that way as well.

I was thinking both, but yeah I think that makes sense to only look at the original request, because the point of the canonicalization redirect is to make sure that from the browser's perspective, the URL does have a trailing / so that relative path links go to the right place. 👍

[mholt]

Thanks for this change, let's try it out.

Edit: This PR turned out to be a regression and has been reverted. See #4205.

Edit 2: I have pushed a more correct implementation of this change in fbd6560, where we only redirect if the base (last element, a.k.a. filename) of the path was not rewritten, that will be released in v2.4.3.