| 1 | //! The catch-all route: map a URL path onto a repository and an action. |
| 2 | //! |
| 3 | //! Repository names may contain `/`, so the longest prefix of the path that |
| 4 | //! names a repository wins; the next segment is the action. Ferrenium's own |
| 5 | //! endpoints live under `/-/`, which never names a repository. |
| 6 | |
| 7 | use std::sync::Arc; |
| 8 | |
| 9 | use axum::extract::{Query, Request, State}; |
| 10 | use axum::http::{StatusCode, header}; |
| 11 | use axum::response::{IntoResponse, Redirect, Response}; |
| 12 | |
| 13 | use crate::handlers::error; |
| 14 | use crate::handlers::repo::{PageQuery, RepoRequest, handle}; |
| 15 | use crate::index::{Repo, RepoIndex}; |
| 16 | use crate::render::{encode_path, percent_decode}; |
| 17 | use crate::state::AppState; |
| 18 | |
| 19 | pub enum Match<'a> { |
| 20 | Repo { |
| 21 | repo: Arc<Repo>, |
| 22 | rest: &'a [Vec<u8>], |
| 23 | }, |
| 24 | /// `/{repo}.git` → `/{repo}`. |
| 25 | DotGit { |
| 26 | name: String, |
| 27 | }, |
| 28 | None, |
| 29 | } |
| 30 | |
| 31 | /// Split a raw URL path into percent-decoded segments, dropping empty ones. |
| 32 | /// `None` for `.`/`..` segments, which are never valid. |
| 33 | pub fn segments(path: &str) -> Option<Vec<Vec<u8>>> { |
| 34 | let segments: Vec<Vec<u8>> = path |
| 35 | .split('/') |
| 36 | .filter(|s| !s.is_empty()) |
| 37 | .map(percent_decode) |
| 38 | .collect(); |
| 39 | let bad = segments |
| 40 | .iter() |
| 41 | .any(|s| s == b"." || s == b".." || s.contains(&b'/') || s.contains(&0)); |
| 42 | (!bad).then_some(segments) |
| 43 | } |
| 44 | |
| 45 | /// Find the repository named by the longest prefix of `segments`. |
| 46 | pub fn find_repo<'a>(index: &RepoIndex, segments: &'a [Vec<u8>]) -> Match<'a> { |
| 47 | if segments.first().is_some_and(|s| s == b"-") { |
| 48 | return Match::None; |
| 49 | } |
| 50 | for n in (1..=segments.len()).rev() { |
| 51 | let parts: Option<Vec<&str>> = segments[..n] |
| 52 | .iter() |
| 53 | .map(|s| std::str::from_utf8(s).ok()) |
| 54 | .collect(); |
| 55 | let Some(parts) = parts else { |
| 56 | continue; |
| 57 | }; |
| 58 | let name = parts.join("/"); |
| 59 | if let Some(repo) = index.repos.get(&name) { |
| 60 | return Match::Repo { |
| 61 | repo: repo.clone(), |
| 62 | rest: &segments[n..], |
| 63 | }; |
| 64 | } |
| 65 | if let Some(stripped) = name.strip_suffix(".git") |
| 66 | && index.repos.contains_key(stripped) |
| 67 | { |
| 68 | // `/{repo}.git/…` is the HTTPS clone endpoint (not served yet). |
| 69 | return if n == segments.len() { |
| 70 | Match::DotGit { |
| 71 | name: stripped.to_string(), |
| 72 | } |
| 73 | } else { |
| 74 | Match::None |
| 75 | }; |
| 76 | } |
| 77 | } |
| 78 | Match::None |
| 79 | } |
| 80 | |
| 81 | pub async fn dispatch(State(state): State<Arc<AppState>>, req: Request) -> Response { |
| 82 | let config = state.config.load_full(); |
| 83 | let index = state.index.load(); |
| 84 | let Some(segments) = segments(req.uri().path()) else { |
| 85 | return error::page(&state, StatusCode::NOT_FOUND, "PAGE NOT FOUND"); |
| 86 | }; |
| 87 | let (repo, rest) = match find_repo(&index, &segments) { |
| 88 | Match::Repo { repo, rest } => (repo, rest.to_vec()), |
| 89 | Match::DotGit { name } => { |
| 90 | let url = format!("{}{}", config.server.base_path, encode_path(&name)); |
| 91 | return Redirect::permanent(&url).into_response(); |
| 92 | } |
| 93 | Match::None => return error::page(&state, StatusCode::NOT_FOUND, "PAGE NOT FOUND"), |
| 94 | }; |
| 95 | let query = Query::<PageQuery>::try_from_uri(req.uri()) |
| 96 | .map(|q| q.0) |
| 97 | .unwrap_or_default(); |
| 98 | let request = RepoRequest { |
| 99 | repo, |
| 100 | segments: rest, |
| 101 | query, |
| 102 | if_none_match: req.headers().get(header::IF_NONE_MATCH).cloned(), |
| 103 | config, |
| 104 | }; |
| 105 | match state.blocking(move |state| handle(state, request)).await { |
| 106 | Ok(response) => response, |
| 107 | Err(e) => { |
| 108 | tracing::error!("request failed: {e:#}"); |
| 109 | error::page(&state, StatusCode::INTERNAL_SERVER_ERROR, "INTERNAL ERROR") |
| 110 | } |
| 111 | } |
| 112 | } |
| 113 | |
| 114 | #[cfg(test)] |
| 115 | mod tests { |
| 116 | use super::*; |
| 117 | use crate::index::RepoIndex; |
| 118 | |
| 119 | #[test] |
| 120 | fn segment_parsing() { |
| 121 | assert_eq!( |
| 122 | segments("/a//b%2Fc/").map(|s| s.len()), |
| 123 | None, |
| 124 | "encoded slash is rejected" |
| 125 | ); |
| 126 | assert_eq!( |
| 127 | segments("/a//b/%C3%BC/").unwrap(), |
| 128 | vec![b"a".to_vec(), b"b".to_vec(), "ü".as_bytes().to_vec()] |
| 129 | ); |
| 130 | assert!(segments("/a/../b").is_none()); |
| 131 | assert!(segments("/a/%2e%2e/b").is_none()); |
| 132 | assert!(segments("/").unwrap().is_empty()); |
| 133 | } |
| 134 | |
| 135 | #[test] |
| 136 | fn no_repos_no_match() { |
| 137 | let index = RepoIndex::default(); |
| 138 | let segs = segments("/-/x").unwrap(); |
| 139 | assert!(matches!(find_repo(&index, &segs), Match::None)); |
| 140 | } |
| 141 | } |