forks/sorcery

git clone https://git.gae.moe/forks/sorcery

Charlotte Somperf: prioritize push-triggered builds over background scans2688dec2026-10-12

//! `sorceryd`: serves generated sites and git data from behind nginx.

use std::fs::{self, Permissions};
use std::os::unix::fs::{FileTypeExt, PermissionsExt};
use std::path::{Path, PathBuf};
use std::sync::Arc;
use std::time::Duration;

use anyhow::{Context, Result, bail};
use arc_swap::ArcSwap;
use axum::Router;
use axum::http::{HeaderValue, StatusCode};
use axum::response::{IntoResponse, Response};
use axum::routing::{any, get, post};
use tokio::net::UnixListener;
use tokio::sync::{Mutex, Notify, OwnedSemaphorePermit, RwLock, Semaphore, watch};

use crate::catalog::Repository;
use crate::highlight::Cache as GrammarCache;
use cache::{BuildQueue, RepoEntry, Snapshot};

mod cache;
mod control;
mod git;
mod objects;
mod site;

#[derive(Clone)]
pub struct Config {
    pub repositories: PathBuf,
    pub cache: PathBuf,
    pub socket: PathBuf,
    pub instance_name: String,
    pub internal_prefix: String,
    /// Internal location rooted at the *repositories* dir, for `.git` serving.
    pub git_internal_prefix: String,
    pub check_interval: Duration,
    pub max_git_processes: usize,
    /// Local socket accepting refresh requests; see `control`.
    pub control_socket: Option<PathBuf>,
    /// Public base URL (e.g. `https://git.example.org`); enables the clone
    /// command on repository pages.
    pub clone_url_base: Option<String>,
}

impl Config {
    fn repo_cache(&self, repository: &Repository) -> PathBuf {
        self.cache.join(&repository.user).join(&repository.name)
    }

    fn site_url(&self, repository: &Repository) -> Option<String> {
        let base = self.clone_url_base.as_ref()?.trim_end_matches('/');
        Some(format!("{base}/{}/{}", repository.user, repository.name))
    }
}

type AppState = Arc<Inner>;

struct Inner {
    config: Config,
    /// Immutable view of the repository catalog, replaced wholesale by the
    /// background scanner. Request handlers only ever read it.
    snapshot: ArcSwap<Snapshot>,
    /// Builds hold this shared; garbage collection needs it exclusively.
    cache_lock: RwLock<()>,
    background: watch::Sender<Vec<RepoEntry>>,
    build_queue: BuildQueue,
    rescan_requested: Notify,
    rescan_gate: Mutex<()>,
    git_processes: Arc<Semaphore>,
    highlights: Arc<GrammarCache>,
}

type WebError = (StatusCode, String);
type WebResult<T> = Result<T, WebError>;

pub async fn run(config: Config) -> Result<()> {
    if config.instance_name.trim().is_empty() {
        bail!("--instance-name must not be empty");
    }
    if !config.internal_prefix.starts_with('/') {
        bail!("--internal-prefix must start with /");
    }
    if config.max_git_processes == 0 {
        bail!("--max-git-processes must be greater than zero");
    }
    fs::create_dir_all(&config.cache)?;
    site::stage_assets(&config.cache)?;
    let listener = bind_unix(&config.socket)?;
    let control = config
        .control_socket
        .as_deref()
        .map(bind_unix)
        .transpose()?;
    if let Some(path) = &config.control_socket {
        fs::set_permissions(path, Permissions::from_mode(0o600))?;
    }

    let highlights = Arc::new(GrammarCache::new(config.cache.join(".arborium")));
    let state = Inner::new(config, highlights);
    tokio::spawn(cache::background_worker(
        state.clone(),
        state.background.subscribe(),
    ));
    state.rescan().await.context("initial repository scan")?;
    tokio::spawn(cache::scan_periodically(state.clone()));
    if let Some(control) = control {
        tokio::spawn(control::serve(state.clone(), control));
    }

    let app = Router::new()
        .route("/", get(site::index))
        .route("/css/style.css", get(site::stylesheet))
        .route("/js/{*path}", get(site::js_asset))
        .route("/fonts/{*path}", get(site::font_asset))
        .route("/{user}", get(site::user_redirect))
        .route("/{user}/", get(site::user_root))
        .route("/{user}/{repo}", get(site::add_trailing_slash))
        .route("/{user}/{repo}/", get(site::repo_root))
        .route("/{user}/{repo}/.git/{*path}", get(git::git_dir))
        .route("/{user}/{repo}/info/refs", get(git::info_refs))
        .route("/{user}/{repo}/git-upload-pack", post(git::upload_pack))
        .route("/{user}/{repo}/obj", any(objects::query))
        .route("/{user}/{repo}/raw/{oid}/{*path}", get(git::raw_blob))
        .route("/{user}/{repo}/{*path}", get(site::repo_path))
        .with_state(state.clone());

    let result = axum::serve(listener, app)
        .with_graceful_shutdown(shutdown_signal())
        .await;
    let sockets = [
        Some(&state.config.socket),
        state.config.control_socket.as_ref(),
    ];
    for socket in sockets.into_iter().flatten() {
        if socket.exists() {
            fs::remove_file(socket)?;
        }
    }
    result?;
    Ok(())
}

/// Bind `path`, replacing a stale socket left behind by an earlier run.
fn bind_unix(path: &Path) -> Result<UnixListener> {
    if let Some(parent) = path.parent() {
        fs::create_dir_all(parent)?;
    }
    if let Ok(metadata) = fs::symlink_metadata(path) {
        if !metadata.file_type().is_socket() {
            bail!("refusing to replace non-socket {}", path.display());
        }
        fs::remove_file(path)?;
    }
    UnixListener::bind(path).with_context(|| format!("binding {}", path.display()))
}

async fn shutdown_signal() {
    let mut terminate = tokio::signal::unix::signal(tokio::signal::unix::SignalKind::terminate())
        .expect("installing SIGTERM handler");
    tokio::select! {
        _ = tokio::signal::ctrl_c() => {}
        _ = terminate.recv() => {}
    }
}

impl Inner {
    fn new(config: Config, highlights: Arc<GrammarCache>) -> AppState {
        Arc::new(Inner {
            snapshot: ArcSwap::from_pointee(Snapshot::default()),
            cache_lock: RwLock::new(()),
            background: watch::channel(Vec::new()).0,
            build_queue: BuildQueue::new(),
            rescan_requested: Notify::new(),
            rescan_gate: Mutex::new(()),
            git_processes: Arc::new(Semaphore::new(config.max_git_processes)),
            highlights,
            config,
        })
    }

    /// Look up a repository in the current snapshot, tolerating a `.git`
    /// suffix on the request (so `git clone .../user/repo.git` works too).
    fn resolve(&self, user: &str, name: &str) -> WebResult<RepoEntry> {
        let name = name.strip_suffix(".git").unwrap_or(name);
        self.snapshot
            .load()
            .repos
            .get(&format!("{user}/{name}"))
            .cloned()
            .ok_or_else(|| (StatusCode::NOT_FOUND, "repository not found".into()))
    }

    fn git_permit(&self) -> WebResult<OwnedSemaphorePermit> {
        self.git_processes.clone().try_acquire_owned().map_err(|_| {
            (
                StatusCode::SERVICE_UNAVAILABLE,
                "git service is busy".into(),
            )
        })
    }
}

/// Hand the (already percent-encoded) path beneath `prefix` to nginx.
fn x_accel(prefix: &str, encoded_path: &str) -> WebResult<Response> {
    let location = format!("{}/{encoded_path}", prefix.trim_end_matches('/'));
    let location = HeaderValue::from_str(&location).map_err(|_| {
        (
            StatusCode::INTERNAL_SERVER_ERROR,
            "invalid internal path".into(),
        )
    })?;
    Ok([("x-accel-redirect", location)].into_response())
}

fn internal(error: impl std::fmt::Display) -> WebError {
    eprintln!("sorceryd: {error}");
    (
        StatusCode::INTERNAL_SERVER_ERROR,
        "internal server error".into(),
    )
}