Reject config values the firewall cannot use

core/src/config/mod.rs

@@ -1,9 +1,11 @@
-use crate::policies::registries::{Registry, BUILT_IN};
+use crate::policies::registries::{host_of, BUILT_IN};
 use directories::ProjectDirs;
+use rama::net::uri::Uri;
 use serde::Deserialize;
 use serde_json::{Map, Value};
 use std::collections::HashMap;
 use std::io::{Error, ErrorKind};
+use std::num::{NonZeroU16, NonZeroU64};
 use std::path::PathBuf;
 
 pub const DEFAULT_PORT: u16 = 8080;
@@ -11,7 +13,7 @@ pub const DEFAULT_PORT: u16 = 8080;
 #[derive(Debug, Deserialize)]
 #[serde(default, deny_unknown_fields)]
 pub struct FirewallConfig {
-    pub port: u16,
+    pub port: NonZeroU16,
     /// Allow a package when a filter could not check it.
     pub fail_open: bool,
     /// 0 turns the age check off.
@@ -20,9 +22,8 @@ pub struct FirewallConfig {
     /// Packages never screened, by ecosystem: `{"npm": [{"name": "left-pad"}]}`.
     pub allow: HashMap<String, Vec<AllowedPackage>>,
     pub telemetry: bool,
-    /// Not describable in the file yet, so this is always the built-in table.
-    #[serde(skip)]
-    pub registries: Vec<&'static Registry>,
+    /// Mirrors laid out like a built-in registry, by its name: `{"npm": ["https://npm.example.com/"]}`.
+    pub registries: HashMap<String, Vec<String>>,
 }
 
 #[derive(Debug, Deserialize)]
@@ -40,19 +41,19 @@ pub struct DenylistSource {
     pub base: String,
     /// Polled for a new commit, which re-fetches every list in use.
     pub revision: String,
-    pub poll_secs: u64,
+    pub poll_secs: NonZeroU64,
 }
 
 impl Default for FirewallConfig {
     fn default() -> Self {
         FirewallConfig {
-            port: DEFAULT_PORT,
+            port: NonZeroU16::new(DEFAULT_PORT).unwrap(),
             fail_open: true,
             min_age_hours: 0,
             denylist: DenylistSource::default(),
             allow: HashMap::new(),
             telemetry: true,
-            registries: BUILT_IN.to_vec(),
+            registries: HashMap::new(),
         }
     }
 }
@@ -63,7 +64,7 @@ impl Default for DenylistSource {
             base: "https://raw.githubusercontent.com/usestrix/firewall-blocklist/main".into(),
             revision: "https://api.github.com/repos/usestrix/firewall-blocklist/commits/main"
                 .into(),
-            poll_secs: 10,
+            poll_secs: NonZeroU64::new(10).unwrap(),
         }
     }
 }
@@ -87,7 +88,8 @@ impl FirewallConfig {
             };
             let invalid = |e| Error::new(ErrorKind::InvalidData, format!("{}: {e}", path.display()));
             let layer: Value = serde_json::from_str(&text).map_err(invalid)?;
-            FirewallConfig::deserialize(&layer).map_err(invalid)?;
+            let config = FirewallConfig::deserialize(&layer).map_err(invalid)?;
+            check(&config).map_err(|e| Error::new(ErrorKind::InvalidData, format!("{}: {e}", path.display())))?;
             let Value::Object(layer) = layer else {
                 unreachable!("a config parses only from an object");
             };
@@ -97,19 +99,7 @@ impl FirewallConfig {
     }
 
     pub fn local_url(&self, path: &str) -> String {
-        crate::local_url(self.port, path)
-    }
-
-    /// Without building a `Firewall`, which needs a runtime.
-    pub fn hosts(&self) -> Vec<&'static str> {
-        let mut hosts: Vec<&'static str> = Vec::new();
-        for registry in &self.registries {
-            let host = registry.host();
-            if !hosts.contains(&host) {
-                hosts.push(host);
-            }
-        }
-        hosts
+        crate::local_url(self.port.get(), path)
     }
 }
 
@@ -129,3 +119,43 @@ fn merge(base: &mut Map<String, Value>, top: Map<String, Value>) {
         }
     }
 }
+
+/// What serde's types cannot say.
+fn check(config: &FirewallConfig) -> Result<(), String> {
+    // `Duration::hours` panics past this.
+    if config.min_age_hours > time::Duration::MAX.whole_hours() as u64 {
+        return Err(format!("{} hours is too long", config.min_age_hours));
+    }
+    for eco in config.allow.keys() {
+        if !BUILT_IN.iter().any(|r| r.eco == eco) {
+            return Err(format!("unknown ecosystem {eco:?} in allow"));
+        }
+    }
+    for url in [&config.denylist.base, &config.denylist.revision] {
+        if parse_url(url, &["http", "https"]).is_none() {
+            return Err(format!("{url:?} is not an http(s) URL"));
+        }
+    }
+    for (name, mirrors) in &config.registries {
+        if !BUILT_IN.iter().any(|r| r.name() == name) {
+            return Err(format!("no built-in registry is named {name}"));
+        }
+        if let Some(url) = mirrors.iter().find(|url| !is_mirror(url)) {
+            return Err(format!("{url}: a mirror is an https:// URL ending in /"));
+        }
+    }
+    Ok(())
+}
+
+/// No port or userinfo, as `host_of` is what gets intercepted.
+fn is_mirror(url: &str) -> bool {
+    let Some(uri) = parse_url(url, &["https"]) else { return false };
+    let bare_host = uri.host_str().is_some_and(|host| host == host_of(url));
+    bare_host && uri.query().is_none() && uri.fragment().is_none() && url.ends_with('/')
+}
+
+fn parse_url(url: &str, schemes: &[&str]) -> Option<Uri> {
+    let uri = Uri::parse(url).ok()?;
+    let has_host = uri.host_str().is_some_and(|host| !host.is_empty());
+    (has_host && schemes.contains(&uri.scheme_str()?)).then_some(uri)
+}

core/src/firewall.rs

@@ -9,6 +9,7 @@ use std::collections::HashMap;
 
 pub struct Firewall {
     pub config: FirewallConfig,
+    registries: Vec<&'static Registry>,
     patterns: HashMap<&'static str, Vec<Regex>>,
     filters: Vec<Box<dyn Filter>>,
 }
@@ -21,18 +22,19 @@ impl Default for Firewall {
 
 impl Firewall {
     pub fn new(config: FirewallConfig) -> Self {
+        let registries = with_mirrors(&config.registries);
         let mut patterns = HashMap::new();
-        for registry in &config.registries {
+        for registry in &registries {
             let compiled = registry.patterns.iter().map(|p| Regex::new(p.re).expect(p.re));
             patterns.insert(registry.prefix, compiled.collect());
         }
-        let ecosystems = config.registries.iter().map(|r| r.osv_name);
+        let ecosystems = registries.iter().map(|r| r.osv_name);
         let denylist = Denylist::new(config.denylist.clone(), ecosystems);
         let mut filters: Vec<Box<dyn Filter>> = vec![Box::new(denylist), Box::new(Osv)];
         if config.min_age_hours > 0 {
             filters.push(Box::new(Age::new(config.min_age_hours)));
         }
-        Firewall { config, patterns, filters }
+        Firewall { config, registries, patterns, filters }
     }
 
     pub fn match_on<'a>(
@@ -40,20 +42,27 @@ impl Firewall {
         url: &'a str,
         key: fn(&Registry) -> &'static str,
     ) -> Option<(&'static Registry, &'a str)> {
-        self.config.registries.iter().find_map(|r| url.strip_prefix(key(r)).map(|rest| (*r, rest)))
+        self.registries.iter().find_map(|r| url.strip_prefix(key(r)).map(|rest| (*r, rest)))
     }
 
     pub fn hosts(&self) -> Vec<&'static str> {
-        self.config.hosts()
+        let mut hosts: Vec<&'static str> = Vec::new();
+        for registry in &self.registries {
+            let host = registry.host();
+            if !hosts.contains(&host) {
+                hosts.push(host);
+            }
+        }
+        hosts
     }
 
     pub fn is_https_url(&self, url: &str) -> bool {
         let host = host_of(url);
-        self.config.registries.iter().any(|r| r.upstream.starts_with("https://") && r.host() == host)
+        self.registries.iter().any(|r| r.upstream.starts_with("https://") && r.host() == host)
     }
 
     pub fn is_index_url(&self, url: &str) -> bool {
-        self.config.registries.iter().any(|r| r.index && r.host() == host_of(url))
+        self.registries.iter().any(|r| r.index && r.host() == host_of(url))
     }
 
     pub fn resolve(&self, registry: &'static Registry, rest: &str) -> Resolved {
@@ -64,3 +73,15 @@ impl Firewall {
         &self.filters
     }
 }
+
+/// The built-ins, then a copy of one per mirror. Leaked, as a `Registry` is `'static` everywhere.
+fn with_mirrors(mirrors: &HashMap<String, Vec<String>>) -> Vec<&'static Registry> {
+    let mut registries = BUILT_IN.to_vec();
+    for built_in in BUILT_IN {
+        for upstream in mirrors.get(built_in.name()).into_iter().flatten() {
+            let upstream = upstream.clone().leak();
+            registries.push(Box::leak(Box::new(Registry { upstream, ..(*built_in).clone() })));
+        }
+    }
+    registries
+}

core/src/policies/filters/denylist.rs

@@ -143,7 +143,7 @@ pub fn describe(advisories: &[&Advisory]) -> String {
 
 /// A failed fetch leaves the last good list in place; a list that never loaded is retried every tick.
 async fn watch(source: DenylistSource, lists: Arc<Lists>) {
-    let period = Duration::from_secs(source.poll_secs);
+    let period = Duration::from_secs(source.poll_secs.get());
     let mut seen = revision(&source, None).await;
     loop {
         tokio::time::sleep(period).await;

core/src/policies/registries/mod.rs

@@ -36,7 +36,7 @@ pub struct Pattern {
 }
 
 /// Requests to the local server match on `prefix`; intercepted URLs match on `upstream`.
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 pub struct Registry {
     pub prefix: &'static str,
     pub eco: &'static str,
@@ -52,7 +52,7 @@ pub struct Registry {
 }
 
 /// A URL to fetch and a JSON pointer to the timestamp in its reply.
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 pub struct Published {
     pub url: &'static str,
     pub pointer: &'static str,
@@ -63,6 +63,11 @@ impl Registry {
     pub fn host(&self) -> &'static str {
         host_of(self.upstream)
     }
+
+    /// What the config file calls it.
+    pub fn name(&self) -> &'static str {
+        self.prefix.trim_matches('/')
+    }
 }
 
 /// The host of an https:// URL, with the path dropped.

core/src/proxy/mod.rs

@@ -15,7 +15,7 @@ use std::sync::Arc;
 use tracing::{info, info_span, warn, Instrument};
 
 pub fn serve(config: FirewallConfig) -> std::io::Result<()> {
-    let bound = std::net::TcpListener::bind(("127.0.0.1", config.port))?;
+    let bound = std::net::TcpListener::bind(("127.0.0.1", config.port.get()))?;
     bound.set_nonblocking(true)?;
     crate::runtime()?.block_on(async move {
         let firewall = Arc::new(Firewall::new(config));

core/tests/config.rs

@@ -7,7 +7,7 @@ use firewall::{Firewall, FirewallConfig};
 fn a_file_only_has_to_set_what_it_changes() {
     let config: FirewallConfig = serde_json::from_str(r#"{"fail_open": false}"#).expect("parses");
     assert!(!config.fail_open);
-    assert_eq!(config.port, firewall::DEFAULT_PORT);
+    assert_eq!(config.port.get(), firewall::DEFAULT_PORT);
 }
 
 #[test]
@@ -15,6 +15,28 @@ fn a_misspelled_setting_is_an_error() {
     assert!(serde_json::from_str::<FirewallConfig>(r#"{"fail-open": false}"#).is_err());
 }
 
+#[test]
+fn nonsense_values_are_errors() {
+    let file = std::env::temp_dir().join(format!("firewall-nonsense-{}.json", std::process::id()));
+    for bad in [
+        r#"{"allow": {"nmp": [{"name": "left-pad"}]}}"#,
+        r#"{"denylist": {"poll_secs": 0}}"#,
+        r#"{"denylist": {"base": "raw.githubusercontent.com/usestrix"}}"#,
+        r#"{"denylist": {"revision": "https://"}}"#,
+        r#"{"port": 0}"#,
+        r#"{"min_age_hours": 18446744073709551615}"#,
+        r#"{"registries": {"npm": ["http://npm.example.com/"]}}"#,
+        r#"{"registries": {"npm": ["https://npm.example.com"]}}"#,
+        r#"{"registries": {"npm": ["https://npm.example.com:8443/"]}}"#,
+        r#"{"registries": {"npm": ["https://[email protected]/"]}}"#,
+        r#"{"registries": {"npm": ["https://npm.example.com/?a=/"]}}"#,
+    ] {
+        std::fs::write(&file, bad).expect("write");
+        assert!(FirewallConfig::load_from(std::slice::from_ref(&file)).is_err(), "{bad}");
+    }
+    std::fs::remove_file(&file).ok();
+}
+
 #[test]
 fn the_config_lives_under_the_config_directory() {
     let path = path().expect("a home directory");
@@ -38,6 +60,21 @@ fn a_later_file_wins_and_allow_lists_add_up() {
     assert_eq!(names, ["a", "b"]);
 }
 
+#[tokio::test]
+async fn a_mirror_is_screened_like_the_registry_it_copies() {
+    let config = r#"{"registries": {"npm": ["https://npm.example.com/"]}}"#;
+    let firewall = Firewall::new(serde_json::from_str(config).expect("parses"));
+    assert!(firewall.hosts().contains(&"npm.example.com"));
+    let url = "https://npm.example.com/left-pad/-/left-pad-1.3.0.tgz";
+    let (registry, rest) = firewall.match_on(url, |r| r.upstream).expect("the mirror");
+    assert_eq!(firewall.resolve(registry, rest).pkg.map(|p| p.label()), Some("[email protected]".into()));
+    let unknown = std::env::temp_dir().join(format!("firewall-mirror-{}.json", std::process::id()));
+    std::fs::write(&unknown, r#"{"registries": {"nmp": ["https://npm.example.com/"]}}"#).expect("write");
+    let loaded = FirewallConfig::load_from(std::slice::from_ref(&unknown));
+    std::fs::remove_file(&unknown).ok();
+    assert!(loaded.is_err());
+}
+
 #[tokio::test]
 async fn an_allowed_version_skips_the_filters() {
     // An unreachable denylist would otherwise leave the package unchecked.

docs/platforms.md

@@ -25,7 +25,10 @@ npm, pnpm, bun, pip, pip3, pipenv, pdm, uv, cargo, go and gem, downloading from:
 | RubyGems | rubygems.org |
 | Go | proxy.golang.org |
 
-Private registries and mirrors are not screened.
+Machine-wide screening also covers mirrors listed in the config file, under the registry they
+copy (`npm`, `pypi`, `pypi-files`, `cargo/index`, `cargo-dl`, `crates-api`, `gem` or `go`):
+`"registries": {"npm": ["https://npm.example.com/"]}`. A mirror must lay out its paths like that
+registry. `firewall <manager>` still sends managers to the public registries.
 
 ## How machine-wide screening works