Drop the GitHub token from denylist fetches

core/src/config/mod.rs

@@ -1,11 +1,11 @@
 use crate::policies::registries::{Registry, BUILT_IN};
 use directories::ProjectDirs;
 use serde::Deserialize;
+use std::collections::HashMap;
 use std::io::{Error, ErrorKind};
 use std::path::PathBuf;
 
 pub const DEFAULT_PORT: u16 = 8080;
-const TOKEN_ENV: &str = "FIREWALL_GITHUB_TOKEN";
 
 #[derive(Debug, Deserialize)]
 #[serde(default, deny_unknown_fields)]
@@ -16,30 +16,32 @@ pub struct FirewallConfig {
     /// 0 turns the age check off.
     pub min_age_hours: u64,
     pub denylist: DenylistSource,
+    /// 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>,
 }
 
+#[derive(Debug, Deserialize)]
+#[serde(deny_unknown_fields)]
+pub struct AllowedPackage {
+    pub name: String,
+    /// Every version when absent.
+    pub version: Option<String>,
+}
+
 #[derive(Debug, Clone, Deserialize)]
 #[serde(default, deny_unknown_fields)]
 pub struct DenylistSource {
     /// One file per ecosystem lives under here, named the way OSV names it, in lowercase.
     pub base: String,
-    /// Falls back to `FIREWALL_GITHUB_TOKEN`, so a token need not be written to disk.
-    pub token: Option<String>,
     /// Polled for a new commit, which re-fetches every list in use.
     pub revision: String,
     pub poll_secs: u64,
 }
 
-impl DenylistSource {
-    pub fn token(&self) -> Option<String> {
-        self.token.clone().or_else(|| std::env::var(TOKEN_ENV).ok())
-    }
-}
-
 impl Default for FirewallConfig {
     fn default() -> Self {
         FirewallConfig {
@@ -47,6 +49,7 @@ impl Default for FirewallConfig {
             fail_open: true,
             min_age_hours: 0,
             denylist: DenylistSource::default(),
+            allow: HashMap::new(),
             telemetry: true,
             registries: BUILT_IN.to_vec(),
         }
@@ -57,7 +60,6 @@ impl Default for DenylistSource {
     fn default() -> Self {
         DenylistSource {
             base: "https://raw.githubusercontent.com/usestrix/malicious-packages/packages".into(),
-            token: None,
             revision: "https://api.github.com/repos/usestrix/malicious-packages/commits/packages"
                 .into(),
             poll_secs: 10,

core/src/firewall.rs

@@ -27,7 +27,7 @@ impl Firewall {
             patterns.insert(registry.prefix, compiled.collect());
         }
         let ecosystems = config.registries.iter().map(|r| r.osv_name);
-        let denylist = Denylist::new(config.denylist.clone(), config.denylist.token(), ecosystems);
+        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)));

core/src/policies/filters/denylist.rs

@@ -3,7 +3,7 @@ use crate::config::DenylistSource;
 use crate::policies::registries::Registry;
 use crate::policies::Package;
 use rama::http::client::EasyHttpWebClient;
-use rama::http::header::{AUTHORIZATION, ETAG, IF_NONE_MATCH, USER_AGENT};
+use rama::http::header::{ETAG, IF_NONE_MATCH, USER_AGENT};
 use rama::http::service::client::HttpClientExt;
 use rama::http::{BodyExtractExt, StatusCode};
 use semver::Version;
@@ -37,20 +37,15 @@ type Lists = HashMap<&'static str, OnceCell<RwLock<Arc<FetchedDenylist>>>>;
 
 pub struct Denylist {
     source: DenylistSource,
-    token: Option<String>,
     lists: Arc<Lists>,
     watch: JoinHandle<()>,
 }
 
 impl Denylist {
-    pub fn new(
-        source: DenylistSource,
-        token: Option<String>,
-        ecosystems: impl Iterator<Item = &'static str>,
-    ) -> Self {
+    pub fn new(source: DenylistSource, ecosystems: impl Iterator<Item = &'static str>) -> Self {
         let lists: Arc<Lists> = Arc::new(ecosystems.map(|eco| (eco, OnceCell::new())).collect());
-        let watch = tokio::spawn(watch(source.clone(), token.clone(), Arc::clone(&lists)));
-        Denylist { source, token, lists, watch }
+        let watch = tokio::spawn(watch(source.clone(), Arc::clone(&lists)));
+        Denylist { source, lists, watch }
     }
 }
 
@@ -69,7 +64,7 @@ impl Filter for Denylist {
         Box::pin(async move {
             let eco = registry.osv_name;
             let load = || async {
-                RwLock::new(Arc::new(load(&self.source, self.token.as_deref(), eco).await))
+                RwLock::new(Arc::new(load(&self.source, eco).await))
             };
             let list = self.lists[eco].get_or_init(load).await;
             let list = Arc::clone(&list.read().expect("denylist lock"));
@@ -147,12 +142,12 @@ 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, token: Option<String>, lists: Arc<Lists>) {
+async fn watch(source: DenylistSource, lists: Arc<Lists>) {
     let period = Duration::from_secs(source.poll_secs);
-    let mut seen = revision(&source, token.as_deref(), None).await;
+    let mut seen = revision(&source, None).await;
     loop {
         tokio::time::sleep(period).await;
-        let changed = revision(&source, token.as_deref(), seen.as_deref()).await;
+        let changed = revision(&source, seen.as_deref()).await;
         if let Some(current) = &changed {
             info!(revision = %current, "denylist source changed");
             seen = changed.clone();
@@ -163,7 +158,7 @@ async fn watch(source: DenylistSource, token: Option<String>, lists: Arc<Lists>)
             if changed.is_none() && !failed {
                 continue;
             }
-            if let Ok(packages) = load(&source, token.as_deref(), eco).await {
+            if let Ok(packages) = load(&source, eco).await {
                 *list.write().expect("denylist lock") = Arc::new(Ok(packages));
             }
         }
@@ -171,16 +166,9 @@ async fn watch(source: DenylistSource, token: Option<String>, lists: Arc<Lists>)
 }
 
 /// GitHub's 304 for an unchanged `If-None-Match` costs no rate limit.
-async fn revision(
-    source: &DenylistSource,
-    token: Option<&str>,
-    seen: Option<&str>,
-) -> Option<String> {
+async fn revision(source: &DenylistSource, seen: Option<&str>) -> Option<String> {
     let client = EasyHttpWebClient::default();
     let mut request = client.get(&source.revision).header(USER_AGENT, AGENT);
-    if let Some(token) = token {
-        request = request.header(AUTHORIZATION, format!("Bearer {token}"));
-    }
     if let Some(seen) = seen {
         request = request.header(IF_NONE_MATCH, seen);
     }
@@ -197,8 +185,8 @@ async fn revision(
     response.headers().get(ETAG)?.to_str().ok().map(str::to_owned)
 }
 
-async fn load(source: &DenylistSource, token: Option<&str>, eco: &'static str) -> FetchedDenylist {
-    let list = fetch(source, token, &eco.to_lowercase()).await;
+async fn load(source: &DenylistSource, eco: &'static str) -> FetchedDenylist {
+    let list = fetch(source, &eco.to_lowercase()).await;
     match &list {
         Ok(packages) => info!(%eco, packages = packages.len(), "denylist loaded"),
         Err(error) => warn!(%eco, %error, "denylist unavailable"),
@@ -206,13 +194,10 @@ async fn load(source: &DenylistSource, token: Option<&str>, eco: &'static str) -
     list
 }
 
-async fn fetch(source: &DenylistSource, token: Option<&str>, eco: &str) -> FetchedDenylist {
+async fn fetch(source: &DenylistSource, eco: &str) -> FetchedDenylist {
     let url = format!("{}/{eco}", source.base);
     let client = EasyHttpWebClient::default();
-    let mut request = client.get(&url).header(USER_AGENT, AGENT);
-    if let Some(token) = token {
-        request = request.header(AUTHORIZATION, format!("Bearer {token}"));
-    }
+    let request = client.get(&url).header(USER_AGENT, AGENT);
     let response = request.send().await.map_err(|e| e.to_string())?;
     if response.status() != StatusCode::OK {
         return Err(format!("{url} replied {}", response.status()));

core/src/policies/filters/mod.rs

@@ -30,6 +30,9 @@ pub async fn consult(
     registry: &'static Registry,
     pkg: &Package,
 ) -> Option<(&'static str, Verdict)> {
+    if allowed(firewall, registry, pkg) {
+        return None;
+    }
     let mut unsure = None;
     for filter in firewall.filters() {
         match filter.check(registry, pkg).await {
@@ -41,3 +44,18 @@ pub async fn consult(
     }
     unsure
 }
+
+fn allowed(firewall: &Firewall, registry: &Registry, pkg: &Package) -> bool {
+    let Some(entries) = firewall.config.allow.get(registry.eco) else {
+        return false;
+    };
+    entries.iter().any(|entry| {
+        let name_matches = match registry.normalize {
+            Some(normalize) => normalize(&entry.name) == pkg.name,
+            None => entry.name == pkg.name,
+        };
+        // Go writes a leading v that the parsed version drops, and PEP 440 ignores one.
+        let version = entry.version.as_deref().map(|v| v.strip_prefix('v').unwrap_or(v));
+        name_matches && (version.is_none() || version == pkg.version.as_deref())
+    })
+}

core/tests/config.rs

@@ -1,12 +1,13 @@
 use firewall::config::path;
-use firewall::FirewallConfig;
+use firewall::policies::filters::consult;
+use firewall::policies::Package;
+use firewall::{Firewall, FirewallConfig};
 
 #[test]
 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.denylist.token, FirewallConfig::default().denylist.token);
 }
 
 #[test]
@@ -19,3 +20,14 @@ fn the_config_lives_under_the_config_directory() {
     let path = path().expect("a home directory");
     assert!(path.ends_with("config.json"), "{}", path.display());
 }
+
+#[tokio::test]
+async fn an_allowed_version_skips_the_filters() {
+    // An unreachable denylist would otherwise leave the package unchecked.
+    let config = r#"{"allow": {"go": [{"name": "golang.org/x/net", "version": "v0.20.0"}]}, "denylist": {"base": "http://127.0.0.1:1"}}"#;
+    let firewall = Firewall::new(serde_json::from_str(config).expect("parses"));
+    let (registry, _) = firewall.match_on("/go/golang.org/x/net/@v/list", |r| r.prefix).expect("go");
+    let pkg = |version: &str| Package { name: "golang.org/x/net".into(), version: Some(version.into()) };
+    assert!(consult(&firewall, registry, &pkg("0.20.0")).await.is_none());
+    assert!(consult(&firewall, registry, &pkg("0.21.0")).await.is_some());
+}