Merge pull request #31 from PredicateSystems/issue_27

Issue 27

Author: rcholic

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "predicate-authorityd"
-version = "0.6.7"
+version = "0.7.0"
 edition = "2021"
 description = "Rust-based sidecar daemon for Predicate Authority"
 license = "MIT"

docs/sidecar-user-manual.md

@@ -1020,8 +1020,9 @@ SSRF protection is **enabled by default**. Blocked requests return:
 
 #### Whitelisting Local Services
 
-To allow specific local endpoints (e.g., local LLM instances, databases), use the `--ssrf-allow` flag with host:port pairs:
+To allow specific local endpoints (e.g., local LLM instances, databases), you have four options:
 
+**Option 1: CLI flag** (highest precedence)
 ```bash
 # Allow local Ollama (WSL2) and Elasticsearch
 ./predicate-authorityd \
@@ -1030,19 +1031,49 @@ To allow specific local endpoints (e.g., local LLM instances, databases), use th
   run
 ```
 
-Or via environment variable (comma-separated):
+**Option 2: Environment variable**
 ```bash
 export PREDICATE_SSRF_ALLOW="172.30.192.1:11434,127.0.0.1:9200"
 ./predicate-authorityd --policy-file policy.json run
 ```
 
-Or in the configuration file:
+**Option 3: TOML configuration file**
 ```toml
 [ssrf]
 allowed_endpoints = ["172.30.192.1:11434", "127.0.0.1:9200"]
 ```
 
-**Important:** The whitelist is host:port specific to limit the exemption surface. Use exact matches only.
+**Option 4: Policy file** (policy-driven, recommended for tenant-scoped deployments)
+
+Add an `ssrf_whitelist` field to your policy JSON/YAML file:
+
+```json
+{
+  "ssrf_whitelist": ["172.30.192.1:11434", "127.0.0.1:9200"],
+  "rules": [
+    ...
+  ]
+}
+```
+
+Or in YAML:
+```yaml
+ssrf_whitelist:
+  - "172.30.192.1:11434"  # Local Ollama on WSL2
+  - "127.0.0.1:9200"      # Local Elasticsearch
+
+rules:
+  - name: allow-llm-calls
+    effect: allow
+    ...
+```
+
+**Precedence and merging:**
+- CLI and environment variables take highest precedence
+- Entries from all sources are merged (deduplicated)
+- If no whitelist is configured anywhere, full SSRF enforcement applies
+
+**Important:** The whitelist uses exact `host:port` matching to limit the exemption surface. Only the specified port is allowed.
 
 #### Disabling SSRF Protection
 

policies/README.md

@@ -289,6 +289,59 @@ Patterns use glob-style matching:
 | `https://*` | Any HTTPS URL |
 | `/home/*/projects/**` | Any file under any user's projects dir |
 
+#### Glob `**` Directory Matching Footgun
+
+**Common mistake:** Using `**` to match a directory itself.
+
+```json
+{
+  "resources": ["model-eval/**"]   // WRONG: matches files INSIDE model-eval, not the directory
+}
+```
+
+The pattern `model-eval/**` matches `model-eval/file.txt` and `model-eval/sub/file.txt`, but it does **NOT** match the directory `model-eval` itself.
+
+**To match both the directory and its contents:**
+
+```json
+{
+  "resources": ["model-eval", "model-eval/**"]   // CORRECT: matches directory AND contents
+}
+```
+
+Or use multiple patterns:
+- `model-eval` - matches the directory itself
+- `model-eval/*` - matches direct children
+- `model-eval/**` - matches all descendants recursively
+
+### SSRF Whitelist (Policy-Driven)
+
+You can include an optional `ssrf_whitelist` field in your policy file to allow specific local endpoints to bypass SSRF protection. This is useful for local LLMs (Ollama), databases, or other services running on private IPs.
+
+```json
+{
+  "ssrf_whitelist": ["172.30.192.1:11434", "127.0.0.1:9200"],
+  "rules": [...]
+}
+```
+
+**Key points:**
+- Whitelist uses exact `host:port` matching for security
+- If CLI `--ssrf-allow` is also provided, entries are merged
+- Defaults to empty (full SSRF enforcement) if omitted
+
+**YAML example:**
+```yaml
+ssrf_whitelist:
+  - "172.30.192.1:11434"  # Local Ollama on WSL2
+  - "127.0.0.1:9200"      # Local Elasticsearch
+
+rules:
+  - name: allow-llm-calls
+    effect: allow
+    # ...
+```
+
 ---
 
 ## Creating Custom Policies

src/main.rs

@@ -484,29 +484,16 @@ async fn main() -> anyhow::Result<()> {
     // Initialize policy engine
     let policy_engine = PolicyEngine::new();
 
-    // Configure SSRF protection
+    // Collect SSRF configuration from CLI and config file
     let ssrf_disabled = cli.ssrf_disabled || file_config.ssrf.disabled;
-    let ssrf_allowed_endpoints: Vec<String> = if !cli.ssrf_allow.is_empty() {
+    let mut ssrf_allowed_endpoints: Vec<String> = if !cli.ssrf_allow.is_empty() {
         cli.ssrf_allow.clone()
     } else {
         file_config.ssrf.allowed_endpoints.clone()
     };
 
-    if ssrf_disabled {
-        policy_engine.set_ssrf_protection(None);
-        warn!("SSRF protection disabled - all endpoints allowed");
-    } else if !ssrf_allowed_endpoints.is_empty() {
-        use predicate_authorityd::ssrf::SsrfProtection;
-        let ssrf = SsrfProtection::new().with_allowed_endpoints(ssrf_allowed_endpoints.clone());
-        policy_engine.set_ssrf_protection(Some(ssrf));
-        info!(
-            "SSRF protection enabled with {} allowed endpoints: {:?}",
-            ssrf_allowed_endpoints.len(),
-            ssrf_allowed_endpoints
-        );
-    }
-
     // Load policy file if specified (supports JSON and YAML formats)
+    // This must happen before SSRF setup to extract ssrf_whitelist from policy
     if let Some(ref policy_path) = policy_file {
         let format = policy_loader::detect_format(policy_path);
         info!(
@@ -526,6 +513,28 @@ async fn main() -> anyhow::Result<()> {
                     info!("Loaded {} policy rules", count);
                 }
 
+                // Merge ssrf_whitelist from policy file (if CLI/config didn't provide any)
+                if !result.ssrf_whitelist.is_empty() {
+                    if ssrf_allowed_endpoints.is_empty() {
+                        ssrf_allowed_endpoints = result.ssrf_whitelist;
+                        info!(
+                            "SSRF whitelist loaded from policy file: {:?}",
+                            ssrf_allowed_endpoints
+                        );
+                    } else {
+                        // CLI/config takes precedence, but we can merge
+                        for endpoint in result.ssrf_whitelist {
+                            if !ssrf_allowed_endpoints.contains(&endpoint) {
+                                ssrf_allowed_endpoints.push(endpoint);
+                            }
+                        }
+                        info!(
+                            "SSRF whitelist merged with policy file entries: {:?}",
+                            ssrf_allowed_endpoints
+                        );
+                    }
+                }
+
                 // Detect audit mode from policy file name
                 let path_lower = policy_path.to_lowercase();
                 if path_lower.contains("audit")
@@ -542,6 +551,21 @@ async fn main() -> anyhow::Result<()> {
         }
     }
 
+    // Configure SSRF protection (after policy loading to include policy-based whitelist)
+    if ssrf_disabled {
+        policy_engine.set_ssrf_protection(None);
+        warn!("SSRF protection disabled - all endpoints allowed");
+    } else if !ssrf_allowed_endpoints.is_empty() {
+        use predicate_authorityd::ssrf::SsrfProtection;
+        let ssrf = SsrfProtection::new().with_whitelist(ssrf_allowed_endpoints.clone());
+        policy_engine.set_ssrf_protection(Some(ssrf));
+        info!(
+            "SSRF protection enabled with {} allowed endpoints: {:?}",
+            ssrf_allowed_endpoints.len(),
+            ssrf_allowed_endpoints
+        );
+    }
+
     // Enable audit mode if explicitly requested via CLI
     if cli.audit_mode {
         policy_engine.set_audit_mode(true);

src/policy_loader.rs

@@ -60,6 +60,9 @@ pub struct PolicyLoadResult {
     pub skipped_rules: usize,
     /// Whether the policy was cryptographically signed
     pub is_signed: bool,
+    /// SSRF whitelist from policy file (optional, host:port format)
+    /// Example: ["172.30.192.1:11434", "127.0.0.1:9200"]
+    pub ssrf_whitelist: Vec<String>,
 }
 
 /// Detect the format of a policy file based on its extension.
@@ -113,11 +116,23 @@ pub fn load_policy_from_string(
 
     let skipped_rules = total_rules - parsed_rules.len();
 
+    // Extract optional ssrf_whitelist array (host:port format)
+    let ssrf_whitelist: Vec<String> = json_value
+        .get("ssrf_whitelist")
+        .and_then(|v| v.as_array())
+        .map(|arr| {
+            arr.iter()
+                .filter_map(|v| v.as_str().map(|s| s.to_string()))
+                .collect()
+        })
+        .unwrap_or_default();
+
     Ok(PolicyLoadResult {
         rules: parsed_rules,
         format,
         skipped_rules,
         is_signed: false,
+        ssrf_whitelist,
     })
 }
 
@@ -425,5 +440,83 @@ mod tests {
 
         assert_eq!(result.rules.len(), 0);
         assert_eq!(result.skipped_rules, 0);
+        assert!(result.ssrf_whitelist.is_empty());
+    }
+
+    // --- SSRF whitelist tests (Issue #27 policy-driven approach) ---
+
+    #[test]
+    fn test_ssrf_whitelist_from_json() {
+        let policy_with_whitelist = r#"{
+            "ssrf_whitelist": ["172.30.192.1:11434", "127.0.0.1:9200"],
+            "rules": [
+                {
+                    "name": "allow-all",
+                    "effect": "allow",
+                    "principals": ["*"],
+                    "actions": ["*"],
+                    "resources": ["*"]
+                }
+            ]
+        }"#;
+
+        let result = load_policy_from_string(policy_with_whitelist, PolicyFormat::Json).unwrap();
+
+        assert_eq!(result.rules.len(), 1);
+        assert_eq!(result.ssrf_whitelist.len(), 2);
+        assert!(result
+            .ssrf_whitelist
+            .contains(&"172.30.192.1:11434".to_string()));
+        assert!(result
+            .ssrf_whitelist
+            .contains(&"127.0.0.1:9200".to_string()));
+    }
+
+    #[test]
+    fn test_ssrf_whitelist_from_yaml() {
+        let yaml_with_whitelist = r#"
+ssrf_whitelist:
+  - "172.30.192.1:11434"
+  - "localhost:8787"
+rules:
+  - name: allow-all
+    effect: allow
+    principals:
+      - "*"
+    actions:
+      - "*"
+    resources:
+      - "*"
+"#;
+
+        let result = load_policy_from_string(yaml_with_whitelist, PolicyFormat::Yaml).unwrap();
+
+        assert_eq!(result.rules.len(), 1);
+        assert_eq!(result.ssrf_whitelist.len(), 2);
+        assert!(result
+            .ssrf_whitelist
+            .contains(&"172.30.192.1:11434".to_string()));
+        assert!(result
+            .ssrf_whitelist
+            .contains(&"localhost:8787".to_string()));
+    }
+
+    #[test]
+    fn test_ssrf_whitelist_missing_defaults_to_empty() {
+        // Policy without ssrf_whitelist field should have empty whitelist
+        let result = load_policy_from_string(SAMPLE_JSON_POLICY, PolicyFormat::Json).unwrap();
+        assert!(result.ssrf_whitelist.is_empty());
+    }
+
+    #[test]
+    fn test_ssrf_whitelist_empty_array() {
+        let policy_with_empty_whitelist = r#"{
+            "ssrf_whitelist": [],
+            "rules": []
+        }"#;
+
+        let result =
+            load_policy_from_string(policy_with_empty_whitelist, PolicyFormat::Json).unwrap();
+        assert!(result.ssrf_whitelist.is_empty());
     }
 }

src/ssrf.rs

@@ -78,6 +78,12 @@ impl SsrfProtection {
         self
     }
 
+    /// Alias for `with_allowed_endpoints` - adds whitelist entries that bypass SSRF checks
+    /// This is the method name used when loading from policy files
+    pub fn with_whitelist(self, whitelist: Vec<String>) -> Self {
+        self.with_allowed_endpoints(whitelist)
+    }
+
     /// Add a single allowed endpoint
     pub fn add_allowed_endpoint(&mut self, endpoint: &str) {
         self.allowed_endpoints.push(endpoint.to_lowercase());
@@ -593,4 +599,20 @@ mod tests {
         assert!(ssrf.block_cloud_metadata);
         assert!(ssrf.block_internal_dns);
     }
+
+    #[test]
+    fn test_with_whitelist_alias() {
+        // Test that with_whitelist() is an alias for with_allowed_endpoints()
+        let ssrf = SsrfProtection::new().with_whitelist(vec!["172.30.192.1:11434".to_string()]);
+
+        // Private IP would normally be blocked
+        assert!(SsrfProtection::new()
+            .check_resource("http://172.30.192.1:11434/api/generate")
+            .is_some());
+
+        // But with_whitelist should allow it
+        assert!(ssrf
+            .check_resource("http://172.30.192.1:11434/api/generate")
+            .is_none());
+    }
 }