Security And Vulnerability Management

priority: critical

Security & Vulnerability Management

CVE disclosure · cargo-audit · cargo-deny · Fuzzing · Unsafe code review · Security testing

Vulnerability Management

• cargo-audit: Run on every commit and in CI; fail build on known vulnerabilities

  cargo audit
  cargo audit --deny warnings  # Fail on advisories
  

• cargo-deny: Detect vulnerable, unmaintained, and unlicensed dependencies

  cargo deny check advisories
  cargo deny check bans
  cargo deny check sources
  

• CVE Disclosure Policy:

  • Report to maintainers privately before public disclosure
  • Allow 90 days for patch before public CVE announcement
  • Document CVE in SECURITY.md with patch version and workarounds
  • Never commit exploits; reference by CVE ID and mitigation
  • Include timeline: discovery → disclosure → patch release

Cargo.lock Management

# Cargo.toml
[dependencies]
serde = "=1.0.150"  # Pin critical deps to known safe versions
serde_json = "=1.0.89"

# Run lock file checks in CI
# .github/workflows/security.yml
- name: Check lock file
  run: cargo tree --locked  # Ensure lock matches Cargo.toml

Fuzzing with cargo-fuzz

• Setup: cargo install cargo-fuzz && cargo fuzz init
• Fuzz targets: fuzz/fuzz_targets/ directory; one target per public API surface
• Continuous fuzzing: Run in CI with timeout limits (e.g., 60 seconds per commit)
• Crash reproduction: Save failing inputs; add regression tests

Fuzzing Example

// fuzz/fuzz_targets/parse_fuzzer.rs
#![no_main]
use libfuzzer_sys::fuzz_target;
use my_crate::parse;

fuzz_target!(|data: &[u8]| {
    // Parse should never crash or panic on arbitrary input
    let _ = parse(data);
});

# Run fuzzing
cargo +nightly fuzz run parse_fuzzer -- -max_len=1024 -timeout=5

# Minimize failing input
cargo +nightly fuzz cmin parse_fuzzer

Unsafe Code Review

• SAFETY comments: EVERY unsafe block MUST have a // SAFETY: comment explaining:

  1. What invariant is being relied upon
  2. Why calling code maintains that invariant
  3. What could go wrong if invariant is violated

• Isolation: Unsafe code in dedicated modules; public API must be safe

• Review checklist:

  • All pointers are valid (not null, not dangling)
  • Pointer dereferencing is properly aligned
  • No use-after-free (object still alive)
  • No double-free (only freed once)
  • No data races (proper synchronization)
  • Type safety maintained across FFI boundaries

Unsafe Code Example

/// SAFETY: This function assumes the pointer points to valid, initialized memory
/// for at least `len` bytes. Caller MUST ensure pointer is:
/// - Non-null and properly aligned
/// - Pointing to valid, initialized T values
/// - Valid for reads of at least `len * sizeof(T)` bytes
/// - No aliasing mutable references exist to this data
unsafe fn copy_from_raw(ptr: *const u8, len: usize) -> Vec<u8> {
    std::slice::from_raw_parts(ptr, len).to_vec()
}

// Calling code MUST validate preconditions
let data = unsafe {
    // SAFETY: C function guarantees ptr is valid, len bytes
    copy_from_raw(c_buffer, buffer_len)
};

Security Testing

• No panics on untrusted input: All public functions should return Result, never panic on bad input

• Test adversarial inputs:

  • Empty data
  • Maximum size data
  • Null/invalid pointers (for FFI)
  • Concurrency stress tests (data races)

• Input validation: Validate all untrusted input at boundaries (FFI, network, files)

• Property-based testing: Use proptest for fuzz-like testing in normal tests

Security Test Example

#[cfg(test)]
mod security_tests {
    use proptest::proptest;

    #[test]
    fn parse_never_panics_on_arbitrary_input() {
        // Test with random bytes
        for _ in 0..1000 {
            let random_bytes = rand::random::<Vec<u8>>();
            let result = my_crate::parse(&random_bytes);
            // Should return Err, never panic
            assert!(result.is_ok() || result.is_err());
        }
    }

    #[test]
    fn handles_max_size_input() {
        let huge_input = vec![0u8; 1024 * 1024 * 100];  // 100MB
        let result = my_crate::parse(&huge_input);
        assert!(result.is_ok() || result.is_err());  // Never panic
    }

    proptest! {
        #[test]
        fn parse_property_never_panics(input in ".*") {
            let _ = my_crate::parse(input.as_bytes());
        }
    }
}

Dependency Management

• Minimize dependencies: Each dependency is a potential vulnerability surface
• Evaluate before adding: Check security history, maintenance status, license
• Keep updated: Regular cargo update but within version constraints
• Audit on CI: Every PR should pass cargo audit and cargo deny

Deny Configuration

# deny.toml
[advisories]
vulnerability = "deny"
unmaintained = "warn"
yanked = "warn"
notice = "warn"
ignore = []

[bans]
multiple-versions = "warn"
wildcards = "warn"
allow = []
deny = [
    { name = "openssl", version = "<0.10" },
    { name = "parking_lot", version = "*" }  # Use std instead
]

[sources]
unknown-registry = "warn"
unknown-git = "warn"
allow-registry = ["https://github.com/rust-lang/crates.io-index"]
allow-git = []

Anti-Patterns

• No SAFETY comments: Unsafe code without explanation is unreviewable and unmaintainable
• Unsafe in public API: Safe wrapper should be enforced at public boundary
• Ignoring cargo-audit warnings: Known vulns should block builds immediately
• Panicking on untrusted input: unwrap() on parsed/network data will crash in production
• No fuzzing of parsers: Parsers are attack surface; fuzzing essential
• Mixing unsafe and concurrency: Unsafe + threading = subtle data races; minimize overlap
• Outdated dependencies: Stale deps accumulate vulns; regular updates required
• No bounds checking on FFI pointers: Dereferencing unchecked FFI pointers is UB

Security Checklist for Release

• cargo audit passes (no known vulns)
• cargo deny check passes (licenses, sources)
• All unsafe blocks have SAFETY comments
• Fuzzing targets pass (no crashes)
• cargo test --release passes (including security tests)
• No panics on arbitrary input
• Dependencies are reviewed for security
• SECURITY.md updated with any CVE fixes