`codeq`: decoding untrusted input can allocate memory from an unchecked length prefix (denial of service)

codeq is a serialization library whose Decode trait reads length-prefixed values. For String and Vec<u8>, the length-prefixed Decode implementations read a 4-byte big-endian length and allocate a buffer of that size before reading the payload. Because the declared length is taken straight from the input with no upper bound and no cross-check against the bytes actually available, a tiny input can force an allocation of up to ~4 GiB. Code that decodes untrusted data through codeq can therefore be made to spike memory use and be OOM-killed (CWE-400, uncontrolled resource consumption).

The affected path, in src/codec/impls/string_impl.rs:

impl Decode for String {
    fn decode<R: io::Read>(mut r: R) -> Result<Self, io::Error> {
        let len = r.read_u32::<BigEndian>()? as usize;
        let mut buf = vec![0; len];   // allocates `len` bytes up front
        r.read_exact(&mut buf)?; 
        String::from_utf8(buf).map_err(|e| io::Error::new(io::ErrorKind::InvalidData, e))
    }
}

Any code that decodes attacker-controlled bytes through codeq can be forcedcto allocate up to ~4 GiB per decode call from a 4-byte length prefix. On a service decoding untrusted input this is a cheap denial of service (crash / OOM-kill). The practical severity depends on how the embedding wires codeq to untrusted bytes.

Remediation

Fixed in 0.6.3, which grows the decode buffers as payload bytes actually arrive instead of pre-allocating the full declared length (commit 9f15313, "fix: codec: grow decode buffers as payload bytes arrive"). Upgrade to codeq >= 0.6.3.