Unbounded memory allocation in `evtx` parsing a `.wevtcache` file

Affected versions of evtx did not bound a per-entry payload length read from a .wevtcache file before allocating a buffer sized from it, allowing a tiny crafted file to force a multi-gigabyte allocation and abort the process (out-of-memory denial of service, CWE-400).

Details

In WevtCacheReader::next_entry (src/wevt_templates/wevtcache.rs), the per-entry payload length is read as a u64 straight from the file and used to size a zeroing buffer before any data is read:

let len = usize::try_from(len_u64)?;   // u64 taken directly from the entry header
let mut buf = vec![0u8; len];          // unbounded allocation, sized from the file
reader.read_exact(&mut buf)?;          // only fails AFTER the allocation

The only check on len is that it fits in usize; there is no check that it fits within the bytes remaining in the file. An entry header declaring a ~4 GiB length therefore makes the crate request a ~4 GiB allocation from a file of only ~25 bytes.

The sink is reachable through the public API evtx::wevt_templates::wevtcache::for_each_crim_blob(path, f) (and the WevtCacheReader::next_entry it drives). The .wevtcache reader first shipped in 0.11.0, and every published version from 0.11.0 through 0.14.0 is affected. The code lives in the wevt_templates feature, which is enabled by default through the evtx_dump feature. The fix landed in 0.15.1: 0.15.0 was tagged but never published to crates.io (its Windows build failed), so 0.15.1 is the first patched release available. The Python bindings (evtx on PyPI) carry the fix as of 0.15.0.

Reproduction

A ~25-byte file is sufficient: the .wevtcache magic (WEVTCACH), a version (u32 LE) of 1, an entry_count (u32 LE) of 1, then one entry consisting of a kind byte and a len (u64 LE) set to a large value (e.g. 0xFFFFFFF0), with no payload bytes following. Passing that file to for_each_crim_blob triggers the oversized allocation before the short read fails.