ZstdDecompressor.decompress: excessive allocation (crash) on 13-byte malformed frame (decompressor.c:335); should raise ZstdError

[plotarmordev]

Repo

indygreg/python-zstandard

Summary

ZstdDecompressor().decompress() pre-allocates the full Frame_Content_Size (FCS) from the frame header WITHOUT capping; a 13-byte malformed frame declaring a corrupt FCS makes it call PyBytes_FromStringAndSize for a ~1.1 TB buffer at c-ext/decompressor.c:335, causing an ASan allocation-size-too-big ABORT (in production: a huge MemoryError) — i.e. a crash/DoS on a tiny untrusted decompression input, rather than the documented ZstdError.

Trigger (13 bytes)

hex: 28b52ffde06c06060663636363

Evidence

ASan (binding C-ext built with -fsanitize=address):

ERROR: AddressSanitizer: requested allocation size 0x66c6c6c00000121 exceeds maximum supported size
    #3 PyBytes_FromStringAndSize
    #4 Decompressor_decompress c-ext/decompressor.c:335
SUMMARY: AddressSanitizer: allocation-size-too-big
ABORTING

Expected

The Frame_Content_Size field of a malformed/short frame must not cause an over-allocation; the decompressor should reject such input (raise ZstdError / cap the buffer & grow dynamically), as single-shot malloc of an attacker-controlled FCS is unsafe (CWE-789).

Note

This binding C-ext path is NOT covered by OSS-Fuzz (only upstream zstd core is) and has no advisory.