GHSA-vvp9-7p8x-rfvv lz4_flex's decompression can leak information from uninitialized memory or reused output buffer

• crates.io lz4_flex: < 0.11.6 → fixed in 0.11.6 • crates.io lz4_flex: >= 0.12.0, < 0.12.1 → fixed in 0.12.1

Summary Decompressing invalid LZ4 data can leak data from uninitialized memory, or can leak content from previous decompression operations when reusing an output buffer. Details The LZ4 block format defines a "match copy operation" which duplicates previously written data or data from the user-supplied dict. The position of that data is defined by an _offset_. The data is copied within the output buffer from the _offset_ to the current output position. However, lz4_flex did not properly detect invalid and out-of-bounds _offset_ values properly, causing it to copy uninitialized data from the output buffer. Only the block based API functions are affected: `lz4_flex::block::{decompress_into, decompress_into_with_dict}` When safe-decode is disabled _additionally_ these functions are affected `lz4_flex::block::{decompress, decompress_with_dict, decompress_size_prepended, decompress_size_prepended_with_dict}` All `frame` APIs are _not_ affected. There are two affected use cases: • decompressing LZ4 data with the `unsafe` implementation (`safe-decode` feature flag disabled, which is enabled by default): can leak content of uninitialized memory as decompressed result • decompressing LZ4 data into a reused, user-supplied `output` buffer (affects the `safe-decode` feature as well): can leak the previous contents of the output buffer as decompressed result Impact Leakage of data from uninitialized memory or content from previous decompression operations, possibly revealing sensitive information and secrets. Mitigation lz4_flex 0.12.1 and 0.11.6 fixes this issue without requiring changes in user code. If you cannot upgrade, you can mitigate this vulnerability by zeroing the output buffer before calling `block::decompress_into` or `block::decompress_into_with_dict` (only block based API is affected, frame API is not affected). Additionally the the `safe-decode` feature flag should be enabled.