CVE-2021-21401

### Impact Decoding a specifically formed message can cause invalid `free()` or `realloc()` calls if the message type contains an `oneof` field, and the `oneof` directly contains both a pointer field and a non-pointer field. If the message data first contains the non-pointer field and then the pointer field, the data of the non-pointer field is incorrectly treated as if it was a pointer value. Such message data rarely occurs in normal messages, but it is a concern when untrusted data is parsed. ### Patches Preliminary patch is available on git for [0.4.x](https://github.com/nanopb/nanopb/commit/e2f0ccf939d9f82931d085acb6df8e9a182a4261) and [0.3.x](https://github.com/nanopb/nanopb/commit/4a375a560651a86726e5283be85a9231fd0efe9c) branches. The fix will be released in versions 0.3.9.8 and 0.4.5 once testing has been completed. ### Workarounds Following workarounds are available: * Set the option `no_unions` for the oneof field. This will generate fields as separate instead of C union, and avoids triggering the problematic code. * Set the type of all fields inside the oneof to `FT_POINTER`. This ensures that the data contained inside the `union` is always a valid pointer. * Heap implementations that guard against invalid `free()` provide a partial mitigation. Depending on the message type, the pointer value may be attacker controlled and can be used to bypass heap protections. ### References Bug report: https://github.com/nanopb/nanopb/issues/647 ### For more information If you have any questions or comments about this advisory, comment on the bug report linked above.

Nanopb is a small code-size Protocol Buffers implementation in ansi C. In Nanopb before versions 0.3.9.8 and 0.4.5, decoding a specifically formed message can cause invalid `free()` or `realloc()` calls if the message type contains an `oneof` field, and the `oneof` directly contains both a pointer field and a non-pointer field. If the message data first contains the non-pointer field and then the pointer field, the data of the non-pointer field is incorrectly treated as if it was a pointer value. Such message data rarely occurs in normal messages, but it is a concern when untrusted data is parsed. This has been fixed in versions 0.3.9.8 and 0.4.5. See referenced GitHub Security Advisory for more information including workarounds.