Matrix Media Repo (MMR) is a highly configurable multi-homeserver media repository for Matrix. If SVG or JPEGXL thumbnailers are enabled (they are disabled by default), a user may upload a file which claims to be either of these types and request a thumbnail to invoke a different decoder in ImageMagick. In some ImageMagick installations, this includes the capability to run Ghostscript to decode the image/file. If MP4 thumbnailers are enabled (also disabled by default), the same issue as above may occur with the ffmpeg installation instead. MMR uses a number of other decoders for all other file types when preparing thumbnails. Theoretical issues are possible with these decoders, however in testing they were not possible to exploit. This is fixed in MMR v1.3.8. MMR now inspects the mimetype of media prior to thumbnailing, and picks a thumbnailer based on those results instead of relying on user-supplied values. This may lead to fewer thumbnails when obscure file shapes are used. This also helps narrow scope of theoretical issues with all decoders MMR uses for thumbnails. Users are advised to upgrade. Users unable to upgrade may disable the SVG, JPEGXL, and MP4 thumbnail types in the MMR config which prevents the decoders from being invoked. Further disabling uncommon file types on the server is recommended to limit risk surface. Containers and other similar technologies may also be used to limit the impact of vulnerabilities in external decoders, like ImageMagick and ffmpeg. Some installations of ImageMagick may disable unsafe file types, like PDFs, already. This option can be replicated to other environments as needed. ffmpeg may be compiled with limited decoders/codecs. The Docker image for MMR disables PDFs and similar formats by default.
The product deserializes untrusted data without sufficiently verifying that the resulting data will be valid.
It is often convenient to serialize objects for communication or to save them for later use. However, deserialized data or code can often be modified without using the provided accessor functions if it does not use cryptography to protect itself. Furthermore, any cryptography would still be client-side security – which is a dangerous security assumption. Data that is untrusted can not be trusted to be well-formed. When developers place no restrictions on “gadget chains,” or series of instances and method invocations that can self-execute during the deserialization process (i.e., before the object is returned to the caller), it is sometimes possible for attackers to leverage them to perform unauthorized actions, like generating a shell.