Repository navigation
What's Changed
Memory safety and security improvements: malformed object detection
This release introduces an important memory safety and security improvement reported to us via AI testing by an MIT security researcher (Thank you @arpitjain099). There is a new Silo library read error code, E_MALFORMED to indicate an object read is corrupted in some way that would lead to memory faults. A lot of validation logic (though none of it involves problem-sized work) was added to the read routines in both the PDB and HDF5 drivers. This feature has been thoroughly tested across files written by previous versions of Silo. However, there is always a chance that something was missed and the new malformed object validation logic prevents reading an object from Silo files that it otherwise should be able to read. There is also a chance that not all memory corrupting issues have been caught in this update. Users encountering issues are encouraged to reach out to the project here on GitHub. For more details read the section of the manual about malformed objects.
Compression fixes and improvements
Support for a new compression METHOD=HDF5_PLUGIN was added. This new method allows Silo applications to use any compression filter available as an actual HDF5 plugin instead of only compression filters directly embedded into the Silo library. See details in the description of DBSetCompression().
Silo now supports compression strategies. An application can specifiy a set of compression strings the library should try to apply for a given write of an object's problem-sized data. Internally, Silo maintains a simple taxonomy of classes of compression methods and which methods are suitable for which kinds of data objects Silo is writing at any given moment. It then loops over the compression strings specified by the application to decide which it can apply for a given write. Read more about this feature in the description of DBSetCompression().
Some issues with HZIP compression were fixed.
Other changes
- A new public API global setting,
DBSetSmallArrayByteSize()was added. - Several object validation methods were added to the public API (
DBValidateZonelist(),DBValidateMaterial(),DBValidateFacelist(). More are coming in future releases. - A problem with the definition of the
char *type in PDB's structure chart was corrected. Older Silo files, however, still have the type incorrectly defined. There is at least one known instance of a PDB-based application that ran into problems when reading older files with this incorrectly defined type. - Fix some leaks and possible memory errors in Silo's browser tool.
- Fix handling of
datatypein generic objects and very, very old Silo files in python module (resolves #567) - Improve CMake finding and linking with HDF5 (Thank you @BradWhitlock)
- Fix inconsistency in strings from
DBGetObjtypeName()andGetObjtypeTage()(Resolves #208) - Resolves issue with funky line endings in some source code files (#520)
- Fixes an issue with
scale_quadvar.pytest when HDF5 is compiled without deflate (zlib) filter. - Improves python module handling of silo library errors
Full Changelog: 4.12.1...4.12.2