Option to attempt reading malformed ISO files

[ookisan]

My employer uses pycdlib as part of an automated toolchain to analyze e-mail attachments. We have started seeing malware distributed in malformed ISO files that pycdlib is capable of reading if certain checks are disabled.

It would be wonderful to have an option to disable these checks when opening an ISO file.

• Checks that version numbers are consistent in the file (e.g. Rock Ridge versions)
• Checks that there is a specific number of things (e.g. El Torito boot catalogs, EFI sections).
• Checks that things are at the correct extents (e.g. 17 for El Torito boot catalog).
• Checks that things match (e.g. anchor points, pvds, big and little endian values).
• Checks for specific values (e.g. UDF volume integrity tag identifier).
• Checks on data structures (e.g. empty UDF file entries for directories, symlink data, duplicate names).

I understand that disabling these checks may result in garbage data, unexpected failures, and all sorts of fun. I understand that it may cause pycdlib to have to make assumptions (e.g. if LE and BE values disagree, assume the LE ones are correct). But sometimes that's OK.

If such an option could be added to pycdlib, I can provide our local patches that disable a number of checks on a per-PyCdlib instance basis. I suspect we may disable checks that can't be disabled and that we've missed some that can. I'm sure things could be done more cleverly and cleanly. I can also provide a few sample ISO files (but not in a public forum since they contain malware).

[clalancette]

Sorry for the delay in answering. Glad to hear that you are using pycdlib.

In general, I've been resistant to adding an option to disable checks and allow more "bad" ISOs through. What it ends up meaning is that users will just always set it to be as relaxed as possible, making it the defacto default mode.

Instead, up until this point my tactic has been to just relax checks as people have run into them. That has seemed to work fairly well until now. My suggestion here is that you open individual pull requests for each of the checks you'd like to disable. We can then evaluate them on a case-by-case basis and see which ones make sense. At the end of that, if we still can't come to an agreement about certain checks, we can consider adding an option. But I consider that a last resort.

Let me know if that makes sense to you.

[clalancette]

No response to this in a year, so I'm going to close this out. I am on board with relaxing the checks that we have as needed, I just need examples of ISOs that violate the standards. When those come up, please do feel free to open issues or pull requests to relax them.