Sandbox escape about asteval HOT 6 CLOSED

@snoopysecurity thanks for the reminder - I just tagged 0.9.23 and pushed to PyPI.

from asteval.

@ross-bradley Thanks! Yes, you are completely correct and this should be addressed.

I think the best approach here is to simply (and always) disallow access to "dunder" attributes entirely but to also keep the other UNSAFE_ATTRS that are not "dunder" methods. A quick check shows that doing that still passes all existing tests. Doing that "no dunder methods" is still sort of a black-list instead of a white-list, but it will adapt better with newly added features.

Does that seem reasonable to you?

from asteval.

That seems to strike an appropriate balance. Whilst yes, it's an extension of the deny-listing approach, it seems to cover the most dangerous attributes and methods in a sufficiently sweeping manner, and does so without requiring an overly onerous effort.

from asteval.

@ross-bradley Thanks - I'll commit that. It turns out that 0.9.22 got released yesterday, but I'll try to push this out as 0.9.23 soon, updating the docs on what is not allowed.

FWIW, I think a whitelist of allowed attributes would be hard to maintain. I guess it does have the advantage of starting with "secure" and moving toward "useful".

from asteval.

@ross-bradley the fix for this is now merged into the master branch.

from asteval.

Hey @newville is it possible to publish a new release with this change? thanks

from asteval.