FIX Prevents segfault in SVC when internals are altered #21336
Conversation
There was a problem hiding this comment.
There is an issue in building the docs, maybe the following suggestion would fix it. Ortherwise LGTM. Thanks @thomasjpfan !
I think checking before calling the low level functions, as done currently, is better, since this is also an issue without pickling. |
There was a problem hiding this comment.
I think checking before calling the low level functions, as done currently, is better, since this is also an issue without pickling.
Or we could do it at the beginning of the native code function itself.
Anyways, I am fine the current fix if it allows to get rid of the CVE.
I find this CVE hypocrite because of course, if you write a program that mutates the internal datastructures of this same program using private API you will always be able to make it segfault.
You can even do that with the Python standard library (e.g. with the ctypes module for instance) without the need for the compiled code of scikit-learn if you wish...
Co-authored-by: Jérémie du Boisberranger <34657725+jeremiedbb@users.noreply.github.com> Co-authored-by: Olivier Grisel <olivier.grisel@gmail.com>
|
I moved the entry of the changelog in 1.0.1 and tag it for backport |
Reference Issues/PRs
Fixes #18891
What does this implement/fix? Explain your changes.
As stated in #18891 (comment), as a CVE (CVE-2020-28975) it would be good to fix this.
We normally do not perform checks on internal attributes that was set during
fit.Any other comments?
The check can also be performed in
__setstate__.CC @rth