Message-ID: <CACT4Y+ZqToLK5R__x8O1ZctsG3wQtRn36JWF2MPRYqY+Zy_CUA@mail.gmail.com>
Date: Fri, 9 May 2025 10:03:13 +0200
From: Dmitry Vyukov <dvyukov@...gle.com>
To: Greg KH <gregkh@...uxfoundation.org>
Cc: cve@...nel.org, linux-cve-announce@...r.kernel.org, 
	linux-kernel@...r.kernel.org
Subject: Re: REJECTED: CVE-2025-0927: heap overflow in the hfs and hfsplus
 filesystems with manually crafted filesystem

On Fri, 9 May 2025 at 09:55, Greg KH <gregkh@...uxfoundation.org> wrote:
>
> On Fri, May 09, 2025 at 09:47:20AM +0200, Dmitry Vyukov wrote:
> > On Fri, 9 May 2025 at 09:34, Greg KH <gregkh@...uxfoundation.org> wrote:
> > >
> > > On Fri, May 09, 2025 at 09:20:33AM +0200, Dmitry Vyukov wrote:
> > > > > CVE-2025-0927 has now been rejected and is no longer a valid CVE.
> > > >
> > > > > Filesystem bugs due to corrupt images are not considered a CVE for any
> > > > > filesystem that is only mountable by CAP_SYS_ADMIN in the initial user
> > > > > namespace. That includes delegated mounting.
> > > >
> > > > I wonder if this should be the case only if the image is flagged by fsck
> > > > as corrupted? Otherwise I am not sure what's "trusted". It's not about
> > > > somebody's "honest eyes", right. E.g. in the context of insider risks
> > > > the person providing an image may be considered "trusted", or in the
> > > > context of Zero Trust Architecture nothing at all is considered trusted,
> > > > or a trusted image may be tampered with while stored somewhere.
> > > >
> > > > Without any formal means to classify an image as corrupted or not,
> > > > this approach does not look very practical to me. While flagging by fsck
> > > > gives concrete workflow for any context that requires more security.
> > >
> > > And how do we know of fsck can flag anything,
> >
> > By running fsck on the image. Or what do you mean?
>
> That requires us to attempt to reproduce stuff when assigning CVEs?
>
> And what architecture/target?  How do we do this for all of them?
>
> Remember, we are averaging 13 CVE assignments a day, this has to be
> automated in order for us to be able to manage this with the volunteer
> staff we have.

If we can't prove it does not have security impact in any context,
then the safe default would be to say it's unsafe.
That's the current Kernel CNA approach, right. It creates a super set
of any issues that may be relevant in some context.


> > > AND which version of fsck?
> >
> > This needs to be answered as part of establishing the vulnerability
> > triage process. I would go for a relatively fresh version. That will
> > remove bugs fixed a long time ago, and if users rely on it for
> > security purposes they have to update it.
>
> Remember older kernels are updated but userspace isn't on many
> platforms, so the combinations of userspace tools and the kernel
> versions is not anything we are going to even be aware of.
>
> > > We'll defer to the fs developers as to what they want here, but note, we
> > > do not determine "trusted" or not, that is a use case that is outside of
> > > our scope entirely.
> >
> > I think classification should be tied to users and use cases in the
> > first place. I, as a developer, wouldn't want any CVEs assigned to my
> > code, if I could just wish so :)
>
> This is open source, we can not, and do not, dictate use.  It is up to
> the users of our software to determine if their use case matches up with
> the reported vulnerability or not.  We can not do it the other way
> around, that is impossible from our side.

Indeed. And for that we need to flag it as a potential vulnerability
in the first place.

But the current rule says it's not a vulnerability for any possible
user and use case out there.

Powered by blists - more mailing lists