Re: [PATCH 0/2] mm: fix UAF caused by race between ptdump and vmap pgtable freeing

["Lorenzo Stoakes (ARM)" <ljs@kernel.org>]

+cc Kiryl as ref'd below.

On Mon, Jul 13, 2026 at 05:08:19PM +0300, Mike Rapoport wrote:
> On Mon, Jul 13, 2026 at 12:37:46PM +0100, Will Deacon wrote:
> > On Sun, Jul 12, 2026 at 09:46:46AM +0100, Lorenzo Stoakes wrote:
> > > On Sun, Jul 12, 2026 at 12:50:08PM +0530, Dev Jain wrote:
> > > > Will Deacon had pushed back on a similar approach:
> > > > https://lore.kernel.org/all/20250530123527.GA30463@willie-the-truck/
> > > >
> > > > Although now when I read back that thread, it feels more so like my
> > > > incompetency to convince :) because:
> > >
> > > No haha not so, I think more like this stuff is fiddly.
> >
> > Yup, not disputing that this is hard to get right.
> >
> > Conceptually, adding locking purely to deal with a vanishingly rare,
> > debug reader does turn my head but I'm _far_ less concerned about it if
> > it's done in the core code, as is the case here. x86 needs it and we're
> > recently running into related locking issues with the set_memory_*()
> > APIs if we want to collapse the page-table on arm64 [1]. If the overhead
> > is flagged as an issue, we can see if it's worth generalising the static
> > key trick that the second patch reverts but I definitely wouldn't start
> > from that position.

Yeah I think my solution is not exactly pretty but it's the best fix as a hotfix
thing _right now_.

As Kiryl points out the nicer long-term solution would be RCU freeing, but I
think that's the wrong solution for a _fix_ right now, rather something to look
at as a larger structural change in the future.

>
> I'd say it's worth generalizing the set_memory APIs ;-)
>
> Since it's de-facto machinery for manipulation of the kernel page tables it
> makes sense to have a common code for page table walks with hooks to
> architectures for checking/setting/clearing protection bits.
>
> Coincidentally, I'm working on a POC that lifts x86's CPA into mm/ with
> the intention to later use it on other architectures.

Lovely :)

There's another issue with CPA too (planned to raise this separately already but
been super busy/distracted lately :) - it doesn't actually mark its page tables
as kernel page tables so we actually have a gap if an IOMMU happens to map that.

I am planning to send a patch for that anyway but it's indicative of this
needing to be shared code, and lifting stuff up to be shared is good in general
:)

>
> > Will
> >
> > [1] https://lore.kernel.org/linux-arm-kernel/799181c3-a1a1-4de7-bc6a-576d3282efb0@arm.com/
>
> --
> Sincerely yours,
> Mike.

Cheers, Lorenzo