Re: [PATCH 1/1] mm: restore full accuracy in COW page reuse

[Kirill A. Shutemov]

On Sat, Jan 09, 2021 at 09:54:05PM -0500, Andrea Arcangeli wrote:
> Hello,
> 
> On Sat, Jan 09, 2021 at 07:44:35PM -0500, Andrea Arcangeli wrote:
> > allowing a child to corrupt memory in the parent. That's a problem
> > that could happen not-maliciously too. So the scenario described
> 
> I updated the above partly quoted sentence since in the previous
> version it didn't have full accuracy:
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/andrea/aa.git/commit/?id=fc5a76b1c14e5e6cdc64ece306fc03773662d98a
> 
> "However since a single transient GUP pin on a tail page, would elevate
> the page_count for all other tail pages (unlike the mapcount which is
> subpage granular), the COW page reuse inaccuracy would then cross
> different vmas and the effect would happen at a distance in vma of
> different processes. A single GUP pin taken on a subpage mapped in a
> different process could trigger 511 false positive COWs copies in the
> local process, after a fork()."
> 
> This a best effort to try to document all side effects, but it'd be
> great to hear from Kirill too on the above detail to have
> confirmation.

Yes, this side effect is possible. But I wouldn't worry about too much. If
it routinely happens in a real workloads (I doubt it does), the workload
can tune it with MADV_NOHUGEPAGE/MADV_DONTFORK/MADV_WIPEONFORK or
something.

-- 
 Kirill A. Shutemov