Re: [PATCH 0/5] 4.14 backports of fixes for "CoW after fork() issue"

[Suren Baghdasaryan]

On Wed, Apr 21, 2021 at 3:59 PM Vlastimil Babka <vbabka@xxxxxxx> wrote:
>
> On 4/21/21 10:01 PM, Suren Baghdasaryan wrote:
> > On Wed, Apr 7, 2021 at 2:53 PM Suren Baghdasaryan <surenb@xxxxxxxxxx> wrote:
> >>
> >> On Wed, Apr 7, 2021 at 12:23 PM Linus Torvalds
> >> <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> >> >
> >> > On Wed, Apr 7, 2021 at 11:47 AM Mikulas Patocka <mpatocka@xxxxxxxxxx> wrote:
> >> > >
> >> > > So, we fixed it, but we don't know why.
> >> > >
> >> > > Peter Xu's patchset that fixed it is here:
> >> > > https://lore.kernel.org/lkml/20200821234958.7896-1-peterx@xxxxxxxxxx/
> >> >
> >> > Yeah, that's the part that ends up being really painful to backport
> >> > (with all the subsequent fixes too), so the 4.14 people would prefer
> >> > to avoid it.
> >> >
> >> > But I think that if it's a "requires dax pmem and ptrace on top", it
> >> > may simply be a non-issue for those users. Although who knows - maybe
> >> > that ends up being a real issue on Android..
> >>
> >> A lot to digest, so I need to do some reading now. Thanks everyone!
> >
> > After a delay due to vacation I prepared backports of 17839856fd58
> > ("gup: document and work around "COW can break either way" issue") for
> > 4.14 and 4.19 kernels. As Linus pointed out, uffd-wp was introduced
> > later in 5.7, so is not an issue for 4.x kernels. The issue with THPs
> > is still unresolved, so with or without this patch it's still there
> > (Android is not affected by this since we do not use THPs with older
> > kernels).
>
> Which THP issue do you mean here? The race that was part of the same Project
> zero report and was solved by a different patch adding some locking? Or the
> vmsplice info leak but applied to THP's? Because if it's the latter then I
> believe 17839856fd58 did solve that too. It was the later switch of approach to
> rely just on page_count() that left THP side unfixed.

I meant the "vmsplice info leak applied to THP's" but now I realize
that 17839856fd58 does not use elevated reference count, so indeed
that should not be a problem. Thanks for the note!

>
> > Andrea pointed out that there are other issues and to properly fix
> > them his COR approach is needed. However it has not been accepted yet,
> > so I can't really backport it. I'll be happy to do that though if it
> > is accepted in the future.
> >
> > Peter, you mentioned https://lkml.org/lkml/2020/8/10/439 patch to
> > distinguish real writes vs enforced COW read requests, however I also
> > see that you had a later version of this patch here:
> > https://lore.kernel.org/patchwork/patch/1286506/. Which one should I
> > backport? Or is it not needed in the absence of uffd-wp support in the
> > earlier kernels?
> > Thanks,
> > Suren.
> >
> >>
> >> >
> >> >             Linus
> >
>