Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 7 Dec 2022 15:14:22 -0500
From:   Peter Xu <peterx@redhat.com>
To:     David Hildenbrand <david@redhat.com>
Cc:     Hugh Dickins <hughd@google.com>, linux-kernel@vger.kernel.org,
        linux-mm@kvack.org, Ives van Hoorne <ives@codesandbox.io>,
        Andrew Morton <akpm@linux-foundation.org>,
        Alistair Popple <apopple@nvidia.com>,
        Mike Rapoport <rppt@linux.vnet.ibm.com>,
        Nadav Amit <nadav.amit@gmail.com>,
        Andrea Arcangeli <aarcange@redhat.com>
Subject: Re: [PATCH RFC] mm/userfaultfd: enable writenotify while
 userfaultfd-wp is enabled for a VMA
Message-ID: <Y5D0HsBAfQiswoKV@x1n>
References: <Y4oo6cN1a4Yz5prh@x1n>
 <690afe0f-c9a0-9631-b365-d11d98fdf56f@redhat.com>
 <19800718-9cb6-9355-da1c-c7961b01e922@redhat.com>
 <Y45duzmGGUT0+u8t@x1n>
 <92173bad-caa3-6b43-9d1e-9a471fdbc184@redhat.com>
 <22d8e8ac-d75-a66-2650-b4d59f89855e@google.com>
 <Y4+xpCRQCazCymdS@x1n>
 <be1a2244-24a8-480a-a38b-e4ebb9868ce5@redhat.com>
 <Y5DQrHqc7RPYWpIA@x1n>
 <37a9442e-f6e5-35f5-0d51-669d60936b5f@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <37a9442e-f6e5-35f5-0d51-669d60936b5f@redhat.com>
Precedence: bulk

On Wed, Dec 07, 2022 at 08:53:36PM +0100, David Hildenbrand wrote:
> Once user space successfully placed an uffd-wp marker, and e.g., verified
> using pagemap that it is indeed placed, the system should not silently drop
> it.

Note that the anon path doesn't use pte markers.  We won't lose a pte
marker, hopefully, if we do that's a more severe one.

> 
> The behavior between an ordinary THP and a huge zeropage differs. For THP,
> we handle the split correctly and don't lose the marker. Assuming the huge
> zeropage woud be disabled, the behavior would be (IMHO) correct. The test
> case would pass.
> 
> For example, QEMU with uffd-wp based snapshotting will make sure that all
> virtual addresses are populated (e.g., mapping the shared, eventually the
> huge zeropage -- populate_read_range()), before protecting using uffd-wp.
> Losing a uffd-wp marker would be problematic.
> 
> The good news is that we barely will end up PTE-mapping the huge zeropage
> unless there is real user-space interaction (mprotect(), mremap(), mmap()),
> so this shouldn't trigger in the QEMU use-case.

Ah yes, I forgot that part.  If it's not affected then it's better.

> 
> 
> Anyhow, I'll send a patch in a couple of days and we can discuss further.
> It's independent of the other discussion, just wanted to report my findings
> after staring at that code for way too long today.

Thanks, that works for me.

-- 
Peter Xu