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;
MIME-Version: 1.0
References: <20230405155120.3608140-1-peterx@redhat.com>
In-Reply-To: <20230405155120.3608140-1-peterx@redhat.com>
From:   Yang Shi <shy828301@gmail.com>
Date:   Wed, 5 Apr 2023 09:59:15 -0700
Message-ID: <CAHbLzkqKE-TE9Od1E=PQDGuhoR+r-TOz4LP8WQgucm_6ZVYTRA@mail.gmail.com>
Subject: Re: [PATCH] mm/khugepaged: Check again on anon uffd-wp during isolation
To:     Peter Xu <peterx@redhat.com>
Cc:     linux-mm@kvack.org, linux-kernel@vger.kernel.org,
        Axel Rasmussen <axelrasmussen@google.com>,
        Nadav Amit <nadav.amit@gmail.com>,
        David Hildenbrand <david@redhat.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Andrea Arcangeli <aarcange@redhat.com>,
        Mike Rapoport <rppt@linux.vnet.ibm.com>,
        linux-stable <stable@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Wed, Apr 5, 2023 at 8:51=E2=80=AFAM Peter Xu <peterx@redhat.com> wrote:
>
> Khugepaged collapse an anonymous thp in two rounds of scans.  The 2nd rou=
nd
> done in __collapse_huge_page_isolate() after hpage_collapse_scan_pmd(),
> during which all the locks will be released temporarily. It means the
> pgtable can change during this phase before 2nd round starts.
>
> It's logically possible some ptes got wr-protected during this phase, and
> we can errornously collapse a thp without noticing some ptes are
> wr-protected by userfault.  e1e267c7928f wanted to avoid it but it only d=
id
> that for the 1st phase, not the 2nd phase.
>
> Since __collapse_huge_page_isolate() happens after a round of small page
> swapins, we don't need to worry on any !present ptes - if it existed
> khugepaged will already bail out.  So we only need to check present ptes
> with uffd-wp bit set there.
>
> This is something I found only but never had a reproducer, I thought it w=
as
> one caused a bug in Muhammad's recent pagemap new ioctl work, but it turn=
s
> out it's not the cause of that but an userspace bug.  However this seems =
to
> still be a real bug even with a very small race window, still worth to ha=
ve
> it fixed and copy stable.

Yeah, I agree. But I got confused by userfaultfd_wp(vma) and
pte_uffd_wp(pte). If a vma is armed with uffd wp, shall we skip the
whole vma? If so, whether it is better to just check vma? We do
revalidate vma once reacquiring mmap_lock, so we should be able to
bail out earlier.

>
> Cc: linux-stable <stable@vger.kernel.org>
> Fixes: e1e267c7928f ("khugepaged: skip collapse if uffd-wp detected")
> Signed-off-by: Peter Xu <peterx@redhat.com>
> ---
>  mm/khugepaged.c | 4 ++++
>  1 file changed, 4 insertions(+)
>
> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> index a19aa140fd52..42ac93b4bd87 100644
> --- a/mm/khugepaged.c
> +++ b/mm/khugepaged.c
> @@ -575,6 +575,10 @@ static int __collapse_huge_page_isolate(struct vm_ar=
ea_struct *vma,
>                         result =3D SCAN_PTE_NON_PRESENT;
>                         goto out;
>                 }
> +               if (pte_uffd_wp(pteval)) {
> +                       result =3D SCAN_PTE_UFFD_WP;
> +                       goto out;
> +               }
>                 page =3D vm_normal_page(vma, address, pteval);
>                 if (unlikely(!page) || unlikely(is_zone_device_page(page)=
)) {
>                         result =3D SCAN_PAGE_NULL;
> --
> 2.39.1
>