Feedback-ID: ie3994620:Fastmail
Date:   Thu, 2 Feb 2023 02:09:43 +0300
From:   "Kirill A. Shutemov" <kirill@shutemov.name>
To:     David Stevens <stevensd@chromium.org>
Cc:     linux-mm@kvack.org, Andrew Morton <akpm@linux-foundation.org>,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] mm/khugepaged: skip shmem with armed userfaultfd
Message-ID: <20230201230943.fg2q6fmvu7gggxar@box.shutemov.name>
References: <20230201034137.2463113-1-stevensd@google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230201034137.2463113-1-stevensd@google.com>
Precedence: bulk

On Wed, Feb 01, 2023 at 12:41:37PM +0900, David Stevens wrote:
> From: David Stevens <stevensd@chromium.org>
> 
> Collapsing memory in a vma that has an armed userfaultfd results in
> zero-filling any missing pages, which breaks user-space paging for those
> filled pages. Avoid khugepage bypassing userfaultfd by not collapsing
> pages in shmem reached via scanning a vma with an armed userfaultfd if
> doing so would zero-fill any pages.

Could you elaborate on the failure? Will zero-filling the page prevent
userfaultfd from catching future access?

A test-case would help a lot.

And what prevents the same pages be filled (with zeros or otherwise) via
write(2) bypassing VMA checks? I cannot immediately see it.

BTW, there's already a check that prevent establishing PMD in the place if
VM_UFFD_WP is set.

Maybe just an update of the check in retract_page_tables() from
userfaultfd_wp() to userfaultfd_armed() would be enough?

I have very limited understanding of userfaultfd(). Sorry in advance for
stupid questions.

-- 
  Kiryl Shutsemau / Kirill A. Shutemov