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:   Sun, 26 Mar 2023 10:46:22 -0400
From:   Peter Xu <peterx@redhat.com>
To:     David Hildenbrand <david@redhat.com>
Cc:     Mike Kravetz <mike.kravetz@oracle.com>, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org,
        Andrew Morton <akpm@linux-foundation.org>,
        Mike Rapoport <rppt@linux.vnet.ibm.com>,
        Nadav Amit <nadav.amit@gmail.com>,
        Axel Rasmussen <axelrasmussen@google.com>,
        Andrea Arcangeli <aarcange@redhat.com>,
        Muhammad Usama Anjum <usama.anjum@collabora.com>,
        linux-stable <stable@vger.kernel.org>
Subject: Re: [PATCH v3] mm/hugetlb: Fix uffd wr-protection for CoW
 optimization path
Message-ID: <ZCBavqZE2cyVOzaW@x1n>
References: <20230324142620.2344140-1-peterx@redhat.com>
 <20230324222707.GA3046@monkey>
 <8a06be33-1b44-b992-f80a-8764810ebf3f@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <8a06be33-1b44-b992-f80a-8764810ebf3f@redhat.com>
Precedence: bulk

On Fri, Mar 24, 2023 at 11:36:53PM +0100, David Hildenbrand wrote:
> > > @@ -5487,6 +5487,17 @@ static vm_fault_t hugetlb_wp(struct mm_struct *mm, struct vm_area_struct *vma,
> > >   	unsigned long haddr = address & huge_page_mask(h);
> > >   	struct mmu_notifier_range range;
> > > +	/*
> > > +	 * Never handle CoW for uffd-wp protected pages.  It should be only
> > > +	 * handled when the uffd-wp protection is removed.
> > > +	 *
> > > +	 * Note that only the CoW optimization path (in hugetlb_no_page())
> > > +	 * can trigger this, because hugetlb_fault() will always resolve
> > > +	 * uffd-wp bit first.
> > > +	 */
> > > +	if (!unshare && huge_pte_uffd_wp(pte))
> > > +		return 0;
> > 
> > This looks correct.  However, since the previous version looked correct I must
> > ask.  Can we have unshare set and huge_pte_uffd_wp true?  If so, then it seems
> > we would need to possibly propogate that uffd_wp to the new pte as in v2

Good point, thanks for spotting!

> 
> We can. A reproducer would share an anon hugetlb page because parent and
> child. In the parent, we would uffd-wp that page. We could trigger unsharing
> by R/O-pinning that page.

Right.  This seems to be a separate bug..  It should be triggered in
totally different context and much harder due to rare use of RO pins,
meanwhile used with userfault-wp.

If both of you agree, I can prepare a separate patch for this bug, and I'll
better prepare a reproducer/selftest with it.

-- 
Peter Xu