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:   Thu, 27 Apr 2023 13:46:49 +0200
From:   Claudio Imbrenda <imbrenda@linux.ibm.com>
To:     Heiko Carstens <hca@linux.ibm.com>
Cc:     kvm@vger.kernel.org, linux-kernel@vger.kernel.org,
        linux-s390@vger.kernel.org, frankja@linux.ibm.com,
        borntraeger@de.ibm.com, nrb@linux.ibm.com, nsg@linux.ibm.com,
        seiden@linux.ibm.com, jgg@nvidia.com
Subject: Re: [PATCH v2 1/1] KVM: s390: fix race in gmap_make_secure
Message-ID: <20230427134649.1e482d91@p-imbrenda>
In-Reply-To: <ZEpUEF7H86E9vVfS@osiris>
References: <20230426134834.35199-1-imbrenda@linux.ibm.com>
        <20230426134834.35199-2-imbrenda@linux.ibm.com>
        <ZEpUEF7H86E9vVfS@osiris>
Organization: IBM
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Thu, 27 Apr 2023 12:53:04 +0200
Heiko Carstens <hca@linux.ibm.com> wrote:

> On Wed, Apr 26, 2023 at 03:48:34PM +0200, Claudio Imbrenda wrote:
> > This patch fixes a potential race in gmap_make_secure and removes the
> > last user of follow_page without FOLL_GET.
> > 
> > Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
> > Suggested-by: Jason Gunthorpe <jgg@nvidia.com>
> > Fixes: 214d9bbcd3a6 ("s390/mm: provide memory management functions for protected KVM guests")
> > ---
> >  arch/s390/kernel/uv.c | 32 +++++++++++---------------------
> >  1 file changed, 11 insertions(+), 21 deletions(-)  
> 
> It would be helpful if this would be a bit more descriptive. "Fix
> race" is not very helpful :)
> 
> What race does this fix?
> When can this happen?
> What are the consequences if the race window is being hit?

We are locking something we don't have a reference to, and as explained
by Jason and David in this thread <Y9J4P/RNvY1Ztn0Q@nvidia.com> it can
lead to all kind of bad things, including the page getting
unmapped (MADV_DONTNEED), freed, reallocated as a larger folio and the
unlock_page() would target the wrong bit.

Also there is another race with the FOLL_WRITE, which could race
between the follow_page and the get_locked_pte.

The main point of the patch is to remove the last follow_page without
FOLL_GET or FOLL_PIN, removing the races can be considered a nice bonus.