Re: upstream kernel crashes

From: "Michael S. Tsirkin" <mst@redhat.com>
To: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Cc: axboe@kernel.dk, martin.petersen@oracle.com,
	gregkh@linuxfoundation.org, linux-kernel@vger.kernel.org,
	kasan-dev@googlegroups.com,
	virtualization@lists.linux-foundation.org,
	James.Bottomley@hansenpartnership.com,
	torvalds@linux-foundation.org, edumazet@google.com,
	linux@roeck-us.net, netdev@vger.kernel.org, c@redhat.com,
	kuba@kernel.org, pabeni@redhat.com, andres@anarazel.de,
	davem@davemloft.net, Dmitry Vyukov <dvyukov@google.com>
Subject: Re: upstream kernel crashes
Date: Wed, 17 Aug 2022 06:53:53 -0400	[thread overview]
Message-ID: <20220817065207-mutt-send-email-mst@kernel.org> (raw)
In-Reply-To: <1660718191.3631961-1-xuanzhuo@linux.alibaba.com>

On Wed, Aug 17, 2022 at 02:36:31PM +0800, Xuan Zhuo wrote:
> On Wed, 17 Aug 2022 08:13:59 +0200, Dmitry Vyukov <dvyukov@google.com> wrote:
> > On Mon, 15 Aug 2022 17:32:06 -0400, Michael wrote:
> > > So if you pass the size parameter for a legacy device it will
> > > try to make the ring smaller and that is not legal with
> > > legacy at all. But the driver treats legacy and modern
> > > the same, it allocates a smaller queue anyway.
> > >
> > > Lo and behold, I pass disable-modern=on to qemu and it happily
> > > corrupts memory exactly the same as GCP does.
> >
> > Ouch!
> >
> > I understand that the host does the actual corruption,
> > but could you think of any additional debug checking in the guest
> > that would caught this in future? Potentially only when KASAN
> > is enabled which can verify validity of memory ranges.
> > Some kind of additional layer of sanity checking.
> >
> > This caused a bit of a havoc for syzbot with almost 100 unique
> > crash signatures, so would be useful to catch such issues more
> > reliably in future.
> 
> We can add a check to vring size before calling vp_legacy_set_queue_address().
> Checking the memory range directly is a bit cumbersome.
> 
> Thanks.

With a comment along the lines of

/* Legacy virtio pci has no way to communicate a change in vq size to
 * the hypervisor. If ring sizes don't match hypervisor will happily
 * corrupt memory.
 */

> diff --git a/drivers/virtio/virtio_pci_legacy.c b/drivers/virtio/virtio_pci_legacy.c
> index 2257f1b3d8ae..0673831f45b6 100644
> --- a/drivers/virtio/virtio_pci_legacy.c
> +++ b/drivers/virtio/virtio_pci_legacy.c
> @@ -146,6 +146,8 @@ static struct virtqueue *setup_vq(struct virtio_pci_device *vp_dev,
>                 goto out_del_vq;
>         }
> 
> +       BUG_ON(num != virtqueue_get_vring_size(vq));
> +
>         /* activate the queue */
>         vp_legacy_set_queue_address(&vp_dev->ldev, index, q_pfn);
> 
> 
> >
> > Thanks

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization