On Fri, 23 Sep 2016 13:54:17 +0100 Stefan Hajnoczi <stefa...@redhat.com> wrote:
> On Wed, Sep 21, 2016 at 06:57:12PM +0200, Greg Kurz wrote: > > A broken guest may send a request with only non-empty out buffers > > or only non-empty in buffers, virtqueue_pop() will then return a > > VirtQueueElement with out_num == 0 or in_num == 0 respectively. > > > > All 9P requests are expected to start with the following 7-byte header: > > > > uint32_t size_le; > > uint8_t id; > > uint16_t tag_le; > > > > If iov_to_buf() fails to return these 7 bytes, then something is wrong in > > the guest. > > > > In both cases, it is wrong to crash QEMU, since the root cause lies in the > > guest. Let's switch the device to the broken state instead. > > > > Signed-off-by: Greg Kurz <gr...@kaod.org> > > --- > > v2: - added out_free_pdu: label for errors or when virtqueue is empty > > --- > > hw/9pfs/virtio-9p-device.c | 20 ++++++++++++++++---- > > 1 file changed, 16 insertions(+), 4 deletions(-) > > > > diff --git a/hw/9pfs/virtio-9p-device.c b/hw/9pfs/virtio-9p-device.c > > index e7ea0e45f3dd..5f3a67cfc717 100644 > > --- a/hw/9pfs/virtio-9p-device.c > > +++ b/hw/9pfs/virtio-9p-device.c > > @@ -52,17 +52,24 @@ static void handle_9p_output(VirtIODevice *vdev, > > VirtQueue *vq) > > > > elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); > > if (!elem) { > > - pdu_free(pdu); > > - break; > > + goto out_free_pdu; > > } > > > > - BUG_ON(elem->out_num == 0 || elem->in_num == 0); > > + if (elem->out_num == 0 || elem->in_num == 0) { > > + virtio_error(vdev, > > + "The guest sent a VirtFS request without > > headers"); > > + goto out_free_pdu; > > + } > > I'm not sure this check is even necessary since we check > iov_to_buf(elem->out_sg[]) and pdu_marshall() avoids overflowing > elem->in_sg[]. > I agree that the difference between absent and empty out headers is quite subtile indeed. It would probably be acceptable to rely on iov_to_buf(), the same way virtio-net relies on iov_size() I guess. The case of elem->in_sg[] seems different though: unless I missed something, v9fs_packunpack() will simply return -ENOBUFS. The error will be propagated to pdu_complete(), which does not even check for errors (see the comment). And then we'll push a broken element to the guest... I'd rather raise the flag here. > Either way, your change is correct: > > Reviewed-by: Stefan Hajnoczi <stefa...@redhat.com>
pgpQ8J0wC46MK.pgp
Description: OpenPGP digital signature