Re: [PATCH] vhost: force spec specified alignment on types

From: Michael S. Tsirkin
Date: Mon Apr 06 2020 - 10:34:50 EST


On Mon, Apr 06, 2020 at 10:09:27PM +0800, Jason Wang wrote:
>
> On 2020/4/6 äå9:55, Michael S. Tsirkin wrote:
> > On Mon, Apr 06, 2020 at 09:34:00PM +0800, Jason Wang wrote:
> > > On 2020/4/6 äå8:50, Michael S. Tsirkin wrote:
> > > > The ring element addresses are passed between components with different
> > > > alignments assumptions. Thus, if guest/userspace selects a pointer and
> > > > host then gets and dereferences it, we might need to decrease the
> > > > compiler-selected alignment to prevent compiler on the host from
> > > > assuming pointer is aligned.
> > > >
> > > > This actually triggers on ARM with -mabi=apcs-gnu - which is a
> > > > deprecated configuration, but it seems safer to handle this
> > > > generally.
> > > >
> > > > I verified that the produced binary is exactly identical on x86.
> > > >
> > > > Signed-off-by: Michael S. Tsirkin <mst@xxxxxxxxxx>
> > > > ---
> > > >
> > > > This is my preferred way to handle the ARM incompatibility issues
> > > > (in preference to kconfig hacks).
> > > > I will push this into next now.
> > > > Comments?
> > >
> > > I'm not sure if it's too late to fix. It would still be still problematic
> > > for the userspace that is using old uapi headers?
> > >
> > > Thanks
> > It's not a problem in userspace. The problem is when
> > userspace/guest uses 2 byte alignment and passes it to kernel
> > assuming 8 byte alignment. The fix is for host not to
> > make these assumptions.
>
>
> Yes, but I meant when userspace is complied with apcs-gnu, then it still
> assumes 8 byte alignment?
>
> Thanks


That's not a problem since with vhost userspace is doing the allocation.
So it can increase alignment with no bad effect.

I agree it's probably safest not to touch struct vring at all though.


>
> >
> > > > drivers/vhost/vhost.h | 6 ++---
> > > > include/uapi/linux/virtio_ring.h | 41 ++++++++++++++++++++++++--------
> > > > 2 files changed, 34 insertions(+), 13 deletions(-)
> > > >
> > > > diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h
> > > > index cc82918158d2..a67bda9792ec 100644
> > > > --- a/drivers/vhost/vhost.h
> > > > +++ b/drivers/vhost/vhost.h
> > > > @@ -74,9 +74,9 @@ struct vhost_virtqueue {
> > > > /* The actual ring of buffers. */
> > > > struct mutex mutex;
> > > > unsigned int num;
> > > > - struct vring_desc __user *desc;
> > > > - struct vring_avail __user *avail;
> > > > - struct vring_used __user *used;
> > > > + vring_desc_t __user *desc;
> > > > + vring_avail_t __user *avail;
> > > > + vring_used_t __user *used;
> > > > const struct vhost_iotlb_map *meta_iotlb[VHOST_NUM_ADDRS];
> > > > struct vhost_desc *descs;
> > > > diff --git a/include/uapi/linux/virtio_ring.h b/include/uapi/linux/virtio_ring.h
> > > > index 559f42e73315..cd6e0b2eaf2f 100644
> > > > --- a/include/uapi/linux/virtio_ring.h
> > > > +++ b/include/uapi/linux/virtio_ring.h
> > > > @@ -118,16 +118,6 @@ struct vring_used {
> > > > struct vring_used_elem ring[];
> > > > };
> > > > -struct vring {
> > > > - unsigned int num;
> > > > -
> > > > - struct vring_desc *desc;
> > > > -
> > > > - struct vring_avail *avail;
> > > > -
> > > > - struct vring_used *used;
> > > > -};
> > > > -
> > > > /* Alignment requirements for vring elements.
> > > > * When using pre-virtio 1.0 layout, these fall out naturally.
> > > > */
> > > > @@ -164,6 +154,37 @@ struct vring {
> > > > #define vring_used_event(vr) ((vr)->avail->ring[(vr)->num])
> > > > #define vring_avail_event(vr) (*(__virtio16 *)&(vr)->used->ring[(vr)->num])
> > > > +/*
> > > > + * The ring element addresses are passed between components with different
> > > > + * alignments assumptions. Thus, we might need to decrease the compiler-selected
> > > > + * alignment, and so must use a typedef to make sure the __aligned attribute
> > > > + * actually takes hold:
> > > > + *
> > > > + * https://gcc.gnu.org/onlinedocs//gcc/Common-Type-Attributes.html#Common-Type-Attributes
> > > > + *
> > > > + * When used on a struct, or struct member, the aligned attribute can only
> > > > + * increase the alignment; in order to decrease it, the packed attribute must
> > > > + * be specified as well. When used as part of a typedef, the aligned attribute
> > > > + * can both increase and decrease alignment, and specifying the packed
> > > > + * attribute generates a warning.
> > > > + */
> > > > +typedef struct vring_desc __attribute__((aligned(VRING_DESC_ALIGN_SIZE)))
> > > > + vring_desc_t;
> > > > +typedef struct vring_avail __attribute__((aligned(VRING_AVAIL_ALIGN_SIZE)))
> > > > + vring_avail_t;
> > > > +typedef struct vring_used __attribute__((aligned(VRING_USED_ALIGN_SIZE)))
> > > > + vring_used_t;
> > > > +
> > > > +struct vring {
> > > > + unsigned int num;
> > > > +
> > > > + vring_desc_t *desc;
> > > > +
> > > > + vring_avail_t *avail;
> > > > +
> > > > + vring_used_t *used;
> > > > +};
> > > > +
> > > > static inline void vring_init(struct vring *vr, unsigned int num, void *p,
> > > > unsigned long align)
> > > > {