| Message ID | 160139703153.162128.16860679176471296230.stgit@bahia.lan |
|---|---|
| State | New |
| Headers | show |
| Series | vhost: Skip access checks on GIOVAs | expand |
On 2020/9/30 上午12:30, Greg Kurz wrote: > When the IOTLB device is enabled, the vring addresses we get > from userspace are GIOVAs. It is thus wrong to pass them down > to access_ok() which only takes HVAs. > > Access validation is done at prefetch time with IOTLB. Teach > vq_access_ok() about that by moving the (vq->iotlb) check > from vhost_vq_access_ok() to vq_access_ok(). This prevents > vhost_vring_set_addr() to fail when verifying the accesses. > No behavior change for vhost_vq_access_ok(). > > BugLink: https://bugzilla.redhat.com/show_bug.cgi?id=1883084 > Fixes: 6b1e6cc7855b ("vhost: new device IOTLB API") > Cc: jasowang@redhat.com > CC: stable@vger.kernel.org # 4.14+ > Signed-off-by: Greg Kurz <groug@kaod.org> > --- > drivers/vhost/vhost.c | 9 +++++---- > 1 file changed, 5 insertions(+), 4 deletions(-) > > diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c > index b45519ca66a7..c3b49975dc28 100644 > --- a/drivers/vhost/vhost.c > +++ b/drivers/vhost/vhost.c > @@ -1290,6 +1290,11 @@ static bool vq_access_ok(struct vhost_virtqueue *vq, unsigned int num, > vring_used_t __user *used) > > { > + /* If an IOTLB device is present, the vring addresses are > + * GIOVAs. Access validation occurs at prefetch time. */ > + if (vq->iotlb) > + return true; > + > return access_ok(desc, vhost_get_desc_size(vq, num)) && > access_ok(avail, vhost_get_avail_size(vq, num)) && > access_ok(used, vhost_get_used_size(vq, num)); > @@ -1383,10 +1388,6 @@ bool vhost_vq_access_ok(struct vhost_virtqueue *vq) > if (!vq_log_access_ok(vq, vq->log_base)) > return false; > > - /* Access validation occurs at prefetch time with IOTLB */ > - if (vq->iotlb) > - return true; > - > return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used); > } > EXPORT_SYMBOL_GPL(vhost_vq_access_ok); > Acked-by: Jason Wang <jasowang@redhat.com> Thanks
diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c index b45519ca66a7..c3b49975dc28 100644 --- a/drivers/vhost/vhost.c +++ b/drivers/vhost/vhost.c @@ -1290,6 +1290,11 @@ static bool vq_access_ok(struct vhost_virtqueue *vq, unsigned int num, vring_used_t __user *used) { + /* If an IOTLB device is present, the vring addresses are + * GIOVAs. Access validation occurs at prefetch time. */ + if (vq->iotlb) + return true; + return access_ok(desc, vhost_get_desc_size(vq, num)) && access_ok(avail, vhost_get_avail_size(vq, num)) && access_ok(used, vhost_get_used_size(vq, num)); @@ -1383,10 +1388,6 @@ bool vhost_vq_access_ok(struct vhost_virtqueue *vq) if (!vq_log_access_ok(vq, vq->log_base)) return false; - /* Access validation occurs at prefetch time with IOTLB */ - if (vq->iotlb) - return true; - return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used); } EXPORT_SYMBOL_GPL(vhost_vq_access_ok);
When the IOTLB device is enabled, the vring addresses we get from userspace are GIOVAs. It is thus wrong to pass them down to access_ok() which only takes HVAs. Access validation is done at prefetch time with IOTLB. Teach vq_access_ok() about that by moving the (vq->iotlb) check from vhost_vq_access_ok() to vq_access_ok(). This prevents vhost_vring_set_addr() to fail when verifying the accesses. No behavior change for vhost_vq_access_ok(). BugLink: https://bugzilla.redhat.com/show_bug.cgi?id=1883084 Fixes: 6b1e6cc7855b ("vhost: new device IOTLB API") Cc: jasowang@redhat.com CC: stable@vger.kernel.org # 4.14+ Signed-off-by: Greg Kurz <groug@kaod.org> --- drivers/vhost/vhost.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-)