While this is not expected to happen, it could still
be that a vhost_dev did not set its nvqs member.

Since `vhost_dev_start` access the device's vqs array
later without checking its size, it would cause a
Segmentation fault when nvqs is 0.

To avoid this `rare` case and made the code safer,
add a clause that ensures nvqs has been set, and
warn the user if it has not.

Signed-off-by: Albert Esteve <aesteve@redhat.com>
---
 hw/virtio/vhost.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/hw/virtio/vhost.c b/hw/virtio/vhost.c
index e2f6ffb446..78805fe5b7 100644
--- a/hw/virtio/vhost.c
+++ b/hw/virtio/vhost.c
@@ -1935,6 +1935,11 @@ int vhost_dev_start(struct vhost_dev *hdev, VirtIODevice *vdev, bool vrings)
     hdev->started = true;
     hdev->vdev = vdev;
 
+    if (!hdev->nvqs) {
+        error_report("device nvqs not set");
+        goto fail_nvqs;
+    }
+
     r = vhost_dev_set_features(hdev, hdev->log_enabled);
     if (r < 0) {
         goto fail_features;
@@ -2028,6 +2033,7 @@ fail_mem:
     if (vhost_dev_has_iommu(hdev)) {
         memory_listener_unregister(&hdev->iommu_listener);
     }
+fail_nvqs:
 fail_features:
     vdev->vhost_started = false;
     hdev->started = false;
-- 
2.41.0

On Fri, Sep 01, 2023 at 02:23:23PM +0200, Albert Esteve wrote:
> While this is not expected to happen, it could still
> be that a vhost_dev did not set its nvqs member.
> 
> Since `vhost_dev_start` access the device's vqs array
> later without checking its size, it would cause a
> Segmentation fault when nvqs is 0.
> 
> To avoid this `rare` case and made the code safer,
> add a clause that ensures nvqs has been set, and
> warn the user if it has not.
> 
> Signed-off-by: Albert Esteve <aesteve@redhat.com>
> ---
>  hw/virtio/vhost.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/hw/virtio/vhost.c b/hw/virtio/vhost.c
> index e2f6ffb446..78805fe5b7 100644
> --- a/hw/virtio/vhost.c
> +++ b/hw/virtio/vhost.c
> @@ -1935,6 +1935,11 @@ int vhost_dev_start(struct vhost_dev *hdev, VirtIODevice *vdev, bool vrings)
>      hdev->started = true;
>      hdev->vdev = vdev;
>  
> +    if (!hdev->nvqs) {
> +        error_report("device nvqs not set");
> +        goto fail_nvqs;
> +    }
> +
>      r = vhost_dev_set_features(hdev, hdev->log_enabled);
>      if (r < 0) {
>          goto fail_features;
> @@ -2028,6 +2033,7 @@ fail_mem:
>      if (vhost_dev_has_iommu(hdev)) {
>          memory_listener_unregister(&hdev->iommu_listener);
>      }
> +fail_nvqs:
>  fail_features:
>      vdev->vhost_started = false;
>      hdev->started = false;

What do we want to return in this case?
ATM the value we return (r) will be uninitialized.

> -- 
> 2.41.0

Ah I see, I wanted to move the fail check as early as possible, and went a
bit too far ahead, before initialisation.

But is ok, it needs its own value either way. What about returning -EFAULT?
Or maybe -EINVAL? I think they would fit for this error.
And then I can use `VHOST_OPS_DEBUG` to make it consistent and print the
error number.

On Mon, Oct 2, 2023 at 11:27 AM Michael S. Tsirkin <mst@redhat.com> wrote:

> On Fri, Sep 01, 2023 at 02:23:23PM +0200, Albert Esteve wrote:
> > While this is not expected to happen, it could still
> > be that a vhost_dev did not set its nvqs member.
> >
> > Since `vhost_dev_start` access the device's vqs array
> > later without checking its size, it would cause a
> > Segmentation fault when nvqs is 0.
> >
> > To avoid this `rare` case and made the code safer,
> > add a clause that ensures nvqs has been set, and
> > warn the user if it has not.
> >
> > Signed-off-by: Albert Esteve <aesteve@redhat.com>
> > ---
> >  hw/virtio/vhost.c | 6 ++++++
> >  1 file changed, 6 insertions(+)
> >
> > diff --git a/hw/virtio/vhost.c b/hw/virtio/vhost.c
> > index e2f6ffb446..78805fe5b7 100644
> > --- a/hw/virtio/vhost.c
> > +++ b/hw/virtio/vhost.c
> > @@ -1935,6 +1935,11 @@ int vhost_dev_start(struct vhost_dev *hdev,
> VirtIODevice *vdev, bool vrings)
> >      hdev->started = true;
> >      hdev->vdev = vdev;
> >
> > +    if (!hdev->nvqs) {
> > +        error_report("device nvqs not set");
> > +        goto fail_nvqs;
> > +    }
> > +
> >      r = vhost_dev_set_features(hdev, hdev->log_enabled);
> >      if (r < 0) {
> >          goto fail_features;
> > @@ -2028,6 +2033,7 @@ fail_mem:
> >      if (vhost_dev_has_iommu(hdev)) {
> >          memory_listener_unregister(&hdev->iommu_listener);
> >      }
> > +fail_nvqs:
> >  fail_features:
> >      vdev->vhost_started = false;
> >      hdev->started = false;
>
> What do we want to return in this case?
> ATM the value we return (r) will be uninitialized.
>
> > --
> > 2.41.0
>
>