On Tue, Oct 15, 2024 at 6:19 PM Michael S. Tsirkin <mst@xxxxxxxxxx> wrote:
>
> On Mon, Oct 14, 2024 at 03:56:33PM -0500, Mike Christie wrote:
> > On 10/3/24 8:58 PM, Cindy Lu wrote:
> > > Add a new UAPI to support setting the vhost device to
> > > use task mode. The user space application needs to use
> > > VHOST_SET_INHERIT_FROM_OWNER to set the mode.
> > > This setting must be set before VHOST_SET_OWNER is set.
> > >
> > > Signed-off-by: Cindy Lu <lulu@xxxxxxxxxx>
> > > ---
> > > drivers/vhost/vhost.c | 18 +++++++++++++++++-
> > > include/uapi/linux/vhost.h | 2 ++
> > > 2 files changed, 19 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
> > > index 08c9e77916ca..0e5c81026acd 100644
> > > --- a/drivers/vhost/vhost.c
> > > +++ b/drivers/vhost/vhost.c
> > > @@ -2341,8 +2341,24 @@ long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp)
> > > {
> > > struct eventfd_ctx *ctx;
> > > u64 p;
> > > - long r;
> > > + long r = 0;
> > > int i, fd;
> > > + bool inherit_owner;
> > > +
> > > + if (ioctl == VHOST_SET_INHERIT_FROM_OWNER) {
> >
> > Maybe instead of a modparam and this ioctl we just want a new ioctl:
> >
> > /*
> > * This will setup the owner based on the calling thread instead of
> > * using kthread.
> > */
> > #define VHOST_INHERIT_OWNER _IO(VHOST_VIRTIO, 0x83)
>
> I feel this is not good because it is insecure -
> process should not normally have a say in whether
> namespaces work correctly.
Note there's still a lot of kthread users, so the "problem" is not
specific to vhost.
> So we want the system admin to be able to block the
> old mode.
Then we will break the userspace silently which seems not good.
>
> > It would initially be used by vhost-scsi when worker_per_virtqueue=true
> > since that is a new use case and there will be no regressions.
> >
> > For the other cases we default to VHOST_SET_OWNER. Other QEMU cases or
> > tool XYZ can use the new ioctl when they are ready.
>
> I do not like it that we switched default now we apparently will be
> switching it back. Will break some userspace whatever we do.
>
> --
> MST
>
Thanks
