Re: [PATCH] vhost/vsock: specify module version

["Michael S. Tsirkin" <mst@redhat.com>]

On Sun, Sep 29, 2024 at 07:35:35PM +0200, Aleksandr Mikhalitsyn wrote:
> On Sun, Sep 29, 2024 at 6:56 PM Michael S. Tsirkin <mst@redhat.com> wrote:
> >
> > On Thu, Sep 26, 2024 at 06:16:40PM +0200, Alexander Mikhalitsyn wrote:
> > > Add an explicit MODULE_VERSION("0.0.1") specification
> > > for a vhost_vsock module. It is useful because it allows
> > > userspace to check if vhost_vsock is there when it is
> > > configured as a built-in.
> > >
> > > Without this change, there is no /sys/module/vhost_vsock directory.
> > >
> > > With this change:
> > > $ ls -la /sys/module/vhost_vsock/
> > > total 0
> > > drwxr-xr-x   2 root root    0 Sep 26 15:59 .
> > > drwxr-xr-x 100 root root    0 Sep 26 15:59 ..
> > > --w-------   1 root root 4096 Sep 26 15:59 uevent
> > > -r--r--r--   1 root root 4096 Sep 26 15:59 version
> > >
> > > Signed-off-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@canonical.com>
> >
> >
> 
> Dear Michael,
> 
> > Why not check that the misc device is registered?
> 
> It is possible to read /proc/misc and check if "241 vhost-vsock" is
> there, but it means that userspace
> needs to have a specific logic for vsock. At the same time, it's quite
> convenient to do something like:
>     if [ ! -d /sys/modules/vhost_vsock ]; then
>         modprobe vhost_vsock
>     fi
> 
> > I'd rather not add a new UAPI until actually necessary.
> 
> I don't insist. I decided to send this patch because, while I was
> debugging a non-related kernel issue
> on my local dev environment I accidentally discovered that LXD
> (containers and VM manager)
> fails to run VMs because it fails to load the vhost_vsock module (but
> it was built-in in my debug kernel
> and the module file didn't exist). Then I discovered that before
> trying to load a module we
> check if /sys/module/<module name> exists. And found that, for some
> reason /sys/module/vhost_vsock
> does not exist when vhost_vsock is configured as a built-in, and
> /sys/module/vhost_vsock *does* exist when
> vhost_vsock is loaded as a module. It looks like an inconsistency and
> I also checked that other modules in
> drivers/vhost have MODULE_VERSION specified and version is 0.0.1. I
> thought that this change looks legitimate
> and convenient for userspace consumers.
> 
> Kind regards,
> Alex

Aha, that's a different matter.
Given userspace already depends on this UAPI, it's easier
to fix it in the kernel.


> >
> > > ---
> > >  drivers/vhost/vsock.c | 1 +
> > >  1 file changed, 1 insertion(+)
> > >
> > > diff --git a/drivers/vhost/vsock.c b/drivers/vhost/vsock.c
> > > index 802153e23073..287ea8e480b5 100644
> > > --- a/drivers/vhost/vsock.c
> > > +++ b/drivers/vhost/vsock.c
> > > @@ -956,6 +956,7 @@ static void __exit vhost_vsock_exit(void)
> > >
> > >  module_init(vhost_vsock_init);
> > >  module_exit(vhost_vsock_exit);
> > > +MODULE_VERSION("0.0.1");
> > >  MODULE_LICENSE("GPL v2");
> > >  MODULE_AUTHOR("Asias He");
> > >  MODULE_DESCRIPTION("vhost transport for vsock ");
> > > --
> > > 2.34.1
> >