Re: [PATCH 0/3] Autoloading for RDMA netlink modules

[Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>]

[-- Attachment #1: Type: text/plain, Size: 1410 bytes --]

On Tue, Jul 25, 2017 at 08:34:20AM -0600, Jason Gunthorpe wrote:
> On Tue, Jul 25, 2017 at 09:32:36AM +0300, Leon Romanovsky wrote:
>
> > First, it interferes badly with my RDMA netlink series posted here in
> > previous cycle and I'm expecting to see it in, so I can move forward
> > with RDMAtool implementation.
>
> I'm sure we can fix a few conflicts...

It is more than a couple of conflicts, the implementation will be
slightly different.

Let's wait till Doug creates for-next with RDMA netlink patches, because
I assume that 2 out of 3 patches will be completely different.

>
> > Second, from brief review, it is unclear to me, how do you protect
> > autoload caused by regular user who sends netlink message to ib_core.
>
> You don't, that isn't how kernel auto loading works. Any user can
> attempt to create any kind of socket and the core kernel code will
> load the correct module to support that socket. That is the standard
> way kernel autoloading works, particularly for net families.

Ok, I'll take a look on it.

>
> > Third, it is ib_core who should autoload modules by default, and not some
> > random daemon in user space.
>
> Kernel autoloading of userspace facing kernel services is always
> triggered by a user process attempting to use the service, so this
> comment makes no sense to me.

My concern that "regular" user will trigger "admin" operation without
any checks.

>
> Jason

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]