Re: [RFC] PCI: allow sysfs file owner to read the config space with CAP_SYS_RAWIO
From: Greg KH
Date: Mon Oct 19 2020 - 09:46:47 EST
On Mon, Oct 19, 2020 at 06:51:39PM +0530, Allen wrote:
> > > > >
> > > > > Access to pci config space is explictly checked with CAP_SYS_ADMIN
> > > > > in order to read configuration space past the frist 64B.
> > > > >
> > > > > Since the path is only for reading, could we use CAP_SYS_RAWIO?
> > > >
> > > > Why? What needs this reduced capability?
> > >
> > > Thanks for the review.
> > >
> > > We need read access to /sys/bus/pci/devices/, We need write access to config,
> > > remove, rescan & enable files under the device directory for each PCIe
> > > functions & the downstream PCIe port.
> > >
> > > We need r/w access to sysfs to unbind and rebind the root complex.
> >
> > That didn't answer my question at all.
>
> Sorry about that, breaking it down:
>
> When the machine first boots, the VFIO device bindings under /dev/vfio
> are not present.
>
> root@localhost:/tmp# ls -l /dev/vfio/
> total 0
> crw-rw-rw-. 1 root root 10, 196 Jan 5 01:47 vfio
>
> We have an agent which needs to run the following commands (We get
> access denied here and need permissions to do this).
> echo -n xxxx yyyy > /sys/module/vfio_pci/drivers/pci:vfio-pci/new_id
> echo -n xxxx yyyy > /sys/module/vfio_pci/drivers/pci:vfio-pci/new_id
>
> And we want to avoid handing CAP_SYS_ADMIN here. Which is why the
> thought about CAP_SYS_RAWIO.
But that is not what you were asking this patch to do at all. So why
bring it up?
new_id is NOT for "raw io" control, that should be only for admin
priviliges.
And just because the vfio driver "abuses" this
traditionally-debug-functionality doesn't mean you get to abuse the
permission levels either.
> > Why can't you have the process that wants to do all of the above, have
> > admin rights as well? Doing all of that is _very_ low-level and can
> > cause all sorts of horrible things to happen to your machine, and is not
> > really "raw io" in the traditional sense at all, right?
>
>
> If the above approach is going to cause the system to do horrible things,
> then I'll drop the idea.
Of course it can cause the system to do horrible things, try it yourself
and see!
greg k-h