Re: [PATCH] PCI/sysfs: Fix read permissions for VPD attributes

["Leon Romanovsky" <leon@kernel.org>]

On Tue, Nov 5, 2024, at 18:26, Leon Romanovsky wrote:
> On Tue, Nov 05, 2024 at 09:24:55AM -0600, Bjorn Helgaas wrote:
>> On Tue, Nov 05, 2024 at 09:51:30AM +0200, Leon Romanovsky wrote:
>> > On Mon, Nov 04, 2024 at 06:10:27PM -0600, Bjorn Helgaas wrote:
>> > > On Sun, Nov 03, 2024 at 02:33:44PM +0200, Leon Romanovsky wrote:
>> > > > On Fri, Nov 01, 2024 at 11:47:37AM -0500, Bjorn Helgaas wrote:
>> > > > > On Fri, Nov 01, 2024 at 04:33:00PM +0200, Leon Romanovsky wrote:
>> > > > > > On Thu, Oct 31, 2024 at 06:22:52PM -0500, Bjorn Helgaas wrote:
>> > > > > > > On Tue, Oct 29, 2024 at 07:04:50PM -0500, Bjorn Helgaas wrote:
>> > > > > > > > On Mon, Oct 28, 2024 at 10:05:33AM +0200, Leon Romanovsky wrote:
>> > > > > > > > > From: Leon Romanovsky <leonro@nvidia.com>
>> > > > > > > > > 
>> > > > > > > > > The Virtual Product Data (VPD) attribute is not
>> > > > > > > > > readable by regular user without root permissions.
>> > > > > > > > > Such restriction is not really needed, as data
>> > > > > > > > > presented in that VPD is not sensitive at all.
>> > > > > > > > > 
>> > > > > > > > > This change aligns the permissions of the VPD
>> > > > > > > > > attribute to be accessible for read by all users,
>> > > > > > > > > while write being restricted to root only.
>> > ...
>> 
>> > > What's the use case?  How does an unprivileged user use the VPD
>> > > information?
>> > 
>> > We have to add new field keyword=value in VA section of VPD, which
>> > will indicate very specific sub-model for devices used as a bridge.
>> > 
>> > > I can certainly imagine using VPD for bug reporting, but that
>> > > would typically involve dmesg, dmidecode, lspci -vv, etc, all of
>> > > which already require privilege, so it's not clear to me how
>> > > public VPD info would help in that scenario.
>> > 
>> > I'm targeting other scenario - monitoring tool, which doesn't need
>> > root permissions for reading data. It needs to distinguish between
>> > NIC sub-models.
>> 
>> Maybe the driver could expose something in sysfs?  Maybe the driver
>> needs to know the sub-model as well, and reading VPD once in the
>> driver would make subsequent userspace sysfs reads trivial and fast.
>
> Our PCI driver lays in netdev subsystem and they have long-standing
> position do not allow any custom sysfs files. To be fair, we (RDMA)
> don't allow custom sysfs files too.
>
> Driver doesn't need to know this information as it is extra key=value in
> existing [VA] field, while driver relies on multiple FW capabilities
> to enable/disable functionality.
>
> Current [VA] line:
> "[VA] Vendor specific: 
> MLX:MN=MLNX:CSKU=V2:UUID=V3:PCI=V0:MODL=CX713106A"
> Future [VA] line:
> "[VA] Vendor specific: 
> MLX:MN=MLNX:CSKU=V2:UUID=V3:PCI=V0:MODL=CX713106A,SMDL=SOMETHING"
>
> Also the idea that we will duplicate existing functionality doesn't
> sound like a good approach to me, and there is no way that it is
> possible to expose as subsystem specific file.
>
> What about to allow existing VPD sysfs file to be readable for everyone 
> for our devices?
> And if this allow list grows to much, we will open it for all devices 
> in the world?

Bjorn,

I don't see this patch in https://git.kernel.org/pub/scm/linux/kernel/git/pci/pci.git/log/?h=next
So what did you decide? How can we enable existing VPD access to regular users?

Thanks

>
> Thanks
>
>> 
>> Bjorn