Re: [RFC 0/2] of: Add whitelist

[Frank Rowand]

On 12/05/17 11:55, Alan Tull wrote:
> On Thu, Nov 30, 2017 at 6:18 AM, Frank Rowand <frowand.list@xxxxxxxxx> wrote:
>> On 11/29/17 08:31, Rob Herring wrote:
>>> On Wed, Nov 29, 2017 at 3:20 AM, Frank Rowand <frowand.list@xxxxxxxxx> wrote:
>>>> On 11/27/17 15:58, Alan Tull wrote:
>>>>> Here's a proposal for a whitelist to lock down the dynamic device tree.
>>>>>
>>>>> For an overlay to be accepted, all of its targets are required to be
>>>>> on a target node whitelist.
>>>>>
>>>>> Currently the only way I have to get on the whitelist is calling a
>>>>> function to add a node.  That works for fpga regions, but I think
>>>>> other uses will need a way of having adding specific nodes from the
>>>>> base device tree, such as by adding a property like 'allow-overlay;'
>>>>> or 'allow-overlay = "okay";' If that is acceptable, I could use some
>>>>> advice on where that particular code should go.
>>>>>
>>>>> Alan
>>>>>
>>>>> Alan Tull (2):
>>>>>   of: overlay: add whitelist
>>>>>   fpga: of region: add of-fpga-region to whitelist
>>>>>
>>>>>  drivers/fpga/of-fpga-region.c |  9 ++++++
>>>>>  drivers/of/overlay.c          | 73 +++++++++++++++++++++++++++++++++++++++++++
>>>>>  include/linux/of.h            | 12 +++++++
>>>>>  3 files changed, 94 insertions(+)
>>>>>
>>>>
>>>> The plan was to use connectors to restrict where an overlay could be applied.
>>>> I would prefer not to have multiple methods for accomplishing the same thing
>>>> unless there is a compelling reason to do so.
>>>
>>> Connector nodes need a mechanism to enable themselves, too. I don't
>>> think connector nodes are going to solve every usecase.
>>>
>>> Rob
>>>
>>
>> The overlay code related to connectors does not exist yet, so my comment
>> is going to be theoretical.
>>
>> I would expect the overlay code to check that the target of the overlay
>> fragment is a connector node, so there is no need to explicitly "enable"
>> applying an overlay to a connector node.
> 
> This will depend on how connectors are implemented.  My proposal in v1
> is that device nodes can have a flag bit.  If its not set, then an
> overlay that contains fragments that target that node can't be
> applied.  There's probably other ways a connector node could be marked
> as different from other nodes, but a flag bit seems simple.  The
> advantage to this scheme is that it gives me something I can use while
> connectors don't exist yet and it will still will be useful later for
> the implementation of connectors (giving connector drivers a way of
> marking their device nodes as valid targets).

I think it is premature to add this code to the kernel when we don't
have an agreed upon architecture for what we are trying to achieve.


> 
>>
>> -Frank
>