Re: [PATCH net-next v2 1/6] net: dcb: add new pcp selector to app object

[Petr Machata]

Petr Machata <petrm@xxxxxxxxxx> writes:

> Jakub Kicinski <kuba@xxxxxxxxxx> writes:
>
>> On Mon, 3 Oct 2022 09:52:59 +0200 Petr Machata wrote:
>>> I assumed the policy is much more strict with changes like this. If you
>>> think it's OK, I'm fine with it as well.
>>> 
>>> The userspace (lldpad in particular) is doing the opposite thing BTW:
>>> assuming everything in the nest is a DCB_ATTR_IEEE_APP. When we start
>>> emitting the new attribute, it will get confused.
>>
>> Can you add an attribute or a flag to the request which would turn
>> emitting the new attrs on?
>
> The question is whether it's better to do it anyway. My opinion is that
> if a userspace decides to make assumptions about the contents of a TLV,
> and neglects to validate the actual TLV type, it's on them, and I'm not
> obligated to keep them working. We know about this case, but really any
> attribute addition at all could potentially trip some userspace if they
> expected something else at this offset.

And re the flag: I think struct dcbmsg.dcb_pad was meant to be the place
to keep flags when the need arises, but it is not validated anywhere, so
we cannot use it. It could be a new command, but I'm not a fan. So if we
need to discriminate userspaces, I think it should be a new attribute.