Re: [net-next v3 1/6] net: dcb: add new pcp selector to app object

[Petr Machata]

<Daniel.Machon@xxxxxxxxxxxxx> writes:

>> >> And vice versa: I'm not sure we want to permit sending the standard
>> >> attributes in the DCB encap.
>> >
>> > dcbnl_app_attr_type_get() in dcbnl_ieee_fill() takes care of this. IEEE are
>> > always sent in DCB_ATTR_IEEE and non-std are sent in DCB_ATTR_DCB.
>> 
>> By "sending" I meant userspace sending this to the kernel. So bounce
>> extended opcodes that are wrapped in IEEE and bounce IEEE opcodes
>> wrapped in DCB as well.
>
> Right. Then we only need to decide what to do with any opcode in-between
> (not defined in uapi, neither ieee or extension opcode, 7-254). If they are 
> sent in DCB_ATTR_DCB they should be bounced, because we agreed that we can 
> interpret data in the new attr), _but_ if they are sent in DCB_ATTR_IEEE I 
> guess we should accept them, to not break userspace that is already sending
> them.

I see, it's not currently validating at all. It just relies on the
driver to do the validation, but e.g. bnxt_dcbnl_ieee_setapp(), just
lets nonsense right through.

OK, but this interface is built on standards. The selector has a
well-defined, IEEE-backed meaning with enumerators published in the UAPI
headers. As before, even though this constitutes API breakage, IMHO if
anyone relies on shoving random garbage through this interface, it's on
them...

I think it's kosher to start bouncing undefined selectors.