Re: [PATCH net-next v11 3/4] net sched actions: dump more than TCA_ACT_MAX_PRIO actions per batch

[Jiri Pirko <jiri@resnulli.us>]

Tue, Jul 25, 2017 at 02:34:58PM CEST, jhs@mojatatu.com wrote:
>On 17-07-25 07:33 AM, Jiri Pirko wrote:
>> Tue, Jul 25, 2017 at 01:22:44PM CEST, jhs@mojatatu.com wrote:
>
>> > > fb? bf? nbf? Please make this synced within the patchset.
>> > > 
>> > > 
>> > 
>> > Ok, what do you like best? ;->
>> 
>> "bf"
>> 
>
>Ok.
>
>> 
>> > 
>> > > Don't you need to mask value with selector? In fact, I think that
>> > > nla_get_bitfield_32 could just return u32 which would be (value&selector).
>> > > The validation takes care of unsupported bits.
>> > 
>> > For my use case I dont need any of the above since I dont need to
>> > unset things. In other use cases you will need both selector and
>> > value in case someone wants a bit to be set to 0.
>> > Infact I think i will rename that helper to "nla_get_bitvalue_32"
>> > to be more precise.
>> 
>> The getter should contain name of the type, so "nla_get_bitfield32_val"
>> is much better.
>> 
>
>Actually I mispoke. I was returning the struct not the value. So
>nla_get_bitfield32() is a better name.

ack


>
>> What if I pass val 0x1 and selector 0x0 from userspace. I don't have the
>> bit selected, so you should not process it in kernel, no?
>> 
>
>Yes, valid point. I am not sure - should we reject?

I think that the validation might check this and reject. Makes sense to
me.