Re: [PATCH net-next v10 1/4] net netlink: Add new type NLA_FLAG_BITS

[Jiri Pirko <jiri@resnulli.us>]

Mon, Jun 12, 2017 at 03:51:19PM CEST, jhs@mojatatu.com wrote:
>On 17-06-12 07:43 AM, Jiri Pirko wrote:
>> Mon, Jun 12, 2017 at 01:10:56PM CEST, jhs@mojatatu.com wrote:
>> > On 17-06-12 06:34 AM, Jiri Pirko wrote:
>> > > Sun, Jun 11, 2017 at 08:37:25PM CEST, jhs@mojatatu.com wrote:
>> > > > On 17-06-11 01:38 PM, Jamal Hadi Salim wrote:
>> > > > > On 17-06-11 09:49 AM, Jiri Pirko wrote:
>> > > > > > Sun, Jun 11, 2017 at 01:53:43PM CEST, jhs@mojatatu.com wrote:
>> > > > > > > From: Jamal Hadi Salim <jhs@mojatatu.com>
>> > > > > 
>> > > > > 
>> > > > > > > This patch also provides an extra feature: a validation callback
>> > > > > > > that could be speaciliazed for other types.
>> > > > > > 
>> > > > > > s/speaciliazed/speciliazed/
>> > > > > > 
>> > > > > 
>> > > > > Will fix.
>> > > > > 
>> > > > > 
>> > > > > > > 
>> > > > > > > [ATTR_GOO] = { .type = MYTYPE,
>> > > > > > >              .validation_data = &myvalidation_data,
>> > > > > > >                  .validate_content = mycontent_validator },
>> > > > > > 
>> > > > > > Indent is wrong. (Does not matter really in desc, but anyway)
>> > > > > > 
>> > > > > 
>> > > > > I cant find out how it got indented that way; my source
>> > > > > or email dont show it as such (but really doesnt matter).
>> > > > > 
>> > > > > 
>> > > > > > Suggested-by: Jiri Pirko <jiri@mellanox.com>
>> > > > > > 
>> > > > > 
>> > > > > Will add.
>> > > > > 
>> > > > > > 
>
>> > > I believe that this struct should go into:
>> > > include/uapi/linux/netlink.h
>> > > 
>> > > struct nla_flag_bits {
>> > > 	__u32 nla_flag_bits_values;
>> > > 	__u32 nla_flag_bits_selector;
>> > > };
>> > > 
>> > > Then you can use it from userspace and everywhere in kernel.
>> > > 
>> > That file seems to be very out of place for this stuff.
>> 
>> Howcome? It is a common netlink api.
>> 
>
>Take a look at that file's content. It talks about what goes in
>the netlink header. Adding types in it seems out of place.

Bit less than rtnetlink.h. But I agree it is not optimal. Re optimal,
please see below.


>
>
>> > NLA_XXX are kernel side types.  They are part of net/netlink.h which is
>> > not uapi accessible.
>> > David Ahern has submitted a patch to move all those defines to iproute2.
>> > Will make sense to move these to a uapi/linux/netlink-types.h but that
>> > is waay beyond the scope of this patch set.
>> 
>> Well, I don't think so :)
>> 
>> The thing is, struct nla_flag_bits is tightly coupled with NLA_FLAG_BITS
>> enum value. They should be in the same uapi file. That makes sense to me.
>> 
>
>Sure - they should be in the same file. But is it uapi/linux/netlink.h?

Might be the netlink-types.h you mentioned above.

ccing DavidA.