Fri, Oct 06, 2023 at 09:44:57PM CEST, kuba@xxxxxxxxxx wrote:
>On Fri, 6 Oct 2023 18:53:04 +0200 Jiri Pirko wrote:
>> Fri, Oct 06, 2023 at 04:55:36PM CEST, kuba@xxxxxxxxxx wrote:
>> >> I'm confused. Didn't you say you'll remove this? If not, my question
>> >> from v1 still stands.
>> >
>> >Perhaps we should dis-allow setting version in non-genetlink-legacy
>> >specs? I thought it may be a useful thing to someone, at some point,
>> >but so far the scoreboard is: legit uses: 0, confused uses: 1 :S
>> >
>> >Thoughts?
>>
>> I don't know what the meaning of version is. I just never saw that being
>> touched. Is there any semantics documented for it?
>>
>> Kuba, any opinion?
>
>/me switches the first name in From :P
I messed up a bit. Kuba* confusion, sorry :)
>
>I think it basically predates the op / policy introspection,
>and allows people to break backward compat.
>
>drop_monitor bumped to 2 in 2009:
>
> 683703a26e46 ("drop_monitor: Update netlink protocol to include
>netlink attribute header in alert message")
>
>which breaks backward compat.
>
>genetlink ctrl went to 2 in 2006:
>
> 334c29a64507 ("[GENETLINK]: Move command capabilities to flags.")
>
>which moves some info around in attrs, also breaks backward compat
>if someone depended on the old placement.
>
>ovs did it in 2013:
>
> 44da5ae5fbea ("openvswitch: Drop user features if old user space
>attempted to create datapath")
>
>again, breaks backwards compat.
>
>
>I guess it may still make one day to bump the version for some proto
>which has very tight control over the user space. But it hasn't
>happened for 10 years.
But since by the policy we cannot break uapi compat, version should be
never bumped. I wonder howcome it is legit in the examples you listed
above...
Let's forbid that in genetlink.yaml. I have a patch ready, please ack
this approach.
Thx!
