[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <877eow7gy7.fsf@toke.dk>
Date: Tue, 24 Apr 2018 17:38:24 +0200
From: Toke Høiland-Jørgensen <toke@...e.dk>
To: Stephen Hemminger <stephen@...workplumber.org>
Cc: netdev@...r.kernel.org, cake@...ts.bufferbloat.net
Subject: Re: [Cake] [PATCH iproute2-next v3] Add support for cake qdisc
Stephen Hemminger <stephen@...workplumber.org> writes:
> On Tue, 24 Apr 2018 16:52:57 +0200
> Toke Høiland-Jørgensen <toke@...e.dk> wrote:
>
>> Well, this is leftover from keeping track of different versions of the
>> out-of-tree patch, and we already broke compatibility pretty thoroughly
>> as a preparation for upstreaming. So I'm fine with dropping the version
>> check; will resend.
>>
>> That being said, the versioning comes from the XSTATS API, which does
>> not use netlink attributes for its members, but rather passes through as
>> struct. So what is one supposed to do in this case?
>
> If a structure is likely to change, then it should be decomposed into nested
> netlink attributes. Once you send a structure over userspace API in netlink
> it is fixed forever.
Right. Is decomposing stats into netlink attributes actually possible
within the qdisc dump_stats callback? Could we do something like:
nla_start_nest(d->skb, TCA_CAKE_STATS);
from within cake_dump_stats() and read it in cake_print_xstats in tc?
-Toke
Powered by blists - more mailing lists