[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Wed, 6 Jun 2018 08:12:20 +0300
From: Or Gerlitz <gerlitz.or@...il.com>
To: Jakub Kicinski <kubakici@...pl>, Paul Blakey <paulb@...lanox.com>
Cc: Jiri Pirko <jiri@...lanox.com>,
Cong Wang <xiyou.wangcong@...il.com>,
Jamal Hadi Salim <jhs@...atatu.com>,
David Miller <davem@...emloft.net>,
Linux Netdev List <netdev@...r.kernel.org>,
Yevgeny Kliteynik <kliteyn@...lanox.com>,
Roi Dayan <roid@...lanox.com>,
Shahar Klein <shahark@...lanox.com>,
Mark Bloch <markb@...lanox.com>,
Or Gerlitz <ogerlitz@...lanox.com>
Subject: Re: [PATCH net] net: sched: cls: Fix offloading when ingress dev is vxlan
On Tue, Jun 5, 2018 at 9:59 PM, Jakub Kicinski <kubakici@...pl> wrote:
> On Tue, 5 Jun 2018 11:57:47 -0700, Jakub Kicinski wrote:
>> On Tue, 5 Jun 2018 11:04:03 +0300, Paul Blakey wrote:
>> > When using a vxlan device as the ingress dev, we count it as a
>> > "no offload dev", so when such a rule comes and err stop is true,
>> > we fail early and don't try the egdev route which can offload it
>> > through the egress device.
>> >
>> > Fix that by not calling the block offload if one of the devices
>> > attached to it is not offload capable, but make sure egress on such case
>> > is capable instead.
>> >
>> > Fixes: caa7260156eb ("net: sched: keep track of offloaded filters [..]")
>> > Reviewed-by: Roi Dayan <roid@...lanox.com>
>> > Acked-by: Jiri Pirko <jiri@...lanox.com>
>> > Signed-off-by: Paul Blakey <paulb@...lanox.com>
>>
>> Very poor commit message. What you're doing is re-enabling skip_sw
>> filters on tunnel devices which semantically make no sense and
>> shouldn't have been allowed in the first place.
>>
>> This will breaks block sharing between tunnels and HW netdevs (because
>> you skip the tcf_block_cb_call() completely). The entire egdev idea
>> remains badly broken accepting filters like this:
>>
>> # tc filter add dev vxlan0 ingress \
>> matchall action skip_sw \
>> mirred egress redirect dev lo \
>> mirred egress redirect dev sw1np0
>
> For above we probably need something like this (untested):
>
> diff --git a/net/sched/act_api.c b/net/sched/act_api.c
> index 3f4cf930f809..71e5eebec31a 100644
> --- a/net/sched/act_api.c
> +++ b/net/sched/act_api.c
> @@ -1511,7 +1511,7 @@ int tc_setup_cb_egdev_call(const struct net_device *dev,
> struct tcf_action_egdev *egdev = tcf_action_egdev_lookup(dev);
>
> if (!egdev)
> - return 0;
> + return err_stop ? -EOPNOTSUPP : 0;
> return tcf_action_egdev_cb_call(egdev, type, type_data, err_stop);
> }
> EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_call);
Jakub,
We will test it out and let you know
> But the correct fix is to remove egdev crutch completely IMO.
Not against it, sometimes designs should change and be replaced
with better ones, happens
Powered by blists - more mailing lists