Date:   Tue, 12 Apr 2022 07:09:54 -0400
From:   Jamal Hadi Salim <jhs@...atatu.com>
To:     Boris Sukholitko <boris.sukholitko@...adcom.com>,
        Stephen Hemminger <stephen@...workplumber.org>
Cc:     netdev@...r.kernel.org, "David S . Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Cong Wang <xiyou.wangcong@...il.com>,
        Jiri Pirko <jiri@...nulli.us>,
        "Gustavo A . R . Silva" <gustavoars@...nel.org>,
        Vladimir Oltean <vladimir.oltean@....com>,
        Eric Dumazet <edumazet@...gle.com>,
        zhang kai <zhangkaiheb@....com>,
        Yoshiki Komachi <komachi.yoshiki@...il.com>,
        Ilya Lifshits <ilya.lifshits@...adcom.com>
Subject: Re: [PATCH iproute2-next 0/2] flower: match on the number of vlan
 tags

On 2022-04-12 06:45, Boris Sukholitko wrote:
> On Mon, Apr 11, 2022 at 08:45:36AM -0700, Stephen Hemminger wrote:
>> On Mon, 11 Apr 2022 16:32:00 +0300
>> Boris Sukholitko <boris.sukholitko@...adcom.com> wrote:
>>
>>> Hi,
>>>
>>> Our customers in the fiber telecom world have network configurations
>>> where they would like to control their traffic according to the number
>>> of tags appearing in the packet.
>>>
>>> For example, TR247 GPON conformance test suite specification mostly
>>> talks about untagged, single, double tagged packets and gives lax
>>> guidelines on the vlan protocol vs. number of vlan tags.
>>>
>>> This is different from the common IT networks where 802.1Q and 802.1ad
>>> protocols are usually describe single and double tagged packet. GPON
>>> configurations that we work with have arbitrary mix the above protocols
>>> and number of vlan tags in the packet.
>>>
>>> The following patch series implement number of vlans flower filter. They
>>> add num_of_vlans flower filter as an alternative to vlan ethtype protocol
>>> matching. The end result is that the following command becomes possible:
>>>
>>> tc filter add dev eth1 ingress flower \
>>>    num_of_vlans 1 vlan_prio 5 action drop
>>>
>>> The corresponding kernel patches are being sent separately.
>>>
>>> Thanks,
>>> Boris.
>>
>> Maybe something custom like this is better done by small BPF program?
> 
> I am not sure it is feasible to have BPF match done on the number of
> vlans and have the rest of TC machinery work as expected.
> 
> For example, the flower filters look at the protocol to allow matching
> on the vlan fields. Patch 5 of the kernel part of the series adds number
> of vlans as a different precondition. Having BPF program does nothing
> for it.
> 
> Replicating more of TC functionality in the BPF to alleviate such pain
> points is probably possible but will not be "simple".
> 
> Also (and sorry for the philosophy rant!) there is an issue of UI and
> intended audience here. The TC tools are well known and accessible. I am
> not sure that the same can be said for a custom BPF programs. :)
> 

I hate to use +1 (proverbial death-by-pluse-one in effect) but, damn
couldnt resist. Stephen, this mantra only makes sense if:

a) You are a big cloud vendor with a gazillion developers who will
write, test and maintain your custom code.
b) willing to pay some consultant or other vendor to do the above.

The majority of the world just wants to pay RH or Cannonical for the
basic distro support and then run their bash scripts (the ops part,
_not the dev_).
I wouldnt call what Boris is doing as "custom". The VLAN infrastructure
has some challenges when it comes to multiple tags.

My 2c Canadiana rant:
I am not saying there's no room for custom - in which case ebpf has
a role to play (and we widely use it here when it makes sense), just
that the standard answer shouldnt be "use ebpf" just because.

Rant continued:
As community we now seem to be driven by cloud vendor mentality really.
What happened to "lets contribute back so everyone can benefit"?
There's a lot of value still in upstreaming things.

cheers,
jamal

Powered by blists - more mailing lists