lkml.org 
[lkml]   [2024]   [Apr]   [2]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH RFC net-next 00/10] MC Flood disable and snooping
From
On 4/2/24 20:43, Vladimir Oltean wrote:
> Hi Nikolai,
>
> On Tue, Apr 02, 2024 at 12:28:38PM +0300, Nikolay Aleksandrov wrote:
>> For the bridge patches:
>> Nacked-by: Nikolay Aleksandrov <razor@blackwall.org>
>>
>> You cannot break the multicast flood flag to add support for a custom
>> use-case. This is unacceptable. The current bridge behaviour is correct
>> your patch 02 doesn't fix anything, you should configure the bridge
>> properly to avoid all those problems, not break protocols.
>>
>> Your special use case can easily be solved by a user-space helper or
>> eBPF and nftables. You can set the mcast flood flag and bypass the
>> bridge for these packets. I basically said the same in 2021, if this is
>> going to be in the bridge it should be hidden behind an option that is
>> default off. But in my opinion adding an option to solve such special
>> cases is undesirable, they can be easily solved with what's currently
>> available.
>
> I appreciate your time is limited, but could you please translate your
> suggestion, and detail your proposed alternative a bit, for those of us
> who are not very familiar with IP multicast snooping?
>

My suggestion is not related to snooping really, but to the goal of
patches 01-03. The bridge patches in this set are trying to forward
traffic that is not supposed to be forwarded with the proposed
configuration, so that can be done by a user-space helper that installs
rules to bypass the bridge specifically for those packets while
monitoring the bridge state to implement a policy and manage these rules
in order to keep snooping working.

> Bypass the bridge for which packets? General IGMP/MLD queries? Wouldn't
> that break snooping? And then do what with the packets, forward them in
> another software layer than the bridge?
>

The ones that are not supposed to be forwarded in the proposed config
and are needed for this use case (control traffic and link-local).
Obviously to have proper snooping you'd need to manage these bypass
rules and use them only while needed.

> I also don't quite understand the suggestion of turning on mcast flooding:
> isn't Joseph saying that he wants it off for the unregistered multicast
> data traffic?

Ah my bad, I meant to turn off flooding and bypass the bridge for those
packets and ports while necessary, under necessary can be any policy
that the user-space helper wants to implement.

In any case, if this is going to be yet another kernel solution then it
must be a new option that is default off, and doesn't break current
mcast flood flag behaviour.

In general my opinion is that the whole snooping control must be in
user-space and only have the dataplane in the kernel, but that is beyond
the scope of this set.


\
 
 \ /
  Last update: 2024-05-27 16:19    [W:0.069 / U:1.072 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site