Re: [PATCH 1/3] net: bridge: Implement bridge flag local_receive

From: Ido Schimmel <idosch@idosch.org>
To: mattias.forsblad+netdev@gmail.com
Cc: Nikolay Aleksandrov <razor@blackwall.org>,
	netdev@vger.kernel.org, "David S . Miller" <davem@davemloft.net>,
	Jakub Kicinski <kuba@kernel.org>, Andrew Lunn <andrew@lunn.ch>,
	Florian Fainelli <f.fainelli@gmail.com>,
	Vivien Didelot <vivien.didelot@gmail.com>,
	Roopa Prabhu <roopa@nvidia.com>
Subject: Re: [PATCH 1/3] net: bridge: Implement bridge flag local_receive
Date: Mon, 14 Mar 2022 18:33:32 +0200	[thread overview]
Message-ID: <Yi9uXGnrrOT13kiI@shredder> (raw)
In-Reply-To: <Yi9tgOQ32q2l2TxD@shredder>

On Mon, Mar 14, 2022 at 06:29:58PM +0200, Ido Schimmel wrote:
> On Wed, Mar 02, 2022 at 07:27:25AM +0100, Mattias Forsblad wrote:
> > On 2022-03-01 23:36, Nikolay Aleksandrov wrote:
> > > On 1 March 2022 17:43:27 CET, Ido Schimmel <idosch@idosch.org> wrote:
> > >> On Tue, Mar 01, 2022 at 01:31:02PM +0100, Mattias Forsblad wrote:
> > >>> diff --git a/net/bridge/br_input.c b/net/bridge/br_input.c
> > >>> index e0c13fcc50ed..5864b61157d3 100644
> > >>> --- a/net/bridge/br_input.c
> > >>> +++ b/net/bridge/br_input.c
> > >>> @@ -163,6 +163,9 @@ int br_handle_frame_finish(struct net *net, struct sock *sk, struct sk_buff *skb
> > >>>  		break;
> > >>>  	}
> > >>>  
> > >>> +	if (local_rcv && !br_opt_get(br, BROPT_LOCAL_RECEIVE))
> > >>> +		local_rcv = false;
> > >>> +
> > >>
> > >> I don't think the description in the commit message is accurate:
> > >> "packets received on bridge ports will not be forwarded up". From the
> > >> code it seems that if packets hit a local FDB entry, then they will be
> > >> "forwarded up". Instead, it seems that packets will not be flooded
> > >> towards the bridge. In which case, why not maintain the same granularity
> > >> we have for the rest of the ports and split this into unicast /
> > >> multicast / broadcast?
> > >>
> > > 
> > > Exactly my first thought - why not implement the same control for the bridge?
> > > Also try to minimize the fast-path hit, you can keep the needed changes 
> > > localized only to the cases where they are needed.
> > > I'll send a few more comments in a reply to the patch.
> > > 
> > 
> > Soo, if I understand you correctly, you want to have three different options?
> > local_receive_unicast
> > local_receive_multicast
> > local_receive_broadcast
> 
> My understanding of the feature is that you want to prevent flooding
> towards the bridge. In which case, it makes sense to keep the same
> granularity as for regular bridge ports and also name the options
> similarly. We already have several options that are applicable to both
> the bridge and bridge ports (e.g., 'mcast_router').
> 
> I suggest:
> 
> $ ip link help bridge
> Usage: ... bridge [ fdb_flush ]
>                   ...
>                   [ flood {on | off} ]
>                   [ mcast_flood {on | off} ]
>                   [ bcast_flood {on | off} ]
> 
> This is consistent with "bridge_slave".

And please add a selftest. See this commit for reference:
https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next.git/commit/?id=b2b681a412517bf477238de62b1d227361fa04fe

It should allow you to test both the software data path (using veth
pairs) and the hardware data path (using physical loopbacks).