Date:   Tue, 23 Aug 2022 09:48:17 +0300
From:   Ido Schimmel <idosch@...dia.com>
To:     netdev@...io-technology.com
Cc:     Vladimir Oltean <olteanv@...il.com>, davem@...emloft.net,
        kuba@...nel.org, netdev@...r.kernel.org,
        Andrew Lunn <andrew@...n.ch>,
        Vivien Didelot <vivien.didelot@...il.com>,
        Florian Fainelli <f.fainelli@...il.com>,
        Eric Dumazet <edumazet@...gle.com>,
        Paolo Abeni <pabeni@...hat.com>, Jiri Pirko <jiri@...nulli.us>,
        Ivan Vecera <ivecera@...hat.com>,
        Roopa Prabhu <roopa@...dia.com>,
        Nikolay Aleksandrov <razor@...ckwall.org>,
        Shuah Khan <shuah@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        linux-kernel@...r.kernel.org, bridge@...ts.linux-foundation.org,
        linux-kselftest@...r.kernel.org
Subject: Re: [PATCH v4 net-next 3/6] drivers: net: dsa: add locked fdb entry
 flag to drivers

On Mon, Aug 22, 2022 at 09:49:28AM +0200, netdev@...io-technology.com wrote:
> On 2022-08-22 07:40, Ido Schimmel wrote:
> > On Sun, Aug 21, 2022 at 03:43:04PM +0200, netdev@...io-technology.com
> > wrote:
> > 
> > I personally think that the mv88e6xxx semantics are very weird (e.g., no
> > roaming, traffic blackhole) and I don't want them to determine how the
> > feature works in the pure software bridge or other hardware
> > implementations. On the other hand, I understand your constraints and I
> > don't want to create a situation where user space is unable to
> > understand how the data path works from the bridge FDB dump with
> > mv88e6xxx.
> > 
> > My suggestion is to have mv88e6xxx report the "locked" entry to the
> > bridge driver with additional flags that describe its behavior in terms
> > of roaming, ageing and forwarding.
> > 
> > In terms of roaming, since in mv88e6xxx the entry can't roam you should
> > report the entry with the "sticky" flag.
> 
> As I am not familiar with roaming in this context, I need to know how the SW
> bridge should behave in this case.

I think I wasn't clear enough. The idea is to make the bridge compatible
with mv88e6xxx in a way that is discoverable by user space by having
mv88e6xxx add the locked entry with flags that describe the hardware
behavior. Therefore, it's not a matter of "how the SW bridge should
behave", but having it behave in a way that matches the offloaded data
path.

>From what I was able to understand from you, the "locked" entry cannot
roam at all in mv88e6xxx, which can be described by the "sticky" flag.

> In this I am assuming that roaming is regarding unauthorized entries.

Yes, talking about "locked" entries that are notified by mv88e6xxx to
the bridge.

> In this case, is the roaming only between locked ports or does the
> roaming include that the entry can move to a unlocked port, resulting
> in the locked flag getting removed?

Any two ports. If the "locked" entry in mv88e6xxx cannot move once
installed, then the "sticky" flag accurately describes it.

> 
> > In terms of ageing, since
> > mv88e6xxx is the one doing the ageing and not the bridge driver, report
> > the entry with the "extern_learn" flag.
> 
> Just for the record, I see that entries coming from the driver to the bridge
> will always have the "extern learn" flag set as can be seen from the
> SWITCHDEV_FDB_ADD_TO_BRIDGE events handling in br_switchdev_event() in br.c,
> which I think is the correct behavior.

Yes.

> 
> > In terms of forwarding, in
> > mv88e6xxx the entry discards all matching packets. We can introduce a
> > new FDB flag that instructs the entry to silently discard all matching
> > packets. Like we have with blackhole routes and nexthops.
> 
> Any suggestions to the name of this flag?

I'm not good at naming, but "blackhole" is at least consistent with what
we already have for routes and nexthop objects.

> 
> > 
> > I believe that the above suggestion allows you to fully describe how
> > these entries work in mv88e6xxx while keeping the bridge driver in sync
> > with complete visibility towards user space.
> > 
> > It also frees the pure software implementation from the constraints of
> > mv88e6xxx, allowing "locked" entries to behave like any other
> > dynamically learned entries modulo the fact that they cannot "unlock" a
> > locked port.
> > 
> > Yes, it does mean that user space will get a bit different behavior with
> > mv88e6xxx compared to a pure software solution, but a) It's only the
> > corner cases that act a bit differently. As a whole, the feature works
> > largely the same. b) User space has complete visibility to understand
> > the behavior of the offloaded data path.
> > 
> 
> > > 
> > > I will change it in iproute2 to:
> > > bridge link set dev DEV mab on|off
> > 
> > And s/BR_PORT_MACAUTH/BR_PORT_MAB/ ?
> 
> Sure, I will do that. :-)

Thanks

Powered by blists - more mailing lists