Message-ID: <20151014174152.GB9216@colbert.mtl.com>
Date:	Wed, 14 Oct 2015 20:42:09 +0300
From:	Ido Schimmel <idosch@...lanox.com>
To:	Scott Feldman <sfeldma@...il.com>
CC:	Vivien Didelot <vivien.didelot@...oirfairelinux.com>,
	Nikolay Aleksandrov <razor@...ckwall.org>,
	Netdev <netdev@...r.kernel.org>,
	Jiří Pírko <jiri@...nulli.us>,
	"David S. Miller" <davem@...emloft.net>,
	Nikolay Aleksandrov <nikolay@...ulusnetworks.com>,
	Elad Raz <eladr@...lanox.com>
Subject: Re: [PATCH net-next] switchdev: enforce no pvid flag in vlan ranges

Wed, Oct 14, 2015 at 08:14:24PM IDT, sfeldma@...il.com wrote:
>On Wed, Oct 14, 2015 at 8:25 AM, Vivien Didelot
><vivien.didelot@...oirfairelinux.com> wrote:
>> On Oct. Wednesday 14 (42) 09:14 AM, Ido Schimmel wrote:
>>> Tue, Oct 13, 2015 at 05:32:26PM IDT, vivien.didelot@...oirfairelinux.com wrote:
>>> >On Oct. Tuesday 13 (42) 11:31 AM, Ido Schimmel wrote:
>>> >> Mon, Oct 12, 2015 at 08:36:25PM IDT, vivien.didelot@...oirfairelinux.com wrote:
>>> >> >Hi guys,
>>> >> >
>>> >> >On Oct. Monday 12 (42) 02:01 PM, Nikolay Aleksandrov wrote:
>>> >> >> From: Nikolay Aleksandrov <nikolay@...ulusnetworks.com>
>>> >> >>
>>> >> >> We shouldn't allow BRIDGE_VLAN_INFO_PVID flag in VLAN ranges.
>>> >> >>
>>> >> >> Signed-off-by: Nikolay Aleksandrov <nikolay@...ulusnetworks.com>
>>> >> >> ---
>>> >> >>  net/switchdev/switchdev.c | 3 +++
>>> >> >>  1 file changed, 3 insertions(+)
>>> >> >>
>>> >> >> diff --git a/net/switchdev/switchdev.c b/net/switchdev/switchdev.c
>>> >> >> index 6e4a4f9ad927..256c596de896 100644
>>> >> >> --- a/net/switchdev/switchdev.c
>>> >> >> +++ b/net/switchdev/switchdev.c
>>> >> >> @@ -720,6 +720,9 @@ static int switchdev_port_br_afspec(struct net_device *dev,
>>> >> >>                         if (vlan.vid_begin)
>>> >> >>                                 return -EINVAL;
>>> >> >>                         vlan.vid_begin = vinfo->vid;
>>> >> >> +                       /* don't allow range of pvids */
>>> >> >> +                       if (vlan.flags & BRIDGE_VLAN_INFO_PVID)
>>> >> >> +                               return -EINVAL;
>>> >> >>                 } else if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_END) {
>>> >> >>                         if (!vlan.vid_begin)
>>> >> >>                                 return -EINVAL;
>>> >> >> --
>>> >> >> 2.4.3
>>> >> >>
>>> >> >
>>> >> >Yes the patch looks good, but it is a minor check though. I hope the
>>> >> >subject of this thread is making sense.
>>> >> >
>>> >> >VLAN ranges seem to have been included for an UX purpose (so commands
>>> >> >look like Cisco IOS). We don't want to change any existing interface, so
>>> >> >we pushed that down to drivers, with the only valid reason that, maybe
>>> >> >one day, an hardware can be capable of programming a range on a per-port
>>> >> >basis.
>>> >> Hi,
>>> >>
>>> >> That's actually what we are doing in mlxsw. We can do up to 256 entries in
>>> >> one go. We've yet to submit this part.
>>> >
>>> >Perfect Ido, thanks for pointing this out! I'm OK with the range then.
>>> >
>>> >So there is now a very last question in my head for this, which is more
>>> >a matter of kernel design. Should the user be aware of such underlying
>>> >support? In other words, would it make sense to do this in a driver:
>>> >
>>> >    foo_port_vlan_add(struct net_device *dev,
>>> >                      struct switchdev_obj_port_vlan *vlan)
>>> >    {
>>> >        if (vlan->vid_begin != vlan->vid_end)
>>> >            return -ENOTSUPP; /* or something more relevant for user */
>>> >
>>> >        return foo_port_single_vlan_add(dev, vlan->vid_begin);
>>> >    }
>>> >
>>> >So drivers keep being simple, and we can easily propagate the fact that
>>> >one-or-all VLAN is not supportable, vs. the VLAN feature itself is not
>>> >implemented and must be done in software.
>>> I think that if you want to keep it simple, then Scott's advice from the
>>> previous thread is the most appropriate one. I believe the hardware you
>>> are using is simply not meant to support multiple 802.1Q bridges.
>>
>> You mean allowing only one Linux bridge over an hardware switch?
>>
>> It would for sure simplify how, as developers and users, we represent a
>> physical switch. But I am not sure how to achieve that and I don't have
>> strong opinions on this TBH.
>
>Hi Vivien, I think it's possible to keep switch ports on just one
>bridge if we do a little bit of work on the NETDEV_CHANGEUPPER
>notifier.  This will give you the driver-level control you want.  Do
>you have time to investigate?  The idea is:
>
>1) In your driver's handler for NETDEV_CHANGEUPPER, if switch port is
>being added to a second bridge,then return NOTIFY_BAD.  Your driver
>needs to track the bridge count.
>
>2) In __netdev_upper_dev_link(), check the return code from the
>call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, ...) call, and if
>NOTIFY_BAD, abort the linking operation (goto rollback_xxx).
>
Hi,

We are doing something similar in mlxsw (not upstream yet). Jiri
introduced PRE_CHANGEUPPER, which is called from the function you
mentioned, but before the linking operation (so that you don't need to
rollback).

If the notification is about a linking operation and the master is a
bridge different than the current one, then NOTIFY_BAD is returned.

Vivien, regarding your WAN interface question, this is something we
currently don't do. We don't even flood traffic from bridged ports
to CPU (although we can), as it can saturate the bus. Only control
traffic is supposed to go there.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists