Re: [Bridge] Re :Re: Re :Re: Bridging LACP (802.3ad) frames not working

Re: [Bridge] Re :Re: Re :Re: Bridging LACP (802.3ad) frames not working

[Jean-Michel Hautbois]

2010/1/13 Stephen Hemminger <shemminger@vyatta.com>:
> On Tue, 12 Jan 2010 17:04:48 -0500
> Ross Vandegrift <ross@kallisti.us> wrote:
>
>> On Tue, Jan 12, 2010 at 09:40:30PM +0000, jhautbois@gmail.com wrote:
>> > This is exactly the problem.
>> > But, sounds like it is not possible... ?
>>
>> You could run a custom version of the bridge driver to enable bridging
>> of frames sent to the bridge-management MAC addresses.  Some folks
>> have talked about doing similar things to enable bridging of STP.
>>
>> Doing that with STP makes a bit more sense to me (since there are
>> valid networks that could be constructed that way).  But you'll be
>> breaking a pretty fundamental assumption of LACP....
>
> Recent kernels will forward link local frames as long as STP is disabled.
> Older kernels will not.
>
> The assumption is that if STP is enabled, then it will be managing
> control plane.

Well, I have a 2.6.31 kernel. And LACP is not forwarded, even when STP
is disabled.
I don't understand what you mean ?

JM

Re: [Bridge] Re :Re: Re :Re: Bridging LACP (802.3ad) frames not working

[Stephen Hemminger]

On Wed, 13 Jan 2010 11:29:55 +0100
Jean-Michel Hautbois <jhautbois@gmail.com> wrote:

> 2010/1/13 Stephen Hemminger <shemminger@vyatta.com>:
> > On Tue, 12 Jan 2010 17:04:48 -0500
> > Ross Vandegrift <ross@kallisti.us> wrote:
> >
> >> On Tue, Jan 12, 2010 at 09:40:30PM +0000, jhautbois@gmail.com wrote:
> >> > This is exactly the problem.
> >> > But, sounds like it is not possible... ?
> >>
> >> You could run a custom version of the bridge driver to enable bridging
> >> of frames sent to the bridge-management MAC addresses.  Some folks
> >> have talked about doing similar things to enable bridging of STP.
> >>
> >> Doing that with STP makes a bit more sense to me (since there are
> >> valid networks that could be constructed that way).  But you'll be
> >> breaking a pretty fundamental assumption of LACP....
> >
> > Recent kernels will forward link local frames as long as STP is disabled.
> > Older kernels will not.
> >
> > The assumption is that if STP is enabled, then it will be managing
> > control plane.
> 
> Well, I have a 2.6.31 kernel. And LACP is not forwarded, even when STP
> is disabled.
> I don't understand what you mean ?

I meant they get processed locally, before that they were dropped.

The best solution is to write an ebtables rule to forward them.

[Bridge] Re :Re: Re :Re: Re :Re: Bridging LACP (802.3ad) frames not working

[jhautbois]

[-- Attachment #1: Type: text/plain, Size: 407 bytes --]

(snip)
> I meant they get processed locally, before that they were dropped.

> The best solution is to write an ebtables rule to forward them.

Mmh, I can understand that, but I can't see which kind of rule would do  
that.
Something like :
ebtables -A FORWARDING -p 0x8809 -j ACCEPT ??

But I think this will not do anything, because it is forwarding by default.
Or there is something I can't see :-).

JM

[-- Attachment #2: Type: text/html, Size: 484 bytes --]

Re: [Bridge] Re :Re: Re :Re: Re :Re: Bridging LACP (802.3ad) frames not working

[Stephen Hemminger]

On Wed, 13 Jan 2010 16:47:14 +0000
jhautbois@gmail.com wrote:

> (snip)
> > I meant they get processed locally, before that they were dropped.
> 
> > The best solution is to write an ebtables rule to forward them.
> 
> Mmh, I can understand that, but I can't see which kind of rule would do  
> that.
> Something like :
> ebtables -A FORWARDING -p 0x8809 -j ACCEPT ??
> 
> But I think this will not do anything, because it is forwarding by default.
> Or there is something I can't see :-).
> 

You need to put the rule on the local input not forwarding chain.

Re: [Bridge] Re :Re: Re :Re: Bridging LACP (802.3ad) frames not working

[richardvoigt]

On Tue, Jan 12, 2010 at 4:04 PM, Ross Vandegrift <ross@kallisti.us> wrote:
> On Tue, Jan 12, 2010 at 09:40:30PM +0000, jhautbois@gmail.com wrote:
>> This is exactly the problem.
>> But, sounds like it is not possible... ?
>
> You could run a custom version of the bridge driver to enable bridging
> of frames sent to the bridge-management MAC addresses.  Some folks
> have talked about doing similar things to enable bridging of STP.
>
> Doing that with STP makes a bit more sense to me (since there are
> valid networks that could be constructed that way).  But you'll be
> breaking a pretty fundamental assumption of LACP....

LACP is between peers, not to the nearest connected device (generally
an ethernet cable).  As long as the intermediate link acts just like a
wire and passes everything, LACP shouldn't care.  And there are some
reasonable cases to want a Linux box to look like a piece of cable
(e.g. wiretap, timed lockdown, satellite network simulator which
inserts delay and errors, etc.) to the surrounding network.  Of
course, whether seeing just one link out of an aggregation bundle is
useful is debatable, but Linux ought to be able to support it.  I've
been bitten before by adding a new switch into an STP setup and having
it eat STP packets even though STP processing was disabled.  So count
me as another vote for (at least the possibility of) layer 1-esque
transparent bridging.

>
> --
> Ross Vandegrift
> ross@kallisti.us
>
> "If the fight gets hot, the songs get hotter.  If the going gets tough,
> the songs get tougher."
>        --Woody Guthrie
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.9 (GNU/Linux)
>
> iEYEARECAAYFAktM8gAACgkQMlMoONfO+HC6UwCgkLsC/ZE4ovcxX1e+A/NV3ELz
> Rp4AoIFyCMtgrMYkxK2k81b9M3n6/v0U
> =A5kn
> -----END PGP SIGNATURE-----
>
> _______________________________________________
> Bridge mailing list
> Bridge@lists.linux-foundation.org
> https://lists.linux-foundation.org/mailman/listinfo/bridge
>

Re: [Bridge] Re :Re: Re :Re: Bridging LACP (802.3ad) frames not working

[Jean-Michel Hautbois]

2010/1/12 richardvoigt@gmail.com <richardvoigt@gmail.com>:
> On Tue, Jan 12, 2010 at 4:04 PM, Ross Vandegrift <ross@kallisti.us> wrote:
>> On Tue, Jan 12, 2010 at 09:40:30PM +0000, jhautbois@gmail.com wrote:
>>> This is exactly the problem.
>>> But, sounds like it is not possible... ?
>>
>> You could run a custom version of the bridge driver to enable bridging
>> of frames sent to the bridge-management MAC addresses.  Some folks
>> have talked about doing similar things to enable bridging of STP.
>>
>> Doing that with STP makes a bit more sense to me (since there are
>> valid networks that could be constructed that way).  But you'll be
>> breaking a pretty fundamental assumption of LACP....
>
> LACP is between peers, not to the nearest connected device (generally
> an ethernet cable).  As long as the intermediate link acts just like a
> wire and passes everything, LACP shouldn't care.  And there are some
> reasonable cases to want a Linux box to look like a piece of cable
> (e.g. wiretap, timed lockdown, satellite network simulator which
> inserts delay and errors, etc.) to the surrounding network.  Of
> course, whether seeing just one link out of an aggregation bundle is
> useful is debatable, but Linux ought to be able to support it.  I've
> been bitten before by adding a new switch into an STP setup and having
> it eat STP packets even though STP processing was disabled.  So count
> me as another vote for (at least the possibility of) layer 1-esque
> transparent bridging.

I think it would be interesting to have the ability to activate this
functionality.
AFAIK, this could be done in the br_handle_frame function, testing the
skb->protocol and comparing it with ETH_P_SLOW. This could be
conditionned by a flag, and this flag would be unabled by default.

Why is this a problem ?
If you tell this is not good, unless you know what you are doing,
where is the problem ?
This is a user problem, it is a functionality. If you misuse it, this
is not a kernel problem...

Best Regards,
JM

Re: [Bridge] Re :Re: Bridging LACP (802.3ad) frames not working

[Ross Vandegrift]

[-- Attachment #1: Type: text/plain, Size: 1208 bytes --]

On Tue, Jan 12, 2010 at 03:28:53PM +0000, jhautbois@gmail.com wrote:
> OK, I agree, it doesn't make sense, but I want to be transparent using  
> bridging. I can't explain my idea :).

You can't be transparent with respect to LACP if you're using ethernet
to bridge.  It doesn't make sense - LACP is negotiated between bridge
ports on a point-to-point ethernet interface.

> You mean I need to have four physical interfaces ?
> I can't understand how too accomplish this.
> Currently, I am doing :
> brctl addif eth1
> brctl addif eth2

No - you don't need four physical interfaces.  I was trying to
extrapolate what problem you might be trying to solve, and I think I
guessed wrong.  I imagined that you needed a multi-interface bundle
through a Linux bridge.

If you just want your box to speak LACP to the adjacent switches, you
could easily make bond interfaces with single ethernet interfaces.
But it sounds like you don't want to speak LACP to the switches, you
want to pass it as if you were a layer 1 device.

Ross

-- 
Ross Vandegrift
ross@kallisti.us

"If the fight gets hot, the songs get hotter.  If the going gets tough,
the songs get tougher."
	--Woody Guthrie

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 197 bytes --]

[Bridge] Re :Re: Re :Re: Bridging LACP (802.3ad) frames not working

[jhautbois]

[-- Attachment #1: Type: text/plain, Size: 1250 bytes --]

Le , Ross Vandegrift <ross@kallisti.us> a écrit :
> On Tue, Jan 12, 2010 at 03:28:53PM +0000, jhautbois@gmail.com wrote:

> > OK, I agree, it doesn't make sense, but I want to be transparent using

> > bridging. I can't explain my idea :).



> You can't be transparent with respect to LACP if you're using ethernet

> to bridge. It doesn't make sense - LACP is negotiated between bridge

> ports on a point-to-point ethernet interface.



> > You mean I need to have four physical interfaces ?

> > I can't understand how too accomplish this.

> > Currently, I am doing :

> > brctl addif eth1

> > brctl addif eth2



> No - you don't need four physical interfaces. I was trying to

> extrapolate what problem you might be trying to solve, and I think I

> guessed wrong. I imagined that you needed a multi-interface bundle

> through a Linux bridge.



> If you just want your box to speak LACP to the adjacent switches, you

> could easily make bond interfaces with single ethernet interfaces.

> But it sounds like you don't want to speak LACP to the switches, you

> want to pass it as if you were a layer 1 device.


This is exactly the problem.
But, sounds like it is not possible... ?

JM

[-- Attachment #2: Type: text/html, Size: 1706 bytes --]

Re: [Bridge] Re :Re: Re :Re: Bridging LACP (802.3ad) frames not working

[Ross Vandegrift]

[-- Attachment #1: Type: text/plain, Size: 706 bytes --]

On Tue, Jan 12, 2010 at 09:40:30PM +0000, jhautbois@gmail.com wrote:
> This is exactly the problem.
> But, sounds like it is not possible... ?

You could run a custom version of the bridge driver to enable bridging
of frames sent to the bridge-management MAC addresses.  Some folks
have talked about doing similar things to enable bridging of STP.

Doing that with STP makes a bit more sense to me (since there are
valid networks that could be constructed that way).  But you'll be
breaking a pretty fundamental assumption of LACP....

-- 
Ross Vandegrift
ross@kallisti.us

"If the fight gets hot, the songs get hotter.  If the going gets tough,
the songs get tougher."
	--Woody Guthrie

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 197 bytes --]

Re: [Bridge] Re :Re: Re :Re: Bridging LACP (802.3ad) frames not working

[Stephen Hemminger]

On Tue, 12 Jan 2010 17:04:48 -0500
Ross Vandegrift <ross@kallisti.us> wrote:

> On Tue, Jan 12, 2010 at 09:40:30PM +0000, jhautbois@gmail.com wrote:
> > This is exactly the problem.
> > But, sounds like it is not possible... ?
> 
> You could run a custom version of the bridge driver to enable bridging
> of frames sent to the bridge-management MAC addresses.  Some folks
> have talked about doing similar things to enable bridging of STP.
> 
> Doing that with STP makes a bit more sense to me (since there are
> valid networks that could be constructed that way).  But you'll be
> breaking a pretty fundamental assumption of LACP....

Recent kernels will forward link local frames as long as STP is disabled.
Older kernels will not.

The assumption is that if STP is enabled, then it will be managing
control plane.

Author: Stephen Hemminger <shemminger@vyatta.com>  2009-05-14 23:10:13
Committer: David S. Miller <davem@davemloft.net>  2009-05-17 21:12:54
Parent: 43aa1920117801fe9ae3d1fad886b62511e09bee (bridge: handle process all link-local frames)
Branches: master, remotes/origin/master
Follows: v2.6.30-rc6
Precedes: v2.6.30-rc7

    bridge: relay bridge multicast pkgs if !STP
    
    Currently the bridge catches all STP packets; even if STP is turned
    off.  This prevents other systems (which do have STP turned on)
    from being able to detect loops in the network.
    
    With this patch, if STP is off, then any packet sent to the STP
    multicast group address is forwarded to all ports.
    
    Based on earlier patch by Joakim Tjernlund with changes
    to go through forwarding (not local chain), and optimization
    that only last octet needs to be checked.
    
    Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
    Signed-off-by: David S. Miller <davem@davemloft.net>


--