[Bridge] Flushing MAC-tables(?)

[Bridge] Flushing MAC-tables(?)

[Dennis Borgmann]

Dear list-members!

I am using OpenWRT (www.openwrt.org) on two accesspoints. Within
OpenWRT, the wireless interface and the ethernet interface are bridged
to one interface (br0).

Those two accesspoints are standing quite far from one another. Their
WLAN-cells do not cross each other, meaning I have an area in between
the two accesspoints, where there is no accesspoint of the two
accessible: a dead spot.

Now I use two WLAN-telephones (Cisco 7920). I start calling one of the
phones inside the same radio cell and I answer this call. Now I start
moving away to the other accesspoint. As soon as I see myself assiciated
to the "new" accesspoint, I am able to hear my mate on the other phone
in the "old" cell, but he cannot hear me. This stays this way, until a
certain point of time is reached. In our tests, he has been able to hear
my voice recurring at :30 each minute. Take these examples:

05:40:20 I roamed to the "new" AP
05:40:30 my mate could hear me

05:43:35 I roamed to the "new" AP
05:44:30 my mate could hear me

05:47:03 I roamed to the "new" AP
05:47:30 my mate could hear me

...and so on.

So at a certain point of time, there seems to be a flushing of some
bridge-tables, but I don't know, how I could influence this behaviour.
Is this "flushing" somehow to be done manually? Or is there a timer,
which sets this behaviour?

I sniffed the WLAN a little while my tests have been done and the result
is, that the "old" accesspoint seems to still believe me in his radio
cell. So he sends out data for me all of the time, but since my
telephone has moved away, there is no more client to ACK these packages.
After the time mentioned before has passed, these packet are away and
the communication goes on the way it should. So I assume, there is a
timeout, that tells the AP to flush its old bridge-MAC tables of
clients, that he did not hear of since XX seconds...

Any help would be appreciated. Thanks a lot in advance,

Dennis

Re: [Bridge] Flushing MAC-tables(?)

[Francesco Dolcini]

Dennis Borgmann wrote:
> So at a certain point of time, there seems to be a flushing of some
> bridge-tables, but I don't know, how I could influence this behaviour.
> Is this "flushing" somehow to be done manually? Or is there a timer,
> which sets this behaviour?
in recent kernel (not in 2.4, and maybe only in recent 2.6) you can 
flush the FDB using sysfs:
/sys/class/net/%s/bridge/flush (whole bridge)
/sys/class/net/%s/brif/%s/flush (single interface)

If you want you can lower the ageing timer from the default 300sec using 
brcl setageing

Maybe you can just force your phone to register more frequently so the 
bridge will just learn the new port when you move.

Francesco

Re: [Bridge] Flushing MAC-tables(?)

[Dennis Borgmann]

Hi Francesco,

I already tried playing around with the ageing timer, but it did not
help me.

Since I am using Kernel 2.4: is there no way of flushing the FDB?

Thanks so far,
Dennis


Francesco Dolcini schrieb:
> Dennis Borgmann wrote:
>> So at a certain point of time, there seems to be a flushing of some
>> bridge-tables, but I don't know, how I could influence this behaviour.
>> Is this "flushing" somehow to be done manually? Or is there a timer,
>> which sets this behaviour?
> in recent kernel (not in 2.4, and maybe only in recent 2.6) you can
> flush the FDB using sysfs:
> /sys/class/net/%s/bridge/flush (whole bridge)
> /sys/class/net/%s/brif/%s/flush (single interface)
>
> If you want you can lower the ageing timer from the default 300sec
> using brcl setageing
>
> Maybe you can just force your phone to register more frequently so the
> bridge will just learn the new port when you move.
>
> Francesco
>

Re: [Bridge] Flushing MAC-tables(?)

[Mark S. Mathews]

There's a simpler solution: at association-time the wlan stack should  
emit (up to linux) a bcast frame w/ the source address of the  
associating station.  That frame will serve to notify the local and  
remote bridges of the move.

-Mark

Quoting Dennis Borgmann <dennis.borgmann@googlemail.com>:

> Hi Francesco,
>
> I already tried playing around with the ageing timer, but it did not
> help me.
>
> Since I am using Kernel 2.4: is there no way of flushing the FDB?
>
> Thanks so far,
> Dennis
>
>
> Francesco Dolcini schrieb:
>> Dennis Borgmann wrote:
>>> So at a certain point of time, there seems to be a flushing of some
>>> bridge-tables, but I don't know, how I could influence this behaviour.
>>> Is this "flushing" somehow to be done manually? Or is there a timer,
>>> which sets this behaviour?
>> in recent kernel (not in 2.4, and maybe only in recent 2.6) you can
>> flush the FDB using sysfs:
>> /sys/class/net/%s/bridge/flush (whole bridge)
>> /sys/class/net/%s/brif/%s/flush (single interface)
>>
>> If you want you can lower the ageing timer from the default 300sec
>> using brcl setageing
>>
>> Maybe you can just force your phone to register more frequently so the
>> bridge will just learn the new port when you move.
>>
>> Francesco
>>
>
> _______________________________________________
> Bridge mailing list
> Bridge@lists.linux-foundation.org
> https://lists.linux-foundation.org/mailman/listinfo/bridge
>
>



Mark S. Mathews

AbsoluteValue Systems      Web:    http://www.linux-wlan.com
721-D North Drive          e-mail: mark@linux-wlan.com
Melbourne, FL 32934        Phone:  321.259.0737
USA                        Fax:    321.259.0286

Re: [Bridge] Flushing MAC-tables(?)

[Stephen Hemminger]

On Fri, 28 Sep 2007 10:38:37 -0400
"Mark S. Mathews" <mark@linux-wlan.com> wrote:

> There's a simpler solution: at association-time the wlan stack should  
> emit (up to linux) a bcast frame w/ the source address of the  
> associating station.  That frame will serve to notify the local and  
> remote bridges of the move.
> 
> -Mark
> 
With 2.6 if you set the ageing time to 0 then no forwarding table
entry is made (it will always flood).  Or you could set it real
short time.

2.4 code is fossilized at this point and your limited to setting
a real short ageing time.  You can go down to 1 HZ value so that
should work.

-- 
Stephen Hemminger <shemminger@linux-foundation.org>

Re: [Bridge] Flushing MAC-tables(?)

[richardvoigt]

[-- Attachment #1: Type: text/plain, Size: 3474 bytes --]

On 9/28/07, Dennis Borgmann <dennis.borgmann@googlemail.com> wrote:
>
> Dear list-members!
>
> I am using OpenWRT (www.openwrt.org) on two accesspoints. Within
> OpenWRT, the wireless interface and the ethernet interface are bridged
> to one interface (br0).
>
> Those two accesspoints are standing quite far from one another. Their
> WLAN-cells do not cross each other, meaning I have an area in between
> the two accesspoints, where there is no accesspoint of the two
> accessible: a dead spot.
>
> Now I use two WLAN-telephones (Cisco 7920). I start calling one of the
> phones inside the same radio cell and I answer this call. Now I start
> moving away to the other accesspoint. As soon as I see myself assiciated
> to the "new" accesspoint, I am able to hear my mate on the other phone
> in the "old" cell, but he cannot hear me. This stays this way, until a



This seems not to be a bridging problem then, since your pal didn't move to
a new access point, his information should be correct in all bridges (either
correct interface or no entry).  Therefore packets sent to him should be
correctly delivered.  OTOH his outgoing packets could be blocked by his
access point as long as it thinks you are locally attached (the first packet
coming in via the other access point should disillusion it without need for
a timer).

The behavior you are seeing is opposite, therefore I think you have
misidentified the cause.  Can you use VOIP software on a real computer, and
run packet capture software?  It might be very informative to have a
connection phone <-> computer and also computer <-> computer in order to see
if any packets are actually being missed, and in which direction.

Most likely your phone clears the routing table when you dissociate from the
other access point, and  the routing table isn't rebuilt until dhcp runs at
the new location.  This affects outgoing packets more than incoming, since
your adapter still may honor packets destined to your MAC address and
broadcast within its range.  Of course it may not be the routing table
either.


certain point of time is reached. In our tests, he has been able to hear
> my voice recurring at :30 each minute. Take these examples:
>
> 05:40:20 I roamed to the "new" AP
> 05:40:30 my mate could hear me
>
> 05:43:35 I roamed to the "new" AP
> 05:44:30 my mate could hear me
>
> 05:47:03 I roamed to the "new" AP
> 05:47:30 my mate could hear me
>
> ...and so on.
>
> So at a certain point of time, there seems to be a flushing of some
> bridge-tables, but I don't know, how I could influence this behaviour.
> Is this "flushing" somehow to be done manually? Or is there a timer,
> which sets this behaviour?
>
> I sniffed the WLAN a little while my tests have been done and the result
> is, that the "old" accesspoint seems to still believe me in his radio
> cell. So he sends out data for me all of the time, but since my
> telephone has moved away, there is no more client to ACK these packages.
> After the time mentioned before has passed, these packet are away and
> the communication goes on the way it should. So I assume, there is a
> timeout, that tells the AP to flush its old bridge-MAC tables of
> clients, that he did not hear of since XX seconds...
>
> Any help would be appreciated. Thanks a lot in advance,
>
> Dennis
> _______________________________________________
> Bridge mailing list
> Bridge@lists.linux-foundation.org
> https://lists.linux-foundation.org/mailman/listinfo/bridge
>

[-- Attachment #2: Type: text/html, Size: 4307 bytes --]

Re: [Bridge] Flushing MAC-tables(?)

[Mark S. Mathews]

Good catch Richard, I forgot about the stale association problem.

Actually, my note about having the 'new AP' wlan stack generate a  
bcast frame in reponse to the association event should take care of  
this problem as well.  When the old AP's wlan stack sees a frame  
coming in from above (i.e. off the wire) with a source address equal  
to one of its associated stations, it _should_ immediately disassoc  
that station thus breaking the mac-level STA-to-STA frame path.

I think this 'send a bcast to the wire upon assoc' behavior is listed  
in a recommended-practices somewhere (been a long time).  It's  
basically a trick to enforce the 'STA shall be associated to  
one-and-only-one BSS within an ESS' rule.

-Mark

Quoting "richardvoigt@gmail.com" <richardvoigt@gmail.com>:

> On 9/28/07, Dennis Borgmann <dennis.borgmann@googlemail.com> wrote:
>>
>> Dear list-members!
>>
>> I am using OpenWRT (www.openwrt.org) on two accesspoints. Within
>> OpenWRT, the wireless interface and the ethernet interface are bridged
>> to one interface (br0).
>>
>> Those two accesspoints are standing quite far from one another. Their
>> WLAN-cells do not cross each other, meaning I have an area in between
>> the two accesspoints, where there is no accesspoint of the two
>> accessible: a dead spot.
>>
>> Now I use two WLAN-telephones (Cisco 7920). I start calling one of the
>> phones inside the same radio cell and I answer this call. Now I start
>> moving away to the other accesspoint. As soon as I see myself assiciated
>> to the "new" accesspoint, I am able to hear my mate on the other phone
>> in the "old" cell, but he cannot hear me. This stays this way, until a
>
> This seems not to be a bridging problem then, since your pal didn't move to
> a new access point, his information should be correct in all bridges (either
> correct interface or no entry).  Therefore packets sent to him should be
> correctly delivered.  OTOH his outgoing packets could be blocked by his
> access point as long as it thinks you are locally attached (the first packet
> coming in via the other access point should disillusion it without need for
> a timer).
>
> The behavior you are seeing is opposite, therefore I think you have
> misidentified the cause.  Can you use VOIP software on a real computer, and
> run packet capture software?  It might be very informative to have a
> connection phone <-> computer and also computer <-> computer in order to see
> if any packets are actually being missed, and in which direction.
>
> Most likely your phone clears the routing table when you dissociate from the
> other access point, and  the routing table isn't rebuilt until dhcp runs at
> the new location.  This affects outgoing packets more than incoming, since
> your adapter still may honor packets destined to your MAC address and
> broadcast within its range.  Of course it may not be the routing table
> either.
>
>
> certain point of time is reached. In our tests, he has been able to hear
>> my voice recurring at :30 each minute. Take these examples:
>>
>> 05:40:20 I roamed to the "new" AP
>> 05:40:30 my mate could hear me
>>
>> 05:43:35 I roamed to the "new" AP
>> 05:44:30 my mate could hear me
>>
>> 05:47:03 I roamed to the "new" AP
>> 05:47:30 my mate could hear me
>>
>> ...and so on.
>>
>> So at a certain point of time, there seems to be a flushing of some
>> bridge-tables, but I don't know, how I could influence this behaviour.
>> Is this "flushing" somehow to be done manually? Or is there a timer,
>> which sets this behaviour?
>>
>> I sniffed the WLAN a little while my tests have been done and the result
>> is, that the "old" accesspoint seems to still believe me in his radio
>> cell. So he sends out data for me all of the time, but since my
>> telephone has moved away, there is no more client to ACK these packages.
>> After the time mentioned before has passed, these packet are away and
>> the communication goes on the way it should. So I assume, there is a
>> timeout, that tells the AP to flush its old bridge-MAC tables of
>> clients, that he did not hear of since XX seconds...
>>
>> Any help would be appreciated. Thanks a lot in advance,
>>
>> Dennis
>> _______________________________________________
>> Bridge mailing list
>> Bridge@lists.linux-foundation.org
>> https://lists.linux-foundation.org/mailman/listinfo/bridge
>>
>



Mark S. Mathews

AbsoluteValue Systems      Web:    http://www.linux-wlan.com
721-D North Drive          e-mail: mark@linux-wlan.com
Melbourne, FL 32934        Phone:  321.259.0737
USA                        Fax:    321.259.0286

Re: [Bridge] Flushing MAC-tables(?)

[richardvoigt]

[-- Attachment #1: Type: text/plain, Size: 1292 bytes --]

On 9/28/07, Mark S. Mathews <mark@linux-wlan.com> wrote:
>
> Good catch Richard, I forgot about the stale association problem.
>
> Actually, my note about having the 'new AP' wlan stack generate a
> bcast frame in reponse to the association event should take care of
> this problem as well.  When the old AP's wlan stack sees a frame
> coming in from above (i.e. off the wire) with a source address equal
> to one of its associated stations, it _should_ immediately disassoc
> that station thus breaking the mac-level STA-to-STA frame path.
>
> I think this 'send a bcast to the wire upon assoc' behavior is listed
> in a recommended-practices somewhere (been a long time).  It's
> basically a trick to enforce the 'STA shall be associated to
> one-and-only-one BSS within an ESS' rule.


Most stations will send a broadcast upon association even without help from
the AP (DHCP-REQUEST for example).

In this case, the station that moved can still hear the other station, which
makes me believe there is not a stale association, or stale MAC entries,
because the stationary node obviously found the new location of the mobile
node correctly.  It's the mobile node that can't transmit.  That's backwards
from the breakage you'd get from a stale association/stale bridge learn
table.


-Mark
>

[-- Attachment #2: Type: text/html, Size: 1832 bytes --]

Re: [Bridge] Flushing MAC-tables(?)

[Mark S. Mathews]

Quoting "richardvoigt@gmail.com" <richardvoigt@gmail.com>:

> On 9/28/07, Mark S. Mathews <mark@linux-wlan.com> wrote:
>>
>> Good catch Richard, I forgot about the stale association problem.
>>
>> Actually, my note about having the 'new AP' wlan stack generate a
>> bcast frame in reponse to the association event should take care of
>> this problem as well.  When the old AP's wlan stack sees a frame
>> coming in from above (i.e. off the wire) with a source address equal
>> to one of its associated stations, it _should_ immediately disassoc
>> that station thus breaking the mac-level STA-to-STA frame path.
>>
>> I think this 'send a bcast to the wire upon assoc' behavior is listed
>> in a recommended-practices somewhere (been a long time).  It's
>> basically a trick to enforce the 'STA shall be associated to
>> one-and-only-one BSS within an ESS' rule.
>
> Most stations will send a broadcast upon association even without help from
> the AP (DHCP-REQUEST for example).
>
> In this case, the station that moved can still hear the other station, which
> makes me believe there is not a stale association, or stale MAC entries,
> because the stationary node obviously found the new location of the mobile
> node correctly.  It's the mobile node that can't transmit.  That's backwards
> from the breakage you'd get from a stale association/stale bridge learn
> table.

Too true.  I had to reread the problem description and you're quite  
correct, sorry for the confusion.

The 'most stations' note though...;-) we've learned the hard way that  
there are plenty of mac/driver/host combinations that only issue DHCP  
requests on ESS transitions.

-Mark


Mark S. Mathews

AbsoluteValue Systems      Web:    http://www.linux-wlan.com
721-D North Drive          e-mail: mark@linux-wlan.com
Melbourne, FL 32934        Phone:  321.259.0737
USA                        Fax:    321.259.0286

Re: [Bridge] Flushing MAC-tables(?)

[richardvoigt]

[-- Attachment #1: Type: text/plain, Size: 302 bytes --]

> The 'most stations' note though...;-) we've learned the hard way that
> there are plenty of mac/driver/host combinations that only issue DHCP
> requests on ESS transitions.
>
>
Didn't think of that.  I guess seamless roaming would require that handoff
isn't considered as a disconnect and reconnect.

[-- Attachment #2: Type: text/html, Size: 497 bytes --]