Linux Ethernet Bridging - Dec 2010 - [Bridge] bridge problem with with ICMP fragment reassembly

Hi,

I have a bridge (br0) with two enslaved interfaces (int0 and ext0). From the
ext0 side, with a windows device, I ping with packet size 1932:

ping -n 1 -l 1932 10.64.160.94

ping request and reply is bridged.

Now if I do the same with packet size 1933:

ping -n 1 -l 1933 10.64.160.94

the request never seems to make it to the bridge interface

If I sniff the ping from the windows device on the ext0 side, I see two frames
sent, request + fragment, as expected.
In the working case, if I sniff ext0 on the bridge device I see the same two
frames received.
In the working case, if I sniff br0 on the bridge I see one frame (I assume the
bridge reassembles the request + fragment).
In the non working case,  if I sniff ext0 on the bridge device I see the same
two frames received.
In the non working case, if I sniff br0 I see nothing.

It seems there is a bug with bridging and reassembly of  packets or fragments of
a certain size, and am wondering If there is a patch.

The MTU of all interfaces is 1500.
I am using connection tracking.
The problem is seen on kernels 2.6.11 and 2.6.24

Any help would be much appreciated.

Regards,
Shawn





-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.linux-foundation.org/pipermail/bridge/attachments/20101202/616797db/attachment.html

Le 02/12/2010 23:19, Shawn Delaney a ?crit :
> Hi,
> I have a bridge (br0) with two enslaved interfaces (int0 and ext0). From
> the ext0 side, with a windows device, I ping with packet size 1932:
> ping ?n 1 ?l 1932 10.64.160.94
> ping request and reply is bridged.
> Now if I do the same with packet size 1933:
> ping ?n 1 ?l 1933 10.64.160.94
> the request never seems to make it to the bridge interface
> If I sniff the ping from the windows device on the ext0 side, I see two
> frames sent, request + fragment, as expected.
> In the working case, if I sniff ext0 on the bridge device I see the same
> two frames received.
> In the working case, if I sniff br0 on the bridge I see one frame (I
> assume the bridge reassembles the request + fragment).
> In the non working case, if I sniff ext0 on the bridge device I see the
> same two frames received.
> In the non working case, if I sniff br0 I see nothing.
> It seems there is a bug with bridging and reassembly of packets or
> fragments of a certain size, and am wondering If there is a patch.
Bridges are not expected to change frames in any way. This is not specific to
the bridging function
of Linux.

No reassembly occurs at the bridge level. If one host send two frames, the
bridge will forward two
frames.

By the way, it means that the MTU for all interfaces of the bridge should be
identical.

	Nicolas.
> The MTU of all interfaces is 1500.
> I am using connection tracking.
> The problem is seen on kernels 2.6.11 and 2.6.24
> Any help would be much appreciated.
> Regards,
> Shawn
>
>
>
> _______________________________________________
> Bridge mailing list
> Bridge at lists.linux-foundation.org
> https://lists.linux-foundation.org/mailman/listinfo/bridge