Mesh freeze on node shutdown

[Francesco]

I have four nodes running ubuntu linux, kernel 2.6.27-7, atheros chipset
wifi cards (ath5k driver dated 20-10-2009).
Mesh points share the same mesh_id, same channel, and they are spread in 4
different rooms in the lab.
I enabled 802.11s as shown in the howto.
Name A B C D the mesh points, A is generating UDP traffic, D is configurated
as sink node.
To force multi hop routes I used the "iw...plink_action block" command on
first tests..then as I experimented the problem, I tried also to spread far
enough nodes A and D to force traffic between them being routed through one
intermediate  node (B or C)  and reach destination node D.

I see which node is used to forward traffic sniffing packets with wireshark
patched, the sniffer is running on another mesh node (name it E) with his
wifi interface in monitor mode, set on the same channel as the mesh.


To reproduce the bug, just start UDP traffic from A to D, look in the
sniffer which is the node forwarding packets at a date time and kill it.
Please be careful to kille the correct node, i faced also fast path hopping
between intermediate nodes, and sometimes it is really fast to change route
so...be sure you kill the right node on time.

By kill I mean fire "$ sudo halt" on that machine.
You can also remove ath5k driver or ifconfig down the mesh interface..sould
not change much.

Analyzing wireshark traces i see a bunch of retransmission sent by the
source node A for a couple of seconds starting at the same time the
intermediate node is killed. But no "action frames -Mesh path request -
response"  being generated from A and this is weird.
After that time node A freeze, and same does the receiver. Sometimes also
intermediate nodes freeze.

Actually i am not sure if it is a 802.11s problem or a driver problem.

Another piece of information: i tried the same tests back in April,
everything worked fine.

Hope I made myself clear..and this could help to debug something.



On Tue, Dec 1, 2009 at 1:41 AM, Rui Paulo <rpaulo at gmail.com> wrote:

> On 30 Nov 2009, at 18:45, Andrey Yurovsky wrote:
>
> > On Mon, Nov 30, 2009 at 10:29 AM, Cristian Nasuti
> > <nasuti.cristian at gmail.com> wrote:
> >> I think that I have the same problem.
> >>
> >> What I do is to configure 4 machine as mesh point and force paths (using
> the
> >> commands mpath mpath and new) for this topology:
> >>
> >> MP1 --- MP2 --- MP3 --- MP4
> >>
> >> Then I run iperf: suppose MP1 as client and MP2 as a server.
> >>
> >> All work fine, but if after a while I turn off one of the intermediate
> >> nodes, such as MP3 (with ifconfig MP3 down or just turning off the
> machine)
> >> after a couple of seconds client machine (MP1) freezing and needs
> rebooting.
> >
> > Thanks Cristian.  This seems to be a recent regression and we'll look in
> to it.
>
> I can't reproduce this problem. Everything works when I ifconfig down the
> intermediate node. No nodes crash and I'm able to resume the transmission if
> I ifconfig up the intermediate node.
>
> Can you both describe in more detail your setup?
>
> Thanks,
> --
> Rui Paulo
>
>


-- 
Francesco Cappuccio
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://open80211s.com/pipermail/devel/attachments/20091202/db90f434/attachment.html