Thomas, MingAnn,

Sorry for the late response.  See inline.

On Wed, Mar 6, 2013 at 8:38 PM, Thomas Pedersen <[email protected]> wrote:
> On Wed, Mar 6, 2013 at 4:59 PM, MingAnn Ng <[email protected]> wrote:
>>  Hi Thomas and Yeoh,
>>
>>
>> I have successfully compile compat-driver-3.8-1 and run it on my system. I'm
>> having 3 nodes near to each other and the signal strength  between them  are
>> about 45dBm to 65dBm. From a particular node, I continuously ping the other
>> two nodes for more than 14hours. Ping result to one of the nodes has no
>> duplication and packet loss. Ping result for the other node are: out of
>> 50453 packet received, there are 590 duplicated without any packet loss.
>
> It is possible that this will happen if the path selection algorithm
> determines a multi-hop path is in fact "better". This bit of code in
> net/mac80211/rx.c:
>
>         /* Frame has reached destination.  Don't forward */
>         if (!is_multicast_ether_addr(hdr->addr1) &&
>             ether_addr_equal(sdata->vif.addr, hdr->addr3))
>                 return RX_CONTINUE;
>
> will accept frames as long as the DA matches the local address. Then
> the receiving node will get both the original and forwarded frame (if
> it is in range, of course). Maybe this is an argument for only
> matching against TA? Except IEEE 802.11-2012 9.32.4.2(b) actually
> states that this is the right address to check.

I'm afraid I don't understand your explanation.  Can you draw a simple
message sequence chart that shows the scenario where a node would
receive duplicate unicast frames?  Have we been able to replicate?

MingAnn, do you have a sniffer capture that we can use to validate
Thomas' hypothesis?

I would only recommend introducing the recently uncast cache Yeoh is
proposing if we understand the root cause why duplicate unicast frames
occur.

Cheers,

Javier

-- 
Javier Cardona
cozybit Inc.
http://www.cozybit.com
_______________________________________________
Devel mailing list
[email protected]
http://lists.open80211s.org/cgi-bin/mailman/listinfo/devel

Reply via email to