[Serdev] SIP UDP / dont fragment issue

[Martin Koenig]

Hello Maxim,

it is implemented via ICMP type 3 (dest. unreachable) code 4 
(fragmentation needed), and the MTU of the next hop specified in this 
ICMP message.

It is very correct to set the dont fragment bit here, because the 
message is already fragmented and an endpoint would not be able to 
aggregate these further fragmented messages.

Thats why the packet will be dropped at the router and ICMP dest. 
unreachable must be responded.

Regards,
Martin

Maxim Sobolev wrote:

>No, he is not quite correct. We are talking about UDP, not TCP.
>UDP doesn't have path MTU discovery mechanism, so that kernel
>should never set DF flag on UDP packets.
>
>-Maxim
>
>On Wed, Jun 01, 2005 at 11:22:53AM +0200, Martin Koenig wrote:
>  
>
>>Hello Ben,
>>
>>you are correct, of course. The problem originates from a Router on the 
>>way to the destination (Japan) that drops / does not reply with the 
>>correct ICMP destination unreachable.
>>
>>Sorry for the disturbance,
>>
>>regards,
>>Martin
>>
>>Benjamin Black wrote:
>>
>>    
>>
>>>That's the kernel implementing TCP path MTU discovery.  The first hop  
>>>before the low MTU link should be sending back an ICMP message  
>>>(fragmentation required/DF set) that includes the MTU of the outgoing  
>>>link.  So, the problem is not really SER, which doesn't control  
>>>setting the DF bit, but the router in front of the slow link that  
>>>isn't sending back the ICMP messages needed for PMTUD to function (or  
>>>perhaps you are filtering them).
>>>
>>>Try disabling PMTUD (I forget if this is the proper way under current  
>>>Linux systems):
>>>
>>>   echo 1 >/proc/sys/net/ipv4/ip_no_pmtu_disc
>>>
>>>
>>>
>>>Ben
>>>
>>>On May 28, 2005, at 5:22 PM, Martin Koenig wrote:
>>>
>>>      
>>>
>>>>Hello,
>>>>
>>>>
>>>>        
>>>>
>>>>>Martin Koenig wrote:
>>>>>
>>>>>
>>>>>          
>>>>>
>>>>>>Hello,
>>>>>>
>>>>>>we are having issues piping SIP to asia via a low MTU transit  
>>>>>>point. The problem is that SER sets the "dont fragement" bit for  
>>>>>>outgoing messages, which will then be dropped at the low MTU  
>>>>>>transit because it may not fragment the too big packets.
>>>>>>
>>>>>>A quick search on the archive did not put up anything useful. Has  
>>>>>>anyone ever had similar problems?
>>>>>>
>>>>>>            
>>>>>>
>>>>>I don't think that SER is responsible for that. Actually, if my  
>>>>>memory serves BSD sockets API doesn't provide any portable way for  
>>>>>the application to set DF flag. I suspect that either your  
>>>>>operating system does it, or the problem is elsewhere.
>>>>>
>>>>>Setting DF on UDP packets is stupid thing to do since UDP doesn't  
>>>>>have any path MTU discovery mechanism. Therefore, sending UDP with  
>>>>>DF is a very good way to shoot yourself in the foot.
>>>>>
>>>>>          
>>>>>
>>>>This System is RH9. I verified with tcpdump that outgoing UDP has  
>>>>dont fragment bit set. Argh.
>>>>
>>>>Is there anything that can be done about this issue except lowering  
>>>>the MTU of the corresponding NIC?
>>>>
>>>>Regards,
>>>>Martin
>>>>
>>>>_______________________________________________
>>>>Serdev mailing list
>>>>serdev at lists.iptel.org
>>>>http://lists.iptel.org/mailman/listinfo/serdev
>>>>
>>>>
>>>>        
>>>>
>>>_______________________________________________
>>>Serdev mailing list
>>>serdev at lists.iptel.org
>>>http://lists.iptel.org/mailman/listinfo/serdev
>>>      
>>>
>>
>>    
>>