Re: [tcpm] TCP zero window timeout?

[Mahesh Jethanandani <[e-mail]>]

Caitlin Bestler wrote:

>>Are u suggesting that the application query TCP to determine
>>that the connection is stuck in persist state and how much of
>>the data has been sent or not and so on? That sounds a little
>>roundabout and complicated...
>>    
>>
>
>And/or the application convey an application specific tolerance
>for such a state through a local interface.
>  
>
We agree that application can help TCP by suggesting how tolerant  it 
wants to be.

>If the behavior can be corrected through a local interface,
>and there is no IETF language barring or discouraging such
>an interface, then why is this something that TCPM should
>consider?
>  
>
Information of how long the connection has been stuck in persist state, 
how many retry probes have been made and whether each of them has 
resulted in a zero window is all in TCP. So TCP(m) has to be involved. 
Who actually issues the reset on the connection is a implementation 
level detail which we can get into once we agree that we have a DOS 
situation.

>The nature of the Denial of Service attack presented does
>not seem to require real-time defenses and definitely does
>not present a strong case for any network element modifying
>their "fast path" handling of packets.
>
If we accept that it is a DOS attack, then the problem needs to be 
addressed. Our suggestion only says that today's behavior of infinite 
TCP probes in persist state lends itself open to a DOS and so the probes 
should have the option to be capped with a timeout.

_______________________________________________
tcpm mailing list
[e-mail]
https://www1.ietf.org/mailman/listinfo/tcpm