RE: [lwip-users] Dropping existing TCP connections to service new ones

[Art R.]

OR will someone someday 'fix' the MAX value, setting it to 255? Is it
intentional that 'MAX' is 127? Should TCP_PRIO_MAX be renamed to
TCP_PRIO_HALF_MAX (<smile>)?


Art R. wrote:
> 
> Setting to max+1 would probably work. I was thinking that the MAX value of
> 127 is the limit of the prio value (an int), but it's an unsigned 8 so 128
> would be ok.
> 
> 
> 
> Bill Auerbach wrote:
>> 
>> 
>>> -----Original Message-----
>>> From: address@hidden
>>> [mailto:address@hidden On
>>> Behalf
>>> Of Art R.
>>> Sent: Wednesday, February 20, 2008 2:04 PM
>>> To: address@hidden
>>> Subject: [lwip-users] Dropping existing TCP connections to service new
>>> ones
>>> 
>>> 
>>> The tcp.c module has function tcp_kill_prio which is used to kill
>>> existing
>>> TCP connections when a new connection is being attempted and there is an
>>> 'out of PCBs' situation. The comment in that function says "kill the
>>> oldest
>>> active connection that has lower priority than prio", but the test it
>>> performs is "...pcb->prio <= prio". This appears to allow a new
>>> connection
>>> to kill an existing one of the same priority.
>>> 
>>> The default case is that all pcbs will have the same priority
>>> ("normal"),
>>> so
>>> the oldest is killed.
>>> 
>>> Is this a bug? Should the code read "if (pcb->prio < prio && ..."  (less
>>> than instead of less or equal)?
>>> Or is it intentional?
>>> 
>>> What would be the best way to disable the killing of active connections?
>>> (Preferably without modifying the lwIP source code.)
>> 
>> Can't you call tcp_setprio with a value of TCP_PRIO_MAX+1 for each pcb?
>> 
>> Bill
>> 
>> 
>> 
>> 
>> _______________________________________________
>> lwip-users mailing list
>> address@hidden
>> http://lists.nongnu.org/mailman/listinfo/lwip-users
>> 
>> 
> 
> 

-- 
View this message in context: 
http://www.nabble.com/Dropping-existing-TCP-connections-to-service-new-ones-tp15596047p15598166.html
Sent from the lwip-users mailing list archive at Nabble.com.