LRO causing stretch ACK violations interacts badly with delayed ACKing

Julian Elischer julian at freebsd.org
Wed Oct 23 03:32:45 UTC 2013


On 10/22/13 4:01 PM, Andre Oppermann wrote:
> On 21.10.2013 22:58, Colin Percival wrote:
>> On 10/21/13 13:11, Andre Oppermann wrote:
>>> On 21.10.2013 21:57, Andre Oppermann wrote:
>>>> This is an excellent observation!  Our tcp doesn't know about LRO
>>>> and I prepared the mbuf header to carry information about the number
>>>> of merged LRO segments.  That's not done yet again.  However a small
>>>> heuristic in tcp_input looking for segment > mss should be 
>>>> sufficient
>>>> for now.  Let me have a look at patching it into a suitable place.
>>>
>>> Please check out the patch below.  Haven't tested it myself yet 
>>> though.
>>
>> Yes, this works:
>>> 00:00:00.000000 IP 10.148.229.78.24405 > 176.32.98.166.443: Flags 
>>> [S], seq 3220740500, win 65535, options [mss 1460,nop,wscale 
>>> 6,sackOK,TS val 350742 ecr 0], length 0
>>> 00:00:00.000613 IP 176.32.98.166.443 > 10.148.229.78.24405: Flags 
>>> [S.], seq 1783557911, ack 3220740501, win 8190, options [mss 
>>> 1460,nop,wscale 6], length 0
>>> 00:00:00.000657 IP 10.148.229.78.24405 > 176.32.98.166.443: Flags 
>>> [.], ack 1, win 1026, length 0
>>> 00:00:00.001842 IP 176.32.98.166.443 > 10.148.229.78.24405: Flags 
>>> [.], ack 1, win 127, length 0
>>> 00:00:00.032269 IP 10.148.229.78.24405 > 176.32.98.166.443: Flags 
>>> [P.], seq 1:318, ack 1, win 1026, length 317
>>> 00:00:00.033080 IP 176.32.98.166.443 > 10.148.229.78.24405: Flags 
>>> [.], ack 318, win 108, length 0
>>> 00:00:00.033115 IP 176.32.98.166.443 > 10.148.229.78.24405: Flags 
>>> [.], seq 1:4097, ack 318, win 108, length 4096
>>> 00:00:00.033129 IP 10.148.229.78.24405 > 176.32.98.166.443: Flags 
>>> [.], ack 4097, win 962, length 0
>>
>> Please commit this fix and get it merged for 10.0-RELEASE!
>
> I committed it to my staging branch pending review:
>  http://svnweb.freebsd.org/changeset/base/256879
>
> It should go into HEAD later this week and the MFC to 10.x then depends
> on re@'s take on it.
>

I think it's true for 9.2 as well.

not sure about 8 or 9.1



More information about the freebsd-net mailing list