Re: Latest -current complete lockup (tcp changes?)

Giorgos Keramidas wrote:
> Hi Andre,
> 
> The attached thread below is a reply from Nicolas Blais which shows some
> problems with the recent SACK changes.  Any idea how we can proceed from
> this point?

Please update to rev. 1.36 of sys/netinet/tcp_sack.c.  The KASSERT() was
too tight but actually caught a bug that would have caused a crash anyway.

-- 
Andre

> - Giorgos
> 
> On 2007-03-25 16:20, Nicolas Blais <nb_root_at_videotron.ca> wrote:
>> On March 25, 2007 03:49:46 pm Giorgos Keramidas wrote:
>>> On 2007-03-25 13:48, Nicolas Blais <nb_root_at_videotron.ca> wrote:
>>>>> The panic happens after a small amount of network activity, inside
>>>>> the TCP SACK code:
>>>>>
>>>>> -----------------------------------------------------------------------
>>>>> ---- Disabling SACK lets me use my laptop again:
>>>>>
>>>>>     net.inet.tcp.sack.enable=0
>>>>>
>>>>> without any panics.
>>>> Same here. If you want to quickly crash your system,
>>>> net.inet.tcp.sack.enable=1 and start a torrent :)
>>>>
>>>> I couldn't find what was causing my desktop to crash until I STOPPED
>>>> trying to restart my torrent after boot :) . Though
>>>> net.inet.tcp.sack.enable=0 fixed it.
>>> Rebuilding a kernel will take a while here.  If you can try reverting
>>> the following commit before I do, let me know if it fixes the panic.
>>>
>>> % andre       2007-03-23 18:33:21 UTC
>>> %
>>> %   FreeBSD src repository
>>> %
>>> %   Modified files:
>>> %     sys/netinet          tcp_input.c tcp_sack.c
>>> %   Log:
>>> %   Bring SACK option handling in tcp_dooptions() in line with all other
>>> %   options and ajust users accordingly.
>>> %
>>> %   Revision  Changes    Path
>>> %   1.326     +7 -4      src/sys/netinet/tcp_input.c
>>> %   1.35      +1 -0      src/sys/netinet/tcp_sack.c
>> Done. 
>>
>> I reverted tcp_input.c to 1.325 and tcp_sack.c to 1.34. I was able to start 
>> many torrents without a panic. I believe the revert fixed the problem, as 
>> previously, just starting 1 torrent would panic (unless  
>> net.inet.tcp.sack.enable=0) , now I have 5 running fine.
>>
>> Nicolas.
> 
>