tcp failing to recover from a packet loss under 8.2-RELEASE?

[Lawrence Stewart]

Sorry for the delay in responding.

On 08/29/11 21:30, Andre Oppermann wrote:
> On 29.08.2011 02:08, Lawrence Stewart wrote:
>> On 08/14/11 23:53, Steven Hartland wrote:
>>> ----- Original Message ----- From: "Lawrence Stewart"
>>> <lstewart at freebsd.org>
>>>>
>>>> Here's my tweaked version of Andre's patch:
>>>> http://people.freebsd.org/~lstewart/patches/misctcp/tcp_reass.c-logdebug%2bmissingsegment-20110811-lstewart.diff
>>>>
>>>>
>>>>
>>>
>>> Still testing this and just noticed that the patch fails to
>>> compile when INVARIANTS is enabled. The KASSERT calls need
>>> ()'s around the split strings.
>>
>> oops, sorry. Compile tested on 8-STABLE without INVARIANTS enabled :/
>>
>> Any further feedback with respect to the patch? Plan to submit it to
>> re@ later this week for
>> inclusion in 9.0.
>
> I'm not sure these excessive KASSERT's are really necessary. Probably one
> covering those cases at the start of the function is sufficient.

I don't see how the logic currently captured in the KASSERTs I've added 
could be easily captured in a single KASSERT at the top of the function. 
I also think KASSERTs are most usefully placed next to the code which is 
making the assumption you wish to sanity check.

Could you please give me a few pointers as to what you had in mind?

> I was about to send the original patch to re@ for approval as well now
> that I'm back from vacation and fully available again. Don't mind if
> you do it if you've got the time.

I didn't realise you'd gone on holidays and planned to deal with this on 
your return so I already started a dialogue with re@ and might as well 
follow it through now.

Cheers,
Lawrence