Re: [PATCH v5 1/2] sctp: check the rto_min and rto_max

[Vlad Yasevich <vyasevich@gmail.com>]

On 12/07/2013 04:12 PM, Daniel Borkmann wrote:
> On 12/07/2013 08:01 PM, Vlad Yasevich wrote:
>> On 12/07/2013 07:43 AM, Daniel Borkmann wrote:
>>> On 12/07/2013 08:17 AM, Wang Weidong wrote:
>>>> rto_min should be smaller than rto_max while rto_max should be larger
>>>> than rto_min. Add two proc_handler for the checking. Add the check in
>>>> sctp_setsockopt_rtoinfo.
>>>>
>>>> Suggested-by: Vlad Yasevich <vyasevich@gmail.com>
>>>> Signed-off-by: Wang Weidong <wangweidong1@huawei.com>
>>>> ---
>>>
>>> Thanks Wang, also for your second patch.
>>>
>>> Second one looks good to me, thanks for the cleanup!
>>>
>>> I was wondering where 86400000 comes from? Looking through the git
>>> history didn't give much clues and the RFC4960 neither. Clearly,
>>> section 15 of RFC4960 *recommends* as initial values ...
>>>
>>>    RTO.Initial - 3 seconds
>>>    RTO.Min - 1 second
>>>    RTO.Max - 60 seconds
>>>
>>> ... which we have as constants in [1] and are assigned to globals
>>> initially in [2,3] with those recommended values. That's all good.
>>>
>>> But still [not *directly* related to your patch though], where does
>>> 86400000 come from? I expect that's for the max SCTP heartbeat
>>> interval or max cookie lifetime?
>>
>> No, initially it was defined as rto_timer_max and was the upper bound
>> for the rto timer.   When you think about it, it's a bit ridiculous
>> really.  What you are saying is that your rto timer is allowed to
>> grow as long as 1 day, so you would at an absolute maximum retransmit
>> one packet per day :)
> 
> Exactly, maybe initial SCTP implementors already took into account
> we could have SCTP connections to other galaxies, but clearly untested
> so far? :)
> 
> I think we should be absolutely fine with a max configurable upper
> limit of twice the recommended RTO.Max value from the RFC.

That would only put at at 2 min...  That may not be enough in certain
deployments.  Yes, they always have the option to set it per socket,
but that becomes a pain and harder to tune the same app to different
links.  I think 2 hours might be a safer maximum RTO.Max, but I find
it hard to justify shrinking this value.  It isn't really broken
and if someone does set it to 6 hours, they really are shooting
themselves in the foot :)

-vlad

> 
>> I don't think this limit is specified anywhere as is though.  It
>> was something that's been there since the 2.5 days.
>>
>> -vlad