Re: connbytes & 64bit counters

[Patrick McHardy <kaber@trash.net>]

Pablo Neira Ayuso wrote:
> Patrick McHardy wrote:
> 
>>> What is wrong in sending 64 bit counters to userspace if we already have
>>> 64 bit counters in kernel?
>>
>>
>> Nothing - but changing the API based on config options is bad design.
>> I am fine with sending 64 bit unconditionally. But you need to make
>> sure you don't send (32 bit) overflow events to userspace anymore.
> 
> 
> They could be used to notify 64 bits overflow events. Userspace can
> differenciate if the kernel is using 32 or 64 bits and interpret the
> overflow event appropiately.


Mhh .. even with 100gbit it will take about 50 years to overflow.
I don't think we really need this.

> Hm, why not default on 64 bits counters but still give the choice to
> select 32 bits at compilation time? Some advanced users could still want
> to compile 32 bits to save some memory (perhaps embedded stuff guys), so
> I have the impression that this issue will be revisited sooner or later.
> I can give a hand to Krzysztof and improve the patch to make it look
> cleaner, although we will still need some ifdef's. Still not convinced?


Config options for things like this are silly in my opinion, if we
really want to save memory these counters should be selectable at
runtime, most people don't need them, but distributions will probably
enable them anyway (if not already then once programs using them
appear). And people using them don't save anything, they have to
keep copies in userspace to handle overflows, which needs even more
memory.

Given that their use of the 32 bit counters in libnetfilter_conntrack
is broken anyway (incorrect byte order conversion) I think we should
just get rid of them.