Re: [RFC 4/9] snet: introduce snet_core.c and snet.h

[Samir Bellabes <sam@synack.fr>]

I'm sorry for the delay I missed this one.

Evgeniy Polyakov <zbr@ioremap.net> writes:
> On Wed, Jan 06, 2010 at 08:46:35PM +0100, Samir Bellabes (sam@synack.fr) wrote:
>> Patrick McHardy <kaber@trash.net> writes:
>> 
>> >> +struct snet_event {
>> >> +	enum snet_syscall syscall;
>> >> +	u8 protocol;
>> >> +} __attribute__ ((packed));
>> >
>> > Does this really need to be packed? You're using it in a
>> > struct snet_event_entry, which is padded anyways.
>> 
>> I think that that members needs to be aligned, because this struct is
>> used for the jhash() computation.
>> when testing on x86_64, I discovered that jhash value wasn't the same,
>> for a same struct snet_event. Then I thougth about misaligned data, and
>> possible 'hole' between syscall and protocol.
>
> Without padding it will eat additional bytes after 'protocol' field,
> since enum is 32-bit long.

thanks for this help.


>> anyway, I patched the code to use jhash_2words() or jhash_1word().
>> here is a patch, which apply on top of initial serie
>
> What's the purpose of hashing verdict_id?

indeed, there is no specific purpose, I will fix this now.

thanks, and again sorry for the delay.
sam