[PATCH] Per bucket conntrack hash locking patch in p-o-m

[PATCH] Per bucket conntrack hash locking patch in p-o-m

[Jozsef Kadlecsik]

Hi,

I have added three patches to patch-o-matic/extra:

conntrack_arefcount.patch:
	The revision of the refcounting/locking in conntrack. First, the
	original methods was not perfect at all (see Patrick's efforts in
	cleaning it up...), secondly it did not make possible to
	introduce per bucket lockings at all.
	As an outcome of the revision, init_conntrack could be made a
	little bit faster (read-lock against expectations instead of
	write-locking).

conntrack_locking.patch:
	The patch implements per bucket locking in the conntrack
	hash. The only task remained for the ip_conntrack lock is
	to protect the protocol/helper lists :-).

conntrack_nonat.patch:
	The patch implements an optimization in the per bucket
	locking for the case when there is no NAT support compiled
	into the code.

Any comments are highly welcomed.

Best regards,
Jozsef
-
E-mail  : kadlec@blackhole.kfki.hu, kadlec@sunserv.kfki.hu
PGP key : http://www.kfki.hu/~kadlec/pgp_public_key.txt
Address : KFKI Research Institute for Particle and Nuclear Physics
          H-1525 Budapest 114, POB. 49, Hungary

Re: [PATCH] Per bucket conntrack hash locking patch in p-o-m

[Patrick McHardy]

Hi Joszef,
I've just had a look at the diffs until now (I tried to apply them
to numerous kernels until reading the helpfiles ;)), but one thought:


Jozsef Kadlecsik wrote:

>conntrack_locking.patch:
>	The patch implements per bucket locking in the conntrack
>	hash. The only task remained for the ip_conntrack lock is
>	to protect the protocol/helper lists :-).
>  
>

Why not remove the lock for protocol_list completely ? It can never change
anyways ..

Best regards,
Patrick

Re: [PATCH] Per bucket conntrack hash locking patch in p-o-m

[Patrick McHardy]

Patrick McHardy wrote:

> Why not remove the lock for protocol_list completely ? It can never 
> change
> anyways ..


Ok now I feel stupid ;) I just looked inside ip_conntrack_core until 
noticing
that the list is not static.

Sorry for the noise ..
Patrick