Re: Minor non-blocking "sleep" bugs

On Tue, 27 Mar 2007, Ismaël BALLO wrote:

> I use ipset 2.2.9a on kernel 2.6.19.7 compiled (from
kernel.org) with
> these options
> (iptables 1.3.6)
> I have a minor bug (non blocking) when I load ipsets (
it seems it
> happens when I have a large numbers of bindings)
>
> Mar 23 14:45:10 fwa01 kernel: BUG: sleeping function
called from
> invalid context at mm/slab.c:3007
> Mar 23 14:45:10 fwa01 kernel: in_atomic():1,
irqs_disabled():0
> Mar 23 14:45:10 fwa01 kernel:  [<c0158e6c>]
kmem_cache_alloc+0x1b/0x55
> Mar 23 14:45:10 fwa01 kernel:  [<f89a77b8>]
ip_set_hash_add+0xe7/0x142 
> [ip_set]

That's due to a stupid bug of mine in the flag of kmalloc.
The fixed 
kernel source can be downloaded from the svn repository or
as the 
patch-o-matic-ng-20070328.tar.bz2 snapshot from the ipset
webpage. Thank 
you for the bugreport.

> and other : What can we chose best parameters for
hashsize, probes, resize ?

There is no golden path: either you pay by memory for speed
(i.e. large 
hashsize, small probes, large resize percentage) or
reversed. The defaults 
of the iphash and nethash types are towards the other end,
i.e. spare with 
the memory requirement and loose thus speed.

Best regards,
Jozsef
-
E-mail  : kadlecblackhole.kfki.hu, kadlecsunserv.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