netdev
[Top] [All Lists]

Re: Allowing netlink_family to be any integer (was: [PATCH 2.6] iptables

To: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>, Evgeniy Polyakov <johnpol@xxxxxxxxxxx>
Subject: Re: Allowing netlink_family to be any integer (was: [PATCH 2.6] iptables CLUSTERIP target)
From: jamal <hadi@xxxxxxxxxx>
Date: 22 Oct 2004 07:29:39 -0400
Cc: Harald Welte <laforge@xxxxxxxxxxxxx>, lmb@xxxxxxx, ahu@xxxxxxx, hch@xxxxxxxxxxxxx, netdev@xxxxxxxxxxx, "David S. Miller" <davem@xxxxxxxxxxxxx>
In-reply-to: <E1CKkWZ-0005x5-00@gondolin.me.apana.org.au>
Organization: jamalopolous
References: <E1CKkWZ-0005x5-00@gondolin.me.apana.org.au>
Reply-to: hadi@xxxxxxxxxx
Sender: netdev-bounce@xxxxxxxxxxx
Evgeniy Polyakov(aka mr. Sean Paul) posted code a while back for
something that uses netlink that he calls  "kernel Konnector". I think
thats a good idea which will ease the use of those limited numbers. It
also has potential for generic kernel-kernel as well as kernel-userspace
messaging subsystem.
I printed the code, got a large cup of brazillian-derived cappucino but
alas got preempted before finishing the rewiew. Maybe you could work
with him Herbert?
I still plan to continue looking at it.

This does not exclude the use of the netlink numbers, but should ease
them.

cheers,
jamal

On Thu, 2004-10-21 at 17:31, Herbert Xu wrote:
> Harald Welte <laforge@xxxxxxxxxxxxx> wrote:
> >
> > For supporting two primitive operation, adding a new netlink address
> > family also isn't worthwhile, especially since we're short of netlink
> > families.
> 
> That's something I'm looking into as well.  The current strategy of
> either creating a new family or tacking random things into RTNETLINK
> is simply not going to scale.
> 
> Initially I considered an interface where kernel users can register
> themselves using a string as the key.  But I soon realised that we
> could simply allow the netlink_family field to be an arbitrary integer
> that is used as a key to a hash table.
> 
> The CPU cost of the hash table isn't too bad since you'll only be
> looking it up when the socket is created.
> 
> Comments anyone?


<Prev in Thread] Current Thread [Next in Thread>