[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <1245275693.31588.55.camel@johannes.local>
Date: Wed, 17 Jun 2009 23:54:53 +0200
From: Johannes Berg <johannes@...solutions.net>
To: "Eric W. Biederman" <ebiederm@...ssion.com>
Cc: Netdev <netdev@...r.kernel.org>, Thomas Graf <tgraf@...g.ch>,
"Eric W. Biederman" <ebiederm@...stanetworks.com>,
Alexey Dobriyan <adobriyan@...il.com>
Subject: Re: [PATCH] genetlink: make netns aware
On Wed, 2009-06-17 at 14:49 -0700, Eric W. Biederman wrote:
> Johannes Berg <johannes@...solutions.net> writes:
>
> > On Wed, 2009-06-17 at 14:19 -0700, Eric W. Biederman wrote:
> >
> >> Nothing at first glance looks wrong.
> >
> > Will review locking :) Not sure I hold enough locks at the correct
> > places here.
>
> grr. At first glance NOTHING looks wrong.
> I have a bad habit of skipping the most important word in my sentences
> sometimes.
My autocorrection kicked in and did actually read "nothing looks wrong",
but locking is still wrong :)
> Probably. As even genetlink has those kinds of users. The kernel
> receive path is easy now. Yeah for synchronous processing. It is
> which socket do you transmit on that is interesting.
Indeed. But genlmsg_reply() will automatically reply to the correct
netns now, so no big deal there. Multicast is more interesting, one
thing I may need to figure out is multicasting to _all_ namespaces.
Actually, hmm, let me change the API so that genlmsg_multicast() sends
to all network namespaces, and genmlsg_multicast_ns() sends to just a
specific netns.
I'm thinking that most genl users should enable netns but not do
anything different, so their services can be used regardless of what
netns the tool using them lives in, since many genl services don't
actually care about networking. Those that do, like nl80211, will have
to be updated differently.
johannes
Download attachment "signature.asc" of type "application/pgp-signature" (802 bytes)
Powered by blists - more mailing lists