[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20110222093425.GX22310@pengutronix.de>
Date: Tue, 22 Feb 2011 10:34:25 +0100
From: Uwe Kleine-König
<u.kleine-koenig@...gutronix.de>
To: David Miller <davem@...emloft.net>
Cc: netdev@...r.kernel.org, Marco d'Itri <md@...ux.it>
Subject: Re: sysctls below net.ipv[46].conf.all not working as expected
Hello David,
On Thu, Dec 30, 2010 at 09:43:19PM +0100, Uwe Kleine-König wrote:
> On Thu, Dec 30, 2010 at 12:30:23PM -0800, David Miller wrote:
> > From: Uwe Kleine-König <u.kleine-koenig@...gutronix.de>
> > Date: Wed, 29 Dec 2010 17:06:07 +0100
> >
> > > I did the following:
> > >
> > > cassiopeia:~# sysctl net.ipv6.conf.all.use_tempaddr
> > > net.ipv6.conf.all.use_tempaddr = 0
> > >
> > > cassiopeia:~# sysctl net.ipv6.conf.eth0.use_tempaddr
> > > net.ipv6.conf.eth0.use_tempaddr = 0
> > >
> > > cassiopeia:~# sysctl -w net.ipv6.conf.all.use_tempaddr=1
> > > net.ipv6.conf.all.use_tempaddr = 1
> > >
> > > cassiopeia:~# sysctl net.ipv6.conf.all.use_tempaddr
> > > net.ipv6.conf.all.use_tempaddr = 1
> > >
> > > cassiopeia:~# sysctl net.ipv6.conf.eth0.use_tempaddr
> > > net.ipv6.conf.eth0.use_tempaddr = 0
> > >
> > > Here I would have expected that eth0's use_tempaddr is 1, too. The
> > > problem is not that this entry isn't writeable:
> >
> > The "all" value is propagated at the first moment that the ipv6 device
> > private is created, usually that is when the device is first brought up
> > which means it can happen as early as the exact moment the device is
> > registered.
> >
> > Therefore, if you want "all" to apply to "eth0", you must make sure
> > the sysctl is set properly before the device is registered.
> I thought this is what "default" was used for?!
I took the time now to dig into the source for ipv4/{all,default}.
Unless I'm mistaken I found:
- inetdev_init copies dev_net(dev)->ipv4.devconf_dflt to &in_dev->cnf
That's called from inetdev_event if there is no in_dev.
- devinet_copy_dflt_conf copies to all devices that have !test_bit(i,
in_dev->cnf.state)
devinet_copy_dflt_conf is called when /proc/sys/net/ipv4/conf/default
is written to. I didn't found where the in_dev->cnf.state bits are
set.
- net->ipv4.devconf_all is never propagated to the devices.
- The data behind /proc/sys/net/ipv4/conf/all is used in registering
net->ipv4.forw_hdr and seems only have to do with ip_forward
So I think that it's really "default" that propagates to a new device as
I expected and the documentation suggests.
So IMHO a better name for "all" would be "global" ...
> If you are really right, the documentation is (IMHO) misleading. e.g.
> Documentation/networking/ip-sysctl.txt tells:
>
> conf/default/*:
> Change the interface-specific default settings.
>
> conf/all/*:
> Change all the interface-specific settings.
... and this is misleading. I wonder if it would be considered OK, to
create a new config directory called "global" that acts like "all" now
and let "all" propagate all changes to all devices, "default" and
"global". If not the documentation needs clearification.
I guess for ipv6 it's similar, though I didn't check that.
Best regards
Uwe
--
Pengutronix e.K. | Uwe Kleine-König |
Industrial Linux Solutions | http://www.pengutronix.de/ |
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists