Xen devel - Oct 2007 - PATCH: default number of netloop devices to number of network devices

Right now, unless specifically overridden, loopback_init() in 
drivers/xen/netback/loopback.c creates 4 netloop devices.  If your 
default network device is eth4 or higher, you''re out of luck creating a
xenbrN device with /etc/xen/scripts/network-bridge unless you explicitly 
set netloop.nloopbacks to something higher.  This patch attempts to set 
a more reasonable default: the number of network devices that exist, as 
reported by dev_get_by_index().

This isn''t a perfect solution-- in particular, there will generally be
a
few more devices than there are physical network devices, but looking 
around and consulting with Paul Moore (netlabel maintainer), there 
doesn''t seem to be a good way to tell real network devices from pseudo 
ones.  My initial attempt at a patch checked tmp->name for whether it 
began with ''eth'', but it''s completely reasonable to
have network devices
named anything-- an old laptop used to have its wireless interface named 
''ath0'', for instance, and another used
''wlan0''.  So while this approach
may waste a little bit of memory, at module load time, we can ensure 
that no matter how many devices you have in your system, at least at 
boot time, you can create that many xenbrN devices.

Signed-off-by: Eric Schwartz <eric.schwartz@hp.com>

-=Eric


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

The loopback device is deprecated in xen-unstable (not needed any more) and
this isn''t really critical enough for 3.1.2.

 -- Keir

On 25/10/07 19:38, "Eric Schwartz" <eric.schwartz@hp.com> wrote:
> Right now, unless specifically overridden, loopback_init() in
> drivers/xen/netback/loopback.c creates 4 netloop devices.  If your
> default network device is eth4 or higher, you''re out of luck
creating a
> xenbrN device with /etc/xen/scripts/network-bridge unless you explicitly
> set netloop.nloopbacks to something higher.  This patch attempts to set
> a more reasonable default: the number of network devices that exist, as
> reported by dev_get_by_index().
> 
> This isn''t a perfect solution-- in particular, there will
generally be a
> few more devices than there are physical network devices, but looking
> around and consulting with Paul Moore (netlabel maintainer), there
> doesn''t seem to be a good way to tell real network devices from
pseudo
> ones.  My initial attempt at a patch checked tmp->name for whether it
> began with ''eth'', but it''s completely reasonable
to have network devices
> named anything-- an old laptop used to have its wireless interface named
> ''ath0'', for instance, and another used
''wlan0''.  So while this approach
> may waste a little bit of memory, at module load time, we can ensure
> that no matter how many devices you have in your system, at least at
> boot time, you can create that many xenbrN devices.
> 
> Signed-off-by: Eric Schwartz <eric.schwartz@hp.com>
> 
> -=Eric
> --- linux-2.6.18.ia64/drivers/xen/netback/loopback.c.old 2007-10-25
> 13:05:30.716516696 -0400
> +++ linux-2.6.18.ia64/drivers/xen/netback/loopback.c 2007-10-25
> 14:14:07.835744809 -0400
> @@ -298,9 +298,18 @@
>  static int __init loopback_init(void)
>  {
> int i, err = 0;
> + struct net_device *tmp;
>  
> - if (nloopbacks == -1)
> -  nloopbacks = is_initial_xendomain() ? 4 : 0;
> + if (nloopbacks == -1) {
> +  nloopbacks = 0;
> +
> +  if (is_initial_xendomain()) {
> +   for(i = 1; tmp = dev_get_by_index(i); i++) {
> +    nloopbacks++;
> +    dev_put(tmp);
> +   } 
> +  } 
> + }
>  
> for (i = 0; i < nloopbacks; i++)
> if ((err = make_loopback(i)) != 0)
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel