svn commit: r253346 - in head/sys: kern net netgraph netgraph/bluetooth/socket

[Marko Zec]

On Thursday 25 July 2013 11:36:46 Craig Rodrigues wrote:
> On Thu, Jul 25, 2013 at 1:07 AM, Marius Strobl 
<marius at alchemy.franken.de>wrote:
> > Uhm - do we really need to have that layering violation in subr_bus.c?
> > Wouldn't it be sufficient to set curthread->td_vnet to vnet0 in
> > if_alloc(9) or if_attach(9) at least instead?
>
> There was some discussion about this involving Marko Zec,  Adrian Chadd,
> and myself
> starting in this thread:
>
> http://lists.freebsd.org/pipermail/svn-src-all/2013-July/071798.html
>
> Adrian and Marko converged on similar patches:
>
> http://lists.freebsd.org/pipermail/freebsd-hackers/2012-November/041120.h
>tml
> http://people.freebsd.org/~adrian/ath/20130712-vimage-default-attach-deta
>ch.diff
>
>
> As Marko mentioned in another e-mail on this thread, the patch as it is
> necessary in
> order to fix VIMAGE related kernel panics in many different scenarios,
> including
> ones involving Netgraph nodes.

Moreover, unconditionally setting curvnet to vnet0 in if_alloc(), 
if_attach() or similar places as suggested my Marius simply couldn't work, 
because that would break creation of pseudo-interfaces inside non-vnet0 
contexts (such as vlan, ng_ether, ng_eiface etc.).

Marko