Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 19 Oct 2024 19:22:08 -0400
From:      Mark Johnston <markj@freebsd.org>
To:        Konstantin Belousov <kostikbel@gmail.com>
Cc:        John Baldwin <jhb@freebsd.org>, src-committers@freebsd.org, dev-commits-src-all@freebsd.org, dev-commits-src-main@freebsd.org
Subject:   Re: git: f4e35c044c89 - main - bus: Set the current VNET in device_attach()
Message-ID:  <ZxQ_IALrOAVAf09i@nuc>
In-Reply-To: <ZxPaDc7jAORhM5O6@kib.kiev.ua>
References:  <202410191304.49JD4JoM084001@gitrepo.freebsd.org> <ed434405-26cc-4363-834e-9f31d7912589@FreeBSD.org> <ZxPaDc7jAORhM5O6@kib.kiev.ua>

next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, Oct 19, 2024 at 07:10:53PM +0300, Konstantin Belousov wrote:
> On Sat, Oct 19, 2024 at 11:36:32AM -0400, John Baldwin wrote:
> > On 10/19/24 09:04, Mark Johnston wrote:
> > > The branch main has been updated by markj:
> > > 
> > > URL: https://cgit.FreeBSD.org/src/commit/?id=f4e35c044c8988b7452cefbdcc417f5fd723e021
> > > 
> > > commit f4e35c044c8988b7452cefbdcc417f5fd723e021
> > > Author:     Mark Johnston <markj@FreeBSD.org>
> > > AuthorDate: 2024-10-19 13:03:56 +0000
> > > Commit:     Mark Johnston <markj@FreeBSD.org>
> > > CommitDate: 2024-10-19 13:03:56 +0000
> > > 
> > >      bus: Set the current VNET in device_attach()
> > >      Some drivers, in particular anything which creates an ifnet during
> > >      attach, need to have the current VNET set, as if_attach_internal() and
> > >      its callees access VNET-global variables.
> > >      device_probe_and_attach() handles this, but this is not the only way to
> > >      arrive in DEVICE_ATTACH.  In particular, bus drivers may invoke
> > >      device_attach() directly, as does devctl2's DEV_ENABLE ioctl handler.
> > >      So, set the current VNET in device_attach() instead.
> > >      I believe it is always safe to use vnet0, as devctl2 ioctls are not
> > >      permitted within a jail.
> > >      PR:             282168
> > >      Reviewed by:    zlei, kevans, bz, imp, glebius
> > >      MFC after:      1 week
> > >      Differential Revision:  https://reviews.freebsd.org/D47174
> > 
> > Hmm, there was some other review I thought that had a completely different change.
> > That change removed all the vnet stuff from new-bus and instead handled it in
> > if.c.  Specifically, that if_attach would set a default vnet to vnet0 if there
> > wasn't an active vnet at the time.  See all the discussion in
> > https://reviews.freebsd.org/D42678 which has a patch that I think is correct
> > in the comments.

Gleb's proposal, described a bit in D47147, is to require device-based
ifnet drivers to fully detach themselves from the parent bus in order to
change VNETs.  The intent is to eliminate the need for if_vmove() and
all the complexity it entails.  This would also eliminate the need for a
"home" VNET, referenced in the patch that you reference here.

> In fact, I think that bus level is better.  At least, I know that detach
> also might need something by vnet (e.g. mce(4) needs to clear the IPSEC
> offload database on detach, although it still does not do).

Shouldn't something like this be handled by
if_detach()/ether_ifdetach()?  Posed another way, why does device detach
itself need to care about the VNET?

I tend to agree that having VNET knowledge in subr_bus.c is a hack.  My
aim was just to fix the panic without making the hack worse.

> It sounds as if bus_topo_lock() is the right place.  May be some other
> name for it is better, like bus_topo_changes_enter().



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?ZxQ_IALrOAVAf09i>