VIMAGE + PF crash in mbuf destructor
Marko Zec
zec at fer.hr
Mon Jul 22 21:38:12 UTC 2013
On Monday 22 July 2013 21:01:31 Adrian Chadd wrote:
> Well I'm worried about _other_ stuff causing issues here.
>
> So - what's the "right" behaviour? Does vnet/vimage make the
> assumption that for all the mbuf processing/free operations, the vnet
> tag/state is set?
To the best of my knowledge, there's nothing vnet-specific in any of the
mbuf-handling routines, i.e. they should all work fine on a VIMAGE kernel
even if curvnet isn't set.
My original motivation behind keeping separate UMA zones for each vnet was
solely to capture resource leaks between vnets in the early days of VIMAGE
prototyping. Nothing prevents a single UMA zone to be shared by multiple
vnets, unless we want to enforce per-vnet limitations on the number of
items in a zone.
Marko
>
> -adrian
>
> On 22 July 2013 11:59, Craig Rodrigues <rodrigc at freebsd.org> wrote:
> > On Mon, Jul 22, 2013 at 10:11 AM, Adrian Chadd <adrian at freebsd.org>
wrote:
> >> I don't think the default vnet context is the correct behaviour there.
> >> We'd need to figure out what the vnet context of the mbuf is and set
> >> that.
> >
> > What do you think about Marko's suggestion to de-virtualize
> > V_pf_mtag_z? What would be the down side of that?
> >
> > I don't understand enough of the PF code to understand which variables
> > need to
> > be virtual and which don't.
> >
> > --
> > Craig
>
> _______________________________________________
> freebsd-virtualization at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-virtualization
> To unsubscribe, send any mail to
> "freebsd-virtualization-unsubscribe at freebsd.org"
More information about the freebsd-virtualization
mailing list