> > In response to the PSARC request that the IP Tunneling case be split in
> > two, here is the specification for the new "Datalink Administration from
> > Non-Global Zones" case to be filed shortly. This is basically just the
> > zones sections from the IP Tunneling design document, shoved into a new
> > document with some slight modifications to make the text more
> > self-contained.
> >
> > http://www.opensolaris.org/os/project/clearview/iptun/dl-zones-spec.pdf
> >
> > I'll be submitting these materials ASAP and getting on the earliest
> > agenda slot for this. The IP Tunneling case now depends on this new
> > case, so this is now in the critical path for the project as the IP
> > Tunneling case can't go for Commitment until this new case gets
> > approved.
Seb,
At the implementation-level, one issue that will need tending to is the
assumption of a single link namespace in the kernel. For instance,
ipmp_phyint_get_kstats() has:
/*
* NOTE: ALL_ZONES here assumes that there's at most one link
* with a given name on a given system (safe for now).
*/
ksp = kstat_hold_byname("link", 0, phyi->phyint_name, ALL_ZONES);
if (ksp == NULL)
return;
As I recall, ALL_ZONES was used ratheer than getzoneid() out of fear that
we might be processing this as part of a kernel worker thread that may not
be associated with any particular zone (also keep in mind that e.g. the
code above needs to work for both shared-stack and exclusive-stack zones).
If that fear is founded, then it seems we need a way to get the proper
zone context down to these sorts of call sites.
--
meem
