On Fri, Mar 09, 2007 at 01:47:51PM +0800, Cathy Zhou wrote: > David Edmondson wrote: > >On Wed, Mar 07, 2007 at 08:13:55PM +0800, Cathy Zhou wrote: > >>What do you think? > > > >I think that we agree on how the system should behave. > > > >What's left, perhaps, is to figure out the pieces in which this can be > >delivered. > > > >Modifying the management of link namespace has to be done in one > >chunk. That is, we shouldn't do the "ip.tun0 is a per-zone link name" > >without also doing "physical links assigned a non-global zone > >disappear from the global zone namespace". > > > What exactly you mean by "disappear from the global zone > namespace". Does you mean the global zone cannot see it at all.
Visible only when a "fully qualified" link name is used (i.e. one that includes the zone, such as 'zone1/bge0'). > Even > by dladm show-link or show-linkprop? Note that it can be seen by > both today and show-linkprop can actually be used to set zoneid of a > specific link. > > ... and although I don't disagree, I want to understand why you feel > strongly that these two has to be done in on chunk. Is this because the > inconsistency caused by the per-zone ip.tun0 link that if the tunnel is > created in the local zone, it cannot be shown in the global zone, but if it > is created in the global zone, it can be shown in the global zone? I think that the separation of the namespace should be done in one hit. dme.
