Peter Memishian wrote: > > If you do "ifconfig -a", and see "vni0", aren't you > > going to expect "snoop -d vni0" to intercept those > > packets and not those from some other interface? > >Given that "ifconfig -a" gives me an IP interface view and "snoop -d" >works on the link layer, certainly not. > > > The expectation is that once a device is created (and > > especially once it is visible) is that all use of that > > name should refer to that name, irrespective of whether > > or not it is DLPI or something else and that the system > > should operate in such a way as to preserve the uniqueness > > of that name. > >Given that the link namespace and IP namespace are distinct, that's an >unreasonable expectation. All we can guarantee is that a given object >will have the same name at every level it exists at (and in fact, this is >stated in the original Clearview charter) -- not that the same name at >every level will refer to the same object. > > > It would appear that there's a serious architectural > > problem here that needs to be addressed. > >I don't see an architectural issue, just a natural consequence of multiple >namespaces. The namespaces are themselves a consequence of a core design >decision made long ago on Solaris (and one that I very much agree with) to >separate the link-layer and the IP-layer rather than weld them together. > >
Whatever the internals may be, we've always presented a unified and sensible appearance to systems administrators. So even though the name of the interface at the IP layer may be set to be something different, the standard tools that everyone use keep everything to be one to one. When I look at the changes UV is delivering, apart from this rename-link problem, it does not appear to change this. It will not be possible to do "ifconfig foo0 plumb" and for foo0 to be actually using bar0. This is not a bad thing, this is good. So while the design of the subsystems and the code might support different name spaces, until ifconfig allows me to either rename an IP interface or plumb an IP interface from a device that is different to the name I wish to give it, the administration model (for both IP and the MAC layers) appears to continue to be driven by the name of the device at the MAC layer. The end result of this is that most people out there expect bge0 in ifconfig's output to be the same thing as bge0 inside, even if they're technically opening different devices. The UV project is by and large in keeping with what people expect - except for the edge case you've raised. Darren
