On Fri, 2009-06-12 at 11:21 -0400, James Carlson wrote: > Sebastien Roy writes: > > Only one request for clarification: > > > > On Mon, 2009-06-08 at 13:47 -0400, James Carlson wrote: > > > - /dev/bridge/ is gone > > > > > > This is no longer needed; Clearview's /dev/net/ observability > > > nodes serve all the needs of the bridging design. > > > > There is some inconsistency at least in the specification (I assume that > > the implementation works). /dev/net nodes are not observability nodes, > > but full-fledged DLPI link nodes. > > Sorry about the nomenclature. > > Yes, they're "real" DLPI link nodes, using plain old Ethernet, with > the exception that they're not treated as bridge links and you can't > transmit on them. Both the design and the 'bridging-spec.txt' > architectural document cover this.
That's what I thought. > > > /dev/ipnet nodes are IP observability > > nodes. > > Right; we're not using those at all. We're not delivering IP changes. Indeed. > > The new materials state: > > > > + using the /dev/net/ facility implemented for "Clearview: IP > > + Observability Devices" (PSARC 2006/475). These nodes will be > > + named by the bridge name plus a trailing "0". > > > > /dev/net nodes were added by 2006/499, and /dev/ipnet nodes by 2006/475. > > The inconsistency should be fixed in the spec. > > Ah, ok. Sorry about that; I somehow thought they were all one thing. > (See? I said before that I was confused about these two classes of > nodes ...) I think that the BPF project's general solution to this kind of problem is quite slick (one device, with a mechanism to specify the "layer" at which the consumer wishes to operate). The BPF and PF_PACKET functionality will eventually make /dev/ipnet obsolete, thus hopefully removing the confusion. > > Consider the reference updated. Excellent, thanks. +1 -Seb
