Peter Memishian wrote: > > > I agree it's gross -- but I think the strcmp() stuff would be worse. > > > > > > What would it take it get all of the DLPI devices in /dev/net? > > > > Sorry, can you rephrase your question? > > I mean: we don't need a fallback mechanism if all of the devices that need > to be accessed are in /dev/net. But it seems some devices, such as > /dev/vni or /dev/sppp, cannot be in /dev/net. What are the reasons? > As disccussed, there are ibd and clprivnet too on the list. But I don't think there are cases that /dev/net/sppp is ever accessed (ppp is not changed to use libdlpi). I didn't find places which really open /dev/vni devices either.
Anyway, the reason for not having them in /dev/net is: According to our design, /dev/net only keeps nodes of data-links, and more strictly, for GlDv3 data-links (and the ones unified to GLDv3 by softmac). But vni, sppp and ibd (together with others) are either IP level interfaces or non-GLDv3 data-links. Although it seems it is possible to resolve the /dev/net nodes to the right dev_t, as /dev/net is based on DBNR (directory based name resolution), there are at least several problems I can think of now: a. are we going to reserve these names (ibdN, sppp and vni) in the /dev/net namespace? b. So far /dev/net only support style-1 open, but the list of "exception" devices might only support style-2. Thanks - Cathy
