James Carlson wrote: > Artem Kachitchkine writes: > >> nv_68-based brussels-gate, ferrari laptop: >> >> # time dladm show-linkprop lo0 >> dladm: warning: cannot open lo0: DLPI operation timed out >> >> real 0m10.021s >> > > Gotcha! > > On BFU'd systems (only), there's a /dev/lo device. That node is most > definitely not DLPI -- it's a STREAMS SVr4 test driver used for the > validation (SVVS) suite. It's not related to networking at all. > > The DLPI library doesn't realize that this node is spurious, and it > tries to open it and talk to it as though it were a DLPI Style 2 > driver ... and it times out when it doesn't respond. > > 10 seconds seems a bit extreme for a timeout here, but in any event, > the library in use here should probably be a bit smarter, and should > avoid trying to talk to the SVVS driver at all. > The default timeout for libdlpi operations is 5 seconds. However, in the case above it took 10 seconds as it first thought it was DLPI style1 device and was trying to process it and failed, then libdlpi tries to open it as style2. So it took 10seconds.
But more importantly, since IP observability will be implementing /dev/lo to be consistent with other unix variants, what would be a way for libdlpi to distinguish the difference? Sagun
