> John has run into somewhat of a road-block while performing stress tests > with ipnet, in that he is continuously running into existing Nevada bug > 6606991: > > 6606991 panic assertion failure !ill->ill_join_allmulti for multicast router > > Our ipnet code exposes this existing bug more easily than did the > previous test case which involved starting a multicast routing daemon. > In addition to our stress tests, it's also possible to hit this bug with > ipobs by running "snoop -I <intf>" while changing the address of the > interface (without first bringing the interface down). In any case, I > decided to finally get to the bottom of this bug in order to facilitate > our stress testing. The fix is here: > > http://zhadum.east/ws/seb/seb-ipobs/webrev.6606991 > > Please review, we need to include this fix with the ipobs putback.
Why does only ipif_up_done() need the fix, and not ipif_up_done_v6() as well? Separately from that, I'm not wild about introducing even more ill_t variables; why can't the existing ill_dl_up() function join allmulti (which would mirror the ill_dl_down() behavior)? -- meem
