In message <OFF79C4263.91C48EE0-ON88256D5B.0025FE92@us.ibm.com>,David Stevens w rites: >But you didn't try reproducing the problem, apparently... just tried. still have the problem. this is 2.5.74... >Try it out. I predict it'll work, but I'm certainly interested in hearing >if you have a test case that doesn't. :-) the lane driver is very similar to a 'stupid' ethernet card. it returns all the multicast traffic. the igmp stack seems to keep track of all the learned groups, so i my network i typically see: IPv6/IPv4 Group Memberships Interface RefCnt Group --------------- ------ --------------------- lo 1 224.0.0.1 lec0 1 224.0.1.1 lec0 1 224.0.0.1 so the learned address 224.0.1.1 is causing trouble on shutdown and removal of the network interface. both ip_mc_down() and ip_mc_destroy_dev() call igmp_group_dropped(). as stated before, igmp_group_dropped() will start the timers in an attempt to leave the multicast group. looking at igmp_group_dropped(), you can see someone fixed the dropping of the ALL_HOSTS group, by simply testing for and ignoring it. i dont know if a host isnt allowed to issue a leave for the all hosts. but a better test might be to check to see if the underyling network device is IFF_UP before continuing on with the drop operation. igmp_group_dropped() really cant do anything if the network interface isnt able to transmit. also, i dont believe one would need to have a 'stupid' interface for the above condition. simply listen on a multicast address and try to ifdown and remove the active network interface. - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
