Created attachment 340283 [details] patch to ifup-eth that requires setting LINKDELAY Description of problem: The physical interfaces on a Linux bridge should not have any IP address. On an IPv6-enabled host, however, interfaces have at least a Link-Local address, which is undesirable, at the very least. Version-Release number of selected component (if applicable): initscripts-8.94-1.i586 How reproducible: Always Steps to Reproduce: 1. Configure eth1 to have no address and "BRIDGE=br0" (e.g.) 2. ifup eth1 3. ifconfig eth1 (or "ip addr show eth1", if you prefer) Actual results: eth1 Link encap:Ethernet HWaddr 00:24:81:23:D2:71 inet6 addr: fe80::224:81ff:fe23:d271/64 Scope:Link UP BROADCAST RUNNING PROMISC MULTICAST MTU:1500 Metric:1 Expected results: eth1 Link encap:Ethernet HWaddr 00:24:81:23:D2:71 UP BROADCAST RUNNING PROMISC MULTICAST MTU:1500 Metric:1 Additional info: /etc/sysconfig/network-scripts/ifup-eth essentially short-circuits most interface processing if the ifcfg file has a BRIDGE= line in it. It attempts to clear any address, add the interface to the specified bridge group, then aborts any further processing on it. To that extent, it does not disable IPv6 on the interface. I've got two alternatives for patches. One condition is that disabling IPv6 only appears to work after IPv6 has initialized on the interface to begin with. That usually requires setting LINKDELAY in the ifcfg file as well. In my testing, I found 3 to be the largest required delay across a couple different machines, however it varied from machine to machine. Alternative #1 requires the admin to set LINKDELAY in the ifcfg file. Alternative #2 does not require setting LINKDELAY. I like it a little better for that reason, but it's also a little riskier, with a while loop that waits whatever time, short or long, for IPv6 to stabilize to be cleared.
Created attachment 340284 [details] patch to ifup-eth that does NOT require setting LINKDELAY
Created attachment 340287 [details] patch to ifup-eth that does NOT require setting LINKDELAY edit an indentation change sleep to /bin/sleep
Comment on attachment 340287 [details] patch to ifup-eth that does NOT require setting LINKDELAY Ah, the acid test: rebooting. Upon initially loading the ipv6 module, the existence the while loop tests for is there, but it's useless. There's no getting away from using LINKDELAY. Scratch the second alternative. Folks just have to make sure they make LINKDELAY large enough.
Thanks for your report. Assigning it Fedora Bugzappers Team
I've got a better patch that doesn't require LINKDELAY. I'll polish it a bit and send it your way shortly.
This bug appears to have been reported against 'rawhide' during the Fedora 11 development cycle. Changing version to '11'. More information and reason for this action is here: http://fedoraproject.org/wiki/BugZappers/HouseKeeping
I had some trouble reproducing the problem. Then I noticed I was testing on a box where I had already applied the patch for Bug 496444. Since applying that patch solves this problem, too, I'm closing this report. If folks have put "IPV6INIT=yes" in /etc/sysconfig/network (wrongly, I think) instead of in the individual ifcfg files, they just have to make sure that they put "IPV6INIT=no" in ifcfg-br0 to override the "yes" in network. *** This bug has been marked as a duplicate of bug 496444 ***