Red Hat Bugzilla – Bug 587513
Unable to reconnect to a dual-stacked network
Last modified: 2010-05-02 17:35:54 EDT
Created attachment 410312 [details]
Description of problem:
When NM is started in a network environment with IPv4 DHCP and IPv6 RA/SLAAC (with RDNSS), it will connect after boot just fine the first time, and both the IPv6 and the IPv4 name servers wil make it into /etc/resolv.conf.
However, if I attempt to re-connect to the network from the GNOME applet drop-down menu, it doesn't ever finish. The applet sits there with two dots green and a spinning blue thing.
Version-Release number of selected component (if applicable):
Steps to Reproduce:
1. Boot up
2. Select "System eth0" in nm-applet drop-down menu to trigger a reconnection
Reconnection process never finishes. Network unuseable.
Reconnection finished quickly.
See attached log. The first reconnection I attempted at 07:55:40, by selecting "System eth0" in the drop-down menu while I was still connected. At 08:02:10 I gave up waiting and deselected first "disable networking" and then selecting it again at 08:02:14. Same results. At 08:04:36 I disabled networking again, saved the log, and rebooted (which successfully reconnected me to the network)
I'll reboot in a little bit just in case it matters, but I can't reproduce Tore's success: I've done:
# /etc/init.d/NetworkManager stop
# rmmod iwlagn
# > /etc/resolv.conf # I'll file a bug for this
## at this point, as far I as I know, I should have reset any state that would affect NetworkManager as far as first time vs Nth time, but I could be wrong
# modprobe iwlagn
# /etc/init.d/NetworkManager start
## sits at the two dots, spinning, with SLAAC IPv6, no IPv4, and empty /etc/resolv.conf (no RDNSS info)
I've seen NM being dependent on what is done by the kernel prior to what NM is doing before (see bug #530670 for example). Since I'm testing with wired ethernet, SLAAC will likely have been performed by the kernel even before NM has started, while in your case with a WLAN, probably not (since NM needs to program the SSID, possibly start the wpa_supplicant, and so on).
I thought that might be the case here, too. Perhaps you can test using wired Ethernet, too?
I rebooted, and it changed nothing. But I'm using wifi while Tore's using Ethernet.
So I went over and plugged in the Ethernet cable, and it duplicates Tore's initial success. I then unplugged the Ethernet cable and reconnected it. It sits at two dots and swirls. However, there is a difference: eth0 only has the link local address, whereas in the wireless case (connect or reconnect), I would always have all the SLAAC addresses from RA.
If I follow the recipe I showed above...it reverts back to the "initial connect" behavior. I don't really even need to rmmod/modprobe either. A simple NetworkManager restart works fine.
So this happens to be the wireless analog to the wifi case I've been noticing for a while, but haven't gotten around to filing because I don't know how to reproduce it--sometimes the (wifi) network will stall out and stop being responsive (low signal strength? not sure). This is clearly a driver problem, I would think, and that seems confirmed in that if I disconnect and reconnect, it doesn't work (sits and spins). But what contradicts that theory is that if I restart NetworkManager (but leave the driver alone), it connects without a hitch.
Could these behaviors be related? Honestly, there is no reason for NetworkManager to behave differently for initial connect vs. reconnect. It seems like something isn't getting reset properly during a reconnect.
As best I can tell, this bug is fixed. I've tried wired & wireless via auto DHCP/auto. I've also tried wireless auto DHCP/ignore & disabled/auto and they connect correctly as well. (IPv6 only I am able to ping6 ipv6.google.com so the routes are right too).
I've disconnected, reconnected, etc.
This bug appears to be fixed in NM git20100502, closing.