Hide Forgot
Created attachment 1151368 [details] logs from reproducer Description of problem: Problematic system has following interfaces setup. +-----------------+ +----------------+ | | | | | eth0 | | eth1 | +-------+---------+ +---------+------+ | | | | +----+---------------------------+---+ | | | | | bond1 | +-------------+-------------+ | | | +-------------+-------------+ | | | | | Vlan0 | +-------------+-------------+ | | +-------------+-------------+ | | | 10.10.10.2/24 | | br1 | +---------------------------+ After system boot everything seems to be fine and all devices come UP as expected. However once NetworkManager is restarted (systemctl restart NetworkManager) vlan0 device and bond1 will not come back up. Version-Release number of selected component (if applicable): I've tried this on 2 versions symptoms seems to be identical: NetworkManager-1.0.6-29.el7_2.x86_64 NetworkManager-1.0.6-27.el7_2.x86_64 data are aquired while running 1.0.6-27 How reproducible: It happens every time on configuration mentioned in description Steps to Reproduce: 1. Setup network devices in configuration as shown in description. I've used nmtui. 2. Restart NetworkManager. systemctl restart NetworkManager Actual results: bond1 and vlan0 devices are down after restart. Expected results: bond1 and vlan0 should come up after restart. Additional info: I'm attaching messages file containing logs generated while I was reproducing issue. I cut the file so it starts with system bootup (with problematic configuration already at place) Then after 5 min I've restarted NetworkManager. I've attached also ip addr & ip link ... before/after NM restart. If sosreport will be needed I can provide it.
Created attachment 1151382 [details] ip link taken before NM restart
Created attachment 1151383 [details] ip link after nm restart
Created attachment 1151384 [details] ip addr before nm restart
Created attachment 1151385 [details] ip addr taken after NM restart
I can reproduce this. There are some messages in logs suggesting that the connection matching logic isn't picking up the existing connections for vlan and bond: Connection 'bond1' differs from candidate 'bond-bond1' in 802-3-ethernet.s390-options, 802-3-ethernet.s390-subchannels, 802-3-ethernet.s390-nettype Connection 'vlan0' differs from candidate 'vlan-vlan0' in 802-3-ethernet.s390-options, 802-3-ethernet.s390-subchannels, 802-3-ethernet.s390-nettype I'm investigating the issue.
The main issue in my tests is that when the bond is configured without IP addresses, the interface is brought down after NM is stopped and not activated on restart. After assigning a static IP to the bond (nmcli con mod <con-name> ipv4.method manual ipv4.address 172.31.255.255/32) the bond stays up. Which IP method did you set for the bond connection? Can you please show the output of 'nmcli con show <con-name>'? I tested this also with upstream git master of NM and the issue is still reproducible there; probably we should improve how software devices without IP configuration are handled upon restart.
Created attachment 1152145 [details] all NetworkManager connection profiles
Hi, I've tried simulate customers setting so I've set following on bond connection: IPv4 disabled IPv6 ignore I've attached contents all connection profiles from my reproducer
The scenario is similar to the one of bug 1333983 (vlan over bond/team that is lost on restart) and there is already a patch there, so I'm setting this a duplicate. *** This bug has been marked as a duplicate of bug 1333983 ***