Bug 31842

Summary:	Mac address of bonding devices modified during boot
Product:	Mageia	Reporter:	Chris Denice <eatdirt>
Component:	RPM Packages	Assignee:	Thomas Backlund <tmb>
Status:	NEW ---	QA Contact:
Severity:	major
Priority:	Normal
Version:	Cauldron
Target Milestone:	---
Hardware:	All
OS:	Linux
Whiteboard:
Source RPM:	systemd-253.3-2.mga9.src.rpm	CVE:
Status comment:

Description Chris Denice 2023-04-25 16:41:51 CEST

Description of problem:

When configuring a bonding ethernet device, a new ethernet mac address is created for the stack. This is unexpected as the kernel documentation rather says that the stack will inherit the mac address of one of the slave.

https://www.kernel.org/doc/html/latest/networking/bonding.html

This seems to concern only mga9.
Notice: networkmanager is off, systemd-networkd is not used, but this seems to be a systemd bug when the bonding device is created. This error is indeed reported upstream since a long time and still not fixed:

https://github.com/systemd/systemd/issues/15208


This is actually a catastrophic bug for servers configured with bonding + dhcp. Because the mac address of all servers is changing during an update of systemd, machines are no longer able to get an IP address. For this reason, I've put the severity to High.

There is a workaround mentioned in https://github.com/systemd/systemd/issues/15208, which works. It consists in creating a user-land file:

cat /etc/systemd/network/98-bond-mac.link
[Match]
Type=bond

[Link]
MACAddressPolicy=none

This file, basically, simply asks systemd to not screw the mac address...

Chris Denice 2023-04-25 16:42:02 CEST

Summary: Mac address of bonding device modified during boot => Mac address of bonding devices modified during boot

Comment 1 Lewis Smith 2023-04-25 21:27:10 CEST

Thank you for the detailed report.
"This seems to concern only mga9" begs the question: why did it not happen in M8?

Assigning directly to Thomas who handles systemd.

Assignee: bugsquad => tmb

Comment 2 Chris Denice 2023-04-26 19:03:42 CEST

>why did it not happen in M8?

I did not have the opportunity to test, I have upgraded my servers straight from mga7->mga8->mga9 without doing a reboot between mga7->mga8. Possibly there is a problem for mga8, but I cannot tell.