Bin*_*rus 5 debian bridge mac-address
我刚刚将两台服务器从Debian 10 (Buster) 升级到Debian 11 (Bullseye)。此后,我再也无法通过网络联系到他们。经过一番排查,发现问题如下:
Both machines have a bridge device configured. Obviously, the algorithm which Debian uses to assign MAC addresses to bridge devices has changed from version 10 to 11. After the upgrade, the bridge device on the first server had the same MAC address as the bridge device on the second server, which for sure has not been the case before.
One of the answers there claims that a bridge is a purely internal device and that therefore a bridge's MAC address does not matter. However, this is obviously wrong. At least in my case, packets from both machines were outgoing with the hardware source address being the bridge's MAC address, and the network ports on both machines were processing incoming packets only if they were destined for the bridge's MAC address.
Since that MAC address was the same on both machines, the network became unusable, which is completely logical and understandable.
How can I make Debian generate different MAC addresses for bridge devices which are on different machines (or even on the same machine, but that's currently not my issue)?
Browsing in Internet I found this bug report on systemd-udev related to Debian 11 bridges: systemd-udev interferes with MAC addresses of interfaces it's not supposed to do #21185:
Run Code Online (Sandbox Code Playgroud)ash.in.ffho.net:~# for n in 0 1 2 3; do ip l add br$n type bridge; done ash.in.ffho.net:~# ip -br l br0 DOWN d2:9e:b3:32:53:42 <BROADCAST,MULTICAST> br1 DOWN e2:00:44:2c:5b:70 <BROADCAST,MULTICAST> br2 DOWN 0e:99:b7:42:f0:25 <BROADCAST,MULTICAST> br3 DOWN a6:3f:5f:b5:9a:d6 <BROADCAST,MULTICAST> ash.in.ffho.net:~# for n in 0 1 2 3; do ip link del br${n}; done ash.in.ffho.net:~# for n in 0 1 2 3; do ip l add br$n type bridge; done ash.in.ffho.net:~# ip -br l br0 DOWN d2:9e:b3:32:53:42 <BROADCAST,MULTICAST> br1 DOWN e2:00:44:2c:5b:70 <BROADCAST,MULTICAST> br2 DOWN 0e:99:b7:42:f0:25 <BROADCAST,MULTICAST> br3 DOWN a6:3f:5f:b5:9a:d6 <BROADCAST,MULTICAST>
As you can see, the bridges were created with low-level commands, but they always inherit the same MAC address value: a systemd
component interferes and sets the MAC address.
One can see this in action using ip monitor link
:
22: brtest0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN group default
link/ether 0a:ae:c3:0d:ec:68 brd ff:ff:ff:ff:ff:ff
22: brtest0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN group default
link/ether 1a:d0:fc:63:c1:71 brd ff:ff:ff:ff:ff:ff
Deleted 22: brtest0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN group default
link/ether 1a:d0:fc:63:c1:71 brd ff:ff:ff:ff:ff:ff
23: brtest0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN group default
link/ether 4e:e9:11:dd:a5:aa brd ff:ff:ff:ff:ff:ff
23: brtest0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN group default
link/ether 1a:d0:fc:63:c1:71 brd ff:ff:ff:ff:ff:ff
Run Code Online (Sandbox Code Playgroud)
You can see how the MAC address initially random is overwritten to a fixed one, twice to the same value for a given bridge name.
An other side effect is that when interface is set administratively UP, the bridge operational status becomes DOWN instead of UNKNOWN initially because of this (see these answers of mine on SU and SF mentioning behaviors about DOWN and UNKNOWN: How does Linux determine the default MAC address of a bridge device? , linux ipv6 bridge address does not work when mac address is forced). Anyway this doesn't matter anymore once its first bridge port is attached.
Doing the same experiment inside a network namespace (eg: ip add netns experiment
and ip netns exec experiment bash -l
before running above commands twice) where systemd-udevd
does not interfere will show the usual behavior of having different random addresses each time.
This is an effect of systemd ecosystem and doesn't happen on systems not running systemd (or older versions of systemd). One proposed fix is to use:
Run Code Online (Sandbox Code Playgroud)# /etc/systemd/network/90-bridge.link [Match] OriginalName=br* [Link] MACAddressPolicy=random
but it appears the real fix is to change the file that participates in generating this "stable random" value, as described there: https://wiki.debian.org/MachineId
Each machine should have a different value. This is especially important for cloned VMs from a base template. The relation between machine-id
and the way the bridge "stable" MAC address is generated is mentioned in the patch having implemented the (quite breaking) change:
=== This patch
This patch means that we will set a "stable" MAC for pretty much any virtual device by default, where "stable" means keyed off the machine-id and interface name.
It was also mentioned that this would be having impacts , but this was shrugged off.
This is not limited to interfaces of type bridge but to any interface that would generate a random MAC address: for example types veth
, macvlan
tuntap
are also affected.
I could verify that the same bridge name would get a different "stable random" value after doing the operations described in Debian's link:
Run Code Online (Sandbox Code Playgroud)rm -f /etc/machine-id /var/lib/dbus/machine-id dbus-uuidgen --ensure=/etc/machine-id dbus-uuidgen --ensure
现在ip monitor
在删除和重新创建时为相同的网桥名称提供一个新的 MAC 地址:32:ee:c8:92:9f:e8 而不是 1a:d0:fc:63:c1:71 brtest0
。
Deleted 23: brtest0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN group default
link/ether 1a:d0:fc:63:c1:71 brd ff:ff:ff:ff:ff:ff
24: brtest0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN group default
link/ether da:72:b6:63:23:e5 brd ff:ff:ff:ff:ff:ff
24: brtest0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN group default
link/ether 32:ee:c8:92:9f:e8 brd ff:ff:ff:ff:ff:ff
Run Code Online (Sandbox Code Playgroud)
结论:
由于桥接 MAC 地址现在是手动设置的,桥接器将不再继承设置为桥接端口的其他接口的 MAC 地址之一,包括通常的永久(物理或虚拟机)接口,这些接口预计具有不同的 MAC 地址。两个系统使用相同machine-id
且相同的网桥名称(例如:br0
),且此类网桥参与路由(即:网桥上配置了 IP 地址,但即使没有,网桥也可以根据其设置发出与桥接相关的其他帧)同一 LAN 上的网络将发出具有相同源 MAC 地址(网桥)的帧,可能会中断路径中的交换机,并且无论如何都会忽略来自对等方的相同源 MAC 地址。
归档时间: |
|
查看次数: |
3766 次 |
最近记录: |