[Bug 1770082] Re: systemd-networkd not renaming devices on boot

Wed May 16 00:55:28 UTC 2018

> Note, that uvt-kvm is going to use cloud-init; how are you making
> sure that cloud-init isn't doing the rename itself?

I instrumented the kernel. I added a call to dump_stack() in the
function that does interface renaming: dev_change_name() in
net/core/dev.c. That showed that the process that sent the netlink
message was systemd-udevd:

[    1.007200] virtio_net virtio3 ens16: renamed from eth0
[    1.008453] CPU: 0 PID: 124 Comm: systemd-udevd Not tainted 4.4.59+ #1
[    1.009887] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1ubuntu1 04/01/2014
[    1.011938]  0000000000000286 00000000d38795d9 ffff88001ed3b970 ffffffff813f3f63
[    1.012435]  ffff88001f2d0000 0000000000000000 ffff88001ed3b9d8 ffffffff8170669b
[    1.012435]  ffff88001ed3b9c8 ffff88001ed3b9a0 ffffffff812255af 0000000030687465
[    1.012435] Call Trace:
[    1.012435]  [<ffffffff813f3f63>] dump_stack+0x63/0x90
[    1.012435]  [<ffffffff8170669b>] dev_change_name+0x25b/0x2e0
<call chain back to syscall snipped>

...

[    2.336114] virtio_net virtio3 ens3: renamed from ens16
[    2.336121] CPU: 0 PID: 454 Comm: systemd-udevd Not tainted 4.4.59+ #1
[    2.336122] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1ubuntu1 04/01/2014
[    2.336124]  0000000000000286 0000000027d63c85 ffff88001c59b970 ffffffff813f3f63
[    2.336126]  ffff88001f2d0000 0000000000000000 ffff88001c59b9d8 ffffffff8170669b
[    2.336128]  0000000000000001 ffff88001c59b9b0 ffffffff810b3f1c 0000003631736e65
[    2.336130] Call Trace:
[    2.336135]  [<ffffffff813f3f63>] dump_stack+0x63/0x90
[    2.336138]  [<ffffffff8170669b>] dev_change_name+0x25b/0x2e0
<call chain back to syscall snipped>

Notice in both cases: "Comm: systemd-udevd". Also notice the change in
PID, suggesting that the first one is in initrd and the second one is
not.

I then grepped the source code for netlink stuff and quickly narrowed in
on rename_if in udev-events.c.

> Xenial by default doesn't use netplan; it still uses eni; the network
> configuration generation in cloud-init using eni
> does create a 70-persistent-net.rules file that handles renames on
> subsequent reboots; in a netplan world (bionic)
> the .link file is meant to be the equivalent of a .rules file.

I am aware of this. (I've hacked on netplan quite a bit.)

The issue is that a .link file does not seem to be functionally
equivalent to a .rules file. I don't know if this difference is
deliberate or an oversight. I will open an issue upstream and ask.

> I've not attempted to determine if systemd-udev in bionic would
> respect renames from .rules file; it certainly seems
> odd to have .rules files allow renames independent of name_assign_type
> value where .link files do.

The end-user tested with a .rules file on Bionic and reported that it
worked. I have also just verified that it does respect renames from a
rules file regardless of name_assign_type. (The codepaths are unchanged
between systemd-229 and systemd-237.)

-- 
You received this bug notification because you are a member of Ubuntu
Sponsors Team, which is subscribed to the bug report.
https://bugs.launchpad.net/bugs/1770082

Title:
  systemd-networkd not renaming devices on boot

Status in netplan:
  Incomplete
Status in systemd package in Ubuntu:
  New

Bug description:
  === systemd issue ===

  Renaming devices doesn't seem to work.

  If I disable all other network configuration and create
  /etc/systemd/network/10-network.link with:

  [Match]
  MACAddress=52:54:00:c1:c9:bb

  [Link]
  Name=myiface3

  I expect this to cause the device with that MAC address to be renamed
  to  myiface3. However, when I reboot, I instead see:

  $ ip l
  1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN mode DEFAULT group default qlen 1000
      link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
  2: ens3: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN mode DEFAULT group default qlen 1000
      link/ether 52:54:00:c1:c9:bb brd ff:ff:ff:ff:ff:ff

  The device is not renamed.

  This link file is pretty much identical to Example 2 in
  https://www.freedesktop.org/software/systemd/man/systemd.link.html.

  The renaming does work if I boot with net.ifnames=0, and oddly, it
  also works if I unbind the device and rebind it as netplan apply does.
  No setting of NamePolicy seems to help.

  === Original Bug ==

  'set-name:' doesn't change the name of a network interface on boot, it
  only works when you do netplan apply.

  Say I take this 50-cloud-init.yaml file:

  # This file is generated from information provided by
  # the datasource.  Changes to it will not persist across an instance.
  # To disable cloud-init's network configuration capabilities, write a file
  # /etc/cloud/cloud.cfg.d/99-disable-network-config.cfg with the following:
  # network: {config: disabled}
  network:
      version: 2
      ethernets:
          ens3:
              dhcp4: true
              match:
                  macaddress: 52:54:00:de:bd:f6
              set-name: ens3

  Say I change set-name to 'myiface3' and reboot. I expect that the
  device will be called myiface3 and brought up fine with dhcp. However,
  instead I see:

  $ ip a
  1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
      link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
      inet 127.0.0.1/8 scope host lo
         valid_lft forever preferred_lft forever
      inet6 ::1/128 scope host
         valid_lft forever preferred_lft forever
  2: ens3: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN group default qlen 1000
      link/ether 52:54:00:de:bd:f6 brd ff:ff:ff:ff:ff:ff

  The name has not been changed, and the device has not been brought up.

  If I run netplan apply however, I see the following:

  1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
      link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
      inet 127.0.0.1/8 scope host lo
         valid_lft forever preferred_lft forever
      inet6 ::1/128 scope host
         valid_lft forever preferred_lft forever
  3: myiface3: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel state UP group default qlen 1000
      link/ether 52:54:00:de:bd:f6 brd ff:ff:ff:ff:ff:ff
      inet 192.168.122.151/24 brd 192.168.122.255 scope global dynamic myiface3
         valid_lft 3575sec preferred_lft 3575sec
      inet6 fe80::5054:ff:fede:bdf6/64 scope link
         valid_lft forever preferred_lft forever

  So names are successfully changed with netplan apply.

  This seems to be some udev-related timing or priority issue that I'm
  still trying to hunt down.

  This breaks some forms of migration in certain cloud environments.

To manage notifications about this bug go to:
https://bugs.launchpad.net/netplan/+bug/1770082/+subscriptions