[Bug 1631474] Re: No networking with initramfs-tools 0.122ubuntu8.3 and ip=dhcp boot option

[Mathieu Trudel-Lapierre]

Seems to me like we're at the point where considering a straight revert
of the SRU would be in order (and I'm certainly considering it for the
short term), in order to move forward again at a later time when this is
more solid -- not that the end goal is still going to be happening
(though maybe not in 16.04), that is to gradually remove klibc from the
initramfs. That *will* require changes, such as we see here. If you
never fail, it's because you're not innovating...

The changes here were motivated by providing IPv6 support for MaaS
deployments in v6-only and mixed networks.

>From my understanding of things, we've now fixed (in 0.122ubuntu8.4) the
issues with ip=dhcp and some other ip= forms being parsed incorrectly.

There are still two outstanding issues:
 - boot speed (I suspect there is only a limited amount of things we can do about this, given the use of dhclient).
 - "portfast behavior", which is how we handle delays in getting a response from a DHCP server. There was definitely a regression there, for which I have a fix in my ppa at ppa:cyphermox/maas.

Are there any other outstanding issues? If so, what kernel command-line
are you using, and please include the exact messages on screen (or in
logs) so we can know what we're dealing with.

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to initramfs-tools in Ubuntu.
https://bugs.launchpad.net/bugs/1631474

Title:
  No networking with initramfs-tools 0.122ubuntu8.3 and ip=dhcp boot
  option

Status in initramfs-tools package in Ubuntu:
  Fix Committed
Status in initramfs-tools source package in Xenial:
  Fix Committed
Status in initramfs-tools source package in Yakkety:
  Fix Committed
Status in initramfs-tools source package in z-series:
  Fix Committed

Bug description:
  [Impact]

   * 0.122ubuntu8.3 of initramfs-tools no longer correctly processed
  ip=dhcp or ip=:::::eth0:dhcp

   * Regression-updates

   * The fix better parses the ip= command line argument.

  [Test Case]

   * Create a machine that boots using an nfsroot.

   * Use ip=:::::eth0:dhcp on the kernel command line.  To set up
     networking.

   * Discover that the device never comes up because, networking is not
  configured correctly.

  [Regression Potential]

   * Regressions potential is limited to machines using
  ip={""|::::*|on|any|dhcp} on the kernel command line.  As this is
  already broken regression potential is minimal.  This is common on
  machines that use nfsroot or otherwise pxe boot.

  [Other Info]

   * There are a number of other issues in this code base that are not solved by this fix.
     - The ?*:?*:?*:?*: use case falls through to the default case, and likely breaks there.  As such static assignment via ip= appears broken
     -
   * The networking configuration does not strictly follow the kernel documentation as described https://www.kernel.org/doc/Documentation/filesystems/nfs/nfsroot.txt . This should be fixed.

  This bug is a regression of changes made under bug 1628306.

  ====================Original Bug Description Follows==================

  initramfs-tools 0.122ubuntu8.3 introduced a serious regression where
  networking is not initialized when the boot option "ip=dhcp" is
  provided. We are seeing this problem in AWS, but cannot confirm if
  this issue is specific to AWS or will occur with different hardware or
  in different environments.

  Removing "ip=dhcp" from the boot options with 0.122ubuntu8.3 results
  in networking being configured.

  The issue does not occur with 0.122ubuntu8.2 or previous versions when
  "ip=dhcp" is set.

  AWS has no console so debugging is not a trivial task. I do have a
  console log with some output, and will update this bug shortly with
  it.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/initramfs-tools/+bug/1631474/+subscriptions