mount script doesn't poll if root device has been initialized, preventing udev+mdadm/lvm/cryptsetup to do its job

[Dominik Sauer]

script '/usr/share/initramfs-tools/scripts/init-premount/udev' launches cold plugging of devices on background, then terminates immediately. this causes mdadm md array assembly to fail, because it doesn't see it's /dev/sd?? devices yet. Ugly hack is to sleep 20 after launching the coldplug.

The script that mounts the root fs should enter a loop checking if all the root device has come up, and if that has not happened for a while start an md device in degraded mode.

Something like
  If rootfs depends on a md device,
    while timeout not reached, do:
      if root md device has come up, continue booting
      wait 0.5 seconds
    done.
    if timeout has been reached:
      mdadm --run <root md device>

The udev lvm/crypt/ hotplug then goes on, so...
  if rootfs then depends on a lvm device
     while timeout not reached ....

(check https://wiki.ubuntu.com/BootDegradedRaid)

[Daniele Pilenga (daniele-pilenga)]

The scripts in initramfs that should attach the root volume do not find any device as the kernel has not yet finished probing. Adding a little sleep helps, but it's not a solution.

[Nicolas Schmerber (nschmerber)]

Hi,

I have the same problem on raid1 initialization at startup. Config is :

Kernel : 2.6.20-13 of feisty
HW : 2 Sata drive on a ata_piix module dependand card, assembled in soft raid1

At grub prompt, without a break=premount, the kernel launches and tries to assemble the raid1 array, without seeing the 2 drives, the ata_piix module is only loaded 3seconds later but too late.
with the break point i can load the module i need, and just continue without any problems.

Regards

[darthvader (sarmbruster)]

I can confirm this bug. It should be considered as a blocker for the Feisty release.

[Dominik Sauer (dominik-sauer)]

The sleep 20 would not be a solution, but busy waiting, like

wait=yes
while [ $wait == yes ] ; do
  (do some mdstuff)
  if [ $? == 0 ]; then
    wait=no
  else
    sleep 3
  fi
done

would do.
IMHO.

[Dominik Sauer (dominik-sauer)]

The sleep 20 would not be a solution, but busy waiting, like

wait=yes
while [ $wait == yes ] ; do
  (do some mdstuff)
  if [ $? == 0 ]; then
    wait=no
  else
    sleep 3
  fi
done

would do.
IMHO.

[Christoph Rauch (christoph-rauch)]

I get the same effect (mdadm does not start arrays), similar setup as Dominik (ata_piix, raid1), but apparently a different cause for the problem.

According to ps udevd is not started at all when mdadm bails out. If I start the udev start script manually the device nodes appear and the mdadm script can initialize the arrays.

[Christoph Rauch (christoph-rauch)]

I solved my problem by adding "auto=yes" to the device section in mdadm.conf and afterwards rerunning
update-initramfs -u -k all

The system boots now, a pristine Feisty installation although would not boot.

[Dominik Sauer (dominik-sauer)]

well, the udev is started up (and the "coldplug" events triggered) in
/usr/share/initramfs-tools/scripts/init-premount/udev; which is the very
script i add the sleep 20 at the end of.

On 4/11/07, Christoph Rauch <email address hidden> wrote:
>
> I get the same effect (mdadm does not start arrays), similar setup as
> Dominik (ata_piix, raid1), but apparently a different cause for the
> problem.
>
> According to ps udevd is not started at all when mdadm bails out. If I
> start the udev start script manually the device nodes appear and the
> mdadm script can initialize the arrays.
>
>
> ** Changed in: initramfs-tools (Ubuntu)
> Status: Unconfirmed => Confirmed
>
> --
> initramfs-tools script doesn't wait for device initialization, preventing
> mdadm to assemble md arrays
> https://bugs.launchpad.net/bugs/83025
> You received this bug notification because you are a direct subscriber
> of the bug.
>

[ceg (ceg)]

To fix the issue the md, cryptsetup, etc. scripts that set up the root device need to wait for their device to get set up by udev and timeout after a while (possibly running a degraded raid).

https://wiki.ubuntu.com/HotplugRaid
https://wiki.ubuntu.com/BootDegradedRaid

[ceg (ceg)]

md and lvm devices are actually handled by udev rules.

For one, cryptsetup handling needs go into a waiting loop (Bug #247153), or also be handled by udev rules.

For two, the initramfs script that mounts the rootfs needs go into a waiting loop (0.5 sec intervals) for their devices (going on quickly when device gets available) and timeout after a while.

[ceg (ceg)]

/usr/share/initramfs-tools/scripts/local in 8.04 contains a checking loop.

[ceg (ceg)]

initramfs/scripts/local in 8.04 does have a timeout loop for the root device.

The remaining mdadm bug numbers are listed on https://wiki.ubuntu.com/BootDegradedRaid