Udev, LVM, Initramfs & Odd Boot/Shutdown Warnings

sk3l

I had been following the discussion concerning the shift of udev requiring early userspace fs mounting of /usr for some time. I had simply masked udev and decided to defer tackling it until later, but I knew sooner or later it would stabilize, and I would eventually need to adapt, so I thought it best to just get out in front of the changes rather than get left behind. I use an LVM2 setup containing everything except root. I thought a long while about what my strategy would be to play along with the new rules. I considered pushing /usr back into /. I thought about using SteveL's patch to udev. In the end, I settled on biting the bullet and leveraging an initramfs approach. It seems there were a handful of different ways to generate the initramfs, dracut, hand-rolling your own or letting genkernel do the heavy lifting. I chose the genkernel method, because I couldn't conceive of the need to execute anything but the vanilla, generic steps in the initramfs startup process.

After an extended period of nail biting and anxiety, in th end, this migration was not as painful as I thought it could be. I moved udev and all of the related packages to unstable, genkernel'd an initramfs, fixed up grub.conf, and that was that. Everything seems to run fine, with 2 small exceptions.

1) On boot, I notice an error (presumably of lesser severity) that e2fsck cannot continue checking /usr, and it aborts.
2) On shutdown, the Logical Volume Manager complains about attempting to bring down the LV's for usr and tmp. It seems to think those file systems are still in use.

Looking around the forum, I've noticed a few others talking about similar problems. From what I can gather, the initramfs mounts /usr as r/w which accounts for the issue with e2fsck. Using the genkernel'd initramfs, I'm guessing the options for manipulating the early mount of usr is limited. I don't have a solid idea about what could be behind the 2nd problem in bringing all of the LV's offline without incident. For the time being, I can prevent the 1st error by changing the /usr entry in fstab to noauto as well as disabling the file system check (setting last column to 0). This is not really an ideal scenario, as I'd prefer to check usr on boot just like every other filesystem. I don't have any ideas on how to work around the 2nd error on shutdown, or where to begin to look. Here is the state of the relevant packages:

jopeto · Tux's lil' helper Joined: 02 Jul 2012 Posts: 106

A few days ago I noticed similar issues. You can take a look at the forum thread which I started, particularly the last few posts.
https://forums.gentoo.org/viewtopic-t-933008-highlight-.html

greyspoke · Apprentice Joined: 08 Jan 2010 Posts: 171

I have /usr on an lvm as a separate file system. I fsck it by doing it in the initramfs (I use a custom initramfs and init script), or alternatively by using my early mount script - though others who have used it have experienced issues. You've obviously heard of steveL's approach.

I am not sure if mounting /usr read-only will get it fsck'd though, that may only happen with / which has its own init script to handle unmounting, checking and re-mounting read-only.

As for the lvm messsages at shutdown, I haven't been able to find a way round them however I mount /usr. I don't get them for tmp though, I get them for /usr and /var (which I also mount early as at one time some early init scripts wanted to access it before localmount (alsasound was definitely one). I am not sure if the problem is that they are mounted, or that they are unmounted and consequently lvm cannot access something it wants that is on them. Possibly the shutdown scripts recognise a separate /usr and unmount it and re-mount it read-only along with / (once everything is on /usr including shutdown, this will have to happen). I have never had messages about /usr not having been cleanly unmounted, so either it is getting unmounted or it is getting properly flushed and so on before powerdown.

I suppose that if you do all your lvm activation in your initramfs and don't need logical volumes to be started or stopped dynamicall, you don't actually need lvm2 in your init scripts at all. The volumes would not be de-activated before shutdown, but I don't think that is necessary. What do your messages say? I will post mine when I have access to them later.

greyspoke · Apprentice Joined: 08 Jan 2010 Posts: 171

Here's a snippet of my rc.log, it appears it is just /usr that is causing the problem. You'll see there is no message that it is getting unmounted. But the error message look like there is a missing variable after "on", so something else is going wrong apart from a busy device.

ph03 · n00b Joined: 14 Jan 2005 Posts: 39

Hi!

Have you been able to solve this issue? I'm having it too and I am still looking for an solution. I'm on Sabayon 10 btw...

Greets Janick

greyspoke · Apprentice Joined: 08 Jan 2010 Posts: 171

hujuice · Guru Joined: 16 Oct 2007 Posts: 336 Location: Rome, Italy

More than one year later, I run exactly in the same problems where sk3l runs, in vaious flavour.

For example, a laptop where / and /usr are managed by LVM has both the mentioned errors, while a little development server with separate /usr manged by LVM has the shutdown error only.

But maybe things are changed.

I took a look inside my genkernel generated initramfs file.
Critical files are a handful of executable files, for example lvm (and all related symlinks)