Bug#568838: Init script finishes before /dev entries created causing boot to fail

[Alexander Clouter]

Hi,

I also have been battling with this race condition for the past few 
years.  I have a non-complicated non-RAID (although I do see this on my 
software RAID boxen too) where /usr, /var and /home are part of an LVM 
however my root is not.

My /etc/fstab typically looks like:
----
LABEL=swap      none            swap    sw                               0  0

LABEL=root      /               auto    relatime                         0  1

LABEL=usr       /usr            auto    relatime,nodev                   0  2
LABEL=var       /var            auto    relatime,nodev,nosuid            0  2
LABEL=home      /home           auto    relatime,nodev,nosuid,noexec     0  2

tmpfs           /tmp            tmpfs   nodev,nosuid,noexec,size=32M     0  0
tmpfs           /var/tmp        tmpfs   nodev,nosuid,noexec,size=8M      0  0
varrun          /var/run        tmpfs   nodev,nosuid,noexec,size=512k    0  0
varlock         /var/lock       tmpfs   nodev,nosuid,noexec,size=96k     0  0

#/dev/hda       /media/cdrom0   auto    ro,user,noauto                   0  0
#/dev/hdf       /media/cdrom1   auto    ro,user,noauto                   0  0
#/dev/fd0       /media/floppy0  auto    rw,user,noauto                   0  0
----

When my box boots I can often stumble onto:
----
Setting up LVM Volume Groups  Reading all physical volumes.  This may 
take a while...
  Found volume group "lvm-oghma" using metadata type lvm2
  3 logical volume(s) in volume group "lvm-oghma" now active
.
Checking file systems...fsck 1.41.3 (12-Oct-2008)
/sbin/fsck.xfs: XFS file system.
/sbin/fsck.xfs: XFS file system.
/sbin/fsck.xfs: LABEL=home does not exist
fsck died with exit status 8
failed (code 8).
File system check failed. A log is being saved in /var/log/fsck/checkfs if that location is writable. Please repair the file sy
A maintenance shell will now be started. CONTROL-D will terminate this shell and resume system boot. (warning).
Give root password for maintenance
(or type Control-D to continue): 
----

This happens about 80% of the time on some kit, 10% on others.  
Sometimes it grumbles that LABEL=usr or LABEL=var does not exist instead 
so it is very obviously a race condition.  Looking at what could be 
going on, /etc/rcS.d/S26lvm2 shoots through without pausing and so udev 
is in a state where it has not quite created all the softlinks in 
/dev/disk/by-label in time for when /etc/rcS.d/S30checkfs.sh kicks in.

This is trivially fixed by amendment to /etc/init.d/lvm2 that 'StalkR' 
suggested, and is done by the patch I have attached.

Now when my boxes boot I get:
----
Setting up LVM Volume Groups  Reading all physical volumes.  This may 
take a while...
  Found volume group "lvm-oghma" using metadata type lvm2
  3 logical volume(s) in volume group "lvm-oghma" now active
.
Checking file systems...fsck 1.41.3 (12-Oct-2008)
/sbin/fsck.xfs: XFS file system.
/sbin/fsck.xfs: XFS file system.
/sbin/fsck.xfs: XFS file system.
done.
Setting kernel variables (/etc/sysctl.conf)...done.
Mounting local filesystems...Filesystem "dm-0": Disabling barriers, not supported by the underlying device
XFS mounting filesystem dm-0
Filesystem "dm-1": Disabling barriers, not supported by the underlying device
XFS mounting filesystem dm-1
[snipped]
----

Please include this, get it pushed into 'lenny' too and 'squeeze' as it 
really is a nasty showstopper that makes booting of servers potentially 
very unreliable.

Cheers

-- 
Alexander Clouter
.sigmonster says: What foods these morsels be!
-------------- next part --------------
A non-text attachment was scrubbed...
Name: lvm-race-fix.patch
Type: text/x-diff
Size: 303 bytes
Desc: not available
URL: <http://lists.alioth.debian.org/pipermail/pkg-lvm-maintainers/attachments/20100409/467be476/attachment.patch>