Bug 144586

Summary:	Race between udev device creation and boot.lvm vgscan with tmpfs /dev
Product:	[openSUSE] SUSE Linux 10.1	Reporter:	Andreas Kleen <ak>
Component:	Basesystem	Assignee:	Kay Sievers <kasievers>
Status:	RESOLVED FIXED	QA Contact:	E-mail List <qa-bugs>
Severity:	Major
Priority:	P5 - None	CC:	kernel01
Version:	Beta 1
Target Milestone:	Beta 2
Hardware:	Other
OS:	Other
Whiteboard:
Found By:	Other	Services Priority:
Business Priority:		Blocker:	---
Marketing QA Status:	---	IT Deployment:	---

Description Andreas Kleen 2006-01-21 01:12:48 UTC

Booted with a self compiled kernel a system running 10.1beta1 doesn't 
mount file systems on LVs at boot. 

The problem seems to be that when the vgscan in boot.lvm runs the /dev 
in tmpfs is only very sparsely populated and missing most of the block 
devices. Without block devices vgscan doesn't find the volume group. 
Previously the problem was hidden because /dev was on disk and the block 
device nodes needed by vgscan were there anyways from the previous boot.

While I can only reproduce it with a self compiled non modular kernel
right now I suspect the race is there with the modular rpm too and
will break something eventually.

I work arounded it here by adding a sleep 3 on top of boot.lvm,
but it's not a full or good solution of course.

Comment 1 Kay Sievers 2006-01-21 12:59:23 UTC

Does it help, to make /etc/init.d/boot.device-mapper depend on boot.udev instead on boot.rootfsck (run insserv after it, to update the .depend* files)?

Comment 2 Andreas Kleen 2006-01-22 02:29:24 UTC

With that it boots yes.

Comment 3 Thomas Fehr 2006-01-23 10:52:25 UTC

Added boot.udev to Required-Start section of init script

Comment 4 Andreas Kleen 2006-02-07 12:30:25 UTC

I updated to beta3 and the problem is there again. The dependencies 
include udev for device-mapper, so there must be some other issue there too.

Comment 5 Thomas Fehr 2006-02-07 12:36:23 UTC

Reassigne to mintainer of udev, it seems that devices in tmpfs are not 
present even if boot.udev has been finished.

Comment 6 Kay Sievers 2006-04-13 08:53:00 UTC

Please test with current release and update the version or close the bug.

Comment 7 Christoph Thiel 2006-04-23 10:58:48 UTC

Andi? What's the status on this one?

Comment 8 Andreas Kleen 2006-04-24 04:04:36 UTC

I'm retesting with RC2

Comment 9 Andreas Kleen 2006-04-24 07:34:58 UTC

I rebooted a few times and the race didn't trigger so I suppose it's fixed