Bug 237415

Created attachment 153251 [details]
Working initrd image

Created attachment 153252 [details]
Broken initrd (causes kernel panic)

After analyzing the two nash init scripts in the initrd files attached, I
noticed that the working one had lvm commands, and the other did not (as I
suspected).  After studying and debugging the mkinitrd script, I isolated the
problem to the following function call:

lvshow() {
    lvm.static lvs --ignorelockingfailure --noheadings -o vg_name \
        $1 2>/dev/null | egrep -v '^ *(WARNING:|Volume Groups with)'
}

The problem stems from my /etc/fstab having all of the LVM partitions called
with LABEL=/, LABEL=/var, and LABEL=/home.  The lvshow function (lvm.static lvs)
does not support LABEL=/ to locate a volume group.

The work-around was to change LABEL=/ to /dev/mapper/VolumeGroup00-root and this
was enough to output the correct nash script to start LVM at boot.

The permanent fix is to determine if Fedora is going to use LABEL=/ in fstab
(and use it in /etc/crypttab), and if so, fix the mkinitrd function to locate
LVM volume groups by label.

See also
    bug 241949

    bug 242043

Whatever is the cause, it has not been fixed as yet.

It appears that the structure and content of /etc/mdadm.conf has changed between
my (running) FC6 and my attempted installation of FC7 (two different hardware
systems)

The FC6 /etc/mdadm.conf shows 3 raid devices (boot,swap,/dev/rootvg/root)

# mdadm.conf written out by anaconda
DEVICE partitions
MAILADDR root
ARRAY /dev/md2 super-minor=2
ARRAY /dev/md0 super-minor=0
ARRAY /dev/md1 super-minor=1


The F7 mdadm.conf (in the initrd file) shows ONLY TWO raid devices (boot,swap)

# mdadm.conf written out by anaconda
DEVICE partitions
MAILADDR root
ARRAY /dev/md0 level=raid1 num-devices=2 uuid=47bba70b:b76ffd5f:816f55b8:cf2ee184
ARRAY /dev/md1 level=raid1 num-devices=2 uuid=36c22074:b238a704:85d99d8b:e9dafa99
[root@hoho0 etc]# 

(Note that the long lines wrapped)

Since the 3rd raid device was not identified in the mdadm.conf file, this could
be the reason why the lvm device was not found (it is on the md2 device)

I did not see any LABEL s in the /etc/fstab files

This is from the FC6 system
/home/user1/Desktop/initrd/etc

[root@hoho0 etc]# cat /etc/fstab
/dev/rootvg/root        /                       ext3    defaults        1 1
/dev/md0                /boot                   ext3    defaults        1 2
devpts                  /dev/pts                devpts  gid=5,mode=620  0 0
tmpfs                   /dev/shm                tmpfs   defaults        0 0
proc                    /proc                   proc    defaults        0 0
sysfs                   /sys                    sysfs   defaults        0 0
/dev/md1                swap                    swap    defaults        0 0
[root@hoho0 etc]# 

This is from the F7 /mnt/sysimage/fstab

[root@hoho0 Desktop]# cat fstab
/dev/rootvg/root        /                       ext3    defaults        1 1
/dev/md0                /boot                   ext3    defaults        1 2
devpts                  /dev/pts                devpts  gid=5,mode=620  0 0
tmpfs                   /dev/shm                tmpfs   defaults        0 0
proc                    /proc                   proc    defaults        0 0
sysfs                   /sys                    sysfs   defaults        0 0
/dev/md1                swap                    swap    defaults        0 0
[root@hoho0 Desktop]# 

My mdadm.conf missed the last raid array too (only md0, not md1 showing). I'd
already rebuilt the initrd first, so I don't know if there was a problem with
that too.

adding mdadm -Es output to /etc/mdadm.conf via rescue mode allowed me to boot up.

Created attachment 155963 [details]
lvmdump from FC7 install DVD running in 'rescue' mode

This is obviously not from the disk installed FC7 (that crashed on boot), but
rather from the F7 DVD running in 'rescue' mode. Might be useful.

I must have missed your 01:12 EST message last night. Saw it in email this morning.

Yes - Welcome Fedora..

I am writing this bug comment from F7

As well as doing your mdadm -Es trick (UUID in caps doesn't matter), I had to
rebuild the initrd.

Thanks much

Ref: bug 241949 #c14

There definitely seems to be a bug in mkinitrd, specifically failing to
correctly grok LVM configurations, and subsequently included the necessary
commands in nash.

My resolution was basically to completely reconstruct my initrd manually using
cpio, gzip and vim ... reminding me of my Linux From Scratch experiences from
years ago :)

It worked, but I fear the next kernel update like the Devil ;)

Hmmm, mdadm -Es helped me. There were the same UUID's for both disks in
mdadm.conf (I'm using RAID 1).

Just tested a fresh install with today's Fedora 8 candidate with LVM on top of
RAID (with multiple arrays) and it booted fine.  There have been some fixes in
this area, so closing as NEXTRELEASE

*** Bug 300701 has been marked as a duplicate of this bug. ***