Hi,

we have most likely observed the same, or a very similar, bug when
upgrading two test systems from Wheezy to Jessie.

Important: We use LVM for everything but / and /boot.

The default fstab looks like this:
LABEL=boot              /boot   ext3   
defaults,errors=remount-ro,noatime              0 2
LABEL=slash             /       ext3   
defaults,errors=remount-ro,noatime              0 1
proc                    /proc   proc   
defaults                                        0 0
LABEL=tmp               /tmp    xfs    
defaults,noatime,nosuid,nodev                   0 0
LABEL=var               /var    xfs    
defaults,noatime                                0 0
LABEL=usr               /usr    xfs    
defaults,noatime                                0 0
LABEL=home              /home   xfs    
defaults,noatime,nosuid,nodev                   0 0
LABEL=swap              none    swap    sw

blkid output like this:
/dev/sda1: LABEL="boot"
/dev/sda2: LABEL="slash"
/dev/sda3: TYPE="LVM2_member"...
/dev/mapper/vg00-usr: LABEL="usr"...
/dev/mapper/vg00-tmp: LABEL="tmp"...
/dev/mapper/vg00-var: LABEL="var"...
/dev/mapper/vg00-home: LABEL="home"...
/dev/mapper/vg00-swap: LABEL="swap"...
/dev/mapper/vg00-data: LABEL="data"...

The initramfs loops in "Begin: Running /scripts/local-block ... done."
until it gives up.
It seems, it did not recognize that /usr is on a logical volume and did
not attempt to activate it. Indeed, the device nodes are missing and
manually running "vgchange -ay" in the emergency shell allows the boot
to continue successfully.

Modifying fstab like this fixes things (all other filesystems continue
to use LABEL):
...
/dev/mapper/vg00-usr    /usr    xfs    
defaults,noatime                                0 0
...

So, I suspect this bug is triggered by the use of UUID or LABEL in fstab
for a split /usr when using any form of device-mapper (not only dm-crypt).

scripts/local-block/lvm2 has this code, that looks very suspect:
# Make sure that we have a d-m path
dev="${dev#/dev/mapper/}"
if [ "$dev" = "$1" ]; then
    return 1
fi

No idea how to fix this except by doing an unconditional LVM activation.

Regards,
Marc


-- 
To UNSUBSCRIBE, email to debian-kernel-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Archive: https://lists.debian.org/55129b9d.5020...@1und1.de

Reply via email to