Bug 855409

P.S.

The host is a kvm virtual host and the lv was created in virt-manager as a virtio disk for a guest with space allocated from a lvm storage pool.

P.P.S.

Am I correct in inferring from this that deleting a virtual guest will not allow one to return the virtual disk space back into the storage pool?

(In reply to comment #0)
> Error message:   Logical volume vg_vhost01/lv_vm_inet05.harte-lyne.ca_01 is
> used by another device.

Please, try to check the output of "lsblk /dev/vg_vhost01/lv_vm_inet05.harte-lyne.ca_01" and see whether there's really no other device layered on top of this one... it shouldn't, but let's see...

# lsblk /dev/vg_vhost01/lv_vm_inet05.harte-lyne.ca_01
NAME                                                  MAJ:MIN RM   SIZE RO TYPE MOUNTPOINT
vg_vhost01-lv_vm_inet05.harte--lyne.ca_01 (dm-17)     253:17   0   3.1G  0 lvm  
└─vg_vhost01-lv_vm_inet05.harte--lyne.ca_01p1 (dm-18) 253:18   0   3.1G  0 dm

(In reply to comment #5)
> # lsblk /dev/vg_vhost01/lv_vm_inet05.harte-lyne.ca_01
> NAME                                                  MAJ:MIN RM   SIZE RO
> TYPE MOUNTPOINT
> vg_vhost01-lv_vm_inet05.harte--lyne.ca_01 (dm-17)     253:17   0   3.1G  0
> lvm  
> └─vg_vhost01-lv_vm_inet05.harte--lyne.ca_01p1 (dm-18) 253:18   0   3.1G  0 dm

The dm device with the "p1" suffix seems to be a mapping for a partition that is on the logical volume vg_vhost01-lv_vm_inet05.harte--lyne.ca_01. The mapping could be created by directly calling kpartx or partprobe on such a device.

One way the kpartx could be called automatically without notice is within /lib/udev/rules.d/40-multipath.rules if device-mapper-multipath is installed. However this should be done only on multipath devices, which does not seem to be this exact case here. So somewhere else the kpartx/partprobe must be called...

You can try:

  egrep "(kpartx|partprobe)" /etc/udev/rules.d/* /lib/udev/rules.d/*

...to see whether it's called somewhere else besides 40-multipath.rules.

# egrep "(kpartx|partprobe)" /etc/udev/rules.d/* /lib/udev/rules.d/*
/lib/udev/rules.d/40-multipath.rules:RUN+="$env{MPATH_SBIN_PATH}/kpartx -a -p p $tempnode"

It does not appear so.  Would running partprobe manually cause this?  I ask because immediately after I partition the new virtual disk with fdisk I receive an error message telling me to do so, which I have.

These are the relevant entries in the history file:

799      fdisk /dev/vg_vhost01/lv_vm_inet05.harte-lyne.ca_01
800      partprobe
801      /sbin/lvchange -aln /dev/vg_vhost01/lv_vm_inet05.harte-lyne.ca_01

(In reply to comment #7)
> # egrep "(kpartx|partprobe)" /etc/udev/rules.d/* /lib/udev/rules.d/*
> /lib/udev/rules.d/40-multipath.rules:RUN+="$env{MPATH_SBIN_PATH}/kpartx -a
> -p p $tempnode"
> 
> It does not appear so.  Would running partprobe manually cause this?  I ask
> because immediately after I partition the new virtual disk with fdisk I
> receive an error message telling me to do so, which I have.

Yes, that's exactly the cause.

It's just a new mapping created over the original LV which then needs to be removed manually as well, for example by calling:

  dmsetup remove /dev/mapper/vg_vhost01-lv_vm_inet05.harte--lyne.ca_01p1

Or, maybe more proper way would be to remove the partition (with fdisk/parted/...) and just after that calling partprobe which should detect that the partition is removed and it should remove the corresponding mapping then. I think parted can create/remove mappings directly without the extra need to call the partprobe (so its use is more straightforward in this respect).

So this not a bug then, it just needs an extra step to clean up the extra stack above the LV used. I hope this helps.