[linux-lvm] Problem with vgscan and md

[Luca Berra]

Hello,
I found a probable bug in vgscan up to LVM 1.0.5 due to lv device nodes 
renumbering

i found the problem in a shared storage environment, but it can be 
reproduced on a single machine

the problem can be reproduced with 2 md devices
create 2 vg each composed of 1 md device
let's say
VG /dev/vg00
Device 109:0
LV /dev/vg00/lvol1
Device 58:0
LV /dev/vg00/lvol2
Device 58:1
LV /dev/vg00/lvol3
Device 58:2
PV /dev/md0

VG /dev/vg01
Device 109:1
LV /dev/vg01/lvol1
Device 58:3
LV /dev/vg01/lvol2
Device 58:4
LV /dev/vg01/lvol3
Device 58:5
PV /dev/md1

now deactivate all VGs
raidstop /dev/md0
run vgscan -v

you have

VG /dev/vg01
Device 109:0
LV /dev/vg01/lvol1
Device 58:0
LV /dev/vg01/lvol2
Device 58:1
LV /dev/vg01/lvol3
Device 58:2
PV /dev/md1

vgchange -a y /dev/vg01

now raidstart /dev/md0

you need to run vgcan again, to see vg00

vgscan renumbers device node for vg00 so you get

VG /dev/vg00
Device 109:0
LV /dev/vg00/lvol1
Device 58:0
LV /dev/vg00/lvol2
Device 58:1
LV /dev/vg00/lvol3
Device 58:2
PV /dev/md0

vgscan cannot renumber device nodes for vg01 (which is active) so you 
also get

VG /dev/vg01
Device 109:0
LV /dev/vg01/lvol1
Device 58:0
LV /dev/vg01/lvol2
Device 58:1
LV /dev/vg01/lvol3
Device 58:2
PV /dev/md1

so we got two DIFFERENT vgs with the SAME device files

the need to run vgscan is due to having a shared storage so i might want 
to activate the md devices and vg devices on different nodes

Regards,
Luca


-- 
Luca Berra -- bluca at comedia.it