PV corruption: allocated PE > total PE

Ive just noticed that one of my PVs has more allocated PE's than the total available, leading to a negative amount of free space:

vrumpet root # pvdisplay -d /dev/hda5
--- Physical volume ---

  PV Name               /dev/hda5
  VG Name               myvg
  PV Size               11.72 GB / not usable 33.65 MB
  Allocatable           yes
  PE Size (KByte)       32768
  Total PE              374
  Free PE               4294967278
  Allocated PE          392
  PV UUID               TXK2VM-UWdP-dMbY-yz4c-Nq3b-GM32-B56udb

This is on a gentoo linux system, kernel 2.6.6. The latest LVM userland in gentoo is 2.00.08, I havent tried anything more recent, but there doesnt seem anything relevant in the change log. These VGs were originally created using LVM1 on a 2.4 kernel.

Any suggestions?

Thanks in advance,

Definitely try the latest version: there've been a few related fixes. (With lvm1 on-disk format, may need to backup restore/metadata if it's wrong on disk.)

Alasdair

[email protected]

OK, I'm on lvm .25 and device mapper .19.

I removed the newest LV in that PV, then rebooted. Everything looked right, and the metadata backup was as I expected. However using lvcreate again to create a sufficiently large LV (large enough to fill this PV plus a bit more) recreated the original problem in pvdisplay's output.

The following values all seem to be consistent:
* pvdisplay 'Total PE'
* pe_count in /etc/lvm/backup/
* the total of the relevant 'extent_count' fields in /etc/lvm/backup/

But the value reported by pvdisplay 'Allocated PE' is too large. Where does pvdisplay get that value from? Do I need to worry about data corruption, or could this be just a display formatting bug?

I still can't reproduce this.

Please do a metadata backup/restore using 2.00.25 to make sure the on-disk metadata is correct, then try again.

If it still goes wrong, please send me your /etc/lvm directory to investigate.

Alasdair

[email protected]