Description of problem:

Unable to complete or abort pvmove migration when it failed with error ' Logical volume testvg/pvmove0_mimage_0 in use.'

1. pvmove --abort
   Device name for LV testvg/pvmove0 not found in metadata. (unfinished pvmove mirror removal?)

Version-Release number of selected component (if applicable):

RHEL7.9 /lvm2-2.02.187-6.el7_9.5.x86_64
issue reproducible with RHEL9.2 as well [lvm2-2.03.17-7.el9.x86_64]

How reproducible:

Always

Steps to Reproduce:

1. Create volume group and logical volume with multiple PVs.
2. Start migration using pvmove
3. Try to access the pvmove image and keep it as busy/open while pvmove is in progress.
4. After moving all PEs, it failed with the error "LVM command executed by lvmpolld failed."

Actual results:

pvmove migration failed and unable to remove the pvmove segments.

>> Below is my lvm stack and I trying to migrate PEs from sda to sdc.

1. lvs -ao +devices testvg
   LV VG Attr LSize Pool Origin Data% Meta% Move Log Cpy%Sync Convert Devices
   lvol0 testvg wi-a---- 6.00g /dev/sda(0)
   lvol0 testvg wi-a---- 6.00g /dev/sdb(0)
   lvol1 testvg wi-a---- 1.00g /dev/sda(1023)

>> Start migration using pvmove and try to open the device. [ here I used hexdump utility to open the dm-X device].

1. pvmove --atomic /dev/sda /dev/sdc
   /dev/sda: Moved: 0.00%
   /dev/sda: Moved: 24.63%
   /dev/sda: Moved: 52.85%
   /dev/sda: Moved: 80.30%
   /dev/sda: Moved: 100.00%
   LVM command executed by lvmpolld failed.
   For more information see lvmpolld messages in syslog or lvmpolld log file.

>> Logical volumes are still accessible and able to mount, but a few pvmove segments still exist, and unable to clear them from lvm stack. Also, the source disk is in use, and unable to remove it from VG.

1. dmsetup table |grep testvg
   testvg-pvmove0_mimage_1: 0 10477568 error
   testvg-pvmove0_mimage_0: 0 8380416 linear 8:0 2048
   testvg-pvmove0_mimage_0: 8380416 2097152 linear 8:0 8382464
   testvg-lvol1: 0 2097152 linear 253:6 8380416
   testvg-pvmove0: 0 8380416 linear 8:32 2048
   testvg-pvmove0: 8380416 2097152 linear 8:32 8382464
   testvg-lvol0: 0 8380416 linear 253:6 0
   testvg-lvol0: 8380416 4202496 linear 8:16 2048

>> pvmove abort is failing with below error.

1. pvmove --abort
   Device name for LV testvg/pvmove0 not found in metadata. (unfinished pvmove mirror removal?)

1. journalctl -u lvm2-lvmpolld
   • Logs begin at Wed 2023-08-02 12:54:26 EDT, end at Wed 2023-08-02 13:22:17 EDT. –
     Aug 02 13:10:55 node1 systemd[1]: Started LVM2 poll daemon.
     Aug 02 13:11:40 node1 lvmpolld[2501]: W: LVPOLL: PID 2504: STDERR: ' WARNING: This metadata update is NOT backed up.'
     Aug 02 13:14:27 node1 lvmpolld[2501]: W: LVPOLL: PID 2860: STDERR: ' WARNING: This metadata update is NOT backed up.'
     Aug 02 13:16:24 node1 systemd[1]: Started LVM2 poll daemon.
     Aug 02 13:17:24 node1 lvmpolld[3222]: W: LVPOLL: PID 3225: STDERR: ' WARNING: This metadata update is NOT backed up.'
     Aug 02 13:17:29 node1 lvmpolld[3222]: W: LVPOLL: PID 3225: STDERR: ' Logical volume testvg/pvmove0_mimage_0 in use.' <===============
     Aug 02 13:17:29 node1 lvmpolld[3222]: LVMPOLLD: lvm2 cmd (PID 3225) failed (retcode: 5)

Expected results:

Looking for some options to clear the leftover pvmove segments properly. There are options to manually edit the VG metadata and restore it back to disk, but for a very large volume, this should be very complicated and may cause the lv in the LOCKED state.

Additional info:

This issue is reproducible with RHEL9 standalone systems as well.
After reproducing this issue, I tried to disable ( use_lvmpolld = 0), but still unable to abort or complete pvmove.