mdmon: device descriptor is not closed

[Mariusz Tkaczyk]

When device disappears from the system "mdadm -If" is triggered by udev and drive is removed from the corresponding array and container. The missing close() in metadata manager (mdmon) causes that userspace handler to the device is still opened (can be tracked by lsof -p <mdmon_pid>).

For nvme drives it causes weird namespace number after hot-plug (e.g. will appear nvme1n2). The issue affects only IMSM metadata format.
Reproducible on Focal with latest mdadm 4.1-5ubuntu1.

Steps:
1. Create IMSM container:
#mdadm -CR imsm -e imsm -n2 /dev/nvme[01]n1
2. Create IMSM array:
#mdadm -CR vol1 -l1 -n2 /dev/nvme[01]n1
3. remove via sysfs drive:
#echo 1 > /sys/block/nvme0n1/device/device/remove
4. restore the same drive:
#echo 1 > /sys/bus/pci/rescan
5. see mdstat and lsblk:
# cat /proc/mdstat

Result:
/dev/nvme0n2 appears in system.

[Mariusz Tkaczyk (mtkaczyk)]

Hi, issue is fixed by:
https://git.kernel.org/pub/scm/utils/mdadm/mdadm.git/commit/?id=91c97c5432028875db5f8abeddb5cb5f31902001

Thanks,
Mariusz

[Launchpad Janitor (janitor)]

Status changed to 'Confirmed' because the bug affects multiple users.

[Jeff Lane  (bladernr)]

Can this please be targeted for 21.10 in subiquity?

[Michael Hudson-Doyle (mwhudson)]

I don't see how this has anything to do with subiquity?

[Jeff Lane  (bladernr)]

Sorry, that was an overzealous CopyPasta. I meant just for 20.10 for some of these (which are obviously not Subiquity parts).

[Jeff Lane  (bladernr)]

@chihchun, any update?

[Jeff Lane  (bladernr)]

Is this resolved in 4.2-rc2?

[Mariusz Tkaczyk (mtkaczyk)]

yes, it is resolved.