misc/144943: gconcat randomly "loses" all knowledge of JBOD set

Rowan Crowe freebspr at sensation.net.au
Sun Mar 21 23:50:06 UTC 2010

>Number:         144943
>Category:       misc
>Synopsis:       gconcat randomly "loses" all knowledge of JBOD set
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    freebsd-bugs
>State:          open
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Sun Mar 21 23:50:05 UTC 2010
>Originator:     Rowan Crowe
>Release:        6.3-RELEASE i386
FreeBSD ******.sensation.net.au 6.3-RELEASE FreeBSD 6.3-RELEASE #0: Wed Aug 20 02:13:07 EST 2008     rowan@******.sensation.net.au:/usr/src/sys/i386/compile/VID62R  i386

I've recently had a few "events" where gconcat seems to have disowned the drives in a set. For example, I just inserted two drives, powered on, then at boot there's no mention of GCONCAT at all. It's in memory - via loader.conf - but it doesn't recognise the disks as belonging to a gconcat set.

I can restore access to the data if I manually recreate the array, but I'm very concerned that it seems to be deliberately disconnecting drives (and possibly updating their metadata?) for some reason.

# gconcat list
[there is no list output, ie no sets are recognised]
# gconcat create -v permdocs ad3 ad1
# geli attach /dev/concat/permdocs
Enter passphrase:
# mount -o rw,noatime /dev/concat/permdocs.eli /permdocs
.. and from this point everything seems to be fine. fsck_ufs also succeeds with no issues.

There is no output in /var/log/messages.

I have been using gconcat+geli for several months but it's only in the past 2-3 weeks that this has happened. I cannot reconcile this behaviour with any change in software or hardware so it may be coincidence.

I consider this a fairly serious problem because although the data does not appear to be lost this issue can impact on the running of a machine (automated mounts at boot will fail) and may also cause panic by the human trying to fix it!
Unknown. I have set kern.geom.concat.debug > 0 to hopefully show some useful debug output next time it happens.


More information about the freebsd-bugs mailing list