kern/170038: [geom] geom_mirror always starts degraded after reboot

[Frank Reppin]

The following reply was made to PR kern/170038; it has been noted by GNATS.

From: Frank Reppin <frank at undermydesk.org>
To: bug-followup at FreeBSD.org, frankreppin at googlemail.com
Cc:  
Subject: Re: kern/170038: [geom] geom_mirror always starts degraded after
 reboot
Date: Thu, 02 Aug 2012 03:30:13 +0200

 Me again here...
 
 ... another thing that puzzles me...
 
 As long as I had 'journalled softupdates' activated I could see
 that the reported size of my /var partition [/dev/mirror/gm0s1d]
 did (quite often) show disconcerting usage values when using:
 
    df; du -cs {/var,/tmp,/usr}
 
 [means - the shown results of 'df' and 'du' did show different
 usage values for /var - and only /var in my case - no idea why...
 and it even didn't somehow sync back after a couple of minutes or
 after rebooting - the results where 'wrong' all the time].
 
 This came especially apparent at one time where 'df' did show
 me a _negative_ available space on /var (and I wasn't even close
 to the reserved space - 144M used out of 53G). A simple reboot (with 
 hopes that the background fsck would fix that) wasn't enough - I had
 to force fsck the partition in question manually in order to show the 
 correct usage values again...
 
 The fsck 'error' reported was (all the time):
 
    FREE BLOCK COUNTS(S) WRONG IN SUPERBLK
 
 and it only applied to /var (for some unknown reason) - I emphasize
 this, because I too did some decent IO to /usr at this time (like svn 
 co) but /usr was never affected by this.
 [/var does provide cyrus/postfix/amavis/spamassassin and the such]
 
 However - the gmirror rebuild wasn't triggered all the time after
 rebooting (as I later noticed - and in contratiction to the topic
 of this report...) - but quite often... I would assume that 6 out
 of 10 reboots would trigger a gmirror rebuild in my case.
 
 ... last but not least - I've disabled SUJ on all my partitions
 and I've rebooted (for testcases like this several times) and the
 'FREE BLOCK COUNTS(S) WRONG IN SUPERBLK' as well as the gmirror rebuilds
 vanished... no more trouble here - this is so weird...
 
 HTH...
 
 cheers,
 frank\
 
 
 -- 
 43rd Law of Computing:
          Anything that can go wr
 fortune: Segmentation violation -- Core dumped