Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 8 Oct 2000 12:21:37 -0700 (PDT)
From:      Matt Dillon <dillon@earth.backplane.com>
To:        Kirk McKusick <mckusick@mckusick.com>
Cc:        freebsd-stable@freebsd.org
Subject:   Weird (possibly softupdates related) data loss on fsck/reboot
Message-ID:  <200010081921.e98JLbn00668@earth.backplane.com>

next in thread | raw e-mail | index | archive | help
    We had building maintainance that turned off power longer then the
    UPS could handle and one of my machines went down.

    When it came back up it ran fsck and came up with a whole lot of
    "missing '.' in directory" (or something like that, I didn't record
    the fsck output unfortunately), and "directory corrupt" errors.

    It recovered around 128 files and four empty directories, all from
    /usr/ports, which it had disconnected due to believing the governing
    directories were corrupt.  After the fsck the governing directories were
    present in the filesystem but empty.  

    I looked at a recent 'dump' backup of /usr and restored it and the
    files were present in the dump, so the 'dump' program had no problem
    seeing them.  I also looked at my cvsup logs and determined that 
    some of the files in question had been cvsup'd weeks before
    the L0 dump I looked at, and that cvsup obviously saw the files in
    the live filesystem every time it ran in the last week since it didn't
    feel the need to update them.

    I cvsup /usr/ports daily.  /usr is mounted with softupdates on.  There
    was no filesystem activity at the time of the power failure.  /usr on
    this machine is a 2G filesystem on a SCSI partition (nothing fancy).

    Very odd!  I'm not sure what is going on, but since dump saw the files
    correctly and since the files were visible in the live filesystem prior
    to the power failure, I can only guess that 'fsck' somehow believed the
    directories were corrupt when they weren't (or at least, not so corrupt
    that dump would feel it had to blow away the whole directory!).  I don't
    think this is buffer-cache related since the raw device and the live
    filesystem seemed to see a consistent view of the filesystem.

    I'll try to reproduce the problem in a week or two (by taking a dd copy
    of /usr and then taking the machine down hard).  This is just a datapoint
    in case other people have had similar problems.

						-Matt



To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-stable" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200010081921.e98JLbn00668>