On Thu, Apr 01, 2004 at 06:52:43PM -0600, Dmitry Nikiforov wrote:
> 'Corrupted' as in 'incorrect data in them'. Besides, a few files
> were lost (Mozilla bookmarks.html, for example, is one of them).
Incorrect data can/will happen for some files. XFS journals metadata
not data so this is expected unless applications take sufficient care
(MTAs like postfix a good example of applications which do the right
thing).
But _what_ incorrect data are we talking about?
> Well, it looks like a random garbage. Last time it happened to
> /var/log/messages - there was some random binary data at the end of
> it.
That shouldn't happen (nulls might though). Can you reproduce this?
> At the same time, there were no error messages nor warnings when the
> system was being started and fsck was running, which I find very
> odd...
I don't see why there should be.... journal replay and fsck have no
way of knowing what files should look like. They don't look at file
contents.
> Open files, as far as I can tell right now.
Such as? You mentioned log files, what else?
> Some files I find just later that they were corrupted, so I can't
> tell what was happening to them at that very moment :(
It would be good to know what these files are and when they were last
used relative to the crash.
> While I don't have too many crashes (about 4-5 in total so far) - it
> always involves something missing.
What causes these crashes?
> On my office machine (which is 100% linux box with XFS all over it
> :) ) every time it gets unconditionally rebooted, I loose most of my
> KDE settings and most of Mozilla settings (these are that I can
> notice immediately after the reboot).
I've seen this (I use KDE here). As best as I can tell it's KDE and
Mozilla making dumb assumptions. I'm not sure I blame them but they
aren't exactly careful about their configuration files either which
bugs me a little. There might going on here than is apparent so it's
possible there is a filesystem bug, but I've not been able to
reproduce something I can blame XFS when it comes to KDE configuration
files.
I'll try again later.
> And, like I said, there were a couple of times when there was some
> random garbage in /var/log/messages, for example:
[...]
> Mar 31 22:01:00 dniq CROND[24797]: (root) CMDL$|.?..?.??.
[... non-NULL cruft deleted ...]
Is that common and/or reproducible? That doesn't seem right.
--cw
|