Re: Random filesystem corruption

[Sebastian Ude <ude@handshake.de>]

On Sat, 9 Mar 2002, sandeen@sgi.com (Eric Sandeen) wrote:
> Date: Sat, 9 Mar 2002 22:22:17 -0600 (CST)
> To: Sebastian Ude <ude@handshake.de>
> From: sandeen@sgi.com (Eric Sandeen)
> CC: <linux-xfs@oss.sgi.com>
> Subject: Re: Random filesystem corruption

[...]

> Can you do an "strace" on a simple program that tries to open one of
> these files, and send the last bit (the failed open)?

Nothing exiting:

/* [...] */

open("/home/sebastian/WU RB Gebührenstaffels.txt", O_RDONLY) = -1 ENOENT (No such file or directory)

/* [...] */

By the way - I noticed that all files that were damaged *yesterday*
contained German Umlauts (the letters u, o, a with two dots above them:
ü, ö, ä) in their names. However, I am sure that in the past files without
Umlauts in the name got damaged as well.


> Also - are you experiencing this problem on several different machines?

No, it's all on the same machine.


> I assume this is local, not NFS access?

Correct, local.


> When you have a file with this problem, though, I assume the behavior is
> repeatable on that file?

You mean that open () says they do not exist, although they are listed in their parent directory ? Yes, it's always the same.


> Out of curiosity, do either the files or the directories have
> "international" characters in the names?

Errr ... yep, see above. But again, in the past also files or directories without "international" characters in the names got corrupted.


> Decoding the oops through ksymoops would be helpful.  If you could enable
> kdb, that might help us get more information.

This time there is no oops, unfortuantely. It could be that there was only a oops with some kernel versions, but I am not sure.


> > xfs_repair, but the corrupted files were lost.
>
> What does xfs_repair tell you now?  Does it find any problems?

I have not run it yet because it could not recover my data the last time (if I remember correctly).


> If you go back to an older kernel (perhaps the released 1.0.2 kernel)
> does the problem go away?

Well, that problem occours perhaps one time a month - and I have no idea how to reproduce it. So if I would try another kernel, I could perhaps tell you in April or May if there was no data
corruption ...


- Sebastian