xfs
[Top] [All Lists]

Re: problems with restore

To: Lars Soltau <lars.soltau@xxxxxxx>
Subject: Re: problems with restore
From: Tim Shimmin <tes@xxxxxxx>
Date: Mon, 29 Jul 2002 11:24:08 +1000
Cc: Tim Shimmin <tes@xxxxxxx>, linux-xfs@xxxxxxxxxxx
In-reply-to: <3D411D05.9020301@slab.de>; from lars.soltau@slab.de on Fri, Jul 26, 2002 at 11:57:25AM +0200
References: <20020723114554.J2935839@boing.melbourne.sgi.com> <3D411D05.9020301@slab.de>
Sender: owner-linux-xfs@xxxxxxxxxxx
Hi Lars,

On Fri, Jul 26, 2002 at 11:57:25AM +0200, Lars Soltau wrote:
> > 
> > 2. [...] So one is not supposed to copy a dump from tape onto a disk file 
> > and 
> >    then restore it. [...] 
> 
> Well, that's not very helpful. :-)
> 
> I had some old tapes and borrowed a matching streamer to restore them. 
> After the restore directly from tape failed with an assertion failure 
> (see blow), I copied the files to disk because I had to return the 
> streamer before I had a chance to do a full examination of the assertion 
> failure.
> 
> I ended up patching drive_minrmt.c to forget about all SCSI commands and 
> open consecutively numbered files on hard disk instead. 
Ahhh, the beauty of open source :)

> It would be a  nice feature for xfsrestore to be able to open a tape
> backup from disk, wouldn't it?
> 
I guess ;-)
It would be nice if the dump format was independent of the 
dump tape commands - at the moment using a "dump tape" strategy means using
the dump tape format with the tape scsi commands.

> > 3. I don't know why xfsrestore is aborting with an assertion failure for 
> > you.
> 
> I do, now. I single-stepped into the assertion and found out that one 
> 64bit offset in each record header, the first_mark_offset, had been 
> written  highendian instead of littleendian. All the other offsets, for 
> example file_offset that is located directly before first_mark_offset, 
> had been written correctly.
> 
If you check out the cmd/xfsdump/doc/CHANGES file, it was fixed
in December last year:
    xfsdump-1.1.10 (10 December 2001)
            - fix xfsdump to endian convert all of the record header
              fields properly just prior to writing the header out
              (in particular first_mark_offset).
              This caused do_next_mark() assertion failures at some
              sites.

> The backup has been written by a version of xfsdump compiled with egcs 
> on Linux. Personally, I suspect egcs's 64bit arithmetic. Anyway, I 
> patched the source to correct the offset on the fly and the problem 
> vanished.
Good.

The mail archive:
    http://marc.theaimsgroup.com/?l=linux-xfs&m=101435725816823&w=2
had suggestions for restoring from the bad dumps.

Cheers,
--Tim


<Prev in Thread] Current Thread [Next in Thread>