bl4:
> Interesting, if not aufs then what?
>
> Firstly, I can prove it this way: There are two identical copies of root 
> partition. Both are mounted read-only in initramfs. One of them becomes 
> a branch of aufs. On shutdown, I rsync changes to both copies and only 
> the one under control of aufs gets dirty.
>
> Secondly, it doesn't happen when both aufs and unionfs are installed and 
> I replace mount -t aufs with mount -t unionfs.

It could be ext2/ext3, buffer management or something else.
>From my point of view, there is no process to write to aufs between two
fscks. Rsync to /dev/hda1 only, is it? This is my biggest reason.
I could find some fixes in ext3 after linux-2.6.26 which look related to
flushing buffers and the consistency of a filesystem. But I have not
dive into them yet, so I am unsure what is wrong now.
Flushing the pseudo links and udbs=inotify may be related, but they will
not corrupt the filesystem. Additionally you wrote that you had already
tried them.

I got an idea to be worth to try.
- remount aufs and make it readonly (and udba=inotify) before the
  script.
- disable CONFIG_AUFS_FSYNC_SUPER_PATCH.
- run sync(8) before the second fsck (or remount readonly).

I will test them when I got your qemu image.


J. R. Okajima

------------------------------------------------------------------------------
Download Intel® Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev

Reply via email to