On Monday 09 August 2010 19:31:10 Henrik Grindal Bakken wrote:
> Henrik Grindal Bakken <hgb@ifi.uio.no> writes:
> > Henrik Grindal Bakken <hgb@ifi.uio.no> writes:
> >> Matthieu CASTET <matthieu.castet@parrot.com> writes:
> >>> After adding OS trace to yaffs2, I saw that sync does a call to
> >>> yaffs_sync_fs, but it is a nop.
> >>>
> >>> This mean sync syscall doesn't flush the yaffs2 cache and data it
> >>> lost on unclean umount.
> >>>
> >>> Calling yaffs_FlushEntireDeviceCache in yaffs_sync_fs seems to
> >>> make sync working.  What do you think of that ?
> >>
> >> I suffer from the same problem.  If a yaffs2 filesystem is merely
> >> remounted read-only, and never properly unmounted, I seem to lose
> >> data.  This only (I think, at least primarily) happens on
> >> filesystems that are loopback mounted from files in the yaffs2
> >> filesystem.  I'd love to see a fix for sync.
> >
> > Has anyone been able to look at this problem and the proposed fix
> > from Matthieu?
>
> I'm bringing this up again.
>
> I'm using yaffs (2) on a NAND flash, and on the yaffs fs, I have among
> other files some file system images (ext2), which are
> loopback-mounted.  If I touch a file on the loopback-mounted image,
> sync, then poweroff, the file is lost when the box comes up again.
>
> I notice that yaffs_sync_fs is no longer a nop, but it apparently
> doesn't quite do what I need it to.
>
> Does anyone have any idea how one could go about attacking this
> problem?

Hello Hendrik

Have you tried enabling tracing to see what is happening?

#echo +os >/proc/yaffs

I wonder if perhaps the file system syncing order does something like this:

# sync:
	sync yaffs yaffs is now sync'ed
	sync ext2 yaffs is now dirty again

What happens if you sync twice?

# sync; sleep 2; sync

-- Charles