Ext3 users - Jul 2002 - is EXT3 data=journal, 2K block size, 2.5 series kernel stable?

A while back, after 2.5.6, there were various changes made to the kernel
that resulted in some instability with journaled data on ext3
filesystems, as well as small issues with my 2k block size file systems.

Can someone (say, Andrew Morton ;) comment on the stability of the
current release with this sort of configuration?

If it's not full of known bugs then I will give it another shot, but I
don't want to waste time finding out that it's still known to be broken
myself. :)

        Daniel

-- 
The extreme limit of wisdom - that is what the public calls madness. 
        -- Jean Cocteau

Daniel Pittman wrote:
> 
> A while back, after 2.5.6, there were various changes made to the kernel
> that resulted in some instability with journaled data on ext3
> filesystems, as well as small issues with my 2k block size file systems.
> 
> Can someone (say, Andrew Morton ;) comment on the stability of the
> current release with this sort of configuration?
> 
> If it's not full of known bugs then I will give it another shot, but I
> don't want to waste time finding out that it's still known to be
broken
> myself. :)
> 
Nope.  data=journal is bust on 2.5.  It's also bust on 2.4, for
the same reason.

The problem arises from the mixing of regular write(2) operations
and the writepage() function.  In 2.4, writepage() doesn't get
called much.  In 2.5, it gets called all the time, and so the
problem manifests more in 2.5.

I think Stephen and I agreed on the way to fix it last week,
but I haven't got onto coding it yet.  Probably next week.

-

Daniel Pittman wrote:
> 
> A while back, after 2.5.6, there were various changes made to the kernel
> that resulted in some instability with journaled data on ext3
> filesystems, as well as small issues with my 2k block size file systems.
> 
> Can someone (say, Andrew Morton ;) comment on the stability of the
> current release with this sort of configuration?
> 
> If it's not full of known bugs then I will give it another shot, but I
> don't want to waste time finding out that it's still known to be
broken
> myself. :)
> 
But the 2k blocksize shouldn't be an issue.

-