BUG in jffs2 with MTD map using cached mapping

[Matt Witherspoon]

I´m replying to this again from a different address; hopefully I can get the reply-to header right 
this time so it will make it through.

On 9 Jan 2009 at 10:47, Anders Grafström wrote:
> Matt Witherspoon wrote:
> > I performed a bisection and the offending commit is
> > 205c109a7a96d9a3d8ffe64c4068b70811fef5e8. Before this commit, neither the BUG
> > nor the node CRC failure occured.
> I'd say that the BUG_ON is a secondary error. A broken error path.
> http://lists.infradead.org/pipermail/linux-mtd/2008-April/021473.html

You may be right in that it´s related, but that patch doesn´t really fix the root problem (bogus 
CRC errors).

Before commit 205c109a7a96d9a3d8ffe64c4068b70811fef5e8 (jffs2: convert to new aops) 
everything was fine. I never got CRC errors and certainly no BUGs. I never saw any file 
corruption.

After 205c10 (and still in 2.6.28), I see the CRC error and the BUG when I copy a file. Copied 
files end up being truncated.

By applying the patch that Anders is referring to, I no longer get the BUG. However, I do still 
get CRC errors and copied files are still truncated.

If I simply comment out the BUG_ON, I still get the CRC errors but files appear to be copying 
okay. I have done various diffs on copied files (after rebooting, etc) and still have not seen 
any corruption or truncation.

I still don´t know what to make of this. Any help would be great.