Re: ext3: fix ext3_dx_readdir hash collision handling - Regression

[Theodore Tso <tytso@mit.edu>]

On Fri, Oct 24, 2008 at 09:08:26AM -0700, Linus Torvalds wrote:
> 
> Hmm. Ted? I have not tried to revert that commit that Markus pinpointed 
> (6a897cf447a83c9c3fd1b85a1e525c02d6eada7d: "ext3: fix ext3_dx_readdir hash 
> collision handling"), but now that I look at that "git bug", I am getting 
> pretty damn sure that it's exactly the same issue, and it's not a git bug 
> at all.

Yep, I can replicate it now.  It appears to only show up if you are
running with an x86_64 kernel.  I normally run with an x86_32 kernel,
so I didn't notice the problem.  The commit in question avoids
returning duplicate entries when there is a hash collision;
unfortunately, it seems to return duplicate entries for any large
directory if you are running on x86_64 (and possibly/probably other
64-bit platforms).  

I'm working on it, and should hopefully have a fix for you soon.  If
this is too annoying, you can revert it and I'll resubmit a fixed
version once I get a fix.

						- Ted