[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: ext3/ext2 kernel bug with umlauts?



On Fri, Aug 29, 2008 at 09:46:56PM +0200, Michael Glockenstein wrote:
> I first did the debugfs change, then created the ext3 journal with tune2fs,
> then mounted the partition, then as you suggested
> mkdir post.new
> mv post/* post.new
> 
> mv: cannot stat `post/Badfenstert\374rumbau.doc': No such file or directory
> mv: cannot stat `post/d-boxk\374ndigung2.doc': No such file or directory
> and so on
> 
> Are you really sure that this is the same problem, because it only affects
> filenames with umlauts (äöüßÄÖÜ in german), should a hash algorythm
> problem not affect also filenames without them?

This is failing because the kernel isn't able to access the files from
the old directory.  So readdir is returning a filename that can't be
looked up.  This is potentially consistent with hash collisions, and
it may be that TEA is just much more sensitive to these sorts of
failures.

So there are two things to try.  Retry the "mv post/* post.new", and
see if it succeeds after it has moved some portion of the files to
post.new.

Secondly, if you're still getting the failure. I need the listing of
the filenames in the "post" directory, plus the output of this
command:

	dumpe2fs -h /dev/hdXXX | grep "Hash Seed"

Also possibly useful will be:

	debugfs /dev/hdXXXX -R "dump /path/to/the/dir/named/post /tmp/post.dump"

And then send me the file /tmp/post.dump

With those two bits of information I should hopefully be able to
replicate the problem.  I really want to double check the theory that
it might be a hash collision problem, and it's just that the TEA hash
is really bad if you have lots of filenames with the high eighth bit
set.  It's consistent with the observations that not all filenames
with special characters are affected, and that it takes a large number
of filenames to show the problem.  So I'm not 100% sure, but it sure
sounds suspiciously like the other problem.  And even if it isn't
that, I really need to be able to replicate the test case on my
computers, one way or another.

						- Ted


Reply to: