Hello Junjiro,

On Thu, Sep 22, 2005 at 10:21:48PM +0900, [EMAIL PROTECTED] wrote:
> 
> Klaus Knopper:
> > I can't believe this. I have been reading that special part of inode.c
> > several times during my 3 week debugging session, and did not notice the
> 
> I didn't believe that either. :-)
> 
> > "off by one", maybe because the last byte of "name" is overwritten with
> > a zero later. Also, I was searching more for uninitialized pointers or
> > race conditions.
> 
> Actually, in my case, only one byte was overwritten. that was zero as a
> terminator. The address was used a part of hidden dentry pointer in
> unionfs dentry. When the real pointer was 0xabcdabcd, it was overwritten
> as 0xabcdab00. And later atomic_read(&dentry->d_count) or something
> crashed.

That makes absolutely sense to me now, since I had the same kind of crashes
with negative dentry->d_count numbers.

> The string constant ".wh." should be a macro, and probably
> allocating/generating whiteout path is better to be an inline function.

A Macro would waste some space for duplicate string constants, I would
rather vote for something like

const char whiteout_prefix[] = ".wh.";

> > Some bugs are still pending, but at least now it looks like I can make a
> > new release this week.
> 
> If you show me the bug list, I'd like to try addressing them. Especially
> the bug is common to my nfs case.

Yesterday we just talked about some other bugs on IRC.

Example (using /var/www, which is not empty):
mv /var/www /var/www.old (works)
mv /var/www.old /var/www (gives a weird error message about moving the 
directory into itself)
mv /var/www.old /var/www (works this time)

I thought that this bug had been fixed earlier, but apparently, it
wasn't (or my setup with ro iso9660 and rw tmpfs is a problem).

Other example:
As an unprivileged user who has NO writing permission to a file or the 
directory containing the file, do a:
rm -f /usr/share/doc/bash/FAQ (Permission denied, correct)
cat /usr/share/doc/bash/FAQ (Segfault, Null pointer exception)

> > Can I send you something for showing my appreciation of your quick bugfix?
> 
> Surprise!!
> My pleasure!!
> I will receive it. Thanx in advance.

Unfortunately, I can't afford the appropriate $2000 someone mentioned
earlier for adequately honouring your debugging, but will think of
something else. ;-)

Just need your postal address (just to me, not the list). :-)

With kind regards
-Klaus Knopper
_______________________________________________
unionfs mailing list
[email protected]
http://www.fsl.cs.sunysb.edu/mailman/listinfo/unionfs

Reply via email to