Re: Regression introduced by bfcfaa77bdf0f775263e906015982a608df01c76 (vfs: use 'unsigned long' accesses for dcache name comparison and hashing)

[Al Viro <viro@ZenIV.linux.org.uk>]

On Thu, Mar 22, 2012 at 01:38:28PM -0700, Linus Torvalds wrote:
> On Thu, Mar 22, 2012 at 1:24 PM, Al Viro <viro@zeniv.linux.org.uk> wrote:
> >
> > OK, full_name_hash()/hash_name() definitely have a mismatch and it's on the
> > names of length 8*n: trivial experiment shows that we have
> > name hash_name full_name_hash
> 
> Good catch, guys.
> 
> Ugh. And I never noticed despite having run this code on my machines
> for several weeks, because I don't think I have anything that uses the
> "full_name_hash()" function. And it looked so obviously the same.
> 
> > Linus, which way do you prefer to shift it? ?Should hash_name() change to
> > match full_name_hash() or should it be the other way round?
> >
> > What happens is that you get multiplication by 9 and adding 0 in the former,
> > after having added the last full word. ?In the latter we add the last full
> > word, see that there's nothing left and bugger off.
> 
> Yes. I think we should make things match "hash_name()", because that's
> the one that is critical and we want to really generate good code for.
> 
> I think you can just move the "*=9" down in full_name_hash(), so that
> we always "pre-multiply" the hash for the next round.  But I'll have
> to double-check my logic.

See upthread for diff doing just that ;-)  Let's see if that fixes the
crap guys are seeing...  BTW, you have used full_name_hash(), just not
on something 8 char long - devpts uses d_alloc_name(), but pty numbers
tend to be less than ten millions...