Re: [PATCH 2/3] fat: add the msdos_format_name() filename cache

[Al Viro]

On Sun, Aug 29, 2021 at 03:11:22PM +0000, Al Viro wrote:
> On Sun, Aug 29, 2021 at 02:25:29PM +0000, Caleb D.S. Brzezinski wrote:
> > Implement the main msdos_format_name() filename cache. If used as a
> > module, all memory allocated for the cache is freed when the module is
> > de-registered.
> > 
> > Signed-off-by: Caleb D.S. Brzezinski <calebdsb@xxxxxxxxxxxxxx>
> > ---
> >  fs/fat/namei_msdos.c | 35 +++++++++++++++++++++++++++++++++++
> >  1 file changed, 35 insertions(+)
> > 
> > diff --git a/fs/fat/namei_msdos.c b/fs/fat/namei_msdos.c
> > index 7561674b1..f9d4f63c3 100644
> > --- a/fs/fat/namei_msdos.c
> > +++ b/fs/fat/namei_msdos.c
> > @@ -124,6 +124,16 @@ static int msdos_format_name(const unsigned char *name, int len,
> >  	unsigned char *walk;
> >  	unsigned char c;
> >  	int space;
> > +	u64 hash;
> > +	struct msdos_name_node *node;
> > +
> > +	/* check if the name is already in the cache */
> > +
> > +	hash = msdos_fname_hash(name);
> > +	if (find_fname_in_cache(res, hash))
> > +		return 0;
> 
> Huh?  How could that possibly work, seeing that
> 	* your hash function only looks at the first 8 characters
> 	* your find_fname_in_cache() assumes that hash collisions
> are impossible, which is... unlikely, considering the nature of
> that hash function
> 	* find_fname_in_cache(res, hash) copies at most 8 characters
> into res in case of match.  Where does the extension come from?
> 
> Out of curiosity, how have you tested that thing?

While we are at it, your "fast path" doesn't even look at opts
argument...