On Tue, 31 Oct 2023 at 22:45, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
>
> On Wed, Nov 01, 2023 at 06:20:58AM +0000, Al Viro wrote:
> > Calls of retain_dentry() happen immediately after getting false
> > from fast_dput() and getting true from retain_dentry() is
> > treated the same way as non-zero refcount would be treated by
> > fast_dput() - unlock dentry and bugger off.
> >
> > Doing that in fast_dput() itself is simpler.
>
> FWIW, I wonder if it would be better to reorganize it a bit -
Hmm. Yes. Except I don't love how the retaining logic is then duplicated.
Could we perhaps at least try to share the dentry flag tests between
the "real" retain_dentry() code and the lockless version?
> Another thing: would you mind
>
> #if USE_CMPXCHG_LOCKREF
> extern int lockref_put_return(struct lockref *);
> #else
> static inline int lockref_put_return(struct lockref *l)
> {
> return -1;
> }
> #endif
>
> in include/linux/lockref.h? Would be useful on DEBUG_SPINLOCK configs...
The above sounds like a good idea, not only for better code generation
for the debug case, but because it would have possibly made the erofs
misuse more obvious to people.
Linus
