Re: [RFC PATCH v3 14/15] dcache: Implement partial shrink via Slab Movable Objects

["Tobin C. Harding" <me@tobin.cc>]

On Thu, Apr 11, 2019 at 03:33:22AM +0100, Al Viro wrote:
> On Thu, Apr 11, 2019 at 11:34:40AM +1000, Tobin C. Harding wrote:
> > +/*
> > + * d_isolate() - Dentry isolation callback function.
> > + * @s: The dentry cache.
> > + * @v: Vector of pointers to the objects to isolate.
> > + * @nr: Number of objects in @v.
> > + *
> > + * The slab allocator is holding off frees. We can safely examine
> > + * the object without the danger of it vanishing from under us.
> > + */
> > +static void *d_isolate(struct kmem_cache *s, void **v, int nr)
> > +{
> > +	struct dentry *dentry;
> > +	int i;
> > +
> > +	for (i = 0; i < nr; i++) {
> > +		dentry = v[i];
> > +		__dget(dentry);
> > +	}
> > +
> > +	return NULL;		/* No need for private data */
> > +}
> 
> Huh?  This is compeletely wrong; what you need is collecting the ones
> with zero refcount (and not on shrink lists) into a private list.
> *NOT* bumping the refcounts at all.  And do it in your isolate thing.

Oh, so putting entries on a shrink list is enough to pin them?

> 
> > +static void d_partial_shrink(struct kmem_cache *s, void **v, int nr,
> > +		      int node, void *_unused)
> > +{
> > +	struct dentry *dentry;
> > +	LIST_HEAD(dispose);
> > +	int i;
> > +
> > +	for (i = 0; i < nr; i++) {
> > +		dentry = v[i];
> > +		spin_lock(&dentry->d_lock);
> > +		dentry->d_lockref.count--;
> > +
> > +		if (dentry->d_lockref.count > 0 ||
> > +		    dentry->d_flags & DCACHE_SHRINK_LIST) {
> > +			spin_unlock(&dentry->d_lock);
> > +			continue;
> > +		}
> > +
> > +		if (dentry->d_flags & DCACHE_LRU_LIST)
> > +			d_lru_del(dentry);
> > +
> > +		d_shrink_add(dentry, &dispose);
> > +
> > +		spin_unlock(&dentry->d_lock);
> > +	}
> 
> Basically, that loop (sans jerking the refcount up and down) should
> get moved into d_isolate().
> > +
> > +	if (!list_empty(&dispose))
> > +		shrink_dentry_list(&dispose);
> > +}
> 
> ... with this left in d_partial_shrink().  And you obviously need some way
> to pass the list from the former to the latter...

Easy enough, we have a void * return value from the isolate function
just for this purpose.

Thanks Al, hackety hack ...


	Tobin