Re: [PATCH v5 4/5] Add d_for_each_positive_child family of iterators

From: Jan Kara

Date: Fri Nov 20 2026 - 05:09:31 EST


On Fri 20-11-26 10:28:55, NeilBrown wrote:
> On Fri, 20 Nov 2026, Amir Goldstein wrote:
> > On Thu, Nov 19, 2026 at 4:00 AM NeilBrown <neilb@xxxxxxxxxxx> wrote:
>
> > > +
> > > +/**
> > > + * d_next_sibling_sched - return next sibling, but schedule() first
> > > + * @child: current child
> > > + *
> > > + * This behaves like d_next_sibling(), but drops the parent d_lock
> > > + * and calls cond_resched() to avoid any soft-lockup. A cursor
> > > + * is inserted as place holder.
> > > + *
> > > + * Caller must hold the parents d_lock, and must hold a counted
> > > + * reference on the parent.
> > > + *
> > > + * Returns: the next sibling, or %NULL if there is none.
> > > + */
> > > +struct dentry *d_next_sibling_sched(struct dentry *child)
> > > +__must_hold(&child->d_parent->d_lock)
> > > +{
> > > + struct dentry cursor;
> > > +
> > > + /* always skip cursors to ensure we make some real progress */
> > > + while (child && (child->d_flags & DCACHE_DENTRY_CURSOR))
> > > + child = d_next_sibling(child);
> > > + if (!child)
> > > + return child;
> >
> > Why is this skip loop here and not after the resched?
> >
> > As this is written, if the caller passed @child which is another cursor,
> > now child the next real dentry after another cursor and hlist_add_behind()
> > below will insert our cursor after the real dentry.
>
> You are right of course - that is the wrong way to address the problem.
>
> I think moving the test into d_next_positive(), like with this
> incremental patch, will work better.

With this change it looks good to me. Feel free to add:

Reviewed-by: Jan Kara <jack@xxxxxxx>
Honza

> diff --git a/fs/dcache.c b/fs/dcache.c
> index bacb218400a8..46aca619a9a3 100644
> --- a/fs/dcache.c
> +++ b/fs/dcache.c
> @@ -2095,6 +2095,9 @@ static void d_init_cursor(struct dentry *dentry, struct dentry *parent)
> * Caller must hold the parents d_lock, and must hold a counted
> * reference on the parent.
> *
> + * @child should not be a cursor, else two threads could potentially
> + * live-lock, each only stepping over each others cursors.
> + *
> * Returns: the next sibling, or %NULL if there is none.
> */
> struct dentry *d_next_sibling_sched(struct dentry *child)
> @@ -2102,12 +2105,6 @@ __must_hold(&child->d_parent->d_lock)
> {
> struct dentry cursor;
>
> - /* always skip cursors to ensure we make some real progress */
> - while (child && (child->d_flags & DCACHE_DENTRY_CURSOR))
> - child = d_next_sibling(child);
> - if (!child)
> - return child;
> -
> d_init_cursor(&cursor, child->d_parent);
> hlist_add_behind(&cursor.d_sib, &child->d_sib);
> spin_unlock(&cursor.d_parent->d_lock);
> diff --git a/include/linux/dcache.h b/include/linux/dcache.h
> index 7150ecd81d72..c384b0f8743e 100644
> --- a/include/linux/dcache.h
> +++ b/include/linux/dcache.h
> @@ -697,7 +697,10 @@ struct dentry *d_next_sibling_sched(struct dentry *child);
> static inline struct dentry *d_next_positive(struct dentry *child)
> {
> do {
> - if (need_resched())
> + /* Don't schedule on cursors to ensure we make some real
> + * progress between scheduling
> + */
> + if (need_resched() && !(child->d_flags & DCACHE_DENTRY_CURSOR))
> child = d_next_sibling_sched(child);
> else
> child = d_next_sibling(child);
>
--
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR