Re: Need advice with iput() deadlock during writeback

[Mateusz Guzik <mjguzik@gmail.com>]

On Wed, Sep 17, 2025 at 10:14 PM Al Viro <viro@zeniv.linux.org.uk> wrote:
>
> On Wed, Sep 17, 2025 at 10:59:23AM +0200, Mateusz Guzik wrote:
>
> > A sketch, incomplete:
> > static DECLARE_DELAYED_WORK(delayed_ceph_iput_work, delayed_ceph_iput);
> >
> > static void __ceph_iput_async(struct callback_head *work)
> > {
> >         struct ceph_inode_info *ci = container_of(work, struct
> > ceph_inode_info, async_task_work);
> >         iput(&ci->netfs.inode);
> > }
> >
> > void ceph_iput_async(struct ceph_inode_info *ci)
> > {
> >         struct inode *inode = &ci->netfs.inode;
> >
> >         if (atomic_add_unless(&inode->i_count, -1, 1))
> >                 return;
> >
> >         if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
> >                 init_task_work(&ci->async_task_work, __ceph_iput_async);
> >                 if (!task_work_add(task, &ci->async_task_work, TWA_RESUME))
> >                         return;
> >         }
> >
> >         if (llist_add(&ci->async_llist, &delayed_ceph_iput_list))
> >                 schedule_delayed_work(&delayed_ceph_iput_work, 1);
> > }
>
> Looks rather dangerous - what do you do on fs shutdown?

Can you elaborate?

This should be equivalent to some random piece of code holding onto a
reference for a time.

I would expect whatever unmount/other teardown would proceed after it
gets rid of it.

Although for the queue at hand something can force flush it.