Re: [PATCH] eCryptfs: Clean up kthread synchronization

[Michael Halcrow]

On Thu, May 22, 2008 at 12:41:42PM -0700, Andrew Morton wrote:
> On Thu, 22 May 2008 14:31:55 -0500
> Michael Halcrow <mhalcrow@xxxxxxxxxx> wrote:
> 
> > > > +void ecryptfs_destroy_kthread(void)
> > > > +{
> > > > +	struct ecryptfs_open_req tmp_req;
> > > > +	struct ecryptfs_open_req *req;
> > > > +
> > > > +	mutex_lock(&ecryptfs_kthread_ctl.mux);
> > > > +	ecryptfs_kthread_ctl.flags |= ECRYPTFS_KTHREAD_ZOMBIE;
> > > > +	list_for_each_entry(req, &ecryptfs_kthread_ctl.req_list,
> > > > +			    kthread_ctl_list) {
> > > > +		mutex_lock(&req->mux);
> > > > +		req->flags |= ECRYPTFS_REQ_ZOMBIE;
> > > > +		wake_up_process(req->requesting_task);
> > > > +		mutex_unlock(&req->mux);
> > > > +	}
> > > > +	memset(&tmp_req, 0, sizeof(tmp_req));
> > > > +	tmp_req.flags = ECRYPTFS_REQ_ZOMBIE;
> > > > +	list_add_tail(&tmp_req.kthread_ctl_list,
> > > > +		      &ecryptfs_kthread_ctl.req_list);
> > > > +	mutex_unlock(&ecryptfs_kthread_ctl.mux);
> > > > +	wake_up(&ecryptfs_kthread_ctl.wait);
> > > > +}
> > > 
> > > eh?  We attach a local variable to a global list and then return?
> > > That won't last very long.
> > 
> > Adding this dummy entry to the list is just my own way of getting the
> > kthread to wake up and shut down. This actually works, albeit a little
> > ugly. The list and its contents get dropped on the floor at this point
> > because (ecryptfs_kthread_ctl.flags & ECRYPTFS_KTHREAD_ZOMBIE). The
> > only consumer of this list (the kthread) checks for this flag
> > immediately after getting the mux, and if it is there, it just
> > exits. The only producer on this list (ecryptfs_privileged_open())
> > checks for this flag immediately after getting the mux and bows out if
> > it is set. In other words, once this flag is set, the list and its
> > contents become untouchable by anything other than
> > ecryptfs_destroy_kthread().
> 
> Unconvinced.
> 
> As soon as ecryptfs_destroy_kthread() returns, tmp_req is destroyed. 
> But it remains on ecryptfs_kthread_ctl.req_list.

I intend for ecryptfs_kthread_ctl.req_list to be irrelevant once
ecryptfs_destroy_kthread() grabs the mux and sets
(ecryptfs_kthread_ctl.flags |= ECRYPTFS_KTHREAD_ZOMBIE); nobody will
ever do anything with that list any more. The state of the list --
including the dangling list pointer -- simply does not matter any
more.

> > 	memset(&tmp_req, 0, sizeof(tmp_req));
> > 	tmp_req.flags = ECRYPTFS_REQ_ZOMBIE;
> > 	list_add_tail(&tmp_req.kthread_ctl_list,
> > 		      &ecryptfs_kthread_ctl.req_list);
> > 	mutex_unlock(&ecryptfs_kthread_ctl.mux);
> > 	wake_up(&ecryptfs_kthread_ctl.wait);
> 
> -> it's dead.

This is what gets woken up:

---
                wait_event_freezable(
                        ecryptfs_kthread_ctl.wait,
                        !list_empty(&ecryptfs_kthread_ctl.req_list));
                mutex_lock(&ecryptfs_kthread_ctl.mux);
                if (ecryptfs_kthread_ctl.flags &
                        ECRYPTFS_KTHREAD_ZOMBIE) {
                        mutex_unlock(&ecryptfs_kthread_ctl.mux);
                        goto out;
                }
---

So the flag causes the kthread to just quit, ignoring the list
altogether.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/