[PATCH] Allow raw_notifier callouts to unregister themselves

[PATCH] Allow raw_notifier callouts to unregister themselves

[Alan Stern]

Since raw_notifier chains don't benefit from any centralized locking
protections, they shouldn't suffer from the associated limitations.  
Under some circumstances it might make sense for a raw_notifier callout
routine to unregister itself from the notifier chain.  This patch (as678)
changes the notifier core to allow for such things.

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>

---

Index: linux-2.6.17-rc3/kernel/sys.c
===================================================================
--- linux-2.6.17-rc3.orig/kernel/sys.c
+++ linux-2.6.17-rc3/kernel/sys.c
@@ -132,14 +132,15 @@ static int __kprobes notifier_call_chain
 		unsigned long val, void *v)
 {
 	int ret = NOTIFY_DONE;
-	struct notifier_block *nb;
+	struct notifier_block *nb, *next_nb;
 
 	nb = rcu_dereference(*nl);
 	while (nb) {
+		next_nb = rcu_dereference(nb->next);
 		ret = nb->notifier_call(nb, val, v);
 		if ((ret & NOTIFY_STOP_MASK) == NOTIFY_STOP_MASK)
 			break;
-		nb = rcu_dereference(nb->next);
+		nb = next_nb;
 	}
 	return ret;
 }

Re: [PATCH] Allow raw_notifier callouts to unregister themselves

[Andrew Morton]

Alan Stern <stern@rowland.harvard.edu> wrote:
>
> Since raw_notifier chains don't benefit from any centralized locking
> protections, they shouldn't suffer from the associated limitations.  
> Under some circumstances it might make sense for a raw_notifier callout
> routine to unregister itself from the notifier chain.  This patch (as678)
> changes the notifier core to allow for such things.

ok...  Can you see any reason why 2.6.17 needs this?

Re: [PATCH] Allow raw_notifier callouts to unregister themselves

[Alan Stern]

On Thu, 11 May 2006, Andrew Morton wrote:

> Alan Stern <stern@rowland.harvard.edu> wrote:
> >
> > Since raw_notifier chains don't benefit from any centralized locking
> > protections, they shouldn't suffer from the associated limitations.  
> > Under some circumstances it might make sense for a raw_notifier callout
> > routine to unregister itself from the notifier chain.  This patch (as678)
> > changes the notifier core to allow for such things.
> 
> ok...  Can you see any reason why 2.6.17 needs this?

No.  The patch was submitted in preparation for some work that Jes
Sorensen is doing on task notifiers.  It definitely is not needed
immediately.

Alan Stern

Re: [PATCH] Allow raw_notifier callouts to unregister themselves

[David S. Miller]

From: Andrew Morton <akpm@osdl.org>
Date: Thu, 11 May 2006 11:25:09 -0700

> Alan Stern <stern@rowland.harvard.edu> wrote:
> >
> > Since raw_notifier chains don't benefit from any centralized locking
> > protections, they shouldn't suffer from the associated limitations.  
> > Under some circumstances it might make sense for a raw_notifier callout
> > routine to unregister itself from the notifier chain.  This patch (as678)
> > changes the notifier core to allow for such things.
> 
> ok...  Can you see any reason why 2.6.17 needs this?

If this patch makes raw notifiers behave more closely to the
way notifiers did before the notifier patch went into 2.6.17,
we should seriouly consider it.  We've had enough regressions
from that patch, and anything which minimizes any possible other
such regressions would be a plus.

Re: [PATCH] Allow raw_notifier callouts to unregister themselves

[Alan Stern]

On Thu, 11 May 2006, David S. Miller wrote:

> From: Andrew Morton <akpm@osdl.org>
> Date: Thu, 11 May 2006 11:25:09 -0700
> 
> > Alan Stern <stern@rowland.harvard.edu> wrote:
> > >
> > > Since raw_notifier chains don't benefit from any centralized locking
> > > protections, they shouldn't suffer from the associated limitations.  
> > > Under some circumstances it might make sense for a raw_notifier callout
> > > routine to unregister itself from the notifier chain.  This patch (as678)
> > > changes the notifier core to allow for such things.
> > 
> > ok...  Can you see any reason why 2.6.17 needs this?
> 
> If this patch makes raw notifiers behave more closely to the
> way notifiers did before the notifier patch went into 2.6.17,
> we should seriouly consider it.  We've had enough regressions
> from that patch, and anything which minimizes any possible other
> such regressions would be a plus.

Sorry I'm a little late replying to this...

Actually this patch makes raw notifiers behave a little _less_ like the
original pre-2.6.17 notifiers.  With the original code, a notifier callout
routine could unregister itself and deallocate its notifier_block, thereby
causing an oops when it returned.  With this patch, the oops will not
occur.

That's the only difference.

Alan Stern