Re: Top 10 kernel oopses for the week ending January 5th, 2008

On Thursday January 10, viro@ZenIV.linux.org.uk wrote:

On Thu, Jan 10, 2008 at 03:13:48PM +1100, Neil Brown wrote:

quoted

quoted

What guarantees that it doesn't happen before we get to callback?  AFAICS,
nothing whatsoever...

Yes, that's bad isn't it :-)

I think I should be using sysfs_schedule_callback here.  That makes the 
required 'get' and 'put' calls.... but it can fail with -ENOMEM.  I
wonder what I do if -ENOMEM???  Maybe I'll just continue to roll my
one :-( 

How about this instead (completely untested)

	* split failure exits
	* switch to kick_rdev_from_array()
	* fold unbind_rdev_from_array() into it (no other callers anymore)
	* take export_rdev() into failure case in bind_rdev_to_array()
	* in kick_rdev_from_array() do what export_rdev() does sans
kobject_put() and do that before schedule_work().  Take kobject_put() into
delayed_delete().

While there are probably some good ideas in there, I think fixing this
particular bug is much simpler.  Just take a reference to the object
before scheduling the worker, and drop it when the worker has done
its work.

I have a closer look at the idea of no required export_rdev after a
failed bind_rdev_to_array.  On the surface it does seem to make the
code nicer.

Thanks,
NeilBrown