Re: Yet another pipe related oops.

[Greg Kroah-Hartman <gregkh@linuxfoundation.org>]

On Mon, Apr 01, 2013 at 10:21:42PM +0100, Al Viro wrote:
> On Mon, Apr 01, 2013 at 02:00:29PM -0700, Greg Kroah-Hartman wrote:
> 
> > > IOW, how do we deal with a race between attempt to open a debugfs file and
> > > its removal on driver unload?  Greg?
> > 
> > Hm, I thought the i_fop->owner thing would be the needed protection, but
> 
> It will be, if you manage to fetch it...

I agree.

> > I guess you are right, it will not.  I guess we need to do what
> > character devices do and have an "intermediate" fops in order to protect
> > this.  Would that work?
> 
> You mean, with reassigning ->f_op in ->open()?  That'll work, as long as
> we have exclusion between removal and fetching the sucker in primary
> ->open()...  Where would you prefer to stash fops?

Ick, that's not going to work as the current api just uses a fops and
debugfs doesn't keep anything else hanging around that referes to
something "before" that, like 'struct cdev' does.

And, it's even worse, look at the use of DEFINE_SIMPLE_ATTRIBUTE(),
those take a pointer from a random module to read/write from, and use
the fops for the debugfs module.  Hopefully no other user of that macro
has the same problem, and at first glance, I think that's true, but I
might be wrong...

Am I allowed to "punt" and say, "removing a module that uses debugfs is
not recommended?" :)

greg k-h