Re: [PATCH] staging; lirc, zilog: put_ir_rx may free 'rx' which can lead to double free

[Andy Walls <awalls@md.metrocast.net>]

On Sun, 2011-07-31 at 08:42 -0400, Andy Walls wrote:
> On Fri, 2011-07-29 at 09:08 +0300, Dan Carpenter wrote:
> > On Thu, Jul 28, 2011 at 11:49:51PM +0200, Jesper Juhl wrote:
> > > If calling put_ir_rx(rx, true); in
> > > drivers/staging/lirc/lirc_zilog.c::ir_probe() returns true (1) then it
> > > means that it has freed it's first argument. Subsequently jumping to
> > > 'out_put_xx' will cause us to call put_ir_rx() once more since 'rx' is
> > > not zero - leading to a double free.
> > 
> > It would be better to just remove the first call to put_ir_rx().
> 
> Jesper,
> 
> (Emails from you don't seem to make it to me, so I looked at your patch
> in lkml.org archive.)
> 
> Good catch!

Jesper,

Ah crud.  I take that back.

NACK on either form of the patch.

lirc_zilog is a ref counting PITA.  Look at ir_probe() again.

We grab 2 rx refs.

One at the line 

	kref_init(&rx->ref);

and another at

	/* An rx ref goes to the i2c_client */
        i2c_set_clientdata(client, get_ir_rx(ir));


On failure, we need two "put_ir_rx(rx, true)" calls.

That's why I wrote the comment to myself:

         /* Failure exit, so put back rx ref from i2c_client */
         i2c_set_clientdata(client, NULL);
         put_ir_rx(rx, true);

Regards,
Andy