Re: [rtc-linux] Re: [BUG 2.6.30_rc5] Null pointer dereference in rtc-cmos driver

[Kay Sievers <kay.sievers@vrfy.org>]

On Sun, May 17, 2009 at 00:49, Andrew Morton <akpm@linux-foundation.org> wrote:

> This should fix the symptoms:

> +++ a/drivers/rtc/rtc-cmos.c
> @@ -1179,7 +1179,9 @@ static int __init cmos_init(void)
>        int retval = 0;
>
>  #ifdef CONFIG_PNP
> -       pnp_register_driver(&cmos_pnp_driver);
> +       retval = pnp_register_driver(&cmos_pnp_driver);
> +       if (retval < 0)
> +               return retval;
>  #endif

That looks like a good fix.

> Can someone please test it?

I'll try to reproduce it.

>> > > [   10.428728] BUG: unable to handle kernel NULL pointer dereference at 00000018
...
>> > > [   10.459009]  [<c03037e6>] ? driver_remove_file+0xf/0x11

> But we have arguably multiple more bugs here.
> pnp_unregister_driver(unregistered-driver) shouldn't oops the kernel.
> Neither pnp_unregister_driver() nor driver_unregister() nor
> bus_remove_driver() nor driver_remove_file() nor sysfs_remove_file()
> detected this.
>
> Kay, Greg: there are already a pile of checks on that path, but it
> seems there's one missing.  Could you please take a look and decide at
> the design level where this check should be occurring?

The issue is caused by a trying to unregister a driver which never got
registered before. I guess we try to access the private part of the
driver to reach the kobject, but for unregistered drivers there is no
such thing at that time. In:
  driver_remove_file()
we try to access the private part:
  sysfs_remove_file(&drv->p->kobj, ...
and that is NULL, for an unregistered driver, I would expect.

I'll try to reproduce the issue, and see if your fix cover the real problem.
If I can reproduce it, I'll take a look where to add a check for this.

Thanks,
Kay