Re: [CHECKER][PATCH] cmpci user-pointer fix

[Raja R Harinath <harinath@cs.umn.edu>]

Hi,

Hollis Blanchard <hollisb@us.ibm.com> writes:
[snip]
> I believe the attached patch fixes it. cm_write was calling access_ok,
> but after that you must still access user space through the
> get/put/copy*_user functions. It should be safe to return -EFAULT at
> these points in cm_write, since there are other returns already in the
> code above and below that. Compile-tested only.
[snip]
> --- linux-2.5.70/sound/oss/cmpci.c.orig	Sat May 24 19:00:00 2003
> +++ linux-2.5.70/sound/oss/cmpci.c	Wed May 28 14:53:15 2003
> @@ -580,15 +580,17 @@
>  	spin_unlock_irqrestore(&s->lock, flags);
>  }
>  
> -static void trans_ac3(struct cm_state *s, void *dest, const char *source, int size)
> +static int trans_ac3(struct cm_state *s, void *dest, const char *source, int size)

Shouldn't 'source' get the new __user annotation, then:

  const char * __user source

IIRC.

>  {
>  	int   i = size / 2;
> +	int err;
>  	unsigned long data;
>  	unsigned long *dst = (unsigned long *) dest;
>  	unsigned short *src = (unsigned short *)source;

Likewise with 'src'.

- Hari
-- 
Raja R Harinath ------------------------------ harinath@cs.umn.edu