Date:	Mon, 2 Jun 2014 13:10:05 +0300
From:	Dan Carpenter <dan.carpenter@...cle.com>
To:	Rickard Strandqvist <rickard_strandqvist@...ctrumdigital.se>
Cc:	Omar Ramirez Luna <omar.ramirez@...itl.com>,
	Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
	devel@...verdev.osuosl.org,
	Rashika Kheria <rashika.kheria@...il.com>,
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH] staging: tidspbridge: pmgr: dspapi.c: Cleaning up
 uninitialized variables

[ I am writing this at the end after writing the rest of this email.

  After looking at the CP_TO_USR() macro more carefully, I realize that
  the uninitialized variable bugs you are fixing are false positives.
  The information leaks where the max size is not capped are real
  security bugs.

  This code is total garbage.  It makes me irritated to look at it.
  Let's replace the stupid CP_TO_USR() macro with normal copy_to_user()
  calls, for example.  -- dan ]

On Sun, Jun 01, 2014 at 03:33:31PM +0200, Rickard Strandqvist wrote:
> There is a risk that the variable will be used without being initialized.
> 
> This was largely found by using a static code analysis program called cppcheck.
> 
> Signed-off-by: Rickard Strandqvist <rickard_strandqvist@...ctrumdigital.se>
> ---
>  drivers/staging/tidspbridge/pmgr/dspapi.c |    8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/staging/tidspbridge/pmgr/dspapi.c b/drivers/staging/tidspbridge/pmgr/dspapi.c
> index b7d5c8c..4e12a5b 100644
> --- a/drivers/staging/tidspbridge/pmgr/dspapi.c
> +++ b/drivers/staging/tidspbridge/pmgr/dspapi.c
> @@ -340,7 +340,7 @@ int api_init_complete2(void)
>  u32 mgrwrap_enum_node_info(union trapped_args *args, void *pr_ctxt)
>  {
>  	u8 *pndb_props;
> -	u32 num_nodes;
> +	u32 num_nodes = 0;
>  	int status = 0;
>  	u32 size = args->args_mgr_enumnode_info.ndb_props_size;
>  

The error handling in this function really sucks.

The "num_nodes" variable was supposed to be initialized in
mgr_enum_node_info().

The error handling in this function really sucks.  If the kmalloc()
fails then we will hit your uninitialized variable bug and also we will
hit a NULL dereference bug and crash.

Besides those two bugs, there is a third even worse bug which is that if
if "size" is greater than sizeof(struct dsp_ndbprops) then we leak the
extra information to the user.  It is a security problem.

Please could you send something to clean this up completely?

1) Add a check:
	if (size > sizeof(struct dsp_ndbprops))
		size = sizeof(struct dsp_ndbprops);

2) Return immediately on kmalloc() failure

3) if mgr_enum_node_info() fails then goto free_props, do not copy bogus
   data to the user.


> @@ -372,7 +372,7 @@ u32 mgrwrap_enum_node_info(union trapped_args *args, void *pr_ctxt)
>  u32 mgrwrap_enum_proc_info(union trapped_args *args, void *pr_ctxt)
>  {
>  	u8 *processor_info;
> -	u8 num_procs;
> +	u8 num_procs = 0;
>  	int status = 0;
>  	u32 size = args->args_mgr_enumproc_info.processor_info_size;
>  

This function has all the same problems as the previous function but
fixing it is more complicated.  The sizeof() check should be:

	if (size > sizeof(struct mgr_processorextinfo))
		size = sizeof(struct mgr_processorextinfo);

And change the kmalloc() to a kzalloc() in case the size is somewhere in
the middle of sizeof(struct dsp_processorinfo) and
sizeof(struct mgr_processorextinfo).

> @@ -475,7 +475,7 @@ u32 mgrwrap_wait_for_bridge_events(union trapped_args *args, void *pr_ctxt)
>  	int status = 0;
>  	struct dsp_notification *anotifications[MAX_EVENTS];
>  	struct dsp_notification notifications[MAX_EVENTS];
> -	u32 index, i;
> +	u32 index = 0, i;
>  	u32 count = args->args_mgr_wait.count;
>  
>  	if (count > MAX_EVENTS)

This function is total garbage as well.  Fix up the error handling.

> @@ -1755,7 +1755,7 @@ u32 strmwrap_register_notify(union trapped_args *args, void *pr_ctxt)
>   */
>  u32 strmwrap_select(union trapped_args *args, void *pr_ctxt)
>  {
> -	u32 mask;
> +	u32 mask = 0;
>  	struct strm_object *strm_tab[MAX_STREAMS];
>  	int status = 0;
>  	struct strm_res_object *strm_res;



> -- 
> 1.7.10.4
> 
> _______________________________________________
> devel mailing list
> devel@...uxdriverproject.org
> http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists