[PATCH] pinctrl: devicetree: fix refcount leak in pinctrl_dt_to_map()

[PATCH] pinctrl: devicetree: fix refcount leak in pinctrl_dt_to_map()

[Zeng Heng]

If we fail to allocate propname buffer, we need to drop the reference
count we just took. Because the pinctrl_dt_free_maps() includes the
droping operation, here we call it directly.

Fixes: 91d5c5060ee2 ("pinctrl: devicetree: fix null pointer dereferencing in pinctrl_dt_to_map")
Suggested-by: Dan Carpenter <dan.carpenter@linaro.org>
Signed-off-by: Zeng Heng <zengheng4@huawei.com>
---
 drivers/pinctrl/devicetree.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/drivers/pinctrl/devicetree.c b/drivers/pinctrl/devicetree.c
index df1efc2e5202..6a94ecd6a8de 100644
--- a/drivers/pinctrl/devicetree.c
+++ b/drivers/pinctrl/devicetree.c
@@ -220,14 +220,16 @@ int pinctrl_dt_to_map(struct pinctrl *p, struct pinctrl_dev *pctldev)
 	for (state = 0; ; state++) {
 		/* Retrieve the pinctrl-* property */
 		propname = kasprintf(GFP_KERNEL, "pinctrl-%d", state);
-		if (!propname)
-			return -ENOMEM;
+		if (!propname) {
+			ret = -ENOMEM;
+			goto err;
+		}
 		prop = of_find_property(np, propname, &size);
 		kfree(propname);
 		if (!prop) {
 			if (state == 0) {
-				of_node_put(np);
-				return -ENODEV;
+				ret = -ENODEV;
+				goto err;
 			}
 			break;
 		}
-- 
2.25.1

Re: [PATCH] pinctrl: devicetree: fix refcount leak in pinctrl_dt_to_map()

[Dan Carpenter]

On Mon, Apr 15, 2024 at 06:53:28PM +0800, Zeng Heng wrote:
> If we fail to allocate propname buffer, we need to drop the reference
> count we just took. Because the pinctrl_dt_free_maps() includes the
> droping operation, here we call it directly.
> 
> Fixes: 91d5c5060ee2 ("pinctrl: devicetree: fix null pointer dereferencing in pinctrl_dt_to_map")
> Suggested-by: Dan Carpenter <dan.carpenter@linaro.org>
> Signed-off-by: Zeng Heng <zengheng4@huawei.com>
> ---

Thanks!

Reviewed-by: Dan Carpenter <dan.carpenter@linaro.org>

regards,
dan carpenter

Re: [PATCH] pinctrl: devicetree: fix refcount leak in pinctrl_dt_to_map()

[Markus Elfring]

>                   … Because the pinctrl_dt_free_maps() includes the
> droping operation, here we call it directly.

I find this change description improvable.

* How do you think about to avoid a typo?

* Would another imperative wording be more desirable?

Regards,
Markus

Re: [PATCH] pinctrl: devicetree: fix refcount leak in pinctrl_dt_to_map()

[Linus Walleij]

On Mon, Apr 15, 2024 at 12:54 PM Zeng Heng <zengheng4@huawei.com> wrote:

> If we fail to allocate propname buffer, we need to drop the reference
> count we just took. Because the pinctrl_dt_free_maps() includes the
> droping operation, here we call it directly.
>
> Fixes: 91d5c5060ee2 ("pinctrl: devicetree: fix null pointer dereferencing in pinctrl_dt_to_map")
> Suggested-by: Dan Carpenter <dan.carpenter@linaro.org>
> Signed-off-by: Zeng Heng <zengheng4@huawei.com>

Patch applied for fixes.

Yours,
Linus Walleij

Re: [PATCH] pinctrl: devicetree: fix refcount leak in pinctrl_dt_to_map()

[Andy Shevchenko]

On Mon, Apr 15, 2024 at 06:53:28PM +0800, Zeng Heng wrote:
> If we fail to allocate propname buffer, we need to drop the reference
> count we just took. Because the pinctrl_dt_free_maps() includes the
> droping operation, here we call it directly.

...

>  	for (state = 0; ; state++) {
>  		/* Retrieve the pinctrl-* property */
>  		propname = kasprintf(GFP_KERNEL, "pinctrl-%d", state);
> -		if (!propname)
> -			return -ENOMEM;
> +		if (!propname) {
> +			ret = -ENOMEM;
> +			goto err;
> +		}
>  		prop = of_find_property(np, propname, &size);
>  		kfree(propname);
>  		if (!prop) {
>  			if (state == 0) {
> -				of_node_put(np);
> -				return -ENODEV;
> +				ret = -ENODEV;
> +				goto err;

Has it been tested? How on earth is this a correct change?

We iterate over state numbers until we have properties available. This chunk is
_successful_ exit path, we may not free parsed maps! Am I wrong?

>  			}
>  			break;
>  		}

-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH] pinctrl: devicetree: fix refcount leak in pinctrl_dt_to_map()

[Dan Carpenter]

On Wed, Apr 17, 2024 at 06:30:59PM +0300, Andy Shevchenko wrote:
> On Mon, Apr 15, 2024 at 06:53:28PM +0800, Zeng Heng wrote:
> > If we fail to allocate propname buffer, we need to drop the reference
> > count we just took. Because the pinctrl_dt_free_maps() includes the
> > droping operation, here we call it directly.
> 
> ...
> 
> >  	for (state = 0; ; state++) {
> >  		/* Retrieve the pinctrl-* property */
> >  		propname = kasprintf(GFP_KERNEL, "pinctrl-%d", state);
> > -		if (!propname)
> > -			return -ENOMEM;
> > +		if (!propname) {
> > +			ret = -ENOMEM;
> > +			goto err;
> > +		}
> >  		prop = of_find_property(np, propname, &size);
> >  		kfree(propname);
> >  		if (!prop) {
> >  			if (state == 0) {
> > -				of_node_put(np);
> > -				return -ENODEV;
> > +				ret = -ENODEV;
> > +				goto err;
> 
> Has it been tested? How on earth is this a correct change?
> 
> We iterate over state numbers until we have properties available. This chunk is
> _successful_ exit path, we may not free parsed maps! Am I wrong?

In this path state == 0 so we haven't had a successful iteration yet.

regards,
dan carpenter

Re: [PATCH] pinctrl: devicetree: fix refcount leak in pinctrl_dt_to_map()

[Andy Shevchenko]

On Wed, Apr 17, 2024 at 06:38:46PM +0300, Dan Carpenter wrote:
> On Wed, Apr 17, 2024 at 06:30:59PM +0300, Andy Shevchenko wrote:
> > On Mon, Apr 15, 2024 at 06:53:28PM +0800, Zeng Heng wrote:

...

> > >  	for (state = 0; ; state++) {
> > >  		/* Retrieve the pinctrl-* property */
> > >  		propname = kasprintf(GFP_KERNEL, "pinctrl-%d", state);
> > > -		if (!propname)
> > > -			return -ENOMEM;
> > > +		if (!propname) {
> > > +			ret = -ENOMEM;
> > > +			goto err;
> > > +		}
> > >  		prop = of_find_property(np, propname, &size);
> > >  		kfree(propname);
> > >  		if (!prop) {
> > >  			if (state == 0) {
> > > -				of_node_put(np);
> > > -				return -ENODEV;
> > > +				ret = -ENODEV;
> > > +				goto err;
> > 
> > Has it been tested? How on earth is this a correct change?
> > 
> > We iterate over state numbers until we have properties available. This chunk is
> > _successful_ exit path, we may not free parsed maps! Am I wrong?
> 
> In this path state == 0 so we haven't had a successful iteration yet.

Ah, indeed, this is not a status. Okay, makes sense, but calling that free
function for the purpose of the putting of_node seems an overkill...

-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH] pinctrl: devicetree: fix refcount leak in pinctrl_dt_to_map()

[Dan Carpenter]

On Wed, Apr 17, 2024 at 08:12:23PM +0300, Andy Shevchenko wrote:
> On Wed, Apr 17, 2024 at 06:38:46PM +0300, Dan Carpenter wrote:
> > On Wed, Apr 17, 2024 at 06:30:59PM +0300, Andy Shevchenko wrote:
> > > On Mon, Apr 15, 2024 at 06:53:28PM +0800, Zeng Heng wrote:
> 
> ...
> 
> > > >  	for (state = 0; ; state++) {
> > > >  		/* Retrieve the pinctrl-* property */
> > > >  		propname = kasprintf(GFP_KERNEL, "pinctrl-%d", state);
> > > > -		if (!propname)
> > > > -			return -ENOMEM;
> > > > +		if (!propname) {
> > > > +			ret = -ENOMEM;
> > > > +			goto err;
> > > > +		}
> > > >  		prop = of_find_property(np, propname, &size);
> > > >  		kfree(propname);
> > > >  		if (!prop) {
> > > >  			if (state == 0) {
> > > > -				of_node_put(np);
> > > > -				return -ENODEV;
> > > > +				ret = -ENODEV;
> > > > +				goto err;
> > > 
> > > Has it been tested? How on earth is this a correct change?
> > > 
> > > We iterate over state numbers until we have properties available. This chunk is
> > > _successful_ exit path, we may not free parsed maps! Am I wrong?
> > 
> > In this path state == 0 so we haven't had a successful iteration yet.
> 
> Ah, indeed, this is not a status. Okay, makes sense, but calling that free
> function for the purpose of the putting of_node seems an overkill...

Sure, that's one way to look at it, but it's suspicious looking when
there is a direct return which is surrounded by gotos.  As I write this,
I remember that Smatch has a warning for code like that.

Probably we should add a comment to say:

	/* Return -ENODEV if the property 'pinctrl-0' is not present. */

regards,
dan carpenter

Re: [PATCH] pinctrl: devicetree: fix refcount leak in pinctrl_dt_to_map()

[Andy Shevchenko]

On Wed, Apr 17, 2024 at 08:49:42PM +0300, Dan Carpenter wrote:
> On Wed, Apr 17, 2024 at 08:12:23PM +0300, Andy Shevchenko wrote:
> > On Wed, Apr 17, 2024 at 06:38:46PM +0300, Dan Carpenter wrote:
> > > On Wed, Apr 17, 2024 at 06:30:59PM +0300, Andy Shevchenko wrote:
> > > > On Mon, Apr 15, 2024 at 06:53:28PM +0800, Zeng Heng wrote:

...

> > > > >  	for (state = 0; ; state++) {
> > > > >  		/* Retrieve the pinctrl-* property */
> > > > >  		propname = kasprintf(GFP_KERNEL, "pinctrl-%d", state);
> > > > > -		if (!propname)
> > > > > -			return -ENOMEM;
> > > > > +		if (!propname) {
> > > > > +			ret = -ENOMEM;
> > > > > +			goto err;
> > > > > +		}
> > > > >  		prop = of_find_property(np, propname, &size);
> > > > >  		kfree(propname);
> > > > >  		if (!prop) {
> > > > >  			if (state == 0) {
> > > > > -				of_node_put(np);
> > > > > -				return -ENODEV;
> > > > > +				ret = -ENODEV;
> > > > > +				goto err;
> > > > 
> > > > Has it been tested? How on earth is this a correct change?
> > > > 
> > > > We iterate over state numbers until we have properties available. This chunk is
> > > > _successful_ exit path, we may not free parsed maps! Am I wrong?
> > > 
> > > In this path state == 0 so we haven't had a successful iteration yet.
> > 
> > Ah, indeed, this is not a status. Okay, makes sense, but calling that free
> > function for the purpose of the putting of_node seems an overkill...
> 
> Sure, that's one way to look at it, but it's suspicious looking when
> there is a direct return which is surrounded by gotos.  As I write this,
> I remember that Smatch has a warning for code like that.
> 
> Probably we should add a comment to say:
> 
> 	/* Return -ENODEV if the property 'pinctrl-0' is not present. */

Good idea, go for it!

-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH] pinctrl: devicetree: fix refcount leak in pinctrl_dt_to_map()

[Linus Walleij]

On Wed, Apr 17, 2024 at 7:49 PM Dan Carpenter <dan.carpenter@linaro.org> wrote:

> Probably we should add a comment to say:
>
>         /* Return -ENODEV if the property 'pinctrl-0' is not present. */

Would you mind sending a oneliner on top to fix this?

Yours,
Linus Walleij