2015-05-06 21:49:02

by Alexey Khoroshilov

Subject: [PATCH] iio: hid-sensors: Fix memory leak on failure path in hid_prox_probe()

If prox_parse_report() fails, memory allocated for channels is not
deallocated, since it is still in local variable channels
while kfree() is called with indio_dev->channels.

Found by Linux Driver Verification project (linuxtesting.org).

Signed-off-by: Alexey Khoroshilov <[email protected]>
---
drivers/iio/light/hid-sensor-prox.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/iio/light/hid-sensor-prox.c b/drivers/iio/light/hid-sensor-prox.c
index 91ecc46ffeaa..d0d188108a11 100644
--- a/drivers/iio/light/hid-sensor-prox.c
+++ b/drivers/iio/light/hid-sensor-prox.c
@@ -281,8 +281,9 @@ static int hid_prox_probe(struct platform_device *pdev)
ret = prox_parse_report(pdev, hsdev, channels,
HID_USAGE_SENSOR_PROX, prox_state);
if (ret) {
+ kfree(channels);
dev_err(&pdev->dev, "failed to setup attributes\n");
- goto error_free_dev_mem;
+ return ret;
}

indio_dev->channels = channels;
--
1.9.1

2015-05-06 22:01:02

by Fabio Estevam

Subject: Re: [PATCH] iio: hid-sensors: Fix memory leak on failure path in hid_prox_probe()

On Wed, May 6, 2015 at 6:48 PM, Alexey Khoroshilov
<[email protected]> wrote:
> If prox_parse_report() fails, memory allocated for channels is not
> deallocated, since it is still in local variable channels
> while kfree() is called with indio_dev->channels.
>
> Found by Linux Driver Verification project (linuxtesting.org).
>
> Signed-off-by: Alexey Khoroshilov <[email protected]>
> ---
> drivers/iio/light/hid-sensor-prox.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/iio/light/hid-sensor-prox.c b/drivers/iio/light/hid-sensor-prox.c
> index 91ecc46ffeaa..d0d188108a11 100644
> --- a/drivers/iio/light/hid-sensor-prox.c
> +++ b/drivers/iio/light/hid-sensor-prox.c
> @@ -281,8 +281,9 @@ static int hid_prox_probe(struct platform_device *pdev)
> ret = prox_parse_report(pdev, hsdev, channels,
> HID_USAGE_SENSOR_PROX, prox_state);
> if (ret) {
> + kfree(channels);
> dev_err(&pdev->dev, "failed to setup attributes\n");
> - goto error_free_dev_mem;
> + return ret;

Then the other calls to error_free_dev_mem will also miss to call
'kfree(channels)'.

What about this fix instead?

--- a/drivers/iio/light/hid-sensor-prox.c
+++ b/drivers/iio/light/hid-sensor-prox.c
@@ -332,7 +332,7 @@ error_remove_trigger:
error_unreg_buffer_funcs:
iio_triggered_buffer_cleanup(indio_dev);
error_free_dev_mem:
- kfree(indio_dev->channels);
+ kfree(channels);
return ret;
}

Regards,

Fabio Estevam

2015-05-06 22:14:29

by Alexey Khoroshilov

Subject: Re: [PATCH] iio: hid-sensors: Fix memory leak on failure path in hid_prox_probe()

On 07.05.2015 01:00, Fabio Estevam wrote:
> On Wed, May 6, 2015 at 6:48 PM, Alexey Khoroshilov
> <[email protected]> wrote:
>> If prox_parse_report() fails, memory allocated for channels is not
>> deallocated, since it is still in local variable channels
>> while kfree() is called with indio_dev->channels.
>>
>> Found by Linux Driver Verification project (linuxtesting.org).
>>
>> Signed-off-by: Alexey Khoroshilov <[email protected]>
>> ---
>> drivers/iio/light/hid-sensor-prox.c | 3 ++-
>> 1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/iio/light/hid-sensor-prox.c b/drivers/iio/light/hid-sensor-prox.c
>> index 91ecc46ffeaa..d0d188108a11 100644
>> --- a/drivers/iio/light/hid-sensor-prox.c
>> +++ b/drivers/iio/light/hid-sensor-prox.c
>> @@ -281,8 +281,9 @@ static int hid_prox_probe(struct platform_device *pdev)
>> ret = prox_parse_report(pdev, hsdev, channels,
>> HID_USAGE_SENSOR_PROX, prox_state);
>> if (ret) {
>> + kfree(channels);
>> dev_err(&pdev->dev, "failed to setup attributes\n");
>> - goto error_free_dev_mem;
>> + return ret;
>
> Then the other calls to error_free_dev_mem will also miss to call
> 'kfree(channels)'.
>

Not exactly. Other calls are after
indio_dev->channels = channels;
So,
error_free_dev_mem:
kfree(indio_dev->channels);
works for them well.

--
Alexey

2015-05-06 22:32:48

by Fabio Estevam

Subject: Re: [PATCH] iio: hid-sensors: Fix memory leak on failure path in hid_prox_probe()

On Wed, May 6, 2015 at 7:14 PM, Alexey Khoroshilov
<[email protected]> wrote:

> Not exactly. Other calls are after
> indio_dev->channels = channels;
> So,
> error_free_dev_mem:
> kfree(indio_dev->channels);
> works for them well.

indio_dev is allocated using devm_ , so you don't need to free it.

Your patch is not correct because you only kfree(channels) in the
prox_parse_report() error case, but you missed the other subsequent
functions.

2015-05-06 23:14:06

by Alexey Khoroshilov

Subject: Re: [PATCH] iio: hid-sensors: Fix memory leak on failure path in hid_prox_probe()

On 07.05.2015 01:32, Fabio Estevam wrote:
> On Wed, May 6, 2015 at 7:14 PM, Alexey Khoroshilov
> <[email protected]> wrote:
>
>> Not exactly. Other calls are after
>> indio_dev->channels = channels;
>> So,
>> error_free_dev_mem:
>> kfree(indio_dev->channels);
>> works for them well.
>
> indio_dev is allocated using devm_ , so you don't need to free it.
>
> Your patch is not correct because you only kfree(channels) in the
> prox_parse_report() error case, but you missed the other subsequent
> functions.
>

No! The other subsequent functions are AFTER (prox_parse_report() error
case is the only BEFORE)
indio_dev->channels = channels;

and all consequent error cases comes to error_free_dev_mem, where

error_free_dev_mem:
kfree(indio_dev->channels);

that is equivalent to kfree(channels);

--
Alexey

2015-05-07 07:43:43

by Daniel Baluta

Subject: Re: [PATCH] iio: hid-sensors: Fix memory leak on failure path in hid_prox_probe()

On Thu, May 7, 2015 at 1:00 AM, Fabio Estevam <[email protected]> wrote:
> On Wed, May 6, 2015 at 6:48 PM, Alexey Khoroshilov
> <[email protected]> wrote:
>> If prox_parse_report() fails, memory allocated for channels is not
>> deallocated, since it is still in local variable channels
>> while kfree() is called with indio_dev->channels.
>>
>> Found by Linux Driver Verification project (linuxtesting.org).
>>
>> Signed-off-by: Alexey Khoroshilov <[email protected]>
>> ---
>> drivers/iio/light/hid-sensor-prox.c | 3 ++-
>> 1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/iio/light/hid-sensor-prox.c b/drivers/iio/light/hid-sensor-prox.c
>> index 91ecc46ffeaa..d0d188108a11 100644
>> --- a/drivers/iio/light/hid-sensor-prox.c
>> +++ b/drivers/iio/light/hid-sensor-prox.c
>> @@ -281,8 +281,9 @@ static int hid_prox_probe(struct platform_device *pdev)
>> ret = prox_parse_report(pdev, hsdev, channels,
>> HID_USAGE_SENSOR_PROX, prox_state);
>> if (ret) {
>> + kfree(channels);
>> dev_err(&pdev->dev, "failed to setup attributes\n");
>> - goto error_free_dev_mem;
>> + return ret;
>
> Then the other calls to error_free_dev_mem will also miss to call
> 'kfree(channels)'.
>
> What about this fix instead?
>
> --- a/drivers/iio/light/hid-sensor-prox.c
> +++ b/drivers/iio/light/hid-sensor-prox.c
> @@ -332,7 +332,7 @@ error_remove_trigger:
> error_unreg_buffer_funcs:
> iio_triggered_buffer_cleanup(indio_dev);
> error_free_dev_mem:
> - kfree(indio_dev->channels);
> + kfree(channels);
> return ret;
> }

Both patches are correct and I think we should go
with Fabio's version since it's consistent with the
rest of the code.

thanks,
Daniel.

2015-05-07 09:18:34

by Jonathan Cameron

Subject: Re: [PATCH] iio: hid-sensors: Fix memory leak on failure path in hid_prox_probe()

On 07/05/15 08:43, Daniel Baluta wrote:
> On Thu, May 7, 2015 at 1:00 AM, Fabio Estevam <[email protected]> wrote:
>> On Wed, May 6, 2015 at 6:48 PM, Alexey Khoroshilov
>> <[email protected]> wrote:
>>> If prox_parse_report() fails, memory allocated for channels is not
>>> deallocated, since it is still in local variable channels
>>> while kfree() is called with indio_dev->channels.
>>>
>>> Found by Linux Driver Verification project (linuxtesting.org).
>>>
>>> Signed-off-by: Alexey Khoroshilov <[email protected]>
>>> ---
>>> drivers/iio/light/hid-sensor-prox.c | 3 ++-
>>> 1 file changed, 2 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/iio/light/hid-sensor-prox.c b/drivers/iio/light/hid-sensor-prox.c
>>> index 91ecc46ffeaa..d0d188108a11 100644
>>> --- a/drivers/iio/light/hid-sensor-prox.c
>>> +++ b/drivers/iio/light/hid-sensor-prox.c
>>> @@ -281,8 +281,9 @@ static int hid_prox_probe(struct platform_device *pdev)
>>> ret = prox_parse_report(pdev, hsdev, channels,
>>> HID_USAGE_SENSOR_PROX, prox_state);
>>> if (ret) {
>>> + kfree(channels);
>>> dev_err(&pdev->dev, "failed to setup attributes\n");
>>> - goto error_free_dev_mem;
>>> + return ret;
>>
>> Then the other calls to error_free_dev_mem will also miss to call
>> 'kfree(channels)'.
>>
>> What about this fix instead?
>>
>> --- a/drivers/iio/light/hid-sensor-prox.c
>> +++ b/drivers/iio/light/hid-sensor-prox.c
>> @@ -332,7 +332,7 @@ error_remove_trigger:
>> error_unreg_buffer_funcs:
>> iio_triggered_buffer_cleanup(indio_dev);
>> error_free_dev_mem:
>> - kfree(indio_dev->channels);
>> + kfree(channels);
>> return ret;
>> }
>
> Both patches are correct and I think we should go
> with Fabio's version since it's consistent with the
> rest of the code.
>
> thanks,
> Daniel.
>
Agreed. I'm travelling (again, yawn) for the next few days
but Fabio, could you send a formal version of your patch
with signoffs etc and a reported by for Alexey (or under
the circumstances Alexey, feel free to sign off on it
as well).

Thanks,

Jonathan