[PATCH] mmc: pwrseq: Use kmalloc_array instead of stack VLA

[PATCH] mmc: pwrseq: Use kmalloc_array instead of stack VLA

[Tobin C. Harding]

The use of stack Variable Length Arrays needs to be avoided, as they
can be a vector for stack exhaustion, which can be both a runtime bug
(kernel Oops) or a security flaw (overwriting memory beyond the
stack). Also, in general, as code evolves it is easy to lose track of
how big a VLA can get. Thus, we can end up having runtime failures
that are hard to debug. As part of the directive[1] to remove all VLAs
from the kernel, and build with -Wvla.

Currently driver is using a VLA declared using the number of descriptors.  This
array is used to store integer values and is later used as an argument to
`gpiod_set_array_value_cansleep()` This can be avoided by using
`kmalloc_array()` to allocate memory for the array of integer values.  Memory is
free'd before return from function.

>From the code it appears that it is safe to sleep so we can use GFP_KERNEL
(based _cansleep() suffix of function `gpiod_set_array_value_cansleep()`.

It can be expected that this patch will result in a small increase in overhead
due to the use of `kmalloc_array()`

[1] https://lkml.org/lkml/2018/3/7/621

Signed-off-by: Tobin C. Harding <me@tobin.cc>
---

Patch is untested.  I was not able to fully grok the call chain involved
with this change so was unable to quantify the total degradation
of performance.

Line calling to gpiod_set_array_value_cansleep() is 83 characters long,
patch attempts to leave the code as clean as possible.  Open to
suggestions for improvement.  I tried various forms of while loop and
pointer arithmetic for the setting of `values` but in the end settled
for array indexing, again open to suggestions.

thanks,
Tobin.

 drivers/mmc/core/pwrseq_simple.c | 14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)

diff --git a/drivers/mmc/core/pwrseq_simple.c b/drivers/mmc/core/pwrseq_simple.c
index 13ef162cf066..a8b9fee4d62a 100644
--- a/drivers/mmc/core/pwrseq_simple.c
+++ b/drivers/mmc/core/pwrseq_simple.c
@@ -40,14 +40,18 @@ static void mmc_pwrseq_simple_set_gpios_value(struct mmc_pwrseq_simple *pwrseq,
 	struct gpio_descs *reset_gpios = pwrseq->reset_gpios;
 
 	if (!IS_ERR(reset_gpios)) {
-		int i;
-		int values[reset_gpios->ndescs];
+		int i, *values;
+		int nvalues = reset_gpios->ndescs;
 
-		for (i = 0; i < reset_gpios->ndescs; i++)
+		values = kmalloc_array(nvalues, sizeof(int), GFP_KERNEL);
+		if (!values)
+			return;
+
+		for (i = 0; i < nvalues; i++)
 			values[i] = value;
 
-		gpiod_set_array_value_cansleep(
-			reset_gpios->ndescs, reset_gpios->desc, values);
+		gpiod_set_array_value_cansleep(nvalues, reset_gpios->desc, values);
+		kfree(values);
 	}
 }
 
-- 
2.7.4

Re: [PATCH] mmc: pwrseq: Use kmalloc_array instead of stack VLA

[Ulf Hansson]

On 26 March 2018 at 08:33, Tobin C. Harding <me@tobin.cc> wrote:
> The use of stack Variable Length Arrays needs to be avoided, as they
> can be a vector for stack exhaustion, which can be both a runtime bug
> (kernel Oops) or a security flaw (overwriting memory beyond the
> stack). Also, in general, as code evolves it is easy to lose track of
> how big a VLA can get. Thus, we can end up having runtime failures
> that are hard to debug. As part of the directive[1] to remove all VLAs
> from the kernel, and build with -Wvla.
>
> Currently driver is using a VLA declared using the number of descriptors.  This
> array is used to store integer values and is later used as an argument to
> `gpiod_set_array_value_cansleep()` This can be avoided by using
> `kmalloc_array()` to allocate memory for the array of integer values.  Memory is
> free'd before return from function.
>
> From the code it appears that it is safe to sleep so we can use GFP_KERNEL
> (based _cansleep() suffix of function `gpiod_set_array_value_cansleep()`.
>
> It can be expected that this patch will result in a small increase in overhead
> due to the use of `kmalloc_array()`
>
> [1] https://lkml.org/lkml/2018/3/7/621
>
> Signed-off-by: Tobin C. Harding <me@tobin.cc>

Thanks, queued for 3.18!

Kind regards
Uffe

> ---
>
> Patch is untested.  I was not able to fully grok the call chain involved
> with this change so was unable to quantify the total degradation
> of performance.
>
> Line calling to gpiod_set_array_value_cansleep() is 83 characters long,
> patch attempts to leave the code as clean as possible.  Open to
> suggestions for improvement.  I tried various forms of while loop and
> pointer arithmetic for the setting of `values` but in the end settled
> for array indexing, again open to suggestions.
>
> thanks,
> Tobin.
>
>  drivers/mmc/core/pwrseq_simple.c | 14 +++++++++-----
>  1 file changed, 9 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/mmc/core/pwrseq_simple.c b/drivers/mmc/core/pwrseq_simple.c
> index 13ef162cf066..a8b9fee4d62a 100644
> --- a/drivers/mmc/core/pwrseq_simple.c
> +++ b/drivers/mmc/core/pwrseq_simple.c
> @@ -40,14 +40,18 @@ static void mmc_pwrseq_simple_set_gpios_value(struct mmc_pwrseq_simple *pwrseq,
>         struct gpio_descs *reset_gpios = pwrseq->reset_gpios;
>
>         if (!IS_ERR(reset_gpios)) {
> -               int i;
> -               int values[reset_gpios->ndescs];
> +               int i, *values;
> +               int nvalues = reset_gpios->ndescs;
>
> -               for (i = 0; i < reset_gpios->ndescs; i++)
> +               values = kmalloc_array(nvalues, sizeof(int), GFP_KERNEL);
> +               if (!values)
> +                       return;
> +
> +               for (i = 0; i < nvalues; i++)
>                         values[i] = value;
>
> -               gpiod_set_array_value_cansleep(
> -                       reset_gpios->ndescs, reset_gpios->desc, values);
> +               gpiod_set_array_value_cansleep(nvalues, reset_gpios->desc, values);
> +               kfree(values);
>         }
>  }
>
> --
> 2.7.4
>

Re: [PATCH] mmc: pwrseq: Use kmalloc_array instead of stack VLA

[Kees Cook]

On Wed, Apr 4, 2018 at 5:45 AM, Ulf Hansson <ulf.hansson@linaro.org> wrote:
> On 26 March 2018 at 08:33, Tobin C. Harding <me@tobin.cc> wrote:
>> The use of stack Variable Length Arrays needs to be avoided, as they
>> can be a vector for stack exhaustion, which can be both a runtime bug
>> (kernel Oops) or a security flaw (overwriting memory beyond the
>> stack). Also, in general, as code evolves it is easy to lose track of
>> how big a VLA can get. Thus, we can end up having runtime failures
>> that are hard to debug. As part of the directive[1] to remove all VLAs
>> from the kernel, and build with -Wvla.
>>
>> Currently driver is using a VLA declared using the number of descriptors.  This
>> array is used to store integer values and is later used as an argument to
>> `gpiod_set_array_value_cansleep()` This can be avoided by using
>> `kmalloc_array()` to allocate memory for the array of integer values.  Memory is
>> free'd before return from function.
>>
>> From the code it appears that it is safe to sleep so we can use GFP_KERNEL
>> (based _cansleep() suffix of function `gpiod_set_array_value_cansleep()`.
>>
>> It can be expected that this patch will result in a small increase in overhead
>> due to the use of `kmalloc_array()`
>>
>> [1] https://lkml.org/lkml/2018/3/7/621
>>
>> Signed-off-by: Tobin C. Harding <me@tobin.cc>
>
> Thanks, queued for 3.18!

Time travel! ;)

-Kees

-- 
Kees Cook
Pixel Security