Re: [drivers/gpio] Question about `ljca_gpio_config`: misuse of __counted_by

[Gustavo A. R. Silva]

Hi all,

On 26/07/24 14:07, Linus Walleij wrote:
> Hi Haoyu,
>
> On Wed, Jul 24, 2024 at 11:12 AM Haoyu Li <lihaoyu499@xxxxxxxxx> wrote:
>
> > Dear Linux Developers for GPIO SUBSYSTEM,
> >
> > We are curious about the use of `struct ljca_gpio_packet *packet` in the function `ljca_gpio_config` (https://elixir.bootlin.com/linux/v6.10/source/drivers/gpio/gpio-ljca.c#L80).
> > ```
> > static int ljca_gpio_config(struct ljca_gpio_dev *ljca_gpio, u8 gpio_id,
> >     u8 config)
> > {
> > struct ljca_gpio_packet *packet =
> > (struct ljca_gpio_packet *)ljca_gpio->obuf;
> > int ret;
> >
> > mutex_lock(&ljca_gpio->trans_lock);
> > packet->item[0].index = gpio_id;
> > packet->item[0].value = config | ljca_gpio->connect_mode[gpio_id];
> > packet->num = 1;
> >
> > ret = ljca_transfer(ljca_gpio->ljca, LJCA_GPIO_CONFIG, (u8 *)packet,
> >     struct_size(packet, item, packet->num), NULL, 0);
> > mutex_unlock(&ljca_gpio->trans_lock);
> >
> > return ret < 0 ? ret : 0;
> > }
> > ```
> > The definition of `struct ljca_gpio_packet` is at https://elixir.bootlin.com/linux/v6.10/source/drivers/gpio/gpio-ljca.c#L53.
> > ```
> > struct ljca_gpio_packet {
> > u8 num;
> > struct ljca_gpio_op item[] __counted_by(num);
> > } __packed;
> > ```
> >
> > Our question is: The `item` member of `struct ljca_gpio_packet` is annotated with "__counted_by". Only if we set `packet->num = 1` before accessing `packet->item[0]`, the flexible member `item` can be properly bounds-checked at run-time when enabling CONFIG_UBSAN_BOUNDS and CONFIG_FORTIFY_SOURCE. Or there will be a warning from each access prior to the initialization because the number of elements is zero.
> > So we think relocating `packet->num = 1` before accessing `packet->item[0]` is needed.
> >
> > Here is a fix example of a similar situation : https://lore.kernel.org/stable/20240613113225.898955993@xxxxxxxxxxxxxxxxxxx/.
> >
> > Please kindly correct us if we missed any key information. Looking forward to your response!
>
> This is a Gustavo AR Silvia question, so let's loop him in.
> (I think you're right, and we should make a patch.)

Yes! `packet->num = 1;` should be relocated:

diff --git a/drivers/gpio/gpio-ljca.c b/drivers/gpio/gpio-ljca.c
index dfec9fbfc7a9..c2a9b4253974 100644
--- a/drivers/gpio/gpio-ljca.c
+++ b/drivers/gpio/gpio-ljca.c
@@ -82,9 +82,9 @@ static int ljca_gpio_config(struct ljca_gpio_dev *ljca_gpio, u8 gpio_id,
        int ret;

        mutex_lock(&ljca_gpio->trans_lock);
+       packet->num = 1;
        packet->item[0].index = gpio_id;
        packet->item[0].value = config | ljca_gpio->connect_mode[gpio_id];
-       packet->num = 1;

        ret = ljca_transfer(ljca_gpio->ljca, LJCA_GPIO_CONFIG, (u8 *)packet,
                            struct_size(packet, item, packet->num), NULL, 0);

stable should be CC'd and the following tag included:

Fixes: 1034cc423f1b ("1034cc423f1b4a7a9a56d310ca980fcd2753e11d")

Thanks for catching this! :)
--
Gustavo