Re: [rft, PATCH v5 1/1] gpiolib: fix memory leak in gpiochip_setup_dev()
From: Bartosz Golaszewski
Date: Fri Nov 25 2022 - 16:09:05 EST
On Fri, Nov 18, 2022 at 8:44 PM Andy Shevchenko
<andriy.shevchenko@xxxxxxxxxxxxxxx> wrote:
>
> From: Zeng Heng <zengheng4@xxxxxxxxxx>
>
> Here is a backtrace report about memory leak detected in
> gpiochip_setup_dev():
>
> unreferenced object 0xffff88810b406400 (size 512):
> comm "python3", pid 1682, jiffies 4295346908 (age 24.090s)
> backtrace:
> kmalloc_trace
> device_add device_private_init at drivers/base/core.c:3361
> (inlined by) device_add at drivers/base/core.c:3411
> cdev_device_add
> gpiolib_cdev_register
> gpiochip_setup_dev
> gpiochip_add_data_with_key
>
> gcdev_register() & gcdev_unregister() would call device_add() &
> device_del() (no matter CONFIG_GPIO_CDEV is enabled or not) to
> register/unregister device.
>
> However, if device_add() succeeds, some resource (like
> struct device_private allocated by device_private_init())
> is not released by device_del().
>
> Therefore, after device_add() succeeds by gcdev_register(), it
> needs to call put_device() to release resource in the error handle
> path.
>
> Here we move forward the register of release function, and let it
> release every piece of resource by put_device() instead of kfree().
>
> While at it, fix another subtle issue, i.e. when gc->ngpio is equal
> to 0, we still call kcalloc() and, in case of further error, kfree()
> on the ZERO_PTR pointer, which is not NULL. It's not a bug per se,
> but rather waste of the resources and potentially wrong expectation
> about contents of the gdev->descs variable.
>
> Fixes: 159f3cd92f17 ("gpiolib: Defer gpio device setup until after gpiolib initialization")
> Signed-off-by: Zeng Heng <zengheng4@xxxxxxxxxx>
> Co-developed-by: Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>
> Signed-off-by: Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>
> ---
> changes in v5 (done by Andy):
> - refactored to avoid double checks on the same conditionals
> - moved kcalloc() after validating its parameter
> - compiled tested only by me (Andy)
> changes in v4:
> - add gpiochip_print_register_fail()
> changes in v3:
> - use put_device() instead of kfree() explicitly
> changes in v2:
> - correct fixes tag
>
> drivers/gpio/gpiolib.c | 42 ++++++++++++++++++++++++++----------------
> 1 file changed, 26 insertions(+), 16 deletions(-)
>
> diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
> index caa747fdcb72..119c9c3a2a50 100644
> --- a/drivers/gpio/gpiolib.c
> +++ b/drivers/gpio/gpiolib.c
> @@ -580,12 +580,13 @@ static int gpiochip_setup_dev(struct gpio_device *gdev)
> if (ret)
> return ret;
>
> + /* From this point, the .release() function cleans up gpio_device */
> + gdev->dev.release = gpiodevice_release;
> +
> ret = gpiochip_sysfs_register(gdev);
> if (ret)
> goto err_remove_device;
>
> - /* From this point, the .release() function cleans up gpio_device */
> - gdev->dev.release = gpiodevice_release;
> dev_dbg(&gdev->dev, "registered GPIOs %d to %d on %s\n", gdev->base,
> gdev->base + gdev->ngpio - 1, gdev->chip->label ? : "generic");
>
> @@ -651,10 +652,10 @@ int gpiochip_add_data_with_key(struct gpio_chip *gc, void *data,
> struct fwnode_handle *fwnode = NULL;
> struct gpio_device *gdev;
> unsigned long flags;
> - int base = gc->base;
> unsigned int i;
> + u32 ngpios = 0;
> + int base = 0;
> int ret = 0;
> - u32 ngpios;
>
> /* If the calling driver did not initialize firmware node, do it here */
> if (gc->fwnode)
> @@ -696,17 +697,12 @@ int gpiochip_add_data_with_key(struct gpio_chip *gc, void *data,
> else
> gdev->owner = THIS_MODULE;
>
> - gdev->descs = kcalloc(gc->ngpio, sizeof(gdev->descs[0]), GFP_KERNEL);
> - if (!gdev->descs) {
> - ret = -ENOMEM;
> - goto err_free_dev_name;
> - }
> -
> /*
> * Try the device properties if the driver didn't supply the number
> * of GPIO lines.
> */
> - if (gc->ngpio == 0) {
> + ngpios = gc->ngpio;
> + if (ngpios == 0) {
> ret = device_property_read_u32(&gdev->dev, "ngpios", &ngpios);
> if (ret == -ENODATA)
> /*
> @@ -717,7 +713,7 @@ int gpiochip_add_data_with_key(struct gpio_chip *gc, void *data,
> */
> ngpios = 0;
> else if (ret)
> - goto err_free_descs;
> + goto err_free_dev_name;
>
> gc->ngpio = ngpios;
> }
> @@ -725,13 +721,19 @@ int gpiochip_add_data_with_key(struct gpio_chip *gc, void *data,
> if (gc->ngpio == 0) {
> chip_err(gc, "tried to insert a GPIO chip with zero lines\n");
> ret = -EINVAL;
> - goto err_free_descs;
> + goto err_free_dev_name;
> }
>
> if (gc->ngpio > FASTPATH_NGPIO)
> chip_warn(gc, "line cnt %u is greater than fast path cnt %u\n",
> gc->ngpio, FASTPATH_NGPIO);
>
> + gdev->descs = kcalloc(gc->ngpio, sizeof(*gdev->descs), GFP_KERNEL);
> + if (!gdev->descs) {
> + ret = -ENOMEM;
> + goto err_free_dev_name;
> + }
> +
> gdev->label = kstrdup_const(gc->label ?: "unknown", GFP_KERNEL);
> if (!gdev->label) {
> ret = -ENOMEM;
> @@ -750,11 +752,13 @@ int gpiochip_add_data_with_key(struct gpio_chip *gc, void *data,
> * it may be a pipe dream. It will not happen before we get rid
> * of the sysfs interface anyways.
> */
> + base = gc->base;
> if (base < 0) {
> base = gpiochip_find_base(gc->ngpio);
> if (base < 0) {
> - ret = base;
> spin_unlock_irqrestore(&gpio_lock, flags);
> + ret = base;
> + base = 0;
> goto err_free_label;
> }
> /*
> @@ -868,6 +872,11 @@ int gpiochip_add_data_with_key(struct gpio_chip *gc, void *data,
> err_free_gpiochip_mask:
> gpiochip_remove_pin_ranges(gc);
> gpiochip_free_valid_mask(gc);
> + if (gdev->dev.release) {
> + /* release() has been registered by gpiochip_setup_dev() */
> + put_device(&gdev->dev);
> + goto err_print_message;
> + }
> err_remove_from_list:
> spin_lock_irqsave(&gpio_lock, flags);
> list_del(&gdev->list);
> @@ -881,13 +890,14 @@ int gpiochip_add_data_with_key(struct gpio_chip *gc, void *data,
> err_free_ida:
> ida_free(&gpio_ida, gdev->id);
> err_free_gdev:
> + kfree(gdev);
> +err_print_message:
> /* failures here can mean systems won't boot... */
> if (ret != -EPROBE_DEFER) {
> pr_err("%s: GPIOs %d..%d (%s) failed to register, %d\n", __func__,
> - gdev->base, gdev->base + gdev->ngpio - 1,
> + base, base + (int)ngpios - 1,
> gc->label ? : "generic", ret);
> }
> - kfree(gdev);
> return ret;
> }
> EXPORT_SYMBOL_GPL(gpiochip_add_data_with_key);
> --
> 2.35.1
>
Queued for fixes, thanks!
Bartosz