Re: [PATCH v2] gpiolib: fix memory leak in gpiochip_setup_dev

[Andy Shevchenko <andriy.shevchenko@intel.com>]

On Wed, Nov 09, 2022 at 05:31:20PM +0800, Zeng Heng wrote:
> gcdev_register & gcdev_unregister call device_add & device_del to
> request/release source. But in device_add, the dev->p allocated by
> device_private_init is not released by device_del.

First of all, we refer to the functions like func().

> So when calling gcdev_unregister to release gdev, it needs put_device
> to release dev in the following.
> 
> Otherwise, kmemleak would report memory leak such as below:
> 
> unreferenced object 0xffff88810b406400 (size 512):
>   comm "python3", pid 1682, jiffies 4295346908 (age 24.090s)
>   hex dump (first 32 bytes):
>     00 00 00 00 ad 4e ad de ff ff ff ff 00 00 00 00  .....N..........
>     ff ff ff ff ff ff ff ff a0 5e 23 90 ff ff ff ff  .........^#.....
>   backtrace:

Second, read Submitting Patches on how to provide your backtraces in the
message body.

>     [<00000000a58ee5fe>] kmalloc_trace+0x22/0x110
>     [<0000000045fe2058>] device_add+0xb34/0x1130
>     [<00000000d778b45f>] cdev_device_add+0x83/0xe0
>     [<0000000089f948ed>] gpiolib_cdev_register+0x73/0xa0
>     [<00000000a3a8a316>] gpiochip_setup_dev+0x1c/0x70
>     [<00000000787227b4>] gpiochip_add_data_with_key+0x10f6/0x1bf0
>     [<000000009ac5742c>] devm_gpiochip_add_data_with_key+0x2e/0x80
>     [<00000000bf2b23d9>] xra1403_probe+0x192/0x1b0 [gpio_xra1403]
>     [<000000005b5ef2d4>] spi_probe+0xe1/0x140
>     [<000000002b26f6f1>] really_probe+0x17c/0x3f0
>     [<00000000dd2dad9c>] __driver_probe_device+0xe3/0x170
>     [<000000005ca60d2a>] device_driver_attach+0x34/0x80
>     [<00000000e9db90db>] bind_store+0x10b/0x1a0
>     [<00000000e2650f8a>] drv_attr_store+0x49/0x70
>     [<0000000080a80b2b>] sysfs_kf_write+0x8c/0xb0
>     [<00000000a28b45b9>] kernfs_fop_write_iter+0x216/0x2e0
> 
> unreferenced object 0xffff888100de9800 (size 512):
>   comm "python3", pid 264, jiffies 4294737615 (age 33.514s)
>   hex dump (first 32 bytes):
>     00 00 00 00 ad 4e ad de ff ff ff ff 00 00 00 00  .....N..........
>     ff ff ff ff ff ff ff ff a0 5e 63 a1 ff ff ff ff  .........^c.....
>   backtrace:
>     [<00000000bcc571d0>] kmalloc_trace+0x22/0x110
>     [<00000000eeb06124>] device_add+0xb34/0x1130
>     [<000000007e5cd2fd>] cdev_device_add+0x83/0xe0
>     [<000000008f6bcd3a>] gpiolib_cdev_register+0x73/0xa0
>     [<0000000012c93b24>] gpiochip_setup_dev+0x1c/0x70
>     [<00000000a24b646a>] gpiochip_add_data_with_key+0x10f6/0x1bf0
>     [<000000000c225212>] tpic2810_probe+0x16e/0x196 [gpio_tpic2810]
>     [<00000000b52d04ff>] i2c_device_probe+0x651/0x680
>     [<0000000058d3ff6b>] really_probe+0x17c/0x3f0
>     [<00000000586f43d3>] __driver_probe_device+0xe3/0x170
>     [<000000003f428602>] device_driver_attach+0x34/0x80
>     [<0000000040e91a1b>] bind_store+0x10b/0x1a0
>     [<00000000c1d990b9>] drv_attr_store+0x49/0x70
>     [<00000000a23bfc22>] sysfs_kf_write+0x8c/0xb0
>     [<00000000064e6572>] kernfs_fop_write_iter+0x216/0x2e0
>     [<00000000026ce093>] vfs_write+0x658/0x810
> 
> Because at the point of gpiochip_setup_dev here, where dev.release
> does not set yet, calling put_device would cause the warning of
> no release function and double-free in the following fault handler
> route (when kfree dev_name). So directly calling kfree to release
> dev->p here in case of memory leak.

...

> @@ -539,6 +539,7 @@ static int gpiochip_setup_dev(struct gpio_device *gdev)
>  
>  err_remove_device:
>  	gcdev_unregister(gdev);
> +	kfree(gdev->dev.p);
>  	return ret;

Third, I do not believe it's a correct fix.
Have you read comments around device_del() / etc.?

NAK.

-- 
With Best Regards,
Andy Shevchenko