Re: [PATCH v8 0/2] gpiolib: don't allow user-space to crash the kernel with hot-unplugs

From: Bartosz Golaszewski
Date: Wed Dec 07 2022 - 03:37:41 EST


On Mon, Dec 5, 2022 at 1:39 PM Bartosz Golaszewski <brgl@xxxxxxxx> wrote:
>
> From: Bartosz Golaszewski <bartosz.golaszewski@xxxxxxxxxx>
>
> Linus Torvalds pointed out that using trylock here is wrong. This iteration
> drops it in favor of unconditional locks but keeps all the fixes that came
> later.
>
> I will also not send it for this release but make it part of the updates PR
> for v6.2 to give it some time in next.
>
> v7 -> v8:
> - don't use down_read_trylock(), just go straight for a full lock
>
> v6 -> v7:
> - fix a build issue with CDEV_V1 code disabled (giving credit to Nick Hainke)
> - protect the gdev->chip also in gpio_chrdev_open()
>
> v5 -> v6:
> - signal an error in poll callbacks instead of returning 0 which would make
> the user-space assume a timeout occurred (which could lead to user-space
> spinning a timeout loop forever)
>
> v4 -> v5:
> - try to acquire the semaphore for reading and bail out of syscall callbacks
> immediately in case of lock contention
>
> v3 -> v4:
> - use function typedefs to make code cleaner
> - add a blank line after down_write()
>
> v2 -> v3:
> - drop the helper variable in patch 1/2 as we won't be using it in 2/2
> - refactor patch 2/2 to use locking wrappers around the syscall callbacks
>
> v1 -> v2:
> - add missing gdev->chip checks in patch 1/2
> - add a second patch that protects the structures that can be accessed
> by user-space calls against concurrent removal
>
> Bartosz Golaszewski (2):
> gpiolib: cdev: fix NULL-pointer dereferences
> gpiolib: protect the GPIO device against being dropped while in use by
> user-space
>
> drivers/gpio/gpiolib-cdev.c | 204 +++++++++++++++++++++++++++++++-----
> drivers/gpio/gpiolib.c | 4 +
> drivers/gpio/gpiolib.h | 5 +
> 3 files changed, 188 insertions(+), 25 deletions(-)
>
> --
> 2.37.2
>

Reapplied again to get it back into next for testing.

Bart