Re: [PATCH] iio: accel: fxls8962af: Fix use after free in fxls8962af_fifo_flush

[David Lechner]

On 6/2/25 5:50 AM, Sean Nyekjaer wrote:
> Hi Jonathan,
> 
> On Sat, May 31, 2025 at 05:53:02PM +0100, Jonathan Cameron wrote:
>> On Sat, 24 May 2025 12:34:09 +0200
>> Sean Nyekjaer <sean@xxxxxxxxxx> wrote:
>>
>>> fxls8962af_fifo_flush() uses indio_dev->active_scan_mask (with
>>> iio_for_each_active_channel()) without making sure the indio_dev
>>> stays in buffer mode.
>>> There is a race if indio_dev exits buffer mode in the middle of the
>>> interrupt that flushes the fifo. Fix this by calling
>>> iio_device_claim_buffer_mode() to ensure indio_dev can't exit buffer
>>> mode during the flush.
>>>
>>> Unable to handle kernel NULL pointer dereference at virtual address 00000000 when read
>>> [...]
>>> _find_first_bit_le from fxls8962af_fifo_flush+0x17c/0x290
>>> fxls8962af_fifo_flush from fxls8962af_interrupt+0x80/0x178
>>> fxls8962af_interrupt from irq_thread_fn+0x1c/0x7c
>>> irq_thread_fn from irq_thread+0x110/0x1f4
>>> irq_thread from kthread+0xe0/0xfc
>>> kthread from ret_from_fork+0x14/0x2c
>>>
>>> Fixes: 79e3a5bdd9ef ("iio: accel: fxls8962af: add hw buffered sampling")
>>> Cc: stable@xxxxxxxxxxxxxxx
>>> Signed-off-by: Sean Nyekjaer <sean@xxxxxxxxxx>
>>
>> That's nasty and a case I'd never thought about.  Most of the
>> races around disabling end up with an extra sample or two which then gets
>> dropped because there are no buffers enabled.
>>
>> We need to consider the active scan mask as part of the buffer state.
>> So effectively taking mlock if we enter this code will delay the state
>> transition (and change of active_scan_mask until after this interrupt is done).
>>
>> If David's synchronize_irq() is enough maybe that's a lighter weight path?
> 
> I agree if David's proposal is sufficient, I can try it.
> It's something we have seen once in some unrelated testing, so it's
> quite hard to reproduce :/
> 
> /Sean

Maybe temporarily adding a time delay of one sample period towards the
beginning of fxls8962af_interrupt() could make it easier to reproduce?