Re: [PATCH] HID: multitouch: prevent memleak with the allocated name

Hi Andy,

On Wed, May 29, 2013 at 10:12 PM, Andy Shevchenko
[off-list ref] wrote:

On Wed, May 29, 2013 at 11:45 AM, Benjamin Tissoires
[off-list ref] wrote:

quoted

mt_free_input_name() was never called during .remove():
hid_hw_stop() removes the hid_input items in hdev->inputs, and so the
list is therefore empty after the call. In the end, we never free the
special names that has been allocated during .probe().

Restore the original name before freeing it to avoid acessing already
freed pointer.

quoted

I just spotted this one yesterday... My guess is that this way is safe (without
a locking mechanism to prevent accessing hi->input->name), but I'm not 100% sure.

Hi Jiri, Benjamin.

What do you think about patch I just sent?

Thanks for looking at this. My very first concern is that none of the
HID part has been devm-ized (this is on my todo-if-I-have-some-time
list). The input susbsystem has been devm-ized quite recently, so this
should be possible now.
So, basically, I honestly don't know if using part of devm mallocs in
hid-multitouch is safe and if it will work as expected. Maybe Jiri
will have a better idea.

I have a few comments on your patch if Jiri wants to include it.

P.S. Benjamin, I re-used your commit message, I think you have no objections.

No objections :)

Cheers,
Benjamin