On Wed, Nov 11, 2020 at 1:35 AM Johannes Berg <johannes@xxxxxxxxxxxxxxxx> wrote: > > On Tue, 2020-11-10 at 16:49 +0800, Claire Chang wrote: > > If a device is getting removed or reprobed during resume, use-after-free > > might happen. For example, h5_btrtl_resume()[drivers/bluetooth/hci_h5.c] > > schedules a work queue for device reprobing. During the reprobing, if > > rfkill_set_block() in rfkill_resume() is called after the corresponding > > *_unregister() and kfree() are called, there will be an use-after-free > > in hci_rfkill_set_block()[net/bluetooth/hci_core.c]. > > > Not sure I understand. So you're saying > > * something (h5_btrtl_resume) schedules a worker > * said worker run, when it runs, calls rfkill_unregister() > * somehow rfkill_resume() still gets called after this > > But that can't really be right, device_del() removes it from the PM > lists? If device_del() is called right before the device_lock() in device_resume()[1], it's possible the rfkill device is unregistered, but rfkill_resume is still called. We actually hit this during the suspend/resume stress test, although it's rare. I also have a patch with multiple msleep that can 100% reproduce this use-after-free. Happy to share here if needed. [1] https://elixir.bootlin.com/linux/v5.10-rc3/source/drivers/base/power/main.c#L919 Thanks, Claire > > > johannes > >
