Imagine the following code flow:
1. `spdk_put_io_channel();`
2. `spdk_io_device_unregister();`
Since putting an io_channel is always deferred,
it is likely that spdk_io_device_unregister will
lock the io_channel mutex first. It will set
dev->unregistered flag and then quickly realize
there are still open channels for this device
(refcnt > 0) - so it'll return. All fine here.
However, if the deferred put_io_channel happens to
lock the mutex first from other thread, it will
decrement the refcnt and attempt to free the device.
Both the decrementation and freeing are done under
a mutex, but there is a slight window inbetween
where the mutex is re-locked. spdk_io_device_unregister
is already sleeping on a mutex_lock(), so it might
strike now. It'll see there are no more io_channels
for this device and free the device. Once
put_io_channels regains the lock again, it will attempt
freeing the device for a second time.
This patch removes the slight window mentioned above.
Related to #278
Change-Id: I6c0f6014353529028d658211135196d97f1d8547
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Reviewed-on: https://review.gerrithub.io/408193
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
a7aa9d5737