zram: fix possible races between sysfs use and bdev access

Its possible to run into a race where a sysfs knob is being used,
we get preempted, and a zram device is removed before we complete
use of the sysfs knob. This can happen for instance on block
devices, where for instance the zram block devices just part of
the private data of the block device. To ensure the private
data pointer is valid we must bdget() / bdput() in between
access.

For instance this can happen in the following two situations
as examples to illustrate this better:

        CPU 1                            CPU 2
destroy_devices
...
                                 compact_store()
                                 zram = dev_to_zram(dev);
idr_for_each(zram_remove_cb
  zram_remove
  ...
  kfree(zram)
                                 down_read(&zram->init_lock);

        CPU 1                            CPU 2
hot_remove_store
                                 compact_store()
                                 zram = dev_to_zram(dev);
  zram_remove
    kfree(zram)
                                 down_read(&zram->init_lock);

This issue does not fix a known crash, however this race was
spotted by Minchan Kim through code inspection upon code review
of the sysfs deadlock patch.

A generic solution for sysfs has been suggested and implemented (refer
to the dev_type_get() / dev_type_put() part of the patch) [0], however
there is no interest at this point in time in fixing this in sysfs itself.

[0] https://lkml.kernel.org/r/20210401235925.GR4332@42.do-not-panic.com

Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>