Dear Logan,


Am 15.03.23 um 07:18 schrieb Yu Kuai:

From: Yu Kuai <redacted>

Changes in v2:
  - fix a compile error for for md-cluster in patch 2
  - replace spin_lock/unlock with spin_lock/unlock_irq in patch 5
  - don't wake up inside the new lock in md wakeup_thread in patch 5

Our test reports a uaf for 'mddev->sync_thread':

T1                      T2
md_start_sync
  md_register_thread
			raid1d
			 md_check_recovery
			  md_reap_sync_thread
			   md_unregister_thread
			    kfree

  md_wakeup_thread
   wake_up
   ->sync_thread was freed

Currently, a global spinlock 'pers_lock' is borrowed to protect
'mddev->thread', this problem can be fixed likewise, however, there might
be similar problem for other md_thread, and I really don't like the idea to
borrow a global lock.

This patchset do some refactor, and then use a disk level spinlock to
protect md_thread in relevant apis.

I tested this pathset with mdadm tests, and there are no new regression,
by the way, following test will failed with or without this patchset:

01raid6integ
04r1update
05r6tor0
10ddf-create
10ddf-fail-spare
10ddf-fail-stop-readd
10ddf-geometry

As you improved the tests in the past, can you confirm, these failed on 
your test systems too and are fixed now?

[…]


Kind regards,

Paul