Re: [PATCH] livepatch: fix race between enabled_store() and klp_unregister_patch()

[Josh Poimboeuf <jpoimboe@redhat.com>]

On Mon, Nov 30, 2015 at 11:54:37AM +0800, Li Bin wrote:
> There is a potential race as following:
> 
> CPU0                         |  CPU1
> -----------------------------|-----------------------------------
> enabled_store()              |  klp_unregister_patch()
>                              |  |-mutex_lock(&klp_mutex);
> |-mutex_lock(&klp_mutex);    |  |-klp_free_patch();
>                              |  |-mutex_unlock(&klp_mutex);
> |-[process the patch's state]|
> |-mutex_unlock(&klp_mutex)   |
> 
> Fix this race condition by adding klp_is_patch_registered() check in
> enabled_store() after get the lock klp_mutex.

I'm thinking this race isn't possible, and that instead it would
deadlock.

When I try to recreate something similar by putting a delay in
enabled_store(), klp_free_patch() just sleeps on its call to
kobject_put() until enabled_store() returns.  The unregister stack looks
like:

  [<ffffffff812e966b>] __kernfs_remove+0x1fb/0x380
  [<ffffffff812ea273>] kernfs_remove+0x23/0x40
  [<ffffffff812ec601>] sysfs_remove_dir+0x51/0x80
  [<ffffffff81407fb8>] kobject_del+0x18/0x50
  [<ffffffff8140804a>] kobject_release+0x5a/0x190
  [<ffffffff81407f27>] kobject_put+0x27/0x50
  [<ffffffff81128d41>] klp_unregister_patch+0x71/0x80

It seems to be waiting on a lock which is held by the kernfs code which
called enabled_store().  So when enabled_store() tries to get the
klp_mutex, it deadlocks.

Miroslav and I previously discussed a few options to fix this:

 https://lkml.kernel.org/r/alpine.LNX.2.00.1502171728520.29490@pobox.suse.cz

-- 
Josh