On Jun 14, 2016, at 6:52 PM, Jeff Layton wrote: > I think I'd still prefer to have it unlock the mutex in the event that > it's not going to use it after all. While that kind of thing is ok for > now, it's stuff like that that can turn into a subtle source of bugs > later. > > Also, I think I'd be more comfortable with this being split into (at > least) two patches. Do one patch as a straight conversion from rwsem to > mutex, and then another that changes the code to take the mutex before > hashing the new stateid.
Ok, I guess that could be arranged too. And then there's this Bruce's patch to pull more stuff into the init_open_stateid
