On Fri, 2012-11-16 at 17:34 +0000, Al Viro wrote:

On Fri, Nov 16, 2012 at 08:43:28AM -0800, Linus Torvalds wrote:

quoted

On Fri, Nov 16, 2012 at 8:36 AM, Ian Kent [off-list ref] wrote:

quoted

Sure, are you recommending I alter the fs/libfs.c functions to add a
function that doesn't have the outer lock, and have simple_empty() call
that, then use it in autofs?

Yup. That's the standard pattern, although usually we *strive* to make
the unlocked versions be static to the internal code, and then use
them there for the various helpers. In your case that seems
impossible, since you do depend on holding the d_lock in the caller
after the tests. But at least we don't have to duplicate the code and
have it in two unrelated places.

Al? Comments?

The thing is, I'm not convinced we really need ->d_lock held downstream.
E.g.  __autofs4_add_expiring() ought to be OK with just sbi->lookup_lock.
Not sure about the situation in autofs4_d_automount() - the thing is messy
as hell ;-/

Ian, do we really need that __simple_empty() variant in either caller?  What
is getting protected by ->d_lock after it and do we really need ->d_lock
continuously held for that?

Yeah, I've thought about that a few times now but haven't gone so far as
to change it.

I'll have another look.

Ian