Re: [PATCH] rust: sync: add accessor for the lock behind a given guard

[Boqun Feng]

On Thu, Jan 30, 2025 at 04:43:22PM +0100, Alice Ryhl wrote:
> On Thu, Jan 30, 2025 at 4:33 PM Boqun Feng <boqun.feng@xxxxxxxxx> wrote:
> >
> > Hi Alice,
> >
> > The patch looks reasonable to me, however...
> >
> > On Thu, Jan 30, 2025 at 11:39:32AM +0000, Alice Ryhl wrote:
> > > Binder has some methods where the caller provides a lock guard, and
> > > Binder needs to be able to assert that the guard is associated with the
> > > right lock. To enable this, add an accessor to obtain a reference to the
> > > underlying lock that you can pass to `ptr::eq`.
> > >
> >
> > ... could you provide more details on this usage, for example, why do
> > you need the assertion, is it for debug purposes? Does the current C
> > code have the same or similar logic? Besides...
> 
> I added the assertion because it makes the SAFETY comment on a call to
> kernel::list::List::remove simpler. The C driver does not have the
> assertion.
> 

Ok. Make sense.

> > > Signed-off-by: Alice Ryhl <aliceryhl@xxxxxxxxxx>
> > > ---
> > >  rust/kernel/sync/lock.rs | 7 ++++++-
> > >  1 file changed, 6 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/rust/kernel/sync/lock.rs b/rust/kernel/sync/lock.rs
> > > index 41dcddac69e2..681d67275b49 100644
> > > --- a/rust/kernel/sync/lock.rs
> > > +++ b/rust/kernel/sync/lock.rs
> > > @@ -169,7 +169,12 @@ pub struct Guard<'a, T: ?Sized, B: Backend> {
> > >  // SAFETY: `Guard` is sync when the data protected by the lock is also sync.
> > >  unsafe impl<T: Sync + ?Sized, B: Backend> Sync for Guard<'_, T, B> {}
> > >
> > > -impl<T: ?Sized, B: Backend> Guard<'_, T, B> {
> > > +impl<'a, T: ?Sized, B: Backend> Guard<'a, T, B> {
> > > +    /// Returns the lock that this guard originates from.
> > > +    pub fn lock(&self) -> &'a Lock<T, B> {
> >
> > How about we name this as `lock_ref()` or something else, because
> > `lock()` itself is already used by `Lock` for the lock *operation*, and
> > this is just an accessor, I would like we don't confuse code readers
> > when they see code like "let b = a.lock()".
> 
> The usual name for this operation in userspace is "mutex".
> https://docs.rs/lock_api/0.4.12/lock_api/struct.MutexGuard.html#method.mutex
> 
> But since our code is generic over many lock types, I went for "lock".
> But I guess it could make sense to rename it.
> 

Got it. The good thing about the naming of lock_api is that the name
"mutex" is not used for other purpose, while "lock" is a bit different.

> > Moreover, if the only usage
> > of this is for asserting the right lock, maybe we can instead add an:
> >
> >     pub fn assert_lock_associated(&self, lock_ptr: *const Lock<T, B>)
> 
> I guess, though there is precedent for having a method that gets the
> underlying lock, so I think it makes sense. If we had an assertion, it

I don't disagree, but I just feel we should be careful about introducing
two "lock()" that one is an operation and the other is an accessor.

> would probably take an &Lock<T,B>.
> 

How about:

    impl<T, B: Backend> Lock<T, B> {
        pub fn assert_held_then<O>(
            &self, proof: &Guard<'_, T, B>, f: FnOnce() -> O
        ) -> O {
            <assert `proof` is associated with `&self`>
	    f()
	}
    }

In this way, not only we can assert the correct lock is held, but we can
also guarantee `f()` is called with the lock held. Thoughts?

Regards,
Boqun

> Alice