The new SAFETY comment style is taken from existing comments in `deref`
and `drop.
Signed-off-by: Tamir Duberstein <tamird@gmail.com>
---
rust/kernel/sync/arc.rs | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/rust/kernel/sync/arc.rs b/rust/kernel/sync/arc.rs
index af383bcd003e1122ebe1b62a49fe40279458e379..9adea755a5ad1a7b03f7fc30a7abc76c1f966c6c 100644
--- a/rust/kernel/sync/arc.rs
+++ b/rust/kernel/sync/arc.rs
@@ -377,10 +377,14 @@ fn as_ref(&self) -> &T {
impl<T: ?Sized> Clone for Arc<T> {
fn clone(&self) -> Self {
+ // SAFETY: By the type invariant, there is necessarily a reference to the object, so it is
+ // safe to dereference it.
+ let refcount = unsafe { self.ptr.as_ref() }.refcount.get();
+
// INVARIANT: C `refcount_inc` saturates the refcount, so it cannot overflow to zero.
// SAFETY: By the type invariant, there is necessarily a reference to the object, so it is
// safe to increment the refcount.
- unsafe { bindings::refcount_inc(self.ptr.as_ref().refcount.get()) };
+ unsafe { bindings::refcount_inc(refcount) };
// SAFETY: We just incremented the refcount. This increment is now owned by the new `Arc`.
unsafe { Self::from_inner(self.ptr) }
--
2.47.0
On Mon, Nov 4, 2024 at 10:20 PM Tamir Duberstein <tamird@gmail.com> wrote:
>
> The new SAFETY comment style is taken from existing comments in `deref`
> and `drop.
>
> Signed-off-by: Tamir Duberstein <tamird@gmail.com>
Reviewed-by: Alice Ryhl <aliceryhl@google.com>
> diff --git a/rust/kernel/sync/arc.rs b/rust/kernel/sync/arc.rs
> index af383bcd003e1122ebe1b62a49fe40279458e379..9adea755a5ad1a7b03f7fc30a7abc76c1f966c6c 100644
> --- a/rust/kernel/sync/arc.rs
> +++ b/rust/kernel/sync/arc.rs
> @@ -377,10 +377,14 @@ fn as_ref(&self) -> &T {
>
> impl<T: ?Sized> Clone for Arc<T> {
> fn clone(&self) -> Self {
> + // SAFETY: By the type invariant, there is necessarily a reference to the object, so it is
> + // safe to dereference it.
> + let refcount = unsafe { self.ptr.as_ref() }.refcount.get();
I would normally prefer to avoid creating a reference to the entire
ArcInner, but in this particular case it is okay due to the specifics
of how Arc works.
Alice
On Fri, Nov 8, 2024 at 6:38 AM Alice Ryhl <aliceryhl@google.com> wrote:
> > diff --git a/rust/kernel/sync/arc.rs b/rust/kernel/sync/arc.rs
> > index af383bcd003e1122ebe1b62a49fe40279458e379..9adea755a5ad1a7b03f7fc30a7abc76c1f966c6c 100644
> > --- a/rust/kernel/sync/arc.rs
> > +++ b/rust/kernel/sync/arc.rs
> > @@ -377,10 +377,14 @@ fn as_ref(&self) -> &T {
> >
> > impl<T: ?Sized> Clone for Arc<T> {
> > fn clone(&self) -> Self {
> > + // SAFETY: By the type invariant, there is necessarily a reference to the object, so it is
> > + // safe to dereference it.
> > + let refcount = unsafe { self.ptr.as_ref() }.refcount.get();
>
> I would normally prefer to avoid creating a reference to the entire
> ArcInner, but in this particular case it is okay due to the specifics
> of how Arc works.
Note that this particular line appears also in the Drop impl just
below. That said, can you help me understand the concern with creating
a reference to the entire ArcInner?
On Fri, Nov 8, 2024 at 1:30 PM Tamir Duberstein <tamird@gmail.com> wrote:
>
> On Fri, Nov 8, 2024 at 6:38 AM Alice Ryhl <aliceryhl@google.com> wrote:
> > > diff --git a/rust/kernel/sync/arc.rs b/rust/kernel/sync/arc.rs
> > > index af383bcd003e1122ebe1b62a49fe40279458e379..9adea755a5ad1a7b03f7fc30a7abc76c1f966c6c 100644
> > > --- a/rust/kernel/sync/arc.rs
> > > +++ b/rust/kernel/sync/arc.rs
> > > @@ -377,10 +377,14 @@ fn as_ref(&self) -> &T {
> > >
> > > impl<T: ?Sized> Clone for Arc<T> {
> > > fn clone(&self) -> Self {
> > > + // SAFETY: By the type invariant, there is necessarily a reference to the object, so it is
> > > + // safe to dereference it.
> > > + let refcount = unsafe { self.ptr.as_ref() }.refcount.get();
> >
> > I would normally prefer to avoid creating a reference to the entire
> > ArcInner, but in this particular case it is okay due to the specifics
> > of how Arc works.
>
> Note that this particular line appears also in the Drop impl just
> below. That said, can you help me understand the concern with creating
> a reference to the entire ArcInner?
Creating a shared reference to the entire ArcInner is an assertion
that no &mut exists to any part of the ArcInner, even though you only
access the refcount field. In this particular case, asserting shared
access to the `data` field is not a problem due to how Arc works, so
it's okay, but the pattern is problematic in other cases.
You could write `unsafe { (*self.ptr.as_ptr()).refcount.get() }` to
avoid the as_ref call.
Alice
On Fri, Nov 8, 2024 at 7:37 AM Alice Ryhl <aliceryhl@google.com> wrote:
> You could write `unsafe { (*self.ptr.as_ptr()).refcount.get() }` to
> avoid the as_ref call.
Doesn't `(*self.ptr.as_ptr())` have the same problem?
On Fri, Nov 8, 2024 at 1:41 PM Tamir Duberstein <tamird@gmail.com> wrote:
>
> On Fri, Nov 8, 2024 at 7:37 AM Alice Ryhl <aliceryhl@google.com> wrote:
> > You could write `unsafe { (*self.ptr.as_ptr()).refcount.get() }` to
> > avoid the as_ref call.
>
> Doesn't `(*self.ptr.as_ptr())` have the same problem?
It's not quite the same. Both `*ptr` and `(*ptr).field` are place
expressions which do not inherently involve the creation of a
reference in the way that `NonNull::as_ref` does.
Alice
On Fri, Nov 8, 2024 at 7:47 AM Alice Ryhl <aliceryhl@google.com> wrote:
>
> On Fri, Nov 8, 2024 at 1:41 PM Tamir Duberstein <tamird@gmail.com> wrote:
> >
> > On Fri, Nov 8, 2024 at 7:37 AM Alice Ryhl <aliceryhl@google.com> wrote:
> > > You could write `unsafe { (*self.ptr.as_ptr()).refcount.get() }` to
> > > avoid the as_ref call.
> >
> > Doesn't `(*self.ptr.as_ptr())` have the same problem?
>
> It's not quite the same. Both `*ptr` and `(*ptr).field` are place
> expressions which do not inherently involve the creation of a
> reference in the way that `NonNull::as_ref` does.
Thanks for explaining!
Tamir
© 2016 - 2026 Red Hat, Inc.