container_of() macro has one major drawback. It does not check whether
the passed ptr has a const pointer, the result will always be a
non-const pointer. Use a _Generic() construct (supported since gcc 4.9
and Clang 3.0) to teach container_of that if converting a const pointer,
the returned pointer should also have the const modifier.

Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
---
 include/linux/container_of.h | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/include/linux/container_of.h b/include/linux/container_of.h
index 2f4944b791b8..269f64e27b09 100644
--- a/include/linux/container_of.h
+++ b/include/linux/container_of.h
@@ -19,7 +19,11 @@
 	static_assert(__same_type(*(ptr), ((type *)0)->member) ||	\
 		      __same_type(*(ptr), void),			\
 		      "pointer type mismatch in container_of()");	\
-	((type *)(__mptr - offsetof(type, member))); })
+	_Generic((ptr),							\
+		const typeof(((type *)0)->member)*:			\
+			((const type *)(__mptr - offsetof(type, member))), \
+		default: ((type *)(__mptr - offsetof(type, member)))	\
+	); })
 
 /**
  * container_of_safe - cast a member of a structure out to the containing structure
@@ -35,6 +39,10 @@
 		      __same_type(*(ptr), void),			\
 		      "pointer type mismatch in container_of_safe()");	\
 	IS_ERR_OR_NULL(__mptr) ? ERR_CAST(__mptr) :			\
-		((type *)(__mptr - offsetof(type, member))); })
+	_Generic((ptr),							\
+		const typeof(((type *)0)->member)*:			\
+			((const type *)(__mptr - offsetof(type, member))), \
+		default: ((type *)(__mptr - offsetof(type, member)))	\
+	); })
 
 #endif	/* _LINUX_CONTAINER_OF_H */
-- 
2.34.1

On Thu, 10 Feb 2022 20:04:34 +0300 Dmitry Baryshkov <dmitry.baryshkov@linaro.org> wrote:

> container_of() macro has one major drawback. It does not check whether
> the passed ptr has a const pointer, the result will always be a
> non-const pointer. Use a _Generic() construct (supported since gcc 4.9
> and Clang 3.0) to teach container_of that if converting a const pointer,
> the returned pointer should also have the const modifier.
> 

Nice idea, but my x86_64 allnoconfig build explodes with zillions of
warnings.

In file included from ./include/linux/list.h:5,
                 from ./include/linux/module.h:12,
                 from init/do_mounts.c:2:
./include/net/sock.h: In function 'sk_entry':
./include/linux/container_of.h:17:42: warning: return discards 'const' qualifier from pointer target type [-Wdiscarded-qualifiers]

On Sat, 12 Feb 2022 at 01:05, Andrew Morton <akpm@linux-foundation.org> wrote:
>
> On Thu, 10 Feb 2022 20:04:34 +0300 Dmitry Baryshkov <dmitry.baryshkov@linaro.org> wrote:
>
> > container_of() macro has one major drawback. It does not check whether
> > the passed ptr has a const pointer, the result will always be a
> > non-const pointer. Use a _Generic() construct (supported since gcc 4.9
> > and Clang 3.0) to teach container_of that if converting a const pointer,
> > the returned pointer should also have the const modifier.
> >
>
> Nice idea, but my x86_64 allnoconfig build explodes with zillions of
> warnings.
>
> In file included from ./include/linux/list.h:5,
>                  from ./include/linux/module.h:12,
>                  from init/do_mounts.c:2:
> ./include/net/sock.h: In function 'sk_entry':
> ./include/linux/container_of.h:17:42: warning: return discards 'const' qualifier from pointer target type [-Wdiscarded-qualifiers]

Are these warnings legit? In other words, is the code converting a
const pointer into a modifiable one? I've stumbled upon this in drm
drivers.
How do we proceed with this patch?


-- 
With best wishes
Dmitry

On Sat, 12 Feb 2022 01:10:09 +0300 Dmitry Baryshkov <dmitry.baryshkov@linaro.org> wrote:

> On Sat, 12 Feb 2022 at 01:05, Andrew Morton <akpm@linux-foundation.org> wrote:
> >
> > On Thu, 10 Feb 2022 20:04:34 +0300 Dmitry Baryshkov <dmitry.baryshkov@linaro.org> wrote:
> >
> > > container_of() macro has one major drawback. It does not check whether
> > > the passed ptr has a const pointer, the result will always be a
> > > non-const pointer. Use a _Generic() construct (supported since gcc 4.9
> > > and Clang 3.0) to teach container_of that if converting a const pointer,
> > > the returned pointer should also have the const modifier.
> > >
> >
> > Nice idea, but my x86_64 allnoconfig build explodes with zillions of
> > warnings.
> >
> > In file included from ./include/linux/list.h:5,
> >                  from ./include/linux/module.h:12,
> >                  from init/do_mounts.c:2:
> > ./include/net/sock.h: In function 'sk_entry':
> > ./include/linux/container_of.h:17:42: warning: return discards 'const' qualifier from pointer target type [-Wdiscarded-qualifiers]
> 
> Are these warnings legit? In other words, is the code converting a
> const pointer into a modifiable one?

It is indeed doing that.

static inline struct sock *sk_entry(const struct hlist_node *node)
{
	return hlist_entry(node, struct sock, sk_node);
}

Which is a pretty rude thing to do.  What was the point in that const arg?

> I've stumbled upon this in drm
> drivers.
> How do we proceed with this patch?

They all need to be fixed up :(

You could take a shot, if you have the time.  If not, bug the relevant
code's owners about it.  But we can't add a patch like this, blow up
everyone's builds and wait for various people to fix up the fallout!