[PATCH trivial] kernel.h: Make container_of_safe type safe

[PATCH trivial] kernel.h: Make container_of_safe type safe

[Jason Gunthorpe]

The final ternary expression is given void * and type * as arguments, so
the resulting type of the macro ends up being void *.

container_of returns a type which maches the argument, this provides type
safety and allows the expression to be immediately used as the target
type.

Explicitly cast ERR_CAST() to get the correct type.

Fixes: 05e6557b8ed8 ("staging: lustre: add container_of_safe()")
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
---
 include/linux/kernel.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Jiri, can you take this? Thanks

Jason

diff --git a/include/linux/kernel.h b/include/linux/kernel.h
index 8f0e68e250a760..7b85aa607eb876 100644
--- a/include/linux/kernel.h
+++ b/include/linux/kernel.h
@@ -1013,7 +1013,7 @@ static inline void ftrace_dump(enum ftrace_dump_mode oops_dump_mode) { }
 	BUILD_BUG_ON_MSG(!__same_type(*(ptr), ((type *)0)->member) &&	\
 			 !__same_type(*(ptr), void),			\
 			 "pointer type mismatch in container_of()");	\
-	IS_ERR_OR_NULL(__mptr) ? ERR_CAST(__mptr) :			\
+	IS_ERR_OR_NULL(__mptr) ? ((type *)ERR_CAST(__mptr)) :		\
 		((type *)(__mptr - offsetof(type, member))); })
 
 /* Rebuild everything on CONFIG_FTRACE_MCOUNT_RECORD */
-- 
2.20.1

Re: [PATCH trivial] kernel.h: Make container_of_safe type safe

[Christoph Hellwig]

On Thu, Jan 24, 2019 at 08:55:05PM +0000, Jason Gunthorpe wrote:
> The final ternary expression is given void * and type * as arguments, so
> the resulting type of the macro ends up being void *.
> 
> container_of returns a type which maches the argument, this provides type
> safety and allows the expression to be immediately used as the target
> type.
> 
> Explicitly cast ERR_CAST() to get the correct type.

I think we should just kill off container_of_safe - it has no user
in the tree, is grossly misnamed and not really all that useful to start
with.

Re: [PATCH trivial] kernel.h: Make container_of_safe type safe

[Jason Gunthorpe]

On Mon, Jan 28, 2019 at 05:42:42AM -0800, Christoph Hellwig wrote:
> On Thu, Jan 24, 2019 at 08:55:05PM +0000, Jason Gunthorpe wrote:
> > The final ternary expression is given void * and type * as arguments, so
> > the resulting type of the macro ends up being void *.
> > 
> > container_of returns a type which maches the argument, this provides type
> > safety and allows the expression to be immediately used as the target
> > type.
> > 
> > Explicitly cast ERR_CAST() to get the correct type.
> 
> I think we should just kill off container_of_safe - it has no user
> in the tree, is grossly misnamed and not really all that useful to start
> with.

I see it open coded from time to time, here someone sent a patch
open-coding it yesterday (with the same bug even):

https://patchwork.kernel.org/patch/10783535/

I noticed it didn't work right because I had a use for it..

Jason

Re: [PATCH trivial] kernel.h: Make container_of_safe type safe

[Christoph Hellwig]

On Mon, Jan 28, 2019 at 06:44:24PM +0000, Jason Gunthorpe wrote:
> On Mon, Jan 28, 2019 at 05:42:42AM -0800, Christoph Hellwig wrote:
> > On Thu, Jan 24, 2019 at 08:55:05PM +0000, Jason Gunthorpe wrote:
> > > The final ternary expression is given void * and type * as arguments, so
> > > the resulting type of the macro ends up being void *.
> > > 
> > > container_of returns a type which maches the argument, this provides type
> > > safety and allows the expression to be immediately used as the target
> > > type.
> > > 
> > > Explicitly cast ERR_CAST() to get the correct type.
> > 
> > I think we should just kill off container_of_safe - it has no user
> > in the tree, is grossly misnamed and not really all that useful to start
> > with.
> 
> I see it open coded from time to time, here someone sent a patch
> open-coding it yesterday (with the same bug even):
> 
> https://patchwork.kernel.org/patch/10783535/
> 
> I noticed it didn't work right because I had a use for it..

And the patch is a good example for an anti-pattern to avoid, I'll
chime in there..