Re: security/landlock/ruleset.c:96:9: warning: 'memcpy' accessing 4294967295 bytes at offsets 20 and 0 overlaps 6442450943 bytes at offset -2147483648

[Mickaël Salaün]

Cc csky developers.

See https://lore.kernel.org/r/202501040747.S3LYfvYq-lkp@xxxxxxxxx

On Mon, Jan 06, 2025 at 04:27:26PM -0800, Kees Cook wrote:
> On Mon, Jan 06, 2025 at 06:04:56PM +0100, Mickaël Salaün wrote:
> > On Sat, Jan 04, 2025 at 07:26:27AM +0800, kernel test robot wrote:
> > >    security/landlock/ruleset.c: In function 'create_rule':
> > > >> security/landlock/ruleset.c:96:9: warning: 'memcpy' accessing 4294967295 bytes at offsets 20 and 0 overlaps 6442450943 bytes at offset -2147483648 [-Wrestrict]
> > >       96 |         memcpy(new_rule->layers, layers,
> > >          |         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > >       97 |                         flex_array_size(new_rule, layers, num_layers));
> > >          |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

This warning seems to only show with the csky architecture.

> >
> > I guess the GCC warning is a false positive?
> > See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116494
> 
> Does the warning go away if flex_array_size() is open-coded or if the
> create_rule prototype uses "*" instead of "[]" syntax for the "layers"
> argument?

But that would not be correct right?

> 
> The warning is about the "read" part ("accessing"). Using tip-of-tree
> gcc with -fdiagnostics-details might show more details on the calling
> path.
> 
> -Kees
> 
> -- 
> Kees Cook
>