Re: linux-next: Tree for May 16 (security/landlock/ruleset.c)

[Kees Cook]

On Mon, May 19, 2025 at 08:41:17PM +0200, Mickaël Salaün wrote:
> On Mon, May 19, 2025 at 11:19:53AM -0700, Kees Cook wrote:
> > On Mon, May 19, 2025 at 05:29:30PM +0200, Mickaël Salaün wrote:
> > > On Fri, May 16, 2025 at 07:54:14PM -0700, Randy Dunlap wrote:
> > > > 
> > > > 
> > > > On 5/16/25 3:24 AM, Stephen Rothwell wrote:
> > > > > Hi all,
> > > > > 
> > > > > Changes since 20250515:
> > > 
> > > Thanks for the report.
> > > 
> > > It is the same warning as reported here:
> > > https://lore.kernel.org/all/202501040747.S3LYfvYq-lkp@xxxxxxxxx/
> > > 
> > > I don't know what the actual issue is though.
> > > 
> > > > 
> > > > on i386:
> > > > 
> > > > In file included from ../arch/x86/include/asm/string.h:3,
> > > >                  from ../include/linux/string.h:65,
> > > >                  from ../include/linux/bitmap.h:13,
> > > >                  from ../include/linux/cpumask.h:12,
> > > >                  from ../include/linux/smp.h:13,
> > > >                  from ../include/linux/lockdep.h:14,
> > > >                  from ../security/landlock/ruleset.c:16:
> > > > ../security/landlock/ruleset.c: In function 'create_rule':
> > > > ../arch/x86/include/asm/string_32.h:150:25: warning: '__builtin_memcpy' accessing 4294967295 bytes at offsets 20 and 0 overlaps 6442450943 bytes at offset -2147483648 [-Wrestrict]
> > > >   150 | #define memcpy(t, f, n) __builtin_memcpy(t, f, n)
> > > >       |                         ^~~~~~~~~~~~~~~~~~~~~~~~~
> > > > ../security/landlock/ruleset.c:137:9: note: in expansion of macro 'memcpy'
> > > >   137 |         memcpy(new_rule->layers, layers,
> > > >       |         ^~~~~~
> > > > 
> > > > 
> > > > Full randconfig file is attached.
> > 
> > The trigger appears to be CONFIG_PROFILE_ALL_BRANCHES, and GCC getting
> > tricked into thinking check_mul_overflow() returns true:
> > 
> > In file included from ../arch/x86/include/asm/string.h:3,
> >                  from ../include/linux/string.h:65,
> >                  from ../include/linux/bitmap.h:13,
> >                  from ../include/linux/cpumask.h:12,
> >                  from ../include/linux/smp.h:13,
> >                  from ../include/linux/lockdep.h:14,
> >                  from ../security/landlock/ruleset.c:16:
> > ../security/landlock/ruleset.c: In function 'create_rule':
> > ../arch/x86/include/asm/string_32.h:150:25: warning: '__builtin_memcpy' accessing 4294967295 bytes at offsets 0 and 0 overlaps 6442450943 bytes at offset -2147483648 [-Wrestrict]
> >   150 | #define memcpy(t, f, n) __builtin_memcpy(t, f, n)
> >       |                         ^~~~~~~~~~~~~~~~~~~~~~~~~
> > ../security/landlock/ruleset.c:137:9: note: in expansion of macro 'memcpy'
> >   137 |         memcpy(new_rule->layers, layers,
> >       |         ^~~~~~
> >   'create_rule': event 1
> > ../include/linux/compiler.h:69:46:
> >    68 |         (cond) ?                                        \
> >       |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> >    69 |                 (__if_trace.miss_hit[1]++,1) :          \
> >       |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~
> >       |                                              |
> >       |                                              (1) when the condition is evaluated to true
> >    70 |                 (__if_trace.miss_hit[0]++,0);           \
> >       |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
> > ../include/linux/compiler.h:57:69: note: in expansion of macro '__trace_if_value'
> >    57 | #define __trace_if_var(cond) (__builtin_constant_p(cond) ? (cond) : __trace_if_value(cond))
> >       |                                                                     ^~~~~~~~~~~~~~~~
> > ../include/linux/compiler.h:55:28: note: in expansion of macro '__trace_if_var'
> >    55 | #define if(cond, ...) if ( __trace_if_var( !!(cond , ## __VA_ARGS__) ) )
> >       |                            ^~~~~~~~~~~~~~
> > ../include/linux/overflow.h:270:9: note: in expansion of macro 'if'
> >   270 |         if (check_mul_overflow(factor1, factor2, &bytes))
> >       |         ^~
> >   'create_rule': event 2
> > ../arch/x86/include/asm/string_32.h:150:25:
> >   150 | #define memcpy(t, f, n) __builtin_memcpy(t, f, n)
> >       |                         ^~~~~~~~~~~~~~~~~~~~~~~~~
> >       |                         |
> >       |                         (2) out of array bounds here
> > ../security/landlock/ruleset.c:137:9: note: in expansion of macro 'memcpy'
> >   137 |         memcpy(new_rule->layers, layers,
> >       |         ^~~~~~
> > make[1]: Leaving directory '/srv/code/gcc-bug'
> 
> That's interesting...
> 
> > 
> > 
> > I'll take a look at ways to make either the overflow macros or memcpy
> > robust against this kind of weirdness...
> 
> Thanks!

I'm doing some build testing, but the below patch makes GCC happy.
Alternatively we could make CONFIG_PROFILE_ALL_BRANCHES=y depend on
CONFIG_FORTIFY_SOURCE=y ...