Re: [PATCH] kmemcheck: Test the full object in kmemcheck_is_obj_initialized()

[Catalin Marinas <catalin.marinas@arm.com>]

On Mon, 2010-02-08 at 14:38 +0000, roel kluin wrote:
> On Mon, Feb 8, 2010 at 12:16 PM, Catalin Marinas <catalin.marinas@arm.com> wrote:
> > This is a fix for bug #14845 (bugzilla.kernel.org). The
> > update_checksum() function in mm/kmemleak.c calls
> > kmemcheck_is_obj_initialised() before scanning an object. When
> > KMEMCHECK_PARTIAL_OK is enabled, this function returns true. However,
> > the crc32_le() reads smaller intervals (32-bit) for which
> > kmemleak_is_obj_initialised() may be false leading to a kmemcheck
> > warning.
> >
> > Note that kmemcheck_is_obj_initialized() is currently only used by
> > kmemleak before scanning a memory location.
> 
> >  enum kmemcheck_shadow kmemcheck_shadow_test(void *shadow, unsigned int size)
> >  {
> > +#ifdef CONFIG_KMEMCHECK_PARTIAL_OK
> >        uint8_t *x;
> >        unsigned int i;
> >
> >        x = shadow;
> >
> > -#ifdef CONFIG_KMEMCHECK_PARTIAL_OK
> >        /*
> >         * Make sure _some_ bytes are initialized. Gcc frequently generates
> >         * code to access neighboring bytes.
> > @@ -139,13 +139,25 @@ enum kmemcheck_shadow kmemcheck_shadow_test(void *shadow, unsigned int size)
> >                if (x[i] == KMEMCHECK_SHADOW_INITIALIZED)
> >                        return x[i];
> >        }
> > +
> > +       return x[0];
> >  #else
> > +       return kmemcheck_shadow_test_all(shadow, size);
> > +#endif
> > +}
> > +
> > +enum kmemcheck_shadow kmemcheck_shadow_test_all(void *shadow, unsigned int size)
> > +{
> > +       uint8_t *x;
> > +       unsigned int i;
> > +
> > +       x = shadow;
> > +
> >        /* All bytes must be initialized. */
> >        for (i = 0; i < size; ++i) {
> >                if (x[i] != KMEMCHECK_SHADOW_INITIALIZED)
> >                        return x[i];
> >        }
> > -#endif
> >
> >        return x[0];
> >  }
> 
> Are we certain that size cannot be 0 in kmemcheck_shadow_test()
> and kmemcheck_shadow_test_all() or other functions in
> arch/x86/mm/kmemcheck/shadow.c with these unsigned
> comparisons in loops?

At least in the kmemleak use-case, I don't think one can allocate a
zero-size object, so it should be safe. I can't tell about the other
cases but AFAICT, this function is called as a result of a memory access
which is always greater than 0.

Thanks.

-- 
Catalin