On Fri, Oct 07, 2022 at 08:07:58AM -0600, Jason A. Donenfeld wrote:
> On Fri, Oct 07, 2022 at 04:57:24AM +0000, Christophe Leroy wrote:
> >
> >
> > Le 07/10/2022 à 01:36, Jason A. Donenfeld a écrit :
> > > On 10/6/22, Christophe Leroy <christophe.leroy@xxxxxxxxxx> wrote:
> > >>
> > >>
> > >> Le 06/10/2022 à 19:31, Christophe Leroy a écrit :
> > >>>
> > >>>
> > >>> Le 06/10/2022 à 19:24, Jason A. Donenfeld a écrit :
> > >>>> Hi Christophe,
> > >>>>
> > >>>> On Thu, Oct 6, 2022 at 11:21 AM Christophe Leroy
> > >>>> <christophe.leroy@xxxxxxxxxx> wrote:
> > >>>>> Le 06/10/2022 à 18:53, Jason A. Donenfeld a écrit :
> > >>>>>> The prandom_u32() function has been a deprecated inline wrapper around
> > >>>>>> get_random_u32() for several releases now, and compiles down to the
> > >>>>>> exact same code. Replace the deprecated wrapper with a direct call to
> > >>>>>> the real function. The same also applies to get_random_int(), which is
> > >>>>>> just a wrapper around get_random_u32().
> > >>>>>>
> > >>>>>> Reviewed-by: Kees Cook <keescook@xxxxxxxxxxxx>
> > >>>>>> Acked-by: Toke Høiland-Jørgensen <toke@xxxxxxx> # for sch_cake
> > >>>>>> Acked-by: Chuck Lever <chuck.lever@xxxxxxxxxx> # for nfsd
> > >>>>>> Reviewed-by: Jan Kara <jack@xxxxxxx> # for ext4
> > >>>>>> Signed-off-by: Jason A. Donenfeld <Jason@xxxxxxxxx>
> > >>>>>> ---
> > >>>>>
> > >>>>>> diff --git a/arch/powerpc/kernel/process.c
> > >>>>>> b/arch/powerpc/kernel/process.c
> > >>>>>> index 0fbda89cd1bb..9c4c15afbbe8 100644
> > >>>>>> --- a/arch/powerpc/kernel/process.c
> > >>>>>> +++ b/arch/powerpc/kernel/process.c
> > >>>>>> @@ -2308,6 +2308,6 @@ void notrace __ppc64_runlatch_off(void)
> > >>>>>> unsigned long arch_align_stack(unsigned long sp)
> > >>>>>> {
> > >>>>>> if (!(current->personality & ADDR_NO_RANDOMIZE) &&
> > >>>>>> randomize_va_space)
> > >>>>>> - sp -= get_random_int() & ~PAGE_MASK;
> > >>>>>> + sp -= get_random_u32() & ~PAGE_MASK;
> > >>>>>> return sp & ~0xf;
> > >>>>>
> > >>>>> Isn't that a candidate for prandom_u32_max() ?
> > >>>>>
> > >>>>> Note that sp is deemed to be 16 bytes aligned at all time.
> > >>>>
> > >>>> Yes, probably. It seemed non-trivial to think about, so I didn't. But
> > >>>> let's see here... maybe it's not too bad:
> > >>>>
> > >>>> If PAGE_MASK is always ~(PAGE_SIZE-1), then ~PAGE_MASK is
> > >>>> (PAGE_SIZE-1), so prandom_u32_max(PAGE_SIZE) should yield the same
> > >>>> thing? Is that accurate? And holds across platforms (this comes up a
> > >>>> few places)? If so, I'll do that for a v4.
> > >>>>
> > >>>
> > >>> On powerpc it is always (from arch/powerpc/include/asm/page.h) :
> > >>>
> > >>> /*
> > >>> * Subtle: (1 << PAGE_SHIFT) is an int, not an unsigned long. So if we
> > >>> * assign PAGE_MASK to a larger type it gets extended the way we want
> > >>> * (i.e. with 1s in the high bits)
> > >>> */
> > >>> #define PAGE_MASK (~((1 << PAGE_SHIFT) - 1))
> > >>>
> > >>> #define PAGE_SIZE (1UL << PAGE_SHIFT)
> > >>>
> > >>>
> > >>> So it would work I guess.
> > >>
> > >> But taking into account that sp must remain 16 bytes aligned, would it
> > >> be better to do something like ?
> > >>
> > >> sp -= prandom_u32_max(PAGE_SIZE >> 4) << 4;
> > >>
> > >> return sp;
> > >
> > > Does this assume that sp is already aligned at the beginning of the
> > > function? I'd assume from the function's name that this isn't the
> > > case?
> >
> > Ah you are right, I overlooked it.
>
> So I think to stay on the safe side, I'm going to go with
> `prandom_u32_max(PAGE_SIZE)`. Sound good?
Given these kinds of less mechanical changes, it may make sense to split
these from the "trivial" conversions in a treewide patch. The chance of
needing a revert from the simple 1:1 conversions is much lower than the
need to revert by-hand changes.
The Cocci script I suggested in my v1 review gets 80% of the first
patch, for example.
--
Kees Cook
