lkml.org 
[lkml]   [2022]   [May]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v2][next] x86/mm/pgtable: Fix -Wstringop-overflow warnings
On Mon, May 09, 2022 at 01:54:32PM -0700, Kees Cook wrote:
> On Mon, May 09, 2022 at 03:50:56PM -0500, Gustavo A. R. Silva wrote:
> > On Mon, May 09, 2022 at 12:59:15PM -0700, Kees Cook wrote:
> > > On Mon, May 09, 2022 at 02:45:41PM -0500, Gustavo A. R. Silva wrote:
> > > > Fix the following -Wstringop-overflow warnings when building with GCC-12.1:
> > > >
> > > > arch/x86/mm/pgtable.c:437:13: warning: 'preallocate_pmds' accessing 8 bytes in a region of size 0 [-Wstringop-overflow=]
> > > > arch/x86/mm/pgtable.c:440:13: warning: 'preallocate_pmds' accessing 8 bytes in a region of size 0 [-Wstringop-overflow=]
> > > > arch/x86/mm/pgtable.c:462:9: warning: 'free_pmds' accessing 8 bytes in a region of size 0 [-Wstringop-overflow=]
> > > > arch/x86/mm/pgtable.c:454:9: warning: 'pgd_prepopulate_pmd' accessing 8 bytes in a region of size 0 [-Wstringop-overflow=]
> > > > arch/x86/mm/pgtable.c:455:9: warning: 'pgd_prepopulate_user_pmd' accessing 8 bytes in a region of size 0 [-Wstringop-overflow=]
> > > > arch/x86/mm/pgtable.c:464:9: warning: 'free_pmds' accessing 8 bytes in a region of size 0 [-Wstringop-overflow=]
> > > >
> > > > There is a case in which PREALLOCATED_PMDS, MAX_PREALLOCATED_PMDS,
> > > > PREALLOCATED_USER_PMDS and MAX_PREALLOCATED_USER_PMDS are defined as
> > > > zero:
> > > >
> > > > 204 #else /* !CONFIG_X86_PAE */
> > > > 205
> > > > 206 /* No need to prepopulate any pagetable entries in non-PAE modes. */
> > > > 207 #define PREALLOCATED_PMDS 0
> > > > 208 #define MAX_PREALLOCATED_PMDS 0
> > > > 209 #define PREALLOCATED_USER_PMDS 0
> > > > 210 #define MAX_PREALLOCATED_USER_PMDS 0
> > > > 211 #endif /* CONFIG_X86_PAE */
> > > >
> > > > It seems that GCC is legitimately complaining about the fact that, under
> > > > certain circumstances, u_pmds and pmds are declared as zero-length arrays
> > > > in the stack and, of course, they are not flexible arrays.
> > >
> > > Ah yeah, I've run into this a few times. Since the relationship between
> > > the macro pairs can't be seen by GCC, it gets upset (i.e. sizeof(u_pmds)
> > > has no relationship wtih PREALLOCATED_USER_PMDS and the calls weren't
> > > inlined, so it can't see that it'll always be 0 and 0).
> > >
> > > > 424 pgd_t *pgd_alloc(struct mm_struct *mm)
> > > > 425 {
> > > > 426 pgd_t *pgd;
> > > > 427 pmd_t *u_pmds[MAX_PREALLOCATED_USER_PMDS];
> > > > 428 pmd_t *pmds[MAX_PREALLOCATED_PMDS];
> > > > 429
> > > >
> > > > Notice that "Accessing elements of zero-length arrays declared in such
> > > > contexts is undefined and may be diagnosed."[1]
> > > >
> > > > We can fix this by checking that MAX_PREALLOCATED_PMDS and MAX_PREALLOCATED_USER_PMDS
> > > > are different than zero, prior to passing u_pmds amd pmds as arguments to any
> > > > function, in this case to functions preallocate_pmds(), pgd_prepopulate_pmd()
> > > > and free_pmds().
> > > >
> > > > This helps with the ongoing efforts to globally enable
> > > > -Wstringop-overflow.
> > > >
> > > > [1] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
> > > >
> > > > Link: https://github.com/KSPP/linux/issues/181
> > > > Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
> > > > ---
> > > > Changes in v2:
> > > > - Check MAX_PREALLOCATED_PMDS and MAX_PREALLOCATED_USER_PMDS
> > > > instead of using pointer notation.
> > > > Link: https://lore.kernel.org/linux-hardening/20220401005834.GA182932@embeddedor/
> > > > - Update changelog text.
> > > >
> > > > arch/x86/mm/pgtable.c | 16 ++++++++++------
> > > > 1 file changed, 10 insertions(+), 6 deletions(-)
> > > >
> > > > diff --git a/arch/x86/mm/pgtable.c b/arch/x86/mm/pgtable.c
> > > > index f16059e9a85e..96c3f402a1da 100644
> > > > --- a/arch/x86/mm/pgtable.c
> > > > +++ b/arch/x86/mm/pgtable.c
> > > > @@ -434,14 +434,18 @@ pgd_t *pgd_alloc(struct mm_struct *mm)
> > > >
> > > > mm->pgd = pgd;
> > > >
> > > > - if (preallocate_pmds(mm, pmds, PREALLOCATED_PMDS) != 0)
> > > > - goto out_free_pgd;
> > > > + if (MAX_PREALLOCATED_PMDS != 0 && MAX_PREALLOCATED_USER_PMDS != 0) {
> > > > + if (preallocate_pmds(mm, pmds, PREALLOCATED_PMDS) != 0)
> > > > + goto out_free_pgd;
> > > >
> > > > - if (preallocate_pmds(mm, u_pmds, PREALLOCATED_USER_PMDS) != 0)
> > > > - goto out_free_pmds;
> > > > + if (preallocate_pmds(mm, u_pmds, PREALLOCATED_USER_PMDS) != 0)
> > > > + goto out_free_pmds;
> > > >
> > > > - if (paravirt_pgd_alloc(mm) != 0)
> > > > - goto out_free_user_pmds;
> > > > + if (paravirt_pgd_alloc(mm) != 0)
> > > > + goto out_free_user_pmds;
> > > > + } else {
> > > > + goto out_free_pgd;
> > >
> > > The "all 0" case shouldn't be a failure mode; it should just skip the
> > > preallocate_pmds() calls.
> >
> > Do you mean something like this:
> >
> > diff --git a/arch/x86/mm/pgtable.c b/arch/x86/mm/pgtable.c
> > index f16059e9a85e..4dae168408f1 100644
> > --- a/arch/x86/mm/pgtable.c
> > +++ b/arch/x86/mm/pgtable.c
> > @@ -434,11 +434,13 @@ pgd_t *pgd_alloc(struct mm_struct *mm)
> >
> > mm->pgd = pgd;
> >
> > - if (preallocate_pmds(mm, pmds, PREALLOCATED_PMDS) != 0)
> > - goto out_free_pgd;
> > + if (MAX_PREALLOCATED_PMDS != 0 && MAX_PREALLOCATED_USER_PMDS != 0) {
> > + if (preallocate_pmds(mm, pmds, PREALLOCATED_PMDS) != 0)
> > + goto out_free_pgd;
> >
> > - if (preallocate_pmds(mm, u_pmds, PREALLOCATED_USER_PMDS) != 0)
> > - goto out_free_pmds;
> > + if (preallocate_pmds(mm, u_pmds, PREALLOCATED_USER_PMDS) != 0)
> > + goto out_free_pmds;
> > + }
> >
> > if (paravirt_pgd_alloc(mm) != 0)
> > goto out_free_user_pmds;
> >
> > It seems that the above is not enough, because we have the same issue
> > when calling pgd_prepopulate_pmd(), pgd_prepopulate_user_pmd() and
> > free_pmds():
> >
> > CC arch/x86/mm/pgtable.o
> > arch/x86/mm/pgtable.c: In function 'pgd_alloc':
> > arch/x86/mm/pgtable.c:464:9: warning: 'free_pmds' accessing 8 bytes in a region of size 0 [-Wstringop-overflow=]
> > 464 | free_pmds(mm, u_pmds, PREALLOCATED_USER_PMDS);
> > | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>
> Ugh. Perhaps just marking both preallocate_pmds() and free_pmds() as
> inline is enough to let the compiler "see" everything correctly?

It doesn't seem to work... however, the following piece of code implies
that pmds and u_pmds should be first preallocated through preallocate_pmds(),
which cannot happen if (MAX_PREALLOCATED_PMDS != 0 && MAX_PREALLOCATED_USER_PMDS != 0)

448 /*
449 * Make sure that pre-populating the pmds is atomic with
450 * respect to anything walking the pgd_list, so that they
451 * never see a partially populated pgd.
452 */
453 spin_lock(&pgd_lock);
454
455 pgd_ctor(mm, pgd);
456 pgd_prepopulate_pmd(mm, pgd, pmds);
457 pgd_prepopulate_user_pmd(mm, pgd, u_pmds);
458
459 spin_unlock(&pgd_lock);
460
461 return pgd;

So, my question here is why do you think the "all 0" case should only skip the
preallocate_pmds() calls and not the pgd_prepopulate_pmd() calls too?

>
> Otherwise, they'll likely each need the same check that was added to
> pgd_prepopulate_pmd() ages ago for a similar situation...

uhm... that doesn't seem to have an impact nowadays, or at least now
Wstringop-overflow sees the problem first, because now the issue is
detected at the moment of passing the arguments to the the function
and not when actually executing the function?

otherwise, I think we wouldn't see this error:

arch/x86/mm/pgtable.c:454:9: warning: 'pgd_prepopulate_pmd' accessing 8 bytes in a region of size 0 [-Wstringop-overflow=]
454 | pgd_prepopulate_pmd(mm, pgd, pmds);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
arch/x86/mm/pgtable.c:454:9: note: referencing argument 3 of type 'pmd_t *[0]'
arch/x86/mm/pgtable.c:296:13: note: in a call to function 'pgd_prepopulate_pmd'
296 | static void pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd, pmd_t *pmds[])
| ^~~~~~~~~~~~~~~~~~~


Thanks
--
Gustavo

\
 
 \ /
  Last update: 2022-05-10 16:43    [W:0.162 / U:0.080 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site