[PATCH] arm/kasan:fix the arry size of kasan_early_shadow_pte

[PATCH] arm/kasan:fix the arry size of kasan_early_shadow_pte

[Hailong liu]

From: Hailong Liu <liu.hailong6@zte.com.cn>

The size of kasan_early_shadow_pte[] now is PTRS_PER_PTE which defined to
512 for arm architecture. This means that it only covers the prev Linux pte
entries, but not the HWTABLE pte entries for arm.

The reason it works well current is that the symbol kasan_early_shadow_page
immediately following kasan_early_shadow_pte in memory is page aligned,
which makes kasan_early_shadow_pte look like a 4KB size array. But we can't
ensure the order always right with different compiler/linker, nor more bss
symbols be introduced.

We had a test with QEMU + vexpress：put a 512KB-size symbol with attribute
__section(".bss..page_aligned") after kasan_early_shadow_pte, and poison it
after kasan_early_init(). Then enabled CONFIG_KASAN, it failed to boot up.

Signed-off-by: Hailong Liu <liu.hailong6@zte.com.cn>
Signed-off-by: Ziliang Guo <guo.ziliang@zte.com.cn>
---
 include/linux/kasan.h | 6 +++++-
 mm/kasan/init.c       | 3 ++-
 2 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/include/linux/kasan.h b/include/linux/kasan.h
index 5e0655fb2a6f..fe1ae73ff8b5 100644
--- a/include/linux/kasan.h
+++ b/include/linux/kasan.h
@@ -35,8 +35,12 @@ struct kunit_kasan_expectation {
 #define KASAN_SHADOW_INIT 0
 #endif
 
+#ifndef PTE_HWTABLE_PTRS
+#define PTE_HWTABLE_PTRS 0
+#endif
+
 extern unsigned char kasan_early_shadow_page[PAGE_SIZE];
-extern pte_t kasan_early_shadow_pte[PTRS_PER_PTE];
+extern pte_t kasan_early_shadow_pte[PTRS_PER_PTE + PTE_HWTABLE_PTRS];
 extern pmd_t kasan_early_shadow_pmd[PTRS_PER_PMD];
 extern pud_t kasan_early_shadow_pud[PTRS_PER_PUD];
 extern p4d_t kasan_early_shadow_p4d[MAX_PTRS_PER_P4D];
diff --git a/mm/kasan/init.c b/mm/kasan/init.c
index bc0ad208b3a7..7ca0b92d5886 100644
--- a/mm/kasan/init.c
+++ b/mm/kasan/init.c
@@ -64,7 +64,8 @@ static inline bool kasan_pmd_table(pud_t pud)
 	return false;
 }
 #endif
-pte_t kasan_early_shadow_pte[PTRS_PER_PTE] __page_aligned_bss;
+pte_t kasan_early_shadow_pte[PTRS_PER_PTE + PTE_HWTABLE_PTRS]
+	__page_aligned_bss;
 
 static inline bool kasan_pte_table(pmd_t pmd)
 {
-- 
2.17.1

Re: [PATCH] arm/kasan:fix the arry size of kasan_early_shadow_pte

[Ard Biesheuvel]

(+ Linus)

On Sat, 9 Jan 2021 at 05:50, Hailong liu <hailongliiu@yeah.net> wrote:
>
> From: Hailong Liu <liu.hailong6@zte.com.cn>
>
> The size of kasan_early_shadow_pte[] now is PTRS_PER_PTE which defined to
> 512 for arm architecture. This means that it only covers the prev Linux pte
> entries, but not the HWTABLE pte entries for arm.
>
> The reason it works well current is that the symbol kasan_early_shadow_page
> immediately following kasan_early_shadow_pte in memory is page aligned,
> which makes kasan_early_shadow_pte look like a 4KB size array. But we can't
> ensure the order always right with different compiler/linker, nor more bss
> symbols be introduced.
>
> We had a test with QEMU + vexpress：put a 512KB-size symbol with attribute
> __section(".bss..page_aligned") after kasan_early_shadow_pte, and poison it
> after kasan_early_init(). Then enabled CONFIG_KASAN, it failed to boot up.
>
> Signed-off-by: Hailong Liu <liu.hailong6@zte.com.cn>
> Signed-off-by: Ziliang Guo <guo.ziliang@zte.com.cn>
> ---
>  include/linux/kasan.h | 6 +++++-
>  mm/kasan/init.c       | 3 ++-
>  2 files changed, 7 insertions(+), 2 deletions(-)
>
> diff --git a/include/linux/kasan.h b/include/linux/kasan.h
> index 5e0655fb2a6f..fe1ae73ff8b5 100644
> --- a/include/linux/kasan.h
> +++ b/include/linux/kasan.h
> @@ -35,8 +35,12 @@ struct kunit_kasan_expectation {
>  #define KASAN_SHADOW_INIT 0
>  #endif
>
> +#ifndef PTE_HWTABLE_PTRS
> +#define PTE_HWTABLE_PTRS 0
> +#endif
> +
>  extern unsigned char kasan_early_shadow_page[PAGE_SIZE];
> -extern pte_t kasan_early_shadow_pte[PTRS_PER_PTE];
> +extern pte_t kasan_early_shadow_pte[PTRS_PER_PTE + PTE_HWTABLE_PTRS];
>  extern pmd_t kasan_early_shadow_pmd[PTRS_PER_PMD];
>  extern pud_t kasan_early_shadow_pud[PTRS_PER_PUD];
>  extern p4d_t kasan_early_shadow_p4d[MAX_PTRS_PER_P4D];
> diff --git a/mm/kasan/init.c b/mm/kasan/init.c
> index bc0ad208b3a7..7ca0b92d5886 100644
> --- a/mm/kasan/init.c
> +++ b/mm/kasan/init.c
> @@ -64,7 +64,8 @@ static inline bool kasan_pmd_table(pud_t pud)
>         return false;
>  }
>  #endif
> -pte_t kasan_early_shadow_pte[PTRS_PER_PTE] __page_aligned_bss;
> +pte_t kasan_early_shadow_pte[PTRS_PER_PTE + PTE_HWTABLE_PTRS]
> +       __page_aligned_bss;
>
>  static inline bool kasan_pte_table(pmd_t pmd)
>  {
> --
> 2.17.1
>
>
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Re: [PATCH] arm/kasan:fix the arry size of kasan_early_shadow_pte

[Linus Walleij]

On Sat, Jan 9, 2021 at 5:51 AM Hailong liu <hailongliiu@yeah.net> wrote:

> From: Hailong Liu <liu.hailong6@zte.com.cn>
>
> The size of kasan_early_shadow_pte[] now is PTRS_PER_PTE which defined to
> 512 for arm architecture. This means that it only covers the prev Linux pte
> entries, but not the HWTABLE pte entries for arm.
>
> The reason it works well current is that the symbol kasan_early_shadow_page
> immediately following kasan_early_shadow_pte in memory is page aligned,
> which makes kasan_early_shadow_pte look like a 4KB size array. But we can't
> ensure the order always right with different compiler/linker, nor more bss
> symbols be introduced.
>
> We had a test with QEMU + vexpress：put a 512KB-size symbol with attribute
> __section(".bss..page_aligned") after kasan_early_shadow_pte, and poison it
> after kasan_early_init(). Then enabled CONFIG_KASAN, it failed to boot up.
>
> Signed-off-by: Hailong Liu <liu.hailong6@zte.com.cn>
> Signed-off-by: Ziliang Guo <guo.ziliang@zte.com.cn>

OK I see the problem, I think.

> +#ifndef PTE_HWTABLE_PTRS
> +#define PTE_HWTABLE_PTRS 0
> +#endif

Can this even happen? We have either pgtable-2level.h or
pgtable-3level.h, both of which define PTE_HWTABLE_PTRS.

>  extern unsigned char kasan_early_shadow_page[PAGE_SIZE];
> -extern pte_t kasan_early_shadow_pte[PTRS_PER_PTE];
> +extern pte_t kasan_early_shadow_pte[PTRS_PER_PTE + PTE_HWTABLE_PTRS];

Yeah this looks exactly like bm_pte so it makes sense.

If you drop the first ifndef,
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>

Yours,
Linus Walleij

Re:Re: [PATCH] arm/kasan:fix the arry size of kasan_early_shadow_pte

[hailong]

[-- Attachment #1: Type: text/plain, Size: 1886 bytes --]


At 2021-01-10 05:26:08, "Linus Walleij" <linus.walleij@linaro.org> wrote:
>On Sat, Jan 9, 2021 at 5:51 AM Hailong liu <hailongliiu@yeah.net> wrote:
>
>> From: Hailong Liu <liu.hailong6@zte.com.cn>
>>
>> The size of kasan_early_shadow_pte[] now is PTRS_PER_PTE which defined to
>> 512 for arm architecture. This means that it only covers the prev Linux pte
>> entries, but not the HWTABLE pte entries for arm.
>>
>> The reason it works well current is that the symbol kasan_early_shadow_page
>> immediately following kasan_early_shadow_pte in memory is page aligned,
>> which makes kasan_early_shadow_pte look like a 4KB size array. But we can't
>> ensure the order always right with different compiler/linker, nor more bss
>> symbols be introduced.
>>
>> We had a test with QEMU + vexpress：put a 512KB-size symbol with attribute
>> __section(".bss..page_aligned") after kasan_early_shadow_pte, and poison it
>> after kasan_early_init(). Then enabled CONFIG_KASAN, it failed to boot up.
>>
>> Signed-off-by: Hailong Liu <liu.hailong6@zte.com.cn>
>> Signed-off-by: Ziliang Guo <guo.ziliang@zte.com.cn>
>
>OK I see the problem, I think.
>
>> +#ifndef PTE_HWTABLE_PTRS
>> +#define PTE_HWTABLE_PTRS 0
>> +#endif
>
>Can this even happen? We have either pgtable-2level.h or
>pgtable-3level.h, both of which define PTE_HWTABLE_PTRS.
>

I guess not for arm. But I'm not sure for other ARCHs.

>>  extern unsigned char kasan_early_shadow_page[PAGE_SIZE];
>> -extern pte_t kasan_early_shadow_pte[PTRS_PER_PTE];
>> +extern pte_t kasan_early_shadow_pte[PTRS_PER_PTE + PTE_HWTABLE_PTRS];
>
>Yeah this looks exactly like bm_pte so it makes sense.
>

Yes, it comes from prototype of pm_pte. And it does solve the
problems our project encountered and seems to work well for now.

>If you drop the first ifndef,
>Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
>
>Yours,
>Linus Walleij

Thanks.

[-- Attachment #2: Type: text/html, Size: 2278 bytes --]

Re: Re: [PATCH] arm/kasan:fix the arry size of kasan_early_shadow_pte

[Linus Walleij]

On Sun, Jan 10, 2021 at 11:21 AM hailong <hailongliiu@yeah.net> wrote:

> >> +#ifndef PTE_HWTABLE_PTRS
> >> +#define PTE_HWTABLE_PTRS 0
> >> +#endif
> >
> >Can this even happen? We have either pgtable-2level.h or
> >pgtable-3level.h, both of which define PTE_HWTABLE_PTRS.
> >
>
> I guess not for arm. But I'm not sure for other ARCHs.

Oh it's a generic include. Sorry for the confusion.

All good then!

Yours,
Linus Walleij

Re: [PATCH] arm/kasan:fix the arry size of kasan_early_shadow_pte

[Andrew Morton]

On Sun, 10 Jan 2021 13:03:49 +0100 Linus Walleij <linus.walleij@linaro.org> wrote:

> On Sun, Jan 10, 2021 at 11:21 AM hailong <hailongliiu@yeah.net> wrote:
> 
> > >> +#ifndef PTE_HWTABLE_PTRS
> > >> +#define PTE_HWTABLE_PTRS 0
> > >> +#endif
> > >
> > >Can this even happen? We have either pgtable-2level.h or
> > >pgtable-3level.h, both of which define PTE_HWTABLE_PTRS.
> > >
> >
> > I guess not for arm. But I'm not sure for other ARCHs.
> 
> Oh it's a generic include. Sorry for the confusion.
> 
> All good then!
> 

This code is 2+ years old.  Do we think it warrants a cc:stable?

Re: [PATCH] arm/kasan:fix the arry size of kasan_early_shadow_pte

[Ard Biesheuvel]

On Tue, 12 Jan 2021 at 07:19, Andrew Morton <akpm@linux-foundation.org> wrote:
>
> On Sun, 10 Jan 2021 13:03:49 +0100 Linus Walleij <linus.walleij@linaro.org> wrote:
>
> > On Sun, Jan 10, 2021 at 11:21 AM hailong <hailongliiu@yeah.net> wrote:
> >
> > > >> +#ifndef PTE_HWTABLE_PTRS
> > > >> +#define PTE_HWTABLE_PTRS 0
> > > >> +#endif
> > > >
> > > >Can this even happen? We have either pgtable-2level.h or
> > > >pgtable-3level.h, both of which define PTE_HWTABLE_PTRS.
> > > >
> > >
> > > I guess not for arm. But I'm not sure for other ARCHs.
> >
> > Oh it's a generic include. Sorry for the confusion.
> >
> > All good then!
> >
>
> This code is 2+ years old.  Do we think it warrants a cc:stable?
>

Not needed - ARM only gained Kasan support this cycle, and this patch
does not affect any other architectures