mm/page_alloc: silence a KASAN false positive
Subsystems (3)
PATCH 1/1 mm/page_alloc: silence a KASAN false positive
kernel_init_free_pages() will use memset() on s390 to clear all pages from kmalloc_order() which will override KASAN redzones because a redzone was setup from the end of the allocation size to the end of the last page. Silence it by not reporting it there. An example of the report is,
BUG: KASAN: slab-out-of-bounds in __free_pages_ok
Write of size 4096 at addr 000000014beaa000
Call Trace:
show_stack+0x152/0x210
dump_stack+0x1f8/0x248
print_address_description.isra.13+0x5e/0x4d0
kasan_report+0x130/0x178
check_memory_region+0x190/0x218
memset+0x34/0x60
__free_pages_ok+0x894/0x12f0
kfree+0x4f2/0x5e0
unpack_to_rootfs+0x60e/0x650
populate_rootfs+0x56/0x358
do_one_initcall+0x1f4/0xa20
kernel_init_freeable+0x758/0x7e8
kernel_init+0x1c/0x170
ret_from_fork+0x24/0x28
Memory state around the buggy address:
000000014bea9f00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
000000014bea9f80: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
>000000014beaa000: 03 fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe
^
000000014beaa080: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe
000000014beaa100: fe fe fe fe fe fe fe fe fe fe fe fe fe fe--- mm/page_alloc.c | 3 +++ 1 file changed, 3 insertions(+)
mm/page_alloc.c +3 -0
@@ -1164,8 +1164,11 @@ static void kernel_init_free_pages(struct page *page, int numpages) {int i; +/* s390's use of memset() could override KASAN redzones. */ +kasan_disable_current ();for (i =0 ; i < numpages; i++)clear_highpage (page + i); +kasan_enable_current (); } static __always_inline bool free_pages_prepare(struct page *page,
Signature · 1 line
2.21.0 (Apple Git-122.2)
Quoted 6 lines · Qian Cai
kernel_init_free_pages()will usememset()on s390 to clear all pages fromkmalloc_order()which will override KASAN redzones because a redzone was setup from the end of the allocation size to the end of the last page. Silence it by not reporting it there. An example of the report is,
Interesting. The reason why we did not hit it on x86_64 is because clear_page is implemented in asm (arch/x86/lib/clear_page_64.S) and thus is not instrumented. Arm64 probably does the same. However, on s390 clear_page is defined to memset.
clear_[high]page are pretty extensively used in the kernel. We can either do this, or make clear_page non instrumented on s390 as well to match the existing implicit assumption. The benefit of the current approach is that we can find some real use-after-free's and maybe out-of-bounds on clear_page. The downside is that we may need more of these annotations. Thoughts?
Quoted 50 lines
BUG: KASAN: slab-out-of-bounds in __free_pages_ok Write of size 4096 at addr 000000014beaa000 Call Trace: show_stack+0x152/0x210 dump_stack+0x1f8/0x248 print_address_description.isra.13+0x5e/0x4d0 kasan_report+0x130/0x178 check_memory_region+0x190/0x218 memset+0x34/0x60 __free_pages_ok+0x894/0x12f0 kfree+0x4f2/0x5e0 unpack_to_rootfs+0x60e/0x650 populate_rootfs+0x56/0x358 do_one_initcall+0x1f4/0xa20 kernel_init_freeable+0x758/0x7e8 kernel_init+0x1c/0x170 ret_from_fork+0x24/0x28 Memory state around the buggy address: 000000014bea9f00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 000000014bea9f80: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 >000000014beaa000: 03 fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe ^ 000000014beaa080: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe 000000014beaa100: fe fe fe fe fe fe fe fe fe fe fe fe fe feFixes 6471384af2a6 ("mm: security: introduce init_on_alloc=1 and init_on_free=1 boot options")Signed-off-by Qian Cai <[email protected]>--- mm/page_alloc.c | 3 +++ 1 file changed, 3 insertions(+)mm/page_alloc.c +3
-1@@ -1164,8 +1164,11 @@ static void kernel_init_free_pages(struct page *page, int numpages) {int i; +/* s390's use of memset() could override KASAN redzones. */ +kasan_disable_current ();for (i =0 ; i < numpages; i++)clear_highpage (page + i); +kasan_enable_current (); } static __always_inline bool free_pages_prepare(struct page *page, --2.21.0 (Apple Git-122.2)
Quoted 9 lines · Dmitry Vyukov
Quoted 6 lines · Qian Cai
kernel_init_free_pages()will usememset()on s390 to clear all pages fromkmalloc_order()which will override KASAN redzones because a redzone was setup from the end of the allocation size to the end of the last page. Silence it by not reporting it there. An example of the report is,
Quoted 4 lines
Interesting. The reason why we did not hit it on
x86_64is becauseclear_pageis implemented in asm (arch/x86/lib/clear_page_64.S) and thus is not instrumented. Arm64 probably does the same. However, on s390clear_pageis defined to memset.
Can we define it to __memset() instead?__memset() is supposed to be ignored by KASAN, e.g. KASAN runtime uses it in the places where we don't care about bugs.
Quoted 10 lines · Alexander Potapenko
Quoted 9 lines · Dmitry Vyukov
Quoted 6 lines · Qian Cai
kernel_init_free_pages()will usememset()on s390 to clear all pages fromkmalloc_order()which will override KASAN redzones because a redzone was setup from the end of the allocation size to the end of the last page. Silence it by not reporting it there. An example of the report is,
Quoted 8 lines
Quoted 4 lines
Interesting. The reason why we did not hit it on
x86_64is becauseclear_pageis implemented in asm (arch/x86/lib/clear_page_64.S) and thus is not instrumented. Arm64 probably does the same. However, on s390clear_pageis defined to memset.Can we define it to
__memset()instead?__memset()is supposed to be ignored by KASAN, e.g. KASAN runtime uses it in the places where we don't care about bugs.
I suppose that could work if s390 maintains perfer this way.
Quoted 8 lines · Dmitry Vyukov
Quoted 6 lines · Qian Cai
kernel_init_free_pages()will usememset()on s390 to clear all pages fromkmalloc_order()which will override KASAN redzones because a redzone was setup from the end of the allocation size to the end of the last page. Silence it by not reporting it there. An example of the report is,
Quoted 10 lines
Interesting. The reason why we did not hit it on
x86_64is becauseclear_pageis implemented in asm (arch/x86/lib/clear_page_64.S) and thus is not instrumented. Arm64 probably does the same. However, on s390clear_pageis defined to memset.
clear_[high]page are pretty extensively used in the kernel. We can either do this, or makeclear_pagenon instrumented on s390 as well to match the existing implicit assumption. The benefit of the current approach is that we can find some real use-after-free's and maybe out-of-bounds onclear_page. The downside is that we may need more of these annotations. Thoughts?
Since we had already done the same thing in poison_page(), I suppose we could do the same here. Also, clear_page() has been used in many places on s390, and it is not clear to me if those are all safe like this.
There might be more annotations required, so it probably up to s390 maintainers (CC'ed) if they prefer not instrumenting clear_page() like other arches.
Quoted 51 lines
Quoted 51 lines
BUG: KASAN: slab-out-of-bounds in __free_pages_ok Write of size 4096 at addr 000000014beaa000 Call Trace: show_stack+0x152/0x210 dump_stack+0x1f8/0x248 print_address_description.isra.13+0x5e/0x4d0 kasan_report+0x130/0x178 check_memory_region+0x190/0x218 memset+0x34/0x60 __free_pages_ok+0x894/0x12f0 kfree+0x4f2/0x5e0 unpack_to_rootfs+0x60e/0x650 populate_rootfs+0x56/0x358 do_one_initcall+0x1f4/0xa20 kernel_init_freeable+0x758/0x7e8 kernel_init+0x1c/0x170 ret_from_fork+0x24/0x28 Memory state around the buggy address: 000000014bea9f00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 000000014bea9f80: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 >000000014beaa000: 03 fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe ^ 000000014beaa080: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe 000000014beaa100: fe fe fe fe fe fe fe fe fe fe fe fe fe feFixes 6471384af2a6 ("mm: security: introduce init_on_alloc=1 and init_on_free=1 boot options")Signed-off-by Qian Cai <[email protected]>--- mm/page_alloc.c | 3 +++ 1 file changed, 3 insertions(+)mm/page_alloc.c +3
-1@@ -1164,8 +1164,11 @@ static void kernel_init_free_pages(struct page *page, int numpages) {int i; +/* s390's use of memset() could override KASAN redzones. */ +kasan_disable_current ();for (i =0 ; i < numpages; i++)clear_highpage (page + i); +kasan_enable_current (); } static __always_inline bool free_pages_prepare(struct page *page, --2.21.0 (Apple Git-122.2)
Quoted 20 lines · Qian Cai
Quoted 8 lines · Dmitry Vyukov
Quoted 6 lines · Qian Cai
kernel_init_free_pages()will usememset()on s390 to clear all pages fromkmalloc_order()which will override KASAN redzones because a redzone was setup from the end of the allocation size to the end of the last page. Silence it by not reporting it there. An example of the report is,Quoted 10 lines
Interesting. The reason why we did not hit it on
x86_64is becauseclear_pageis implemented in asm (arch/x86/lib/clear_page_64.S) and thus is not instrumented. Arm64 probably does the same. However, on s390clear_pageis defined to memset.
clear_[high]page are pretty extensively used in the kernel. We can either do this, or makeclear_pagenon instrumented on s390 as well to match the existing implicit assumption. The benefit of the current approach is that we can find some real use-after-free's and maybe out-of-bounds onclear_page. The downside is that we may need more of these annotations. Thoughts?
Quoted 7 lines
Since we had already done the same thing in
poison_page(), I suppose we could do the same here. Also,clear_page()has been used in many places on s390, and it is not clear to me if those are all safe like this.There might be more annotations required, so it probably up to s390 maintainers (CC'ed) if they prefer not instrumenting
clear_page()like other arches.
Vasily will look into this and come up with a proper solution for s390.
Quoted 20 lines · Qian Cai
Quoted 8 lines · Dmitry Vyukov
Quoted 6 lines · Qian Cai
kernel_init_free_pages()will usememset()on s390 to clear all pages fromkmalloc_order()which will override KASAN redzones because a redzone was setup from the end of the allocation size to the end of the last page. Silence it by not reporting it there. An example of the report is,Quoted 10 lines
Interesting. The reason why we did not hit it on
x86_64is becauseclear_pageis implemented in asm (arch/x86/lib/clear_page_64.S) and thus is not instrumented. Arm64 probably does the same. However, on s390clear_pageis defined to memset.
clear_[high]page are pretty extensively used in the kernel. We can either do this, or makeclear_pagenon instrumented on s390 as well to match the existing implicit assumption. The benefit of the current approach is that we can find some real use-after-free's and maybe out-of-bounds onclear_page. The downside is that we may need more of these annotations. Thoughts?
Quoted 8 lines
Since we had already done the same thing in
poison_page(), I suppose we could do the same here. Also,clear_page()has been used in many places on s390, and it is not clear to me if those are all safe like this.There might be more annotations required, so it probably up to s390 maintainers (CC'ed) if they prefer not instrumenting
clear_page()like other arches.
Sorry for delay. I assume you tested it without CONFIG_JUMP_LABEL. I had to fix couple of things before I was able to use init_on_alloc=1 and init_on_free=1 boot options on s390 to reproduce KASAN problem:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?h=v5.8-rc3&id=998f5bbe3dbdab81c1cfb1aef7c3892f5d24f6c7
https://git.kernel.org/pub/scm/linux/kernel/git/s390/linux.git/commit/?h=fixes&id=95e61b1b5d6394b53d147c0fcbe2ae70fbe09446
https://git.kernel.org/pub/scm/linux/kernel/git/s390/linux.git/commit/?h=fixes&id=d6df52e9996dcc2062c3d9c9123288468bb95b52
Back to clear_page - we could certainly make it non-instrumented. But it didn't cause any problems so far. And as Dmitry pointed out we could potentially find additional bugs with it. So, I'm leaning towards original solution proposed. For that you have my
Thank you for looking into this!
Andrew, would you pick this change up?
Thank you
Vasily