2021-02-17 21:27:40

by Andrey Konovalov

[permalink] [raw]
Subject: [PATCH] mm, kasan: don't poison boot memory

During boot, all non-reserved memblock memory is exposed to the buddy
allocator. Poisoning all that memory with KASAN lengthens boot time,
especially on systems with large amount of RAM. This patch makes
page_alloc to not call kasan_free_pages() on all new memory.

__free_pages_core() is used when exposing fresh memory during system
boot and when onlining memory during hotplug. This patch adds a new
FPI_SKIP_KASAN_POISON flag and passes it to __free_pages_ok() through
free_pages_prepare() from __free_pages_core().

This has little impact on KASAN memory tracking.

Assuming that there are no references to newly exposed pages before they
are ever allocated, there won't be any intended (but buggy) accesses to
that memory that KASAN would normally detect.

However, with this patch, KASAN stops detecting wild and large
out-of-bounds accesses that happen to land on a fresh memory page that
was never allocated. This is taken as an acceptable trade-off.

All memory allocated normally when the boot is over keeps getting
poisoned as usual.

Signed-off-by: Andrey Konovalov <[email protected]>
Change-Id: Iae6b1e4bb8216955ffc14af255a7eaaa6f35324d
---
mm/page_alloc.c | 43 ++++++++++++++++++++++++++++++++-----------
1 file changed, 32 insertions(+), 11 deletions(-)

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 0b55c9c95364..f10966e3b4a5 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -108,6 +108,17 @@ typedef int __bitwise fpi_t;
*/
#define FPI_TO_TAIL ((__force fpi_t)BIT(1))

+/*
+ * Don't poison memory with KASAN.
+ * During boot, all non-reserved memblock memory is exposed to the buddy
+ * allocator. Poisoning all that memory lengthens boot time, especially on
+ * systems with large amount of RAM. This flag is used to skip that poisoning.
+ * Assuming that there are no references to those newly exposed pages before
+ * they are ever allocated, this has little effect on KASAN memory tracking.
+ * All memory allocated normally after boot gets poisoned as usual.
+ */
+#define FPI_SKIP_KASAN_POISON ((__force fpi_t)BIT(2))
+
/* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
static DEFINE_MUTEX(pcp_batch_high_lock);
#define MIN_PERCPU_PAGELIST_FRACTION (8)
@@ -384,10 +395,14 @@ static DEFINE_STATIC_KEY_TRUE(deferred_pages);
* on-demand allocation and then freed again before the deferred pages
* initialization is done, but this is not likely to happen.
*/
-static inline void kasan_free_nondeferred_pages(struct page *page, int order)
+static inline void kasan_free_nondeferred_pages(struct page *page, int order,
+ fpi_t fpi_flags)
{
- if (!static_branch_unlikely(&deferred_pages))
- kasan_free_pages(page, order);
+ if (static_branch_unlikely(&deferred_pages))
+ return;
+ if (fpi_flags & FPI_SKIP_KASAN_POISON)
+ return;
+ kasan_free_pages(page, order);
}

/* Returns true if the struct page for the pfn is uninitialised */
@@ -438,7 +453,13 @@ defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
return false;
}
#else
-#define kasan_free_nondeferred_pages(p, o) kasan_free_pages(p, o)
+static inline void kasan_free_nondeferred_pages(struct page *page, int order,
+ fpi_t fpi_flags)
+{
+ if (fpi_flags & FPI_SKIP_KASAN_POISON)
+ return;
+ kasan_free_pages(page, order);
+}

static inline bool early_page_uninitialised(unsigned long pfn)
{
@@ -1216,7 +1237,7 @@ static void kernel_init_free_pages(struct page *page, int numpages)
}

static __always_inline bool free_pages_prepare(struct page *page,
- unsigned int order, bool check_free)
+ unsigned int order, bool check_free, fpi_t fpi_flags)
{
int bad = 0;

@@ -1290,7 +1311,7 @@ static __always_inline bool free_pages_prepare(struct page *page,

debug_pagealloc_unmap_pages(page, 1 << order);

- kasan_free_nondeferred_pages(page, order);
+ kasan_free_nondeferred_pages(page, order, fpi_flags);

return true;
}
@@ -1303,7 +1324,7 @@ static __always_inline bool free_pages_prepare(struct page *page,
*/
static bool free_pcp_prepare(struct page *page)
{
- return free_pages_prepare(page, 0, true);
+ return free_pages_prepare(page, 0, true, FPI_NONE);
}

static bool bulkfree_pcp_prepare(struct page *page)
@@ -1323,9 +1344,9 @@ static bool bulkfree_pcp_prepare(struct page *page)
static bool free_pcp_prepare(struct page *page)
{
if (debug_pagealloc_enabled_static())
- return free_pages_prepare(page, 0, true);
+ return free_pages_prepare(page, 0, true, FPI_NONE);
else
- return free_pages_prepare(page, 0, false);
+ return free_pages_prepare(page, 0, false, FPI_NONE);
}

static bool bulkfree_pcp_prepare(struct page *page)
@@ -1533,7 +1554,7 @@ static void __free_pages_ok(struct page *page, unsigned int order,
int migratetype;
unsigned long pfn = page_to_pfn(page);

- if (!free_pages_prepare(page, order, true))
+ if (!free_pages_prepare(page, order, true, fpi_flags))
return;

migratetype = get_pfnblock_migratetype(page, pfn);
@@ -1570,7 +1591,7 @@ void __free_pages_core(struct page *page, unsigned int order)
* Bypass PCP and place fresh pages right to the tail, primarily
* relevant for memory onlining.
*/
- __free_pages_ok(page, order, FPI_TO_TAIL);
+ __free_pages_ok(page, order, FPI_TO_TAIL | FPI_SKIP_KASAN_POISON);
}

#ifdef CONFIG_NEED_MULTIPLE_NODES
--
2.30.0.478.g8a0d178c01-goog


2021-02-18 10:23:53

by David Hildenbrand

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory

On 17.02.21 21:56, Andrey Konovalov wrote:
> During boot, all non-reserved memblock memory is exposed to the buddy
> allocator. Poisoning all that memory with KASAN lengthens boot time,
> especially on systems with large amount of RAM. This patch makes
> page_alloc to not call kasan_free_pages() on all new memory.
>
> __free_pages_core() is used when exposing fresh memory during system
> boot and when onlining memory during hotplug. This patch adds a new
> FPI_SKIP_KASAN_POISON flag and passes it to __free_pages_ok() through
> free_pages_prepare() from __free_pages_core().
>
> This has little impact on KASAN memory tracking.
>
> Assuming that there are no references to newly exposed pages before they
> are ever allocated, there won't be any intended (but buggy) accesses to
> that memory that KASAN would normally detect.
>
> However, with this patch, KASAN stops detecting wild and large
> out-of-bounds accesses that happen to land on a fresh memory page that
> was never allocated. This is taken as an acceptable trade-off.
>
> All memory allocated normally when the boot is over keeps getting
> poisoned as usual.
>
> Signed-off-by: Andrey Konovalov <[email protected]>
> Change-Id: Iae6b1e4bb8216955ffc14af255a7eaaa6f35324d

Not sure this is the right thing to do, see

https://lkml.kernel.org/r/[email protected]

Reversing the order in which memory gets allocated + used during boot
(in a patch by me) might have revealed an invalid memory access during boot.

I suspect that that issue would no longer get detected with your patch,
as the invalid memory access would simply not get detected. Now, I
cannot prove that :)

--
Thanks,

David / dhildenb

2021-02-18 19:59:05

by Andrey Konovalov

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory

On Thu, Feb 18, 2021 at 9:55 AM David Hildenbrand <[email protected]> wrote:
>
> On 17.02.21 21:56, Andrey Konovalov wrote:
> > During boot, all non-reserved memblock memory is exposed to the buddy
> > allocator. Poisoning all that memory with KASAN lengthens boot time,
> > especially on systems with large amount of RAM. This patch makes
> > page_alloc to not call kasan_free_pages() on all new memory.
> >
> > __free_pages_core() is used when exposing fresh memory during system
> > boot and when onlining memory during hotplug. This patch adds a new
> > FPI_SKIP_KASAN_POISON flag and passes it to __free_pages_ok() through
> > free_pages_prepare() from __free_pages_core().
> >
> > This has little impact on KASAN memory tracking.
> >
> > Assuming that there are no references to newly exposed pages before they
> > are ever allocated, there won't be any intended (but buggy) accesses to
> > that memory that KASAN would normally detect.
> >
> > However, with this patch, KASAN stops detecting wild and large
> > out-of-bounds accesses that happen to land on a fresh memory page that
> > was never allocated. This is taken as an acceptable trade-off.
> >
> > All memory allocated normally when the boot is over keeps getting
> > poisoned as usual.
> >
> > Signed-off-by: Andrey Konovalov <[email protected]>
> > Change-Id: Iae6b1e4bb8216955ffc14af255a7eaaa6f35324d
>
> Not sure this is the right thing to do, see
>
> https://lkml.kernel.org/r/[email protected]
>
> Reversing the order in which memory gets allocated + used during boot
> (in a patch by me) might have revealed an invalid memory access during boot.
>
> I suspect that that issue would no longer get detected with your patch,
> as the invalid memory access would simply not get detected. Now, I
> cannot prove that :)

This looks like a good example.

Ok, what we can do is:

1. For KASAN_GENERIC: leave everything as is to be able to detect
these boot-time bugs.

2. For KASAN_SW_TAGS: remove boot-time poisoning via
kasan_free_pages(), but use the "invalid" tag as the default shadow
value. The end result should be the same: bad accesses will be
detected. For unallocated memory as it has the default "invalid" tag,
and for allocated memory as it's poisoned properly when
allocated/freed.

3. For KASAN_HW_TAGS: just remove boot-time poisoning via
kasan_free_pages(). As the memory tags have a random unspecified
value, we'll still have a 15/16 chance to detect a memory corruption.

This also makes sense from the performance perspective: KASAN_GENERIC
isn't meant to be running in production, so having a larger perf
impact is acceptable. The other two modes will be faster.

2021-02-18 20:05:46

by David Hildenbrand

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory

On 18.02.21 20:40, Andrey Konovalov wrote:
> On Thu, Feb 18, 2021 at 9:55 AM David Hildenbrand <[email protected]> wrote:
>>
>> On 17.02.21 21:56, Andrey Konovalov wrote:
>>> During boot, all non-reserved memblock memory is exposed to the buddy
>>> allocator. Poisoning all that memory with KASAN lengthens boot time,
>>> especially on systems with large amount of RAM. This patch makes
>>> page_alloc to not call kasan_free_pages() on all new memory.
>>>
>>> __free_pages_core() is used when exposing fresh memory during system
>>> boot and when onlining memory during hotplug. This patch adds a new
>>> FPI_SKIP_KASAN_POISON flag and passes it to __free_pages_ok() through
>>> free_pages_prepare() from __free_pages_core().
>>>
>>> This has little impact on KASAN memory tracking.
>>>
>>> Assuming that there are no references to newly exposed pages before they
>>> are ever allocated, there won't be any intended (but buggy) accesses to
>>> that memory that KASAN would normally detect.
>>>
>>> However, with this patch, KASAN stops detecting wild and large
>>> out-of-bounds accesses that happen to land on a fresh memory page that
>>> was never allocated. This is taken as an acceptable trade-off.
>>>
>>> All memory allocated normally when the boot is over keeps getting
>>> poisoned as usual.
>>>
>>> Signed-off-by: Andrey Konovalov <[email protected]>
>>> Change-Id: Iae6b1e4bb8216955ffc14af255a7eaaa6f35324d
>>
>> Not sure this is the right thing to do, see
>>
>> https://lkml.kernel.org/r/[email protected]
>>
>> Reversing the order in which memory gets allocated + used during boot
>> (in a patch by me) might have revealed an invalid memory access during boot.
>>
>> I suspect that that issue would no longer get detected with your patch,
>> as the invalid memory access would simply not get detected. Now, I
>> cannot prove that :)
>
> This looks like a good example.
>
> Ok, what we can do is:
>
> 1. For KASAN_GENERIC: leave everything as is to be able to detect
> these boot-time bugs.
>
> 2. For KASAN_SW_TAGS: remove boot-time poisoning via
> kasan_free_pages(), but use the "invalid" tag as the default shadow
> value. The end result should be the same: bad accesses will be
> detected. For unallocated memory as it has the default "invalid" tag,
> and for allocated memory as it's poisoned properly when
> allocated/freed.
>
> 3. For KASAN_HW_TAGS: just remove boot-time poisoning via
> kasan_free_pages(). As the memory tags have a random unspecified
> value, we'll still have a 15/16 chance to detect a memory corruption.
>
> This also makes sense from the performance perspective: KASAN_GENERIC
> isn't meant to be running in production, so having a larger perf
> impact is acceptable. The other two modes will be faster.

Sounds in principle sane to me.

Side note: I am not sure if anybody runs KASAN in production. Memory is
expensive. Feel free to prove me wrong, I'd be very interest in actual
users.

--
Thanks,

David / dhildenb

2021-02-18 20:28:57

by Andrey Konovalov

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory

On Thu, Feb 18, 2021 at 8:46 PM David Hildenbrand <[email protected]> wrote:
>
> > 1. For KASAN_GENERIC: leave everything as is to be able to detect
> > these boot-time bugs.
> >
> > 2. For KASAN_SW_TAGS: remove boot-time poisoning via
> > kasan_free_pages(), but use the "invalid" tag as the default shadow
> > value. The end result should be the same: bad accesses will be
> > detected. For unallocated memory as it has the default "invalid" tag,
> > and for allocated memory as it's poisoned properly when
> > allocated/freed.
> >
> > 3. For KASAN_HW_TAGS: just remove boot-time poisoning via
> > kasan_free_pages(). As the memory tags have a random unspecified
> > value, we'll still have a 15/16 chance to detect a memory corruption.
> >
> > This also makes sense from the performance perspective: KASAN_GENERIC
> > isn't meant to be running in production, so having a larger perf
> > impact is acceptable. The other two modes will be faster.
>
> Sounds in principle sane to me.

I'll post a v2 soon, thanks!

> Side note: I am not sure if anybody runs KASAN in production. Memory is
> expensive. Feel free to prove me wrong, I'd be very interest in actual
> users.

We run KASAN_SW_TAGS on some dogfood testing devices, and
KASAN_HW_TAGS is being developed with the goal to be running in
production.

2021-02-19 00:08:52

by George Kennedy

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory



On 2/18/2021 3:55 AM, David Hildenbrand wrote:
> On 17.02.21 21:56, Andrey Konovalov wrote:
>> During boot, all non-reserved memblock memory is exposed to the buddy
>> allocator. Poisoning all that memory with KASAN lengthens boot time,
>> especially on systems with large amount of RAM. This patch makes
>> page_alloc to not call kasan_free_pages() on all new memory.
>>
>> __free_pages_core() is used when exposing fresh memory during system
>> boot and when onlining memory during hotplug. This patch adds a new
>> FPI_SKIP_KASAN_POISON flag and passes it to __free_pages_ok() through
>> free_pages_prepare() from __free_pages_core().
>>
>> This has little impact on KASAN memory tracking.
>>
>> Assuming that there are no references to newly exposed pages before they
>> are ever allocated, there won't be any intended (but buggy) accesses to
>> that memory that KASAN would normally detect.
>>
>> However, with this patch, KASAN stops detecting wild and large
>> out-of-bounds accesses that happen to land on a fresh memory page that
>> was never allocated. This is taken as an acceptable trade-off.
>>
>> All memory allocated normally when the boot is over keeps getting
>> poisoned as usual.
>>
>> Signed-off-by: Andrey Konovalov <[email protected]>
>> Change-Id: Iae6b1e4bb8216955ffc14af255a7eaaa6f35324d
>
> Not sure this is the right thing to do, see
>
> https://lkml.kernel.org/r/[email protected]
>
> Reversing the order in which memory gets allocated + used during boot
> (in a patch by me) might have revealed an invalid memory access during
> boot.
>
> I suspect that that issue would no longer get detected with your
> patch, as the invalid memory access would simply not get detected.
> Now, I cannot prove that :)

Since David's patch we're having trouble with the iBFT ACPI table, which
is mapped in via kmap() - see acpi_map() in "drivers/acpi/osl.c". KASAN
detects that it is being used after free when ibft_init() accesses the
iBFT table, but as of yet we can't find where it get's freed (we've
instrumented calls to kunmap()).

Thank you,
George

2021-02-19 00:14:09

by Andrey Konovalov

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory

On Fri, Feb 19, 2021 at 1:06 AM George Kennedy
<[email protected]> wrote:
>
>
>
> On 2/18/2021 3:55 AM, David Hildenbrand wrote:
> > On 17.02.21 21:56, Andrey Konovalov wrote:
> >> During boot, all non-reserved memblock memory is exposed to the buddy
> >> allocator. Poisoning all that memory with KASAN lengthens boot time,
> >> especially on systems with large amount of RAM. This patch makes
> >> page_alloc to not call kasan_free_pages() on all new memory.
> >>
> >> __free_pages_core() is used when exposing fresh memory during system
> >> boot and when onlining memory during hotplug. This patch adds a new
> >> FPI_SKIP_KASAN_POISON flag and passes it to __free_pages_ok() through
> >> free_pages_prepare() from __free_pages_core().
> >>
> >> This has little impact on KASAN memory tracking.
> >>
> >> Assuming that there are no references to newly exposed pages before they
> >> are ever allocated, there won't be any intended (but buggy) accesses to
> >> that memory that KASAN would normally detect.
> >>
> >> However, with this patch, KASAN stops detecting wild and large
> >> out-of-bounds accesses that happen to land on a fresh memory page that
> >> was never allocated. This is taken as an acceptable trade-off.
> >>
> >> All memory allocated normally when the boot is over keeps getting
> >> poisoned as usual.
> >>
> >> Signed-off-by: Andrey Konovalov <[email protected]>
> >> Change-Id: Iae6b1e4bb8216955ffc14af255a7eaaa6f35324d
> >
> > Not sure this is the right thing to do, see
> >
> > https://lkml.kernel.org/r/[email protected]
> >
> > Reversing the order in which memory gets allocated + used during boot
> > (in a patch by me) might have revealed an invalid memory access during
> > boot.
> >
> > I suspect that that issue would no longer get detected with your
> > patch, as the invalid memory access would simply not get detected.
> > Now, I cannot prove that :)
>
> Since David's patch we're having trouble with the iBFT ACPI table, which
> is mapped in via kmap() - see acpi_map() in "drivers/acpi/osl.c". KASAN
> detects that it is being used after free when ibft_init() accesses the
> iBFT table, but as of yet we can't find where it get's freed (we've
> instrumented calls to kunmap()).

Maybe it doesn't get freed, but what you see is a wild or a large
out-of-bounds access. Since KASAN marks all memory as freed during the
memblock->page_alloc transition, such bugs can manifest as
use-after-frees.

2021-02-19 16:48:37

by George Kennedy

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory



On 2/18/2021 7:09 PM, Andrey Konovalov wrote:
> On Fri, Feb 19, 2021 at 1:06 AM George Kennedy
> <[email protected]> wrote:
>>
>>
>> On 2/18/2021 3:55 AM, David Hildenbrand wrote:
>>> On 17.02.21 21:56, Andrey Konovalov wrote:
>>>> During boot, all non-reserved memblock memory is exposed to the buddy
>>>> allocator. Poisoning all that memory with KASAN lengthens boot time,
>>>> especially on systems with large amount of RAM. This patch makes
>>>> page_alloc to not call kasan_free_pages() on all new memory.
>>>>
>>>> __free_pages_core() is used when exposing fresh memory during system
>>>> boot and when onlining memory during hotplug. This patch adds a new
>>>> FPI_SKIP_KASAN_POISON flag and passes it to __free_pages_ok() through
>>>> free_pages_prepare() from __free_pages_core().
>>>>
>>>> This has little impact on KASAN memory tracking.
>>>>
>>>> Assuming that there are no references to newly exposed pages before they
>>>> are ever allocated, there won't be any intended (but buggy) accesses to
>>>> that memory that KASAN would normally detect.
>>>>
>>>> However, with this patch, KASAN stops detecting wild and large
>>>> out-of-bounds accesses that happen to land on a fresh memory page that
>>>> was never allocated. This is taken as an acceptable trade-off.
>>>>
>>>> All memory allocated normally when the boot is over keeps getting
>>>> poisoned as usual.
>>>>
>>>> Signed-off-by: Andrey Konovalov <[email protected]>
>>>> Change-Id: Iae6b1e4bb8216955ffc14af255a7eaaa6f35324d
>>> Not sure this is the right thing to do, see
>>>
>>> https://lkml.kernel.org/r/[email protected]
>>>
>>> Reversing the order in which memory gets allocated + used during boot
>>> (in a patch by me) might have revealed an invalid memory access during
>>> boot.
>>>
>>> I suspect that that issue would no longer get detected with your
>>> patch, as the invalid memory access would simply not get detected.
>>> Now, I cannot prove that :)
>> Since David's patch we're having trouble with the iBFT ACPI table, which
>> is mapped in via kmap() - see acpi_map() in "drivers/acpi/osl.c". KASAN
>> detects that it is being used after free when ibft_init() accesses the
>> iBFT table, but as of yet we can't find where it get's freed (we've
>> instrumented calls to kunmap()).
> Maybe it doesn't get freed, but what you see is a wild or a large
> out-of-bounds access. Since KASAN marks all memory as freed during the
> memblock->page_alloc transition, such bugs can manifest as
> use-after-frees.

It gets freed and re-used. By the time the iBFT table is accessed by
ibft_init() the page has been over-written.

Setting page flags like the following before the call to kmap() prevents
the iBFT table page from being freed:

diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c
index 0418feb..41c1bbd 100644
--- a/drivers/acpi/osl.c
+++ b/drivers/acpi/osl.c
@@ -287,9 +287,14 @@ static void __iomem *acpi_map(acpi_physical_address
pg_off, unsigned long pg_sz)

        pfn = pg_off >> PAGE_SHIFT;
        if (should_use_kmap(pfn)) {
+               struct page *page =  pfn_to_page(pfn);
+
                if (pg_sz > PAGE_SIZE)
                        return NULL;
-               return (void __iomem __force *)kmap(pfn_to_page(pfn));
+
+               page->flags |= ((1UL << PG_unevictable) | (1UL <<
PG_reserved) | (1UL << PG_locked));
+
+               return (void __iomem __force *)kmap(page);
        } else
                return acpi_os_ioremap(pg_off, pg_sz);
 }

Just not sure of the correct way to set the page flags.

George

2021-02-19 23:08:24

by George Kennedy

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory



On 2/19/2021 11:45 AM, George Kennedy wrote:
>
>
> On 2/18/2021 7:09 PM, Andrey Konovalov wrote:
>> On Fri, Feb 19, 2021 at 1:06 AM George Kennedy
>> <[email protected]> wrote:
>>>
>>>
>>> On 2/18/2021 3:55 AM, David Hildenbrand wrote:
>>>> On 17.02.21 21:56, Andrey Konovalov wrote:
>>>>> During boot, all non-reserved memblock memory is exposed to the buddy
>>>>> allocator. Poisoning all that memory with KASAN lengthens boot time,
>>>>> especially on systems with large amount of RAM. This patch makes
>>>>> page_alloc to not call kasan_free_pages() on all new memory.
>>>>>
>>>>> __free_pages_core() is used when exposing fresh memory during system
>>>>> boot and when onlining memory during hotplug. This patch adds a new
>>>>> FPI_SKIP_KASAN_POISON flag and passes it to __free_pages_ok() through
>>>>> free_pages_prepare() from __free_pages_core().
>>>>>
>>>>> This has little impact on KASAN memory tracking.
>>>>>
>>>>> Assuming that there are no references to newly exposed pages
>>>>> before they
>>>>> are ever allocated, there won't be any intended (but buggy)
>>>>> accesses to
>>>>> that memory that KASAN would normally detect.
>>>>>
>>>>> However, with this patch, KASAN stops detecting wild and large
>>>>> out-of-bounds accesses that happen to land on a fresh memory page
>>>>> that
>>>>> was never allocated. This is taken as an acceptable trade-off.
>>>>>
>>>>> All memory allocated normally when the boot is over keeps getting
>>>>> poisoned as usual.
>>>>>
>>>>> Signed-off-by: Andrey Konovalov <[email protected]>
>>>>> Change-Id: Iae6b1e4bb8216955ffc14af255a7eaaa6f35324d
>>>> Not sure this is the right thing to do, see
>>>>
>>>> https://lkml.kernel.org/r/[email protected]
>>>>
>>>>
>>>> Reversing the order in which memory gets allocated + used during boot
>>>> (in a patch by me) might have revealed an invalid memory access during
>>>> boot.
>>>>
>>>> I suspect that that issue would no longer get detected with your
>>>> patch, as the invalid memory access would simply not get detected.
>>>> Now, I cannot prove that :)
>>> Since David's patch we're having trouble with the iBFT ACPI table,
>>> which
>>> is mapped in via kmap() - see acpi_map() in "drivers/acpi/osl.c". KASAN
>>> detects that it is being used after free when ibft_init() accesses the
>>> iBFT table, but as of yet we can't find where it get's freed (we've
>>> instrumented calls to kunmap()).
>> Maybe it doesn't get freed, but what you see is a wild or a large
>> out-of-bounds access. Since KASAN marks all memory as freed during the
>> memblock->page_alloc transition, such bugs can manifest as
>> use-after-frees.
>
> It gets freed and re-used. By the time the iBFT table is accessed by
> ibft_init() the page has been over-written.
>
> Setting page flags like the following before the call to kmap()
> prevents the iBFT table page from being freed:

Cleaned up version:

diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c
index 0418feb..8f0a8e7 100644
--- a/drivers/acpi/osl.c
+++ b/drivers/acpi/osl.c
@@ -287,9 +287,12 @@ static void __iomem *acpi_map(acpi_physical_address
pg_off, unsigned long pg_sz)

     pfn = pg_off >> PAGE_SHIFT;
     if (should_use_kmap(pfn)) {
+        struct page *page = pfn_to_page(pfn);
+
         if (pg_sz > PAGE_SIZE)
             return NULL;
-        return (void __iomem __force *)kmap(pfn_to_page(pfn));
+        SetPageReserved(page);
+        return (void __iomem __force *)kmap(page);
     } else
         return acpi_os_ioremap(pg_off, pg_sz);
 }
@@ -299,9 +302,12 @@ static void acpi_unmap(acpi_physical_address
pg_off, void __iomem *vaddr)
     unsigned long pfn;

     pfn = pg_off >> PAGE_SHIFT;
-    if (should_use_kmap(pfn))
-        kunmap(pfn_to_page(pfn));
-    else
+    if (should_use_kmap(pfn)) {
+        struct page *page = pfn_to_page(pfn);
+
+        ClearPageReserved(page);
+        kunmap(page);
+    } else
         iounmap(vaddr);
 }

David, the above works, but wondering why it is now necessary. kunmap()
is not hit. What other ways could a page mapped via kmap() be unmapped?

Thank you,
George

>
> diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c
> index 0418feb..41c1bbd 100644
> --- a/drivers/acpi/osl.c
> +++ b/drivers/acpi/osl.c
> @@ -287,9 +287,14 @@ static void __iomem
> *acpi_map(acpi_physical_address pg_off, unsigned long pg_sz)
>
>         pfn = pg_off >> PAGE_SHIFT;
>         if (should_use_kmap(pfn)) {
> +               struct page *page =  pfn_to_page(pfn);
> +
>                 if (pg_sz > PAGE_SIZE)
>                         return NULL;
> -               return (void __iomem __force *)kmap(pfn_to_page(pfn));
> +
> +               page->flags |= ((1UL << PG_unevictable) | (1UL <<
> PG_reserved) | (1UL << PG_locked));
> +
> +               return (void __iomem __force *)kmap(page);
>         } else
>                 return acpi_os_ioremap(pg_off, pg_sz);
>  }
>
> Just not sure of the correct way to set the page flags.
>
> George
>

2021-02-22 09:55:30

by David Hildenbrand

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory

On 20.02.21 00:04, George Kennedy wrote:
>
>
> On 2/19/2021 11:45 AM, George Kennedy wrote:
>>
>>
>> On 2/18/2021 7:09 PM, Andrey Konovalov wrote:
>>> On Fri, Feb 19, 2021 at 1:06 AM George Kennedy
>>> <[email protected]> wrote:
>>>>
>>>>
>>>> On 2/18/2021 3:55 AM, David Hildenbrand wrote:
>>>>> On 17.02.21 21:56, Andrey Konovalov wrote:
>>>>>> During boot, all non-reserved memblock memory is exposed to the buddy
>>>>>> allocator. Poisoning all that memory with KASAN lengthens boot time,
>>>>>> especially on systems with large amount of RAM. This patch makes
>>>>>> page_alloc to not call kasan_free_pages() on all new memory.
>>>>>>
>>>>>> __free_pages_core() is used when exposing fresh memory during system
>>>>>> boot and when onlining memory during hotplug. This patch adds a new
>>>>>> FPI_SKIP_KASAN_POISON flag and passes it to __free_pages_ok() through
>>>>>> free_pages_prepare() from __free_pages_core().
>>>>>>
>>>>>> This has little impact on KASAN memory tracking.
>>>>>>
>>>>>> Assuming that there are no references to newly exposed pages
>>>>>> before they
>>>>>> are ever allocated, there won't be any intended (but buggy)
>>>>>> accesses to
>>>>>> that memory that KASAN would normally detect.
>>>>>>
>>>>>> However, with this patch, KASAN stops detecting wild and large
>>>>>> out-of-bounds accesses that happen to land on a fresh memory page
>>>>>> that
>>>>>> was never allocated. This is taken as an acceptable trade-off.
>>>>>>
>>>>>> All memory allocated normally when the boot is over keeps getting
>>>>>> poisoned as usual.
>>>>>>
>>>>>> Signed-off-by: Andrey Konovalov <[email protected]>
>>>>>> Change-Id: Iae6b1e4bb8216955ffc14af255a7eaaa6f35324d
>>>>> Not sure this is the right thing to do, see
>>>>>
>>>>> https://lkml.kernel.org/r/[email protected]
>>>>>
>>>>>
>>>>> Reversing the order in which memory gets allocated + used during boot
>>>>> (in a patch by me) might have revealed an invalid memory access during
>>>>> boot.
>>>>>
>>>>> I suspect that that issue would no longer get detected with your
>>>>> patch, as the invalid memory access would simply not get detected.
>>>>> Now, I cannot prove that :)
>>>> Since David's patch we're having trouble with the iBFT ACPI table,
>>>> which
>>>> is mapped in via kmap() - see acpi_map() in "drivers/acpi/osl.c". KASAN
>>>> detects that it is being used after free when ibft_init() accesses the
>>>> iBFT table, but as of yet we can't find where it get's freed (we've
>>>> instrumented calls to kunmap()).
>>> Maybe it doesn't get freed, but what you see is a wild or a large
>>> out-of-bounds access. Since KASAN marks all memory as freed during the
>>> memblock->page_alloc transition, such bugs can manifest as
>>> use-after-frees.
>>
>> It gets freed and re-used. By the time the iBFT table is accessed by
>> ibft_init() the page has been over-written.
>>
>> Setting page flags like the following before the call to kmap()
>> prevents the iBFT table page from being freed:
>
> Cleaned up version:
>
> diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c
> index 0418feb..8f0a8e7 100644
> --- a/drivers/acpi/osl.c
> +++ b/drivers/acpi/osl.c
> @@ -287,9 +287,12 @@ static void __iomem *acpi_map(acpi_physical_address
> pg_off, unsigned long pg_sz)
>
>      pfn = pg_off >> PAGE_SHIFT;
>      if (should_use_kmap(pfn)) {
> +        struct page *page = pfn_to_page(pfn);
> +
>          if (pg_sz > PAGE_SIZE)
>              return NULL;
> -        return (void __iomem __force *)kmap(pfn_to_page(pfn));
> +        SetPageReserved(page);
> +        return (void __iomem __force *)kmap(page);
>      } else
>          return acpi_os_ioremap(pg_off, pg_sz);
>  }
> @@ -299,9 +302,12 @@ static void acpi_unmap(acpi_physical_address
> pg_off, void __iomem *vaddr)
>      unsigned long pfn;
>
>      pfn = pg_off >> PAGE_SHIFT;
> -    if (should_use_kmap(pfn))
> -        kunmap(pfn_to_page(pfn));
> -    else
> +    if (should_use_kmap(pfn)) {
> +        struct page *page = pfn_to_page(pfn);
> +
> +        ClearPageReserved(page);
> +        kunmap(page);
> +    } else
>          iounmap(vaddr);
>  }
>
> David, the above works, but wondering why it is now necessary. kunmap()
> is not hit. What other ways could a page mapped via kmap() be unmapped?
>

Let me look into the code ... I have little experience with ACPI
details, so bear with me.

I assume that acpi_map()/acpi_unmap() map some firmware blob that is
provided via firmware/bios/... to us.

should_use_kmap() tells us whether
a) we have a "struct page" and should kmap() that one
b) we don't have a "struct page" and should ioremap.

As it is a blob, the firmware should always reserve that memory region
via memblock (e.g., memblock_reserve()), such that we either
1) don't create a memmap ("struct page") at all (-> case b) )
2) if we have to create e memmap, we mark the page PG_reserved and
*never* expose it to the buddy (-> case a) )


Are you telling me that in this case we might have a memmap for the HW
blob that is *not* PG_reserved? In that case it most probably got
exposed to the buddy where it can happily get allocated/freed.

The latent BUG would be that that blob gets exposed to the system like
ordinary RAM, and not reserved via memblock early during boot. Assuming
that blob has a low physical address, with my patch it will get
allocated/used a lot earlier - which would mean we trigger this latent
BUG now more easily.

There have been similar latent BUGs on ARM boards that my patch
discovered where special RAM regions did not get marked as reserved via
the device tree properly.

Now, this is just a wild guess :) Can you dump the page when mapping
(before PageReserved()) and when unmapping, to see what the state of
that memmap is?

--
Thanks,

David / dhildenb

2021-02-22 15:23:59

by George Kennedy

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory



On 2/22/2021 4:52 AM, David Hildenbrand wrote:
> On 20.02.21 00:04, George Kennedy wrote:
>>
>>
>> On 2/19/2021 11:45 AM, George Kennedy wrote:
>>>
>>>
>>> On 2/18/2021 7:09 PM, Andrey Konovalov wrote:
>>>> On Fri, Feb 19, 2021 at 1:06 AM George Kennedy
>>>> <[email protected]> wrote:
>>>>>
>>>>>
>>>>> On 2/18/2021 3:55 AM, David Hildenbrand wrote:
>>>>>> On 17.02.21 21:56, Andrey Konovalov wrote:
>>>>>>> During boot, all non-reserved memblock memory is exposed to the
>>>>>>> buddy
>>>>>>> allocator. Poisoning all that memory with KASAN lengthens boot
>>>>>>> time,
>>>>>>> especially on systems with large amount of RAM. This patch makes
>>>>>>> page_alloc to not call kasan_free_pages() on all new memory.
>>>>>>>
>>>>>>> __free_pages_core() is used when exposing fresh memory during
>>>>>>> system
>>>>>>> boot and when onlining memory during hotplug. This patch adds a new
>>>>>>> FPI_SKIP_KASAN_POISON flag and passes it to __free_pages_ok()
>>>>>>> through
>>>>>>> free_pages_prepare() from __free_pages_core().
>>>>>>>
>>>>>>> This has little impact on KASAN memory tracking.
>>>>>>>
>>>>>>> Assuming that there are no references to newly exposed pages
>>>>>>> before they
>>>>>>> are ever allocated, there won't be any intended (but buggy)
>>>>>>> accesses to
>>>>>>> that memory that KASAN would normally detect.
>>>>>>>
>>>>>>> However, with this patch, KASAN stops detecting wild and large
>>>>>>> out-of-bounds accesses that happen to land on a fresh memory page
>>>>>>> that
>>>>>>> was never allocated. This is taken as an acceptable trade-off.
>>>>>>>
>>>>>>> All memory allocated normally when the boot is over keeps getting
>>>>>>> poisoned as usual.
>>>>>>>
>>>>>>> Signed-off-by: Andrey Konovalov <[email protected]>
>>>>>>> Change-Id: Iae6b1e4bb8216955ffc14af255a7eaaa6f35324d
>>>>>> Not sure this is the right thing to do, see
>>>>>>
>>>>>> https://lkml.kernel.org/r/[email protected]
>>>>>>
>>>>>>
>>>>>>
>>>>>> Reversing the order in which memory gets allocated + used during
>>>>>> boot
>>>>>> (in a patch by me) might have revealed an invalid memory access
>>>>>> during
>>>>>> boot.
>>>>>>
>>>>>> I suspect that that issue would no longer get detected with your
>>>>>> patch, as the invalid memory access would simply not get detected.
>>>>>> Now, I cannot prove that :)
>>>>> Since David's patch we're having trouble with the iBFT ACPI table,
>>>>> which
>>>>> is mapped in via kmap() - see acpi_map() in "drivers/acpi/osl.c".
>>>>> KASAN
>>>>> detects that it is being used after free when ibft_init() accesses
>>>>> the
>>>>> iBFT table, but as of yet we can't find where it get's freed (we've
>>>>> instrumented calls to kunmap()).
>>>> Maybe it doesn't get freed, but what you see is a wild or a large
>>>> out-of-bounds access. Since KASAN marks all memory as freed during the
>>>> memblock->page_alloc transition, such bugs can manifest as
>>>> use-after-frees.
>>>
>>> It gets freed and re-used. By the time the iBFT table is accessed by
>>> ibft_init() the page has been over-written.
>>>
>>> Setting page flags like the following before the call to kmap()
>>> prevents the iBFT table page from being freed:
>>
>> Cleaned up version:
>>
>> diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c
>> index 0418feb..8f0a8e7 100644
>> --- a/drivers/acpi/osl.c
>> +++ b/drivers/acpi/osl.c
>> @@ -287,9 +287,12 @@ static void __iomem *acpi_map(acpi_physical_address
>> pg_off, unsigned long pg_sz)
>>
>>        pfn = pg_off >> PAGE_SHIFT;
>>        if (should_use_kmap(pfn)) {
>> +        struct page *page = pfn_to_page(pfn);
>> +
>>            if (pg_sz > PAGE_SIZE)
>>                return NULL;
>> -        return (void __iomem __force *)kmap(pfn_to_page(pfn));
>> +        SetPageReserved(page);
>> +        return (void __iomem __force *)kmap(page);
>>        } else
>>            return acpi_os_ioremap(pg_off, pg_sz);
>>    }
>> @@ -299,9 +302,12 @@ static void acpi_unmap(acpi_physical_address
>> pg_off, void __iomem *vaddr)
>>        unsigned long pfn;
>>
>>        pfn = pg_off >> PAGE_SHIFT;
>> -    if (should_use_kmap(pfn))
>> -        kunmap(pfn_to_page(pfn));
>> -    else
>> +    if (should_use_kmap(pfn)) {
>> +        struct page *page = pfn_to_page(pfn);
>> +
>> +        ClearPageReserved(page);
>> +        kunmap(page);
>> +    } else
>>            iounmap(vaddr);
>>    }
>>
>> David, the above works, but wondering why it is now necessary. kunmap()
>> is not hit. What other ways could a page mapped via kmap() be unmapped?
>>
>
> Let me look into the code ... I have little experience with ACPI
> details, so bear with me.
>
> I assume that acpi_map()/acpi_unmap() map some firmware blob that is
> provided via firmware/bios/... to us.
>
> should_use_kmap() tells us whether
> a) we have a "struct page" and should kmap() that one
> b) we don't have a "struct page" and should ioremap.
>
> As it is a blob, the firmware should always reserve that memory region
> via memblock (e.g., memblock_reserve()), such that we either
> 1) don't create a memmap ("struct page") at all (-> case b) )
> 2) if we have to create e memmap, we mark the page PG_reserved and
>    *never* expose it to the buddy (-> case a) )
>
>
> Are you telling me that in this case we might have a memmap for the HW
> blob that is *not* PG_reserved? In that case it most probably got
> exposed to the buddy where it can happily get allocated/freed.
>
> The latent BUG would be that that blob gets exposed to the system like
> ordinary RAM, and not reserved via memblock early during boot.
> Assuming that blob has a low physical address, with my patch it will
> get allocated/used a lot earlier - which would mean we trigger this
> latent BUG now more easily.
>
> There have been similar latent BUGs on ARM boards that my patch
> discovered where special RAM regions did not get marked as reserved
> via the device tree properly.
>
> Now, this is just a wild guess :) Can you dump the page when mapping
> (before PageReserved()) and when unmapping, to see what the state of
> that memmap is?

Thank you David for the explanation and your help on this,

dump_page() before PageReserved and before kmap() in the above patch:

[    1.116480] ACPI: Core revision 20201113
[    1.117628] XXX acpi_map: about to call kmap()...
[    1.118561] page:ffffea0002f914c0 refcount:0 mapcount:0
mapping:0000000000000000 index:0x0 pfn:0xbe453
[    1.120381] flags: 0xfffffc0000000()
[    1.121116] raw: 000fffffc0000000 ffffea0002f914c8 ffffea0002f914c8
0000000000000000
[    1.122638] raw: 0000000000000000 0000000000000000 00000000ffffffff
0000000000000000
[    1.124146] page dumped because: acpi_map pre SetPageReserved

I also added dump_page() before unmapping, but it is not hit. The
following for the same pfn now shows up I believe as a result of setting
PageReserved:

[   28.098208] BUG: Bad page state in process modprobe  pfn:be453
[   28.098394] page:ffffea0002f914c0 refcount:0 mapcount:0
mapping:0000000000000000 index:0x1 pfn:0xbe453
[   28.098394] flags: 0xfffffc0001000(reserved)
[   28.098394] raw: 000fffffc0001000 dead000000000100 dead000000000122
0000000000000000
[   28.098394] raw: 0000000000000001 0000000000000000 00000000ffffffff
0000000000000000
[   28.098394] page dumped because: PAGE_FLAGS_CHECK_AT_PREP flag(s) set
[   28.098394] page_owner info is not present (never set?)
[   28.098394] Modules linked in:
[   28.098394] CPU: 2 PID: 204 Comm: modprobe Not tainted 5.11.0-3dbd5e3 #66
[   28.098394] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
BIOS 0.0.0 02/06/2015
[   28.098394] Call Trace:
[   28.098394]  dump_stack+0xdb/0x120
[   28.098394]  bad_page.cold.108+0xc6/0xcb
[   28.098394]  check_new_page_bad+0x47/0xa0
[   28.098394]  get_page_from_freelist+0x30cd/0x5730
[   28.098394]  ? __isolate_free_page+0x4f0/0x4f0
[   28.098394]  ? init_object+0x7e/0x90
[   28.098394]  __alloc_pages_nodemask+0x2d8/0x650
[   28.098394]  ? write_comp_data+0x2f/0x90
[   28.098394]  ? __alloc_pages_slowpath.constprop.103+0x2110/0x2110
[   28.098394]  ? __sanitizer_cov_trace_pc+0x21/0x50
[   28.098394]  alloc_pages_vma+0xe2/0x560
[   28.098394]  do_fault+0x194/0x12c0
[   28.098394]  ? write_comp_data+0x2f/0x90
[   28.098394]  __handle_mm_fault+0x1650/0x26c0
[   28.098394]  ? copy_page_range+0x1350/0x1350
[   28.098394]  ? write_comp_data+0x2f/0x90
[   28.098394]  ? write_comp_data+0x2f/0x90
[   28.098394]  handle_mm_fault+0x1f9/0x810
[   28.098394]  ? write_comp_data+0x2f/0x90
[   28.098394]  do_user_addr_fault+0x6f7/0xca0
[   28.098394]  exc_page_fault+0xaf/0x1a0
[   28.098394]  asm_exc_page_fault+0x1e/0x30
[   28.098394] RIP: 0010:__clear_user+0x30/0x60

What would be  the correct way to reserve the page so that the above
would not be hit?

BTW, this is running with Konrad's patch that pairs acpi_get_table &
acpi_put_table for the iBFT table which should result in an eventual
call to acpi_unmap() and kunmap(), though that does not occur. Could be
a possible acpi page refcount issue that will have to be looked into.

George

2021-02-22 16:16:54

by David Hildenbrand

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory

On 22.02.21 16:13, George Kennedy wrote:
>
>
> On 2/22/2021 4:52 AM, David Hildenbrand wrote:
>> On 20.02.21 00:04, George Kennedy wrote:
>>>
>>>
>>> On 2/19/2021 11:45 AM, George Kennedy wrote:
>>>>
>>>>
>>>> On 2/18/2021 7:09 PM, Andrey Konovalov wrote:
>>>>> On Fri, Feb 19, 2021 at 1:06 AM George Kennedy
>>>>> <[email protected]> wrote:
>>>>>>
>>>>>>
>>>>>> On 2/18/2021 3:55 AM, David Hildenbrand wrote:
>>>>>>> On 17.02.21 21:56, Andrey Konovalov wrote:
>>>>>>>> During boot, all non-reserved memblock memory is exposed to the
>>>>>>>> buddy
>>>>>>>> allocator. Poisoning all that memory with KASAN lengthens boot
>>>>>>>> time,
>>>>>>>> especially on systems with large amount of RAM. This patch makes
>>>>>>>> page_alloc to not call kasan_free_pages() on all new memory.
>>>>>>>>
>>>>>>>> __free_pages_core() is used when exposing fresh memory during
>>>>>>>> system
>>>>>>>> boot and when onlining memory during hotplug. This patch adds a new
>>>>>>>> FPI_SKIP_KASAN_POISON flag and passes it to __free_pages_ok()
>>>>>>>> through
>>>>>>>> free_pages_prepare() from __free_pages_core().
>>>>>>>>
>>>>>>>> This has little impact on KASAN memory tracking.
>>>>>>>>
>>>>>>>> Assuming that there are no references to newly exposed pages
>>>>>>>> before they
>>>>>>>> are ever allocated, there won't be any intended (but buggy)
>>>>>>>> accesses to
>>>>>>>> that memory that KASAN would normally detect.
>>>>>>>>
>>>>>>>> However, with this patch, KASAN stops detecting wild and large
>>>>>>>> out-of-bounds accesses that happen to land on a fresh memory page
>>>>>>>> that
>>>>>>>> was never allocated. This is taken as an acceptable trade-off.
>>>>>>>>
>>>>>>>> All memory allocated normally when the boot is over keeps getting
>>>>>>>> poisoned as usual.
>>>>>>>>
>>>>>>>> Signed-off-by: Andrey Konovalov <[email protected]>
>>>>>>>> Change-Id: Iae6b1e4bb8216955ffc14af255a7eaaa6f35324d
>>>>>>> Not sure this is the right thing to do, see
>>>>>>>
>>>>>>> https://lkml.kernel.org/r/[email protected]
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> Reversing the order in which memory gets allocated + used during
>>>>>>> boot
>>>>>>> (in a patch by me) might have revealed an invalid memory access
>>>>>>> during
>>>>>>> boot.
>>>>>>>
>>>>>>> I suspect that that issue would no longer get detected with your
>>>>>>> patch, as the invalid memory access would simply not get detected.
>>>>>>> Now, I cannot prove that :)
>>>>>> Since David's patch we're having trouble with the iBFT ACPI table,
>>>>>> which
>>>>>> is mapped in via kmap() - see acpi_map() in "drivers/acpi/osl.c".
>>>>>> KASAN
>>>>>> detects that it is being used after free when ibft_init() accesses
>>>>>> the
>>>>>> iBFT table, but as of yet we can't find where it get's freed (we've
>>>>>> instrumented calls to kunmap()).
>>>>> Maybe it doesn't get freed, but what you see is a wild or a large
>>>>> out-of-bounds access. Since KASAN marks all memory as freed during the
>>>>> memblock->page_alloc transition, such bugs can manifest as
>>>>> use-after-frees.
>>>>
>>>> It gets freed and re-used. By the time the iBFT table is accessed by
>>>> ibft_init() the page has been over-written.
>>>>
>>>> Setting page flags like the following before the call to kmap()
>>>> prevents the iBFT table page from being freed:
>>>
>>> Cleaned up version:
>>>
>>> diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c
>>> index 0418feb..8f0a8e7 100644
>>> --- a/drivers/acpi/osl.c
>>> +++ b/drivers/acpi/osl.c
>>> @@ -287,9 +287,12 @@ static void __iomem *acpi_map(acpi_physical_address
>>> pg_off, unsigned long pg_sz)
>>>
>>>        pfn = pg_off >> PAGE_SHIFT;
>>>        if (should_use_kmap(pfn)) {
>>> +        struct page *page = pfn_to_page(pfn);
>>> +
>>>            if (pg_sz > PAGE_SIZE)
>>>                return NULL;
>>> -        return (void __iomem __force *)kmap(pfn_to_page(pfn));
>>> +        SetPageReserved(page);
>>> +        return (void __iomem __force *)kmap(page);
>>>        } else
>>>            return acpi_os_ioremap(pg_off, pg_sz);
>>>    }
>>> @@ -299,9 +302,12 @@ static void acpi_unmap(acpi_physical_address
>>> pg_off, void __iomem *vaddr)
>>>        unsigned long pfn;
>>>
>>>        pfn = pg_off >> PAGE_SHIFT;
>>> -    if (should_use_kmap(pfn))
>>> -        kunmap(pfn_to_page(pfn));
>>> -    else
>>> +    if (should_use_kmap(pfn)) {
>>> +        struct page *page = pfn_to_page(pfn);
>>> +
>>> +        ClearPageReserved(page);
>>> +        kunmap(page);
>>> +    } else
>>>            iounmap(vaddr);
>>>    }
>>>
>>> David, the above works, but wondering why it is now necessary. kunmap()
>>> is not hit. What other ways could a page mapped via kmap() be unmapped?
>>>
>>
>> Let me look into the code ... I have little experience with ACPI
>> details, so bear with me.
>>
>> I assume that acpi_map()/acpi_unmap() map some firmware blob that is
>> provided via firmware/bios/... to us.
>>
>> should_use_kmap() tells us whether
>> a) we have a "struct page" and should kmap() that one
>> b) we don't have a "struct page" and should ioremap.
>>
>> As it is a blob, the firmware should always reserve that memory region
>> via memblock (e.g., memblock_reserve()), such that we either
>> 1) don't create a memmap ("struct page") at all (-> case b) )
>> 2) if we have to create e memmap, we mark the page PG_reserved and
>>    *never* expose it to the buddy (-> case a) )
>>
>>
>> Are you telling me that in this case we might have a memmap for the HW
>> blob that is *not* PG_reserved? In that case it most probably got
>> exposed to the buddy where it can happily get allocated/freed.
>>
>> The latent BUG would be that that blob gets exposed to the system like
>> ordinary RAM, and not reserved via memblock early during boot.
>> Assuming that blob has a low physical address, with my patch it will
>> get allocated/used a lot earlier - which would mean we trigger this
>> latent BUG now more easily.
>>
>> There have been similar latent BUGs on ARM boards that my patch
>> discovered where special RAM regions did not get marked as reserved
>> via the device tree properly.
>>
>> Now, this is just a wild guess :) Can you dump the page when mapping
>> (before PageReserved()) and when unmapping, to see what the state of
>> that memmap is?
>
> Thank you David for the explanation and your help on this,
>
> dump_page() before PageReserved and before kmap() in the above patch:
>
> [    1.116480] ACPI: Core revision 20201113
> [    1.117628] XXX acpi_map: about to call kmap()...
> [    1.118561] page:ffffea0002f914c0 refcount:0 mapcount:0
> mapping:0000000000000000 index:0x0 pfn:0xbe453
> [    1.120381] flags: 0xfffffc0000000()
> [    1.121116] raw: 000fffffc0000000 ffffea0002f914c8 ffffea0002f914c8
> 0000000000000000
> [    1.122638] raw: 0000000000000000 0000000000000000 00000000ffffffff
> 0000000000000000
> [    1.124146] page dumped because: acpi_map pre SetPageReserved
>
> I also added dump_page() before unmapping, but it is not hit. The
> following for the same pfn now shows up I believe as a result of setting
> PageReserved:
>
> [   28.098208] BUG:Bad page state in process mo dprobe  pfn:be453
> [   28.098394] page:ffffea0002f914c0 refcount:0 mapcount:0
> mapping:0000000000000000 index:0x1 pfn:0xbe453
> [   28.098394] flags: 0xfffffc0001000(reserved)
> [   28.098394] raw: 000fffffc0001000 dead000000000100 dead000000000122
> 0000000000000000
> [   28.098394] raw: 0000000000000001 0000000000000000 00000000ffffffff
> 0000000000000000
> [   28.098394] page dumped because: PAGE_FLAGS_CHECK_AT_PREP flag(s) set
> [   28.098394] page_owner info is not present (never set?)
> [   28.098394] Modules linked in:
> [   28.098394] CPU: 2 PID: 204 Comm: modprobe Not tainted 5.11.0-3dbd5e3 #66
> [   28.098394] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
> BIOS 0.0.0 02/06/2015
> [   28.098394] Call Trace:
> [   28.098394]  dump_stack+0xdb/0x120
> [   28.098394]  bad_page.cold.108+0xc6/0xcb
> [   28.098394]  check_new_page_bad+0x47/0xa0
> [   28.098394]  get_page_from_freelist+0x30cd/0x5730
> [   28.098394]  ? __isolate_free_page+0x4f0/0x4f0
> [   28.098394]  ? init_object+0x7e/0x90
> [   28.098394]  __alloc_pages_nodemask+0x2d8/0x650
> [   28.098394]  ? write_comp_data+0x2f/0x90
> [   28.098394]  ? __alloc_pages_slowpath.constprop.103+0x2110/0x2110
> [   28.098394]  ? __sanitizer_cov_trace_pc+0x21/0x50
> [   28.098394]  alloc_pages_vma+0xe2/0x560
> [   28.098394]  do_fault+0x194/0x12c0
> [   28.098394]  ? write_comp_data+0x2f/0x90
> [   28.098394]  __handle_mm_fault+0x1650/0x26c0
> [   28.098394]  ? copy_page_range+0x1350/0x1350
> [   28.098394]  ? write_comp_data+0x2f/0x90
> [   28.098394]  ? write_comp_data+0x2f/0x90
> [   28.098394]  handle_mm_fault+0x1f9/0x810
> [   28.098394]  ? write_comp_data+0x2f/0x90
> [   28.098394]  do_user_addr_fault+0x6f7/0xca0
> [   28.098394]  exc_page_fault+0xaf/0x1a0
> [   28.098394]  asm_exc_page_fault+0x1e/0x30
> [   28.098394] RIP: 0010:__clear_user+0x30/0x60

I think the PAGE_FLAGS_CHECK_AT_PREP check in this instance means that
someone is trying to allocate that page with the PG_reserved bit set.
This means that the page actually was exposed to the buddy.

However, when you SetPageReserved(), I don't think that PG_buddy is set
and the refcount is 0. That could indicate that the page is on the buddy
PCP list. Could be that it is getting reused a couple of times.

The PFN 0xbe453 looks a little strange, though. Do we expect ACPI tables
close to 3 GiB ? No idea. Could it be that you are trying to map a wrong
table? Just a guess.

>
> What would be  the correct way to reserve the page so that the above
> would not be hit?

I would have assumed that if this is a binary blob, that someone (which
I think would be acpi code) reserved via memblock_reserve() early during
boot.

E.g., see drivers/acpi/tables.c:acpi_table_upgrade()->memblock_reserve().

--
Thanks,

David / dhildenb

2021-02-22 16:43:26

by David Hildenbrand

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory

On 22.02.21 17:13, David Hildenbrand wrote:
> On 22.02.21 16:13, George Kennedy wrote:
>>
>>
>> On 2/22/2021 4:52 AM, David Hildenbrand wrote:
>>> On 20.02.21 00:04, George Kennedy wrote:
>>>>
>>>>
>>>> On 2/19/2021 11:45 AM, George Kennedy wrote:
>>>>>
>>>>>
>>>>> On 2/18/2021 7:09 PM, Andrey Konovalov wrote:
>>>>>> On Fri, Feb 19, 2021 at 1:06 AM George Kennedy
>>>>>> <[email protected]> wrote:
>>>>>>>
>>>>>>>
>>>>>>> On 2/18/2021 3:55 AM, David Hildenbrand wrote:
>>>>>>>> On 17.02.21 21:56, Andrey Konovalov wrote:
>>>>>>>>> During boot, all non-reserved memblock memory is exposed to the
>>>>>>>>> buddy
>>>>>>>>> allocator. Poisoning all that memory with KASAN lengthens boot
>>>>>>>>> time,
>>>>>>>>> especially on systems with large amount of RAM. This patch makes
>>>>>>>>> page_alloc to not call kasan_free_pages() on all new memory.
>>>>>>>>>
>>>>>>>>> __free_pages_core() is used when exposing fresh memory during
>>>>>>>>> system
>>>>>>>>> boot and when onlining memory during hotplug. This patch adds a new
>>>>>>>>> FPI_SKIP_KASAN_POISON flag and passes it to __free_pages_ok()
>>>>>>>>> through
>>>>>>>>> free_pages_prepare() from __free_pages_core().
>>>>>>>>>
>>>>>>>>> This has little impact on KASAN memory tracking.
>>>>>>>>>
>>>>>>>>> Assuming that there are no references to newly exposed pages
>>>>>>>>> before they
>>>>>>>>> are ever allocated, there won't be any intended (but buggy)
>>>>>>>>> accesses to
>>>>>>>>> that memory that KASAN would normally detect.
>>>>>>>>>
>>>>>>>>> However, with this patch, KASAN stops detecting wild and large
>>>>>>>>> out-of-bounds accesses that happen to land on a fresh memory page
>>>>>>>>> that
>>>>>>>>> was never allocated. This is taken as an acceptable trade-off.
>>>>>>>>>
>>>>>>>>> All memory allocated normally when the boot is over keeps getting
>>>>>>>>> poisoned as usual.
>>>>>>>>>
>>>>>>>>> Signed-off-by: Andrey Konovalov <[email protected]>
>>>>>>>>> Change-Id: Iae6b1e4bb8216955ffc14af255a7eaaa6f35324d
>>>>>>>> Not sure this is the right thing to do, see
>>>>>>>>
>>>>>>>> https://lkml.kernel.org/r/[email protected]
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> Reversing the order in which memory gets allocated + used during
>>>>>>>> boot
>>>>>>>> (in a patch by me) might have revealed an invalid memory access
>>>>>>>> during
>>>>>>>> boot.
>>>>>>>>
>>>>>>>> I suspect that that issue would no longer get detected with your
>>>>>>>> patch, as the invalid memory access would simply not get detected.
>>>>>>>> Now, I cannot prove that :)
>>>>>>> Since David's patch we're having trouble with the iBFT ACPI table,
>>>>>>> which
>>>>>>> is mapped in via kmap() - see acpi_map() in "drivers/acpi/osl.c".
>>>>>>> KASAN
>>>>>>> detects that it is being used after free when ibft_init() accesses
>>>>>>> the
>>>>>>> iBFT table, but as of yet we can't find where it get's freed (we've
>>>>>>> instrumented calls to kunmap()).
>>>>>> Maybe it doesn't get freed, but what you see is a wild or a large
>>>>>> out-of-bounds access. Since KASAN marks all memory as freed during the
>>>>>> memblock->page_alloc transition, such bugs can manifest as
>>>>>> use-after-frees.
>>>>>
>>>>> It gets freed and re-used. By the time the iBFT table is accessed by
>>>>> ibft_init() the page has been over-written.
>>>>>
>>>>> Setting page flags like the following before the call to kmap()
>>>>> prevents the iBFT table page from being freed:
>>>>
>>>> Cleaned up version:
>>>>
>>>> diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c
>>>> index 0418feb..8f0a8e7 100644
>>>> --- a/drivers/acpi/osl.c
>>>> +++ b/drivers/acpi/osl.c
>>>> @@ -287,9 +287,12 @@ static void __iomem *acpi_map(acpi_physical_address
>>>> pg_off, unsigned long pg_sz)
>>>>
>>>>        pfn = pg_off >> PAGE_SHIFT;
>>>>        if (should_use_kmap(pfn)) {
>>>> +        struct page *page = pfn_to_page(pfn);
>>>> +
>>>>            if (pg_sz > PAGE_SIZE)
>>>>                return NULL;
>>>> -        return (void __iomem __force *)kmap(pfn_to_page(pfn));
>>>> +        SetPageReserved(page);
>>>> +        return (void __iomem __force *)kmap(page);
>>>>        } else
>>>>            return acpi_os_ioremap(pg_off, pg_sz);
>>>>    }
>>>> @@ -299,9 +302,12 @@ static void acpi_unmap(acpi_physical_address
>>>> pg_off, void __iomem *vaddr)
>>>>        unsigned long pfn;
>>>>
>>>>        pfn = pg_off >> PAGE_SHIFT;
>>>> -    if (should_use_kmap(pfn))
>>>> -        kunmap(pfn_to_page(pfn));
>>>> -    else
>>>> +    if (should_use_kmap(pfn)) {
>>>> +        struct page *page = pfn_to_page(pfn);
>>>> +
>>>> +        ClearPageReserved(page);
>>>> +        kunmap(page);
>>>> +    } else
>>>>            iounmap(vaddr);
>>>>    }
>>>>
>>>> David, the above works, but wondering why it is now necessary. kunmap()
>>>> is not hit. What other ways could a page mapped via kmap() be unmapped?
>>>>
>>>
>>> Let me look into the code ... I have little experience with ACPI
>>> details, so bear with me.
>>>
>>> I assume that acpi_map()/acpi_unmap() map some firmware blob that is
>>> provided via firmware/bios/... to us.
>>>
>>> should_use_kmap() tells us whether
>>> a) we have a "struct page" and should kmap() that one
>>> b) we don't have a "struct page" and should ioremap.
>>>
>>> As it is a blob, the firmware should always reserve that memory region
>>> via memblock (e.g., memblock_reserve()), such that we either
>>> 1) don't create a memmap ("struct page") at all (-> case b) )
>>> 2) if we have to create e memmap, we mark the page PG_reserved and
>>>    *never* expose it to the buddy (-> case a) )
>>>
>>>
>>> Are you telling me that in this case we might have a memmap for the HW
>>> blob that is *not* PG_reserved? In that case it most probably got
>>> exposed to the buddy where it can happily get allocated/freed.
>>>
>>> The latent BUG would be that that blob gets exposed to the system like
>>> ordinary RAM, and not reserved via memblock early during boot.
>>> Assuming that blob has a low physical address, with my patch it will
>>> get allocated/used a lot earlier - which would mean we trigger this
>>> latent BUG now more easily.
>>>
>>> There have been similar latent BUGs on ARM boards that my patch
>>> discovered where special RAM regions did not get marked as reserved
>>> via the device tree properly.
>>>
>>> Now, this is just a wild guess :) Can you dump the page when mapping
>>> (before PageReserved()) and when unmapping, to see what the state of
>>> that memmap is?
>>
>> Thank you David for the explanation and your help on this,
>>
>> dump_page() before PageReserved and before kmap() in the above patch:
>>
>> [    1.116480] ACPI: Core revision 20201113
>> [    1.117628] XXX acpi_map: about to call kmap()...
>> [    1.118561] page:ffffea0002f914c0 refcount:0 mapcount:0
>> mapping:0000000000000000 index:0x0 pfn:0xbe453
>> [    1.120381] flags: 0xfffffc0000000()
>> [    1.121116] raw: 000fffffc0000000 ffffea0002f914c8 ffffea0002f914c8
>> 0000000000000000
>> [    1.122638] raw: 0000000000000000 0000000000000000 00000000ffffffff
>> 0000000000000000
>> [    1.124146] page dumped because: acpi_map pre SetPageReserved
>>
>> I also added dump_page() before unmapping, but it is not hit. The
>> following for the same pfn now shows up I believe as a result of setting
>> PageReserved:
>>
>> [   28.098208] BUG:Bad page state in process mo dprobe  pfn:be453
>> [   28.098394] page:ffffea0002f914c0 refcount:0 mapcount:0
>> mapping:0000000000000000 index:0x1 pfn:0xbe453
>> [   28.098394] flags: 0xfffffc0001000(reserved)
>> [   28.098394] raw: 000fffffc0001000 dead000000000100 dead000000000122
>> 0000000000000000
>> [   28.098394] raw: 0000000000000001 0000000000000000 00000000ffffffff
>> 0000000000000000
>> [   28.098394] page dumped because: PAGE_FLAGS_CHECK_AT_PREP flag(s) set
>> [   28.098394] page_owner info is not present (never set?)
>> [   28.098394] Modules linked in:
>> [   28.098394] CPU: 2 PID: 204 Comm: modprobe Not tainted 5.11.0-3dbd5e3 #66
>> [   28.098394] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
>> BIOS 0.0.0 02/06/2015
>> [   28.098394] Call Trace:
>> [   28.098394]  dump_stack+0xdb/0x120
>> [   28.098394]  bad_page.cold.108+0xc6/0xcb
>> [   28.098394]  check_new_page_bad+0x47/0xa0
>> [   28.098394]  get_page_from_freelist+0x30cd/0x5730
>> [   28.098394]  ? __isolate_free_page+0x4f0/0x4f0
>> [   28.098394]  ? init_object+0x7e/0x90
>> [   28.098394]  __alloc_pages_nodemask+0x2d8/0x650
>> [   28.098394]  ? write_comp_data+0x2f/0x90
>> [   28.098394]  ? __alloc_pages_slowpath.constprop.103+0x2110/0x2110
>> [   28.098394]  ? __sanitizer_cov_trace_pc+0x21/0x50
>> [   28.098394]  alloc_pages_vma+0xe2/0x560
>> [   28.098394]  do_fault+0x194/0x12c0
>> [   28.098394]  ? write_comp_data+0x2f/0x90
>> [   28.098394]  __handle_mm_fault+0x1650/0x26c0
>> [   28.098394]  ? copy_page_range+0x1350/0x1350
>> [   28.098394]  ? write_comp_data+0x2f/0x90
>> [   28.098394]  ? write_comp_data+0x2f/0x90
>> [   28.098394]  handle_mm_fault+0x1f9/0x810
>> [   28.098394]  ? write_comp_data+0x2f/0x90
>> [   28.098394]  do_user_addr_fault+0x6f7/0xca0
>> [   28.098394]  exc_page_fault+0xaf/0x1a0
>> [   28.098394]  asm_exc_page_fault+0x1e/0x30
>> [   28.098394] RIP: 0010:__clear_user+0x30/0x60
>
> I think the PAGE_FLAGS_CHECK_AT_PREP check in this instance means that
> someone is trying to allocate that page with the PG_reserved bit set.
> This means that the page actually was exposed to the buddy.
>
> However, when you SetPageReserved(), I don't think that PG_buddy is set
> and the refcount is 0. That could indicate that the page is on the buddy
> PCP list. Could be that it is getting reused a couple of times.
>
> The PFN 0xbe453 looks a little strange, though. Do we expect ACPI tables
> close to 3 GiB ? No idea. Could it be that you are trying to map a wrong
> table? Just a guess.

... but I assume ibft_check_device() would bail out on an invalid
checksum. So the question is, why is this page not properly marked as
reserved already.

--
Thanks,

David / dhildenb

2021-02-22 17:43:41

by Konrad Rzeszutek Wilk

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory

On Mon, Feb 22, 2021 at 05:39:29PM +0100, David Hildenbrand wrote:
> On 22.02.21 17:13, David Hildenbrand wrote:
> > On 22.02.21 16:13, George Kennedy wrote:
> > >
> > >
> > > On 2/22/2021 4:52 AM, David Hildenbrand wrote:
> > > > On 20.02.21 00:04, George Kennedy wrote:
> > > > >
> > > > >
> > > > > On 2/19/2021 11:45 AM, George Kennedy wrote:
> > > > > >
> > > > > >
> > > > > > On 2/18/2021 7:09 PM, Andrey Konovalov wrote:
> > > > > > > On Fri, Feb 19, 2021 at 1:06 AM George Kennedy
> > > > > > > <[email protected]> wrote:
> > > > > > > >
> > > > > > > >
> > > > > > > > On 2/18/2021 3:55 AM, David Hildenbrand wrote:
> > > > > > > > > On 17.02.21 21:56, Andrey Konovalov wrote:
> > > > > > > > > > During boot, all non-reserved memblock memory is exposed to the
> > > > > > > > > > buddy
> > > > > > > > > > allocator. Poisoning all that memory with KASAN lengthens boot
> > > > > > > > > > time,
> > > > > > > > > > especially on systems with large amount of RAM. This patch makes
> > > > > > > > > > page_alloc to not call kasan_free_pages() on all new memory.
> > > > > > > > > >
> > > > > > > > > > __free_pages_core() is used when exposing fresh memory during
> > > > > > > > > > system
> > > > > > > > > > boot and when onlining memory during hotplug. This patch adds a new
> > > > > > > > > > FPI_SKIP_KASAN_POISON flag and passes it to __free_pages_ok()
> > > > > > > > > > through
> > > > > > > > > > free_pages_prepare() from __free_pages_core().
> > > > > > > > > >
> > > > > > > > > > This has little impact on KASAN memory tracking.
> > > > > > > > > >
> > > > > > > > > > Assuming that there are no references to newly exposed pages
> > > > > > > > > > before they
> > > > > > > > > > are ever allocated, there won't be any intended (but buggy)
> > > > > > > > > > accesses to
> > > > > > > > > > that memory that KASAN would normally detect.
> > > > > > > > > >
> > > > > > > > > > However, with this patch, KASAN stops detecting wild and large
> > > > > > > > > > out-of-bounds accesses that happen to land on a fresh memory page
> > > > > > > > > > that
> > > > > > > > > > was never allocated. This is taken as an acceptable trade-off.
> > > > > > > > > >
> > > > > > > > > > All memory allocated normally when the boot is over keeps getting
> > > > > > > > > > poisoned as usual.
> > > > > > > > > >
> > > > > > > > > > Signed-off-by: Andrey Konovalov <[email protected]>
> > > > > > > > > > Change-Id: Iae6b1e4bb8216955ffc14af255a7eaaa6f35324d
> > > > > > > > > Not sure this is the right thing to do, see
> > > > > > > > >
> > > > > > > > > https://lkml.kernel.org/r/[email protected]
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > Reversing the order in which memory gets allocated + used during
> > > > > > > > > boot
> > > > > > > > > (in a patch by me) might have revealed an invalid memory access
> > > > > > > > > during
> > > > > > > > > boot.
> > > > > > > > >
> > > > > > > > > I suspect that that issue would no longer get detected with your
> > > > > > > > > patch, as the invalid memory access would simply not get detected.
> > > > > > > > > Now, I cannot prove that :)
> > > > > > > > Since David's patch we're having trouble with the iBFT ACPI table,
> > > > > > > > which
> > > > > > > > is mapped in via kmap() - see acpi_map() in "drivers/acpi/osl.c".
> > > > > > > > KASAN
> > > > > > > > detects that it is being used after free when ibft_init() accesses
> > > > > > > > the
> > > > > > > > iBFT table, but as of yet we can't find where it get's freed (we've
> > > > > > > > instrumented calls to kunmap()).
> > > > > > > Maybe it doesn't get freed, but what you see is a wild or a large
> > > > > > > out-of-bounds access. Since KASAN marks all memory as freed during the
> > > > > > > memblock->page_alloc transition, such bugs can manifest as
> > > > > > > use-after-frees.
> > > > > >
> > > > > > It gets freed and re-used. By the time the iBFT table is accessed by
> > > > > > ibft_init() the page has been over-written.
> > > > > >
> > > > > > Setting page flags like the following before the call to kmap()
> > > > > > prevents the iBFT table page from being freed:
> > > > >
> > > > > Cleaned up version:
> > > > >
> > > > > diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c
> > > > > index 0418feb..8f0a8e7 100644
> > > > > --- a/drivers/acpi/osl.c
> > > > > +++ b/drivers/acpi/osl.c
> > > > > @@ -287,9 +287,12 @@ static void __iomem *acpi_map(acpi_physical_address
> > > > > pg_off, unsigned long pg_sz)
> > > > >
> > > > > ? ???? pfn = pg_off >> PAGE_SHIFT;
> > > > > ? ???? if (should_use_kmap(pfn)) {
> > > > > +??? ??? struct page *page = pfn_to_page(pfn);
> > > > > +
> > > > > ? ???? ??? if (pg_sz > PAGE_SIZE)
> > > > > ? ???? ??? ??? return NULL;
> > > > > -??? ??? return (void __iomem __force *)kmap(pfn_to_page(pfn));
> > > > > +??? ??? SetPageReserved(page);
> > > > > +??? ??? return (void __iomem __force *)kmap(page);
> > > > > ? ???? } else
> > > > > ? ???? ??? return acpi_os_ioremap(pg_off, pg_sz);
> > > > > ? ?}
> > > > > @@ -299,9 +302,12 @@ static void acpi_unmap(acpi_physical_address
> > > > > pg_off, void __iomem *vaddr)
> > > > > ? ???? unsigned long pfn;
> > > > >
> > > > > ? ???? pfn = pg_off >> PAGE_SHIFT;
> > > > > -??? if (should_use_kmap(pfn))
> > > > > -??? ??? kunmap(pfn_to_page(pfn));
> > > > > -??? else
> > > > > +??? if (should_use_kmap(pfn)) {
> > > > > +??? ??? struct page *page = pfn_to_page(pfn);
> > > > > +
> > > > > +??? ??? ClearPageReserved(page);
> > > > > +??? ??? kunmap(page);
> > > > > +??? } else
> > > > > ? ???? ??? iounmap(vaddr);
> > > > > ? ?}
> > > > >
> > > > > David, the above works, but wondering why it is now necessary. kunmap()
> > > > > is not hit. What other ways could a page mapped via kmap() be unmapped?
> > > > >
> > > >
> > > > Let me look into the code ... I have little experience with ACPI
> > > > details, so bear with me.
> > > >
> > > > I assume that acpi_map()/acpi_unmap() map some firmware blob that is
> > > > provided via firmware/bios/... to us.
> > > >
> > > > should_use_kmap() tells us whether
> > > > a) we have a "struct page" and should kmap() that one
> > > > b) we don't have a "struct page" and should ioremap.
> > > >
> > > > As it is a blob, the firmware should always reserve that memory region
> > > > via memblock (e.g., memblock_reserve()), such that we either
> > > > 1) don't create a memmap ("struct page") at all (-> case b) )
> > > > 2) if we have to create e memmap, we mark the page PG_reserved and
> > > > ?? *never* expose it to the buddy (-> case a) )
> > > >
> > > >
> > > > Are you telling me that in this case we might have a memmap for the HW
> > > > blob that is *not* PG_reserved? In that case it most probably got
> > > > exposed to the buddy where it can happily get allocated/freed.
> > > >
> > > > The latent BUG would be that that blob gets exposed to the system like
> > > > ordinary RAM, and not reserved via memblock early during boot.
> > > > Assuming that blob has a low physical address, with my patch it will
> > > > get allocated/used a lot earlier - which would mean we trigger this
> > > > latent BUG now more easily.
> > > >
> > > > There have been similar latent BUGs on ARM boards that my patch
> > > > discovered where special RAM regions did not get marked as reserved
> > > > via the device tree properly.
> > > >
> > > > Now, this is just a wild guess :) Can you dump the page when mapping
> > > > (before PageReserved()) and when unmapping, to see what the state of
> > > > that memmap is?
> > >
> > > Thank you David for the explanation and your help on this,
> > >
> > > dump_page() before PageReserved and before kmap() in the above patch:
> > >
> > > [??? 1.116480] ACPI: Core revision 20201113
> > > [??? 1.117628] XXX acpi_map: about to call kmap()...
> > > [??? 1.118561] page:ffffea0002f914c0 refcount:0 mapcount:0
> > > mapping:0000000000000000 index:0x0 pfn:0xbe453
> > > [??? 1.120381] flags: 0xfffffc0000000()
> > > [??? 1.121116] raw: 000fffffc0000000 ffffea0002f914c8 ffffea0002f914c8
> > > 0000000000000000
> > > [??? 1.122638] raw: 0000000000000000 0000000000000000 00000000ffffffff
> > > 0000000000000000
> > > [??? 1.124146] page dumped because: acpi_map pre SetPageReserved
> > >
> > > I also added dump_page() before unmapping, but it is not hit. The
> > > following for the same pfn now shows up I believe as a result of setting
> > > PageReserved:
> > >
> > > [?? 28.098208] BUG:Bad page state in process mo dprobe? pfn:be453
> > > [?? 28.098394] page:ffffea0002f914c0 refcount:0 mapcount:0
> > > mapping:0000000000000000 index:0x1 pfn:0xbe453
> > > [?? 28.098394] flags: 0xfffffc0001000(reserved)
> > > [?? 28.098394] raw: 000fffffc0001000 dead000000000100 dead000000000122
> > > 0000000000000000
> > > [?? 28.098394] raw: 0000000000000001 0000000000000000 00000000ffffffff
> > > 0000000000000000
> > > [?? 28.098394] page dumped because: PAGE_FLAGS_CHECK_AT_PREP flag(s) set
> > > [?? 28.098394] page_owner info is not present (never set?)
> > > [?? 28.098394] Modules linked in:
> > > [?? 28.098394] CPU: 2 PID: 204 Comm: modprobe Not tainted 5.11.0-3dbd5e3 #66
> > > [?? 28.098394] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
> > > BIOS 0.0.0 02/06/2015
> > > [?? 28.098394] Call Trace:
> > > [?? 28.098394]? dump_stack+0xdb/0x120
> > > [?? 28.098394]? bad_page.cold.108+0xc6/0xcb
> > > [?? 28.098394]? check_new_page_bad+0x47/0xa0
> > > [?? 28.098394]? get_page_from_freelist+0x30cd/0x5730
> > > [?? 28.098394]? ? __isolate_free_page+0x4f0/0x4f0
> > > [?? 28.098394]? ? init_object+0x7e/0x90
> > > [?? 28.098394]? __alloc_pages_nodemask+0x2d8/0x650
> > > [?? 28.098394]? ? write_comp_data+0x2f/0x90
> > > [?? 28.098394]? ? __alloc_pages_slowpath.constprop.103+0x2110/0x2110
> > > [?? 28.098394]? ? __sanitizer_cov_trace_pc+0x21/0x50
> > > [?? 28.098394]? alloc_pages_vma+0xe2/0x560
> > > [?? 28.098394]? do_fault+0x194/0x12c0
> > > [?? 28.098394]? ? write_comp_data+0x2f/0x90
> > > [?? 28.098394]? __handle_mm_fault+0x1650/0x26c0
> > > [?? 28.098394]? ? copy_page_range+0x1350/0x1350
> > > [?? 28.098394]? ? write_comp_data+0x2f/0x90
> > > [?? 28.098394]? ? write_comp_data+0x2f/0x90
> > > [?? 28.098394]? handle_mm_fault+0x1f9/0x810
> > > [?? 28.098394]? ? write_comp_data+0x2f/0x90
> > > [?? 28.098394]? do_user_addr_fault+0x6f7/0xca0
> > > [?? 28.098394]? exc_page_fault+0xaf/0x1a0
> > > [?? 28.098394]? asm_exc_page_fault+0x1e/0x30
> > > [?? 28.098394] RIP: 0010:__clear_user+0x30/0x60
> >
> > I think the PAGE_FLAGS_CHECK_AT_PREP check in this instance means that
> > someone is trying to allocate that page with the PG_reserved bit set.
> > This means that the page actually was exposed to the buddy.
> >
> > However, when you SetPageReserved(), I don't think that PG_buddy is set
> > and the refcount is 0. That could indicate that the page is on the buddy
> > PCP list. Could be that it is getting reused a couple of times.
> >
> > The PFN 0xbe453 looks a little strange, though. Do we expect ACPI tables
> > close to 3 GiB ? No idea. Could it be that you are trying to map a wrong
> > table? Just a guess.

Nah, ACPI MADT enumerates the table and that is the proper location of it.
>
> ... but I assume ibft_check_device() would bail out on an invalid checksum.
> So the question is, why is this page not properly marked as reserved
> already.

The ibft_check_device ends up being called as module way way after the
kernel has cleaned the memory.

The funny thing about iBFT is that (it is also mentioned in the spec)
that the table can resize in memory .. or in the ACPI regions (which
have no E820_RAM and are considered "MMIO" regions).

Either place is fine, so it can be in either RAM or MMIO :-(

>
> --
> Thanks,
>
> David / dhildenb
>

2021-02-22 18:48:57

by George Kennedy

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory



On 2/22/2021 11:13 AM, David Hildenbrand wrote:
> On 22.02.21 16:13, George Kennedy wrote:
>>
>>
>> On 2/22/2021 4:52 AM, David Hildenbrand wrote:
>>> On 20.02.21 00:04, George Kennedy wrote:
>>>>
>>>>
>>>> On 2/19/2021 11:45 AM, George Kennedy wrote:
>>>>>
>>>>>
>>>>> On 2/18/2021 7:09 PM, Andrey Konovalov wrote:
>>>>>> On Fri, Feb 19, 2021 at 1:06 AM George Kennedy
>>>>>> <[email protected]> wrote:
>>>>>>>
>>>>>>>
>>>>>>> On 2/18/2021 3:55 AM, David Hildenbrand wrote:
>>>>>>>> On 17.02.21 21:56, Andrey Konovalov wrote:
>>>>>>>>> During boot, all non-reserved memblock memory is exposed to the
>>>>>>>>> buddy
>>>>>>>>> allocator. Poisoning all that memory with KASAN lengthens boot
>>>>>>>>> time,
>>>>>>>>> especially on systems with large amount of RAM. This patch makes
>>>>>>>>> page_alloc to not call kasan_free_pages() on all new memory.
>>>>>>>>>
>>>>>>>>> __free_pages_core() is used when exposing fresh memory during
>>>>>>>>> system
>>>>>>>>> boot and when onlining memory during hotplug. This patch adds
>>>>>>>>> a new
>>>>>>>>> FPI_SKIP_KASAN_POISON flag and passes it to __free_pages_ok()
>>>>>>>>> through
>>>>>>>>> free_pages_prepare() from __free_pages_core().
>>>>>>>>>
>>>>>>>>> This has little impact on KASAN memory tracking.
>>>>>>>>>
>>>>>>>>> Assuming that there are no references to newly exposed pages
>>>>>>>>> before they
>>>>>>>>> are ever allocated, there won't be any intended (but buggy)
>>>>>>>>> accesses to
>>>>>>>>> that memory that KASAN would normally detect.
>>>>>>>>>
>>>>>>>>> However, with this patch, KASAN stops detecting wild and large
>>>>>>>>> out-of-bounds accesses that happen to land on a fresh memory page
>>>>>>>>> that
>>>>>>>>> was never allocated. This is taken as an acceptable trade-off.
>>>>>>>>>
>>>>>>>>> All memory allocated normally when the boot is over keeps getting
>>>>>>>>> poisoned as usual.
>>>>>>>>>
>>>>>>>>> Signed-off-by: Andrey Konovalov <[email protected]>
>>>>>>>>> Change-Id: Iae6b1e4bb8216955ffc14af255a7eaaa6f35324d
>>>>>>>> Not sure this is the right thing to do, see
>>>>>>>>
>>>>>>>> https://lkml.kernel.org/r/[email protected]
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> Reversing the order in which memory gets allocated + used during
>>>>>>>> boot
>>>>>>>> (in a patch by me) might have revealed an invalid memory access
>>>>>>>> during
>>>>>>>> boot.
>>>>>>>>
>>>>>>>> I suspect that that issue would no longer get detected with your
>>>>>>>> patch, as the invalid memory access would simply not get detected.
>>>>>>>> Now, I cannot prove that :)
>>>>>>> Since David's patch we're having trouble with the iBFT ACPI table,
>>>>>>> which
>>>>>>> is mapped in via kmap() - see acpi_map() in "drivers/acpi/osl.c".
>>>>>>> KASAN
>>>>>>> detects that it is being used after free when ibft_init() accesses
>>>>>>> the
>>>>>>> iBFT table, but as of yet we can't find where it get's freed (we've
>>>>>>> instrumented calls to kunmap()).
>>>>>> Maybe it doesn't get freed, but what you see is a wild or a large
>>>>>> out-of-bounds access. Since KASAN marks all memory as freed
>>>>>> during the
>>>>>> memblock->page_alloc transition, such bugs can manifest as
>>>>>> use-after-frees.
>>>>>
>>>>> It gets freed and re-used. By the time the iBFT table is accessed by
>>>>> ibft_init() the page has been over-written.
>>>>>
>>>>> Setting page flags like the following before the call to kmap()
>>>>> prevents the iBFT table page from being freed:
>>>>
>>>> Cleaned up version:
>>>>
>>>> diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c
>>>> index 0418feb..8f0a8e7 100644
>>>> --- a/drivers/acpi/osl.c
>>>> +++ b/drivers/acpi/osl.c
>>>> @@ -287,9 +287,12 @@ static void __iomem
>>>> *acpi_map(acpi_physical_address
>>>> pg_off, unsigned long pg_sz)
>>>>
>>>>         pfn = pg_off >> PAGE_SHIFT;
>>>>         if (should_use_kmap(pfn)) {
>>>> +        struct page *page = pfn_to_page(pfn);
>>>> +
>>>>             if (pg_sz > PAGE_SIZE)
>>>>                 return NULL;
>>>> -        return (void __iomem __force *)kmap(pfn_to_page(pfn));
>>>> +        SetPageReserved(page);
>>>> +        return (void __iomem __force *)kmap(page);
>>>>         } else
>>>>             return acpi_os_ioremap(pg_off, pg_sz);
>>>>     }
>>>> @@ -299,9 +302,12 @@ static void acpi_unmap(acpi_physical_address
>>>> pg_off, void __iomem *vaddr)
>>>>         unsigned long pfn;
>>>>
>>>>         pfn = pg_off >> PAGE_SHIFT;
>>>> -    if (should_use_kmap(pfn))
>>>> -        kunmap(pfn_to_page(pfn));
>>>> -    else
>>>> +    if (should_use_kmap(pfn)) {
>>>> +        struct page *page = pfn_to_page(pfn);
>>>> +
>>>> +        ClearPageReserved(page);
>>>> +        kunmap(page);
>>>> +    } else
>>>>             iounmap(vaddr);
>>>>     }
>>>>
>>>> David, the above works, but wondering why it is now necessary.
>>>> kunmap()
>>>> is not hit. What other ways could a page mapped via kmap() be
>>>> unmapped?
>>>>
>>>
>>> Let me look into the code ... I have little experience with ACPI
>>> details, so bear with me.
>>>
>>> I assume that acpi_map()/acpi_unmap() map some firmware blob that is
>>> provided via firmware/bios/... to us.
>>>
>>> should_use_kmap() tells us whether
>>> a) we have a "struct page" and should kmap() that one
>>> b) we don't have a "struct page" and should ioremap.
>>>
>>> As it is a blob, the firmware should always reserve that memory region
>>> via memblock (e.g., memblock_reserve()), such that we either
>>> 1) don't create a memmap ("struct page") at all (-> case b) )
>>> 2) if we have to create e memmap, we mark the page PG_reserved and
>>>     *never* expose it to the buddy (-> case a) )
>>>
>>>
>>> Are you telling me that in this case we might have a memmap for the HW
>>> blob that is *not* PG_reserved? In that case it most probably got
>>> exposed to the buddy where it can happily get allocated/freed.
>>>
>>> The latent BUG would be that that blob gets exposed to the system like
>>> ordinary RAM, and not reserved via memblock early during boot.
>>> Assuming that blob has a low physical address, with my patch it will
>>> get allocated/used a lot earlier - which would mean we trigger this
>>> latent BUG now more easily.
>>>
>>> There have been similar latent BUGs on ARM boards that my patch
>>> discovered where special RAM regions did not get marked as reserved
>>> via the device tree properly.
>>>
>>> Now, this is just a wild guess :) Can you dump the page when mapping
>>> (before PageReserved()) and when unmapping, to see what the state of
>>> that memmap is?
>>
>> Thank you David for the explanation and your help on this,
>>
>> dump_page() before PageReserved and before kmap() in the above patch:
>>
>> [    1.116480] ACPI: Core revision 20201113
>> [    1.117628] XXX acpi_map: about to call kmap()...
>> [    1.118561] page:ffffea0002f914c0 refcount:0 mapcount:0
>> mapping:0000000000000000 index:0x0 pfn:0xbe453
>> [    1.120381] flags: 0xfffffc0000000()
>> [    1.121116] raw: 000fffffc0000000 ffffea0002f914c8 ffffea0002f914c8
>> 0000000000000000
>> [    1.122638] raw: 0000000000000000 0000000000000000 00000000ffffffff
>> 0000000000000000
>> [    1.124146] page dumped because: acpi_map pre SetPageReserved
>>
>> I also added dump_page() before unmapping, but it is not hit. The
>> following for the same pfn now shows up I believe as a result of setting
>> PageReserved:
>>
>> [   28.098208] BUG:Bad page state in process mo dprobe pfn:be453
>> [   28.098394] page:ffffea0002f914c0 refcount:0 mapcount:0
>> mapping:0000000000000000 index:0x1 pfn:0xbe453
>> [   28.098394] flags: 0xfffffc0001000(reserved)
>> [   28.098394] raw: 000fffffc0001000 dead000000000100 dead000000000122
>> 0000000000000000
>> [   28.098394] raw: 0000000000000001 0000000000000000 00000000ffffffff
>> 0000000000000000
>> [   28.098394] page dumped because: PAGE_FLAGS_CHECK_AT_PREP flag(s) set
>> [   28.098394] page_owner info is not present (never set?)
>> [   28.098394] Modules linked in:
>> [   28.098394] CPU: 2 PID: 204 Comm: modprobe Not tainted
>> 5.11.0-3dbd5e3 #66
>> [   28.098394] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
>> BIOS 0.0.0 02/06/2015
>> [   28.098394] Call Trace:
>> [   28.098394]  dump_stack+0xdb/0x120
>> [   28.098394]  bad_page.cold.108+0xc6/0xcb
>> [   28.098394]  check_new_page_bad+0x47/0xa0
>> [   28.098394]  get_page_from_freelist+0x30cd/0x5730
>> [   28.098394]  ? __isolate_free_page+0x4f0/0x4f0
>> [   28.098394]  ? init_object+0x7e/0x90
>> [   28.098394]  __alloc_pages_nodemask+0x2d8/0x650
>> [   28.098394]  ? write_comp_data+0x2f/0x90
>> [   28.098394]  ? __alloc_pages_slowpath.constprop.103+0x2110/0x2110
>> [   28.098394]  ? __sanitizer_cov_trace_pc+0x21/0x50
>> [   28.098394]  alloc_pages_vma+0xe2/0x560
>> [   28.098394]  do_fault+0x194/0x12c0
>> [   28.098394]  ? write_comp_data+0x2f/0x90
>> [   28.098394]  __handle_mm_fault+0x1650/0x26c0
>> [   28.098394]  ? copy_page_range+0x1350/0x1350
>> [   28.098394]  ? write_comp_data+0x2f/0x90
>> [   28.098394]  ? write_comp_data+0x2f/0x90
>> [   28.098394]  handle_mm_fault+0x1f9/0x810
>> [   28.098394]  ? write_comp_data+0x2f/0x90
>> [   28.098394]  do_user_addr_fault+0x6f7/0xca0
>> [   28.098394]  exc_page_fault+0xaf/0x1a0
>> [   28.098394]  asm_exc_page_fault+0x1e/0x30
>> [   28.098394] RIP: 0010:__clear_user+0x30/0x60
>
> I think the PAGE_FLAGS_CHECK_AT_PREP check in this instance means that
> someone is trying to allocate that page with the PG_reserved bit set.
> This means that the page actually was exposed to the buddy.
>
> However, when you SetPageReserved(), I don't think that PG_buddy is
> set and the refcount is 0. That could indicate that the page is on the
> buddy PCP list. Could be that it is getting reused a couple of times.
>
> The PFN 0xbe453 looks a little strange, though. Do we expect ACPI
> tables close to 3 GiB ? No idea. Could it be that you are trying to
> map a wrong table? Just a guess.
>
>>
>> What would be  the correct way to reserve the page so that the above
>> would not be hit?
>
> I would have assumed that if this is a binary blob, that someone
> (which I think would be acpi code) reserved via memblock_reserve()
> early during boot.
>
> E.g., see drivers/acpi/tables.c:acpi_table_upgrade()->memblock_reserve().

acpi_table_upgrade() gets called, but bails out before
memblock_reserve() is called. Thus, it appears no pages are getting
reserved.

    503 void __init acpi_table_upgrade(void)
    504 {
    505         void *data;
    506         size_t size;
    507         int sig, no, table_nr = 0, total_offset = 0;
    508         long offset = 0;
    509         struct acpi_table_header *table;
    510         char cpio_path[32] = "kernel/firmware/acpi/";
    511         struct cpio_data file;
    512
    513         if
(IS_ENABLED(CONFIG_ACPI_TABLE_OVERRIDE_VIA_BUILTIN_INITRD)) {
    514                 data = __initramfs_start;
    515                 size = __initramfs_size;
    516         } else {
    517                 data = (void *)initrd_start;
    518                 size = initrd_end - initrd_start;
    519         }
    520
    521         if (data == NULL || size == 0)
    522                 return;
    523
    524         for (no = 0; no < NR_ACPI_INITRD_TABLES; no++) {
    525                 file = find_cpio_data(cpio_path, data, size,
&offset);
    526                 if (!file.data)
    527                         break;
...
    563                 all_tables_size += table->length;
    564                 acpi_initrd_files[table_nr].data = file.data;
    565                 acpi_initrd_files[table_nr].size = file.size;
    566                 table_nr++;
    567         }
    568         if (table_nr == 0)
    569                 return;                                 <--
bails out here
"drivers/acpi/tables.c"

George

2021-02-22 18:51:06

by Mike Rapoport

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory

On Mon, Feb 22, 2021 at 12:40:36PM -0500, Konrad Rzeszutek Wilk wrote:
> On Mon, Feb 22, 2021 at 05:39:29PM +0100, David Hildenbrand wrote:
> > On 22.02.21 17:13, David Hildenbrand wrote:
> > > On 22.02.21 16:13, George Kennedy wrote:
> > > >
> > > >
> > > > On 2/22/2021 4:52 AM, David Hildenbrand wrote:
> > > > > On 20.02.21 00:04, George Kennedy wrote:
> > > > > >
> > > > > >
> > > > > > On 2/19/2021 11:45 AM, George Kennedy wrote:
> > > > > > >
> > > > > > >
> > > > > > > On 2/18/2021 7:09 PM, Andrey Konovalov wrote:
> > > > > > > > On Fri, Feb 19, 2021 at 1:06 AM George Kennedy
> > > > > > > > <[email protected]> wrote:
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > On 2/18/2021 3:55 AM, David Hildenbrand wrote:
> > > > > > > > > > On 17.02.21 21:56, Andrey Konovalov wrote:
> > > > > > > > > > > During boot, all non-reserved memblock memory is exposed to the
> > > > > > > > > > > buddy
> > > > > > > > > > > allocator. Poisoning all that memory with KASAN lengthens boot
> > > > > > > > > > > time,
> > > > > > > > > > > especially on systems with large amount of RAM. This patch makes
> > > > > > > > > > > page_alloc to not call kasan_free_pages() on all new memory.
> > > > > > > > > > >
> > > > > > > > > > > __free_pages_core() is used when exposing fresh memory during
> > > > > > > > > > > system
> > > > > > > > > > > boot and when onlining memory during hotplug. This patch adds a new
> > > > > > > > > > > FPI_SKIP_KASAN_POISON flag and passes it to __free_pages_ok()
> > > > > > > > > > > through
> > > > > > > > > > > free_pages_prepare() from __free_pages_core().
> > > > > > > > > > >
> > > > > > > > > > > This has little impact on KASAN memory tracking.
> > > > > > > > > > >
> > > > > > > > > > > Assuming that there are no references to newly exposed pages
> > > > > > > > > > > before they
> > > > > > > > > > > are ever allocated, there won't be any intended (but buggy)
> > > > > > > > > > > accesses to
> > > > > > > > > > > that memory that KASAN would normally detect.
> > > > > > > > > > >
> > > > > > > > > > > However, with this patch, KASAN stops detecting wild and large
> > > > > > > > > > > out-of-bounds accesses that happen to land on a fresh memory page
> > > > > > > > > > > that
> > > > > > > > > > > was never allocated. This is taken as an acceptable trade-off.
> > > > > > > > > > >
> > > > > > > > > > > All memory allocated normally when the boot is over keeps getting
> > > > > > > > > > > poisoned as usual.
> > > > > > > > > > >
> > > > > > > > > > > Signed-off-by: Andrey Konovalov <[email protected]>
> > > > > > > > > > > Change-Id: Iae6b1e4bb8216955ffc14af255a7eaaa6f35324d
> > > > > > > > > > Not sure this is the right thing to do, see
> > > > > > > > > >
> > > > > > > > > > https://lkml.kernel.org/r/[email protected]
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > Reversing the order in which memory gets allocated + used during
> > > > > > > > > > boot
> > > > > > > > > > (in a patch by me) might have revealed an invalid memory access
> > > > > > > > > > during
> > > > > > > > > > boot.
> > > > > > > > > >
> > > > > > > > > > I suspect that that issue would no longer get detected with your
> > > > > > > > > > patch, as the invalid memory access would simply not get detected.
> > > > > > > > > > Now, I cannot prove that :)
> > > > > > > > > Since David's patch we're having trouble with the iBFT ACPI table,
> > > > > > > > > which
> > > > > > > > > is mapped in via kmap() - see acpi_map() in "drivers/acpi/osl.c".
> > > > > > > > > KASAN
> > > > > > > > > detects that it is being used after free when ibft_init() accesses
> > > > > > > > > the
> > > > > > > > > iBFT table, but as of yet we can't find where it get's freed (we've
> > > > > > > > > instrumented calls to kunmap()).
> > > > > > > > Maybe it doesn't get freed, but what you see is a wild or a large
> > > > > > > > out-of-bounds access. Since KASAN marks all memory as freed during the
> > > > > > > > memblock->page_alloc transition, such bugs can manifest as
> > > > > > > > use-after-frees.
> > > > > > >
> > > > > > > It gets freed and re-used. By the time the iBFT table is accessed by
> > > > > > > ibft_init() the page has been over-written.
> > > > > > >
> > > > > > > Setting page flags like the following before the call to kmap()
> > > > > > > prevents the iBFT table page from being freed:
> > > > > >
> > > > > > Cleaned up version:
> > > > > >
> > > > > > diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c
> > > > > > index 0418feb..8f0a8e7 100644
> > > > > > --- a/drivers/acpi/osl.c
> > > > > > +++ b/drivers/acpi/osl.c
> > > > > > @@ -287,9 +287,12 @@ static void __iomem *acpi_map(acpi_physical_address
> > > > > > pg_off, unsigned long pg_sz)
> > > > > >
> > > > > > ? ???? pfn = pg_off >> PAGE_SHIFT;
> > > > > > ? ???? if (should_use_kmap(pfn)) {
> > > > > > +??? ??? struct page *page = pfn_to_page(pfn);
> > > > > > +
> > > > > > ? ???? ??? if (pg_sz > PAGE_SIZE)
> > > > > > ? ???? ??? ??? return NULL;
> > > > > > -??? ??? return (void __iomem __force *)kmap(pfn_to_page(pfn));
> > > > > > +??? ??? SetPageReserved(page);
> > > > > > +??? ??? return (void __iomem __force *)kmap(page);
> > > > > > ? ???? } else
> > > > > > ? ???? ??? return acpi_os_ioremap(pg_off, pg_sz);
> > > > > > ? ?}
> > > > > > @@ -299,9 +302,12 @@ static void acpi_unmap(acpi_physical_address
> > > > > > pg_off, void __iomem *vaddr)
> > > > > > ? ???? unsigned long pfn;
> > > > > >
> > > > > > ? ???? pfn = pg_off >> PAGE_SHIFT;
> > > > > > -??? if (should_use_kmap(pfn))
> > > > > > -??? ??? kunmap(pfn_to_page(pfn));
> > > > > > -??? else
> > > > > > +??? if (should_use_kmap(pfn)) {
> > > > > > +??? ??? struct page *page = pfn_to_page(pfn);
> > > > > > +
> > > > > > +??? ??? ClearPageReserved(page);
> > > > > > +??? ??? kunmap(page);
> > > > > > +??? } else
> > > > > > ? ???? ??? iounmap(vaddr);
> > > > > > ? ?}
> > > > > >
> > > > > > David, the above works, but wondering why it is now necessary. kunmap()
> > > > > > is not hit. What other ways could a page mapped via kmap() be unmapped?
> > > > > >
> > > > >
> > > > > Let me look into the code ... I have little experience with ACPI
> > > > > details, so bear with me.
> > > > >
> > > > > I assume that acpi_map()/acpi_unmap() map some firmware blob that is
> > > > > provided via firmware/bios/... to us.
> > > > >
> > > > > should_use_kmap() tells us whether
> > > > > a) we have a "struct page" and should kmap() that one
> > > > > b) we don't have a "struct page" and should ioremap.
> > > > >
> > > > > As it is a blob, the firmware should always reserve that memory region
> > > > > via memblock (e.g., memblock_reserve()), such that we either
> > > > > 1) don't create a memmap ("struct page") at all (-> case b) )
> > > > > 2) if we have to create e memmap, we mark the page PG_reserved and
> > > > > ?? *never* expose it to the buddy (-> case a) )
> > > > >
> > > > >
> > > > > Are you telling me that in this case we might have a memmap for the HW
> > > > > blob that is *not* PG_reserved? In that case it most probably got
> > > > > exposed to the buddy where it can happily get allocated/freed.
> > > > >
> > > > > The latent BUG would be that that blob gets exposed to the system like
> > > > > ordinary RAM, and not reserved via memblock early during boot.
> > > > > Assuming that blob has a low physical address, with my patch it will
> > > > > get allocated/used a lot earlier - which would mean we trigger this
> > > > > latent BUG now more easily.
> > > > >
> > > > > There have been similar latent BUGs on ARM boards that my patch
> > > > > discovered where special RAM regions did not get marked as reserved
> > > > > via the device tree properly.
> > > > >
> > > > > Now, this is just a wild guess :) Can you dump the page when mapping
> > > > > (before PageReserved()) and when unmapping, to see what the state of
> > > > > that memmap is?
> > > >
> > > > Thank you David for the explanation and your help on this,
> > > >
> > > > dump_page() before PageReserved and before kmap() in the above patch:
> > > >
> > > > [??? 1.116480] ACPI: Core revision 20201113
> > > > [??? 1.117628] XXX acpi_map: about to call kmap()...
> > > > [??? 1.118561] page:ffffea0002f914c0 refcount:0 mapcount:0
> > > > mapping:0000000000000000 index:0x0 pfn:0xbe453
> > > > [??? 1.120381] flags: 0xfffffc0000000()
> > > > [??? 1.121116] raw: 000fffffc0000000 ffffea0002f914c8 ffffea0002f914c8
> > > > 0000000000000000
> > > > [??? 1.122638] raw: 0000000000000000 0000000000000000 00000000ffffffff
> > > > 0000000000000000
> > > > [??? 1.124146] page dumped because: acpi_map pre SetPageReserved
> > > >
> > > > I also added dump_page() before unmapping, but it is not hit. The
> > > > following for the same pfn now shows up I believe as a result of setting
> > > > PageReserved:
> > > >
> > > > [?? 28.098208] BUG:Bad page state in process mo dprobe? pfn:be453
> > > > [?? 28.098394] page:ffffea0002f914c0 refcount:0 mapcount:0
> > > > mapping:0000000000000000 index:0x1 pfn:0xbe453
> > > > [?? 28.098394] flags: 0xfffffc0001000(reserved)
> > > > [?? 28.098394] raw: 000fffffc0001000 dead000000000100 dead000000000122
> > > > 0000000000000000
> > > > [?? 28.098394] raw: 0000000000000001 0000000000000000 00000000ffffffff
> > > > 0000000000000000
> > > > [?? 28.098394] page dumped because: PAGE_FLAGS_CHECK_AT_PREP flag(s) set
> > > > [?? 28.098394] page_owner info is not present (never set?)
> > > > [?? 28.098394] Modules linked in:
> > > > [?? 28.098394] CPU: 2 PID: 204 Comm: modprobe Not tainted 5.11.0-3dbd5e3 #66
> > > > [?? 28.098394] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
> > > > BIOS 0.0.0 02/06/2015
> > > > [?? 28.098394] Call Trace:
> > > > [?? 28.098394]? dump_stack+0xdb/0x120
> > > > [?? 28.098394]? bad_page.cold.108+0xc6/0xcb
> > > > [?? 28.098394]? check_new_page_bad+0x47/0xa0
> > > > [?? 28.098394]? get_page_from_freelist+0x30cd/0x5730
> > > > [?? 28.098394]? ? __isolate_free_page+0x4f0/0x4f0
> > > > [?? 28.098394]? ? init_object+0x7e/0x90
> > > > [?? 28.098394]? __alloc_pages_nodemask+0x2d8/0x650
> > > > [?? 28.098394]? ? write_comp_data+0x2f/0x90
> > > > [?? 28.098394]? ? __alloc_pages_slowpath.constprop.103+0x2110/0x2110
> > > > [?? 28.098394]? ? __sanitizer_cov_trace_pc+0x21/0x50
> > > > [?? 28.098394]? alloc_pages_vma+0xe2/0x560
> > > > [?? 28.098394]? do_fault+0x194/0x12c0
> > > > [?? 28.098394]? ? write_comp_data+0x2f/0x90
> > > > [?? 28.098394]? __handle_mm_fault+0x1650/0x26c0
> > > > [?? 28.098394]? ? copy_page_range+0x1350/0x1350
> > > > [?? 28.098394]? ? write_comp_data+0x2f/0x90
> > > > [?? 28.098394]? ? write_comp_data+0x2f/0x90
> > > > [?? 28.098394]? handle_mm_fault+0x1f9/0x810
> > > > [?? 28.098394]? ? write_comp_data+0x2f/0x90
> > > > [?? 28.098394]? do_user_addr_fault+0x6f7/0xca0
> > > > [?? 28.098394]? exc_page_fault+0xaf/0x1a0
> > > > [?? 28.098394]? asm_exc_page_fault+0x1e/0x30
> > > > [?? 28.098394] RIP: 0010:__clear_user+0x30/0x60
> > >
> > > I think the PAGE_FLAGS_CHECK_AT_PREP check in this instance means that
> > > someone is trying to allocate that page with the PG_reserved bit set.
> > > This means that the page actually was exposed to the buddy.
> > >
> > > However, when you SetPageReserved(), I don't think that PG_buddy is set
> > > and the refcount is 0. That could indicate that the page is on the buddy
> > > PCP list. Could be that it is getting reused a couple of times.
> > >
> > > The PFN 0xbe453 looks a little strange, though. Do we expect ACPI tables
> > > close to 3 GiB ? No idea. Could it be that you are trying to map a wrong
> > > table? Just a guess.
>
> Nah, ACPI MADT enumerates the table and that is the proper location of it.
> >
> > ... but I assume ibft_check_device() would bail out on an invalid checksum.
> > So the question is, why is this page not properly marked as reserved
> > already.
>
> The ibft_check_device ends up being called as module way way after the
> kernel has cleaned the memory.
>
> The funny thing about iBFT is that (it is also mentioned in the spec)
> that the table can resize in memory .. or in the ACPI regions (which

^ reside I presume?

> have no E820_RAM and are considered "MMIO" regions).
>
> Either place is fine, so it can be in either RAM or MMIO :-(

I'd say that the tables in this case are in E820_RAM, because with MMIO we
wouldn't get to kmap() at the first place.
It can be easily confirmed by comparing the problematic address with
/proc/iomem.

Can't say I have a clue about what's going on there, but the theory that
somehow iBFT table does not get PG_Reserved during boot makes sense.

Do you see "iBFT found at 0x<addr>" early in the kernel log?

I don't know if ACPI relocates the tables, but I could not find anywhere
that it reserves the original ones. The memblock_reserve() in
acpi_table_upgrade() is merely a part of open coded memblock allocation.

--
Sincerely yours,
Mike.

2021-02-22 21:58:41

by Mike Rapoport

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory

On Mon, Feb 22, 2021 at 01:42:56PM -0500, George Kennedy wrote:
>
> On 2/22/2021 11:13 AM, David Hildenbrand wrote:
> > On 22.02.21 16:13, George Kennedy wrote:
> > >
> > > On 2/22/2021 4:52 AM, David Hildenbrand wrote:
> > > >
> > > > Let me look into the code ... I have little experience with ACPI
> > > > details, so bear with me.
> > > >
> > > > I assume that acpi_map()/acpi_unmap() map some firmware blob that is
> > > > provided via firmware/bios/... to us.
> > > >
> > > > should_use_kmap() tells us whether
> > > > a) we have a "struct page" and should kmap() that one
> > > > b) we don't have a "struct page" and should ioremap.
> > > >
> > > > As it is a blob, the firmware should always reserve that memory region
> > > > via memblock (e.g., memblock_reserve()), such that we either
> > > > 1) don't create a memmap ("struct page") at all (-> case b) )
> > > > 2) if we have to create e memmap, we mark the page PG_reserved and
> > > > ??? *never* expose it to the buddy (-> case a) )
> > > >
> > > >
> > > > Are you telling me that in this case we might have a memmap for the HW
> > > > blob that is *not* PG_reserved? In that case it most probably got
> > > > exposed to the buddy where it can happily get allocated/freed.
> > > >
> > > > The latent BUG would be that that blob gets exposed to the system like
> > > > ordinary RAM, and not reserved via memblock early during boot.
> > > > Assuming that blob has a low physical address, with my patch it will
> > > > get allocated/used a lot earlier - which would mean we trigger this
> > > > latent BUG now more easily.
> > > >
> > > > There have been similar latent BUGs on ARM boards that my patch
> > > > discovered where special RAM regions did not get marked as reserved
> > > > via the device tree properly.
> > > >
> > > > Now, this is just a wild guess :) Can you dump the page when mapping
> > > > (before PageReserved()) and when unmapping, to see what the state of
> > > > that memmap is?
> > >
> > > Thank you David for the explanation and your help on this,
> > >
> > > dump_page() before PageReserved and before kmap() in the above patch:
> > >
> > > [??? 1.116480] ACPI: Core revision 20201113
> > > [??? 1.117628] XXX acpi_map: about to call kmap()...
> > > [??? 1.118561] page:ffffea0002f914c0 refcount:0 mapcount:0
> > > mapping:0000000000000000 index:0x0 pfn:0xbe453
> > > [??? 1.120381] flags: 0xfffffc0000000()
> > > [??? 1.121116] raw: 000fffffc0000000 ffffea0002f914c8 ffffea0002f914c8
> > > 0000000000000000
> > > [??? 1.122638] raw: 0000000000000000 0000000000000000 00000000ffffffff
> > > 0000000000000000
> > > [??? 1.124146] page dumped because: acpi_map pre SetPageReserved
> > >
> > > I also added dump_page() before unmapping, but it is not hit. The
> > > following for the same pfn now shows up I believe as a result of setting
> > > PageReserved:
> > >
> > > [?? 28.098208] BUG:Bad page state in process mo dprobe pfn:be453
> > > [?? 28.098394] page:ffffea0002f914c0 refcount:0 mapcount:0
> > > mapping:0000000000000000 index:0x1 pfn:0xbe453
> > > [?? 28.098394] flags: 0xfffffc0001000(reserved)
> > > [?? 28.098394] raw: 000fffffc0001000 dead000000000100 dead000000000122
> > > 0000000000000000
> > > [?? 28.098394] raw: 0000000000000001 0000000000000000 00000000ffffffff
> > > 0000000000000000
> > > [?? 28.098394] page dumped because: PAGE_FLAGS_CHECK_AT_PREP flag(s) set
> > > [?? 28.098394] page_owner info is not present (never set?)
> > > [?? 28.098394] Modules linked in:
> > > [?? 28.098394] CPU: 2 PID: 204 Comm: modprobe Not tainted
> > > 5.11.0-3dbd5e3 #66
> > > [?? 28.098394] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
> > > BIOS 0.0.0 02/06/2015
> > > [?? 28.098394] Call Trace:
> > > [?? 28.098394]? dump_stack+0xdb/0x120
> > > [?? 28.098394]? bad_page.cold.108+0xc6/0xcb
> > > [?? 28.098394]? check_new_page_bad+0x47/0xa0
> > > [?? 28.098394]? get_page_from_freelist+0x30cd/0x5730
> > > [?? 28.098394]? ? __isolate_free_page+0x4f0/0x4f0
> > > [?? 28.098394]? ? init_object+0x7e/0x90
> > > [?? 28.098394]? __alloc_pages_nodemask+0x2d8/0x650
> > > [?? 28.098394]? ? write_comp_data+0x2f/0x90
> > > [?? 28.098394]? ? __alloc_pages_slowpath.constprop.103+0x2110/0x2110
> > > [?? 28.098394]? ? __sanitizer_cov_trace_pc+0x21/0x50
> > > [?? 28.098394]? alloc_pages_vma+0xe2/0x560
> > > [?? 28.098394]? do_fault+0x194/0x12c0
> > > [?? 28.098394]? ? write_comp_data+0x2f/0x90
> > > [?? 28.098394]? __handle_mm_fault+0x1650/0x26c0
> > > [?? 28.098394]? ? copy_page_range+0x1350/0x1350
> > > [?? 28.098394]? ? write_comp_data+0x2f/0x90
> > > [?? 28.098394]? ? write_comp_data+0x2f/0x90
> > > [?? 28.098394]? handle_mm_fault+0x1f9/0x810
> > > [?? 28.098394]? ? write_comp_data+0x2f/0x90
> > > [?? 28.098394]? do_user_addr_fault+0x6f7/0xca0
> > > [?? 28.098394]? exc_page_fault+0xaf/0x1a0
> > > [?? 28.098394]? asm_exc_page_fault+0x1e/0x30
> > > [?? 28.098394] RIP: 0010:__clear_user+0x30/0x60
> >
> > I think the PAGE_FLAGS_CHECK_AT_PREP check in this instance means that
> > someone is trying to allocate that page with the PG_reserved bit set.
> > This means that the page actually was exposed to the buddy.
> >
> > However, when you SetPageReserved(), I don't think that PG_buddy is set
> > and the refcount is 0. That could indicate that the page is on the buddy
> > PCP list. Could be that it is getting reused a couple of times.
> >
> > The PFN 0xbe453 looks a little strange, though. Do we expect ACPI tables
> > close to 3 GiB ? No idea. Could it be that you are trying to map a wrong
> > table? Just a guess.
> >
> > >
> > > What would be? the correct way to reserve the page so that the above
> > > would not be hit?
> >
> > I would have assumed that if this is a binary blob, that someone (which
> > I think would be acpi code) reserved via memblock_reserve() early during
> > boot.
> >
> > E.g., see drivers/acpi/tables.c:acpi_table_upgrade()->memblock_reserve().
>
> acpi_table_upgrade() gets called, but bails out before memblock_reserve() is
> called. Thus, it appears no pages are getting reserved.

acpi_table_upgrade() does not actually reserve memory but rather open
codes memblock allocation with memblock_find_in_range() +
memblock_reserve(), so it does not seem related anyway.

Do you have by chance a full boot log handy?

> ??? 503 void __init acpi_table_upgrade(void)
> ??? 504 {

...

> ??? 568???????? if (table_nr == 0)
> ??? 569???????????????? return; ??? ??? ??? ??? ??? ??? ??? ??? <-- bails
> out here
> "drivers/acpi/tables.c"
>
> George
>

--
Sincerely yours,
Mike.

2021-02-23 10:38:34

by Mike Rapoport

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory

(re-added CC)

On Mon, Feb 22, 2021 at 08:24:59PM -0500, George Kennedy wrote:
>
> On 2/22/2021 4:55 PM, Mike Rapoport wrote:
> > On Mon, Feb 22, 2021 at 01:42:56PM -0500, George Kennedy wrote:
> > > On 2/22/2021 11:13 AM, David Hildenbrand wrote:
> > > > On 22.02.21 16:13, George Kennedy wrote:
> > > >
> > > > The PFN 0xbe453 looks a little strange, though. Do we expect ACPI tables
> > > > close to 3 GiB ? No idea. Could it be that you are trying to map a wrong
> > > > table? Just a guess.
> > > >
> > > > > What would be? the correct way to reserve the page so that the above
> > > > > would not be hit?
> > > > I would have assumed that if this is a binary blob, that someone (which
> > > > I think would be acpi code) reserved via memblock_reserve() early during
> > > > boot.
> > > >
> > > > E.g., see drivers/acpi/tables.c:acpi_table_upgrade()->memblock_reserve().
> > > acpi_table_upgrade() gets called, but bails out before memblock_reserve() is
> > > called. Thus, it appears no pages are getting reserved.
> > acpi_table_upgrade() does not actually reserve memory but rather open
> > codes memblock allocation with memblock_find_in_range() +
> > memblock_reserve(), so it does not seem related anyway.
> >
> > Do you have by chance a full boot log handy?
>
> Hello Mike,
>
> Are you after the console output? See attached.
>
> It includes my patch to set PG_Reserved along with the dump_page() debug
> that David asked for - see: "page:"

So, iBFT is indeed at pfn 0xbe453:

[ 0.077698] ACPI: iBFT 0x00000000BE453000 000800 (v01 BOCHS BXPCFACP 00000000 00000000)

and it's in E820_TYPE_RAM region rather than in ACPI data:

[ 0.000000] BIOS-e820: [mem 0x0000000000810000-0x00000000008fffff] ACPI NVS
[ 0.000000] BIOS-e820: [mem 0x0000000000900000-0x00000000be49afff] usable
[ 0.000000] BIOS-e820: [mem 0x00000000be49b000-0x00000000be49bfff] ACPI data

I could not find anywhere in x86 setup or in ACPI tables parsing the code
that reserves this memory or any other ACPI data for that matter. It could
be that I've missed some copying of the data to statically allocated
initial_tables, but AFAICS any ACPI data that was not marked as such in
e820 tables by BIOS resides in memory that is considered as free.

Can you please check if this hack (entirely untested) changes anything:

diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c
index 7bdc0239a943..c118dd54a747 100644
--- a/arch/x86/kernel/acpi/boot.c
+++ b/arch/x86/kernel/acpi/boot.c
@@ -1551,6 +1551,7 @@ void __init acpi_boot_table_init(void)
if (acpi_disabled)
return;

+#if 0
/*
* Initialize the ACPI boot-time table parser.
*/
@@ -1558,6 +1559,7 @@ void __init acpi_boot_table_init(void)
disable_acpi();
return;
}
+#endif

acpi_table_parse(ACPI_SIG_BOOT, acpi_parse_sbf);

diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
index d883176ef2ce..c8a07a7b9577 100644
--- a/arch/x86/kernel/setup.c
+++ b/arch/x86/kernel/setup.c
@@ -1032,6 +1032,14 @@ void __init setup_arch(char **cmdline_p)
*/
find_smp_config();

+ /*
+ * Initialize the ACPI boot-time table parser.
+ */
+ if (acpi_table_init()) {
+ disable_acpi();
+ return;
+ }
+
reserve_ibft_region();

early_alloc_pgt_buf();
diff --git a/drivers/firmware/iscsi_ibft_find.c b/drivers/firmware/iscsi_ibft_find.c
index 64bb94523281..2e5e04090fe2 100644
--- a/drivers/firmware/iscsi_ibft_find.c
+++ b/drivers/firmware/iscsi_ibft_find.c
@@ -80,6 +80,21 @@ static int __init find_ibft_in_mem(void)
done:
return len;
}
+
+static void __init acpi_find_ibft_region(void)
+{
+ int i;
+ struct acpi_table_header *table = NULL;
+
+ if (acpi_disabled)
+ return;
+
+ for (i = 0; i < ARRAY_SIZE(ibft_signs) && !ibft_addr; i++) {
+ acpi_get_table(ibft_signs[i].sign, 0, &table);
+ ibft_addr = (struct acpi_table_ibft *)table;
+ }
+}
+
/*
* Routine used to find the iSCSI Boot Format Table. The logical
* kernel address is set in the ibft_addr global variable.
@@ -93,6 +108,8 @@ unsigned long __init find_ibft_region(unsigned long *sizep)

if (!efi_enabled(EFI_BOOT))
find_ibft_in_mem();
+ else
+ acpi_find_ibft_region();

if (ibft_addr) {
*sizep = PAGE_ALIGN(ibft_addr->header.length);

> Thank you,
> George

--
Sincerely yours,
Mike.

2021-02-23 20:12:32

by Mike Rapoport

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory

Hi George,

On Tue, Feb 23, 2021 at 09:35:32AM -0500, George Kennedy wrote:
>
> On 2/23/2021 5:33 AM, Mike Rapoport wrote:
> > (re-added CC)
> >
> > On Mon, Feb 22, 2021 at 08:24:59PM -0500, George Kennedy wrote:
> > > On 2/22/2021 4:55 PM, Mike Rapoport wrote:
> > > > On Mon, Feb 22, 2021 at 01:42:56PM -0500, George Kennedy wrote:
> > > > > On 2/22/2021 11:13 AM, David Hildenbrand wrote:
> > > > > > On 22.02.21 16:13, George Kennedy wrote:
> > > > > >
> > > > > > The PFN 0xbe453 looks a little strange, though. Do we expect ACPI tables
> > > > > > close to 3 GiB ? No idea. Could it be that you are trying to map a wrong
> > > > > > table? Just a guess.
> > > > > >
> > > > > > > What would be? the correct way to reserve the page so that the above
> > > > > > > would not be hit?
> > > > > > I would have assumed that if this is a binary blob, that someone (which
> > > > > > I think would be acpi code) reserved via memblock_reserve() early during
> > > > > > boot.
> > > > > >
> > > > > > E.g., see drivers/acpi/tables.c:acpi_table_upgrade()->memblock_reserve().
> > > > > acpi_table_upgrade() gets called, but bails out before memblock_reserve() is
> > > > > called. Thus, it appears no pages are getting reserved.
> > > > acpi_table_upgrade() does not actually reserve memory but rather open
> > > > codes memblock allocation with memblock_find_in_range() +
> > > > memblock_reserve(), so it does not seem related anyway.
> > > >
> > > > Do you have by chance a full boot log handy?
> > > Hello Mike,
> > >
> > > Are you after the console output? See attached.
> > >
> > > It includes my patch to set PG_Reserved along with the dump_page() debug
> > > that David asked for - see: "page:"
> > So, iBFT is indeed at pfn 0xbe453:
> >
> > [ 0.077698] ACPI: iBFT 0x00000000BE453000 000800 (v01 BOCHS BXPCFACP 00000000 00000000)
> > and it's in E820_TYPE_RAM region rather than in ACPI data:
> >
> > [ 0.000000] BIOS-e820: [mem 0x0000000000810000-0x00000000008fffff] ACPI NVS
> > [ 0.000000] BIOS-e820: [mem 0x0000000000900000-0x00000000be49afff] usable
> > [ 0.000000] BIOS-e820: [mem 0x00000000be49b000-0x00000000be49bfff] ACPI data
> >
> > I could not find anywhere in x86 setup or in ACPI tables parsing the code
> > that reserves this memory or any other ACPI data for that matter. It could
> > be that I've missed some copying of the data to statically allocated
> > initial_tables, but AFAICS any ACPI data that was not marked as such in
> > e820 tables by BIOS resides in memory that is considered as free.
> >
>
> Close...
>
> Applied the patch, see "[?? 30.136157] iBFT detected.", but now hit the
> following (missing iounmap()? see full console output attached):
>
> diff --git a/drivers/firmware/iscsi_ibft_find.c
> b/drivers/firmware/iscsi_ibft_find.c
> index 64bb945..2e5e040 100644
> --- a/drivers/firmware/iscsi_ibft_find.c
> +++ b/drivers/firmware/iscsi_ibft_find.c
> @@ -80,6 +80,21 @@ static int __init find_ibft_in_mem(void)
> ?done:
> ??????? return len;
> ?}
> +
> +static void __init acpi_find_ibft_region(void)
> +{
> +?????? int i;
> +?????? struct acpi_table_header *table = NULL;
> +
> +?????? if (acpi_disabled)
> +?????????????? return;
> +
> +?????? for (i = 0; i < ARRAY_SIZE(ibft_signs) && !ibft_addr; i++) {
> +?????????????? acpi_get_table(ibft_signs[i].sign, 0, &table);
> +?????????????? ibft_addr = (struct acpi_table_ibft *)table;

Can you try adding

acpi_put_table(table);

here?

> +?????? }
> +}
> +

--
Sincerely yours,
Mike.

2021-02-23 20:36:11

by George Kennedy

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory



On 2/23/2021 10:47 AM, Mike Rapoport wrote:
> Hi George,
>
> On Tue, Feb 23, 2021 at 09:35:32AM -0500, George Kennedy wrote:
>> On 2/23/2021 5:33 AM, Mike Rapoport wrote:
>>> (re-added CC)
>>>
>>> On Mon, Feb 22, 2021 at 08:24:59PM -0500, George Kennedy wrote:
>>>> On 2/22/2021 4:55 PM, Mike Rapoport wrote:
>>>>> On Mon, Feb 22, 2021 at 01:42:56PM -0500, George Kennedy wrote:
>>>>>> On 2/22/2021 11:13 AM, David Hildenbrand wrote:
>>>>>>> On 22.02.21 16:13, George Kennedy wrote:
>>>>>>>
>>>>>>> The PFN 0xbe453 looks a little strange, though. Do we expect ACPI tables
>>>>>>> close to 3 GiB ? No idea. Could it be that you are trying to map a wrong
>>>>>>> table? Just a guess.
>>>>>>>
>>>>>>>> What would be  the correct way to reserve the page so that the above
>>>>>>>> would not be hit?
>>>>>>> I would have assumed that if this is a binary blob, that someone (which
>>>>>>> I think would be acpi code) reserved via memblock_reserve() early during
>>>>>>> boot.
>>>>>>>
>>>>>>> E.g., see drivers/acpi/tables.c:acpi_table_upgrade()->memblock_reserve().
>>>>>> acpi_table_upgrade() gets called, but bails out before memblock_reserve() is
>>>>>> called. Thus, it appears no pages are getting reserved.
>>>>> acpi_table_upgrade() does not actually reserve memory but rather open
>>>>> codes memblock allocation with memblock_find_in_range() +
>>>>> memblock_reserve(), so it does not seem related anyway.
>>>>>
>>>>> Do you have by chance a full boot log handy?
>>>> Hello Mike,
>>>>
>>>> Are you after the console output? See attached.
>>>>
>>>> It includes my patch to set PG_Reserved along with the dump_page() debug
>>>> that David asked for - see: "page:"
>>> So, iBFT is indeed at pfn 0xbe453:
>>>
>>> [ 0.077698] ACPI: iBFT 0x00000000BE453000 000800 (v01 BOCHS BXPCFACP 00000000 00000000)
>>> and it's in E820_TYPE_RAM region rather than in ACPI data:
>>>
>>> [ 0.000000] BIOS-e820: [mem 0x0000000000810000-0x00000000008fffff] ACPI NVS
>>> [ 0.000000] BIOS-e820: [mem 0x0000000000900000-0x00000000be49afff] usable
>>> [ 0.000000] BIOS-e820: [mem 0x00000000be49b000-0x00000000be49bfff] ACPI data
>>>
>>> I could not find anywhere in x86 setup or in ACPI tables parsing the code
>>> that reserves this memory or any other ACPI data for that matter. It could
>>> be that I've missed some copying of the data to statically allocated
>>> initial_tables, but AFAICS any ACPI data that was not marked as such in
>>> e820 tables by BIOS resides in memory that is considered as free.
>>>
>> Close...
>>
>> Applied the patch, see "[   30.136157] iBFT detected.", but now hit the
>> following (missing iounmap()? see full console output attached):
>>
>> diff --git a/drivers/firmware/iscsi_ibft_find.c
>> b/drivers/firmware/iscsi_ibft_find.c
>> index 64bb945..2e5e040 100644
>> --- a/drivers/firmware/iscsi_ibft_find.c
>> +++ b/drivers/firmware/iscsi_ibft_find.c
>> @@ -80,6 +80,21 @@ static int __init find_ibft_in_mem(void)
>>  done:
>>         return len;
>>  }
>> +
>> +static void __init acpi_find_ibft_region(void)
>> +{
>> +       int i;
>> +       struct acpi_table_header *table = NULL;
>> +
>> +       if (acpi_disabled)
>> +               return;
>> +
>> +       for (i = 0; i < ARRAY_SIZE(ibft_signs) && !ibft_addr; i++) {
>> +               acpi_get_table(ibft_signs[i].sign, 0, &table);
>> +               ibft_addr = (struct acpi_table_ibft *)table;
> Can you try adding
>
> acpi_put_table(table);
>
> here?
Mike,

It now crashes here:

[    0.051019] ACPI: Early table checksum verification disabled
[    0.056721] ACPI: RSDP 0x00000000BFBFA014 000024 (v02 BOCHS )
[    0.057874] ACPI: XSDT 0x00000000BFBF90E8 00004C (v01 BOCHS BXPCFACP
00000001      01000013)
[    0.059590] ACPI: FACP 0x00000000BFBF5000 000074 (v01 BOCHS BXPCFACP
00000001 BXPC 00000001)
[    0.061306] ACPI: DSDT 0x00000000BFBF6000 00238D (v01 BOCHS BXPCDSDT
00000001 BXPC 00000001)
[    0.063006] ACPI: FACS 0x00000000BFBFD000 000040
[    0.063938] ACPI: APIC 0x00000000BFBF4000 000090 (v01 BOCHS BXPCAPIC
00000001 BXPC 00000001)
[    0.065638] ACPI: HPET 0x00000000BFBF3000 000038 (v01 BOCHS BXPCHPET
00000001 BXPC 00000001)
[    0.067335] ACPI: BGRT 0x00000000BE49B000 000038 (v01 INTEL EDK2    
00000002      01000013)
[    0.069030] ACPI: iBFT 0x00000000BE453000 000800 (v01 BOCHS BXPCFACP
00000000      00000000)
[    0.070734] XXX acpi_find_ibft_region:
[    0.071468] XXX iBFT, status=0
[    0.072073] XXX about to call acpi_put_table()...
ibft_addr=ffffffffff240000
[    0.073449] XXX acpi_find_ibft_region(EXIT):
PANIC: early exception 0x0e IP 10:ffffffff9259f439 error 0 cr2
0xffffffffff240004
[    0.075711] CPU: 0 PID: 0 Comm: swapper Not tainted 5.11.0-34a2105 #8
[    0.076983] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
BIOS 0.0.0 02/06/2015
[    0.078579] RIP: 0010:find_ibft_region+0x470/0x577
[    0.079541] Code: f1 40 0f 9e c6 84 c9 0f 95 c1 40 84 ce 75 11 83 e0
07 38 c2 0f 9e c1 84 d2 0f 95 c0 84 c1 74 0a be 04 00 00 00 e8 37 f8 5f
ef <8b> 5b 04 4c 89 fa b8 ff ff 37 00 48 c1 ea 03 48 c1 e0 2a 81 c3 ff
[    0.083207] RSP: 0000:ffffffff8fe07ca8 EFLAGS: 00010046 ORIG_RAX:
0000000000000000
[    0.084709] RAX: 0000000000000000 RBX: ffffffffff240000 RCX:
ffffffff815fcf01
[    0.086109] RDX: dffffc0000000000 RSI: 0000000000000001 RDI:
ffffffffff240004
[    0.087509] RBP: ffffffff8fe07d60 R08: fffffbfff1fc0f21 R09:
fffffbfff1fc0f21
[    0.088911] R10: ffffffff8fe07907 R11: fffffbfff1fc0f20 R12:
ffffffff8fe07d38
[    0.090310] R13: 0000000000000001 R14: 0000000000000001 R15:
ffffffff8fe07e80
[    0.091716] FS:  0000000000000000(0000) GS:ffffffff92409000(0000)
knlGS:0000000000000000
[    0.093304] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[    0.094435] CR2: ffffffffff240004 CR3: 0000000027630000 CR4:
00000000000006a0
[    0.095843] Call Trace:
[    0.096345]  ? acpi_table_init+0x3eb/0x428
[    0.097164]  ? dmi_id_init+0x871/0x871
[    0.097912]  ? early_memunmap+0x22/0x27
[    0.098683]  ? smp_scan_config+0x20e/0x230
[    0.099500]  setup_arch+0xd3e/0x181d
[    0.100221]  ? reserve_standard_io_resources+0x3e/0x3e
[    0.101265]  ? __sanitizer_cov_trace_pc+0x21/0x50
[    0.102203]  ? vprintk_func+0xe9/0x200
[    0.102953]  ? printk+0xac/0xd4
[    0.103589]  ? record_print_text.cold.38+0x16/0x16
[    0.104540]  ? write_comp_data+0x2f/0x90
[    0.105325]  ? __sanitizer_cov_trace_pc+0x21/0x50
[    0.106262]  start_kernel+0x6c/0x474
[    0.106981]  x86_64_start_reservations+0x37/0x39
[    0.107902]  x86_64_start_kernel+0x7b/0x7e
[    0.108722]  secondary_startup_64_no_verify+0xb0/0xbb


Added debug to dump out the ibft_addr:

[[email protected] linux-upwork]# git diff
diff --git a/drivers/firmware/iscsi_ibft_find.c
b/drivers/firmware/iscsi_ibft_find.c
index 2e5e040..a246373 100644
--- a/drivers/firmware/iscsi_ibft_find.c
+++ b/drivers/firmware/iscsi_ibft_find.c
@@ -83,16 +83,22 @@ static int __init find_ibft_in_mem(void)

 static void __init acpi_find_ibft_region(void)
 {
-       int i;
+       int i, status;
        struct acpi_table_header *table = NULL;
-
+printk(KERN_ERR "XXX acpi_find_ibft_region:\n");
        if (acpi_disabled)
                return;

        for (i = 0; i < ARRAY_SIZE(ibft_signs) && !ibft_addr; i++) {
-               acpi_get_table(ibft_signs[i].sign, 0, &table);
-               ibft_addr = (struct acpi_table_ibft *)table;
+               status = acpi_get_table(ibft_signs[i].sign, 0, &table);
+               printk(KERN_ERR "XXX %s, status=%x\n",
ibft_signs[i].sign, status);
+               if (ACPI_SUCCESS(status)) {
+                       ibft_addr = (struct acpi_table_ibft *)table;
+                       printk(KERN_ERR "XXX about to call
acpi_put_table()... ibft_addr=%llx\n", (u64)ibft_addr);
+                       acpi_put_table(table);
+               }
        }
+printk(KERN_ERR "XXX acpi_find_ibft_region(EXIT):\n");
 }

 /*
(END)

George
>
>> +       }
>> +}
>> +

2021-02-23 21:32:05

by George Kennedy

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory



On 2/23/2021 3:09 PM, Mike Rapoport wrote:
> On Tue, Feb 23, 2021 at 01:05:05PM -0500, George Kennedy wrote:
>> On 2/23/2021 10:47 AM, Mike Rapoport wrote:
>>
>> It now crashes here:
>>
>> [    0.051019] ACPI: Early table checksum verification disabled
>> [    0.056721] ACPI: RSDP 0x00000000BFBFA014 000024 (v02 BOCHS )
>> [    0.057874] ACPI: XSDT 0x00000000BFBF90E8 00004C (v01 BOCHS BXPCFACP
>> 00000001      01000013)
>> [    0.059590] ACPI: FACP 0x00000000BFBF5000 000074 (v01 BOCHS BXPCFACP
>> 00000001 BXPC 00000001)
>> [    0.061306] ACPI: DSDT 0x00000000BFBF6000 00238D (v01 BOCHS BXPCDSDT
>> 00000001 BXPC 00000001)
>> [    0.063006] ACPI: FACS 0x00000000BFBFD000 000040
>> [    0.063938] ACPI: APIC 0x00000000BFBF4000 000090 (v01 BOCHS BXPCAPIC
>> 00000001 BXPC 00000001)
>> [    0.065638] ACPI: HPET 0x00000000BFBF3000 000038 (v01 BOCHS BXPCHPET
>> 00000001 BXPC 00000001)
>> [    0.067335] ACPI: BGRT 0x00000000BE49B000 000038 (v01 INTEL EDK2
>> 00000002      01000013)
>> [    0.069030] ACPI: iBFT 0x00000000BE453000 000800 (v01 BOCHS BXPCFACP
>> 00000000      00000000)
>> [    0.070734] XXX acpi_find_ibft_region:
>> [    0.071468] XXX iBFT, status=0
>> [    0.072073] XXX about to call acpi_put_table()...
>> ibft_addr=ffffffffff240000
>> [    0.073449] XXX acpi_find_ibft_region(EXIT):
>> PANIC: early exception 0x0e IP 10:ffffffff9259f439 error 0 cr2
>> 0xffffffffff240004
> Right, I've missed the dereference of the ibft_addr after
> acpi_find_ibft_region().
>
> With this change to iscsi_ibft_find.c instead of the previous one it should
> be better:
>
> diff --git a/drivers/firmware/iscsi_ibft_find.c b/drivers/firmware/iscsi_ibft_find.c
> index 64bb94523281..1be7481d5c69 100644
> --- a/drivers/firmware/iscsi_ibft_find.c
> +++ b/drivers/firmware/iscsi_ibft_find.c
> @@ -80,6 +80,27 @@ static int __init find_ibft_in_mem(void)
> done:
> return len;
> }
> +
> +static void __init acpi_find_ibft_region(unsigned long *sizep)
> +{
> + int i;
> + struct acpi_table_header *table = NULL;
> + acpi_status status;
> +
> + if (acpi_disabled)
> + return;
> +
> + for (i = 0; i < ARRAY_SIZE(ibft_signs) && !ibft_addr; i++) {
> + status = acpi_get_table(ibft_signs[i].sign, 0, &table);
> + if (ACPI_SUCCESS(status)) {
> + ibft_addr = (struct acpi_table_ibft *)table;
> + *sizep = PAGE_ALIGN(ibft_addr->header.length);
> + acpi_put_table(table);
> + break;
> + }
> + }
> +}
> +
> /*
> * Routine used to find the iSCSI Boot Format Table. The logical
> * kernel address is set in the ibft_addr global variable.
> @@ -91,14 +112,16 @@ unsigned long __init find_ibft_region(unsigned long *sizep)
> /* iBFT 1.03 section 1.4.3.1 mandates that UEFI machines will
> * only use ACPI for this */
>
> - if (!efi_enabled(EFI_BOOT))
> + if (!efi_enabled(EFI_BOOT)) {
> find_ibft_in_mem();
> -
> - if (ibft_addr) {
> *sizep = PAGE_ALIGN(ibft_addr->header.length);
> - return (u64)virt_to_phys(ibft_addr);
> + } else {
> + acpi_find_ibft_region(sizep);
> }
>
> + if (ibft_addr)
> + return (u64)virt_to_phys(ibft_addr);
> +
> *sizep = 0;
> return 0;
> }
Mike,

No luck. Back to the original KASAN ibft_init crash.

I ran with only the above patch from you. Was that what you wanted? Your
previous patch had a section defined out by #if 0. Was that supposed to
be in there as well?

If you need the console output let me know. Got bounced because it was
too large.

[   30.124650] iBFT detected.
[   30.125228]
==================================================================
[   30.126201] BUG: KASAN: use-after-free in ibft_init+0x134/0xc33
[   30.126201] Read of size 4 at addr ffff8880be453004 by task swapper/0/1
[   30.126201]
[   30.126201] CPU: 2 PID: 1 Comm: swapper/0 Not tainted 5.11.0-f9593a0 #9
[   30.126201] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
BIOS 0.0.0 02/06/2015
[   30.126201] Call Trace:
[   30.126201]  dump_stack+0xdb/0x120
[   30.126201]  ? ibft_init+0x134/0xc33
[   30.126201]  print_address_description.constprop.7+0x41/0x60
[   30.126201]  ? ibft_init+0x134/0xc33
[   30.126201]  ? ibft_init+0x134/0xc33
[   30.126201]  kasan_report.cold.10+0x78/0xd1
[   30.126201]  ? ibft_init+0x134/0xc33
[   30.126201]  __asan_report_load_n_noabort+0xf/0x20
[   30.126201]  ibft_init+0x134/0xc33
[   30.126201]  ? write_comp_data+0x2f/0x90
[   30.126201]  ? ibft_check_initiator_for+0x159/0x159
[   30.126201]  ? write_comp_data+0x2f/0x90
[   30.126201]  ? ibft_check_initiator_for+0x159/0x159
[   30.126201]  do_one_initcall+0xc4/0x3e0
[   30.126201]  ? perf_trace_initcall_level+0x3e0/0x3e0
[   30.126201]  ? unpoison_range+0x14/0x40
[   30.126201]  ? ____kasan_kmalloc.constprop.5+0x8f/0xc0
[   30.126201]  ? kernel_init_freeable+0x420/0x652
[   30.126201]  ? __kasan_kmalloc+0x9/0x10
[   30.126201]  ? __sanitizer_cov_trace_pc+0x21/0x50
[   30.126201]  kernel_init_freeable+0x596/0x652
[   30.126201]  ? console_on_rootfs+0x7d/0x7d
[   30.126201]  ? __sanitizer_cov_trace_pc+0x21/0x50
[   30.126201]  ? rest_init+0xf0/0xf0
[   30.126201]  kernel_init+0x16/0x1d0
[   30.126201]  ? rest_init+0xf0/0xf0
[   30.126201]  ret_from_fork+0x22/0x30
[   30.126201]
[   30.126201] The buggy address belongs to the page:
[   30.126201] page:0000000091b8f2b4 refcount:0 mapcount:0
mapping:0000000000000000 index:0x1 pfn:0xbe453
[   30.126201] flags: 0xfffffc0000000()
[   30.126201] raw: 000fffffc0000000 ffffea0002fac708 ffffea0002fac748
0000000000000000
[   30.126201] raw: 0000000000000001 0000000000000000 00000000ffffffff
0000000000000000
[   30.126201] page dumped because: kasan: bad access detected
[   30.126201] page_owner tracks the page as freed
[   30.126201] page last allocated via order 0, migratetype Movable,
gfp_mask 0x100dca(GFP_HIGHUSER_MOVABLE|__GFP_ZERO), pid 204, ts 27975563827
[   30.126201]  prep_new_page+0xfb/0x140
[   30.126201]  get_page_from_freelist+0x3503/0x5730
[   30.126201]  __alloc_pages_nodemask+0x2d8/0x650
[   30.126201]  alloc_pages_vma+0xe2/0x560
[   30.126201]  __handle_mm_fault+0x930/0x26c0
[   30.126201]  handle_mm_fault+0x1f9/0x810
[   30.126201]  do_user_addr_fault+0x6f7/0xca0
[   30.126201]  exc_page_fault+0xaf/0x1a0
[   30.126201]  asm_exc_page_fault+0x1e/0x30
[   30.126201] page last free stack trace:
[   30.126201]  free_pcp_prepare+0x122/0x290
[   30.126201]  free_unref_page_list+0xe6/0x490
[   30.126201]  release_pages+0x2ed/0x1270
[   30.126201]  free_pages_and_swap_cache+0x245/0x2e0
[   30.126201]  tlb_flush_mmu+0x11e/0x680
[   30.126201]  tlb_finish_mmu+0xa6/0x3e0
[   30.126201]  exit_mmap+0x2b3/0x540
[   30.126201]  mmput+0x11d/0x450
[   30.126201]  do_exit+0xaa6/0x2d40
[   30.126201]  do_group_exit+0x128/0x340
[   30.126201]  __x64_sys_exit_group+0x43/0x50
[   30.126201]  do_syscall_64+0x37/0x50
[   30.126201]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
[   30.126201]
[   30.126201] Memory state around the buggy address:
[   30.126201]  ffff8880be452f00: ff ff ff ff ff ff ff ff ff ff ff ff ff
ff ff ff
[   30.126201]  ffff8880be452f80: ff ff ff ff ff ff ff ff ff ff ff ff ff
ff ff ff
[   30.126201] >ffff8880be453000: ff ff ff ff ff ff ff ff ff ff ff ff ff
ff ff ff
[   30.126201]                    ^
[   30.126201]  ffff8880be453080: ff ff ff ff ff ff ff ff ff ff ff ff ff
ff ff ff
[   30.126201]  ffff8880be453100: ff ff ff ff ff ff ff ff ff ff ff ff ff
ff ff ff
[   30.126201]
==================================================================


This is all I ran with:

# git diff
diff --git a/drivers/firmware/iscsi_ibft_find.c
b/drivers/firmware/iscsi_ibft_find.c
index 64bb945..1be7481 100644
--- a/drivers/firmware/iscsi_ibft_find.c
+++ b/drivers/firmware/iscsi_ibft_find.c
@@ -80,6 +80,27 @@ static int __init find_ibft_in_mem(void)
 done:
        return len;
 }
+
+static void __init acpi_find_ibft_region(unsigned long *sizep)
+{
+       int i;
+       struct acpi_table_header *table = NULL;
+       acpi_status status;
+
+       if (acpi_disabled)
+               return;
+
+       for (i = 0; i < ARRAY_SIZE(ibft_signs) && !ibft_addr; i++) {
+               status = acpi_get_table(ibft_signs[i].sign, 0, &table);
+               if (ACPI_SUCCESS(status)) {
+                       ibft_addr = (struct acpi_table_ibft *)table;
+                       *sizep = PAGE_ALIGN(ibft_addr->header.length);
+                       acpi_put_table(table);
+                       break;
+               }
+       }
+}
+
 /*
  * Routine used to find the iSCSI Boot Format Table. The logical
  * kernel address is set in the ibft_addr global variable.
@@ -91,14 +112,16 @@ unsigned long __init find_ibft_region(unsigned long
*sizep)
        /* iBFT 1.03 section 1.4.3.1 mandates that UEFI machines will
         * only use ACPI for this */

-       if (!efi_enabled(EFI_BOOT))
+       if (!efi_enabled(EFI_BOOT)) {
                find_ibft_in_mem();
-
-       if (ibft_addr) {
                *sizep = PAGE_ALIGN(ibft_addr->header.length);
-               return (u64)virt_to_phys(ibft_addr);
+       } else {
+               acpi_find_ibft_region(sizep);
        }

+       if (ibft_addr)
+               return (u64)virt_to_phys(ibft_addr);
+
        *sizep = 0;
        return 0;
 }


Thank you,
George
>> [    0.075711] CPU: 0 PID: 0 Comm: swapper Not tainted 5.11.0-34a2105 #8
>> [    0.076983] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
>> 0.0.0 02/06/2015
>> [    0.078579] RIP: 0010:find_ibft_region+0x470/0x577

2021-02-23 21:35:53

by Mike Rapoport

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory

On Tue, Feb 23, 2021 at 04:16:44PM -0500, George Kennedy wrote:
>
>
> On 2/23/2021 3:09 PM, Mike Rapoport wrote:
> > On Tue, Feb 23, 2021 at 01:05:05PM -0500, George Kennedy wrote:
> > > On 2/23/2021 10:47 AM, Mike Rapoport wrote:
> > >
> > > It now crashes here:
> > >
> > > [??? 0.051019] ACPI: Early table checksum verification disabled
> > > [??? 0.056721] ACPI: RSDP 0x00000000BFBFA014 000024 (v02 BOCHS )
> > > [??? 0.057874] ACPI: XSDT 0x00000000BFBF90E8 00004C (v01 BOCHS BXPCFACP
> > > 00000001????? 01000013)
> > > [??? 0.059590] ACPI: FACP 0x00000000BFBF5000 000074 (v01 BOCHS BXPCFACP
> > > 00000001 BXPC 00000001)
> > > [??? 0.061306] ACPI: DSDT 0x00000000BFBF6000 00238D (v01 BOCHS BXPCDSDT
> > > 00000001 BXPC 00000001)
> > > [??? 0.063006] ACPI: FACS 0x00000000BFBFD000 000040
> > > [??? 0.063938] ACPI: APIC 0x00000000BFBF4000 000090 (v01 BOCHS BXPCAPIC
> > > 00000001 BXPC 00000001)
> > > [??? 0.065638] ACPI: HPET 0x00000000BFBF3000 000038 (v01 BOCHS BXPCHPET
> > > 00000001 BXPC 00000001)
> > > [??? 0.067335] ACPI: BGRT 0x00000000BE49B000 000038 (v01 INTEL EDK2
> > > 00000002????? 01000013)
> > > [??? 0.069030] ACPI: iBFT 0x00000000BE453000 000800 (v01 BOCHS BXPCFACP
> > > 00000000????? 00000000)
> > > [??? 0.070734] XXX acpi_find_ibft_region:
> > > [??? 0.071468] XXX iBFT, status=0
> > > [??? 0.072073] XXX about to call acpi_put_table()...
> > > ibft_addr=ffffffffff240000
> > > [??? 0.073449] XXX acpi_find_ibft_region(EXIT):
> > > PANIC: early exception 0x0e IP 10:ffffffff9259f439 error 0 cr2
> > > 0xffffffffff240004
> > Right, I've missed the dereference of the ibft_addr after
> > acpi_find_ibft_region().
> >
> > With this change to iscsi_ibft_find.c instead of the previous one it should
> > be better:
> >
> > diff --git a/drivers/firmware/iscsi_ibft_find.c b/drivers/firmware/iscsi_ibft_find.c
> > index 64bb94523281..1be7481d5c69 100644
> > --- a/drivers/firmware/iscsi_ibft_find.c
> > +++ b/drivers/firmware/iscsi_ibft_find.c
> > @@ -80,6 +80,27 @@ static int __init find_ibft_in_mem(void)
> > done:
> > return len;
> > }
> > +
> > +static void __init acpi_find_ibft_region(unsigned long *sizep)
> > +{
> > + int i;
> > + struct acpi_table_header *table = NULL;
> > + acpi_status status;
> > +
> > + if (acpi_disabled)
> > + return;
> > +
> > + for (i = 0; i < ARRAY_SIZE(ibft_signs) && !ibft_addr; i++) {
> > + status = acpi_get_table(ibft_signs[i].sign, 0, &table);
> > + if (ACPI_SUCCESS(status)) {
> > + ibft_addr = (struct acpi_table_ibft *)table;
> > + *sizep = PAGE_ALIGN(ibft_addr->header.length);
> > + acpi_put_table(table);
> > + break;
> > + }
> > + }
> > +}
> > +
> > /*
> > * Routine used to find the iSCSI Boot Format Table. The logical
> > * kernel address is set in the ibft_addr global variable.
> > @@ -91,14 +112,16 @@ unsigned long __init find_ibft_region(unsigned long *sizep)
> > /* iBFT 1.03 section 1.4.3.1 mandates that UEFI machines will
> > * only use ACPI for this */
> > - if (!efi_enabled(EFI_BOOT))
> > + if (!efi_enabled(EFI_BOOT)) {
> > find_ibft_in_mem();
> > -
> > - if (ibft_addr) {
> > *sizep = PAGE_ALIGN(ibft_addr->header.length);
> > - return (u64)virt_to_phys(ibft_addr);
> > + } else {
> > + acpi_find_ibft_region(sizep);
> > }
> > + if (ibft_addr)
> > + return (u64)virt_to_phys(ibft_addr);
> > +
> > *sizep = 0;
> > return 0;
> > }
> Mike,
>
> No luck. Back to the original KASAN ibft_init crash.
>
> I ran with only the above patch from you. Was that what you wanted? Your
> previous patch had a section defined out by #if 0. Was that supposed to be
> in there as well?

Sorry, I wasn't clear, but I meant to use the first patch and only replace
changes to iscsi_ibft_find.c with the new patch.

Here's the full patch to be sure we're on the same page:

diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c
index 7bdc0239a943..c118dd54a747 100644
--- a/arch/x86/kernel/acpi/boot.c
+++ b/arch/x86/kernel/acpi/boot.c
@@ -1551,6 +1551,7 @@ void __init acpi_boot_table_init(void)
if (acpi_disabled)
return;

+#if 0
/*
* Initialize the ACPI boot-time table parser.
*/
@@ -1558,6 +1559,7 @@ void __init acpi_boot_table_init(void)
disable_acpi();
return;
}
+#endif

acpi_table_parse(ACPI_SIG_BOOT, acpi_parse_sbf);

diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
index d883176ef2ce..c8a07a7b9577 100644
--- a/arch/x86/kernel/setup.c
+++ b/arch/x86/kernel/setup.c
@@ -1032,6 +1032,14 @@ void __init setup_arch(char **cmdline_p)
*/
find_smp_config();

+ /*
+ * Initialize the ACPI boot-time table parser.
+ */
+ if (acpi_table_init()) {
+ disable_acpi();
+ return;
+ }
+
reserve_ibft_region();

early_alloc_pgt_buf();
diff --git a/drivers/firmware/iscsi_ibft_find.c b/drivers/firmware/iscsi_ibft_find.c
index 64bb94523281..1be7481d5c69 100644
--- a/drivers/firmware/iscsi_ibft_find.c
+++ b/drivers/firmware/iscsi_ibft_find.c
@@ -80,6 +80,27 @@ static int __init find_ibft_in_mem(void)
done:
return len;
}
+
+static void __init acpi_find_ibft_region(unsigned long *sizep)
+{
+ int i;
+ struct acpi_table_header *table = NULL;
+ acpi_status status;
+
+ if (acpi_disabled)
+ return;
+
+ for (i = 0; i < ARRAY_SIZE(ibft_signs) && !ibft_addr; i++) {
+ status = acpi_get_table(ibft_signs[i].sign, 0, &table);
+ if (ACPI_SUCCESS(status)) {
+ ibft_addr = (struct acpi_table_ibft *)table;
+ *sizep = PAGE_ALIGN(ibft_addr->header.length);
+ acpi_put_table(table);
+ break;
+ }
+ }
+}
+
/*
* Routine used to find the iSCSI Boot Format Table. The logical
* kernel address is set in the ibft_addr global variable.
@@ -91,14 +112,16 @@ unsigned long __init find_ibft_region(unsigned long *sizep)
/* iBFT 1.03 section 1.4.3.1 mandates that UEFI machines will
* only use ACPI for this */

- if (!efi_enabled(EFI_BOOT))
+ if (!efi_enabled(EFI_BOOT)) {
find_ibft_in_mem();
-
- if (ibft_addr) {
*sizep = PAGE_ALIGN(ibft_addr->header.length);
- return (u64)virt_to_phys(ibft_addr);
+ } else {
+ acpi_find_ibft_region(sizep);
}

+ if (ibft_addr)
+ return (u64)virt_to_phys(ibft_addr);
+
*sizep = 0;
return 0;
}


--
Sincerely yours,
Mike.

2021-02-23 21:53:09

by George Kennedy

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory



On 2/23/2021 4:32 PM, Mike Rapoport wrote:
> diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c
> index 7bdc0239a943..c118dd54a747 100644
> --- a/arch/x86/kernel/acpi/boot.c
> +++ b/arch/x86/kernel/acpi/boot.c
> @@ -1551,6 +1551,7 @@ void __init acpi_boot_table_init(void)
> if (acpi_disabled)
> return;
>
> +#if 0
> /*
> * Initialize the ACPI boot-time table parser.
> */
> @@ -1558,6 +1559,7 @@ void __init acpi_boot_table_init(void)
> disable_acpi();
> return;
> }
> +#endif
>
> acpi_table_parse(ACPI_SIG_BOOT, acpi_parse_sbf);
>
> diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
> index d883176ef2ce..c8a07a7b9577 100644
> --- a/arch/x86/kernel/setup.c
> +++ b/arch/x86/kernel/setup.c
> @@ -1032,6 +1032,14 @@ void __init setup_arch(char **cmdline_p)
> */
> find_smp_config();
>
> + /*
> + * Initialize the ACPI boot-time table parser.
> + */
> + if (acpi_table_init()) {
> + disable_acpi();
> + return;
> + }
> +
> reserve_ibft_region();
>
> early_alloc_pgt_buf();
> diff --git a/drivers/firmware/iscsi_ibft_find.c b/drivers/firmware/iscsi_ibft_find.c
> index 64bb94523281..1be7481d5c69 100644
> --- a/drivers/firmware/iscsi_ibft_find.c
> +++ b/drivers/firmware/iscsi_ibft_find.c
> @@ -80,6 +80,27 @@ static int __init find_ibft_in_mem(void)
> done:
> return len;
> }
> +
> +static void __init acpi_find_ibft_region(unsigned long *sizep)
> +{
> + int i;
> + struct acpi_table_header *table = NULL;
> + acpi_status status;
> +
> + if (acpi_disabled)
> + return;
> +
> + for (i = 0; i < ARRAY_SIZE(ibft_signs) && !ibft_addr; i++) {
> + status = acpi_get_table(ibft_signs[i].sign, 0, &table);
> + if (ACPI_SUCCESS(status)) {
> + ibft_addr = (struct acpi_table_ibft *)table;
> + *sizep = PAGE_ALIGN(ibft_addr->header.length);
> + acpi_put_table(table);
> + break;
> + }
> + }
> +}
> +
> /*
> * Routine used to find the iSCSI Boot Format Table. The logical
> * kernel address is set in the ibft_addr global variable.
> @@ -91,14 +112,16 @@ unsigned long __init find_ibft_region(unsigned long *sizep)
> /* iBFT 1.03 section 1.4.3.1 mandates that UEFI machines will
> * only use ACPI for this */
>
> - if (!efi_enabled(EFI_BOOT))
> + if (!efi_enabled(EFI_BOOT)) {
> find_ibft_in_mem();
> -
> - if (ibft_addr) {
> *sizep = PAGE_ALIGN(ibft_addr->header.length);
> - return (u64)virt_to_phys(ibft_addr);
> + } else {
> + acpi_find_ibft_region(sizep);
> }
>
> + if (ibft_addr)
> + return (u64)virt_to_phys(ibft_addr);
> +
> *sizep = 0;
> return 0;
> }
>
Mike,

Still no luck.

[   30.193723] iscsi: registered transport (iser)
[   30.195970] iBFT detected.
[   30.196571] BUG: unable to handle page fault for address:
ffffffffff240004
[   30.196824] #PF: supervisor read access in kernel mode
[   30.196824] #PF: error_code(0x0000) - not-present page
[   30.196824] PGD 24e34067 P4D 24e34067 PUD 24e36067 PMD 27a0e067 PTE 0
[   30.196824] Oops: 0000 [#1] SMP KASAN PTI
[   30.196824] CPU: 3 PID: 1 Comm: swapper/0 Not tainted 5.11.0-f9593a0 #10
[   30.196824] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
BIOS 0.0.0 02/06/2015
[   30.196824] RIP: 0010:ibft_init+0x13d/0xc33
[   30.196824] Code: c1 40 84 ce 75 11 83 e0 07 38 c2 0f 9e c1 84 d2 0f
95 c0 84 c1 74 0a be 04 00 00 00 e8 77 f2 5f ef 49 8d 7f 08 b8 ff ff 37
00 <4d> 63 6f 04 48 89 fa 48 c1 e0 2a 48 c1 ea 03 8a 04 02 48 89 fa 83
[   30.196824] RSP: 0000:ffff888100fafc30 EFLAGS: 00010246
[   30.196824] RAX: 000000000037ffff RBX: ffffffff937c6fc0 RCX:
ffffffff815fcf01
[   30.196824] RDX: dffffc0000000000 RSI: 0000000000000001 RDI:
ffffffffff240008
[   30.196824] RBP: ffff888100fafcf8 R08: ffffed10201f5f12 R09:
ffffed10201f5f12
[   30.196824] R10: ffff888100faf88f R11: ffffed10201f5f11 R12:
dffffc0000000000
[   30.196824] R13: ffff888100fafdc0 R14: ffff888100fafcd0 R15:
ffffffffff240000
[   30.196824] FS:  0000000000000000(0000) GS:ffff88810ad80000(0000)
knlGS:0000000000000000
[   30.196824] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   30.196824] CR2: ffffffffff240004 CR3: 0000000024e30000 CR4:
00000000000006e0
[   30.196824] DR0: 0000000000000000 DR1: 0000000000000000 DR2:
0000000000000000
[   30.196824] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7:
0000000000000400
[   30.196824] Call Trace:
[   30.196824]  ? write_comp_data+0x2f/0x90
[   30.196824]  ? __sanitizer_cov_trace_pc+0x21/0x50
[   30.196824]  ? ibft_check_initiator_for+0x159/0x159
[   30.196824]  ? dmi_setup+0x46c/0x46c
[   30.196824]  ? write_comp_data+0x2f/0x90
[   30.196824]  ? ibft_check_initiator_for+0x159/0x159
[   30.196824]  do_one_initcall+0xc4/0x3e0
[   30.196824]  ? perf_trace_initcall_level+0x3e0/0x3e0
[   30.196824]  ? asm_sysvec_error_interrupt+0x10/0x20
[   30.196824]  ? do_one_initcall+0x18c/0x3e0
[   30.196824]  kernel_init_freeable+0x596/0x652
[   30.196824]  ? console_on_rootfs+0x7d/0x7d
[   30.196824]  ? __sanitizer_cov_trace_pc+0x21/0x50
[   30.196824]  ? rest_init+0xf0/0xf0
[   30.196824]  kernel_init+0x16/0x1d0
[   30.196824]  ? rest_init+0xf0/0xf0
[   30.196824]  ret_from_fork+0x22/0x30
[   30.196824] Modules linked in:
[   30.196824] Dumping ftrace buffer:
[   30.196824]    (ftrace buffer empty)
[   30.196824] CR2: ffffffffff240004
[   30.196824] ---[ end trace 293eae51adac1398 ]---
[   30.196824] RIP: 0010:ibft_init+0x13d/0xc33
[   30.196824] Code: c1 40 84 ce 75 11 83 e0 07 38 c2 0f 9e c1 84 d2 0f
95 c0 84 c1 74 0a be 04 00 00 00 e8 77 f2 5f ef 49 8d 7f 08 b8 ff ff 37
00 <4d> 63 6f 04 48 89 fa 48 c1 e0 2a 48 c1 ea 03 8a 04 02 48 89 fa 83
[   30.196824] RSP: 0000:ffff888100fafc30 EFLAGS: 00010246
[   30.196824] RAX: 000000000037ffff RBX: ffffffff937c6fc0 RCX:
ffffffff815fcf01
[   30.196824] RDX: dffffc0000000000 RSI: 0000000000000001 RDI:
ffffffffff240008
[   30.196824] RBP: ffff888100fafcf8 R08: ffffed10201f5f12 R09:
ffffed10201f5f12
[   30.196824] R10: ffff888100faf88f R11: ffffed10201f5f11 R12:
dffffc0000000000
[   30.196824] R13: ffff888100fafdc0 R14: ffff888100fafcd0 R15:
ffffffffff240000
[   30.196824] FS:  0000000000000000(0000) GS:ffff88810ad80000(0000)
knlGS:0000000000000000
[   30.196824] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   30.196824] CR2: ffffffffff240004 CR3: 0000000024e30000 CR4:
00000000000006e0
[   30.196824] DR0: 0000000000000000 DR1: 0000000000000000 DR2:
0000000000000000
[   30.196824] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7:
0000000000000400
[   30.196824] Kernel panic - not syncing: Fatal exception
[   30.196824] Dumping ftrace buffer:
[   30.196824]    (ftrace buffer empty)
[   30.196824] Kernel Offset: disabled
[   30.196824] Rebooting in 1 seconds..

George

2021-02-23 23:59:13

by Mike Rapoport

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory

On Tue, Feb 23, 2021 at 01:05:05PM -0500, George Kennedy wrote:
> On 2/23/2021 10:47 AM, Mike Rapoport wrote:
>
> It now crashes here:
>
> [??? 0.051019] ACPI: Early table checksum verification disabled
> [??? 0.056721] ACPI: RSDP 0x00000000BFBFA014 000024 (v02 BOCHS )
> [??? 0.057874] ACPI: XSDT 0x00000000BFBF90E8 00004C (v01 BOCHS BXPCFACP
> 00000001????? 01000013)
> [??? 0.059590] ACPI: FACP 0x00000000BFBF5000 000074 (v01 BOCHS BXPCFACP
> 00000001 BXPC 00000001)
> [??? 0.061306] ACPI: DSDT 0x00000000BFBF6000 00238D (v01 BOCHS BXPCDSDT
> 00000001 BXPC 00000001)
> [??? 0.063006] ACPI: FACS 0x00000000BFBFD000 000040
> [??? 0.063938] ACPI: APIC 0x00000000BFBF4000 000090 (v01 BOCHS BXPCAPIC
> 00000001 BXPC 00000001)
> [??? 0.065638] ACPI: HPET 0x00000000BFBF3000 000038 (v01 BOCHS BXPCHPET
> 00000001 BXPC 00000001)
> [??? 0.067335] ACPI: BGRT 0x00000000BE49B000 000038 (v01 INTEL EDK2????
> 00000002????? 01000013)
> [??? 0.069030] ACPI: iBFT 0x00000000BE453000 000800 (v01 BOCHS BXPCFACP
> 00000000????? 00000000)
> [??? 0.070734] XXX acpi_find_ibft_region:
> [??? 0.071468] XXX iBFT, status=0
> [??? 0.072073] XXX about to call acpi_put_table()...
> ibft_addr=ffffffffff240000
> [??? 0.073449] XXX acpi_find_ibft_region(EXIT):
> PANIC: early exception 0x0e IP 10:ffffffff9259f439 error 0 cr2
> 0xffffffffff240004

Right, I've missed the dereference of the ibft_addr after
acpi_find_ibft_region().

With this change to iscsi_ibft_find.c instead of the previous one it should
be better:

diff --git a/drivers/firmware/iscsi_ibft_find.c b/drivers/firmware/iscsi_ibft_find.c
index 64bb94523281..1be7481d5c69 100644
--- a/drivers/firmware/iscsi_ibft_find.c
+++ b/drivers/firmware/iscsi_ibft_find.c
@@ -80,6 +80,27 @@ static int __init find_ibft_in_mem(void)
done:
return len;
}
+
+static void __init acpi_find_ibft_region(unsigned long *sizep)
+{
+ int i;
+ struct acpi_table_header *table = NULL;
+ acpi_status status;
+
+ if (acpi_disabled)
+ return;
+
+ for (i = 0; i < ARRAY_SIZE(ibft_signs) && !ibft_addr; i++) {
+ status = acpi_get_table(ibft_signs[i].sign, 0, &table);
+ if (ACPI_SUCCESS(status)) {
+ ibft_addr = (struct acpi_table_ibft *)table;
+ *sizep = PAGE_ALIGN(ibft_addr->header.length);
+ acpi_put_table(table);
+ break;
+ }
+ }
+}
+
/*
* Routine used to find the iSCSI Boot Format Table. The logical
* kernel address is set in the ibft_addr global variable.
@@ -91,14 +112,16 @@ unsigned long __init find_ibft_region(unsigned long *sizep)
/* iBFT 1.03 section 1.4.3.1 mandates that UEFI machines will
* only use ACPI for this */

- if (!efi_enabled(EFI_BOOT))
+ if (!efi_enabled(EFI_BOOT)) {
find_ibft_in_mem();
-
- if (ibft_addr) {
*sizep = PAGE_ALIGN(ibft_addr->header.length);
- return (u64)virt_to_phys(ibft_addr);
+ } else {
+ acpi_find_ibft_region(sizep);
}

+ if (ibft_addr)
+ return (u64)virt_to_phys(ibft_addr);
+
*sizep = 0;
return 0;
}

> [??? 0.075711] CPU: 0 PID: 0 Comm: swapper Not tainted 5.11.0-34a2105 #8
> [??? 0.076983] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
> 0.0.0 02/06/2015
> [??? 0.078579] RIP: 0010:find_ibft_region+0x470/0x577

--
Sincerely yours,
Mike.

2021-02-24 00:27:36

by George Kennedy

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory



On 2/23/2021 3:09 PM, Mike Rapoport wrote:
> On Tue, Feb 23, 2021 at 01:05:05PM -0500, George Kennedy wrote:
>> On 2/23/2021 10:47 AM, Mike Rapoport wrote:
>>
>> It now crashes here:
>>
>> [    0.051019] ACPI: Early table checksum verification disabled
>> [    0.056721] ACPI: RSDP 0x00000000BFBFA014 000024 (v02 BOCHS )
>> [    0.057874] ACPI: XSDT 0x00000000BFBF90E8 00004C (v01 BOCHS BXPCFACP
>> 00000001      01000013)
>> [    0.059590] ACPI: FACP 0x00000000BFBF5000 000074 (v01 BOCHS BXPCFACP
>> 00000001 BXPC 00000001)
>> [    0.061306] ACPI: DSDT 0x00000000BFBF6000 00238D (v01 BOCHS BXPCDSDT
>> 00000001 BXPC 00000001)
>> [    0.063006] ACPI: FACS 0x00000000BFBFD000 000040
>> [    0.063938] ACPI: APIC 0x00000000BFBF4000 000090 (v01 BOCHS BXPCAPIC
>> 00000001 BXPC 00000001)
>> [    0.065638] ACPI: HPET 0x00000000BFBF3000 000038 (v01 BOCHS BXPCHPET
>> 00000001 BXPC 00000001)
>> [    0.067335] ACPI: BGRT 0x00000000BE49B000 000038 (v01 INTEL EDK2
>> 00000002      01000013)
>> [    0.069030] ACPI: iBFT 0x00000000BE453000 000800 (v01 BOCHS BXPCFACP
>> 00000000      00000000)
>> [    0.070734] XXX acpi_find_ibft_region:
>> [    0.071468] XXX iBFT, status=0
>> [    0.072073] XXX about to call acpi_put_table()...
>> ibft_addr=ffffffffff240000
>> [    0.073449] XXX acpi_find_ibft_region(EXIT):
>> PANIC: early exception 0x0e IP 10:ffffffff9259f439 error 0 cr2
>> 0xffffffffff240004
> Right, I've missed the dereference of the ibft_addr after
> acpi_find_ibft_region().
>
> With this change to iscsi_ibft_find.c instead of the previous one it should
> be better:
>
> diff --git a/drivers/firmware/iscsi_ibft_find.c b/drivers/firmware/iscsi_ibft_find.c
> index 64bb94523281..1be7481d5c69 100644
> --- a/drivers/firmware/iscsi_ibft_find.c
> +++ b/drivers/firmware/iscsi_ibft_find.c
> @@ -80,6 +80,27 @@ static int __init find_ibft_in_mem(void)
> done:
> return len;
> }
> +
> +static void __init acpi_find_ibft_region(unsigned long *sizep)
> +{
> + int i;
> + struct acpi_table_header *table = NULL;
> + acpi_status status;
> +
> + if (acpi_disabled)
> + return;
> +
> + for (i = 0; i < ARRAY_SIZE(ibft_signs) && !ibft_addr; i++) {
> + status = acpi_get_table(ibft_signs[i].sign, 0, &table);
> + if (ACPI_SUCCESS(status)) {
> + ibft_addr = (struct acpi_table_ibft *)table;
> + *sizep = PAGE_ALIGN(ibft_addr->header.length);
> + acpi_put_table(table);
> + break;
> + }
> + }
> +}
> +
> /*
> * Routine used to find the iSCSI Boot Format Table. The logical
> * kernel address is set in the ibft_addr global variable.
> @@ -91,14 +112,16 @@ unsigned long __init find_ibft_region(unsigned long *sizep)
> /* iBFT 1.03 section 1.4.3.1 mandates that UEFI machines will
> * only use ACPI for this */
>
> - if (!efi_enabled(EFI_BOOT))
> + if (!efi_enabled(EFI_BOOT)) {
> find_ibft_in_mem();
> -
> - if (ibft_addr) {
> *sizep = PAGE_ALIGN(ibft_addr->header.length);
> - return (u64)virt_to_phys(ibft_addr);
> + } else {
> + acpi_find_ibft_region(sizep);
> }
>
> + if (ibft_addr)
> + return (u64)virt_to_phys(ibft_addr);
> +
> *sizep = 0;
> return 0;
> }
Mike,

No luck. Back to the original KASAN ibft_init crash.

I ran with only the above patch from you. Was that what you wanted? Your
previous patch had a section defined out by #if 0. Was that supposed to
be in there as well?

See the attached console output.

This is all I ran with:

# git diff
diff --git a/drivers/firmware/iscsi_ibft_find.c
b/drivers/firmware/iscsi_ibft_find.c
index 64bb945..1be7481 100644
--- a/drivers/firmware/iscsi_ibft_find.c
+++ b/drivers/firmware/iscsi_ibft_find.c
@@ -80,6 +80,27 @@ static int __init find_ibft_in_mem(void)
 done:
        return len;
 }
+
+static void __init acpi_find_ibft_region(unsigned long *sizep)
+{
+       int i;
+       struct acpi_table_header *table = NULL;
+       acpi_status status;
+
+       if (acpi_disabled)
+               return;
+
+       for (i = 0; i < ARRAY_SIZE(ibft_signs) && !ibft_addr; i++) {
+               status = acpi_get_table(ibft_signs[i].sign, 0, &table);
+               if (ACPI_SUCCESS(status)) {
+                       ibft_addr = (struct acpi_table_ibft *)table;
+                       *sizep = PAGE_ALIGN(ibft_addr->header.length);
+                       acpi_put_table(table);
+                       break;
+               }
+       }
+}
+
 /*
  * Routine used to find the iSCSI Boot Format Table. The logical
  * kernel address is set in the ibft_addr global variable.
@@ -91,14 +112,16 @@ unsigned long __init find_ibft_region(unsigned long
*sizep)
        /* iBFT 1.03 section 1.4.3.1 mandates that UEFI machines will
         * only use ACPI for this */

-       if (!efi_enabled(EFI_BOOT))
+       if (!efi_enabled(EFI_BOOT)) {
                find_ibft_in_mem();
-
-       if (ibft_addr) {
                *sizep = PAGE_ALIGN(ibft_addr->header.length);
-               return (u64)virt_to_phys(ibft_addr);
+       } else {
+               acpi_find_ibft_region(sizep);
        }

+       if (ibft_addr)
+               return (u64)virt_to_phys(ibft_addr);
+
        *sizep = 0;
        return 0;
 }


Thank you,
George
>> [    0.075711] CPU: 0 PID: 0 Comm: swapper Not tainted 5.11.0-34a2105 #8
>> [    0.076983] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
>> 0.0.0 02/06/2015
>> [    0.078579] RIP: 0010:find_ibft_region+0x470/0x577


Attachments:
console.out.mike2_patch_fail (99.81 kB)

2021-02-24 10:42:22

by Mike Rapoport

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory

On Tue, Feb 23, 2021 at 04:46:28PM -0500, George Kennedy wrote:
>
> Mike,
>
> Still no luck.
>
> [?? 30.193723] iscsi: registered transport (iser)
> [?? 30.195970] iBFT detected.
> [?? 30.196571] BUG: unable to handle page fault for address: ffffffffff240004

Hmm, we cannot set ibft_addr to early pointer to the ACPI table.
Let's try something more disruptive and move the reservation back to
iscsi_ibft_find.c.

diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c
index 7bdc0239a943..c118dd54a747 100644
--- a/arch/x86/kernel/acpi/boot.c
+++ b/arch/x86/kernel/acpi/boot.c
@@ -1551,6 +1551,7 @@ void __init acpi_boot_table_init(void)
if (acpi_disabled)
return;

+#if 0
/*
* Initialize the ACPI boot-time table parser.
*/
@@ -1558,6 +1559,7 @@ void __init acpi_boot_table_init(void)
disable_acpi();
return;
}
+#endif

acpi_table_parse(ACPI_SIG_BOOT, acpi_parse_sbf);

diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
index d883176ef2ce..c615ce96c9a2 100644
--- a/arch/x86/kernel/setup.c
+++ b/arch/x86/kernel/setup.c
@@ -570,16 +570,6 @@ void __init reserve_standard_io_resources(void)

}

-static __init void reserve_ibft_region(void)
-{
- unsigned long addr, size = 0;
-
- addr = find_ibft_region(&size);
-
- if (size)
- memblock_reserve(addr, size);
-}
-
static bool __init snb_gfx_workaround_needed(void)
{
#ifdef CONFIG_PCI
@@ -1032,6 +1022,12 @@ void __init setup_arch(char **cmdline_p)
*/
find_smp_config();

+ /*
+ * Initialize the ACPI boot-time table parser.
+ */
+ if (acpi_table_init())
+ disable_acpi();
+
reserve_ibft_region();

early_alloc_pgt_buf();
diff --git a/drivers/firmware/iscsi_ibft_find.c b/drivers/firmware/iscsi_ibft_find.c
index 64bb94523281..01be513843d6 100644
--- a/drivers/firmware/iscsi_ibft_find.c
+++ b/drivers/firmware/iscsi_ibft_find.c
@@ -47,7 +47,25 @@ static const struct {
#define VGA_MEM 0xA0000 /* VGA buffer */
#define VGA_SIZE 0x20000 /* 128kB */

-static int __init find_ibft_in_mem(void)
+static void __init *acpi_find_ibft_region(void)
+{
+ int i;
+ struct acpi_table_header *table = NULL;
+ acpi_status status;
+
+ if (acpi_disabled)
+ return NULL;
+
+ for (i = 0; i < ARRAY_SIZE(ibft_signs) && !ibft_addr; i++) {
+ status = acpi_get_table(ibft_signs[i].sign, 0, &table);
+ if (ACPI_SUCCESS(status))
+ return table;
+ }
+
+ return NULL;
+}
+
+static void __init *find_ibft_in_mem(void)
{
unsigned long pos;
unsigned int len = 0;
@@ -70,35 +88,44 @@ static int __init find_ibft_in_mem(void)
/* if the length of the table extends past 1M,
* the table cannot be valid. */
if (pos + len <= (IBFT_END-1)) {
- ibft_addr = (struct acpi_table_ibft *)virt;
pr_info("iBFT found at 0x%lx.\n", pos);
- goto done;
+ return virt;
}
}
}
}
-done:
- return len;
+
+ return NULL;
}
+
+static void __init *find_ibft(void)
+{
+ /* iBFT 1.03 section 1.4.3.1 mandates that UEFI machines will
+ * only use ACPI for this */
+ if (!efi_enabled(EFI_BOOT))
+ return find_ibft_in_mem();
+ else
+ return acpi_find_ibft_region();
+}
+
/*
* Routine used to find the iSCSI Boot Format Table. The logical
* kernel address is set in the ibft_addr global variable.
*/
-unsigned long __init find_ibft_region(unsigned long *sizep)
+void __init reserve_ibft_region(void)
{
- ibft_addr = NULL;
+ struct acpi_table_ibft *table;
+ unsigned long size;

- /* iBFT 1.03 section 1.4.3.1 mandates that UEFI machines will
- * only use ACPI for this */
+ table = find_ibft();
+ if (!table)
+ return;

- if (!efi_enabled(EFI_BOOT))
- find_ibft_in_mem();
-
- if (ibft_addr) {
- *sizep = PAGE_ALIGN(ibft_addr->header.length);
- return (u64)virt_to_phys(ibft_addr);
- }
+ size = PAGE_ALIGN(table->header.length);
+ memblock_reserve(virt_to_phys(table), size);

- *sizep = 0;
- return 0;
+ if (efi_enabled(EFI_BOOT))
+ acpi_put_table(&table->header);
+ else
+ ibft_addr = table;
}
diff --git a/include/linux/iscsi_ibft.h b/include/linux/iscsi_ibft.h
index b7b45ca82bea..da813c891990 100644
--- a/include/linux/iscsi_ibft.h
+++ b/include/linux/iscsi_ibft.h
@@ -26,13 +26,9 @@ extern struct acpi_table_ibft *ibft_addr;
* mapped address is set in the ibft_addr variable.
*/
#ifdef CONFIG_ISCSI_IBFT_FIND
-unsigned long find_ibft_region(unsigned long *sizep);
+void reserve_ibft_region(void);
#else
-static inline unsigned long find_ibft_region(unsigned long *sizep)
-{
- *sizep = 0;
- return 0;
-}
+static inline void reserve_ibft_region(void) {}
#endif

#endif /* ISCSI_IBFT_H */

--
Sincerely yours,
Mike.

2021-02-24 15:17:31

by George Kennedy

[permalink] [raw]
Subject: Re: [PATCH] mm, kasan: don't poison boot memory



On 2/24/2021 5:37 AM, Mike Rapoport wrote:
> On Tue, Feb 23, 2021 at 04:46:28PM -0500, George Kennedy wrote:
>> Mike,
>>
>> Still no luck.
>>
>> [   30.193723] iscsi: registered transport (iser)
>> [   30.195970] iBFT detected.
>> [   30.196571] BUG: unable to handle page fault for address: ffffffffff240004
> Hmm, we cannot set ibft_addr to early pointer to the ACPI table.
> Let's try something more disruptive and move the reservation back to
> iscsi_ibft_find.c.
>
> diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c
> index 7bdc0239a943..c118dd54a747 100644
> --- a/arch/x86/kernel/acpi/boot.c
> +++ b/arch/x86/kernel/acpi/boot.c
> @@ -1551,6 +1551,7 @@ void __init acpi_boot_table_init(void)
> if (acpi_disabled)
> return;
>
> +#if 0
> /*
> * Initialize the ACPI boot-time table parser.
> */
> @@ -1558,6 +1559,7 @@ void __init acpi_boot_table_init(void)
> disable_acpi();
> return;
> }
> +#endif
>
> acpi_table_parse(ACPI_SIG_BOOT, acpi_parse_sbf);
>
> diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
> index d883176ef2ce..c615ce96c9a2 100644
> --- a/arch/x86/kernel/setup.c
> +++ b/arch/x86/kernel/setup.c
> @@ -570,16 +570,6 @@ void __init reserve_standard_io_resources(void)
>
> }
>
> -static __init void reserve_ibft_region(void)
> -{
> - unsigned long addr, size = 0;
> -
> - addr = find_ibft_region(&size);
> -
> - if (size)
> - memblock_reserve(addr, size);
> -}
> -
> static bool __init snb_gfx_workaround_needed(void)
> {
> #ifdef CONFIG_PCI
> @@ -1032,6 +1022,12 @@ void __init setup_arch(char **cmdline_p)
> */
> find_smp_config();
>
> + /*
> + * Initialize the ACPI boot-time table parser.
> + */
> + if (acpi_table_init())
> + disable_acpi();
> +
> reserve_ibft_region();
>
> early_alloc_pgt_buf();
> diff --git a/drivers/firmware/iscsi_ibft_find.c b/drivers/firmware/iscsi_ibft_find.c
> index 64bb94523281..01be513843d6 100644
> --- a/drivers/firmware/iscsi_ibft_find.c
> +++ b/drivers/firmware/iscsi_ibft_find.c
> @@ -47,7 +47,25 @@ static const struct {
> #define VGA_MEM 0xA0000 /* VGA buffer */
> #define VGA_SIZE 0x20000 /* 128kB */
>
> -static int __init find_ibft_in_mem(void)
> +static void __init *acpi_find_ibft_region(void)
> +{
> + int i;
> + struct acpi_table_header *table = NULL;
> + acpi_status status;
> +
> + if (acpi_disabled)
> + return NULL;
> +
> + for (i = 0; i < ARRAY_SIZE(ibft_signs) && !ibft_addr; i++) {
> + status = acpi_get_table(ibft_signs[i].sign, 0, &table);
> + if (ACPI_SUCCESS(status))
> + return table;
> + }
> +
> + return NULL;
> +}
> +
> +static void __init *find_ibft_in_mem(void)
> {
> unsigned long pos;
> unsigned int len = 0;
> @@ -70,35 +88,44 @@ static int __init find_ibft_in_mem(void)
> /* if the length of the table extends past 1M,
> * the table cannot be valid. */
> if (pos + len <= (IBFT_END-1)) {
> - ibft_addr = (struct acpi_table_ibft *)virt;
> pr_info("iBFT found at 0x%lx.\n", pos);
> - goto done;
> + return virt;
> }
> }
> }
> }
> -done:
> - return len;
> +
> + return NULL;
> }
> +
> +static void __init *find_ibft(void)
> +{
> + /* iBFT 1.03 section 1.4.3.1 mandates that UEFI machines will
> + * only use ACPI for this */
> + if (!efi_enabled(EFI_BOOT))
> + return find_ibft_in_mem();
> + else
> + return acpi_find_ibft_region();
> +}
> +
> /*
> * Routine used to find the iSCSI Boot Format Table. The logical
> * kernel address is set in the ibft_addr global variable.
> */
> -unsigned long __init find_ibft_region(unsigned long *sizep)
> +void __init reserve_ibft_region(void)
> {
> - ibft_addr = NULL;
> + struct acpi_table_ibft *table;
> + unsigned long size;
>
> - /* iBFT 1.03 section 1.4.3.1 mandates that UEFI machines will
> - * only use ACPI for this */
> + table = find_ibft();
> + if (!table)
> + return;
>
> - if (!efi_enabled(EFI_BOOT))
> - find_ibft_in_mem();
> -
> - if (ibft_addr) {
> - *sizep = PAGE_ALIGN(ibft_addr->header.length);
> - return (u64)virt_to_phys(ibft_addr);
> - }
> + size = PAGE_ALIGN(table->header.length);
> + memblock_reserve(virt_to_phys(table), size);
>
> - *sizep = 0;
> - return 0;
> + if (efi_enabled(EFI_BOOT))
> + acpi_put_table(&table->header);
> + else
> + ibft_addr = table;
> }
> diff --git a/include/linux/iscsi_ibft.h b/include/linux/iscsi_ibft.h
> index b7b45ca82bea..da813c891990 100644
> --- a/include/linux/iscsi_ibft.h
> +++ b/include/linux/iscsi_ibft.h
> @@ -26,13 +26,9 @@ extern struct acpi_table_ibft *ibft_addr;
> * mapped address is set in the ibft_addr variable.
> */
> #ifdef CONFIG_ISCSI_IBFT_FIND
> -unsigned long find_ibft_region(unsigned long *sizep);
> +void reserve_ibft_region(void);
> #else
> -static inline unsigned long find_ibft_region(unsigned long *sizep)
> -{
> - *sizep = 0;
> - return 0;
> -}
> +static inline void reserve_ibft_region(void) {}
> #endif
>
> #endif /* ISCSI_IBFT_H */

Still no luck Mike,

We're back to the original problem where the only thing that worked was
to run "SetPageReserved(page)" before calling "kmap(page)". The page is
being "freed" before ibft_init() is called as a result of the recent
buddy page freeing changes.

[   30.385207] iscsi: registered transport (iser)
[   30.387462] iBFT detected.
[   30.388042]
==================================================================
[   30.388119] BUG: KASAN: use-after-free in ibft_init+0x134/0xc33
[   30.388119] Read of size 4 at addr ffff8880be453004 by task swapper/0/1
[   30.388119]
[   30.388119] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.11.0-f9593a0 #11
[   30.388119] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
BIOS 0.0.0 02/06/2015
[   30.388119] Call Trace:
[   30.388119]  dump_stack+0xdb/0x120
[   30.388119]  ? ibft_init+0x134/0xc33
[   30.388119]  print_address_description.constprop.7+0x41/0x60
[   30.388119]  ? ibft_init+0x134/0xc33
[   30.388119]  ? ibft_init+0x134/0xc33
[   30.388119]  kasan_report.cold.10+0x78/0xd1
[   30.388119]  ? ibft_init+0x134/0xc33
[   30.388119]  __asan_report_load_n_noabort+0xf/0x20
[   30.388119]  ibft_init+0x134/0xc33
[   30.388119]  ? write_comp_data+0x2f/0x90
[   30.388119]  ? ibft_check_initiator_for+0x159/0x159
[   30.388119]  ? write_comp_data+0x2f/0x90
[   30.388119]  ? ibft_check_initiator_for+0x159/0x159
[   30.388119]  do_one_initcall+0xc4/0x3e0
[   30.388119]  ? perf_trace_initcall_level+0x3e0/0x3e0
[   30.388119]  ? unpoison_range+0x14/0x40
[   30.388119]  ? ____kasan_kmalloc.constprop.5+0x8f/0xc0
[   30.388119]  ? kernel_init_freeable+0x420/0x652

George

>