Commit 5ec8e8ea8b77 ("mm/sparsemem: fix race in accessing
memory_section->usage") changed pfn_section_valid() to add a READ_ONCE()
call around "ms->usage" to fix a race with section_deactivate() where
ms->usage can be cleared.  The READ_ONCE() call, by itself, is not enough
to prevent NULL pointer dereference.  We need to check its value before
dereferencing it.

Fixes: 5ec8e8ea8b77 ("mm/sparsemem: fix race in accessing memory_section->usage")
Signed-off-by: Waiman Long <longman@redhat.com>
---
 include/linux/mmzone.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index 8f9c9590a42c..b1dcf6ddb406 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -1980,8 +1980,9 @@ static inline int subsection_map_index(unsigned long pfn)
 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
 {
 	int idx = subsection_map_index(pfn);
+	struct mem_section_usage *usage = READ_ONCE(ms->usage);
 
-	return test_bit(idx, READ_ONCE(ms->usage)->subsection_map);
+	return usage ? test_bit(idx, usage->subsection_map) : 0;
 }
 #else
 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
-- 
2.39.3

Hi Waiman,

On 6/26/2024 5:46 AM, Waiman Long wrote:
> Commit 5ec8e8ea8b77 ("mm/sparsemem: fix race in accessing
> memory_section->usage") changed pfn_section_valid() to add a READ_ONCE()
> call around "ms->usage" to fix a race with section_deactivate() where
> ms->usage can be cleared.  The READ_ONCE() call, by itself, is not enough
> to prevent NULL pointer dereference.  We need to check its value before
> dereferencing it.

I am unable to see a scenario where ms->usage will be NULL when
pfn_section_valid() is called:

1) In pfn_valid, valid_section() check ensures that pfn_section_valid()
is not called as the section is marked as invalid.

2) In pfn_to_online_page, online_section() check ensures that
pfn_section_valid() is not called.

and in the update path, we do:
     kfree_rcu(ms->usage, rcu);
     WRITE_ONCE(ms->usage, NULL);

Could you help me in understanding about what I am missing here, please?

> 
> Fixes: 5ec8e8ea8b77 ("mm/sparsemem: fix race in accessing memory_section->usage")
> Signed-off-by: Waiman Long <longman@redhat.com>
> ---
>  include/linux/mmzone.h | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
> index 8f9c9590a42c..b1dcf6ddb406 100644
> --- a/include/linux/mmzone.h
> +++ b/include/linux/mmzone.h
> @@ -1980,8 +1980,9 @@ static inline int subsection_map_index(unsigned long pfn)
>  static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
>  {
>  	int idx = subsection_map_index(pfn);
> +	struct mem_section_usage *usage = READ_ONCE(ms->usage);
>  
> -	return test_bit(idx, READ_ONCE(ms->usage)->subsection_map);
> +	return usage ? test_bit(idx, usage->subsection_map) : 0;
>  }
>  #else
>  static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)

On 7/1/24 09:50, Charan Teja Kalla wrote:
> Hi Waiman,
>
> On 6/26/2024 5:46 AM, Waiman Long wrote:
>> Commit 5ec8e8ea8b77 ("mm/sparsemem: fix race in accessing
>> memory_section->usage") changed pfn_section_valid() to add a READ_ONCE()
>> call around "ms->usage" to fix a race with section_deactivate() where
>> ms->usage can be cleared.  The READ_ONCE() call, by itself, is not enough
>> to prevent NULL pointer dereference.  We need to check its value before
>> dereferencing it.
> I am unable to see a scenario where ms->usage will be NULL when
> pfn_section_valid() is called:
>
> 1) In pfn_valid, valid_section() check ensures that pfn_section_valid()
> is not called as the section is marked as invalid.
>
> 2) In pfn_to_online_page, online_section() check ensures that
> pfn_section_valid() is not called.
>
> and in the update path, we do:
>       kfree_rcu(ms->usage, rcu);
>       WRITE_ONCE(ms->usage, NULL);
>
> Could you help me in understanding about what I am missing here, please?
>
With the below timing sequence:

      CPU 0                                      CPU 1
      -----                                      -----
                                         if (!valid_section(ms))
                                             return 0;
  ms->section_mem_map &=
     ~SECTION_HAS_MEM_MAP                <interrupt>

  WRITE_ONCE(ms->usage, NULL);
                                         READ_ONCE(ms->usage)->subsection_map


In the time gap between valid_section() check and accessing ms->usage,
it may have been cleared leading to dereferencing a NULL pointer. That is
why it will be prudent to do a NULL check first.

Regards,
Longman