[PATCH] mm/page_table_check: fix accessing unmapped ptep

[PATCH] mm/page_table_check: fix accessing unmapped ptep

[Miaohe Lin]

ptep is unmapped too early, so ptep will be accessed while it's unmapped.
Fix it by deferring pte_unmap() until page table checking is done.

Fixes: 80110bbfbba6 ("mm/page_table_check: check entries at pmd levels")
Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
---
 mm/page_table_check.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/page_table_check.c b/mm/page_table_check.c
index 3692bea2ea2c..971c3129b0e3 100644
--- a/mm/page_table_check.c
+++ b/mm/page_table_check.c
@@ -234,11 +234,11 @@ void __page_table_check_pte_clear_range(struct mm_struct *mm,
 		pte_t *ptep = pte_offset_map(&pmd, addr);
 		unsigned long i;
 
-		pte_unmap(ptep);
 		for (i = 0; i < PTRS_PER_PTE; i++) {
 			__page_table_check_pte_clear(mm, addr, *ptep);
 			addr += PAGE_SIZE;
 			ptep++;
 		}
+		pte_unmap(ptep);
 	}
 }
-- 
2.23.0

Re: [PATCH] mm/page_table_check: fix accessing unmapped ptep

[Qi Zheng]

On 2022/5/26 7:33 PM, Miaohe Lin wrote:
> ptep is unmapped too early, so ptep will be accessed while it's unmapped.
> Fix it by deferring pte_unmap() until page table checking is done.

In the beginning, page_table_check only supported x86_64, so there
is no problem. But then the commit 3fee229a8eb9 ("riscv/mm: enable
ARCH_SUPPORTS_PAGE_TABLE_CHECK") added support for riscv-32, it is 
indeed a problem in this case.

So:

Reviewed-by: Qi Zheng <zhengqi.arch@bytedance.com>

> 
> Fixes: 80110bbfbba6 ("mm/page_table_check: check entries at pmd levels")
> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
> ---
>   mm/page_table_check.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/mm/page_table_check.c b/mm/page_table_check.c
> index 3692bea2ea2c..971c3129b0e3 100644
> --- a/mm/page_table_check.c
> +++ b/mm/page_table_check.c
> @@ -234,11 +234,11 @@ void __page_table_check_pte_clear_range(struct mm_struct *mm,
>   		pte_t *ptep = pte_offset_map(&pmd, addr);
>   		unsigned long i;
>   
> -		pte_unmap(ptep);
>   		for (i = 0; i < PTRS_PER_PTE; i++) {
>   			__page_table_check_pte_clear(mm, addr, *ptep);
>   			addr += PAGE_SIZE;
>   			ptep++;
>   		}
> +		pte_unmap(ptep);
>   	}
>   }

-- 
Thanks,
Qi

Re: [PATCH] mm/page_table_check: fix accessing unmapped ptep

[Pasha Tatashin]

On Thu, May 26, 2022 at 9:04 AM Qi Zheng <zhengqi.arch@bytedance.com> wrote:
>
>
>
> On 2022/5/26 7:33 PM, Miaohe Lin wrote:
> > ptep is unmapped too early, so ptep will be accessed while it's unmapped.
> > Fix it by deferring pte_unmap() until page table checking is done.
>
> In the beginning, page_table_check only supported x86_64, so there
> is no problem. But then the commit 3fee229a8eb9 ("riscv/mm: enable
> ARCH_SUPPORTS_PAGE_TABLE_CHECK") added support for riscv-32, it is
> indeed a problem in this case.

pte_unmap() is needed only with  CONFIG_HIGHPTE. I do not see this
config for  riskv-32?

Pasha

>
> So:
>
> Reviewed-by: Qi Zheng <zhengqi.arch@bytedance.com>
>
> >
> > Fixes: 80110bbfbba6 ("mm/page_table_check: check entries at pmd levels")
> > Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
> > ---
> >   mm/page_table_check.c | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/mm/page_table_check.c b/mm/page_table_check.c
> > index 3692bea2ea2c..971c3129b0e3 100644
> > --- a/mm/page_table_check.c
> > +++ b/mm/page_table_check.c
> > @@ -234,11 +234,11 @@ void __page_table_check_pte_clear_range(struct mm_struct *mm,
> >               pte_t *ptep = pte_offset_map(&pmd, addr);
> >               unsigned long i;
> >
> > -             pte_unmap(ptep);
> >               for (i = 0; i < PTRS_PER_PTE; i++) {
> >                       __page_table_check_pte_clear(mm, addr, *ptep);
> >                       addr += PAGE_SIZE;
> >                       ptep++;
> >               }
> > +             pte_unmap(ptep);
> >       }
> >   }
>
> --
> Thanks,
> Qi

Re: [PATCH] mm/page_table_check: fix accessing unmapped ptep

[Pasha Tatashin]

On Thu, May 26, 2022 at 7:33 AM Miaohe Lin <linmiaohe@huawei.com> wrote:
>
> ptep is unmapped too early, so ptep will be accessed while it's unmapped.
> Fix it by deferring pte_unmap() until page table checking is done.

I would re-word this as a cleanup. While pte_unmap() is currently
unused, it is still better to call it after we are done with *ptep in
case of future changes in other architectures.

>
> Fixes: 80110bbfbba6 ("mm/page_table_check: check entries at pmd levels")

This is more a clean-up, there is no existing bug, so no need to
backport to stable. Please remove the above.

> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>

With the above changes:
Acked-by: Pasha Tatashin <pasha.tatashin@soleen.com>


> ---
>  mm/page_table_check.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/mm/page_table_check.c b/mm/page_table_check.c
> index 3692bea2ea2c..971c3129b0e3 100644
> --- a/mm/page_table_check.c
> +++ b/mm/page_table_check.c
> @@ -234,11 +234,11 @@ void __page_table_check_pte_clear_range(struct mm_struct *mm,
>                 pte_t *ptep = pte_offset_map(&pmd, addr);
>                 unsigned long i;
>
> -               pte_unmap(ptep);
>                 for (i = 0; i < PTRS_PER_PTE; i++) {
>                         __page_table_check_pte_clear(mm, addr, *ptep);
>                         addr += PAGE_SIZE;
>                         ptep++;
>                 }
> +               pte_unmap(ptep);
>         }
>  }
> --
> 2.23.0
>

Re: [PATCH] mm/page_table_check: fix accessing unmapped ptep

[Qi Zheng]

On 2022/5/26 9:11 PM, Pasha Tatashin wrote:
> On Thu, May 26, 2022 at 9:04 AM Qi Zheng <zhengqi.arch@bytedance.com> wrote:
>>
>>
>>
>> On 2022/5/26 7:33 PM, Miaohe Lin wrote:
>>> ptep is unmapped too early, so ptep will be accessed while it's unmapped.
>>> Fix it by deferring pte_unmap() until page table checking is done.
>>
>> In the beginning, page_table_check only supported x86_64, so there
>> is no problem. But then the commit 3fee229a8eb9 ("riscv/mm: enable
>> ARCH_SUPPORTS_PAGE_TABLE_CHECK") added support for riscv-32, it is
>> indeed a problem in this case.
> 
> pte_unmap() is needed only with  CONFIG_HIGHPTE. I do not see this
> config for  riskv-32?

My bad, but it's better to call pte_unmap() after the check is done.

> 
> Pasha
> 
>>
>> So:
>>
>> Reviewed-by: Qi Zheng <zhengqi.arch@bytedance.com>
>>
>>>
>>> Fixes: 80110bbfbba6 ("mm/page_table_check: check entries at pmd levels")
>>> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
>>> ---
>>>    mm/page_table_check.c | 2 +-
>>>    1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/mm/page_table_check.c b/mm/page_table_check.c
>>> index 3692bea2ea2c..971c3129b0e3 100644
>>> --- a/mm/page_table_check.c
>>> +++ b/mm/page_table_check.c
>>> @@ -234,11 +234,11 @@ void __page_table_check_pte_clear_range(struct mm_struct *mm,
>>>                pte_t *ptep = pte_offset_map(&pmd, addr);
>>>                unsigned long i;
>>>
>>> -             pte_unmap(ptep);
>>>                for (i = 0; i < PTRS_PER_PTE; i++) {
>>>                        __page_table_check_pte_clear(mm, addr, *ptep);
>>>                        addr += PAGE_SIZE;
>>>                        ptep++;
>>>                }
>>> +             pte_unmap(ptep);
>>>        }
>>>    }
>>
>> --
>> Thanks,
>> Qi

-- 
Thanks,
Qi

Re: [PATCH] mm/page_table_check: fix accessing unmapped ptep

[Matthew Wilcox]

On Thu, May 26, 2022 at 07:33:50PM +0800, Miaohe Lin wrote:
> ptep is unmapped too early, so ptep will be accessed while it's unmapped.
> Fix it by deferring pte_unmap() until page table checking is done.
> 
> Fixes: 80110bbfbba6 ("mm/page_table_check: check entries at pmd levels")
> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
> ---
>  mm/page_table_check.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/mm/page_table_check.c b/mm/page_table_check.c
> index 3692bea2ea2c..971c3129b0e3 100644
> --- a/mm/page_table_check.c
> +++ b/mm/page_table_check.c
> @@ -234,11 +234,11 @@ void __page_table_check_pte_clear_range(struct mm_struct *mm,
>  		pte_t *ptep = pte_offset_map(&pmd, addr);
>  		unsigned long i;
>  
> -		pte_unmap(ptep);
>  		for (i = 0; i < PTRS_PER_PTE; i++) {
>  			__page_table_check_pte_clear(mm, addr, *ptep);
>  			addr += PAGE_SIZE;
>  			ptep++;
>  		}
> +		pte_unmap(ptep);

But ptep was mutated in the loop.  So surely this needs to be:

		pte_unmap(ptep - PTRS_PER_PTE);

or you'll be unmapping the wrong page.

Re: [PATCH] mm/page_table_check: fix accessing unmapped ptep

[Pasha Tatashin]

On Thu, May 26, 2022 at 9:34 AM Matthew Wilcox <willy@infradead.org> wrote:
>
> On Thu, May 26, 2022 at 07:33:50PM +0800, Miaohe Lin wrote:
> > ptep is unmapped too early, so ptep will be accessed while it's unmapped.
> > Fix it by deferring pte_unmap() until page table checking is done.
> >
> > Fixes: 80110bbfbba6 ("mm/page_table_check: check entries at pmd levels")
> > Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
> > ---
> >  mm/page_table_check.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/mm/page_table_check.c b/mm/page_table_check.c
> > index 3692bea2ea2c..971c3129b0e3 100644
> > --- a/mm/page_table_check.c
> > +++ b/mm/page_table_check.c
> > @@ -234,11 +234,11 @@ void __page_table_check_pte_clear_range(struct mm_struct *mm,
> >               pte_t *ptep = pte_offset_map(&pmd, addr);
> >               unsigned long i;
> >
> > -             pte_unmap(ptep);
> >               for (i = 0; i < PTRS_PER_PTE; i++) {
> >                       __page_table_check_pte_clear(mm, addr, *ptep);
> >                       addr += PAGE_SIZE;
> >                       ptep++;
> >               }
> > +             pte_unmap(ptep);
>
> But ptep was mutated in the loop.  So surely this needs to be:
>
>                 pte_unmap(ptep - PTRS_PER_PTE);
>
> or you'll be unmapping the wrong page.

Right, thank you Matthew.

Miaohe, please store the ptep, or maybe drop this patch entirely.

Thanks,
Pasha

Re: [PATCH] mm/page_table_check: fix accessing unmapped ptep

[Andrew Morton]

On Thu, 26 May 2022 09:37:43 -0400 Pasha Tatashin <pasha.tatashin@soleen.com> wrote:

> On Thu, May 26, 2022 at 9:34 AM Matthew Wilcox <willy@infradead.org> wrote:
> >
> > On Thu, May 26, 2022 at 07:33:50PM +0800, Miaohe Lin wrote:
> > > ptep is unmapped too early, so ptep will be accessed while it's unmapped.
> > > Fix it by deferring pte_unmap() until page table checking is done.
> > >
> > > Fixes: 80110bbfbba6 ("mm/page_table_check: check entries at pmd levels")
> > > Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
> > > ---
> > >  mm/page_table_check.c | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/mm/page_table_check.c b/mm/page_table_check.c
> > > index 3692bea2ea2c..971c3129b0e3 100644
> > > --- a/mm/page_table_check.c
> > > +++ b/mm/page_table_check.c
> > > @@ -234,11 +234,11 @@ void __page_table_check_pte_clear_range(struct mm_struct *mm,
> > >               pte_t *ptep = pte_offset_map(&pmd, addr);
> > >               unsigned long i;
> > >
> > > -             pte_unmap(ptep);
> > >               for (i = 0; i < PTRS_PER_PTE; i++) {
> > >                       __page_table_check_pte_clear(mm, addr, *ptep);
> > >                       addr += PAGE_SIZE;
> > >                       ptep++;
> > >               }
> > > +             pte_unmap(ptep);
> >
> > But ptep was mutated in the loop.  So surely this needs to be:
> >
> >                 pte_unmap(ptep - PTRS_PER_PTE);
> >
> > or you'll be unmapping the wrong page.
> 
> Right, thank you Matthew.
> 
> Miaohe, please store the ptep, or maybe drop this patch entirely.

I think it's best to fix it.  I rewrote the changelog as

: ptep is unmapped too early, so ptep could theoretically be accessed while
: it's unmapped.  This might become a problem if/when CONFIG_HIGHPTE becomes
: available on riscv.
:
: Fix it by deferring pte_unmap() until page table checking is done.

I'll retain the Fixes:.  This doesn't imply cc:stable in MM, and anyone
who backports the original patchset will want to know about this fixup.

And I queued a fixup for the thing Matthew noticed.

Re: [PATCH] mm/page_table_check: fix accessing unmapped ptep

[Pasha Tatashin]

> > Miaohe, please store the ptep, or maybe drop this patch entirely.
>
> I think it's best to fix it.  I rewrote the changelog as
>
> : ptep is unmapped too early, so ptep could theoretically be accessed while
> : it's unmapped.  This might become a problem if/when CONFIG_HIGHPTE becomes
> : available on riscv.
> :
> : Fix it by deferring pte_unmap() until page table checking is done.
>
> I'll retain the Fixes:.  This doesn't imply cc:stable in MM, and anyone
> who backports the original patchset will want to know about this fixup.
Makes sense.


> And I queued a fixup for the thing Matthew noticed.
Thank you Andrew.

Pasha

Re: [PATCH] mm/page_table_check: fix accessing unmapped ptep

[Miaohe Lin]

On 2022/5/27 2:31, Pasha Tatashin wrote:
>>> Miaohe, please store the ptep, or maybe drop this patch entirely.
>>
>> I think it's best to fix it.  I rewrote the changelog as
>>
>> : ptep is unmapped too early, so ptep could theoretically be accessed while
>> : it's unmapped.  This might become a problem if/when CONFIG_HIGHPTE becomes
>> : available on riscv.
>> :
>> : Fix it by deferring pte_unmap() until page table checking is done.
>>
>> I'll retain the Fixes:.  This doesn't imply cc:stable in MM, and anyone
>> who backports the original patchset will want to know about this fixup.
> Makes sense.
> 
> 
>> And I queued a fixup for the thing Matthew noticed.
> Thank you Andrew.
> 

Many thanks for all of your comments, review and fixup! :)

> Pasha
> .
>