mm: Prevent mapping slab pages to userspace
diff mbox series

Message ID 20190125173827.2658-1-willy@infradead.org
State New
Headers show
Series
  • mm: Prevent mapping slab pages to userspace
Related show

Commit Message

Matthew Wilcox Jan. 25, 2019, 5:38 p.m. UTC
It's never appropriate to map a page allocated by SLAB into userspace.
A buggy device driver might try this, or an attacker might be able to
find a way to make it happen.

Signed-off-by: Matthew Wilcox <willy@infradead.org>
---
 mm/memory.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Kees Cook Jan. 25, 2019, 6:44 p.m. UTC | #1
On Sat, Jan 26, 2019 at 6:38 AM Matthew Wilcox <willy@infradead.org> wrote:
>
> It's never appropriate to map a page allocated by SLAB into userspace.
> A buggy device driver might try this, or an attacker might be able to
> find a way to make it happen.
>
> Signed-off-by: Matthew Wilcox <willy@infradead.org>
> ---
>  mm/memory.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/mm/memory.c b/mm/memory.c
> index e11ca9dd823f..ce8c90b752be 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -1451,7 +1451,7 @@ static int insert_page(struct vm_area_struct *vma, unsigned long addr,
>         spinlock_t *ptl;
>
>         retval = -EINVAL;
> -       if (PageAnon(page))
> +       if (PageAnon(page) || PageSlab(page))

Are there other types that should not get mapped? (Or better yet, is
there a whitelist of those that are okay to be mapped?)

Either way, this sounds good. :)

Reviewed-by: Kees Cook <keescook@chromium.org>

-Kees

>                 goto out;
>         retval = -ENOMEM;
>         flush_dcache_page(page);
> --
> 2.20.1
>
Matthew Wilcox Jan. 25, 2019, 7:30 p.m. UTC | #2
On Sat, Jan 26, 2019 at 07:44:40AM +1300, Kees Cook wrote:
> > -       if (PageAnon(page))
> > +       if (PageAnon(page) || PageSlab(page))
> 
> Are there other types that should not get mapped? (Or better yet, is
> there a whitelist of those that are okay to be mapped?)

Funny you should ask; I think the next patch in this series looks like this:

-       if (PageAnon(page) || PageSlab(page))
+       if (PageAnon(page) || PageSlab(page) || page_has_type(page))

but let's see if there's something I've overlooked with this patch.
Andrew Morton Jan. 28, 2019, 6:20 p.m. UTC | #3
On Fri, 25 Jan 2019 09:38:27 -0800 Matthew Wilcox <willy@infradead.org> wrote:

> It's never appropriate to map a page allocated by SLAB into userspace.
> A buggy device driver might try this, or an attacker might be able to
> find a way to make it happen.

It wouldn't surprise me if someone somewhere is doing this.  Rather
than mysteriously breaking their code, how about we emit a warning and
still permit it to proceed, for a while?
Kees Cook Jan. 28, 2019, 7 p.m. UTC | #4
On Tue, Jan 29, 2019 at 7:21 AM Andrew Morton <akpm@linux-foundation.org> wrote:
>
> On Fri, 25 Jan 2019 09:38:27 -0800 Matthew Wilcox <willy@infradead.org> wrote:
>
> > It's never appropriate to map a page allocated by SLAB into userspace.
> > A buggy device driver might try this, or an attacker might be able to
> > find a way to make it happen.
>
> It wouldn't surprise me if someone somewhere is doing this.  Rather
> than mysteriously breaking their code, how about we emit a warning and
> still permit it to proceed, for a while?

It seems like a fatal condition to me? There's nothing to check that
such a page wouldn't get freed by the slab while still mapped to
userspace, right?

But I'll take warning over not checking. :)
Christopher Lameter Jan. 28, 2019, 8:08 p.m. UTC | #5
On Tue, 29 Jan 2019, Kees Cook wrote:

> It seems like a fatal condition to me? There's nothing to check that
> such a page wouldn't get freed by the slab while still mapped to
> userspace, right?

Lets just fail the code.  Currently this may work with SLUB. But SLAB and
SLOB overlay fields with mapcount. So you would have a corrupted page
struct if you mapped a slab page to user space.
Michael Ellerman Jan. 31, 2019, 12:37 a.m. UTC | #6
Matthew Wilcox <willy@infradead.org> writes:

> It's never appropriate to map a page allocated by SLAB into userspace.
> A buggy device driver might try this, or an attacker might be able to
> find a way to make it happen.
>
> Signed-off-by: Matthew Wilcox <willy@infradead.org>
> ---
>  mm/memory.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/mm/memory.c b/mm/memory.c
> index e11ca9dd823f..ce8c90b752be 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -1451,7 +1451,7 @@ static int insert_page(struct vm_area_struct *vma, unsigned long addr,
>  	spinlock_t *ptl;
>  
>  	retval = -EINVAL;
> -	if (PageAnon(page))
> +	if (PageAnon(page) || PageSlab(page))
>  		goto out;
>  	retval = -ENOMEM;
>  	flush_dcache_page(page);


Thanks for turning this into an actual patch.

cheers
Pekka Enberg Jan. 31, 2019, 6:03 a.m. UTC | #7
On 25/01/2019 19.38, Matthew Wilcox wrote:
> It's never appropriate to map a page allocated by SLAB into userspace.
> A buggy device driver might try this, or an attacker might be able to
> find a way to make it happen.
> 
> Signed-off-by: Matthew Wilcox <willy@infradead.org>

Acked-by: Pekka Enberg <penberg@kernel.org>

A WARN_ON_ONCE() would be nice here to let those buggy drivers know that 
they will no longer work.

> ---
>   mm/memory.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/mm/memory.c b/mm/memory.c
> index e11ca9dd823f..ce8c90b752be 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -1451,7 +1451,7 @@ static int insert_page(struct vm_area_struct *vma, unsigned long addr,
>   	spinlock_t *ptl;
>   
>   	retval = -EINVAL;
> -	if (PageAnon(page))
> +	if (PageAnon(page) || PageSlab(page))
>   		goto out;
>   	retval = -ENOMEM;
>   	flush_dcache_page(page);
>

Patch
diff mbox series

diff --git a/mm/memory.c b/mm/memory.c
index e11ca9dd823f..ce8c90b752be 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -1451,7 +1451,7 @@  static int insert_page(struct vm_area_struct *vma, unsigned long addr,
 	spinlock_t *ptl;
 
 	retval = -EINVAL;
-	if (PageAnon(page))
+	if (PageAnon(page) || PageSlab(page))
 		goto out;
 	retval = -ENOMEM;
 	flush_dcache_page(page);