[PATCH] lkdtm: fix potential use after free

[PATCH] lkdtm: fix potential use after free

[Weikang shi]

From: swkhack <swkhack@gmail.com>

The function lkdtm_WRITE_AFTER_FREE calls kfree(base) to free the memory
of base. However, following kfree(base),
it write the memory which base point to via base[offset] = 0x0abcdef0. This may result in a
use-after-free bug. This patch moves kfree(base) after the write.

Signed-off-by: swkhack <swkhack@gmail.com>
---
 drivers/misc/lkdtm/heap.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/misc/lkdtm/heap.c b/drivers/misc/lkdtm/heap.c
index 65026d7de..0b9141525 100644
--- a/drivers/misc/lkdtm/heap.c
+++ b/drivers/misc/lkdtm/heap.c
@@ -40,8 +40,8 @@ void lkdtm_WRITE_AFTER_FREE(void)
 	pr_info("Allocated memory %p-%p\n", base, &base[offset * 2]);
 	pr_info("Attempting bad write to freed memory at %p\n",
 		&base[offset]);
-	kfree(base);
 	base[offset] = 0x0abcdef0;
+	kfree(base);
 	/* Attempt to notice the overwrite. */
 	again = kmalloc(len, GFP_KERNEL);
 	kfree(again);
-- 
2.17.1

Re: [PATCH] lkdtm: fix potential use after free

[Mark Rutland]

On Wed, Apr 24, 2019 at 06:21:03PM +0800, Weikang shi wrote:
> From: swkhack <swkhack@gmail.com>
> 
> The function lkdtm_WRITE_AFTER_FREE calls kfree(base) to free the memory
> of base. However, following kfree(base),
> it write the memory which base point to via base[offset] = 0x0abcdef0. This may result in a
> use-after-free bug. This patch moves kfree(base) after the write.

As with lkdtm_READ_AFTER_FREE, this is deliberate, and we should not
make this change.

Thanks,
Mark.

> 
> Signed-off-by: swkhack <swkhack@gmail.com>
> ---
>  drivers/misc/lkdtm/heap.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/misc/lkdtm/heap.c b/drivers/misc/lkdtm/heap.c
> index 65026d7de..0b9141525 100644
> --- a/drivers/misc/lkdtm/heap.c
> +++ b/drivers/misc/lkdtm/heap.c
> @@ -40,8 +40,8 @@ void lkdtm_WRITE_AFTER_FREE(void)
>  	pr_info("Allocated memory %p-%p\n", base, &base[offset * 2]);
>  	pr_info("Attempting bad write to freed memory at %p\n",
>  		&base[offset]);
> -	kfree(base);
>  	base[offset] = 0x0abcdef0;
> +	kfree(base);
>  	/* Attempt to notice the overwrite. */
>  	again = kmalloc(len, GFP_KERNEL);
>  	kfree(again);
> -- 
> 2.17.1
> 

[PATCH] lkdtm: fix potential use after free

[Weikang shi]

From: swkhack <swkhack@gmail.com>

The function lkdtm_READ_AFTER_FREE calls kfree(base) to free the memory
of base. However, following kfree(base),
it access the memory which base point to via base[offset]. This may result in a
use-after-free bug. This patch moves kfree(base) after the dereference.

Signed-off-by: swkhack <swkhack@gmail.com>
---
 drivers/misc/lkdtm/heap.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/misc/lkdtm/heap.c b/drivers/misc/lkdtm/heap.c
index 65026d7de..3e2f1580a 100644
--- a/drivers/misc/lkdtm/heap.c
+++ b/drivers/misc/lkdtm/heap.c
@@ -77,7 +77,6 @@ void lkdtm_READ_AFTER_FREE(void)
 	base[offset] = *val;
 	pr_info("Value in memory before free: %x\n", base[offset]);
 
-	kfree(base);
 
 	pr_info("Attempting bad read from freed memory\n");
 	saw = base[offset];
@@ -88,6 +87,7 @@ void lkdtm_READ_AFTER_FREE(void)
 	}
 	pr_info("Memory was not poisoned\n");
 
+	kfree(base);
 	kfree(val);
 }
 
-- 
2.17.1

Re: [PATCH] lkdtm: fix potential use after free

[Mark Rutland]

Hi,

On Wed, Apr 24, 2019 at 05:59:52PM +0800, Weikang shi wrote:
> From: swkhack <swkhack@gmail.com>
> 
> The function lkdtm_READ_AFTER_FREE calls kfree(base) to free the memory
> of base. However, following kfree(base),
> it access the memory which base point to via base[offset]. This may result in a
> use-after-free bug. 

This is deliberate; the test is designed to provoke a use-after-free to
verify that this can be detected by tools such as KASAN.

> This patch moves kfree(base) after the dereference.

Doing this would break the test, so we should not make this change.

Was this something you spotted with a static analysis tool?

> Signed-off-by: swkhack <swkhack@gmail.com>

A signed-off-by line should have your real name rather than an alias.

Thanks,
Mark.

> ---
>  drivers/misc/lkdtm/heap.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/misc/lkdtm/heap.c b/drivers/misc/lkdtm/heap.c
> index 65026d7de..3e2f1580a 100644
> --- a/drivers/misc/lkdtm/heap.c
> +++ b/drivers/misc/lkdtm/heap.c
> @@ -77,7 +77,6 @@ void lkdtm_READ_AFTER_FREE(void)
>  	base[offset] = *val;
>  	pr_info("Value in memory before free: %x\n", base[offset]);
>  
> -	kfree(base);
>  
>  	pr_info("Attempting bad read from freed memory\n");
>  	saw = base[offset];
> @@ -88,6 +87,7 @@ void lkdtm_READ_AFTER_FREE(void)
>  	}
>  	pr_info("Memory was not poisoned\n");
>  
> +	kfree(base);
>  	kfree(val);
>  }
>  
> -- 
> 2.17.1
>