Re: wmi: usercopy: Kernel memory overwrite attempt detected to spans multiple pages (offset 0, size 4104)

From: Kees Cook
Date: Sun Jun 17 2018 - 13:36:58 EST


On Sat, Jun 16, 2018 at 3:04 PM, Mihai DonÈu <mihai.dontu@xxxxxxxxx> wrote:
> On Sun, 2018-06-17 at 00:01 +0300, Mihai DonÈu wrote:
>> While trying to adjust the keyboard backlight mode, I hit this BUG:
>>
>> Jun 16 22:16:07 mdontu-l kernel: usercopy: Kernel memory overwrite attempt detected to spans multiple pages (offset 0, size 4104)!

CONFIG_HARDENED_USERCOPY_PAGESPAN=y is really only useful for
debugging special cases. For now, I recommend leaving it disabled,
since there are a lot of cases it still trips over.

> I eventually sprinkled some printk-s and got this:
>
> 855 if (copy_from_user(buf, input, wblock->req_buf_size)) {
> 856 dev_dbg(&wblock->dev.dev, "Copy %llu from user failed\n",
> 857 wblock->req_buf_size);
> 858 ret = -EFAULT;
> 859 goto out_ioctl;
> 860 }

However, since you tracked this one down, I think this would be fixed
by adjusting the handler_data allocation:


diff --git a/drivers/platform/x86/wmi.c b/drivers/platform/x86/wmi.c
index 8e3d0146ff8c..ea6bf98f197a 100644
--- a/drivers/platform/x86/wmi.c
+++ b/drivers/platform/x86/wmi.c
@@ -918,8 +918,8 @@ static int wmi_dev_probe(struct device *dev)
}

count = get_order(wblock->req_buf_size);
- wblock->handler_data = (void *)__get_free_pages(GFP_KERNEL,
- count);
+ wblock->handler_data = (void *)
+ __get_free_pages(GFP_KERNEL | __GFP_COMP, count);
if (!wblock->handler_data) {
ret = -ENOMEM;
goto probe_failure;


But in looking further, I don't know why this is using
__get_free_pages() instead of kmalloc? In fact, there is a kfree() in
the error path, which looks wrong:

kfree(wblock->handler_data);

I think this should just be converted to using kmalloc/kfree everywhere.

-Kees

--
Kees Cook
Pixel Security