Re: [PATCH] lib: Add test module for CONFIG_DEBUG_VIRTUAL

[Florian Fainelli]

On 08/08/2017 12:58 PM, Kees Cook wrote:
> On Tue, Aug 8, 2017 at 11:35 AM, Luis R. Rodriguez <mcgrof@xxxxxxxxxx> wrote:
>> On Tue, Aug 08, 2017 at 11:04:11AM -0700, Florian Fainelli wrote:
>>> On 08/08/2017 10:57 AM, Luis R. Rodriguez wrote:
>>>> On Tue, Aug 08, 2017 at 09:40:26AM -0700, Florian Fainelli wrote:
>>>>> Add a test module that allows testing that CONFIG_DEBUG_VIRTUAL works
>>>>> correctly, at least that it can catch invalid calls to virt_to_phys()
>>>>> against the non-linear kernel virtual address map.
>>>>>
>>>>> Signed-off-by: Florian Fainelli <f.fainelli@xxxxxxxxx>
>>>>> ---
>>>
>>>>> +static int __init test_debug_virtual_init(void)
>>>>> +{
>>>>> +  phys_addr_t pa;
>>>>> +  void *va;
>>>>> +
>>>>> +  va = (void *)VMALLOC_START;
>>>>> +  pa = virt_to_phys(va);
>>>>> +
>>>>> +  pr_info("PA: %pa for VA: 0x%lx\n", &pa, (unsigned long)va);
>>>>> +
>>>>> +  foo = kzalloc(sizeof(*foo), GFP_KERNEL);
>>>>> +  if (!foo)
>>>>> +          return -ENOMEM;
>>>>> +
>>>>> +  pa = virt_to_phys(foo);
>>>>> +  va = foo;
>>>>> +  pr_info("PA: %pa for VA: 0x%lx\n", &pa, (unsigned long)va);
>>>>
>>>> Should there be a tests here of some sort? When should this fail, why?
>>>
>>> There is no test per-se, the kernel will produce warning with
>>> CONFIG_DEBUG_VIRTUAL telling you that what you are doing is wrong.
>>>
>>>> There is no docs on this self test, could one be added?
>>>
>>> I suppose I could add one even though that just means pointing out the
>>> code that produces the warning?
>>
>> A /* note */ indicating what you just said above would suffice then but
>> typically tests return back to userspace an error, so another option
>> would be to see if one could get a return value that an error happened
>> and return that back to the module init. Grepping just for warning for
>> an error seems error prone.
> 
> If the test depends on the kernel's response (i.e. WARN, BUG, panic)
> that cannot be detected in the test itself, it may be better suited
> for lkdtm (drivers/misc/lkdtm*) which is almost entirely comprised of
> tests like that.

OK, do you have any specific coding styles, naming guidelines or
anything else that you would like to see for these lkdtm* modules?

On second thought it might be possible to produce both the warning and
check the virtual address against PAGE_OFFSET and if it is outside,
return an error during module_init(), I will experiment with that a
little bit.
-- 
Florian