[PATCH v3 12/12] lkdtm: Add a test for function descriptors protection

[Christophe Leroy]

Le 11/02/2022 à 02:09, Kees Cook a écrit :
> On Sun, Oct 17, 2021 at 02:38:25PM +0200, Christophe Leroy wrote:
>> Add WRITE_OPD to check that you can't modify function
>> descriptors.
>>
>> Gives the following result when function descriptors are
>> not protected:
>>
>> 	lkdtm: Performing direct entry WRITE_OPD
>> 	lkdtm: attempting bad 16 bytes write at c00000000269b358
>> 	lkdtm: FAIL: survived bad write
>> 	lkdtm: do_nothing was hijacked!
>>
>> Looks like a standard compiler barrier() is not enough to force
>> GCC to use the modified function descriptor. Had to add a fake empty
>> inline assembly to force GCC to reload the function descriptor.
>>
>> Signed-off-by: Christophe Leroy <christophe.leroy at csgroup.eu>
>> ---
>>   drivers/misc/lkdtm/core.c  |  1 +
>>   drivers/misc/lkdtm/lkdtm.h |  1 +
>>   drivers/misc/lkdtm/perms.c | 22 ++++++++++++++++++++++
>>   3 files changed, 24 insertions(+)
>>
>> diff --git a/drivers/misc/lkdtm/core.c b/drivers/misc/lkdtm/core.c
>> index fe6fd34b8caf..de092aa03b5d 100644
>> --- a/drivers/misc/lkdtm/core.c
>> +++ b/drivers/misc/lkdtm/core.c
>> @@ -148,6 +148,7 @@ static const struct crashtype crashtypes[] = {
>>   	CRASHTYPE(WRITE_RO),
>>   	CRASHTYPE(WRITE_RO_AFTER_INIT),
>>   	CRASHTYPE(WRITE_KERN),
>> +	CRASHTYPE(WRITE_OPD),
>>   	CRASHTYPE(REFCOUNT_INC_OVERFLOW),
>>   	CRASHTYPE(REFCOUNT_ADD_OVERFLOW),
>>   	CRASHTYPE(REFCOUNT_INC_NOT_ZERO_OVERFLOW),
>> diff --git a/drivers/misc/lkdtm/lkdtm.h b/drivers/misc/lkdtm/lkdtm.h
>> index c212a253edde..188bd0fd6575 100644
>> --- a/drivers/misc/lkdtm/lkdtm.h
>> +++ b/drivers/misc/lkdtm/lkdtm.h
>> @@ -105,6 +105,7 @@ void __init lkdtm_perms_init(void);
>>   void lkdtm_WRITE_RO(void);
>>   void lkdtm_WRITE_RO_AFTER_INIT(void);
>>   void lkdtm_WRITE_KERN(void);
>> +void lkdtm_WRITE_OPD(void);
>>   void lkdtm_EXEC_DATA(void);
>>   void lkdtm_EXEC_STACK(void);
>>   void lkdtm_EXEC_KMALLOC(void);
>> diff --git a/drivers/misc/lkdtm/perms.c b/drivers/misc/lkdtm/perms.c
>> index 1cf24c4a79e9..2c6aba3ff32b 100644
>> --- a/drivers/misc/lkdtm/perms.c
>> +++ b/drivers/misc/lkdtm/perms.c
>> @@ -44,6 +44,11 @@ static noinline void do_overwritten(void)
>>   	return;
>>   }
>>   
>> +static noinline void do_almost_nothing(void)
>> +{
>> +	pr_info("do_nothing was hijacked!\n");
>> +}
>> +
>>   static void *setup_function_descriptor(func_desc_t *fdesc, void *dst)
>>   {
>>   	if (!have_function_descriptors())
>> @@ -144,6 +149,23 @@ void lkdtm_WRITE_KERN(void)
>>   	do_overwritten();
>>   }
>>   
>> +void lkdtm_WRITE_OPD(void)
>> +{
>> +	size_t size = sizeof(func_desc_t);
>> +	void (*func)(void) = do_nothing;
>> +
>> +	if (!have_function_descriptors()) {
>> +		pr_info("XFAIL: Platform doesn't use function descriptors.\n");
>> +		return;
>> +	}
>> +	pr_info("attempting bad %zu bytes write at %px\n", size, do_nothing);
>> +	memcpy(do_nothing, do_almost_nothing, size);
>> +	pr_err("FAIL: survived bad write\n");
> 
> Non-function-descriptor architectures would successfully crash at the
> memcpy too, right? (i.e. for them this is just repeating WRITE_KERN)

Yes it should. But not for the good reason.

> 
> I'm pondering the utility of the XFAIL vs just letting is succeed, but I
> think it more accurate to say "hey, no OPD" as you have it.
> 
>> +
>> +	asm("" : "=m"(func));
>> +	func();
>> +}
>> +
>>   void lkdtm_EXEC_DATA(void)
>>   {
>>   	execute_location(data_area, CODE_WRITE);
>> -- 
>> 2.31.1
>>
> 
> One tiny suggestion, since I think you need to respin for the
> EXPORT_SYMBOL_GPL() anyway. Please update the selftests too:
> 
> diff --git a/tools/testing/selftests/lkdtm/tests.txt b/tools/testing/selftests/lkdtm/tests.txt
> index 6b36b7f5dcf9..243c781f0780 100644
> --- a/tools/testing/selftests/lkdtm/tests.txt
> +++ b/tools/testing/selftests/lkdtm/tests.txt
> @@ -44,6 +44,7 @@ ACCESS_NULL
>   WRITE_RO
>   WRITE_RO_AFTER_INIT
>   WRITE_KERN
> +WRITE_OPD
>   REFCOUNT_INC_OVERFLOW
>   REFCOUNT_ADD_OVERFLOW
>   REFCOUNT_INC_NOT_ZERO_OVERFLOW
> 
> (Though for the future I've been considering making the selftests an
> opt-out list so the "normal" stuff doesn't need to keep getting added
> there.)
> 
> Thanks!
> 
> Acked-by: Kees Cook <keescook at chromium.org>
> 

Done.

Thanks
Christophe