Re: [PATCH RFC 2/2] lkdtm: Add heap spraying test

[Alexander Popov]

On 15.08.2020 19:59, Kees Cook wrote:
> On Thu, Aug 13, 2020 at 06:19:22PM +0300, Alexander Popov wrote:
>> Add a simple test for CONFIG_SLAB_QUARANTINE.
>>
>> It performs heap spraying that aims to reallocate the recently freed heap
>> object. This technique is used for exploiting use-after-free
>> vulnerabilities in the kernel code.
>>
>> This test shows that CONFIG_SLAB_QUARANTINE breaks heap spraying
>> exploitation technique.
> 
> Yay tests!

Yes :)
I'm going to improve it to demonstrate the quarantine security properties.

>> Signed-off-by: Alexander Popov <alex.popov@xxxxxxxxx>
>> ---
>>  drivers/misc/lkdtm/core.c  |  1 +
>>  drivers/misc/lkdtm/heap.c  | 40 ++++++++++++++++++++++++++++++++++++++
>>  drivers/misc/lkdtm/lkdtm.h |  1 +
>>  3 files changed, 42 insertions(+)
>>
>> diff --git a/drivers/misc/lkdtm/core.c b/drivers/misc/lkdtm/core.c
>> index a5e344df9166..78b7669c35eb 100644
>> --- a/drivers/misc/lkdtm/core.c
>> +++ b/drivers/misc/lkdtm/core.c
>> @@ -126,6 +126,7 @@ static const struct crashtype crashtypes[] = {
>>  	CRASHTYPE(SLAB_FREE_DOUBLE),
>>  	CRASHTYPE(SLAB_FREE_CROSS),
>>  	CRASHTYPE(SLAB_FREE_PAGE),
>> +	CRASHTYPE(HEAP_SPRAY),
>>  	CRASHTYPE(SOFTLOCKUP),
>>  	CRASHTYPE(HARDLOCKUP),
>>  	CRASHTYPE(SPINLOCKUP),
>> diff --git a/drivers/misc/lkdtm/heap.c b/drivers/misc/lkdtm/heap.c
>> index 1323bc16f113..a72a241e314a 100644
>> --- a/drivers/misc/lkdtm/heap.c
>> +++ b/drivers/misc/lkdtm/heap.c
>> @@ -205,6 +205,46 @@ static void ctor_a(void *region)
>>  static void ctor_b(void *region)
>>  { }
>>  
>> +#define HEAP_SPRAY_SIZE 128
>> +
>> +void lkdtm_HEAP_SPRAY(void)
>> +{
>> +	int *addr;
>> +	int *spray_addrs[HEAP_SPRAY_SIZE] = { 0 };
> 
> (the 0 isn't needed -- and it was left there, it should be NULL)

It is used in tear-down below.
I'll change it to { NULL }.

>> +	unsigned long i = 0;
>> +
>> +	addr = kmem_cache_alloc(a_cache, GFP_KERNEL);
> 
> I would prefer this test add its own cache (e.g. spray_cache), to avoid
> misbehaviors between tests. (e.g. the a and b caches already run the
> risk of getting corrupted weirdly.)

Ok, I'll do that.

>> +	if (!addr) {
>> +		pr_info("Unable to allocate memory in lkdtm-heap-a cache\n");
>> +		return;
>> +	}
>> +
>> +	*addr = 0x31337;
>> +	kmem_cache_free(a_cache, addr);
>> +
>> +	pr_info("Performing heap spraying...\n");
>> +	for (i = 0; i < HEAP_SPRAY_SIZE; i++) {
>> +		spray_addrs[i] = kmem_cache_alloc(a_cache, GFP_KERNEL);
>> +		*spray_addrs[i] = 0x31337;
>> +		pr_info("attempt %lu: spray alloc addr %p vs freed addr %p\n",
>> +						i, spray_addrs[i], addr);
> 
> That's 128 lines spewed into dmesg... I would leave this out.

Ok.

>> +		if (spray_addrs[i] == addr) {
>> +			pr_info("freed addr is reallocated!\n");
>> +			break;
>> +		}
>> +	}
>> +
>> +	if (i < HEAP_SPRAY_SIZE)
>> +		pr_info("FAIL! Heap spraying succeed :(\n");
> 
> I'd move this into the "if (spray_addrs[i] == addr)" test instead of the
> pr_info() that is there.
> 
>> +	else
>> +		pr_info("OK! Heap spraying hasn't succeed :)\n");
> 
> And then make this an "if (i == HEAP_SPRAY_SIZE)" test

Do you mean that I need to avoid the additional line in the test output,
printing only the final result?

>> +
>> +	for (i = 0; i < HEAP_SPRAY_SIZE; i++) {
>> +		if (spray_addrs[i])
>> +			kmem_cache_free(a_cache, spray_addrs[i]);
>> +	}
>> +}
>> +
>>  void __init lkdtm_heap_init(void)
>>  {
>>  	double_free_cache = kmem_cache_create("lkdtm-heap-double_free",
>> diff --git a/drivers/misc/lkdtm/lkdtm.h b/drivers/misc/lkdtm/lkdtm.h
>> index 8878538b2c13..dfafb4ae6f3a 100644
>> --- a/drivers/misc/lkdtm/lkdtm.h
>> +++ b/drivers/misc/lkdtm/lkdtm.h
>> @@ -45,6 +45,7 @@ void lkdtm_READ_BUDDY_AFTER_FREE(void);
>>  void lkdtm_SLAB_FREE_DOUBLE(void);
>>  void lkdtm_SLAB_FREE_CROSS(void);
>>  void lkdtm_SLAB_FREE_PAGE(void);
>> +void lkdtm_HEAP_SPRAY(void);
>>  
>>  /* lkdtm_perms.c */
>>  void __init lkdtm_perms_init(void);
>> -- 
>> 2.26.2
>>
> 
> I assume enabling the quarantine defense also ends up being seen in the
> SLAB_FREE_DOUBLE LKDTM test too, yes?

I'll experiment with that.

Thank you!

Best regards,
Alexander