Re: [PATCH 1/2] mm: krealloc: consider spare memory for __GFP_ZERO

[Vlastimil Babka <vbabka@suse.cz>]

On 7/30/24 10:31 PM, Andrew Morton wrote:
> On Tue, 30 Jul 2024 21:42:05 +0200 Danilo Krummrich <dakr@kernel.org> wrote:
> 
>> As long as krealloc() is called with __GFP_ZERO consistently, starting
>> with the initial memory allocation, __GFP_ZERO should be fully honored.
>> 
>> However, if for an existing allocation krealloc() is called with a
>> decreased size, it is not ensured that the spare portion the allocation
>> is zeroed. Thus, if krealloc() is subsequently called with a larger size
>> again, __GFP_ZERO can't be fully honored, since we don't know the
>> previous size, but only the bucket size.
> 
> Well that's bad.
> 
>> Example:
>> 
>> 	buf = kzalloc(64, GFP_KERNEL);
> 
> If this was kmalloc()

Then already here we have unitialized kernel memory that a buggy user could
expose, no?

>> 	memset(buf, 0xff, 64);
>> 
>> 	buf = krealloc(buf, 48, GFP_KERNEL | __GFP_ZERO);
>> 
>> 	/* After this call the last 16 bytes are still 0xff. */
>> 	buf = krealloc(buf, 64, GFP_KERNEL | __GFP_ZERO);
> 
> then this would expose uninitialized kernel memory to kernel code, with
> a risk that the kernel code will expose that to userspace, yes?
> 
> This does seem rather a trap, and I wonder whether krealloc() should
> just zero out any such data by default.

So unless I'm missing how this differs from plain kmalloc(), relying on
want_init_on_alloc() seems the right way how to opt-in harden against this
potential exposure.

>> Fix this, by explicitly setting spare memory to zero, when shrinking an
>> allocation with __GFP_ZERO flag set or init_on_alloc enabled.
>> 
>> --- a/mm/slab_common.c
>> +++ b/mm/slab_common.c
>> @@ -1273,6 +1273,13 @@ __do_krealloc(const void *p, size_t new_size, gfp_t flags)
>>  
>>  	/* If the object still fits, repoison it precisely. */
>>  	if (ks >= new_size) {
>> +		/* Zero out spare memory. */
>> +		if (want_init_on_alloc(flags)) {
>> +			kasan_disable_current();
>> +			memset((void *)p + new_size, 0, ks - new_size);
> 
> Casting away the constness of `*p'.  This is just misleading everyone,
> really.  It would be better to make argument `p' have type "void *".
> 
>> +			kasan_enable_current();
>> +		}
>> +
>>  		p = kasan_krealloc((void *)p, new_size, flags);
>