Re: [PATCH RFC v10 2/6] x86/entry: Add STACKLEAK erasing the kernel stack at the end of syscalls

[Alexander Popov <alex.popov@linux.com>]

On 29.03.2018 21:38, Shivappa Vikas wrote:
> On Wed, 28 Mar 2018, Alexander Popov wrote:
> 
>> +	const unsigned long check_depth = STACKLEAK_POISON_CHECK_DEPTH /
>> +							sizeof(unsigned long);
>> +
>> +	/*
>> +	 * Let's search for the poison value in the stack.
>> +	 * Start from the lowest_stack and go to the bottom.
>> +	 */
>> +	while (p > boundary && poison <= check_depth) {
>> +		if (*(unsigned long *)p == STACKLEAK_POISON)
>> +			poison++;
>> +		else
>> +			poison = 0;
>> +
>> +		p -= sizeof(unsigned long);
>> +	}
>> +
>> +	/*
>> +	 * One long int at the bottom of the thread stack is reserved and
>> +	 * should not be poisoned (see CONFIG_SCHED_STACK_END_CHECK).
>> +	 */
>> +	if (p == boundary)
>> +		p += sizeof(unsigned long);
>> +
>> +	/*
>> +	 * So let's write the poison value to the kernel stack.
>> +	 * Start from the address in p and move up till the new boundary.
>> +	 */
>> +	if (on_thread_stack())
>> +		boundary = current_stack_pointer;
>> +	else
>> +		boundary = current_top_of_stack();
>> +
>> +	BUG_ON(boundary - p >= THREAD_SIZE);
>> +
>> +	while (p < boundary) {
>> +		*(unsigned long *)p = STACKLEAK_POISON;
>> +		p += sizeof(unsigned long);
>> +	}
> 
> Sorry catching up late on this. 

Hello Vikas,

You are just in time! Thank you for looking at the code.

> can we use a stosq or something here ? wondering 
> if that helps get more performance. since you seem to copy the whole stack with 
> the poison value. If that was already discussed sorry for the spam :)

Yes, the previous version of the patch series had this function written in
assembly. It used scasq for the poison searching and then stosq for writing:
http://www.openwall.com/lists/kernel-hardening/2018/03/03/9

That amount of assembly was blamed by the maintainers. So I've done my best to
rework it in C bypassing the pitfalls:
http://www.openwall.com/lists/kernel-hardening/2018/03/21/4

In fact, this implementation is not much slower than the assembly one, since we
erase only the used part of the thread stack. This is achieved by the
lowest_stack variable, which is updated during the syscall (please see the next
patch of the series).

Best regards,
Alexander