[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 11 Nov 2013 11:42:44 -0800
From: "H. Peter Anvin" <hpa@...or.com>
To: Kees Cook <keescook@...omium.org>
CC: Ingo Molnar <mingo@...nel.org>,
LKML <linux-kernel@...r.kernel.org>,
Thomas Gleixner <tglx@...utronix.de>,
"H. Peter Anvin" <hpa@...ux.intel.com>,
"linux-tip-commits@...r.kernel.org"
<linux-tip-commits@...r.kernel.org>
Subject: Re: [tip:x86/kaslr] x86, kaslr: Provide randomness functions
On 11/11/2013 11:37 AM, Kees Cook wrote:
> On Mon, Nov 11, 2013 at 11:27 AM, H. Peter Anvin <hpa@...or.com> wrote:
>> On 11/11/2013 10:20 AM, Ingo Molnar wrote:
>>>
>>> Why aren't the 3 sources of entropy XOR-ed together?
>>
>> Note we don't want to poke i8254 if we have any other sources, as the
>> i8254 may not be present on newer systems and in some other cases not
>> safe to touch. We can XOR in the TSC with RDRAND although it seems a
>> bit odd.
>
> I await the final decision! :)
>
Use the i8254 if and only if nothing else is available, it isn't safe to
touch on some newer systems.
>> HOWEVER...
>>
>> + /* Only use the low bits of rdtsc. */
>> + random = raw & 0xffff;
>>
>> Why?! All that can do is throw away randomness...
>
> Due to the 2GiB addressing and 2MiB alignment, we can never use more
> than 10 bits of entropy at maximum. The existing patches only use 9
> bits (due to page table layouts) on 64-bit and 8 bits on 32-bit.
The point is that you are already doing that masking somewhere else, it
is just odd.
>> + /* Extend timer bits poorly... */
>> + random |= (random << 16);
>> +#ifdef CONFIG_X86_64
>> + random |= (random << 32);
>> +#endif
>
> I did this so that any portion of the returned value could be used for
> the 10 bit mask.
Why? It is standard practice to only use the low bits, and in light of
the above it is even more bizarre. However, doing a circular multiply
might still spread in some additional entropy at least in the case of a
kexec boot on a non-RDRAND system.
>> For smearing out bits, a better way is usually to multiply with a large
>> prime number (preferably a circular multiply.)
>
> This shouldn't be needed since we're only using the low 10 bits.
None of the above is needed in that case, although a spreading multiply
will make it less likely that you are throwing away the highest entropy
bits.
> How would you like to see this function updated to make that more
> clear? More comments?
-hpa
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists