[Cryptography] Crypto for optimistic transactions ?

[Howard Chu]

Bill Frantz wrote:
> On 1/6/18 at 10:55 PM, hyc at symas.com (Howard Chu) wrote:
> 
>> The Spectre attack is harder to defend against, but it can only reveal memory
>> within a single process's address space, so for the most part I find it a 
>> non-event. It only becomes a problem if you allow hostile code to be 
>> injected into your running processes. Web browsers are the most obviously 
>> vulnerable, particularly when they allow user-loaded extensions and 
>> executing javascript etc. from random web sites. In that respect the attack 
>> surface is nothing new, and we already know about isolating browser 
>> tabs/pages into their own processes to mitigate such types of attacks.
> 
> The Spectre attack can use "Return Oriented Programming", which is used to 
> empower buffer overrun attacks on systems which don't allow execution from 
> read/write pages. See the Spectre paper below. Note that instead of overlaying 
> the stack, the Spectre attack sets the addresses in the branch prediction 
> table in the processor. No write access to victim memory needed.

That's a nonsensical statement. Somebody had to initiate the attack in the 
victim memory in the first place. Turning off javascript in the web browser 
would go a long way toward eliminating this threat.

-- 
   -- Howard Chu
   CTO, Symas Corp.           http://www.symas.com
   Director, Highland Sun     http://highlandsun.com/hyc/
   Chief Architect, OpenLDAP  http://www.openldap.org/project/