In late July; security researchers publicly disclosed (defined) a new set of vulnerabilities within Intel CPUs (defined) (and possibly AMD and ARM; which the researchers also notified). These vulnerabilities are collectively referred to as SpectreRSB (Return Stack Buffer). The purpose of an RSB is explained in this document (PDF) but in summary it is a buffer (defined) that stores multiple return addresses while attempting to predict function (a set of instructions that carries out a specific action within a program) return addresses.
A very short time later nearing the end of July; a separate set of researchers released details of another vulnerability known as NetSpectre. This is an evict and reload cache attack that targets systems remotely to extract data.
How could an attacker exploit these vulnerabilities and what is the result?
For SpectreRSB; an attacker could recover data from the speculative execution feature of the CPU by targeting the Return Stack Buffer and predicting the return address which it stores. By manipulating the data it contains by predicting the return address the CPU will access when it completes a task the attacker can influence the address CPU will jump to and thus jump to an address of the attacker’s choosing. Unfortunately; this buffer is shared among the threads (defined) on the same virtual process thus affecting multiple running processes and virtual machines.
The attacker could alter the RSB to expose and gather data from applications running within the CPU. Another form of manipulation by the researchers resulted in them being able to expose data contained within Intel’s Software Guard Extensions (defined)(PDF).
Separately for the NetSpectre vulnerability; if attackers can send specifically crafted packets (defined) to a vulnerable system they can use the responses they receive to infer data from that systems memory. Currently this can only take place at a very low rate; 15 bits per hour. This means 15 times a zero or a one; in other words true or false (I’m not referring to Boolean logic here; just trying to convey a concept) or even simpler on for 1 and off for zero. This increased to 60 bits per hour for an Intel CPU equipped with AVX2 instructions.
With such a low throughput at this time (although I realise an attack can usually be refined and significantly improved within a short time); this attack is not a practical threat but more a theoretical weakness.
How can I protect myself from these vulnerabilities?
The good news for this SpectreRSB subclass of vulnerabilities is that Intel has already created an update but not for all of it’s CPU (Intel Core i7 Skylake (6th Generation Core models) and later CPUs). The researchers are aware of this patch and are recommending it’s use. When I use the word subclass above; my meaning is that SpectreRSB is a subclass of the original Spectre vulnerabilities from January this year. Red Hat also announced they are reviewing these vulnerabilities.
Intel however have stated that existing mitigations from the vulnerabilities disclosed in January will protect against this new subclass. However this is unconfirmed at this time.
While an APT (defined) could leverage the NetSpectre vulnerability over a period of weeks or months to extract useful data; existing mitigations for Spectre variant 1 and variant 2 mitigate this new vulnerability reinforcing my statement above of being a theoretical weakness.
In summary; to protect against both classes of these vulnerabilities; please continue to roll-out the mitigations for the Spectre vulnerabilities from January 2018 (if you have not already completed them).
For any system which cannot be updated (due to performance or end of life constraints e.g. Intel not providing updates for some CPUs); seek to migrate the responsibilities/roles/duties of these systems to newer CPUs which have received updates. A list of patched and un-patched Intel CPUs is available here (PDF).