Five Things That Should Scare You 2015-10-12
The world of computers is full of many dangerous nooks and crannies. Hackers, viruses, and scammers all sit as the boogeymen of the modern age. Some threats are less well known though, even amongst the keepers of the gates. This is my personal list of the five worst things floating around the chips and waves, based on notability, danger, and most of all the lack of mitigation strategies.
USB is pervasive: almost every computer produced in the past decade will have some form of USB port, most modern portable devices are charged via USB, and some new laptops are even forgoing dedicated charging ports in favor of USB-C. Behind every one of those USB ports is a small controller chip acting handling the low-level details of the USB protocol and signaling the CPU when data is decoded and ready (or vice versa). These controller chips started off as relatively dedicated silicon, but as the USB protocol has become more complex and general-purpose chips have come down in price, often they are generic co-processors with small embedded flash memory and their own firmware. With this change came the possibility to reprogram them remotely, a huge time saver when building circuit boards.
Unfortunately this also brought with it a new attack surface. Many of these USB chipsets don’t adequately protect their firmware. In some cases they simply don’t disable write access to the firmware, and others have vulnerabilities in their write restrictions.
This has led to malware that spreads directly from one USB controller to another, totally outside of the control of the operating system. Each USB controller infected with a BadUSB-style virus will attempt to attack the firmware of every USB port it is connected to. Once the firmware is overwritten, there is generally no way to even detect it other than to connect a USB debugger and monitor the traffic. On its own, this could simply be annoying, but when combined with a dangerous payload this can lead to things like invisible keyloggers, secret webcam monitors, or launching points for more complex attacks.
Protecting yourself from BadUSB is a difficult prospect. The first line of defense is to not use devices vulnerable to firmware attacks, but it can be difficult to find information on which devices are vulnerable. Avoiding untrusted USB devices is a good step, but it only takes one misstep for this to fail. Similarly you can keep power-only USB connectors, sometimes called “USB condoms”, to allow charging with untrusted connections. Overall, as will be a pattern here today, just hope nothing goes wrong really really hard.
In 1936 Alan Turing laid out the structure for what is now known as a Turing Machine. The ensuring 79 years have brought almost unfathomable advances to computer technology, but deep down in the model is the idea of a tape (RAM) which programs can read from and write to. An almost unspoken assumption in this model is that if a memory cell is written to, it will contain the same value later on when we read from it. RowHammer smashed this assumption.
Protecting against RowHammer is a difficult proposition. It does require some level of code execution, so patching local code execution issues and using browser plugins like NoScript can help, but zero-days will always be a threat. At a more fundamental level, RowHammer represents a physical problem with how we build RAM. Error-correcting (ECC) RAM does help by effectively blocking all single-bit flips and some double-bit flips, but this only makes RowHammer attacks take longer, not actually prevent them. Some more recent server hardware allows monitoring the rate of writes which could be used to detect an in-progress RowHammer attack. New types of RAM are being developed which could reduce or eliminate this threat, so keep an eye out for announcements from the memory industry.
Cloud CPU Side-Channels
Building on top of the fail that is “RAM has currently unfixable physical flaws” is the fact that most of us now run our servers on huge, multi-tenant clouds. Side-channel attacks are similar probabilistic or timing based methods, but to extract data from another virtual machine (or process, but we’ll just talk about virtual machine-level attacks). The specific vectors are numerous and continually evolving, but generally exploit hardware features created to make things faster for single-tenant desktops and laptops. Examples include timing attacks against the L2 or L3 CPU caches, CPU power usage analysis, and attacks against RAM technologies like the translation look-aside buffer (TLB). By running code to constantly poke these bits of hardware you can look for patterns that correspond to known software or data. We can do things like learn what programs are running on other VMs, what kind of data they are processing, or in some cases even extract specific data. A recent attack even managed to extract private key material across VM boundaries. This can even extend beyond local attacks, an issue in GPG allowed extracting key material by analyzing the electromagnetic radiation given off by the CPU.
Unfortunately all of these attacks are things that will have to be fixed by a mix of hardware improvements and cloud vendor support for those new features. For now, keep a careful eye on how much sensitive information you put on multi-tenant systems. Consider applying techniques used to mitigate remote timing attacks (constant time compares, etc) to truly critical data.
Originally developed as part of the KARMA exploit toolkit, the WiFi Pineapple attack platform has carried on the torch onwards. To discuss the attack itself, we need to talk about how WiFi works. Most WiFi access points will regularly send out “beacon” packets, containing information about themselves, including the network name and which frequency it is operating on. These beacon broadcasts are used to populate the list of what networks are available, so you can connect to a network without having to type in all the details manually. The downside is that on some access points, beacons can be as slow as once every 30 seconds. To speed up reconnection, when your laptop wakes up it sends out a “probe request” packet for each saved network it knows about asking that if that access point is nearby, please respond.
The attack is similar to things like ARP cache poisoning: the responder simply lies and happily replies to any probe request. This tricks the computer in to joining the attackers access point, believing it to be one of their saved networks. This usually works with unencrypted networks, as if the attacker doesn’t know the encryption key they would be unable to fake the later parts of the authentication cycle.
Some newer operating systems like iOS mitigate this by not sending probe requests with specific network names, rather requesting every access point within range send a new beacon packet immediately. This prevents direct attacks, but some related attacks are still possible by observing the beacon packets and creating a race condition where the victim might be fooled into joining the attacker’s network instead.
For most other OSes (including OS X and most laptop-friendly Linux distros) the best defense is to never leave open/unencrypted networks in your saved networks list. You can prune this list after using open networks, though this can still leave you vulnerable to determined pineapple attackers depending on the specifics of your OS.
Much of modern cryptography is built on the fact that multiplying two very big
numbers is fast, but finding the prime factors for a very big number is slow.
This asymmetry powers the math that allows things like HTTPS and most other
secure communication systems we have today. Shor’s Algorithm
offers a much faster option for finding the prime factors of the huge numbers
used to secure the Internet. Shor’s is a quantum algorithm, meaning
it is designed for a quantum computer. It can factor a number in roughly
time relative to the size of the input number, almost incalculably faster than
the exponential algorithms we have for factorization on traditional silicon.
I mention Shor’s specifically as it is the most well known at this point, but
there are is growing body of research looking at quantum factorization in general.
There is, however, a catch; the largest number ever factored via Shor’s is 21, and the largest via any quantum technique is 56153. As of yet no one has managed to build a large enough quantum computer to factor even marginally non-trivial numbers. This means Shor’s (and related quantum factorization algorithms) are not yet a direct threat, but there is another “but” to add on here. If someone was nefarious and well-funded enough to record large amounts of encrypted traffic now, they could conceivably crack it at some point in the future when more of the engineering issues with building large-scale quantum computers are solved. I will state for the record that the new NSA data center in Utah is believed to have a storage capacity in excess of a trillion terabytes.
Fortunately this is also the easiest thing on this list to defend against. Modern TLS implementations support an increasing number of cryptographic options that aren’t based on prime numbers. While it isn’t inconceivable that a quantum algorithm to solve elliptic curve problems will be developed, it is at least better than a system that we already know how to break. Keep up with modern TLS best practices and your biggest fears will have to be whatever encrypted data was already recorded, sitting somewhere and waiting for the quantum-powered future to arrive.
Hiring Chef engineers or tool developers? I'm looking for a new team! Check out my Looking for Group post for details.