Slashdot Mirror
Can You Trust Chinese Computer Equipment?
Ian Lamont writes "Suspicions about China slipping eavesdropping technology into computer exports have been around for years. But the recent spying attacks, attributed to China, on Google and other Internet companies have revived the hardware spying concerns. An IT World blogger suggests the gear can't be trusted, noting that it wouldn't be hard to add security holes to the firmware of Chinese-made USB memory sticks, computers, hard drives, and cameras. He also implies that running automatic checks for data of interest in the compromised gear would not be difficult." The blog post mentions Ken Thompson's admission in 1983 that he had put a backdoor into the Unix C compiler; he laid out the details in the 1983 Turing Award lecture, Reflections On Trusting Trust: "The moral is obvious. You can't trust code that you did not totally create yourself. (Especially code from companies that employ people like me.) No amount of source-level verification or scrutiny will protect you from using untrusted code. In demonstrating the possibility of this kind of attack, I picked on the C compiler. I could have picked on any program-handling program such as an assembler, a loader, or even hardware microcode. As the level of program gets lower, these bugs will be harder and harder to detect. A well installed microcode bug will be almost impossible to detect."
This is just another reason for me to not want to buy Chinese made goods. Unfortunately, so much is made in China that it is nearly impossible to completely avoid the country.
Some component of your car, cell phone, computer, etc. is going to be made in China. I have a feeling eventually they will catch on that people aren't buying Chinese made stuff and will just put stamps on it from their more friendly neighboring countries.
Posts not to be taken literally. Almost everything is sarcasm.
AFAIK, this is the only CPU still made in America.
the preceding comment is my own and in no way reflects the opinion of the Joint Chiefs of Staff
In a general sense, you really can't trust any computer equipment that you didn't build yourself, pretty much from the ground up(as the issues with compilers and microcode suggest). I'm pretty sure that using somebody else's sand to make your silicon is safe; but that's about it.
Computer gear hasn't quite reached biological levels of complexity, where trust is even harder(one malformed Prion in a batch of millions can end up eating holes in your brain); but, from the perspective of a user who isn't a tech god, it might as well have.
That being so, the question of whether you can trust Chinese computer equipment is basically a political one. China's general enthusiasm for industrial espionage is well known, so if you have data on interesting technology or military stuff, the answer is almost certainly "no". If you are basically just Joe Consumer, though, your data are just noise obscuring what Chinese intelligence really wants. You would do better to be worried about the botnet your PC is part of, Google, ChoicePoint, Equifax, the NSA, and whoever is taking advantage of CALEA at that particular moment. The world of technology is a ghastly morass of potential backdoors, quite a few of them not even hidden, that most of us are constantly vulnerable to, and, in a great many cases, actively being monitored through.
Bugged Chinese chips are definitely something to think about if you are doing military COTS procurement, or doing security for somebody who has data of real interest; but, for most of us, it's all just one more piece of assymetric transparency. I, for one, don't feel any warmer and fuzzier about the Americans spying on me than the Chinese spying on me(worse, in fact, because some sinister chinese intelligence agency is substantially less likely to sell my information to advertisers, make it harder to get medical insurance, or damage my credit rating than some warm, fuzzy, American multinational corporation).
I really hope that this threat leads to a general recognition of the need for sound and open practices for security(both in the sense of novel CS research on how to do maximally verifiable stuff, test blackboxes, build verified bootstrap compilers, etc, etc. and in the sense of market acceptance of the fact that mysterious binary firmwares, and "just trust us" responses from vendors, and blackbox systems in general just aren't good enough). That would make things better for everybody. I get the unpleasant sense, though, that a lot of this concern is less about "We really need to understand how to build highly complex systems that are dependable and verifiable for those who use them." and more about "Goddam chinks, only we are supposed to have backdoors and surveillance capabilities!"
This isn't just for good known to be made in china. This past year we performed an audit of our network infrastructure with Cisco's help. We found almost 10% of our switches were counterfeit. They were all models of layer 2 and layer 3 switches and were virtually indistinguishable from genuine Cisco products down to the enhanced security IOS.
I'll meet you at the intersection of "Should be" and "Reality"
There is a fairly large amount of counterfeit Cisco gear floating around
http://www.networkworld.com/news/2006/102306counterfeit.html
http://www.networkworld.com/community/node/13213
http://www.andovercg.com/services/cisco-counterfeit-wic-1dsu-t1.shtml
And we all know where this stuff is made.
OTOH we just bought a huge pile of new Juniper stuff at work, every single piece "Made in China".
Give a man a fish and you have fed him for today. Teach a man to fish, and he'll say "WHERE'S MY FISH, YOU IDIOT?"
DoD is really worried about this. They're trying to develop ways to efficiently examine ICs to check for unexpected "features". Right now, it's necessary to open up the IC and put it under a scanning electron microscope, then use software that can extract the logic diagram from the scan.
One of the obvious places to put in a "back door" is in Ethernet controllers. Many used in servers already have logic for hardware "remote administration" (turn machine off, reboot, load code, etc.). It is supposed to be disabled by default, and work only when initialized with keys during hardware installation. Just build a set of default remote administration keys into the chip, and everyone using that chip is 0wned. Send the right UDP packets, and you can take over the machine. This would be completely invisible until activated.
Whenever this subject comes up, I post about it and either get a +5 insightful or get flamed to hell and told I don't know what I'm talking about, so let's see what happens this time. I work in semiconductor design. In a CPU or memory chip there are some sections of the chip that have duplicate/spare circuitry that can be brought into play if some of the main circuitry is defective. This is what people refer to when they talk about trimming memory chips. I don't do this sort of stuff so I don't actually know for sure, but people who post on slashdot claiming to know, say that it would be "easy" to jigger some of the spare circuitry to provide added/surreptitious functionality to the chip.
Thing is: I don't see that this is very useful since it's in ram or the cpu, and it seems to me to be possible, maybe even likely, to see surreptitious traffic from them heading outwards to the ethernet controller chip.
I think -- as apparently do you -- that the most likely places to try to put in backdoors are the I/O chips because it's hard for you to determine what they're doing. But then they have to include some serious functionality, to implement at least a little intelligence to decide what to send, unless they want to send everything, which again would be pretty obvious to someone looking at the hardware.
And since I work at a place that *does* design ethernet controller chips, although that's not what *I* do, I can say with at least some assurance that it's really, really, really unlikely that they could be backdoored.
Let me explain why: on analog and small digital chips, die size is *unbelievably* important because it is directly related to your profit margin. I've done chip layout. We will go to any lengths whatsoever to make the die smaller, even if it means completely relaying out the chip. There isn't any space for extra circuitry at all. Every square mil is loaded.
On top of that, we then run our prototype chips on planet runs, where a bunch of proto chips from various designers are all masked onto a chunk of silicon, in either our own local fab or our tiny owned fab in Europe, and then characterize the returned chip, and do metal changes and maybe a complete new mask set, and only *then* does it go out to the big fabs. And when we get *those* back, we spend months characterizing *them*, making sure that every individual pin has the same leakage current and ESD protection characteristics, as the ones we got back from our local fab, to ensure the chips will actually work in the field.
In order for a Chinese fab to put a backdoor into one of our designs they'd have to increase the die area, which would be really amazingly obvious, or remove existing circuitry, which would be really amazingly obvious. Even if they're so incredibly clever as to redesign the chip better than we can design it in the first place, giving them space to add their circuitry, it's very unlikely that the current draw on every pin during operation and when forced into test mode and pushed to failure, would be within 1% of the chips we got from fabs that we control.
With all that said, my company recently closed our Chinese fabs, an
Nostalgia's not what it used to be.
After all we did that to the Russians in the 80's causing one of their large oil pipelines to explode. Does it make you feel better that Microsoft gave China a peek at the full source code for Windows? http://www.builderau.com.au/architect/work/soa/US-software-blew-up-Russian-gas-pipeline-/0,339024596,320283135,00.htm
"I'm not a quack, I'm a mad scientist! There's a difference." - Dr. Cockroach
In order for a Chinese fab to put a backdoor into one of our designs...
If just the IC fab is outsourced, with masks provided, that's true. Many Ethernet chips are designed in Taiwan and fabbed in China, but so far I can't find ones developed entirely on the mainland. That can't be far off; eventually, engineering and design moves near the fab. There are competent IC design houses in China; HiSilicon and C2 Microsystems are sizable design companies. But neither makes an Ethernet controller. The focus of the Chinese design companies tends to be entertainment electronics and portable devices.
I think it would be difficult to do a company like HP. Any additional chip means additional cost, and HP would notice this right away. It would have to be a company that collaborates in the design stage.
Intel has their own network-facing backdoor built into their chips. HP uses them in its laptops - and HP's outsouced-IT service organization supplies these machines to the companies which hire them.
Look up "Intel AMT" on the web. There's lots of stuff on it available there. It's a "feature" intended for large companies' IT operations to use to remotely administer the workers' laptop and desktop machines: Remote update software, detect malware, cut misbehaving machines off the LAN or shut them down, monitor workers' behavior, ...
It is "below" the main CPU(s) and OS. It runs even if the main machine is off. It is a man-in-the-middle on the network interface, accepting its own connections from the "mother ship" and configurable to "phone home" when on the road. It can monitor and twiddle all the network traffic, monitor all the I/O (including keystroke logging), access the hard drive, stop the processor, monitor applications for watchdog events and shut them down if they "misbehave", halt and restart the main processors, yadda yadda yadda.
It can also present one of its own intercepted connections-from-afar to the main processor as if it were a terminal interface on another chip. The recommended way to configure Linux or Unix on the box is for this interface to be given a login process with root login privileges.
How do you know if it's disabled? The BIOS TELLS you it's disabled. (If you believe that, especially after the next BIOS firmware update, would you be interested in some land in Nevada?)
Bantam Dominique roosters crow a four-note song. Once you've heard it as "Happy BIRTHday" you can't NOT hear it that way