Schneier on Security

Nick P • November 14, 2014 10:00 PM

@ Benni

You beat me to my counter in your own post by the end. When you first said the Belgacom situation was evidence the Russian attack was NSA I intended to counter: “with BND’s similar interception capabilities, partnership with NSA, and general amorality, how do you know it’s not them doing it? And would they cooperate with Russia or China?” But, you already answered that and confirmed my suspicions that they’re potentially scarier than NSA.

Nick P • November 14, 2014 10:15 PM

@ JestInCase

“what’s your take on the VPN’s that advertise ‘full encryption from source to destination?”

They all are built with insecure development processes and tools, running on insecure OS’s, running on potentially insecure firmware, running on hardware that makes security hard. With few exceptions, that statement applies to every VPN advertised and you typically can’t buy the exceptions outside of defense sector of certain countries. So, they’re all full of crap if they’re offering it to you personally.

The better of them might protect you from run of the mill blackhats eavesdropping on the network or doing limited endpoint attacks. However, vast majority of black hat attacks (esp nation state) hit software flaws to take over your computer. They then bypass your crypto. So, you need very strong endpoint protection combined with strong crypto. Not exactly a lot of that on the market right now so… good luck.

Tip: the best kinds of VPN’s are end-to-end link encryptors that do security-critical stuff in hardware or custom high security processors (eg Rockwell-Collin’s JANUS). The next are microkernel-based systems like INTEGRITY RTOS, Micro-SINA, or Turaya. Aesec’s GEMSOS crypto seals are similar and achieved high assurance in the past. The next best thing is custom appliances based on high quality monolithic kernels like OpenBSD (eg GeNUA).

Nick P • November 14, 2014 11:18 PM

@ Thoth

“I think I did not catch that message. Too many posts :)”

Yeah, there’s been a ton of them recently. That’s why I said “somewhere above” as I didn’t care to wade through them to find my message lol.

“Pass me a temp. pub key and I will encrypt a challenge to you. Destroy the temp. keypair after use.”

A clever secure comms attempt on a… Windows machine (MingW32)? I’m sure even Singapore can probably rootkit one of those by now or purchase them from various vendors. I’ll still set up a GPG keypair tomorrow and send it to you. 😉

Nick P • November 15, 2014 12:34 PM

@ BoppingAround

Good catch! Although risky, it’s potentially one of the best methods as VPN providers are typically not rolling in cash. Giving them a five to six digit consulting fee in exchange for looking the other way when a device is attached to their network might work fine. It’s basically what the NSA does, but for tens of millions less.

@ Benni

I predicted they’d be using Russian and Chinese proxies here years ago. My reason for that was that most online criminals use those. This means they blend in. Such boxes are also cheap. Finally, they also get to point out on TV that most cyberattacks are coming from China and Russia while neglecting to mention they were running a number of them. 😉

Nick P • November 15, 2014 2:45 PM

Good point on Unity being untrustworthy. However, this is a physically air gapped machine: it won’t be talking to anyone. So spyware is irrelevant, esp if it’s a LiveCD and he encrypts files on storage. Compatibility with hardware, app availability, usability, and reliability are more important.

He can always do a custom distro off Debian or Hardened Gentoo if he’s concerned about such things. Just not as deniable as a common LiveCD.

Nick P • November 15, 2014 11:00 PM

@ Daniel

Very well said. Yeah, I am amazed we don’t see worse. I think the reason is that attackers focus on what’s popular, the good ones want control rather than destruction, there’s a higher security baseline, and the number of shallow bugs has decreased a bit in the main apps. That attackers shifted post-SDL from Windows apps to Flash and Adobe supports my last two points. So, it’s still garbage and we’re coasting along, but only at the grace of black hats that are too lazy or greedy to do more.

This is why I support the clean slate designs and others that make everything outside the host untrusted. That brings me to my one critique:

“The internet is not safe. It’s not safe because it was never designed with a vision to become what it has in fact become. No amount of Tor or VPNs or any of such similar solutions is going to make it safe. ”

There’s been solutions that can defeat the attacks on the Internet and other untrusted networks. NSA themselves depend on them. They’re certainly not mainstream and only rely on a tiny subset of Internet protocol. The Web, on the other hand, is even worse with so many complex and wasteful protocols. I’ve seen web offerings that were possibly secure but they aren’t anything like what most web developers would go with. They’re also very wasteful and expensive if securely implemented. The Web was always garbage with an updated client-server or p2p model with type-safe, native apps being much better.

So, the situation can be a lot better even with current standards. Requires lots of proxies, guards, abandonment of compatibility, custom hardware, etc. Not practical for many and all due to the problem you illustrate: people applying technology designed for a small, benign, personal environment to a large, extremely hostile, and impersonal environment. And then trusting their privacy, money, content, I.P, and business critical operations to it.

I vote that this is the new definition of insanity.

Nick P • November 16, 2014 11:18 AM

@ 65535

I’ll try to keep posting the good shit. Meanwhile, I was reminded of why people often don’t use GPG. It’s a pain in the ass (too manual) that still glitches. That’s all the problem likely was as it worked the second try on the same PC & Internet pipe. Over a decade in and the most mature OSS secure communication tool still doesn’t work easily. And people wonder why encryption by default doesn’t take off for the masses. (sighs)

@ Thoth

With your HSM obsession, you should look up the IBM 4758 for some inspiration on the hardware side. It was the first general purpose HSM to hit FIPS Level 4. There’s a lot of information on it. Design is here and here. Teardown of one. Its tamper methods are probably beatable by now in a lab.

However, my real proposal is building an EMSEC safe with 4758-style tamper detection so a bunch of machines can be put in there. I especially like it because it’s patents age as it ages. Far as the rest of the board goes, either of us can do much better than they did in terms of processor, algorithm, and OS. The only clever thing that might be worth imitating was how they used hardware to enforce software security, esp Guardian processor.

Nick P • November 16, 2014 2:56 PM

Interesting study on bugs and vulnerabilities in projects with varying programming languages
http://www.infoworld.com/article/2844268/application-development/functional-languages-rack-up-best-scores-software-quality.html

That Java still has plenty of memory errors and Go concurrency errors isn’t speaking well for them. And what’s on top? LISP (Clojure) and Haskell. We “LISP weenies” score again. 🙂 Haskell is doing extra well thanks to its allegedly bulletproof concurrency. And the third on top is Scala, a cross between Java and functional programming. I judge things by the results they bring. These languages were all clearly well designed and props to those that contributed to that.

Another article is interesting for its section on Rust:
https://medium.com/@adamhjk/rust-and-go-e18d511fbd95

That the person’s software virtually always worked AND the compilers warnings were clear is a rare combination for a toolset. So, Rust also seems well designed if this person isn’t just a zealot or lucky. If so, then I agree with the author that it’s better to build an ecosystem around such a tool for the long-term benefits. I’d also recommend language security researchers start assessing what vulnerabilities can and can’t happen in Rust. Then start to mitigate them with coding practices, safer libraries, static/dynamic analysis tools, DSL’s, and code generators. That’s my normal recipe for boosting a language’s security.

Nick P • November 17, 2014 10:26 AM

@ Thoth

Actually the solution requires integrity at hardware, mediation at system call layer, a read everything but write nothing permission, a wrapper that allows data exfiltration over various channels, and an authentication method that enables it for key holder (service or product provider).

All of these features but one are in existing high assurance products. That one is the exfiltration wrapper that ensures their transport operates safely. There’s lots of prototypes for that sort of thing but not highly assured yet. Still a cat and mouse game. Exception is for capability secure systems.

Nick P • November 17, 2014 3:48 PM

@ konfi

I can’t believe some people think they’re going to operate anonymously in a country with a global adversary with 0-days for every major platform and app. And partnered with an adversary that can seize assets & imprison people who resist. There’s effectively backdoors in most online services already. And they’re unrestrictive with ability to damage or plant evidence on hardware.

I think a greatly restricted, hardened backdoor is preferrable. And how it’s used can be done differently in each country. Lets you move to a country that isn’t a surveillance state if you want less odds of eavesdropping. Security situation will be similar to current one but more accountable.

Nick P • November 17, 2014 6:29 PM

@ Justin

“Yet as soon as they shut down Silk Road 2, a Silk Road 3 popped up.”

Business’s that last mere years and whose owners face prosecution. Most investors and business owners would think that’s not such a great tradeoff. Gotta have less risk in the equation. I’m also personally not going to prison just so someone can make a purchase or a comment without their name on it for a short time.

“That’s just like in real life: cops have the ability to damage or plant evidence. It all depends on their honesty and competence not to do so. ”

Cops in real life might be recorded and there can be witnesses. The backdoors the NSA uses are deniable enough that even the existence of a search can be concealed, along with its specific activities. There’s considerable difference here. If each device must be specifically targeted via warrant, then a third party knows the search exists & accountability measures can be built on this. Not on covert, widespread search though.

“You’d need to greatly restrict and harden the entire system, not just the back door, as in an open source high-security, high-assurance operating system and web browser. ”

Not the case. A chip, kernel, backdoor and interface are all that’s necessary. Maybe not even the chip & kernel if hardware is designed a certain way. You just need it to be able to pull data from memory or storage yet not write it. There’s been both processors and security kernels built that can do this very robustly. (And simply.) The rest of the system runs on top of it or along side it.

“that still depends on the personal character of those in law enforcement and on politics.”

And the service provider. Remember that part. A service provider might make an argument to a federal court that the alternative backdoor can give police information while maintaining integrity of the process. The judge in the Lavabit case specifically asked for an alternative to the FBI’s black box that wouldn’t compromise users. Lavabit didn’t have anything ready and convincing, so judge ordered FBI’s device used and Lavabit shutdown. That’s where I originally got the idea to develop a high integrity & less risky to us backdoor in the event another judge might OK it. It can also be evaluated by a government certification lab and use tech similar to what’s in defense marketing materials. “If it’s good enough for DOD & FBI, it’s good enough for a court room.”

Right now there is no alternative. There’s the FBI’s black boxes, the NSA’s implants, and 0-days.

Nick P • November 17, 2014 7:53 PM

@ All

What to Do When the FBI Breaks into Your Encrypted Filing Cabinet
https://firstlook.org/theintercept/2014/11/17/newsroom-fbi-breaks-encrypted-filing-cabinet/

Funny stuff.

Nick P • November 17, 2014 8:41 PM

@ Justin

Oh I agree there’s plenty of risks. The difference is we can know and attempt to mitigate them. There’s effectively two models: service provider control and centralized control. Either allow it to be one system/site or a decentralized architecture. The best way to operate is SAP-style OPSEC with a variation of what Clive and I have promoted to deal with hardware subversions: a diverse number of systems that all perform the same function with voting protocols. Each piece is as strong as one can possibly make it, with the keys in specific fixed locations & encrypted if in storage. The physical room is protected against EMSEC and physical attacks. The keys can be killed locally or remotely at a given site upon a command. Local and remote surveillance monitoring.

The overall system is designed to securely produce, store, and use keys. The process is transactional. The nodes communicate over a variety of simple mechanisms. The specifics of the system, especially interconnection, change constantly in a way meant to throw off the attacker. The software itself is vetted by mutually suspicious users, can only perform the intended function, and is very easy to port. The components are supplied similarly. Pre-existing images exist for a number of platforms. The administrator just loads the stuff onto the machine another person prepared for them, loads the pre-existing image, and installs it. This happens under surveillance and reveals no private key material onscreen. The warranted accesses of the keys or signing of black boxes happens via prebuilt software interface of the software and generate audit records. Optionally under surveillance as well.

Further, the method would be optimized to fail-safe where the keys or systems are destroyed in worst case scenario. There’s potentially effective backup strategies that combine a physical device one needs, an encrypted backup, and a threshhold scheme. Yet, a backup increases risk quite a bit. So, fail safe is my default strategy to loose access to the devices is my default strategy. This is my main point of legal and technical risk.

Certain government and private organizations take strong measures to protect their OPSEC and INFOSEC. My methods exceed theirs. So, there’s good odds that it prevents an infiltration and otherwise detects it. The way it does it is simple: reduce trust in… everything. And only trust what many paranoids have vetted: the process and software.

Nick P • November 18, 2014 10:09 PM

@ konfi

“Their highest representatives have lied under oath and made a mockery of the legal system in western democracies. We as tax payers are paying for the big guns and it turns out the guns have been turned on us.”

Yep, but people keep empowering them even knowing this. That’s almost psychotic. And supports my proposal more than others, ironically. They’ll get the access one way or another. So, you want it limited & somewhat accounted for? Or unlimited & in secret? That’s the decision businesses have.

“The idea that one can introduce a backdoor in a system and expect no-one else to figure it out has been proven to be a bad idea over and over again.”

That’s why the computer your typing on keeps deleting text, sending spam, and so on. Because your backdoor (the update program) keeps getting broken by bad guys over and over again. And so do all your service providers’ VPN’s, SSH, etc. Alternatively, you can build backdoors that are secure most of the time, occasionally result in a breakin, are improved, and systems are recovered. And life goes on. Which model corresponds to reality the most?

“Finally, the argument “it’s hard to get privacy right, so we might as well not try at all” is so obviously flawed I won’t even get into the details.”

Who said that? The premise is people using low assurance systems and services by people with no regard to their privacy have no privacy. The second premise is the majority of those same people ensure laws that make this the default and punish anyone who does it differently. So, what legal options are available for increasing privacy in a country where people destroy it directly and indirectly? Much more difficult question to answer and very different from the one you posed. I agree that one is so flawed “I won’t even go into the details.”

@ e’

I’ve met many gun nuts being in the South and some from the West. Many of them push politicians that help create a police state and listen to media outlets that back them. Further, despite many knowing purpose of 2nd amendment, they still aren’t using it to deal with even the proven dirty politicians and corporate types that are effectively immune to prosecution. Many have gone to prison or had their assets seized over various offenses without a single gunshot fired. In short, they aren’t shit to the FBI or NSA. If anything, they’d probably shoot the people trying to help them because their friends, the radio, or TV tells them he’s the enemy.

But, yeah, go ahead and build an impenetrable service here. Let the subversives know you have bullets waiting for them like those Waco folks did. I’ll at least visit you in jail or your grave out of respect for your personal commitment to your beliefs. I surely won’t use your service.

@ Thoth

“You need to also resist the monster appetite of Govt requests. They will never be satisfied.”

That’s the political problem. If the law says so & you refuse, you have to close up shop, go to jail, or die. Depends on their level of escalation. That the public allows that situation is on them. I’ve decided that I won’t judge a business owner negatively if they do everything they legally can for us and no more. Why should we expect them to do more rather than the voters and Congress that enable these things? The situation is ridiculous here.

The key difference (for me) is that I can fight them in a court somewhere. I’m in control of how I build it, how I comply, etc. I can do everything I can do to provide some assurance against them and hopefully much more against others. That doesn’t exist with current proprietary and FOSS offerings for a variety of reasons. And the domestic agencies won’t stop hammering away. So, I like a situation where they might chill a bit, a judge might side with me, and everyone is safer. At least, until Americans decide to give us a chance to return to having civil liberties and I can further improve the offering.

@ Clive Robinson

I thought of that one too. Remember that the process runs on a highly robust TCB with trusted boot, POLA, etc. The warrant becomes another piece of software running on the machine in accordance with a security policy. If it’s read-only, then all you need after that is an expiration date. The system would periodically check it to see if that date was reached. If so, the backdoor is disabled and deleted. The expiration date would be a public (to the escrow holder) part of the request. I particularly like the POLA enforcing hardware architectures because they make it easier to ensure a program doesn’t act maliciously.

Further, the backdoor might even run on a dedicated chip or I.P. core that has high assurance methods applied to it even if system it accesses isn’t highly assured. For instance, the system being accessed might put user data in certain locations of memory or parts of the file system without users being able to alter that. Necessitates memory or file encryption, of course. Then, the high assurance L.I. component is given access to just that part of memory/disk, can only use certain interface features, can’t harm the rest of the system, and can be disabled at will.

Honestly, I think we all hate the concept of backdoors so much that the industry is overestimating how hard they are to make securely. Here’s the function in a nutshell of a basic one: issue a read command to something, perform processing, and write to something for communication. Others, you, and I have all designed strong stuff more complicated. Most compromised examples that have been given weren’t made in a high assurance way. Many they didn’t really even care or have expertise to secure. So, it’s no surprise they don’t work right. Fortunately, security engineers continue to improve on the concept with research into “remote administration,” “port knocking,” “VPN’s,” etc. Secure backdoors basically… 😉

Nick P • November 18, 2014 11:54 PM

@ Justin

” Meanwhile the rest of us aren’t going to buy it if it has a backdoor in it, and we don’t want some trusted computing base with a backdoor mandated in our PCs, either.”

You’re probably typing that message on one. So you’re full of it even if well-intentioned in that message.

“Just ordinary stuff where lives might be at stake if it doesn’t work right. ”

Other people build that stuff already with varying assurance. Anyway, that’s not my thing. I’m very concerned about the ability of people to be themselves and do things without that being a weapon against them. The problems you push are smaller, although some are difficult. The basic liberties are a foundation that even they are built on so I have to keep trying to solve those problems even in a surveillance/police state.

Nick P • November 19, 2014 10:35 AM

@ Justin

“I’m not denying my PC may well be riddled with 0-days and backdoors in both hardware and software. That doesn’t mean I want even more crap mandated on my PC.”

It already is. They’re just lying to you about whether they put it in, the level of access, how many local/foreign organizations have access, and its quality. My proposal is better on all accounts. I understand that you would oppose it, though, and did myself for over a decade. Yet, here we are writing on probably backdoored machines connected to a surveillance platform. You’re already trading off security/privacy for convenience & cost rather than leveraging what strong privacy/encryption tools are available. You’d do it again if I risked my freedom to build a better one.

“Haven’t you heard or read about COINTELPRO under J. Edgar Hoover? Your backdoor would just further enable similar abuses, and there isn’t a good technological solution against those abuses, once they are made easy by your backdoor. ”

It could. That’s a risk. Know what else can? An agency with backdoors in most things that works with police agencies doing parallel construction. What you say is already happening and that’s because the process is totally secret with nobody in the loop (past a few leakers) that gives a shit. My method allows people like us to be in the loop and ensure the integrity of the process. That’s at least a chance at restricting them on some levels until legal reforms show up.

“You don’t seem to have much respect for our 4th Amendment rights, even though the Constitution is the supreme law of the land, ”

Had you read the Fourth Amendment, you wouldn’t think it grants you absolute privacy. Notice that the Framers included that the government has the right to search or seize your property, so long as it’s reasonable and a warrant was issued. Applied to PC’s, this means they either have to have a backdoor or a good assurance they’ll get the user’s key when they ask. They haven’t had the latter assurance. So, being compliant with the Constitution requires that problem to be solved (somehow) or an escrow process that holds them accountable. I’ve worked the other end for almost ten years so I don’t mind investing a few months into escrow.

“You seem to have no qualms about shipping others off to jail (or worse) arbitrarily with your backdoor, yet you sure want to stay out of jail yourself. (From your post above, you’re squealing on anybody and everybody to avoid felony charges yourself.) ”

Straight to a baseless personal attack. Had I been a squeeler, I’d still have money and a business. I stayed out of jail by ceasing operations in a police state. Allowing warranted search in a country with a justice system and the 4th amendment is not “shipping others off to jail.” That would be the voters, Congress, police, and courts deciding that stuff as usual. I don’t even have the legal right to tell them otherwise per the Constitution. I don’t see any trend to change that either. Blame the people creating those problems and building a box around privacy tool builders, not the builders.

“Guess what: everybody else wants to stay out of jail and make money, too.”

Then, they should do something to change the status quo than watching football, partying, online flaming, bullshiting, etc. I’ve made and continue to make sacrifices to help make it happen. Where is “everybody” in this fight? I haven’t seen them. We’re losing it because they’re not present. So, I call bullshit again: you’re projecting yourself onto America as a whole. A group of people that just gripe about problems and wouldn’t risk a finger to change them in a serious effort. Minus a small few.

I won’t give up on liberty or security. But I’ve almost given up on it happening in America or if its people are worth protecting. I’m thinking my efforts would be better served in an active democracy. Can’t relocate my whole extended family so I stay over here for their sake. That’s all that’s keeping me here this long into a one-sided war.

@ Figureitout

Oh, I take a share of the blame. I gave up everything for them and they’re not even trying. That’s what I blame them for. I keep reminding them that the only thing necessary for the triumph of evil is that good men and women do nothing.

re message to Justin

I’m just reminding him that he’s probably willingly using a backdoor to say he will resist backdoors at all costs. He probably didn’t try to foreign source his equipment, use an security driven OS that’s less user-friendly, authenticate his message with crypto, etc. Yet, he’d fight a highly secure solution of mine because a local agency might get his data. And he’s using stuff that enable that. I’m just… pointing out the irony and contradiction of his words to his actions.

I’m using it for a different reason. I already told you what that was. It’s not something I’d recommend to others but it’s what I have to do until some variables in the situation change.

@ Andrew_K

I’m still an outlier. I’m just not getting jailed or shot for nothing. I’ll continue fighting them every legal way I can while trying to increase the security of our systems against foreign and criminal organizations. My scheme also secures the systems against sabotage by the domestic groups. Don’t forget that what I’m proposing will still put me at risk of prison if the secret courts decide it’s not good enough. I’d have to shutdown if I lost the legal battle.

@ BoppingAround

I strongly considered Switzerland and Iceland. I’m still evaluating the others. I’d start with the Democracy Index or whatever its called that rates democracies. Start with the Top 10 looking at how their laws, courts, press, and democratic processes work in general. I got a recent setback when a leak showed most countries in Europe were SIGINT partners of NSA. Even the ones I thought wouldn’t go for that. The subversion level is incredible and Andrew_K was about one thing: their continued, unchallenged success is depressing to me. Several thousand checkmates in a row with many opponents skipping all their turns to play.

@ Clive Robinson

I’m out of time (work calls) and want to think on your post more deeply before I respond. So, I’ll get back to you later today or the next day.

Nick P • November 19, 2014 9:08 PM

@ Justin

“I strongly object to all the moral relativism and wishy-washiness when it comes to the Bill of Rights”

I agree. I’m merely implementing the 4th Amendment on a computer. It allows warranted search and per case law quite a bit more. Remember that Supreme Court rulings are enshrined in the Constitution as well. Full anonymity and crypto schemes on highly assured SOC’s don’t allow warranted search. So, they’ll be interpreted as unconstitutional by most branches of government and they’ll push something worse than I am discussing.

The technical point here: how can they execute a search on a secure container when the keys are protected in the SOC, memory/devices don’t see plaintext, and the owner doesn’t provide the key?

@ Clive

re legal case

You hit the nail on the head of the legal aspects of the discussion. It is a Fourth vs First + Fifth Amendment issue. You’re way off on a safe: they can search a safe with a warrant and we legally must open it for them. That can also search it with probable cause with higher risk of what they find being tossed out if the judge thinks the search was unjustified. That communications are papers (letters, specifically) as mainly case law I agree. We should probably fight that in the courts more often. Meanwhile, that’s what the courts are enforcing.

I think the trick might be to split it up a bit. The main target of the search warrant per Fourth Amendment is stored data. The documents, pictures, videos, conversation logs, etc all seem to fit in the Fourth Amendment perspective. If you don’t delete them or put them off device, they’re fair game. The communications are only searched if a warranted wiretap is in place. Note that this gives a restriction already: a search for a specific type of files or data stored on the system; intercepting communications. The time limit also. So, if we fight for different case law via jury trials, then the wiretapping or types of electronic search might be limited more although not likely eliminated.

So far, I think this is consistent with people’s expectations, case law, and the Constitution’s intent.

“My solution to “papers” for stored data is to split it across foreign jurisdictions such that they can only be considered “comms” within the home jurisdiction and thus the 4th is negated leaving only the 5th.”

Sounds similar to my moving communications through multiple foreign jurisdictions scheme. Although, these days, I think that might be treated as a sign of guilt if it doesn’t look incidental (eg. multinational company). Perhaps do that by making it (on paper) a collaboration between a variety of nationalities which keep their own part of keys/data to maintain mutual trust. Allows a setup like yours in a way that makes sense to a judge or jury. Hoped result: “I could see a bunch of people from different nations wanting to split the trust of their platform in a way where all of their countries’ courts would have to authorize a search.”

Something like that. Sounds like it might be too expensive for an individual to pull off without it looking fishy. One that isn’t rich.

Nick P • November 19, 2014 9:43 PM

@ AlanS

You’re actually both right on this. He was right and worded it well when he pointed out it needed to be legal reforms with public acceptance/support. You’re right that the old reforms have been eroded and no such reforms are happening. There’s also nowhere near the strong public and Congressional support for reforms that there were before. So, he’s right but a key component is missing in this situation: the people and/or Congress pushing hard for the reforms. Neither are really doing it and so the reforms aren’t showing up.

I said once that we needed to see something that would get the public as angry and active as they once were to change this situation. If there’s a coup, the most brilliant aspect of it is that they’ve also realized this and made sure they don’t cross into that territory. Seeing how expertly they do this even with the Snowden revelations makes me doubt The People’s Fight or Flight Response will kick in. They’ll just be alerted, annoyed, grumble, and continue coasting.

Nick P • November 20, 2014 9:06 PM

@ Figureitout

Not in jail. Don’t give Levison too much credit. His sacrifice was sticking to his principles. What happened to him, though, is partly or totally related to his own foolishness: protecting all his users with one key; trying to screw with the Feds and court in obvious ways; offering no credible alternative when he got the rare opportunity to provide one. He unsurprisingly lost and might do time. So, I rarely give him as an example of anything given the extra risk he added to his users and case. He may or may not be how the next group will be treated.

The rest of your points re the chilling effect and mindsets I largely agree with. It’s why I’m pushing for organizations or people with money to fund people who can do the job. They need a guaranteed, steady funding source that can cover at least a highly qualified core team. Volunteers might contribute too and their submissions only included if vetted by the core team. The development process, from procedures to coding guidelines, is designed to make everything low risk and easy to review. Simple tools (eg Wirth’s stuff) are used so they can easily be vetted from their source all the way down to the hardware and on independent implementations by mutually suspicious parties. Total cost at snail pace development might be a few hundred K with considerable development pace possible at several million a year.

Can’t recall if I posted it but I unfortunately missed the multimillion dollar NSF grants I was working on. Couldn’t get a reply from any of the nearby schools I contacted with an INFOSEC problem despite a recommendation by the guy running it. So, I’m looking at businesses again and trying to build a value proposition for them along the lines the open source & hardware movements are doing. Except, it’s not going to be free: just continual financial support of an expanding ROI platform with security, usability, open interfaces/data-storage, future proofing, and low training/maintenance cost.

Will be extra busy at work for next few months so that’s slowed me down on that and posting here far as discussions that require a lot of thought.

@ Clive Robinson

That’s an interesting scheme and legal observation. It’s recommendations are in my high security email advice I posted here in the past, although it was more comprehensive. So, yeah I might be able to work out some details. 😛

The problem on the U.S. side of your scheme is that they would likely subvert the company like they did Crypto AG. If it’s an exemplar subversion (in chip), then it might go undetectable. Blocking the import or seizing it as a denial of service is the next step. Interdiction with an EMSEC attack on internal secrets is another. They’ve been getting innovative recently in that category. Getting the signing key with a black bag job or EMSEC attack on the factory is another. Then there’s the design database attack at the handoff to the fabs that RobertT mentioned. The benefit is that some of these are targeted attacks and low hanging fruit is removed for many high strength attackers.

The bad thing is that what’s left is well within Five Eyes, Russian, Chinese, or Israeli reach. I have some potential solutions far as getting rid of the likely solutions at hardware subversion. That leaves a few attacks: ban of importation unless backdoor; seizure for DOS or sabotage; cancelling of defense contracts of business or academic funding of those using them (important for some); legal attacks on the individual’s freedom. These would have to be addressed. My framework only addressed some of them, as well.

re original post at 11/19/14 4:52am

The mechanism I described can work in secret. Here’s my trick. The device supports covert channels over software (packet headers) or hardware (eg RF). The main chip always sends the keys. The backdoor chip always does an operation equivalent to leaking them: you can’t tell at black box level if it’s running or not because it’s always running on something. The activation doesn’t signal the owner, switches it to “Send it for real” mode, it works for a time, and then it’s deactivated manually or via time limit. Only group’s like Ross Anderson’s people and such equipment can perform any countermeasures you mention.

re your comms scheme

See my above comment to Justin for mine. Yours integrates with it nicely. So, the backdoor gives them nothing on you but plenty on other potential targets. Assuming I leak over straight data rather than side channels. The real challenge for my scheme is is with mobile devices. Like my last subversion-resistant design, I think the newer trick on a mobile device will stretch the word “mobile.” It’s not going to be a slim Galaxy or have great battery life. Manufacturing security would be interesting. Again, I have proposals for this sort of thing but I don’t want to publish them yet.

Either way, I’m basically playing Devil’s advocate and exploring a potential desparation move at improving market security in general against foreign attacks. Whatever I build, if there’s a backdoor, will have it off by default with activation (or permanent deactivation) at the factory over authenticated channel. If I’m forced to enable it, I can imagine many privacy lovers will have a profitable run in the resulting black market on the early models.

In this country, though, I’d probably have to enable it by default and just tell people it’s there. There’s personal risk for me there. I’d also tell them what’s in their other devices with slides to prove it and show a visual comparison of resulting security level against various adversaries. Whether they buy or not, I doubt they’ll feel like they’ve avoided a backdoored product. 😉

Nick P • November 21, 2014 3:39 PM

Techniques for creating pervasive assurance at manageable cost

This is a combination and summary of many of my prior essays. There’s several concepts in my method: TCB reduction; chip reuse; physical isolation; safer languages; diversity. Let me describe several different strategies I used in past designs for stopping High Strength Attackers. Then, I can show how they can be combined.

System of Systems

Based on my PCI backplane scheme. The system is actually a bunch of different card computers connected to each other, with at least one master node. Each card computer performs a logical function such as kernel, interface, graphics, sound, network, or disk. The card computer runs a hardened OS with everything non-essential to its function stripped. Custom drivers are developed so the main node can offload work or receive results from the card computers with input validation optional. The main node has a I/O MMU to restrict what the other DMA engines can access. Main node might have trusted boot, built-in software mitigation, a secure OS, a hardened vanilla OS, etc. The physical isolation & MMU’s contain attacks on various devices while letting you use diversify on the card computers. This allows many performance, cost, and security tradeoffs.

Incremental assurance

This was invented in MLS days. The idea is that you start with a secure design, leverage at least one highly assured component, start selling/using this system, and use that funding to increase assurance of whole system over time. A simple example is a database. You might start with a high assurance guard sitting between an insecure application server and an insecure database that runs a fixed set of commands on restricted data. Then, you might design a tool that takes a query language input, performs security checks, and then converts it to executable code that runs against the database. The system is now more flexible with apps still having restricted access to the database. The next step might be to build secure parallelism or replication into it for higher performance. Alternatively, one might improve the assurance of the guard’s hardware/firmware/software. And so on.

Each step provides a meaningful increase in security over what came before it. Each step might give the user more functionality or performance. Each step is small enough to fund with minimal sales if medium assurance development and reasonable if high assurance. It also reinforces the notion that security is a moving target due to constantly changing requirements.

Reuse of hardware and software

We can’t be developing chips for everything. It will be too expensive. However, most of the components we need have already been developed. A base I.P. core with trusted boot, a processor, hardware-based enforcement of software security, I/O support with security, and optionally onboard crypto are all we need. That can be turned into a number of chips by combining chip specific hardware with the base we’ve already verified the hell out of. The reuse of I.P. in chips or individual chips in boards keeps cost way down. There’s one company that’s gone through several product iterations of cutting edge chips with essentially only a few million in expenses.

At the system level, we’ve seen you can stretch one thing pretty far (eg Linux kernel). Microsoft’s approach in Verve, which they borrowed from old mainframe vendor, is probably the best approach. You create a minimal kernel that implements functionality from memory management to concurrency to exception handling that the rest of the system needs. Like with separation kernels, it has the security critical functionality. It’s built with strong techniques. The rest is built on top of it leveraging its type-safe interfaces and type/capability secure access. Unlike the kernel, these components can be written in high level languages with extensive verification available. Many FOSS and limited funding academics have done all these things in prototypes so it seems cost effective.

Artificial diversity

Diversity forces targeted attacks if done right. There’s a new field of study called software or system diversity that aims to use automated techniques to transform systems into a moving target for the attacker. Old versions are things like changing port numbers or address space randomization. Newer techniques include randomizing the instruction sets, internal data layout, primitive operations used for equivalent functionality, and so on. These techniques can have strong, open implementations built to be used on whatever systems are designed. Users could validate and generate their systems from source with these techniques automatically applied to make their system unique.

Subversion resistant development

Development & review will be done by mutually suspicious parties. These parties will keep their own development machines, repository, and guards to protect them. A build system such as Aegis or OpenCM will be used to ensure only signed submissions are allowed, their integrity is preserved, automated checks are run, and so on. Nothing is final until it’s been reviewed and signed by members of at least 3 other parties. The parties collaborate and chat to keep on the same page to prevent issues that can stall the project.

Design and coding standards are employed to help. An EAL6-style development process occurs where the software is kept structurally simple, there’s little looping around in the code, the ins/outputs of each function are easy to understand, risky or hard to analyze constructs are avoided, everything must pass basic static/dynamic checking, a basic covert channel analysis is used, and so on. Modern tools actually make this easier to do than one might expect except for maybe covert channel analysis. There’s methods like Kemmerer’s Shared Resource Matrix to make that easier and automated tools in development. The resulting components are easy to vet, have been vetted, and should have few defects.

Combining it all

It ideally starts with hardware development. Academics will build the hardware and come from nations unlikely to cooperate on a backdoor. If possible, the hardware starts with an existing clean slate design (eg SAFE or CHERI), adds features needed for practical use, adds secure boot, and implements on FPGA’s and an ASIC prototype. The design work is partitioned among them, although all end up reviewing the results. The reviews are done quite regularly so each review can cover a small number of changes or claims at a time. Makes it easy, consistent, and embedded into the process. The resulting source, synthesized results, and so on are published by all parties with signatures. This is either freely available or licensed at just low enough cost to fund further development, with reviewers or academics being able to study it freely.

The core software development runs in parallel. The nucleus of the system with the most serious features are developed using tools that allow strong analysis. The multinational team codes the stuff up, validates each others work, and so on. The team keeps a software and FPGA model of the hardware that is regularly updated for easy prototyping and verification. Much of the team’s early activities will be acquiring tools and learning to leverage them in a design flow suitable to cost-effective, higher assurance development. Empirical evidence shows this will have a high upfront cost that goes way down in future iterations that reuse tools, techniques, and/or code.

Diversification tool development runs in parallel. This is already happening in academia. The techniques can be prototyped for the system language in use by academics using purpose-built tools like Racket. They could throw together a dozen or more diversity techniques in a year. The best one’s can be implemented in standalone, assured code. That will be done incrementally either by them or others. The aim will be that the Racket-style versions will be reference implementations showing anybody how to incorporate it in their systems or use it on their own code. The assured implementations have lower TCB, leverage the strong analysis tools, and are mainly for efforts integrating into the assured platform.

Software and tool development above firmware begins once the kernel is a bit stable or runs in parallel. This will be drivers, OS functions, middleware, application services, certifying compilers, etc. Any hobbyist, company, or university might contribute to the repository. The contributions have strict guidelines for what must be submitted and how for verification purposes much as in other development. The verification status of submissions is listed publicly with what verifications were done for each submission and by who. The system might automatically include new componts into a build after their testing and verification meets a certain minimal standard. There might be different builds with different feature and trust levels.

Integration is where this gets turned into useful stuff. The integrators architect a whole system with a specific set of required functionality. They specify various pieces from the verified repository. They add their new functionality with various safety/security tests or claims. They create build instructions ideal for their project that a third party can run to generate the final system for evaluation or use. They then distribute that who collection of stuff with their signature to end users. End users (Linux-style) get the pieces from the repository, many tests are run automatically on everything involved, the build process turns it all into an executable, and it optionally can be installed permanently onto the system.

So, let’s do a web server for someone’s blog. They might pull in a certifying compiler, the kernel, networking components, HTTP parser, network/application firewall component, a concurrent garbage collector, and a hardened Python engine ported to these. They wrote the web server in Python and with various tests for reliability/security. Regardless of their web server code, the result should run correctly at everything below their code with no code injection mechanisms so long as such things are disabled in the Python interpreter. The work for the user was getting repository access set up, maybe buying/configuring some dedicated hardware, doing Linux-style repo download, learning a subset of Python, and following some secure configuration/coding guidelines. I’ve seen lay people do equivalent IT work with good instructions and naturally techies can do way more. Lots of security ROI without much impact on development, maintenance, or usability.

The whole process is design to continuously, incrementally, and sometimes massively improve the assurance of usable systems. The work, both development and review, can be done in parallel. The scheme, esp distributed repositories, are designed so progress on a component stalls unless work is put into reviewing it. Biggest risk here is bullshit reviews so that’s why it should be something many involved actually use for themselves, esp for national or corporate secrets. A subscription and licensing model is used to keep entry barrier very low while providing continuous funding for multinational organization maintaining the repository & doing baseline work. Donations of money/code, academic grants, and corporate coders can be used to get more functionality in.

Over time, such methods can build highly secure and useful systems at manageable cost. Previous projects using each mini strategy achieved the objectives I described. Empirical evidence indicates the combination of them should achieve the overall goals I describe. And this can all be done with a lightweight organization with minimal funding for the basic deliverables. More funding and sponsorship delivers more results.

Nick P • November 21, 2014 9:11 PM

@ Justin

“Well, no shit. ”

It’s so obvious that about nobody in industry, academia, or FOSS is doing it. 😛

Ironically, the best two at the moment are from U.S. and U.K., respectively. They’re also funded by DARPA. So, I’d have France, Japan, China, or Russia do the reviews on them. We could simultaneously have the Chinese and Japanese engineers developing the hardware additions and French and Russians doing kernel software. The U.S. and U.K. people would review that. Independents could review any of it as well.

Single biggest attack on this is if the main countries all agree in secret to kill the project because it would hurt all their SIGINT efforts. Honestly, though, I think China and Russia particularly won’t do that because they can count on Western companies using insecure stuff. If ground up secure stuff were available, Russia and China would be more likely to benefit in SIGINT defense without loosing offense. So, they might support a project even if others would want it gone.

And then they would demand backdoors in any domestic implementation. 😉

Nick P • November 22, 2014 4:35 PM

@ Chris

re Tripwire on Windows

You can try using this one which is multiplatform:

http://www.ossec.net/?page_id=169

It also has a Splunk app to integrate them. They’re both free.

Nick P • November 23, 2014 9:05 PM

@ Figureitout

re grants

I contacted schools within a few hours drive with talented INFOSEC or hardware people. It needed to be an organization with a research track record, prior grant approvals, and the minimal amount of staff needed to do the work. None were interested. (shrugs)

re assurance scheme

There is either no isolation on the backplane or great isolation. Depends on the backplane or how you connect to it. A system wide policy enforced by IO/MMU’s each card connects to and input validation are a simple combo to limit damage by compromised nodes. The main benefit is the hardware, firmware, and software can be picked for the logical function. You can reduce TCB, trade security for compatibility/cost, etc. The biggest advantage for me was the “unsolved” problem of driver security: I just got them the hell off my main node and problem solved.

Initial dev takes place in a shared repository everyone uses with their own local copy. I’m talking air gaps, guards, whatever. The funding organization would provide the guards or vettable instructions on how to build them. Signatures are used for verification purposes, submissions are sent in batches in zip files so individual files don’t need sig checks, and plenty of communication (eg hardened chat or IM) to keep things in sync. The language used for tools will be cross-platform C/C++, Oberon, or a ML-like language. These all have implementations on Windows and Linux. Oberon has the advantage of being extremely easy to port to… anything.

I’d probably run it jointly with another whose more diplomatic and could manage day-to-day operations. I’d hire, part-time or full-time, a number of experts in various areas of high assurance system design. They’d vet submissions to the site, wiki, etc for safety, quality, and relevance. This is just to ensure people going into a section will see what might benefit them most. Diverse ideas will be allowed. At least one person will be dedicated to the moderated discussion forum which has general chat, sections for common categories, and threads to manage different discussions. Moderation is, similar to here, focused on limiting abusive behavior, limiting the unintentional flamewar effect of debates, and keeping people on topic if it’s a narrow one. There would also be a place for the source and binaries probably done about the same as the articles.

It’s funny you mentioned microcontrollers because I recently posted something like that. (See last part of comment.) The modern microcontrollers are quite capable, have device support, and so on. A custom motherboard with a main CPU, a microcontroller for each device (running its driver), a way to secure the physical I/O, and then a regular microkernel approach on top might be a very easy start. I think one custom chip for a secure I/O processor would be useful. It works side-by-side with either a COTS or custom processor that use its standardized interface for chip-to-chip or chip-to-memory connection. In a nutshell, early mainframes worked like this except the word “micro” didn’t apply to their physical footprint in any way.

In instruction set randomization, you still have control of it. You just make sure you backup (onto paper) the key that drives the process, reinter it into the tool, and validate whatever it created before you trust your system to it. And, of course, strongly assure the implementation of the scheme if its failure would be catastrophic to your data. Things above the CPU/firmware would likely just crash the OS or an app in a way that allows recovery. This kind of stuff has been done for years now without any huge problems cropping up outside the security claims for a specific scheme not panning out.

Subversion resistant development starts with centralized storage controlled by my organization, but verified in a distributed way. The best resource on this sort of thing is Wheeler’s page. It’s a simple enough problem that’s been pretty well hashed out. There’s been exemplarly designs that could be converted into strong implementations on fairly secure TCB’s. Shapiro, the guy behind EROS OS, helped develop the OpenCM system that implemented most of what you would want and in a quite modular way. Corrupt hardware dealt with by many independent implementations of an unambiguous standard for it by different people on different hardware, guards, air gaps, recoverable OS’s, etc. Simple shit: just a lot of work to get set up, vet, etc. Good news is you do it once for the basic tools and repo, then can leverage them for the rest of the process with the change list going back to very first files. Similar to Bitcoin.

Licensing will start in the $250,000-$1mil range for this access to use whatever organization creates: free copies of any software with source with improvements required to be sent back to organization; secure hardware at just above cost; free access to all papers, code, training material, tools, and so on in repo’s; support for all the above. Enterprises might get VPN’s, databases, web servers, and so on for the price of one high end proprietary software. This can be discounted on basis of the organization’s income to the point a small business might get it for a few grand. Pricing for non-commercial use, esp individuals, will be lower: maybe $100-200 a year. Contributors might get discounts of a free week to year for contributions to the repo. Services offered to individuals would be based on the resources they use plus a basic fee for each service. Unless I built a platform of all kinds of services. In that case, individuals or business subscribe to the platform at probably a fixed rate where we just keep developing more stuff with what comes in and charge extra for extra resource use. Businesses would pay the most with it being cheap enough for individuals to encourage mass adoption.

Nick P • November 24, 2014 12:22 PM

@ Figureitout

Using a subversion resistant development process helps solve a lot of what you describe. The people all operate on a trust-but-verify principle. Diverse hardware, diverse minds, and multinational. I’ll help any bootstrap their activities with something trustworthy if they need and preferrably parts are couriered.

The experts are potentially subversive. So far, many academics, businessmen, and FOSS players are consistently producing stuff that I approve of. If your theory was right, they shouldn’t be doing that. So, picking some of them to review the work and having others test a sample of their reviews should suffice. Plus, we must remember that their reputation is tied to this: their choices are immortalized in history. The last thing a good academic or senior scientist wants to be immortalized for is someone holding back the good shit. Odds are in our favor on this, especially if in a good geographical location.

re specifics on securing the computers

I already have quite a few options for this part. I’m just not posting them. The reason is, at this stage, obfuscation is a huge part of security and sharing details of one’s site security plan just helps HSA’s out.

re site security

That’s easier than you think. I’m just not doing a web site. :O Well, there will be a web site but it’s just mirroring the real data store. I’m going back to client server model where the client and server are written in safe code. The protocols, interfaces, and data structures will be simple. A guard handles transport security sitting in between the Internet and intranet. The intranet might use TCP/IP with TCPcrypt or IPsec. However, I might also throw TCP/IP out for another protocol stack or obfuscate it. Robust systems will handle network monitoring, authentication, repository, backups, updates to untrusted systems, and various security critical functions. These will likely be cheap embedded boards running microkernels, etc.

re devs’ security

We don’t trust them: we verify them. How many times I gotta say this? 😛 The simplest route for them is to use my old MLS setup: two or more computers, a KVM switch, and a guard. They might substitute laptops/netbooks to avoid the KVM switch. They can use whatever they want so long as one computer is airgapped, data going into it can be vetted, data going out of it can be vetted, and it can be recovered in event of problems. Network connection is handled by an untrusted machine.

And good luck on your hardware work. Yeah, ADC coding might be a bitch to learn. I’m still thinking you’re better off just getting something running first (even a Forth or BASIC system) and incrementally improving your setup’s security/functionality. Then, you can actually use it for development and as a root of trust. But hey I’m just goofy like that.