Schneier on Security

Clive Robinson • August 5, 2010 3:42 AM

A few points,

One, AES256 is not in of it’s self secure outside of a single block of encryption. Depending on what mode you use and how is actually a lot more important.

Two, it may not be a single AES key for the container it could be many (think AES in CTR mode as the key gen to the AES mode that encrypts the container and gets rekeyed from the AES-CTR every file block)

Three, the container it’s self may actually provide the key in some way (say hash the first half of the file gives the key to the second half which contains a file with a key to the first half).

There are a number of ways you can play with the encrypted container. For instance at what point do you start decrypting? the file might have been subject to “russian coupling” in some form.

Further let us say that a 512byte block buried in the middle is a “bit map” to reorder some of the file blocks. Another block might likewise be a bit map to “nul” blocks that need to be removed. This process it’s self is obviously critical in it’s ordering and can be repeated over as multiple rounds…

Or you could use a system such as Ross Anderson’s “Dancing Bear”.

So you may need to know one heck of a sight more than the AES256 key before you get going.

But we are again assuming just the one “container” and once opened all the cards are shown.

How do we know that it is not a “Russian Doll” with containers within containers, as you open a container you get one or two “taster files” and the next container.

The keys to each container being produced from one or more BBS generators for instance?

This gives wikileaks the way to inflict “death by 1000 cuts” onto both the administration and the Afghanistanies.

Which gives the administration a problem, they push wikileaks and the outer container gets opened they get a black eye for one of their intel files but also a kick in the happy sack because a file with Afghan-US colaborators is revealed.

They DOD might be prepared to take repeated black eyes but are they Administration going to take repeated and very public kicks in the happy sack from the public via the press / internet over “wontanly revealing the names of Afghan’s fighting to bring democracy aganst terrorists”…

Thus a cautious and moderatly clever person could keep the US off their back. If they feel they are being threatend they give a single warning to “back off” if the threat does not decrease a key goes out and a couple of more files become public. This could go on and on many many times with containers in containers.

As was once said,

Such is the price of freedom from tyranny.

Clive Robinson • August 5, 2010 12:36 PM

@ BF Skinner,

“No it sounds like a re-voicing of the suspicion that there’s a back door in AES. Tell me how you can put a back door into an algorithm will you?”

First you have to state the conditions…

It is well known that the AES competition did not consider certain “dynamic” aspects and within a very short time it was found that many AES implementations had “timing side channels”. Subsequently it turns out you have to be an extraordinary software engineer to prevent side channels on the majority of modern CPU’s with AES (cache timing attacks).

It is well known how to put “covert channels” into stream ciphers and there has been a paper or two (at the most) about what would be required to put a covert channel into a block cipher.

AS Nick P and myself have discussed in the past the NSA has “previous” when it comes to “back dooring” designs not just by them selves but also by Crypto companies in other nations (where clocks pop out little birds on the hour every hour 😉

I first came across this years ago when writing up some stuff on how to crack some early machine ciphers such as the “coin counting mechanism”.

It turns out that when you look at all the keys they have different strengths ranging from very weak to strong. Only very few in the latter group (say 1 sixth or 17%).

If you know this AND you are responsible for producing the daily “key settings” then you stick within that 17%. However if you lose one of these machines and it gets copied, the chances are your adversary does not know about the relative key strength differences therefore they will use keys from all 100%. Which means as the NSA you have been gifted something like 1/3 of the messages by your adversary…

If you come a little further upto date do you remember the “crypto escrow” arguments and “Capstone”?

Well when the “secret design” was looked at it was found to be strong but very brittle, in that even tiny non obvious changes would render it a weak system.

Lastly go back and have a look at AES certification by the NSA, They don’t actually “blanket certify” AES… Only AES in approved equipment operated under certain conditions. One of which is usualy only for systems where the ciphertext is at rest (see the inline media encryptor).

My advice to anyone with reason to be even a little paranoid (ie anyone with secrets worth more than 100K USD) is NEVER use AES online and take good EmSec precautions with the machine you do it on…

If you have to use AES online, remember “Clock the Inputs and Clock the Outputs” and “fail hard on any error for a long timeout”. This should hopefully throttle any time based side channels to a very very low bandwidth (but it won’t stop them so re-key often).

Also don’t use standard formats such as Word PDF ZIP etc. (they have standard headers) unless you use “Russian coupling” or equivalent “transposition”.

The devil is as they say in the details, which is why the NSA have more strings to using AES than there are on the harps of those angels dancing on the head of that pin…

Clive Robinson • August 6, 2010 2:54 AM

@ cornerstone,

Not being American I’m not 100% sure what a “nutbar” is or why it would be “raving” (It brings up images of a “Snickers” Bar dancing the night away to techno / drum-n-base 😉

As for entertaining, aside from the occasional oddbeat comment to raise a wry smile, I aim to provide usefull information or give pause for thought (the first step of inovation). If I also make peoples time pass more easily or provide them with other entertainment, then consider it a bonus.

@ Peter Maxwell,

“If you are suggesting any assets worth over a certain value should not rely on protection by AES”

Absolutly not.

What I’m talking about is useing AES in “on line” or “off line” mode.

In on line mode use the time based side channels from the use of AES can be seen remotly and have been practicaly demonstrated as such.

Removing these AES side channels in a software only implementations on modern CPU’s is extreamly difficult for a number of reasons.

Further software changes only limit passive attacks not active attacks. Eliminating active timing attacks on a general purpose multi tasking computer where an attacker can influence such things as load, cache and memory utilisation and a whole host of other factors opens timing side channels back up again.

Also there are other timeing side channels involved with communications that can leak information through AES or any other crypto system because they are not a consiquence of the crypto algorithm.

Thus timing side channels either inherant within the system or induced in the system by an attacker are very real issues not just in the future but currently, and are very difficult to spot (I’m not aware of any IDS or other off the shelf software system that can detect them).

Obviously for an attacker irrespective of if it is a passive or active timing side channel attack there is a certain degree of technical and other difficulties involved which requires a commitment of resources.

Thus there is a tipping point at which the commitment of such resources will (probably) show a return above the cost of commiting them.

Which brings me onto your question,

“So are you saying for $100k you can break a channel encrypted with properly implemented AES”

As I said you have to define your parameters to make a definate answer to that. We do know that previous published timing side channel attacks have involved little more than a network connection and a high end PC.

Therefore what I am saying is that if you routinely handle secrets worth over 100K USD you should realisticaly consider that you may well become a target now or in the future to some kind of attack against your encrypted communications.

As Bruce notes attacks only get more sophisticated and practical with time, and people learn with (sometimes painfull) experiance not to fall for simple attacks. Thus those criminals that have made a good living out of various phishing and malware attacks will migrate upwards to this sort of attack as and when their existing attacks start to fail to bring in the money.

And as a consiquence a responsable user/organisation should take steps to eliminate the potential vulnerability of timing side channels.

There are various ways this can be done but the simplest currently is to use AES “off line” in a way that EmSec attacks are also reduced.

That is your plain text and crypto software are on an “air gaped” and possibly TEMPEST hardend system and only cipher text files get put on the machine responsable for network communications. This way there should be no timing side channels based on the plain text.

However for obvious reasons this does not mitigate traffic flow analysis which may reveal much more information than a single plain text message.

@ running around in circles,

@ Nick P asked if that was me some time ago and the answer is still the same.

Though why people want to know who I am puzzels me unless they have large handfulls of money they want to give me in which case they can give it to Bruce and he can pass it on minus a handeling fee 😉

Clive Robinson • August 6, 2010 8:21 AM

@ Yvon Sauvageau,

“The best encryption for a one-time message is the one-time pad. It’s 100% uncrackable.”

Only in theory…

As you note,

“So if they’ve done it right, their key (pad) is 1.4 GB in size.”

The weasel words being “if they’ve done it right”

Generating 1.4GByte or more of really random is not easy, and it has it’s own issues.

The same theory that proves the OTP security also proves that an unending string of zero’s or ones or some “plain text revealing” pattern thereof is equi-probable with one that shows no discernible pattern (ie just one of the myriad possible in 11,000,000,000 bits)

Although the full length string is highly improbable shorter strings of plain text revealing “patterns” are very probable in such large collections of bits.

Which means you have to take steps to decide which strings of bits are unacceptable and why and then mitigate against them (yes you can do Fourier, Walsh, Wavelet and compressibility analysis but that only gets some of them).

So it is far from being as simple as just waiting for your TRNG to pop them out…

In practice you would end up with a deterministic process with a random input, with two major questions arising,

1, What is the acceptable level of determanisum to “sufficiently” prevent revealing plain text.

and,

2, How much True Random input is required to produce the required 1.4Gbytes of OTP key to an appropriate level of security…

In real life you might well be better off using a 256 bit counter with start offset and AES (ie CTR mode) where the AES key is updated at some lesser rate by another crypto algorithm in CTR mode. You then select a subset of the output (ie bottom 32bits or some such)

However at the end of the day it is still effectively a stream cipher (as is the OTP) with all it’s inherent defects.

So maybe it should use another block cipher to replace the XOR function, but not in ECB if the…

And so you slowly spiral down in “what if’s” 😉

Clive Robinson • August 7, 2010 4:47 AM

@ David Thornley,

@ Nick P’s comments about “sharing -v- seperation” come from the then Director of Education at the NSA Brian Snow (Christopher snow did something altogether different) in a keynot speach he gave back in 2005,

http://www.acsac.org/2005/papers/Snow.pdf

In which he does a 5 year and 10 year forcast, even though his forcast was dismal for the time he was woefully over optomistic (in many ways we. have actually regressed).

The paper is well worth a read and I agree with most of it including the DRM point (you need to seperate the purpose of the technology from the intent of the paymasters to see why).

I have been banging on about “efficiency -v- security” for many years (publicly so since the late 1990’s). Brian Snow’s “share -v- seperation” is but a small subset of it. Sometimes you have to look at the motivation not the technology to see where the real issues and solutions repose.

Clive Robinson • August 8, 2010 5:11 AM

@ Nick P,

Yes Brian Snow has been a Director of quite a few areas in the NSA at one time or another. And I have no issue with what he said in that Key Note Speech (other than it did not go far enough).

Yes the NSA does have a tarnished image as do the likes of IBM Microsoft et al.

The problem the NSA has is it’s almost schizophrenic or Multiple Personality Disorder of it’s original mission statment.

It was original tasked with two main objectives, the first being “to protect the communications of the US” the second “to intercept and report the communications of other nations”.

At first sight it appears to be two seperate functions with a lot of common ground (to be secure you first have to know what you are securing against, which you find out by attacking others systems for weaknesses).

However an issue arises when the first mission gets extended via the likes of NATO, BRUSA etc to “protecting the communications of the US and her Allies” esspecially when you are activly attacking the allies private communications (for whatever reason). You get a schism simply because any system you use with your allies leakes information about it’s inner workings which enables those you are evesdropping on to improve their communications beyond the point where you can usefully do anything with any intercepts.

Back in the 1960’s it became abundantly clear that National Security was not limited to just matters pertaining to Diplomats and the Military but commerce as well. This point became brutaly clear later when Russia used espionage techniques to manipulate grain prices so they could get large quantities of US grain at prices they would not have been able to without using them.

Thus the NSA had more mission creap on it’s hands it now had to provide protection to comercial interests as well which is why we ended up with DES. But the NSA had miscalculated they never intended for DES to be anything other than a hardware implementation (one of the reasons for the Initial Permutation).

Unfortunatly for the NSA DES leaked way to many secrets to enquiring minds, the other side of their business took a major hit. So the NSA went into Public Denial about DES for many many years with easily disprovable statments about it’s strength etc.

Back in the 1980’s it became clear to some that software had major issues with side channels and software implementation of crypto algorthms was very bad news for security. But very good news for the other side of the NSA’s mission as it allowed them to deduce key information and thus stay in business.

However it was bad news for the protection of the comms side. If you have ever had the misfortune to work on NSA or BID equipment you will know what an engineering night mare it is. More so if you ever had to design the bl**dy stuff. It was not just TEMPEST design but SEGREGATION it all had to be done in a non obvious way (“give away” paranoia). The result was the 500lb gold plated Gorilla doing the job of a 10 ounce carrier pigeon.

The equipment the NSA was producing was not fit for purpose and the cost was unacceptable. With the end of the cold war the NSA had to “get with the program” of cost effectivness. They tried a myriad of wrigles (CapStone / key escrow etc) but during the Clinton era they realised that it was better for their interests to be seen to doing something usefull for the US people with all the money they got given.

The NSA actually for a while became proactive in many ways. Unfortunatly the Hawks got back in the driving seat and with no real external enemy they had to treat the civilians as the enemy thus the NSA got a third mission that was most definatly outside it’s charter but it came with a nice big bundle of cash.

Then along came 9/11 as a gift from heaven and the NSA illegal third mission went compleatly rouge. And due to the administration at the time it’s access to private individuals within the US went just about “total”. However this involved so many resources and people it leaked out fairly quickly and hit the fan and the whole organisation got compleatly “dipped in 5h1t”.

Sadly the good work (that they still do) as part of their mission to protect the comms of the US et al is now effectivly hidden from view or treated with deep suspicion.

Mostly what this side of the NSA says can be regarded as spot on. However you have to pay attention and read between the lines. The old game between the two original missions still applies only at a different level.

So as I said the NSA approve of AES under certain conditions only… they don’t give it blanket approval. It’s working out the why’s of the conditions that provides a significant challenge to the astute.

As I’ve said befor on this blog the area of concern currently is “side channels” in all their forms and “known plaintext” in the likes of standard document headers.

Side channels can only be dealt with by “segregation” techniques and strict interface monitoring and control to reduce channel bandwidth to very minimal amounts, then effectivly encrypt the channel with noise such that even that narrow bandwidth is unusable in a realistic time frame.

As for “known” plain text “compression” is not effective thus after compressing a file you need to pre-encrypt it before using the main encryption.

One early way to hide headers (Russian Coupling) was to split the text into two at a random point and swap their position thereby hidding the position of the message header in the body of the message (you then inserted the position of the split at a known point in the text).

In modern times one split is inadiquate and the length of splits needs to be kept as short as possible, whilst being spread across the entire message length. Thus a byte level transposition encryption will break up a “known text”. However the “known text” frequency statistics will still remain thus the application of a substitution (stream cipher) after the transposition would not go amiss 😉

Have a look at Ross Anderson’s “Dancing Bear” system for some good ideas,

http://www.cl.cam.ac.uk/~rja14/Papers/grizzle.pdf

Clive Robinson • August 8, 2010 7:53 AM

@ Cornerstone,

“You last posts are far more understandable. If you are about sharing information then clarity and explanation are far more helpful than a tornado of technical terms that only a very few readers wil make any sense of.”

One flip side of the problem is keeping the post short, compress it to far via techical terms etc and well “snickers all round”.

However to answer you question,

“What’s the simplest way to make a notebook”EmSec”?”

The simple answer as you have already guesed is you cann’t it’s the environment around it AND the user.

The things you have to consider are energy and bandwidth in channels.

Most people immediatly think of EMC type protection and forget that a hidden CCTV camera can watch the user typing and possibly even the screen.

It has been shown that you don’t even have to be in line with the screen opticaly to use optics to reconstruct it. Some bods at Camb Labs used a telescope and a photo multiplier to pick the serial scan signal of a generalised reflection on the wall from across the street.

Likewise acoustic energy has been used to pick up the keyboard sounds and that of the fan and hard drive to leak information.

In theory (and probably in practice) strain guages mounted at appropriate points will be able to tell which key was pressed if the mechanical coupling is sufficient, Thus a laptop put down ontop of a suitable preasure mat would each time a key is pressed give four different strength signals from the four rubber pads underneath the laptop…

Thus in “theory” you have a lot to consider and mitigate against.

As I said (badly 😉 above there is a tipping point at which each type of attack becomes viable to an adversary.

There are two ways of dealing with energy emissions, suppression and masking.

Neither is totaly effective for a number of reasons and their cost tends to rise faster than their effectivness (same as car seat belts and airbags).

However in combination as in many engineering problems there is a “sweet spot” where their combind efforts produce results greater than the sum of their individual effects.

But… and it’s a big but you need to consider the transmission channels available to an attacker and how you suppress them. As you note,

“you pull the LAN cable out.”

To remove one channel (and also turn off the wifi bluetooth and unplug the power cable).

But some you cannot see like re-bar in the wall, steel joists, cables etc where an EM signal “close couples” and gets carried considerable distances.

In some cases a loop antenna might well be put in an environment and not immediatly be recognised as such (outer of the TV antenna cable and ground in mains wiring forming a closed loop around a hotel room likewise metal framed windows). Thus distance from such objects is desirable as is using your laptop in random places in public areas (but never the same place twice and keep moving).

But that only deals to a certain extent with low frequency EM there is still IR and visable light to be considered. Which brings me around to your comment,

“A big foil bag tented over it, or perhap a copper mesh box with openings for hands”.

A big foil bag over you and the laptop will help suppress IR and Visable light but not real be effective with low frequency EM which an appropriate metal mesh would be better at.

With care most laptops can be taken appart and appropriate metal mesh put inside the outer clamshell casing and be connected back to the battery ground. However care needs to be taken as it could also act as a radiating antenna if not done properly. With more care the mesh can be put across the keyboard area (take the key caps off and punch holes in the meash to alow the plastic key cap sprues to go through) and with the right mesh put across the LCD screen and still be seen through (this has the extra advantage of vastly reducing the screen viewing angles). Further suppression can be gained by the use of ferrite and carbon loaded foam and low density plastics, these can usually be easily molded to the case as well (but not the screen) and can be optained as “100 ohm foam” for ESD protection for storing IC’s on or anechoic material for RF test chambers.

There are also software solutions such as “TEMPEST fonts” to reduce screen emmision and disabaling the loading of peripheral drivers.

Sound and vibration can usually suppressed via the use of graded materials that absorb achostic energy.

But these are all only partialy effective thus you therefore need to consider “masking” the signals your laptop produces.

To do this you need to find out what sort of signals your laptop emits this involves the use of an anechoic test chamber and suitable sensors for both EM and acoustic energy and test equipment to analyse the individual signals. Having done this you then make a suitable generator that produces about 100 times (+20dB) the output.

There are also software solutions that produce random (white) noise in the speaker and “champing jobs” that cause random CPU, memory and I/O activity.

However masking has it’s downside effectivly you are adding a random signal to the signal an attacker is looking for. And the attacker should not be able to predict it or null this noise signal otherwise they can remove it from the signal the actually want.

An early example of predicting was the cars provided by the Russian authorities for visiting dignitaries, they helpfully put in a radio knowing that the dignitaries had been told that talking quietly with a loud music station playing would “confuse the hidden mics” and stop the conversation being recorded. However the Russians used multiple track recorders where one track was wired in from the radios speaker terminals to provide a canceling signal. And the three or more mic channels could be put through various processess to remove other audio interferance.

Which brings me on to your second question,

“How close does someone need to be to detect side channel information?”

The answer is difficult because of the complexities involved with such things as “near field” effects and coupling to transmission channels as well as the actual level of emissions from your laptop.

Also what the attacker can do out of sight. For instance if you are in a hotel room and your laptop emits a usefull VHF or UHF signal (and the all do these days) the attacker can use two or more high gain antennas and use “space diversity” effects to pull your laptop signal out of the general background noise. Like wise with audio and traveling wave mics.

As for your past friend,

“I once had a friend who actually made a small tent sized wood box lined with copper sheeting to sit in. He didn’t use computers- this was just to prevent cosmic rays from getting him.”

Most “cosmic rays” and quite a bit of other EM / magnetic energy would pass through copper sheeting and him(?)….

Clive Robinson • August 9, 2010 5:21 AM

@ pdf23ds,

“Clive, I’m disappointed you took the OTP idea seriously.”

As a hypotheses I have taken it no more seriously or less seriously than any other presented here and have added an evaluation (not that favourable to OTP) which you will see if you go back and read it again.

The problem we have is,

1, A large bag of bits with a name attached that supposadly “is” something to do with WiKiLeaks, everything else is hypotheses which we cannot test (ie unsound scientific method).

2, We further have assumptions about how many documents or other data (17,000 to 200,000 cables etc) the “front man” of WiKiLeaks might or might not have.

3, Likewise the assumptions as to if the “front man” behind WiKiLeaks actualy believes his life / liberty is in danger from atlest three different sets of people (US Gov, Taliban, Families of Afgan agents) and possibly various non governmental entities in the US and other places.

4, There are also statments made by others that the “front man” behind WiKiLeaks has seen his funding dry up.

5, And likewise supposadly the monies sofar donated to WiKiLeaks has paid for the “front man’s” (lavish?) life style abroad.

As a series of not necessarily true or related items you can spin and weave a whole heap of unfinished broad cloth and then cut your sails to suite from it.

As the Newtonian Scientific method can not be applied (no measurands to test) you are left with the methods of Aristotle and his predecessors, which as a scientific method was so lacking we got a whole load of subsiquent limiters such as Occam’s Razor.

However when dealing with humans and their endevors the likes of Occam’s Razor do not apply (just apply it to just about any game man has ever invented to see why).

However the way of Aristotle is supposadly also the way of the detective with different adagies applied such as the fictional Sherlock Holm’s “after eliminating the impossible…”

So you have to examin each and every hypotheses no matter how improbable and only disgard it if you can show it’s impossible.

Then you rank each surviving hypotheses against a set of measures to see which is more probable than others.

You then set about trying to confirm or refute each hypothes, whilst also adjusting your measures against the light of new evidence or hypotheses.

As we know what is considered to be “evidence” is at the best of times a probable fact open to interpretation or at worst pure invention with no basis in reality we have considerable problems.

It thus descends into the joys and entertainment of a “who dun it” weekend or “rag chew over a brewski on the front step”, and it would be inadvisable to treat as anything more serious.

Clive Robinson • August 11, 2010 5:25 AM

@ Nick P,

“It’s basically the security through diversity concept that’s been getting traction in the academic community”

About time it has been known and proved over and over again (especially with 3-DES).

Claude Shannon pointed out that multiple uses of “transposition followed by substitution” would strengthan a cipher by more than the sum of their individual parts. Latterly the ideas of “Confusion and diffusion” where realised to apply in a more general purpose way.

Ultimatly the only known way to crack any cipher (including OTP) is to mount a “British Museum” or “Brut Force” attack and go through every possible key.

However it’ success depends on knowing when you have the correct key from examining the deciphered output. Thus simplisticaly if you cannot recognise the correct deciphered text “Brut Force” fails.

The question of “recognition” then arises in two ways,

Firstly even a very simple cipher either on the plain text or cipher text renders the recognition process to expensive to make a Brut Force attack viable.

Secondly is the issue of differentiating multiple plain texts.

The usual assumption is the first deciphered text that makes sense is the correct key and you can stop (however with OTP all plain texts are equi probable so…).

All of which has been known if not painfully obvious to the Open Academic community since the DES competition (and almost certainly since the second world war for a great number of people many in academia).

Which begs the question “why now” with the academic community?

I’ve certainly been very well aware of the implications since the mid 1970’s and have applied the concepts where required in many designs. Which is unusual for an engineer where simplicity and clarity are usualy strived for beyond all else (guess you could say either I’m an unusuall engineer or have worked on unusuall engineering projects or both 😉

As you say,

“Diversity is how. nature builds survivable systems.”

Importantly is the consequence of this with respect to “efficiency”…

Evolutionarily the more efficient an organism is the less tolerant it is to changes in it’s environment and the less able it is to respond. We call this an “evolutionary culdersac” and the much over given (and incorrectly attributed) example is the “saber tooth tiger” being an “over specialised hunter”.

As many are aware I’ve been banging on about “Efficiency-v-security” for a very very long time 😉

I can assure any budding engineers out there that efficiency is a very double edged sword…

In one respect it results in over specialisation a lack of resiliance and as a consequence gives rise to “brittle” designs (think about the longevity of Victorian steem engines -v- modern very short life high efficient engines).

In another it can result in systems that are over lumbering designs where inertia makes them slow to respond and thus very susceptable to more agile systems.

The trick as in all things to find a balance or “sweet spot” where you get a desired response in a sensible way.

As a rough rule of thumb small light general purpose standard parts (like nuts and bolts) are the best way to go. But not to general purpose this results in unnecassary waste. Thus you have a range of “standard parts” from which to select (as seen with various types of fastenings not just nutsand bolts).

Which brings me nicely onto your point,

“I’ve found this useful in designing security systems, as I can increase the assurance of a system with low assurance COTS components”

It’s a case of “sweet spot” finding in any given environment at a point in time. The COTS parts can be easily changed should a more demanding or different requirment arise from the change in environment.

And the best way to ensure you get “sweet spots” is “standardization” and well designed “frame works” into which “standard parts” fit easily and conveniently.