Schneier on Security

Clive Robinson • May 19, 2018 7:46 PM

@ Alyer Babtu,

Is it even possible/how would one measure the entropy of things like dabs of glitter etc.

Yes it’s not only possible but sounds almost trivial when you explain.

You start off by working out the uniform density from measuring volume of epoxy to volume of glitter, to get a baseline.

You then “assume” a uniform distribution and alignment of glitter. That is as though they were atoms in a crystal structure.

You then have to decide where you are going to have a zero-zero-zero refrence point and put in place measures refrence for vector analysise and . Which will be fairly arbitrary much as the earths zero-zero is selegted. Then the rotational refrences for each piece of glitter…

You can actually measure all of this with a low power laser, this is repeatable providing you scratch a positional refrence for it to go on.

Clive Robinson • May 20, 2018 5:53 AM

@ RockLobster, Czerno,

I think Natalya Kaspersky is probably accurate, it makes sense, just as the internet and tor were designed as a communications channel for the intelligence services while allowing public access to obfuscate the traffic,

There is a third parameter you have to think through, which is “The NOBUS Advantage”.

Both the Internet and Tor have a NOBUS Advantage you can see in their design if you look at them the right way…

The original design of what would eventually become the Internet was Gordon Welchman who worked at Bletchley Park during WWII and he was intemently aquainted with Traffic Analysis and other “indistrial surveillance” techniques having either invented them or promoted their use. People talk abou Alan Turing in awed tones, but Gorden Welchman was a step or two up from him, and he went on to become fairly senior consultant wise to the US Dept of Defence and ARPA through his job at Rand. Rand was somewhat like “operation paperclip” in purpose which was to get the best of the best science and technology wise from the beligerants in Europe post WWII to give the US another form of NOBUS advantage. Whilst paperclip was designed to get German and other non allies personnel, Rand was a choice bait offering better jobs and better standards of living to get allied personnel.

The main military radio network system still in use was also designed by Gordon Welchman whilst working at RAND. If you can get to see some of the design info you will see great similarity to the Internet but also some important differences. The first Iraq war revealed the fact that the original ideas behind both systems were problematic when they got into the hands of opposing beligerants, which is one reason the Internet in effect got “toned down” on some of it’s more interesting survivability capabilities (which is one reason why certain types of DDoS attack work on one network but not the other).

The main point to notice about the Internet is that whilst the lower layers have survivability still there higher level protocols such as DNS are actually quite centralised –in the US– easily surveiled –in the US and Five Eye choke points– and even when encrypted easily susceptable to quite basic traffic analysis by those who have the center and choke points instrumented.

When you look at Tor it at first looks great, but you will notice that whilst it can give content security to packets inside the network that is lost when you are a client or server which by design are connected to known gateways with plaintext in and out at the server connected gateway. Even if encryption is used into the server the traffic is still easily subject to traffic analysis, especially if you control Tor design and choke points via instrumentation.

There is a very large misconception about the Internet and the way it is physically routed and controlled on the “backbone” side via the Border Gateway Protocol (BGP) etc.

On the physical side if you actually map all the interconnects out to physical locations on maps you will see that the US is at the center. Likewise for most traffic by far the majority of nodes it passes through –which act as choke points to be easily instrumented– are in areas easily visable to Five Eyes nations. Both directly and indirectly be it by subsea cable or satellite link. Other traffic can get “diverted” just by minor tweaks to BGP information

Thus one of Tor’s Achilles heals which is a great NOBUS Advantage is it’s mixing function traffic all gets to be snagged at the instrumented choke points. Because Tor is “a known network” its traffic is very susceptable to traffic analysis if the mixnet is setup incorrectly. Which in Tor’s case it is, as it goes for low latency and has no traffic or channel padding or store and forward…

Thus whilst the US and Five Eyes can map Tor traffic from the user’s connection to the Tor gateway and cross correlate it to the traffic at the server connection to a gateway to remove the mix-net obsfication, they can also see the size of the data being sent/received that alows them to cross correlate with other users to map out first and second communicating parties or groups.

The same trick works with all web based services including email and shared edit spaces facebook walls etc, and the mapping software works just as well by watching Signaling System Seven (SS7) traffic passively so snags all phone calls… Which is basically all the electronic communications available to most people…

Oh and one of the reasons Google got upset with the NSA grabbing their unencrypted backhaul traffic was the realisation as to what was going on. However just as with Tor’s mix-net obsfication being stripped by Traffic Analysis the NSA will still be mapping traffic on those back hauls and identifying first and second party users and any groups they might form and chucking it all into the same mapping software…

After a moments thought you will realise that whilst the US and FiveEyes etc have that NOBUS Advantage, other Super Powers and nations do not… Though we know China has “accidently” sent BGP commands that caused traffic from other countries to get routed through China unsuprisingly it was picked up very rapidly… Supprise supprise the US likewise manipulates BGP but it does not get picked up in the same way, thus few talk about it and it does not get into the public consciousness.

Worse even the likes of the FBI now have ways to get at Tor traffic thanks to a certain university and it’s researchers. Which must be of some annoyance to the SigInt agencies as it intrudes on their NOBUS Advantage.

As our host and other respected security researchers have pointed out NOBUS Advantage will always fail and others will get an increasing share of the “Advantage”.

Is there anything you can do as an individual against this level of surveillance advantage. Probably not unless you know sufficient about covert radio communications and the OpSec involved with their use.

A modern update to the OpSec is to keep a carefull eye on Automatic Dependent Surveillance — Broadcast (ADS–B) traffic and know how to do passive “offset radar reception”[1] with the likes of home computers and Software Defined Radio equipment that can be as little as 10USD each. Thus proving that a supposed NOBUS Advantage by “technology gap” is likewise quickly eroded.

The need to update the OpSec is due to the smaller versions of Boeing’s RC-135V/W RIVET JOINT surveillance aircraft flying around looking for low power covert communications. Some “air spotting” enthusiasts have proved reasonably adept at spotting their flight patterns and tracking them down to airfields and front companies for certain Intel agencies…

When you combine ADS-B signals info with radar returns you gain an extra dimension of information. Thus if the aricraft does not have ADS-B or has it turned off to try to be a little more stealthy… The results of comparing the data kind of acts as a big red flag, actually drawing more attention to them…

As I occasionaly mention “technology is agnostic to use”, the IC and SigInt agencies have for years “backdoored standards” for their purposes via faux “health and safety” arguments. It is kind of ironic that technology with a very real safety purpose can be used to “out” some of the agencies surveillance activities by “hobbyists” and very very loe cost tech and the Internet…

[1] The idea of passively receiving radar signals has been around for some time now with various radar signals. Oddly perhaps it’s the same age more or less as cracking the encryption on cable and early satellite television, and “van Eck” monitor image reconstruction. Because these “hobbies” share a lot of electronic circuits and ideas. For radar in the simple case you have a pulse transmitter and a rotating antenna shared with the receiver. Because the antenna is shared sync signals for thr receiver display can be pulled from a simple angular encoder on the antenna. The simplest of which is a micro switch at the “0 degree” point and a simple saw tooth oscillator triggered by it giving the CRT defection signals. If you know where the radar transmitter is located and you have a direct path signal from it you can with a little processing treat that as your trigger signal. From this you can work out what direction the radar antenna is pointing as accurately as the radar system it’s self. Thus you can use a moderate gain antenna to receive not just the radar outwards pulse, but also the reflected return pulse from aircraft. Thus with a little simple trig you can calculate where the aircraft is almost as easily as the radar. If you can see multiple radars or you have multiple offset receivers connected to gether you can get quite an advantage via “synthetic appature” techniques, which are also used in the likes of astronomy on “Very Long Baseline” radio telescopes and now MIMO type WiFi systems. If you want a modernish view of such radar techniques,

https://patents.google.com/patent/US20140049423

But you can also find in the archives of “Wirless World” a circuit level implementation from the last century. Or you can go direct to an article on hoe to do it with a cheap SDR,

https://www.rtl-sdr.com/building-a-passive-radar-system-with-an-rtl-sdr/

Oh guess what using more than two receivers you do not need the radar transmitter, you could use the omnidirectional pulsed signals from a WiFi access point to map out your home and people moving in it…

Clive Robinson • May 20, 2018 5:27 PM

@ Thoth,

I am wondering if my following description is do-able in chip manufacturing arena to insert a covert and deniable PUF based backdoor.

I’m still having a think on it.

But from previous conversations with people who have had to deal with amongst other things meyastability issues in chips the answer would be that it is possible to make a PUF that can be backdoored but would get through most inspections.

From my point of view if I can not effectively verify such circuits, I have to ask what use they are to me? Currently I see no compelling use case for PUFs as an end user. Which makes me wonder why some manufacturers are so keen on them. That is what compelling use case is there for them to fit such circuits?

I think you can guess where that thought took me 😉

Clive Robinson • May 21, 2018 3:17 AM

@ Thoth,

my previous thoughts of controlling the SRAM cell gate production in a way that they emit the 1s and 0s as predicted by the semi-trusted manufacturer would be detected if someone manages to get a whole bunch of the chips and test them all and find that the output are almost consistent between the batch of chips.

SRAM cells have always been fairly predictable in start up value when new, and they can also suffer a form of “burn in” over time if they hold the same value for long enough.

A couple of decades ago back last century I was involved with the design of Fast Moving Consumer Electronics (FMCE) in the mobile communications markets. The company I was working for covered from childrens walkie-talkie toys through CB, Amateur radio, Private Mobile Radio, Mil/Pol/Gov rugidized, specialised packet radios for utility companies, and the whole range of radio phones with ranges from 50meters to 5000meters HF to UHF analogue “cordless” through digital such as DECT and celular/mobile phones and satellite and maritime units.

Basicaly if you had a finger to press a PTT or dial with, we had something to fit in your communications requirments, and we even had products for those with no fingers or other disabilities or infirmities including dementia and rescue beacons.

We used to ship so many items we were far from joking when talking about “shipping by weight” not “by numbers of units”. Thus a ton of phones to the UK would be say 600 units and those in the post production shipping side would talk weight on pallets more than they would units and you had to talk the same way to them… Thus a lot of effort was made to reduce weight of the equipment and packaging as weight had a significant cost and time penalty measured in a significant fractions of the per unit BOM price. I still remember the jaw dropped shocked look on a British Telecom project managers face when he was told the cost of changing shipping carton size from six units down to five units to make “stock control” in their warehouse easier. In effect they were getting the sixth unit virtually free as the cost difference between a nonstandard sized carton of five units and a standard sized carton with six units was just tens of cents.

Every one of those units had a microprocessor in it… So we used to get to see very good customer service from one of the worlds largest chip manufacturers at the time, especially when we discovered “silicon errors” which we did from time to time.

On the phone and high end product side each and every unit had an electronic serial number, that back then was set by “cutting diodes” on a far east production line. This was thousands of hours per month wasted production time and had a high “rework rate” both of which were highly undesirable. It was still “the done way” because “that’s the way we’ve always done it” syndrome.

The company was split into several parts and did design in various countries around the world. But solving problems caused by their “problem solving” often came the UK office way, and I would get a note on a box on my desk to greet me first thing or just a note to call someone before it was late in some other time zone half way around the globe… Although nobody actually put a ticking stop watch on my desk with a note you knew there was one running somewhere on such problems (I even got a call from the big boss to help his daughter with her maths degree project, which was kind of scary). Thus quite a chunk of my work in this area produced not just patentable but very valuable trade secret wise ideas (sadly they decided not to patent and the culture in other offices and even customers like British Telecom led to the ideas being “given away” to others, which might be one of the reasons the company has gone).

Any way in one of the far east design offices they were designing a cordless phone for the French market and the French regulators had decided 16bit IDs were insufficient they should be a minimum of 20bits with 24bits being looked on favourably… The production cost would take a major blow with that many diodes, it was time to change the “thats the way we’ve always done it” syndrome, and in a hurry (which is never good with such fundemental changes).

So an engineer in that far eastetn office came up with the idea of generating the ID randomly the first time the unit was powered up, by reading a bunch of uninitialised SRAM locations in the microcontroler and mixing them up. So a kind of precursor to on chip PUFs of the sort you are describing.

The engineer obviously did –understandably– insufficient testing because rather than a simple linear mixing function they did some hair brained nonlinear mix. Put simply they had found a lack of randomness in the SRAM from chip to chip on the production line and had rapidly built some wierd algorithm to correct their deficient test results as the unit had been halted in production… Unfortunatly it did not work hence the call not a note from the big boss…

The problem is if you do read out a bunch of uninitialized SRAM locations from a single microcontroler they do look cell to cell quite random (more detailed analysis shows they partially reflect internal layout structure). Likewise when you test chips from different steppings they might not have correlations. As the engineer was working with a new chip their test parts were from different steppings and some had had the chips zaped by rework tools on the metallization layers etc. Thus their initial test results looked suitably random…

However when the masked programed parts arrived for production they were not only all from the same stepping but also the same mask programing machine. Nearly all of that “randomness” had vanished like morning mist after sunrise…

The engineer’s panicked quick nonlinear mixing fix with hindsight was not going to work so a compleate rework had to be done. Worse no changes could be made to the PCBs etc and any code change had to fit in the same space or smaller than that the engineer had used… Oh and I had around four hours to diagnose the problem and come up with a proposed solution and cut the code blind and if green lighted get it sent off to the chip manufacturer who was holding open staff and machines, so no preasure then…

With hindsight my solution sounds easy if not trivial, but it was actually not for various reasons.

As you may be aware a good source of randomness is the time between pressing buttons by an “untrained” human, the faster the timer runs the more random it’s lower bits will be (roulet wheel effect). The problem is on a production line the operators quickly get into a rhythm and their responses are thus “trained” and the randomness shrinks away as they become like automotons.

Thus the hard part was “disrupting the rhythm or training” which ment making them operators do multiple key presses in response to a randomish stimulus, with lockouts to prevent cheating etc… This product had no LCD just a couple of LEDs. I came up with a solution involving three button presses and a diagnostic code for the LEDs which was actually longer than the timer setup, reading and mixing function code. Then dashed out a three page report on what was wrong with the existing system and a four page report with mix diagram and highly commented source code to replace the old. Dropped it under the big bosses nose and a copy and list of fax numbers to his personal assistant to get the ball rolling. Five minutes later (the boss was a fast reader) phone calls were made by him to five different time zones whilst his PA was faxing and I loaded the source code to a dial in “bulletin board” (the Internet was still to get out of the west back then). After just enough time to get a fresh cup of tea a short phone conference was held. The unfortunate engineer was asked questions about if the code was loaded into a prototype and ready to be tested they responded yes. Other engineers including one from the chip manufacture gave me a short grilling on the code then as you see in those movies about NASA mission control for a launch, a “go / no go” call was made and the ball started to roll. Within an hour or two a hundred EPROM parts were blown and put through production and extended test. Things were as expected and the first few thousand phones rolled off with EPROM parts untill the fresh masked parts were hand carried into the factory from the chip manufacturer… All things worked out and product got shipped on time. Which considering it was a “Just In Time” production and ship process was a miracle.

My reward… I got to do the software for the next French phone that was all singing and dancing with LCD display and multiple cordless handsets… Which when it went to market got a “best product award” from the premiere French consumer product test magazine…

Anyway the long term lesson I took away from it which is relevant to security, is just how “unrandom” uninitialized SRAM in chips can be when they are from the same stepping and production batch.

@RobertT found similar problems with on chip TRNGS having only a few bits of entropy, if you look far enough back in this blog.

Which in the case of PUFs and their close cousins TRNGs means the chip manufacture is mixing in some “secret sauce” somewhere at the very bottom of the computing stack at the FAB plant.

As I’ve noted occasionaly –prior to Intel’s Xmass special– the further down the computing stack a security vunerability is the more devastating the resulting attacks will be due to the bubbling up effect. Worse the lower it is the more difficult it is to find under test. But also as you approach the device physics level the number of people who could spot it is rapidly diminishing…

@Nick P tried thinking up a way to control a chip production process to stop backdoors getting put in by SigInT and other IC entities but he dropped it. He might have some choice comments and suggestions to make if you ask.