Schneier on Security

Clive Robinson • October 28, 2010 12:30 PM

@ BF Skinner,

“Here’s why I think it likely they weren’t encrypted inflight. Cost. Key management costs a lot. Each device would have to have a key, each key would have to accounted for through it’s life cycle(creation, use, escrow, change, destruction). Think in terms of thousands and tens of thousands of clients”

There are also the technical issues of HD encryption.

Without going into messy details it is a difficult problem for a number of reasons.

Often the simplest solution used is an Inline Media Encryptor (which the NSA make available these days) However these have big disadvantages as they are usually “effectivly one (master) key” in operation and usually have absolutly no knowledge of how data is stored on the disk above sector level.

Idealy any HD encryption should be as a consiquence of the application & the user ID and the OS & the File System in use as well as the individual sectors.

It can be done (I’ve done it) but it is a real pain to get right, oh and of course a certain well known and popular OS manufacture does not make sufficient information available (to mear mortals) in a manner to make it viable as a product (and believe me they will send a flock of legal vulture that makes Hitchcoks “the Birds” look tame in comparison if you do try…).

Clive Robinson • October 31, 2010 5:07 AM

@ Nick P,

“How much does the equipment and software needed for your attack cost?”

There is more than one type of attack.

A quick bit of ancient (in computing terms) history. there where many different types of storage media tried before the use of integrated circuit memory. Two that you may know about are “ferrite core” memory and “CRT Storage tube”. Both suffered from “residual data” issues that fell into two proad categories,

1, Non permanent retention.
2, Permanent deformation.

In the case of ferrite core memory the first was partial magnetisation of the core and was an integral part of the design. In effect you used the core like a tiny soft iron magnet by placing a large current pulse in one direction it became a weak magnet pointing in one axsis or it’s opposit depending on which direction the write current pulse went around the core. The direction of magnetisation could be read out later via the sense wire mechanism. However If you repeatedly wrote the same direction to the core you would eventualy turn it from being a very soft magnet to an almost permanent magnet in that direction at which point the core could not be pulled back by the ordinary write wire mechanism, and the data state was effectivly permanently writen in.

Likewise Storage tube CRT’s these held an illuminated spot on the screen by the use of what amounts to static charge and low level flooding by electron beam. They had a tiny grid inside the tube just behind the phoshor screen that could hold discreet packets of charge almost like miniture capacitors that leaked their charge to the phosphor screen and kept it illuminated after the main electron beam had passed. In order to maintain the bright spot the charge had to be “topped up” by a flood beam that provided about the same charge into the grid as was leaking out to the phosphor.

Even if you turned the flood beam off the charge would remain for a very long time and in a dark room the phosphor could still be seen vaguly emmiting light many minutes and sometimes hours after the power was removed from the tube and could with care be recorded with a camera with it’s shutter deliberatly held open for several seconds (and yes such screen recording cameras where quite common in electronics research labs up into the 1980’s). However there was a downside to CRT storage tubes the suffered very easily to a from of permanent “burn in” all CRT tubes suffer from often called “screen burn”. That is the grid or the phosphor would become permanently damaged and could be seen from then onwards as a change in brightness on the screen as the electron beam went over the effected areas.

Now the important thing to remember is to some extent all memory devices and many transducers suffer from both these two problems. For instance incandescent (filament) light bulbs suffer from metal migration, run them on DC and they will (on average) fail a lot sooner than if you powered them from AC. Silicon chips suffer from both problems as well.

Silicon memory storage elements generaly come in one of two conceptual forms,

1, A latch made from two NOR gates with crossed over feedback (ie output of gat A to input of gate B).
2, An insulation gate Field Effect Transistor (FET) with a capacitor between the gate and source on which charge is held.

Due to silicon “real estate” issues the first is not realy used much these days.

So the first question that arises of just how long can the charge be held on the gate capacitor?

The answer depends on the design but in the case of EPROM / EEPROM / FLASH the answer appears to be tens of years. So the residual charge time on a RAM chip can be a very long time. For instance back in the early 1980’s I had designed a small Z80 CPU card and it used Dynamic RAM (DRAM) which requires to be refreshed every few milliseconds.

The third prototype had a problem it appeared to work fine during testing but would go wrong after about 3/4 of an hour for no apparent reason. The fault was eventually found to be a break in the PCB track on the refresh circuit. Now even alowing for the fact that ordinary memory writes effectivly did a memory refresh we found that the data appeared to remain good for atleast ten minutes…

Also Remember this was without any spescialised equipment and at around 25 degrees C or ~77F (it was summer and in the UK we did not have “AirCon” back in the 80’s you were lucky if you had a window you could actually see out of due to that frosted safety glass with wires in and where I was working for security reasons they had even put a “white wash” paint coat on the glass…)

So (1) Non permanent retention can be a lot longer than you would think even without the use of Liquid Nitrogen. And RoberT has detailed some of the ways it can be got at.

But (1) presupposes that the Chinese where ready and waiting with all the bit’s required to take advantage of the “time window” of the aircraft landing. Does this sound likley if it was an accident that brought the plane down?

Which kind of suggests that (1) was not the route they used (I don’t know if CO2 fire extinquishers would lengthen the time significantly).

But what of (2) Permanent deformation, well it’s not just light bulbs that suffer from metal migration sillicon chips do as well. As a simple rule of thumb all normal conductors have IR losses which means they are potential fuses, that is at some current the conductor will actually melt. Long before this point however the metal will actually “go with the flow” of current how much depends on a number of things, one of which is the total charge transfere (effectivly current multipled by duration).

However the net effect is under a DC current the conducter will get thinner at on end, and fatter at the other, this is effectivly permanant deformation as sillicon chips as a rule don’t like having AC applied to the power rails ( at the very least it kind of messes things up internaly as it goes through the memory cell bias zero crossing point which can and has be used as an attack vector).

Importantly at the point where the conductor is getting thinner the resistance and therefore IR loss goes up. And there are a number of ways this can be seen/exploited (the simplest for ordinary conductors is by the slight increase in temprature at that point seen as an infrared hot spot). Temprature is not the only method and I suspect it is not practical with modern memory chips.

However as RobertT noted metal migration in the drain and source circuits etc is not the only permanent damage a “static” data state inflicts on a memory cell within a memory chip.

One method that has be used by “hackers” in the past on the “supposadly erased” SRAM in “Satellite” and “cable” set top boxes and the like is to power the chip up slowly at different tempratures and read the data out. You can do this over and over and average out random state changes and find the state a memory cell prefers to be in (which might be indicative of the state it was almost permanently in or just the chip layout).

Again I don’t know how well this will work with modern memory chips but this is not the only way it can be done. One simple enhancment to this simple method was to add a small AC waveform to the DC power supply (google “A-D Ditthering” for why it works).

If it works sufficiently reliably this is a very very cheap in equipment terms attack, requiring little physical complexity thus could be done by an undergraduate (it is however time expensive).

Now the questions that follow are what are the other methods open to exploitation? and are they known to the US and/or the Chinese?

If you have a number of people activly designing memory chips and activly researching the physics of the materials involved what other methods can they come up with for reading the (2) Permanent Deformation.

It is known that in times past the NSA was the leading light in semiconductor research, and some of their employees / associates left to form their own companies to further develop technology for them. Super Computers are perhaps the most well known area but telecomms as well (like CELP voice codecs etc).

Importantly it is suspected that the NSA are still the worlds experts on data storage simply due to their own rapacious needs (Never Say Anything, but store everything you see and hear ;).

But you also need to remember that you find a large number of Chinese and ethnic Chinese researchers all over the globe in organisations researching into both security and chip development. And further that both China’s have used such people in the past to “steal secrets” at all levels (think about where the expression “Chinese knock off” comes from and how “China got the bomb”).

So it might well be that the Chinese have the same or more knowledge in this area with regards to the US and other WASP nations.

All of which muddies the water quite a lot as to how China might have got the data together.