Schneier on Security

Clive Robinson • June 20, 2008 8:38 AM

Not the only way to be underhanded with RC4 or any other stream cipher and Coding review bosses that do not know much about being sneeky 😉

I once wrote a stream cipher system that leaked the key in the parity bit of an ASCII Octet.

The main trick involved using a minor problem with most C compilers libs in that when you get memory for a buffer it almost never gets cleared befor or after use.

So if you have a sub in which you create a buffer to get/build the “key” in then before returning you free the buffer all looks well to most code reviewers…

However in most C compilers the lib implementation for memory heap handaling does not effect the memory contents (efficiency strikes at security again 😉

Therefore the original buffer contents stay in memory (heap) untill the programer either makes a mistake (freed memory reuse) or requests another piece of memory that is the same size or less than the original buffer size.

Therfore (easiest way) if you then ask for exactly the same amount of memory you get handed the same block of memory back with it’s contents unmodified.

Which means that as the original buffer held the “key” you have just passed it off by “sleight of hand” into another unrelated buffer (Opps 8)

The hard part was coming up with a convincing argument to modify the parity bit “randomly” that the code boss would swallow without question (hint show them lots of stuff on breaking LRNGs by examining the output and pointing out that the parity bit effectivly is known output from the stream generator)

Oh the reason for using the parity bit, most coders never check it, they just mask it off, therfore even when leaking the key it remained compatable with an older existing implementation (regresion testing can be a malware coders friend)…

The whole point of the excercise was to show to another party that unlike the coding bosses assertion proper “security reviews” where not the same as “code reviews” and one heck of a sight more important.

Clive Robinson • June 21, 2008 9:23 PM

@ Andrew Garland, B-Con,

Your initial lack of understanding indicates that perhaps you are part of a younger generation of programers that have been fed “high level” programing…

If you have read some of my posts to this blog in the past you will know that I am of the opinion that “software engineer” is a false statment and that “software artisan” or “code cutter” is more appropriate for modern Comp Sci grads.

The reason for this is that it used to be the case that engineering subjects where taught from the base fundamentals upwards. And that programing was (pre 90s) likewise taught from the bottom up.

That is knowledge of basic logic, direct and indirect pointers, structures, single and doubly link lists, circular buffers, fence post problems and passing by ref / val and their merits that once where the staple of Prog101.

However the need to “bring on” programers now means that Prog101 is at best about partial knowledge of abstract ideas such as objects, inheritance, overloading and code reuse but primarily about how to drive the latest IDE.

The result is that the knowledge of your average graduate Comp Sci programmer is extreamly brittle and usually non portable from IDE to IDE let alone from programing language to language.

The reason is that your “code cutter” graduate has been trained in the latest CV “must have” language and it’s marketable features of the IDE and other support tools,not the fundementals. Presumably under preasure from industry or politicos (both true in the UK).

Thus the journyman “code cutter” lacks the ability to “stand back” from something that is new and radicaly different and relate it to what they know simply because they cannot take what they know back to the fundamentals and build it back up again in the new “methodology”.

A further and realy significant issue is that they are all but usless at debuging below the “compiler warning” level.

I have stated that I would rather take on a physics maths or electronics graduate that has done a couple of years “real time” assembler than a comp science graduate with 5+ years in a high level language, and not even interview them if they only had “object” language skills on their CV.

Unfortunatly during the 90s the rot seemed to set in to other engineering subjects such as electronics marine and aviation engineering and laterly civil and process engineering subjects where familiarity of the latest CAD/CAM tools not engineering fundimentals was the educational priority.

Subsiquently the rot has also spread downwards to things like primary school mathmatics where it is more important to show the correct use of the calculator or computer than basic addition or subtraction, and a child will be lucky to be shown how to do basic longhand multiplication let alone division with fractional numbers and differing signs.

Part of this is that teachers do not actually have knowledge of the subject over and above the books they teach from (I recently found out that the maths teacher at my sons school did not know about the rule of nines or other mental math short cuts and had never heard of Napiers bones or a slide rule).

Worse I know of quite a few Comp Sci graduates who cannot show why the floating point multiplier in a general purpose processor and a digital signal processor work in exactly the same way (ie integer multiplier and it is just the way you select which output bits to use) just because they have never been shown how to multiply numbers long hand. And worse have no knowledge of the difference between logical and arithmetic shifts and why you would use them.

And appalingly when asked to do simple mental math some engineering graduates do not apear to understand how to deal with numbers with different signs.

Oh and the most ignoble claim to fame that most modern Comp Sci graduates appear to suffer from is explaining how you do subtraction by addition only (look up ones / twos complement). And why the side effects are so important to the likes of computer security.

I guess that is why the more fun programing jobs go to graduates from other subjects such as physics maths and to a lesser extent electronics.

Any way my appologies for being so long winded on this but the problem realy gets under my skin.

Clive Robinson • June 23, 2008 7:56 AM

@ GordonS,

I’m well aware of just how different the Open University is as a pioneer of distance learning it is substantialy different from attending Universities.

Of attending universities if you go to Camb/Oxford and quite a few of the “original unis” then they derive a substantial part of their income from research grants and their teaching is aimed at those who are going on to do research. They would see “dumbing down” courses as a disaster and they see it as an absolute must to maintain their reputation as “research” institutions.

The newer Unis (Old Polys etc) have a very different financial model which is “bums on seats” and vanila teaching, and they derive very little of their income from research funding. They need a high turnover of students or other non Government derived income to survive.

As I noted the Open University is in an almost uniques position, it trys to maintain a very high accademic teaching standard similar to that of the Old Unis for all students as well as doing a disproportianatly large amount of research for a University that is bassed substantialy around distance learning.

However at the end of the day the majority of places on CompSci degrees in the U.K. are in the “new Unis”. That as I said above derive little or none of their income from research grants and do not aim to attract the cream of students and only realy offer a qualification. Often their finances are (of necesity) topped up by either forign students or courses sponsored by major employers who obviously have a say on course content. As part of the process they need to “Market” to attract in students and often the only difference they offer is what “buzz word” teaching they do.

I fully expect this situation to get worse as Tony Blairs dream of getting 50% of people degrees continues. Basically all but a few Unis will need to pander to the whims of those providing them with their income…

Clive Robinson • June 24, 2008 8:05 AM

@ JimFive

It was not actualy me that posed the evenness question but Andrew Garland,

And Jim if you gave me your first answer of,

x & 1

I would ask you,

What happens if the number to be tested is negative?

Judging by the blank / puzzled / scared looks I have seen 80-90% of candidates would fail without a further prompt of,

If the CPU is a 1’s complement machine?

If you answer correctly without the prompt it tells me you have actually done some low level programing, with the prompt it tells me you are aware of one of the limitations of C and if not (back to your mainstream MS code-cutter shop 😉

When you get close to the metal C is not architecture nutral and the assumption of behaviour only holds true for unsigned integers. Having worked on PDPs and some other odd architectures K&R where well aware of this which is why they left it open.

You are now thinking something like,

“In reality what are your chances of meeting a one’s complement machine these days it must be close to zero with general purpose CPUs.”

However when you get to custom CPUs made from FPGA’s etc to do specific transforms efficiently (in say signal processing or graphics) you do stand quite a chance of bumping into non 2’s complement behaviour.

The reason being that most A-Ds output for zero volts is actualy 0x10… or 0x7F…, 0xF… is the maximum positive value and 0x0… is the maximum negitive value, and D-As output voltage in the same way.

For speed and efficiency the extra operation to normalise to 2’s complement is just not justified, as the overhead becomes excesive…

Likewise the “grey code twiddle” needed for some DSP algs etc might not work with the negative number representation used on your CPU.

“Real Time Hardware” and “Device Driver” programers realy are a breed apart from the norm, and often do a big chunk of their work in assembler to get the I/O or Interupt response times to acceptable levels.

Final question hold your hands up the distance light travels in 1/1GHz (app 30cms or a foot is close enough 8) which is one of the real limits on how you can design electronics and therefore computers (Moore’s Law is just an assumption 8)

Clive Robinson • June 25, 2008 1:42 AM

@ Andrew Garland,

“It may even be harder to answer a simple question”

Definatly, not just for the reason you stated, think in degrees of freedom…

A difficult question is either drilling down for a fairly specific answer or the person asking the question may not know enough to answer the question simply (a failing I have seen at many interviews which an applicant might well take advantage of to spin you a line).

A deliberatly simple question however “is a licence to roam” in that it should test the bredth of knowledge and more importantly give an insite into the applicants reasoning process. Hints are helpfull to both the applicant and you especialy as it enables you to shepard the discussion in the direction you want it to go.

To use an analagy, A pencil is a simple tool, a CAD machine is not. The pencil can be used to draw in many ways you might not have thought of (ie shading by cross hatch / dots / dashes / smudging). The CAD machine only gives a limited set of options that the designer included and only a very very limited way to access any particular feature.

When interviewing people I have a very simple process. I ask an individual applicant about what is on their CV that I have knowledge about, and two simple questions that I ask all the applicants. The CV questions are realy only to alow them to relax and talk about what they have done and give me an idea of their style of answering and warm them up for the two questions .

The first question will be speciffic to the domain, such as the Evenness question above which gives them an oportunity to show their stuff and their reasoning ability within the domain (drill down question). It will have one or two simple answers you would expect all applicants to know and many other ways of achieving the same objective (but you the questioner realy must be very very conversant with the subject).

The second question is to test their reasoning ability and it has little or nothing to do with the domain it intentionaly has no right answer. It needs to be very general and not have limiting social overtones (politics, faith or current events etc) and might be as follows (but please make your own otherwise it will not be as effective),

“You have an important meeting with a client in the morning. When you get in your car and turn the key the engine does not turn over. What do you do next?”

This is usually so open and unexpected that it should give quite an insite into their general way of thinking. Oh and if they answer immediatly then either they have heard the question before or they are a sales person 😉

Seriously expect a slightly puzzeled or initialy blank look or even a slightly unfocused start and ask them other gentle questions to get them into the swing of it. The results are usually worth more than the rest of the interview.

Finaly I always give the applicant feedback on how they did in general. It is probably the most difficult part of the interview process. I do it for a couple of reasons firstly because if I have had a negative feeling about them for some reason it’s an oportunity to sort it out, secondly I have sat there as an applicant myself and I have always appreciated the feedback good or bad on the few occasions it has been given.

The easiest way is to go back to their CV and very gently go through the good and bad bits with them again.