OWASP AppSec DC 2012/Dan Geer
Application Security Matters Daniel E. Geer, Jr., Sc.D.
I am here today to talk about application security. By the time I'm done, you may think that I have talked more around application security than about it. Perhaps you are right, but bear with me. I am assuming that each of you have heard all the exhortations to work smarter and harder many times before, and have heard (and will be hearing) lot of talks on secure methods as this meeting goes on.
Application security is a pervasive need because applications are themselves now pervasive. Application security is a critical need because applications are themselves now critical. Over the last three years, much bile has been spilled in the press over institutions that are too big to fail and how it must have been a conspiracy of dunces that allowed those institutions to get that big. If application software in the aggregate can be said to be an institution in and of itself, then it is too big to fail. One only wonders what conspirators the press will nominate when they get around to looking for a new set of dunces to ridicule.
We are at or near the point where it is no longer possible to live your life without having a critical dependence on software, even if you live at the end of a dirt road but still occasionally buy nails or gasoline.
Everyone knows that the amount of data in the world is growing. Where does that data come from? Data are the outputs of software, data are nothing without software, and data volume is now so great that the data that matter are too big to move -- meaning that your only knowledge of those data will be what software tells you when you ask politely. Data-centric activities that are too big to move seem naturally to be too big to fail.
If biologic analogies move you, then I'd ask you to consider data as the body and application software as the body's biggest organ, that is to say application software is data's skin. It is the shape and color of the skin that gets our attention when we are looking for a mate, and, for many, how hard they are looking can be gauged by how much skin they expose. The application software that is data's skin is likewise; the most massaged data is the data with the most exposed application software.
Many say that we know how to "build security in[to]" application software, that it is simply some kind of weak will that explains why we have insecure applications all around us. Others say that if and only if the market likes the application's functionality is it worth retrofitting some security into the application, that is to say that security only matters once you have a customer base.
Applications are the only reason to have an Internet; without them, who would care? At the same time, the Internet was not designed for security -- and may I say "Thank God" for that. If the Internet had been designed for security, we wouldn't be here not because the problem of application security would have been solved at the outset but because the innovation would not have come. The Internet was designed for resistance to random faults, and that design worked. It worked so spectacularly that innovation followed simply because the Internet did not depend on the flawless functioning of every one of its moving parts. It was not designed for resistance to targetted faults, which, as Laszlo Barabasi showed, cannot be done at the same time as you are designing for resistance to random faults. Further, there was no gatekeeper you had to ask permission of to put new services on that Internet. There's no government like no government.
The end-to-end principle was the single most important technical decision made in building out the Internet. By putting no control policy in the network, only transport functionality, the network became useful. The present day drumbeat to put control policy into the network fabric itself is so blatantly stupid that it isn't even wrong. Those who propose making the network itself contain security policy are just another breed of Communists, this time with the effete subtlety that neither our Chief Executive nor our Congress has to nationalize critical infrastructure, they just have to deputize it, by force and in private.
But, for the moment at least, the end-to-end principle generally holds. It is, however, no longer the only principle we need as the nature of applications on the Internet makes defining an "end" less clear than once it had been. Under an end-to-end design rule, the two ends of any conversation get to negotiate whatever security policy they like just as they get to pick any network protocol that meets their needs.
That means application security is an end-to-end design issue, does it not? Well of course it does, but those of us who want to retain the freedom to tinker are running into a headwind and that headwind is the increasing difficulty of defining what an "end" in "end-to-end" means. The reasons for this are several, and all serious.
- First, there are proxies; are they the end to which your security
regime guarantees trustworthy connection? If don't know about a certain proxy exists, does that change your answer?
- Second, what one sees on one's screen may be a complicated merging
of many applications and data sources; which one of them is the end? Are they all ends? Does it matter that you can't tell where the bits came from? Is plural marriage a good idea?
- Third, in the original construction, the word "end" implicitly
meant "that which is trusted," and everthing between the ends was not trusted. In applications today, this is not the case -- trust may be all over the place, some of it misplaced, perhaps, but that is besides the point.
Let me acknowledge here Marjory Blumenthal and Dave Clark whose thinking on the end-to-end principle is especially instructive. As they pointed out, the end-to-end principle made the assumption that the communications system was not trustworthy whereas the endpoints were. That the communications system might be unreliable was and is an assumption that leads to good design choices. However, the idea that endpoints are trustworthy needs some updating.
It may well be that the reframing should be not end-to-end but trust-to-trust. In the original formulation, end-to-end meant machine-to-machine or human-to-human. The idea was that the ends could be trusted but the rest could not. Let's say that because there are new attacks every day I don't trust my PC any more, then my PC can't be an "end" in the end-to-end formulation. Put differently, what began as "You're OK, I'm OK, but the network is dangerous" has become "I hope I'm OK, I have to assume that you are hosed, and the network may make this worse."
This, of course, brings us to the question of what is trust. My definition of (a state of) "trust" is this: Confident anticipation backed by effective recourse. The "confident anticipation" is the day to day operational reality; I commit a job or some data to your computing queue and I am confident that it will return in the manner I anticipate, confident enough that I deploy no armed guards. The "effective recourse" is that I actually do know enough about where that job or data is going that should something go wrong I then would know what next to do to force you to make me whole -- to get an effective recourse. This is the antithesis of saying that everyone is my friend. Rudyard Kipling's poem "If" is a jewel, but the cybersecurity practitioner's couplet is:
If neither foes nor loving friends can hurt you, If all men count with you, but none too much;
Summing up what I have said so far, data is where the value is, winners have the most data in motion whereas losers have too much, applications are the skin on the data -- some more erotic than others, the Internet's existence gives applications their universality and some applications their raison d'etre, the end-to-end principle is about trust placement, trust is not for sissies, and the central discipline of secure design is that of choosing what failure modes you are prepared to tolerate.
At the same time, a design and its implementation can diverge. At least since Hugh Thompson published his famous Venn diagram, we've known that security flaw is in that part of the implementation that was not in the design. It is from that realization that I give you what I know to be the paramount rule of all security engineering: No Silent Failure.
We have nowhere to go but up with respect to a rule of "no silent failure." The Verizon Data Breach Investigations Report shows that data loss is overwhelmingly silent. Part of that silence is digital physics -- if I steal your data, then you still have them, unlike when I steal your underpants -- but the majority of that silence is that there is no programmatic indicator of the data's cloning; it is like a (UNIX) _fork_ operation, fast and cheap. [As an historical aside, the late Dennis Ritchie wrote that the "PDP-7's fork call required precisely 27 lines of assembly code."]
But the most telling legacy of Dennis Ritchie was that C had data structures, data structures that operated at a level that was just barely high enough. I've come to view parsimony of expressiveness as a talisman against silent failure. Let me quote Don Davis, whose code is probably running on every computer in this room:
The network-security industry has produced lots of examples of over-rich expressiveness: RACF, firewall rules, and .htaccess are my favorite examples. I argue that in computer security applications, a language or UI should present a little _less_ expressiveness than expert administrators will find necessary, so as not to help normal administrators to confuse themselves. The problem is that every security rule-set has to be long-lived and to change steadily. If the rule-set's syntax allows for subtlety, then each rule-set's size and complexity tends only to grow, never to shrink. This is because each security administrator will tend to avoid analyzing whatever subtleties have accumulated, and will instead blindly add special-case allowances and constraints, so as to avoid breaking whatever came before. The typical result is an unwieldy rule-set that no human can understand, with unpredictable security holes. Here, as remedy, are two rules of thumb: for security, avoid designing order-dependent syntax, and avoid recursive features, like groups of groups. Such features seem useful and innocuous, but when administrators use them heavily, complexity mounts destructively.
Don wrote that seven years ago. His distinction between programming language and what an administrator uses may now be a distinction without a difference, but that does not disable his point; it strengthens it. PERL and Ruby and Java have too many ways to express the same thing, to do the same thing, and they brag about how "there is always another way." Each of the three are Turing complete. Greater expressiveness seems to be the way things are going.
The work being done by Sergey Bratus, et al., at Dartmouth in the "Langsec" group is instructive here. Quoting from their home page,
The Language-theoretic approach (LANGSEC) regards the Internet insecurity epidemic as a consequence of ad hoc programming of input handling at all layers of network stacks, and in other kinds of software stacks. LANGSEC posits that the only path to trustworthy software that takes untrusted inputs is treating all valid or expected inputs as a formal language, and the respective input-handling routines as a recognizer for that language. The recognition must be feasible, and the recognizer must match the language in required computation power.
When input handling is done in ad hoc way, the de facto recognizer, i.e., the input recognition and validation code ends up scattered throughout the program, does not match the programmers' assumptions about safety and validity of data, and thus provides ample opportunities for exploitation. Moreover, for complex input languages the problem of full recognition of valid or expected inputs may be UNDECIDABLE, in which case no amount of input-checking code or testing will suffice to secure the program. Many popular protocols and formats fell into this trap, the empirical fact with which security practitioners are all too familiar.
Viewed from the venerable perspective of Least Privilege, ... computational power is privilege, and should be given as sparingly as any other kind of privilege to reduce the attack surface. We call this ... the Minimal Computational Power Principle.
We note that recent developments in common protocols run contrary to these principles. In our opinion, this heralds a bumpy road ahead. In particular, HTML5 is Turing-complete, whereas HTML4 was not.
There is a parallel between Bratus' "weird machine" construct and my own training in statistical computation. When you look at the numerical stability of a statistical computation you posit that the result of the computation is correct for some problem and you measure how different the problem that you gave is from the problem whose solution you got. As Young, Boebert, & Kain said, "If a program has not been specified, it cannot be incorrect; it can only be surprising." Those "weird machines" have not been specified, they've been discovered, and that includes discovery of weird machines in standard protocols, which Marsh Ray and I enumerated in our "Vulnerable Compliance" talk.
There is a growing interest in DevOps. As a strategy, DevOps intentionally merges programming and administration, though this merger is not designed for limiting expressiveness but rather to "reduce the scope of changes" based on "the idea that all elements of a technology infrastructure can be controlled through code." If true, the latter -- that all elements of an infrastructure can be controlled through code -- makes cybersecurity the only game in town. As it happens, Sandy Clark's thesis work under Matt Blaze analyzes frequent deployment as a security tool, that is to say how releasing code often enough that your opponents' ability to analyze your code is thwarted by what she calls "the Honeymoon Effect" -- which can be restated as when attack development has a cycle time then you just modify your apps using a shorter cycle time. And then you win. Later today, Josh Corman will work to convince you that it is possible to write and deploy secure code at, as he says, ludicrous speed. Perhaps as you listen to Josh's talk you can ask whether constant sprints at ludicrous speed are in fact (1) inherent limitations on expressiveness thus precluding error, (2) proof by demonstration of the Honeymoon Effect, (3) a way to simply minimize the otherwise long wait for code fixes to vulnerabilities found in the field, or (4) something else altogether.
This brings us to cloud computing which I can't help but think of as a variation on the theme of timesharing. In a blog about how Seattle is building a planet-wide dominance in cloud computing, the author made this assertion:
[There is an] important, often overlooked difference between cloud computing and old-school time-share computing: minimal bureaucracy between users and the cloud. Timeshare typically allocated scarce resources via a bureaucratic process. In the cloud, anybody with $5 can be the IT manager of their own massive compute facility, at least for a little while.
While it is completely unfair of me to pick on that one paragraph, here's where an economist would say: Price Allocates Scarcity. It is not as if cloud computing is forgoing bureaucracy out of some social principle. But to the point of this meeting, I'd be very interested in some one of you doing a thoughtful, real-numbers analysis of the impact of cloud computing's cheapness on code bloat. Chris Wysopal, co-founder of Veracode, has observed that the size of applications tends to rise after they are moved into the cloud precisely because space becomes too cheap to meter -- developers link against any library that contains even one call they care about and, in any case, it's faster to just include everything. Bloat like that doesn't matter to anyone except, perhaps, us security people, and how we are supposed to measure it seems a research grade problem to me.
Of course, the Software as a Service (SaaS) model has something important going for it from our security point of view: when your users stop using a copy of your software you get to stop pleading with them to take their updates -- you can just force it down their throat. Think of that as fluoridating the water supply instead of begging people to not eat sweets. The security possibilities of Software as a Service are real and unmistakable. Marcus Ranum's position these days is that if we were really serious, then we would
Switch the whole planet from "you own this software" to "all software is a service" and put in place an app store model for everything (with the proviso that mission critical systems could opt out if payment was escrowed far enough in advance). Get the whole planet running a small common set of codebases instead of the chaos of crap we're wading around in.
As it happens, I pretty much disagree with all of his comment but it *is* a coherent idea that *does* lead in an identifiable direction that *can* be imagined. It may even be the direction that we (the capital-W "We") are going as by now everyone has seen the figures for smartphone sales versus computer sales. Remember that a cell phone network is not a public Internet and that a device that only works on such a network is not a general purpose computer. This triplet of ideas, that the smartphone is the new endpoint, that appstores are the only suppliers, and that software is a service not a product leads to the end of the general purpose computer as a consumer durable.
I've written about this as has Cory Doctorow; let me read a litte of Cory's commentary from his speech to the Chaos Computer Congress this past December in Berlin:
The triviality of [the] copyright [battles] tell you that when other sectors of the economy start to evince concerns about the Internet and the PC, that copyright will be revealed for a minor skirmish, and not a war. Why would other sectors nurse grudges against computers? Well, because the world we live in today is made of computers. We don't have cars anymore, we have computers we ride in; we don't have airplanes anymore, we have flying Solaris boxes with a big bucketful of SCADA controllers; a 3D printer is not a device, it's a peripheral, and it only works connected to a computer; a radio is no longer a crystal, it's a general-purpose computer with a fast ADC and a fast DAC and some software.
The grievances that arose from unauthorized copying are trivial, when compared to the calls for action that our new computer-embroidered reality will create. Think of radio for a minute. The entire basis for radio regulation up until today was based on the idea that the properties of a radio are fixed at the time of manufacture, and can't be easily altered. You can't just flip a switch on your baby monitor, and turn it into something that interferes with air traffic control signals. But powerful software-defined radios can change from baby monitor to emergency services dispatcher to air traffic controller just by loading and executing different software, which is why the first time the ... FCC considered what would happen when we put SDRs in the field, they asked for comment on whether it should mandate that all software-defined radios should be embedded in trusted computing machines, ... whether every PC should be locked, so that the programs they run are strictly regulated by central authorities.
Cory is laying out for us yet another evidence that cyber security, or at least its tools, matter. In a sense, our longstanding wish to be taken seriously has come; we will soon reflect on whether we really wanted that.
Remember, security is about control and governments everywhere want more of it. Michael Jay Gross goes further in his "World War 3.0" column. Though he does not remind us that the Internet is a ecosystem of applications, he does show that at the nation state level the coming treatment of the Internet will embody the majority of the seven deadly sins. As both Cory and Michael suggest, if you think security technology is important now, you ain't seen nothing yet.
I would suggest that cyber security is like any other ecosystem and that predators and prey evolve in response to each other. Under the usual terminology of evolutionary biology, evolution shows itself as a series of long quiet periods separated by shorter periods of change, what is called "punctuated equilibria." Sentience on the part of the competing life forms only makes the clock run faster; the overall dynamic remains the same. There have been, to my count, three great punctuations that bring us here today. You and I owe the existence of our field to the first, which was the sudden appearance of a TCP/IP stack in Microsoft Windows. That stack was all in all a good thing to have, but it exposed an operating system designed for a single owner/operator on, at most, a private net to a world in which every sociopath is your next door neighbor. The attack rate had a sudden and profound acceleration discontinuity much like that when they light the solid fuel on the Space Shuttle.
The second punctuation came perhaps as much as five years ago when the population of attackers changed over from braggarts to professionals. Because braggarts are paid in bragging rights, their discoveries become common knowledge quickly. Professionals, however, are paid in money and their discoveries are intellectual property, ergo their discoveries do not become common knowledge until they have been milked of all the income they can produce. Put differently, the better our opponents are the less we know about how they work -- a fact true both in cybersecurity and in the intelligence game.
The third punctuation, in which we are presently swimming, is the very thing that Cory and I talked about -- the death of the general purpose computer. This death does very much make Marcus's prescription a likely outcome, but let me remind you that this is absolutely an example of trading freedom for security and if you are widely read, then you will have absolutely zero confusion as to how that trade will eventually play out whether your muse is Benjamin Franklin, Emiliano Zapata, or Edward Gibbon.
If there is to be a fourth punctuation, it is the turning over of our protections entirely to machines. I spoke about this at length in February at the Suits and Spooks meeting and wrote about it in two recent IEEE Security & Privacy columns. The core argument is simply that when everything is connected all the time, the human cybersecurity practitioner is largely a liability, not a failsafe.
The source of risk is dependence, especially dependence on expectations of system state. As a term of biology, the virulence of a disease is a measure of how fast it moves from me to you. Virulence in an infectious organism is, interestingly, a marker for how good your immune system is insofar as if your immune is quick and efficient at killing Microbe X, then evolutionary selection pressure on Microbe X will cause it to move faster (be more virulent) in proportion to how good your immune system is.
Several years ago, I looked up the major virus attacks on cybersecurity and, in particular, looked for the doubling time of the infected population. I plotted the doubling time on a calendar timeline and found a hyperbolic fall-off, i.e., virus pandemic events became ever rarer but for those that did occur, their doubling times fell away as if on a hyperbolic curve. This is exactly what one would expect in the biologic world where an infectious agent and its target species co-evolve. On the one hand, the immune system of the target gets better so episodes of infection become locally rarer over time. Simultaneously, those episodes that do occur involve a mutated infectious agent that displays much higher virulence with each pandemic outbreak.
If you argue that evolution is a force that does not depend on the choice of its participants to play ball but is, instead, operating without either their cooperation or their sentience, then one must conclude that computer infections will tend to follow the same general life cycle of decreasing frequency balanced by increasing virulence.
In other words, by the time a truly virulent strain of something appears, our dependence on the to-be-infected infrastructure will be so complete as to guarantee collapse.
But let us flatly assume that Software as a Service is the future. As someone who sees a lot of business plans, I can assure you that step 5 in nearly all of them involves renting space from Amazon, Microsoft, or somebody and that nearly none of them include in their capital requirements the cost of building out data centers. Software as a Service is very much more likely to be closed source than open -- and may I add that you will have little way to determine if a claim of open-ness is valid when, for example, you don't have a way to compile that source or look at the binaries that the SaaS vendor is actually running.
Closed source can be helpful to security if the programmers take security seriously, and harmful if they don't. While Eric Raymond is all but entirely correct that "given enough eyeballs, all bugs are shallow," he is nevertheless wrong that this is a fact-of-Nature requirement on how to build code if, in particular, the user cannot actually see the code. SaaS, in other words, can single-handedly demote the decompiler as the best attack tool and put the fuzzer in its place.
When you believe that your code won't actually be seen by its users because they are only buying it as a service, your tendency will be to compete not on ease of installation, update, field supportability or integrability, but rather on performance and the latency of re-configuration. Your code will get more idiosyncratic and possibly more clever. Two engineers who have built big systems have something to say about this; first, Mike O'Dell, the founding Chief Scientist of UUNet and now a venture capitalist, said:
Left to themselves, creative engineers will deliver the most complicated system they think they can debug.
while Brian Kernighan, the co-inventor of C, said:
Debugging is twice as hard as writing the code in the first place. Therefore, if you write the code as cleverly as possible, you are, by definition, not smart enough to debug it.
If I were to walk up to you and say that you must open your machine to RPCs just because I say so, then you would kick me out of your office. But that is precisely what is being asked -- to accept RPCs all day, every day, so that the showmen in both industry and government can deliver their vision of the all singing, all dancing "user experience." A cavalcade of RPCs can certainly be configured to be not only performance enhancing but security enhancing, but security enhancement is not the default outcome.
Instead of calling me a crank, at least for the moment, remind yourself that when, in the name of security, we "lock down" an operating system, we do so by removing functions, by reducing the choice set of what might be running, by shrinking the attack surface. The reason that the Web browser is the leading entry point for malware is the number of choices that a browser offers up to whomever is at the other end.
Lockdown is just one aspect of what someone might do in the name of hardening. Let me make a point about that, however. Using the terminology of metallurgy, we can harden in one of two ways. On the one hand, we can pick a layer in the stack or wherever and case harden that. If we make the whole product hard, it becomes brittle -- we have to make only the surface hard. On the other hand, we can make the metal shatter resistant but at the price of it being malleable enough that if I hit it with a hammer it certainly will not shatter but it will ding. Which kind of hardening we want depends on the kinds of impacts you expect it to be exposed to when in actual use.
Auto-update of software is a great thing so long as it is capable of dealing with local anomalies. Generally speaking, you get a better result if you don't try to analyze too much, just replace the whole software package. This is a strategy that is, to a degree, hardening of the embrittlement sort as a regular auto-update that does what is in effect synchronization means that both update and repair can use the same tool and you don't really have to know whether you are updating or repairing. However, if anyone else that doesn't like you ever gets control of your auto-update mechanism, then it will be hard to ever pick up the pieces well enough to truthfully say that you got back to where you were before the strike.
Just don't forget that total cycle time for a round of updates matters. The coming Smart Grid, is, after all, an application layered on top of the biggest machine in the world. Kelly Ziegler's numbers indicate that should it be necessary to do a total update of the firmware for all households on a fully deployed Smart Grid it would take a year or so. What might we do differently?
A strategy of intrusion tolerance is different. It begins with the assumption that your software package will be dinged often but, if you did a good job in design, the dings won't make the package unusable. This is an uglier process in practice, but under some scenarios it is more survivable. Any of you who work for large firms will have had some visibility if not input into your disaster recovery plan. If that includes your software base, which I hope it does, then the DR plan probably includes mechanisms for diminished operation, which is precisely what I am talking about. In a way, the Microsoft Address Space Layout Randomization (ASLR) is exactly a strategy for intrusion tolerance; you may still get dinged but the attacker will have less probability of getting a shatter. Just an ugly ding.
Perhaps the most important thing you can do for us all is to ensure that there is no silent failure. This means instrumentation, it means well designed surveillance regimes, it means an attention to the kind of metrics that come out of an airplane's black box, it means keeping things simple enough that, well, there are fewer surprises, and it may mean changing how you think about how you make tradeoffs. Repeating Kernighan, if you write the code as cleverly as possible, you are, by definition, not smart enough to debug it.
Finally, because security is not composable (and may never be), be very careful where the code you reuse comes from. Every time I see a page larded up with more domains than I have fingers, I plan never to visit them again. I know you have to compete; I ask that you not end up in a race to the bottom.
And, perhaps above all else, remember that all security technology is dual use.
There is never enough time. Thank you for yours.
"Basics of the Unix Philosophy," www.faqs.org/docs/artu/ch01s06.html
"Build Security In," buildsecurityin.us-cert.gov
Barabasi L & Albert R, "Emergence of scaling in random networks," Science, v286 pp509-512, 15 October 1999.
Bratus S, Locasto ME, Patterson ML, Sassaman L, & Shubina A, "Exploit Programming: from Buffer Overflows to Weird Machines and Theory of Computation," USENIX Association ;login:, v36 n6, December 2011.
Clark DD & Blumenthal MS, "The End-to-End Argument and Application Design: The Role of Trust," TPRC, 2007.
Clark S, Blaze B, Frei S & Smith J, "Familiarity Breeds Contempt: The Honeymoon Effect and the Role of Legacy Code in Zero-Day Vulnerabilities," ACSAC, 6-10 December 2010.
DARPA internet program, RFC 793, "Transmission Control Protocol Specification," September 1981.
Davis, DT, personal communication, 27 January 2005.
Doctorow C, "Lockdown: The Coming War on General-Purpose Computing," keynote speech, Chaos Computer Congress, Berlin, 26 December 2011.
Farrow, R, "Internet Explorer is the most dangerous program ever written," 30 June 2004.
Freedom to Tinker is hosted by Princeton's Center for Information Technology Policy. freedom-to-tinker.com
Geer D, "People in the Loop: Are They a Failsafe or a Liability?," Invited Talk, Suits and Spooks, Washington DC, 8 February 2012. Geer D, "Power. Law.," For Good Measure, IEEE Security & Privacy, v10 n1 pp94-95, January/February 2012. Geer D, "More or Less," Clear Text, IEEE Security & Privacy, v10 n1 p96, January/February 2012.
Geer DE & Ray M, "Vulnerable Compliance," USENIX Security Symposium, Washington, D.C., August 12, 2010. www.usenix.org/events/sec10/tech/tech.html#Geer
Gross MJ, "World War 3.0," Vanity Fair, May 2012.
Kolsek M, personal communication, 10 February 2012.
Nilsson E, "Seattle's Growing Advantage in The Cloud," 3 August 2010. www.xconomy.com/seattle/2010/08/03/seattles-growing-advantage-in-the-cloud
Raymond E, "The Cathedral and the Bazaar," O'Reilly Media, 1999.
Ritchie DM, "The Evolution of the Unix Time-sharing System," Bell Laboratories Technical Journal, v63 n6/2 pp1577-1593, October 1984.
Saltzer J, Reed D, & Clark DD, "End-to-End Arguments in System Design," ACM Transactions on Computer Systems, v2 n4 pp277-288, November 1984.
Thompson HH & Whittaker JA, "Testing for Software Security," Dr. Dobbs Journal, v342 pp24-34, November 2002.
"Web Pages Are Getting More Bloated, and Here's Why," royal.pingdom.com/2011/11/21/web-pages-getting-bloated-here-is-why ( above based on "Trends," www.httparchive.org/trends.php )
"Verizon 2012 Data Breach Investigations Report," 22 March 2012.
Wysopal C, personal communication, 12 January 2012
Young WD, Boebert WE, & Kain RY "Proving a Computer System Secure," Scientific Honeyweller, v6 n2 pp18-27, July 1985.
Ziegler K, "Smart Grid, Cyber Security, and the Future of Keeping the Lights On," USENIX Security Symposium, August 13, 2010.
They that can give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety. -- Benjamin Franklin
Better to die on one's feet than to live on one's knees. -- Emiliano Zapata
In the end, more than freedom, they wanted security. They wanted a comfortable life, and they lost it all -- security, comfort, and freedom. When the Athenians finally wanted not to give to society but for society to give to them, when the freedom they wished for most was freedom from responsibility, then Athens ceased to be free and was never free again. -- Edward Gibbon