(A)bort, (R)etry, (I)gnore: Perspectives on Information Security

To the tune of U2′s “God Part II”:

I don’t believe in the ACM
I don’t believe in the IEEE
I — I’m going to set my mind free

I don’t believe in publication counts
Don’t beli’ve in acceptance rates
I — I’m going to ignore PC gates

I don’t want to bloat my CV
Don’t want to have a Dean to please
There are outcomes other than these

Rewrite without GOTO’s:

http://lxr.linux.no/linux+v3.1.3/mm/mmap.c#L1181

Go ahead, I’ll wait.

http://www.cbc.ca/news/canada/british-columbia/story/2011/07/18/bc-nexus-pass-smuggling-border.html?ref=rss

“I’m shocked; shocked to discover gambling in this establishment!”

If you mark people as trusted, some of them will take advantage. It’s human nature. And the answer isn’t more enforcement. If your protective trade tariffs and domestic prices are much too high, people will shop elsewhere. And you can catch people buying engagement rings, but I’m not sure you’ll catch real smugglers, human traffickers, and drug runners.

I recently made and posted a Youtube video of my 2 year old son getting a pat down at airport security.

I figured I should provide a few words of context here to clarify my intent and the circumstances surrounding the video — and what security lessons we should draw from this incident and others like it.

Security researchers know, as commonly accepted wisdom, that no security system, digital or otherwise, is 100% foolproof or secure. Bad things will happen. Malicious activity is impossible to prevent.

Meaningful security is therefore about managing risk. We, particularly as citizens of a free and open government, should actively question the cost of security mechanisms imposed for the purposes of managing that risk and keeping us safe. If the risk is low and the cost is high, we should find a better, alternative security mechanism (or change our policy to make better use of the existing mechanism, or change the policy to even consider other mechanisms, or none).

In any event, the video I recorded is meant to help us ask this question: is patting down toddlers a useful security mechanism? If not, what policy changes should we consider to improve the value we are getting for the investment in airport security measures?

The video shows 44 seconds of what was about a 30 minute episode, so a lot of context is missing, including the ho-hum standard screening that led up to the pat-down and the discussion with various CATSA employees and law enforcement that followed asking me to delete the footage (and finally deciding that it was OK for me to have recorded it and retained it). We submitted to routine screening and were cooperative the entire time (modulo my refusing to hand over my already-scanned laptop with the raw footage while the matter of whether I was allowed to record that footage was still pending).

One of the chemical sensors detected something (no idea what substance or in what concentration) on our baby food jars. This alert triggered an automatic escalation to re-screening and a choice of a patdown or the X-ray machines (it is impossible to tell what X-ray technology they are using, how it is calibrated, whether rigorous independent testing is performed, etc., so we chose the pat-down, knowing that we would be denied boarding if we didn’t comply). At this point, a CATSA employee (for those in the US, the Canadian equivalent of the TSA) gave my two year old son a quick pat-down. Knowing the pat-down was coming, I opened my laptop and started recording.

I emphasize that the agent was quick and courteous, and did not hurt my son (he later gave me a pat down and was also quick and friendly). I have no problem with the CATSA employees — they are just being asked to complete their jobs and carry out the security mechanisms that policy puts in place.

However, I still think something is deeply amiss if we consider patting down a toddler (who was in toddler PJs, which any parent can tell you is fairly impossible to hide something in) a valid, high-efficacy security mechanism. So I did the only thing I could to retain some measure of control, and that was to record the incident.

Why did I do this?

Because sometimes we get so used to something (and as a FF, I’ve been through about 60 screenings a year for the past 5 years) that we just come to accept it as good and proper. We shuffle through a line at 6:30 in the morning, half-awake, and comply with requests that are, in retrospect, totally absurd. That ground starts to get slippery and slope pretty quickly. Humans are designed to obey authority (see Milgram and Stanford experiments/incidents).

Is our current approach to physical safety in commercial airline travel useful? Does it work? Is it consistent with our values? Are there safer, more effective mechanisms that preserve our dignity? Is there an open process of public calibration and testing for these mechanisms?

A lot of people are uncomfortable with the line we seem to be crossing to defend against a constantly moving, amorphous, low-risk threat. But only a few people seem to actually want to say something. As the signs on MTA transit says: “If you see something, say something.”…the phrase could easily be applied to citizen oversight of security measures, not just “weird stuff” that citizens should report to local law enforcement. You have a right to speak up if you don’t like something.

I was asked to delete the video and was told that it was illegal for me to record checkpoints. Here is a list of evidence to the contrary (kudos to P. Mocek for blazing a trail here):

CATSA FAQ: http://www.catsa.gc.ca/Page.aspx?ID=26&pname=TravellerFAQs_FAQVoyageurs&lang=en&sid=7&sname=Pre-Board-Screening-Experience_Processus-de-controle-preembarquement

http://blog.tsa.gov/2009/03/can-i-take-photos-at-checkpoint-and.html

http://www.papersplease.org/wp/mocek/

http://articles.cnn.com/2010-11-25/tech/shooting.video.tsa_1_tsa-s-office-tsa-checkpoints-shooting-video?_s=PM:TECH

http://www.krages.com/phoright.htm

http://www.freerepublic.com/focus/f-bloggers/2632673/posts

http://www.flyertalk.com/forum/travel-safety-security/938543-pv-alert-can-i-take-photos-checkpoint-airport-13.html

http://www.boingboing.net/2011/01/24/flier-beats-tsa-vide.html

The TSA also says: “We recognize that using video and photography equipment is a constitutionally protected activity unless it interferes with the screening process at our checkpoints.” (see here)

Update 5 April 2011: This is a big story. A big IT services FAIL: http://www.cbc.ca/news/business/story/2011/04/05/business-data-breach.html?ref=rss

I’m getting ready for the wave of spam. BestBuy’s direct email service (i.e., legitimate spam) was hacked and the attackers got real, live email addresses:

http://www.thestreet.com/story/11070689/1/retailers-victims-of-e-mail-hackers.html?CM_VEN=AD|TWR|JC

On the one hand, this is no big deal. On the other, it’s kind of annoying — not the incident, but the typical response from a BestBuy corporate VP of — Marketing. Yes, the customer notification doesn’t come from tech staff, but rather from the spin machine. (see email in entirety below)

I love the vagueness in these notification emails assuring users that nothing else of value was disclosed and that the “appropriate authorities” were notified.

Just once, I’ve love an honest message: “We have no idea what they actually got, and neither do the 3rd-party consulting firm or the FBI. I’m sorry we left the default password as ‘password’ on our firewall. We’ll give you $10 of BestBuy credit toward your next purchase.”

What I find lovely is that our (by which I mean the information security community) best advice boils down to six relatively useless recommendations:

http://www.geeksquad.com/do-it-yourself/tech-tip/six-steps-to-keeping-your-data-safe.aspx

—

Dear Valued Best Buy Customer,

On March 31, we were informed by Epsilon, a company we use to send emails to our customers, that files containing the email addresses of some Best Buy customers were accessed without authorization.

We have been assured by Epsilon that the only information that may have been obtained was your email address and that the accessed files did not include any other information. A rigorous assessment by Epsilon determined that no other information is at risk. We are actively investigating to confirm this.

For your security, however, we wanted to call this matter to your attention. We ask that you remain alert to any unusual or suspicious emails. As our experts at Geek Squad would tell you, be very cautious when opening links or attachments from unknown senders.

In keeping with best industry security practices, Best Buy will never ask you to provide or confirm any information, including credit card numbers, unless you are on our secure e-commerce site, www.bestbuy.com. If you receive an email asking for personal information, delete it. It did not come from Best Buy.

Our service provider has reported this incident to the appropriate authorities.

We regret this has taken place and for any inconvenience this may have caused you. We take your privacy very seriously, and we will continue to work diligently to protect your personal information. For more information on keeping your data safe, please visit:

http://www.geeksquad.com/do-it-yourself/tech-tip/six-steps-to-keeping-your-data-safe.aspx.

Sincerely,

Barry Judge
Executive Vice President & Chief Marketing Officer
Best Buy

Update 4 April 2011: It looks like Marriott used the same service. This is a cloud failure mode. Think about it. All I need to do is attack a single vendor, and I get multiple information streams.

April 4, 2011

Dear Marriott Customer,

We were recently notified by Epsilon, a marketing vendor used by Marriott International, Inc. to manage customer emails, that an unauthorized third party gained access to a number of Epsilon’s accounts including Marriott’s email list.

In all likelihood, this will not impact you. However, we recommend that you continue to be on the alert for spam emails requesting personal or sensitive information. Please understand and be assured that Marriott does not send emails requesting customers to verify personal information.

We take your privacy very seriously. Marriott has a long-standing commitment to protecting the privacy of the personal information that our guests entrust to us. We regret this has taken place and apologize for any inconvenience.

Please visit our FAQ to learn more.

Sincerely,

Marriott International, Inc.

and the FAQ has this content:

April 4, 2011

What happened?
Marriott International Inc. was recently notified by Epsilon, a marketing vendor used by Marriott to manage customer emails, that an unauthorized third party gained access to a number of Epsilon’s accounts including Marriott’s email list.

What Marriott information was accessed?
The unauthorized person(s) had access to names and email addresses only. They did not have access to sensitive customer information, such as physical addresses, loyalty program point balances, account logins and passwords, credit card information or other personal data.

How does this affect Marriott customers?
There is a possibility that customers whose email addresses were obtained may receive unsolicited emails (i.e., spam or phishing). Marriott has a long-standing commitment to protecting the privacy of the personal information that our guests entrust to us. As a result, Marriott never sells or rents member lists.

What is Epsilon doing about this?
Epsilon has notified Marriott that they have identified where the breach took place and has taken the necessary steps to prevent additional data loss. Epsilon has further informed us that they have notified law enforcement and that additional security precautions have been put in place to help prevent future incidents.

I was recently posed a question by a psychologist friend of mine. She needed to enumerate all possible answers to a set of 20 questions. Each individual question could have three answers. She then planned to count the actual frequency of answers from the test sample (say, N~=100) and run this result through some other statistical software. The problem was that the statistical software’s input format required the permutations of answers, even if they had a frequency of zero.

I could easily write a procedural iterative program to enumerate all possible binary choices for 20 questions. (Wait, can I?) Wait – what about three possible choices?

The difficulty is deciding just what type of problem this actually is. Is it a permutation thing (i.e., order matters)? Combination (i.e., order doesn’t matter)? Enumeration (iterate each possible permutation)? Using what number base (3)? Should I use an iterative or recursive solution, or perhaps a clever workaround? How long would a reasonably efficient program take?

In any event, I knew that the output set was potentially large, and it wasn’t immediately apparent how to write a piece of C code to enumerate all possible permutations of answers for 20 questions. Let’s define what I’m looking at: I want all possible combinations (i.e., I don’t care about order specifically) of answer strings to a list of 20 questions where each question could be answered in 1 of 3 ways. If I were just interested in permuting the test questions themselves, I would have a set of 20! or 2,432,902,008,176,640,000 or 2.4×10¹⁸ (as Unix `bc’ tells me). But I’m not; I don’t care what order the questions are in, but rather producing all possible ways to answer the test itself for one arbitrary ordering.

For 1 question, there are three possible answers: 0, 1, 2 (let’s say they correspond to “yes”, “maybe”, “no”). You might also consider a null answer in addition to this set, but for now, let’s ignore that possibility. So, there are three possible ways that this one question test could be answered. What if we added more questions?

Q1: ?
0, 1, 2
Q2: ?
0, 1, 2

So we have all possible answer strings (of length 2) as:

00
01
02
10
11
12
20
21
22

With just 2 questions, there are 9 possible answer permutations.

Thus, the number of questions becomes the length of our string. The number of response options become the possible values of each location in the bitstring. In binary (base 2) number systems, the possibilities proceed in powers of 2 (e.g., for a 100 question test with only two possible response options, the total number of answer combinations is 2^100). With base 3 and 20 questions, the total number of unique answer combinations is 3^20; that is, someone could answer the test itself in any one of 3^20 ways. For those of us who are computationally challenged, this number has a value of 3,486,784,401 or about 3.4 billion (as an aside, even if 10,000 people responded to this test, the sets of answers they produce would very sparsely populate this space of 3.4 billion strings).

So how do you generate 3.4 billion unique strings of length 20 where each position could be one of three values? Actually, how do you write a program to do that for you?

It turns out that we can take a couple of interesting detours into mathematics [1,2], but I’d prefer to stay on the CS side of the house for now. Let’s think about symbol representation. Above, I abandoned ‘y’, ‘n’, and ‘m’ for representing the choices because those three characters have no directly mathematical relationship. Representing the answers as 0, 1, and 2 (on a binary machine) really doesn’t seem to have much advantage either.

There is in fact a ternary number system with some interesting properties. According to the American Scientist article “Third Base” (Nov/Dec 2001), the ternary number system “is the most efficient of all integer bases; it offers the most economical way of representing numbers.”

Perhaps I could simply write a program to count to 3,486,784,401, converting each integer value to a bitstring and then write a routine for converting that bitstring to a tritstring (ternary string) and printing it to stdout. I’ll try that, and see what happens. But let’s consider the heart of that approach: the translation of binary bitstrings to tritstrings. Let’s translate the binary representation of decimal 12 to ternary.

12 contains two powers of 2: 8 + 4. So in a 4-bit string, let’s turn on the leftmost two bits to get: 1100 (i.e., 1*2³ + 1*2² + 0*2¹ + 0*2⁰, or 1*8 + 1*4 + 0 + 0).

Now how do we convert from base 2 to base 3? How do you convert from one base to another? Well, I just did it for base 10 to base 2, albiet informally. You keep dividing the number to be translated by the target base and keep track of the remainder. The sequence of remainders becomes the digits of your translated number. To wit:

12 / 2 = 6 remainder 0
6 / 2 = 3 remainder 0
3 /2 = 1 remainder 1
1 / 2 = 0 remainder 1

Once the division results in a zero value, we have all the remainders we need (read that right most column of remainders upwards…see how it equals 1100?).

Armed with this knowledge, need we convert from decimal to binary to ternary at all, or can we just directly convert?

12 / 3 = 4 remainder 0
4 / 3 = 1 remainder 1
1 / 3 = 0 remainder 1

Is 110 (ternary) really equal to 12? Yes: 1*3^2 + 1*3^1 + 0 or 1*9 + 1*3 or 12.

Code

Ternary Counting

Postscript

Along the way to investigating this, I stumbled across two other interesting American Scientist articles: “E Pluribus Confusion“, about “fair” seat allocation in the US House of Representatives (a process with a series of varied hacky tie-breaking rules over the years) and “The Bootstrap” about the limits of our common statistical assumptions.

References

[1] http://lin-ear-th-inking.blogspot.com/2012/11/enumerating-permutations-using.html
[2] http://www.roard.com/docs/cookbook/cbsu2.html

I just received some spam from the ACM touting some material about Cloud Computing. The email says, in part:

“Cloud computing promises to radically change the way computer
applications and services are constructed, delivered, and managed.
It is a fundamental new paradigm in which computing is migrating from
personal computers sitting on a desk to large, centrally managed
datacenters.”

This got me to thinking: there are actually two paradigm shifts happening in computing, and the other is the move to smaller, powerful, mobile personal devices. Each seems to be heading in the opposite direction, and the desk looks like the loser.

But of course these paradigm shifts are complementary; there is no reason why a smaller, more natural human interface in the form of a mobile device cannot take advantage of the compute and storage power of the cloud while retaining the ability to perform some computationally heavy tasks if called upon (e.g., if the cloud is unavailable, or if you don’t trust the cloud with the data/inputs to perform the calculation).

In fact, we can think of these two shifts as a factoring of the traditional desktop role; you still have a personal computing device, but it is now mobile, and its power is greatly enhanced by remote storage and compute capabilities. Of course, shortcomings exist: in interfaces (thumb typing? come on), in power (battery life, anyone?), in security (store my medical info in the cloud???), in display size (watching movies on 4 inches…eye strain), etc.

Debugging the OS, even with full source, full debugging symbols, and a framework like kprobes or DTrace is a very hard exercise.

Debugging OS X from a user perspective (without access to source, and with skimpy information in a crash report) is almost impossible.

I’m willing to speculate a bit, though.

My recent kernel panics are likely the result of two possible conditions: (1) an actual race condition bug in the kernel; or (2) interference with the kernel data structures that manage locking by some 3rd-party software loaded into the OS as a module. On my system, the only “foreign” code loaded into the OS are drivers for my Microsoft Mouse and Keyboard (let’s face it, Apple keyboard suck, and the one-button-trackpad-click-multi-touch interface is just a nightmare as a pointing device) and VMware:

loaded kexts:
com.microsoft.driver.MicrosoftMouseUSB 6.2.2
com.microsoft.driver.MicrosoftMouse 6.2.2
com.microsoft.driver.MicrosoftKeyboardUSB 6.2.2
com.microsoft.driver.MicrosoftKeyboard 6.2.2
com.vmware.kext.vmnet 2.0.4
com.vmware.kext.vmioplug 2.0.4
com.vmware.kext.vmci 2.0.4
com.vmware.kext.vmx86 2.0.4

I’m willing to believe VMware may be the culprit here, particularly since I’ve started using it heavily since Migrating to the new notebook. My copy of VMware Fusion is a few revisions behind (I think), and OS X is up to date. I didn’t have problems with it on my previous Macbook Pro (although I did have other problems with that notebook). I’ll try updating VMware Fusion and see if that resolves the problem (although it is naturally tough to test the absence of something).

I should rename this blog “Complaints about Apple Mac OS X”.

Apple has their own alternative to Microsoft’s “Blue Screen of Death.” I call it the “Slow Curtain of Death” — the machine suddenly stops responding, and a line proceeds from the top of the screen to the bottom, dimming the display and displaying a typically sexy-looking Apple message box saying “You must restart your computer. Press…”

My new Macbook Pro crashed (i.e., the kernel had a “panic” — yes, that’s a technical term for when the underlying operating software encounters an unrecoverable error) for the fourth time in a week from the exact same line of code (it looks like an error in handling thread locking conditions in the kernel scheduler). I have sent all four of these reports to Apple via their automated reporting mechanism.

I suspect the culprit is some incompatibility between my version of VMWare Fusion and Snow Leopard (but this is complete speculation). I append the start of the most recent crash report below. Note a few things. First, the error output line itself is quite detailed, even including a possible cause (unlocking an already unlocked mutex or spinlock? why would that cause a problem?) It even includes the file sched_prim.c and line number of the error. Too bad I don’t have Mac OS X source code. The report also includes a dump of the stack, but it lacks reporting the standard dump of CPU registers.

Interval Since Last Panic Report: 27770 sec
Panics Since Last Report: 1
Anonymous UUID: B22098C3-CA08-4230-A115-AFDF431CCB8B

Tue Jan 18 14:50:12 2011
panic(cpu 2 caller 0x226b53): “thread_invoke: preemption_level -1, possible cause: unlocking an unlocked mutex or spinlock”@/SourceCache/xnu/xnu-1504.9.26/osfmk/kern/sched_prim.c:1471
Backtrace (CPU 2), Frame : Return Address (4 potential args on stack)
0xbe56be18 : 0x21b50c (0x5d4438 0xbe56be4c 0×223974 0×0)
0xbe56be68 : 0x226b53 (0x58babc 0xffffffff 0x58ba54 0×226714)
0xbe56bee8 : 0×227259 (0xde9db98 0×0 0x6bc9f000 0×1)
0xbe56bf58 : 0x2272c4 (0x22fc20 0x863ea0 0×0 0x2a358d)
0xbe56bf78 : 0x22fdba (0x22fc20 0x863ea0 0×0 0×0)
0xbe56bfc8 : 0x2a06cc (0x863ea0 0×0 0×10 0xe1933c0)

BSD process name corresponding to current thread: kernel_task

Mac OS version:
10J567

Kernel version:
Darwin Kernel Version 10.6.0: Wed Nov 10 18:13:17 PST 2010; root:xnu-1504.9.26~3/RELEASE_I386
System model name: MacBookPro6,2 (Mac-F22586C8)

System uptime in nanoseconds: 38460313523585

Sad story about influence on wikileaks’s ability to distribute content (see link below).

Wikileaks runs into two problems:

1. government pressure against companies providing hosting or directory
   services to wikileaks material
2. “market” pressure arising from DDoS attack against shared
   infrastructure that has enough collateral damage to force the
   infrastructure owner to stop providing hosting or directory services

What are the design criteria for a world-wide information network that
can resist these forms of pressure? In other words, given their own
laptop and a reliable source of energy, how can individuals create a
truly global P2P information infrastructure independent of government or
business influence? (And things like Tor are a red herring here…the
point is to be totally divorced from running on top of existing Internet
infrastructure). Yes, this raises all sorts of security issues dealing
with namespace control and issues about capacity and DoS…but
it seems like an interesting thought experiment.

http://arstechnica.com/tech-policy/news/2010/12/wheres-wikileaks-the-infowar-is-on-as-site-hops-servers.ars