Loophole in Windows Random Number Generator

Invisible Pink Unicorn writes "A security loophole in the pseudo-random number generator used by Windows was recently detailed in a paper presented by researchers at the University of Haifa. The team found a way to decipher how the number generator works, and thus compute previous and future encryption keys used by the computer, and eavesdrop on private communication. Their conclusion is that Microsoft needs to improve the way it encodes information. They recommend that Microsoft publish the code of their random number generators as well as of other elements of the Windows security system to enable computer security experts outside Microsoft to evaluate their effectiveness. Although they only checked Windows 2000, they assume that XP and Vista use similar random number generators and may also be vulnerable. The full text of the paper is available in PDF format."

31784

[FooAtWFU (699187)]

129775, 80123133, 5580012. 6740091, 6558, 42!

Re:

[Spokehedz (599285)]

All I get on mine is '8675309'

Damn you Johnny Two-Tone!

Re:

[jejones (115979)]

RANDU! Save us, RANDU!

Oh, wait, that was Landru.

Hardware RNG

[CRCulver (715279)]

I assume this is only a problem for those whose motherboard doesn't have a hardware random-number generator?

Re:Hardware RNG

[$RANDOMLUSER (804576)]

Now why would you assume Microsoft would use the hardware RNG when they have thier own, much better, proprietary RNG available?

Re:Hardware RNG

[defnoz (1128875)]

Now why would you assume Microsoft would use the hardware RNG when they have thier own, much better, proprietary RNG available?

After all, they spent so much time perfecting it in Excel 2007!

Re:Hardware RNG

[thePsychologist (1062886)]

It might only be a problem for 2000 users:

According to the researchers, who have already notified the Microsoft security response team about their discovery, although they only checked "Windows 2000" (which is currently the third most popular operating system in use) they assume that newer versions of "Windows", XP and Vista, use similar random number generators and may also be vulnerable.

Re:Hardware RNG

[thePsychologist (1062886)]

This is classic behaviour on Slashdot. I point out this might not be a big of a problem as it seems (as they only tested Windows 2000, and not XP or Vista, both combined are far more used than 2000), and I'm modded as troll, only because (I presume) that I'm providing evidence that a problem with Microsoft isn't as serious as it seems (i.e. I'm getting in the way of MS bashing).

Re:Hardware RNG

[somersault (912633)]

Yeah because every time Windows is updated, it's a really high priority to write a new random number generator? XP is based off of 2000 even if Vista was meant to be a rewrite.

"Hey guys, I dont think the random number generator is random enough today - it came up with 2 prime numbers in a row! Anyone feel like taking a few days to rewrite it, test it, introduce a few bugs, document it, seal off the documentation to make sure nobody finds it, and go take it up to Steve? I hear he's out of chairs right now so it should be okay".

Re:Hardware RNG

[Tim Browse (9263)]

Unfortunately, some people might believe that's really how it happens. Cryptographically secure RNGs are a widely known issue in the field (hell, even I know about it, and I'm not in the field), and you can be sure that the Crypto programmers at MS are at least aware of the issue. It wouldn't surprise me, at any rate, if implementing a new RNG had been considered a priority for XP or Vista if they had discovered the existing one to be vulnerable.

If they had time in between cocking up all the WGA stuff, that is.

Re:Hardware RNG

[yukk (638002)]

What makes you think that MS has "Crypto programmers" ? I'm sure that part of development went something like this.
Okay, module 14537r Random Number Generator. Teams, who wants do do this ? No, it's not boring. Come on. Okay, draw straws. Jones, you win. Yes, sure you can get the intern to write it. You carry on with the Clippy enhancements.

Re:Hardware RNG

[thebdj (768618)]

A new RNG is not really a selling point, the only way it will help their bottom line is if enough people know about flaws in the old one that it's profitable to replace it.

Actually it can be, since it would be necessary to use a FIPS compliant PRNG to perform certain operations, they would need to have one. I suspect (see my other posts) that this is from a deprecated cryptographic service provider that MS no longer providers (DSS_BASE). If you check out the information on the CMVP website for the RNG Validation Lists [nist.gov], you will see they implement FIPS 186-2 PRNGs, which the paper itself admits (Appendix B) has some forward security and is not the PRNG they are attacking here.

Re:Hardware RNG

[Goaway (82658)]

What is this, "proof by sarcasm"?

Re:Hardware RNG

[somersault (912633)]

I prefer to call it Sarca'm's Razor

the number of affected users enbiggens the problem

[doti (966971)]

only tested Windows 2000, and not XP or Vista, both combined are far more used than 2000

Still, 2000 has more (desktop) users than Linux. By your logic, if there were a similar problem in Linux, it would be less of a problem?

Re:Hardware RNG

[Belial6 (794905)]

You actually didn't provide any evidence that the problem doesn't affect XP or Vista, you just suggested that the two newer version should be trusted immediately after finding out that 2000 has a bug in an unlikely to be updated part of the system. The non-troll way of highlighting this information would be:

That is a problem. I am eagerly awaiting the tests of XP and Vista to see if this was fixed for them.

You could probably even slip a little bias in there without being called a troll with:

They are going to test with XP and Vista aren't they? After all, it should be trivial to test this on the newer systems if the cryptography hasn't been changed. I mean what kind of security researcher just assumes the functionality of a security system?

Of course, it would be a little silly to assume that this does not affect at least XP, as 2000 was still under maintenance when XP was released, so if the bug was found during the development of XP, it should have been fixed in 2000. It would look far worse for Microsoft if they KNEW about a security hole in 2000 while it was still under maintanace, and did not bother to back port the fix from XP.

Re:

[operagost (62405)]

I recently discovered that Windows is not Y2K compliant! Although I only checked Windows 3.1, I assume that newer versions of Windows, 2000, XP, and Vista, use similar 2-digit dates and may also be vulnerable.

Re:Hardware RNG

[lgw (121541)]

Windows RNG collects "entropy" (that is, non-pseudo-randomness) from many sources, including drive timing, network timing, keyboard and mouse timing, temperature information, etc. However, there are only so many "really random" bits per second available.

Any good RNG combines sources of entropy with a cryptographically secure PRNG. The researchers are attacking the PRNG portion of the Windows RNG. If you only generate keys (or other random numbers) infrequently, this is a non-issue, as the hardware sources of entropy provide enough "really random" bits to generate a "really random" number.

However, if you generate a fast series of keys (or other random numbers), you quickly use up all of the "really random" bits that the RNG has cached, and you only have the PRNG on your side, and therefor the key is merely "pseudo random". TFA is an attack on the "psuedo random" portion of the Windows RNG.

Interestingly, the much-reviewed TrueCrypt engine seems to slow to a crawl if you create a bunch of files (and therefore keys) in a hurry - presumably it has an RNG that actually blocks waiting until it has enough new "really random" bits for each new key. This is a cool idea for a crypto library, but not usable for a general-purpose RNG, which suggests that the system libraries should probably provide *two* RNGs.

Re:Hardware RNG

[EsbenMoseHansen (731150)]

Interestingly, the much-reviewed TrueCrypt engine seems to slow to a crawl if you create a bunch of files (and therefore keys) in a hurry - presumably it has an RNG that actually blocks waiting until it has enough new "really random" bits for each new key. This is a cool idea for a crypto library, but not usable for a general-purpose RNG, which suggests that the system libraries should probably provide *two* RNGs.

Brilliant idea! Let's call one of them /dev/urandom and the other one /dev/random. ;)

Re:

[larry bagina (561269)]

intel's fpu is a random number generator (unintentionally).

Seed time

[EaglemanBSA (950534)]

How accurate would they have to be with predicting the generator seed times for the keys to work? Would that be a hitch? I'm not an expert in the field, so I honestly don't know.

Re:Seed time

[EaglemanBSA (950534)]

Looks like if you can use their method to find the current state fast enough, windows doesn't do a great job of reseeding very quickly: I read through the PDF and found this comparison of the LRNG to WRNG (p. 18) - "Reseeding timeout. The LRNG is feeding the state with system based entropy in every iteration and whenever system events happen, while the WRNG is reseeding its state only after generating 128 KBytes of output. Synchronization. The collection of entropy in the LRNG is asynchronous: whenever there is an entropy event the data is accumulated in the state of the generator. In the WRNG the entropy is collected only for a short period of time before the state is reseeded. In the long period between reseedings there is no entropy collection. Security implication: The impact of the previous four properties is that forward and backward security attacks are more severe when applied to the WRNG. The attacks are more e±cient by twelve orders of magnitude. They reveal the outputs of the generator between consecutive reseedings, and these reseedings are much more rare in the case of the WRNG. In some cases, reseeding the LRNG happens every few seconds, while the WRNG is reseeded every few days, if it is reseeded at all."

Huh?

[mrseigen (518390)]

Maybe it's just me, but I didn't think anyone would be stupid enough to use rand for SSL like the article is implying.

From what I can see, this is an old article anyway.

Re:Huh?

[Hatta (162192)]

No, but they might use it for encrypting windows passwords.

Re:Huh?

[mrseigen (518390)]

Whoops -- it's not rand, it's CryptGenRandom.

Re:Huh?

[Mantaar (1139339)]

From TFA:

Date: received 4 Nov 2007

Old indeed. 8 days. That's a lot, Microsoft might have already fixed it, you see, they fix things fast!

Where's the white noise generator?

[tjstork (137384)]

I am still at a loss to wonder why a PC does not have a white noise generator built into it yet. Even the best random number algorithms are pseudo random, so blasting Microsoft for their algorithm is a little like blasting the kid for not carrying enough of a bucket when the dam is the thing that broke.

Put white noise hardware and real random number hardware on PCs, and this whole problem goes away.

Re:Where's the white noise generator?

[OrangeCowHide (810076)]

A white noise generator? Bah... What systems need are pop-o-matic bubbles with m * 2^n sided dice to generate m * n bits. It could even put a window up saying, "The entropy pool is depleted. Please press the pop-o-matic bubble to generate more."

That would be awesome

Re:Where's the white noise generator?

[ConceptJunkie (24823)]

Perhaps you could pour hot tea into it instead.

Re:

[pesc (147035)]

Like the VIA C3 [via.com.tw] processor?

Re:Where's the white noise generator?

[palladiate (1018086)]

The Commodore had one too, on the sound chip. The old P3 i810 and VIA C3 chipsets had RNGs built in. They relied on thermal noise. Some AMD chipsets still have it. But for the most part, no modern motherboard comes integrated with a hardware RNG.

USB Hardware RND

[CustomDesigned (250089)]

Buy one of those $25 toy digital cameras. Keep the lens cap on, or put black tape over the lens. Connect to USB port. Add script to snap a "picture" every few minutes to prng. (Is there a way for userland to feed entropy to kernel based /dev/random?) With no light, digital cameras return thermal noise - which looks like "snow" on an analog TV. I've done this with a toy camera I bought for my daughter. The camera feeds raw pixels to the linux driver, and the post processing done by the Windows software was never implemented in Linux, making it useless as a camera (plus it has 256M ram, but no flash memory). But it works great for this application. I haven't done a mathematical analysis of exactly how much entropy is in the signal. I'll leave that for the stat geeks.

I got the idea from a project that used a webcam snapping pictures of a Lava Lamp® as a hardware RNG.

Re:Where's the white noise generator?

[palladiate (1018086)]

No, Intel no longer provides a hardware RNG on most chipsets. The last is the i810.

Some AMD64 chipsets still do though. You generally don't find hardware RNG on any chipset below the "Major Enterprise Purchase" mark.

Which could be bettered, easily.

The Vista RNG

[by Anonymous Coward]

Although they only checked Windows 2000, they assume that XP and Vista use similar random number generators and may also be vulnerable.

Your system must meet the requirements to be able to run the Windows Random Number Generator on Vista. Otherwise, you will need to use Windows Number Generator Basic. The only number WNGB can generate is 4.

Re:The Vista RNG

[eln (21727)]

Yes, but that 4 was generated via a fair dice roll, and is guaranteed to be random. You can't say that about the numbers the Vista RNG spits out. So you see, what the WNGB lacks in quantity it makes up for in quality.

ob XKCD reference

[wren337 (182018)]

http://xkcd.com/221/ [xkcd.com] // chosen by fair dice roll // guaranteed to be random

Re:The Vista RNG

[secPM_MS (1081961)]

The random number generator for XP and 2K3 server was substantially improved over that of Win 2000. Additional work was done for Vista. These systems are used in highly secure military deployments and due to its importance to system security, the random number generator was subjected to extensive analysis and was updated to deal with issues uncovered. When evaluating "random number generators" you need to consider not only the "random number" generator, but entropy harvesting from the system and other issues relating to usage. I assume the bulk of the readers are not MS developers, but if you need a good random number on a Windows platform, call CryptGenRandom. Equivalent functionality is provided for managed code as well.

Win 2K is a very legacy product and its crypto functionality is very limited compared to 2K3 and Vista.

Re:The Vista RNG

[secPM_MS (1081961)]

I didn't, but I know the people who did the enhancements, and they are very competent and well known cryptographers.

Novell

[by Anonymous Coward]

In other news, Miguel de Icaza said that he believes that the random number generator is a good idea. Linux should have one because Microsoft is going to win anyway, so linux would better be prepared if it doesn't want to be locked out of the future markets, and presented a beta version of the algorithm. Members of the GNOME foundation are participating in the standarization: ''it's better to provide our own insecure random number generator'' said ownen taylor.

Fixed in Vista?

[adonoman (624929)]

http://msdn.microsoft.com/msdnmag/issues/07/07/Security/default.aspx [microsoft.com] has the new API, including a RNG

that meets Federal Information Processing Standards (FIPS) for use with the Digital Signature Algorithm (DSA).

There's a lot I don't like about Vista, but for security researchers to "assume that XP and Vista use similar random number generators and may also be vulnerable" without a basic google search is a bit much!

Re:Fixed in Vista?

[by Anonymous Coward]

Actually go further back;

http://www.microsoft.com/technet/archive/security/topics/issues/fipseval.mspx?mfr=true [microsoft.com]

You'll note that Windows 2000 passed FIPS-140-1.

Re:Fixed in Vista?

[CastrTroy (595695)]

Just because they have a new API for getting the random numbers, it doesn't mean that they are using different algorithms for generating those random numbers. Also, they much still have the old APIs in there, otherwise, a lot of programs would fail to work. Since most of the software out there was written pre-Vista, and written to run on Vista, XP, and 2000, it's conceivable that applications on these operating systems are using the vulnerable code.

Publication iffy

[cdrguru (88047)]

The only benefit that could possibly be derived by publishing algorithms and/or code for Windows security would be if (a) changes proposed would be implemented quickly and (b) everyone planet-wide upgraded.

If both of these did not happen, especially if (b) didn't happen, what you would be doing is exposing all non-upgrading users to the full brunt of whatever flaws their might be. Would this really be productive? Does this remind you of various failures in Linux code that led to rootkits being developed for it. Did the victims of such attacks think it was all for the best because they didn't upgrade in a timely manner?

Yes, relying on people not reverse-engineering code to protect users isn't a great plan. But the current situation - as regrettable as it is - is this is the only plan. There are no fallbacks, there are no alternatives. Most of the running copies of Windows aren't going to be "fixed" in any way whatsoever.

Re:

[IkeTo (27776)]

This sounds *really* wrong. You can say white-hats should have waited for a few days or even a few weeks after notifying the vendors before disclosing problems, but they should be disclosed eventually, and should be disclosed after giving vendors a reasonable amount of time. There bound to be people not upgrading their Windows, and there bound to be people not upgrading their Redhat or Fedora or Ubuntu or SuSE or FreeBSD or whatever operating system you name (not to mention whatever Firewalls, protocols,

Does this mean...

[physicsphairy (720718)]

That it will be possible to predict what values Excel will give us in our spreadsheets?

Similar but different?

[QuietLagoon (813062)]

I wonder if this [coredump.cx] is a similar problem?

Is there a list of slots machines that run windows

[Joe The Dragon (967727)]

Is there a list of slots machines that run windows?

Hardware RNG

[SamMichaels (213605)]

You'd think that computers would have built-in hardware based RNGs by now. On-board sound, video, network, etc.......where is the radioactive decay RNG [fourmilab.ch]? After all, in 1985 plutonium should be available in every corner drugstore [wikipedia.org].

Not so severe

[SiliconEntity (448450)]

IMO the attack is not so severe as they make it sound. While this is a nice piece of reverse engineering and cryptanalysis, in practice the security implications are small.

The bottom line is that every process has its own copy of the RNG state. That means that breaking into one process will not help you deduce the random numbers being used by another. (The authors comment that there may be similarities between the two states, but they don't have any way to turn that into a practical attack.) So the only thing this does is it lets an attacker who compromises a certain process or program, such as IE, be able to learn the random number state. From that he can deduce old random numbers that were used, as well as deduce new random numbers that will be created in the future.

That second part is hard to avoid, but the first part, running the state backward (confusingly called forward security by cryptographers), is a sign of bad design of the RNG. Okay, Microsoft messed that up. But what are the security implications?

The implication is that if someone breaks into your computer, here is something more he can do. Not only can he take over going forward, he can learn a certain amount of data about the past. If you had an SSL protected session in the past, then he could go back and figure out what they keys were back then and decrypt the data.

But how bad is this, really? Compared to the harm he can already do by breaking into your computer? Given that he's there, he can learn all of your future SSL keys anyway. Anywhere you go in the future, your bank, paypal, ebay, any site he can learn all of your passwords and account numbers. He doesn't need to compromise the RNG for this, he can just watch your keystrokes. Basically, you are totally screwed if this happens.

Given the enormous magnitude of the security lost, the additional harm from being able to decrypt a few old requests is quite small. You are basically owned from then on. If you have insecure software that is vulnerable to such attacks, you're screwed anyway. A weakness in the RNG state means you are slightly more screwed, that's all. It's not a major change in the security equation.

The bottom line is that most of the damage comes from the break-in. Again, not to take anything away from these guys' work, but the attack they describe is at worst just the icing on a very nasty cake. Microsoft should fix it, and it sounds like they probably have in Vista, but nobody needs to change their security practices because of this flaw.

Re:

[Cheesey (70139)]

That sort of attack could probably be used against online Nethack servers such as nethack.alt.org. You could predict what set of items you'd get if you generated a character at a specific value of time(NULL). You'd also be able to predict the future for that character. You'd try out sequences of moves on your PC, and then send the sequence that got you the best results.

Unfortunately extra non-determinism would be introduced by bones files, and you'd get a new random sequence if you logged out. The server ad