Slashdot Mirror
Ask Slashdot: Linux Security, In Light of NSA Crypto-Subverting Attacks?
New submitter deepdive writes "I have a basic question: What is the privacy/security health of the Linux kernel (and indeed other FOSS OSes) given all the recent stories about the NSA going in and deliberately subverting various parts of the privacy/security sub-systems? Basically, can one still sleep soundly thinking that the most recent latest/greatest Ubuntu/OpenSUSE/what-have-you distro she/he downloaded is still pretty safe?"
The big worry is not building from source, but builds delivered by companies like Ubuntu, which you have absolutely no guarantee are actually built from the same source that they publish. Ditto Microsquishy, iOS, Android, et. al.
The big concern is back doors built into distributed binaries.
I do not fail; I succeed at finding out what does not work.
You can not add security, later.
In Unix systems, there’s a program named “login“. login is the code that takes your username and password, verifies that the password you gave is the correct one for the username you gave, and if so, logs you in to the system.
For debugging purposes, Thompson put a back-door into “login”. The way he did it was by modifying the C compiler. He took the code pattern for password verification, and embedded it into the C compiler, so that when it saw that pattern, it would actually generate code
that accepted either the correct password for the username, or Thompson’s special debugging password. In pseudo-Python:
def compile(code):
if (looksLikeLoginCode(code)):
generateLoginWithBackDoor()
else:
compileNormally(code)
With that in the C compiler, any time that anyone compiles login,
the code generated by the compiler will include Ritchie’s back door.
Now comes the really clever part. Obviously, if anyone saw code like what’s in that
example, they’d throw a fit. That’s insanely insecure, and any manager who saw that would immediately demand that it be removed. So, how can you keep the back door, but get rid of the danger of someone noticing it in the source code for the C compiler? You hack the C compiler itself:
def compile(code):
if (looksLikeLoginCode(code)):
generateLoginWithBackDoor(code)
elif (looksLikeCompilerCode(code)):
generateCompilerWithBackDoorDetection(code)
else:
compileNormally(code)
What happens here is that you modify the C compiler code so that when it compiles itelf, it inserts the back-door code. So now when the C compiler compiles login, it will insert the back door code; and when it compiles
the C compiler, it will insert the code that inserts the code into both login and the C compiler.
Now, you compile the C compiler with itself – getting a C compiler that includes the back-door generation code explicitly. Then you delete the back-door code from the C compiler source. But it’s in the binary. So when you use that binary to produce a new version of the compiler from the source, it will insert the back-door code into
the new version.
So you’ve now got a C compiler that inserts back-door code when it compiles itself – and that code appears nowhere in the source code of the compiler. It did exist in the code at one point – but then it got deleted. But because the C compiler is written in C, and always compiled with itself, that means thats each successive new version of the C compiler will pass along the back-door – and it will continue to appear in both login and in the C compiler, without any trace in the source code of either.
http://scienceblogs.com/goodmath/2007/04/15/strange-loops-dennis-ritchie-a/
"Flyin' in just a sweet place,
Never been known to fail..."
No, but there's no reason to think that Linux is worse than anything else, and it's probably easier to fix.
If I were Linus I'd be putting together a small team of people who have been with Linux for years to begin assessing things. From Gilmour's posting it seems clear that IPsec and VPN functionality will need major change. Other things to audit include crypto libraries, both in Linux and the browsers, and the random number generators.
But certainly some examination of SELinux and other portions are also needed.
I don't see how anyone can answer the original question without doing some serious assessment. However I'm a bit skpetical whether this problem can actually be fixed at all. We don't know what things have been subverted, and what level of access the NSA and their equivalents in other countries have had to be code and algorithm design. They probably have access to more resources than the Linux community does.
I believe you can trust OpenBSD totally but it lacks many of the features and much of the convenience of the main Linux distros. It is rock solid and utterly secure though, and the man pages are actually better than any Linux distro I've ever seen.
Three points:
1) See the above discussion: you cannot trust anything that you did not create and compile yourself. With a compiler you wrote yourself. On a machine you created yourself from the ground up, that is not connected to any network in any way. OpenBSD does not make any difference if your compiler or toolchain is compromised.
2) Speaking of which, I cannot but note that OpenBSD had a little kerfuffle a while back, about a backdoot planted by the FBI in the OS? (Source 1) (Source 2). I am willing to bet that (a) it's perfectly possible (though not likely), (b) if it was done, it was not by the FBI and (c) that the dev @openbsd.org are, right now, taking another long and hard look at the incriminated code.
3) Finally OpenBSD lacking features and convenience? Care to support that statement? I have a couple of computers running OpenBSD here, and they are just as nice - or even nicer - to use than any Linux. Besides, you don't choose OpenBSD for convenience - you use it for its security. Period.
The possibly bigger problem is that no matter what OS you use you can't trust SSL's broken certificate system either because the public certificate authorities are corruptible. And before someone says create your own CA, sure, for internal sites, but you can't do that for someone else's website.
This goes way beyond a simple question of OpenSSL certificates - think OpenSSH and VPN security being compromised, and you will have a small idea of the sh*tstorm brewing right now.
The right to offend is far more important than the right not to be offended. (Rowan Atkinson)
Even that's no good if the problem is flaws in the spec rather than how it's implemented by OSs. If the NSA did things correctly they didn't have to muddle with actual Linux/BSD/etc src, they got flaws into the crypto definition itself that reduces the work needed to crack it. The better an OS follows the spec... the easier for the NSA to punch through.
Hmmm - all of a sudden this looks interesting again:
http://news.cnet.com/8301-31921_3-20025767-281.html
Is there any particular reason why people don't strengthen AES (or any other symmetric encryption) by just reencrypting 1000 times? Perhaps interleaving each encryption with encrypting with the first 1, then 2 etc. It would make next to no difference for the end user, who's going to decrypt just once, but I imagine it would add a lot more time to the cracking of the encrypted data than increasing the size of the key.
It seems that encrypting the file multiple times with the same key is not safe, and tends to expose the flaws in the encryption method. It will be less secure. Hashing the password with a random key multiple times (like keepass uses 5000 rounds), and then using that string to encrypt the file, however should work. I'm not an expert on this matter, just repeat what someone else replied me when asking the same question.
Digitial Forensics for Prosecutors presentation suggests Truecrypt has a backdoor.
http://www.techarp.com/showarticle.aspx?artno=770&pgno=0
That won't even make it through the casual review. Most project maintainers don't like code that's impenetrable. Unless it's a fix for a critical bug that nobody else even has a proposal for a fix for, they're going to take one look at obfuscated code and toss it back with a "No thanks.". Especially if it's coming from a source they don't recognize, because messy complex obfuscated code also tends to be buggy unreliable unmaintainable code and they don't want the headache.
I'd like to find a more recent report of what they found.
The reason you can boot from a raid card or network is because the BIOS loads and runs BIOS modules from those cards. You may be familiar with the Linux kernel, where most of the functionallity is in modules that become part of the kernel. BIOS is the same. One differentiator between a server motherboard and a consumer one is how much BIOS memory it has, to load modules from many different pieces of hardware. I have one machine with at least four different pieces of hardware that include BIOS. MOST of the BIOS on that machine didn't come with the motherboard.
For the Linux kernel, that's how development is done already, for quality control and bloat reduction. Nobody can commit by themselves, it takes at least three people to get a change into mainline. Each developer has their own copy of the tree into which changes are pulled, so they can see all changes that are made, and who made them.
For each part of the kernel, there are a number of people particularly interested in that bit who watch it and work on it. For example, the people making NAS and SAN devices and services keep a close eye on the storage subsystems. Myself, I watch the cm storage stack generally, more specifically LVM, and even more specifically snapshots. There are a few dozen people around the world with special interest in that particular part of the code. No backdoors will come in without some of us spotting it. What COULD happen is that some code could come in that isn't quite as secure as it could be.
It just so happens that I'm a security professional who uses advanced Linux storage systems for a security product called Clonebox, so that's at least one security professional closely watching that part of the code. Thousands of others watch the other parts.
It's convenient that a lot of the development is done by companies like Netapp, Amazon (S3) and Google. You can bet that when Amazon submits code, Netapp and Google are looking closely at it. When RedHat submits something, Canonical will point out any reasons it shouldn't be accepted.