Hacker News Reader

> (it's the TLS library for Firefox and Chrome)

The False Start bug in NSS that you are talking about (CVE-2013-1740) never affected a released version of Firefox; we found the bug during testing of False Start in preparation for enabling it in Firefox 26. We pushed back the enabling of False Start to Firefox 28 (which will be released next month) so that we could fix this bug first.

You have an odd definition of "less exploitable". On what planet is compromise of all trusted communications "less exploitable" just because it doesn't immediately lead to code execution?

Once you are MITM'ing, you get passwords. You get personal details. You get everything you need to do far, far more than run code on someone's phone.

All the vulnerabilities you list are bad. Sometimes I wonder if you post your consistently contrarian posts on HN just to stay visible.

You could achive a full MITM and much more with the abillity to run arbitary code on a device. It is categoricly more exploitable.

This affects every single application using TLS, even when sandboxed. Is that more, or less, exploitable?

How many angels fit on a head of a pin?

The example provided was an arbitrary code execution in the same TLS library.

Tptacek isnt saying the issue isn't serious, it quite self evidently is, simply that other issues that appear regularly are also of equal or greater exploitability, and this being the bug that gave the NSA the keys to the kingdom is unlikely.

Code execution requires active compromise of target hosts. MITM'ing on the NSA scale can be done without risking any noticeable intrusion on the target hosts.

So what's "greater"?

Code execution.

My iPhone apps are sandboxed, but thanks for simmering down a complex, nuanced topic into an authoritative statement of contrarian truthiness.

Only on HN is this "contrarian". Exactly what does sandboxing do for you here? Consider the actual vulnerabilities carefully.

Sandboxing the app gives me a buffer between single application-level arbitrary code execution and full access to all application data and full phone code execution.

Broken TLS means all my sensitive data from all my apps are exposed.

Which one is worse? With my passwords, you could access our servers, cloud storage, corporate data. You could leverage this into VPN access, and from there, easy server-side code execution.

All without ever having to actually run code on my phone and potentially trigger the notice of observant parties.

I don't know the answer so certainly as you seem to, because it seems much more complicated a question to me.

You're not reading carefully. Memory corruption in PKCS padding impacts every program that uses TLS.

Hold up. Regardless of whether the media appreciates this point, a memory corruption bug in a relatively restricted (compared to the wealth of JavaScript) protocol like TLS has a decent chance of being very difficult to exploit reliably on modern systems, probably including the NSS bug you mentioned. The NSA might be able to do it, but it's not magic, and failure has some possibility (though low) of being detected; a bug like the current one, an exploit for which is basically 100% reliable if set up correctly and is probably less likely to be detected if not, is going to be a lot more appealing, and sniffing data rather than executing code is probably usually good enough for its purposes. Many other adversaries won't even try difficult exploits, but can easily set up this one now that it's been disclosed.

Admittedly, the above leaves aside the elephant in the room - regular old browser bugs, which are vastly more common, more likely to be reliable, and for the aforementioned other adversaries, relatively easily leveraged with exploit kits. As I assume you know, browser sandboxes provide protection against renderer bugs on MitMed or attacker-controlled origins being used to hijack a HTTPS site (and OS-level sandboxes protect non-browser SSL applications), but sandboxes get popped all the time and one of the two targets of this bug, iOS Safari, is single-process, so meh.

I agree the coverage of this bug is overblown (...depressingly, since that's just saying that modern computing is drastically insecure); I strongly doubt it was introduced intentionally, or that the NSA, if it knew about it, was "relying on" it in the sense that the disclosure is a huge loss. But it's still a particularly good bug.

I don't disagree with any of this. We're now litigating a thought experiment I proposed upthread. In particular: I agree that straight up browser bugs are more common, more exploitable, and more appealing than NSS parsing bugs; I just liked the symmetry of comparing a TLS logic flaw to a different TLS parsing bug in a codebase people weren't as familiar with.

Sandboxing isn't going to save you here, because a arbitrary code execution bug (via say the TLS library) inside the same process that has sensitive data leaves everything fully exposed.

Given that we know from leaked slides that the NSA has a policy of restricting the use of exploits in order to avoid information about them being compromised, and that obtaining the same level of access through code execution would involve leaving code behind that's at risk of being detected whereas this leaves no trace - which is actually the bigger compromise?

It's obviously less exploitable, because a memory corruption flaw that affected the TLS handshake would be devastating regardless of whether your application used a TLS connection to update your email password or refresh the available episodes on the Nightvale podcast feed, and the certificate validation bug isn't.

I would expect memory corruption bugs to be harder to exploit, because their exploiting might require knowledge of some particulars of the environment (maybe we need to know the exact version of the buggy library used, so that we know the memory layout/location of some symbols; maybe we need to know some particulars of the interaction between library and application like how long are the buffers that the application passes to the library). Am I completely wrong or are such problems easy enough to overcome?

You think this is the bug they're relying on?

Actually... yes. Memory corruption bugs are harder to exploit quietly, because if you get something wrong you'll get a crash instead of code execution. Missing certificate validation is much safer to exploit.

Tptacek, it is scary because of exactly that -- it is so simple. How could such a secure platform built by such a massively large cap tech company do something so incredibly stupid in its most critical security stack.

If this trivial bug can happen and pushed into production .. Good lord, anything can happen.

The foundation of our mobile economy has just been proven to be on very very shaky legs.

It would be nice if massive meant "more secure," but the real relationship is not between an organization's size and its security, rather it's about how much that organization tries to minimize, or displace, risk. Very large organizations tend to not specifically care that something is provably secure so much as they want contractor X to be liable if something bad happens because of it.

Some organizations--and even though it's easy to mock Apple here, I do believe they're one of them--sufficiently appreciate that improving security minimizes risk, and so spend a decent amount of money on it. You could argue they failed here, but even Adam Langley who essentially "is SSL" at Google admits he's not sure they have a test that simulates an attack on a possible similar implementation error in their own code (though he does argue that such an error would have been caught in Google's code review process.)

In any case, an attention to security is certainly not something that is borne out of an organization's revenue or number of employees.

I'm under the impression that Apple prides itself on having a smaller number of very dedicated employees that are perfectionists.

I can totally see this being "innocent 2AM mis-judgement call by a single employee", judging from what my peers tell me of Apple's corporate culture. I do think that Adam Langley's suggestion that code review would help is plausible, but it merely just means that more than one person has to make the same mistake in a judgement call. (It reduces the probability of such an error happening; it doesn't theoretically eliminate it.)

I can totally see this being "innocent 2AM mis-judgement call by a single employee", judging from what my peers tell me of Apple's corporate culture.

https://support.apple.com/library/APPLE/APPLECARE_ALLGEOS/HT...

We begin therefore where they are determined not to end, with the question whether any form of democratic self-government, anywhere, is consistent with the kind of massive, pervasive, surveillance into which the Unites Sta tes government has led not only us but the world.

This should not actually be a complicated inquiry.

http://snowdenandthefuture.info/events.html

FIPS 140-2 certification isn't remotely an indication of correctness of code, for better or worse.

Take, for example, the implementation Dual EC DRBG in the FIPS 140-2 certified OpenSSL module -- it was fatally flawed, and has never worked in practice. (It will be removed from the next version of the module in light of developments in the past year.)

https://lwn.net/Articles/578375/

> The foundation of our economy has just been proven to be on very very shaky legs.

Exactly. That sums up my feelings about it better than I could. Something so critical, something trivially easy to catch in a code review was not caught. And, the only scenario in which a code review might not have caught it: no code review. That's no code review for libssl.

If this is incompetence and not malice, it's incompetence of monumental proportions.

No. No. No.

Code reviews only mean your code is going through two sieves instead of one.

Of course it helps. But there is no guarantee of anything unless the reviewer is incapable of making mistakes, in which case you could just ask him to write the code in the first place.

> But there is no guarantee of anything unless the reviewer is incapable of making mistakes, in which case you could just ask him to write the code in the first place.

Did you look at the bug? I'll quote it here:

    if ((err = SSLHashSHA1.update(&hashCtx, &serverRandom)) != 0)  
        goto fail;
    if ((err = SSLHashSHA1.update(&hashCtx, &signedParams)) != 0)    
        goto fail;  
        goto fail;  
    if ((err = SSLHashSHA1.final(&hashCtx, &hashOut)) != 0)  
        goto fail;

Even somebody with basic programming skills can see that's wrong. And, remember this is libssl. Any checkins to that warrant thoroughness if not paranoia.

In the CR tool I assume it may have looked more like this:

        if ((err = SSLHashSHA1.update(&hashCtx, &serverRandom)) != 0)  
            goto fail;
        if ((err = SSLHashSHA1.update(&hashCtx, &signedParams)) != 0)    
            goto fail;
    -   if ((err = SSLHashSHA1.update(&hashCtx, &somethingElse)) != 0)
            goto fail;  
        if ((err = SSLHashSHA1.final(&hashCtx, &hashOut)) != 0)  
            goto fail;

Which is a little harder to see. Obviously you should always look at it in a side-by-side view (god I wish github would implement this) or at the resulting code, but people are imperfect.

Damn fine point. I really don't think we can conclude this is malice; do all of our own bugs always turn out to be really tricky to spot? Do we never make utterly ridiculous mistakes along these lines?

Doesn't look like any (major) modification to surrounding lines.

    if ((err = ReadyHash(&SSLHashSHA1, &hashCtx, ctx)) != 0)
        goto fail;

changes to:

    if ((err = ReadyHash(&SSLHashSHA1, &hashCtx)) != 0)
        goto fail;

See: http://www.diffnow.com/?report=ob51k

Diff35

We only see the diff between released versions, not intermediate commits. For all we know, Apple developers use a single-pane diff tool, where such bugs are easy to miss.

Except that I don't think this was the original code considering it appears to be cut and paste of same code in same file. I commented on this: https://news.ycombinator.com/item?id=7286582

I seriously don't understand why so many people don't habitually use braces there...

  if (...) {
    goto fail;
  }
  if (...) {
    goto fail;
    goto fail;
  }
  if (...) {
    goto fail;
  }

tada, not actually a bug!

Which is fine, but in my experience flaws like this are often introduced via automated merges that could as easily have resulted in:

  if (...) {
    goto fail;
  }
  if (...) {
    goto fail;
  }
    goto fail;
  if (...) {
    goto fail;
  }

which produces the same bug. (Everyone's been shouting about "braces in single statement if clauses" as though they're an absolute fix; they're not, although they're a good idea in general. And yes, code review, better merge tools, yada yada.)

Agreed. That and blank lines between the if blocks to visually separate the code blocks as well.

Interesting, does this mean if you were using an updated SHA256 hash (as will be required soon for all EV certs) that the exploit would not have occurred?

> do something so incredibly stupid in its most critical security stack.

I disagree. We are human. We are not divine. We do stupid things and we can write stupid infinite loop or off-by-1 and the bug will live for a decade. I am sure you have written a program which can be easily exploited and you know that it is such a trivial bug.

Yes, the programmer is only human. But the vp of engineering should have had a budget and process that would not let this slip through.

Are you saying they should have extra people doing code review - requires two instead of 1 (I don't know the number, but assume that's the norm). Or more rigorous testing technique? I am all for it. Is this something they can easily discover through automated testing?

having extra people doing code review risks something that happens with physical products.

You take the experienced people off the shop floor, and call them inspectors. So now the quality of product coming off the shop floor is worse.

Ann is the first inspector in the chain. When she's busy (and remember, she will be because quality has dropped) she might be tempted to let a few things go because Bob, the second inspector in the chain, is bound to catch them. That's what he's there for.

Bob is the second inspector in the chain. Sometimes Ann really churns the product through. Luckily during those busy times he knows she's already inspected the stuff, so he only needs to give a 10% inspection.

So you have worse product with more errors and leaky inspection.

Ideally you'd have a system with skilled workers and self inspection. That's okay for aerospace (which pays well) but not so great for lower cost product.

Not quite sure how this transfers to coding.

Okay. Good point but I hope if people take the job seriously they will have to be very careful. Though that's ideal..

How about this:

owner/peer of the module has to sign off, and randomly select two more. One must be QA and one is another programmer who works on the module. We can also pick a "junior" level, but then that's probably not going to work since Apple employes programmers who have some years of experience already. Or we can pick someone who isn't directly working on that module, but have some qualification to do review. Mostly just asking "why are you having two goto, why return -100 here)

But I see counterargument: they will just listen to the author if he's senior or the owner who is also a senior. Their words carry weight.

That's assuming they are full time at inspection, another scenario is where one person does it in addition to their current responsibilities. Therefore the overall quantity of what's produced is reduced but the inspection levels are higher.

The vp of engineering should have invested in a good static analysis tool which would have spotted this section in about five seconds. Failing that turning on a sensible set of defaults in the compiler (warning about unreachable code) would have also drawn attention to it.

There are many many ways this should have been caught before it even left the building by numerous automated tools. Heck most modern IDEs will flag unreachable code in the editor so there is no real excuse for this. (Adding -Wunreachable-code to a sample project in Xcode immediately flags the next line).

The programmer is human which is why automated verification tools exist.

(AppCode's default inspections will also flag the issue providing you run them)

I'd be really really surprised they didn't run this through static analysis. I will, to be honest, then maybe people are right about "it's stupid they make such stupid mistake."

I am not familiar with running static analysis and looking output, only did very minimal undefined behavior santisizer detection.

I haven't been keeping up, but a year ago Google wasn't using any static analyzer at all on Chrome. They didn't use a real memory access checker like valgrind even to load a blank page, although they came up with a weaker memcheck that mostly works.

Lax development is par for the course at these mega corps.

I cannot comment on Chrome itself, but internal C++ development at Google has a pretty comprehensive array of checks, both static and dynamic.

We've been using valgrind bots for a long time on the Chromium project - certainly for much longer than a year.

For example, according to the Wayback Machine, our sites page had an entry about it since 2009:

http://web.archive.org/web/20091123132855/http://www.chromiu...

I was thinking of this:

http://www.reddit.com/r/programming/comments/mw2sn/googles_a...

This is a comment that seems to suggest that it's abnormal for dumb bugs to cause huge security problems. It is not. If the foundation of our economy is C software security, well, hate to break it to you, but...

What strikes me is that the bug looks like a typo but it implements a logic error and that logic error pretty much negates everything the library represents itself as doing and what the library represents itself as doing is providing a foundation upon which signals security relies.

Suppose a contractor was hired by the NSA to write an exploit with equivalent function. How could it be crafted more cleverly?

It is absolutely abnormal for companies like Apple to release core security bugs this shallow that could have been easily discovered by straightforward unit tests and static analysis tools.

This is why it's a big deal.

The other reason why this is no big deal (anymore) is that the Snowden leaks have shown that the NSA has total control over all iPhones.

Why should we even bother talking about bugs like this anymore? Pure distraction.

Which is worse: the NSA having total control over all iPhones (citation? I obviously haven't been paying enough attention), or the NSA and all the (other) bad guys in the world having total control ...? Sure, they're both terrible, but I'd take the former over the latter.

> the Snowden leaks have shown that the NSA has total control over all iPhones

You mean: had total control over all the original iPhones that they could get physical access to. (back when jailbreaking was extremely simple and common)

Where does the "physical access" part come from? And if that was the case, why would it be impossible now?

One of the sources: http://www.forbes.com/sites/erikkain/2013/12/30/the-nsa-repo...

From the slide on your very link: "The initial release of DROPOUTJEEP will focus on installing the implant via close access methods. A remote installation capability will be pursued in a future release."

Basically how it worked was they jailbroke your iPhone and installed spyware on it. Is it quite likely that today, 7 years later, they have a remote 0day to do the same? Absolutely. But there's no proof that "the NSA has total control over all iPhones".

The ease with which a bug can be exploited has everything to do with its impact. Exploiting buffer overflows is messy, requires a lot of effort, and it's detectable. Thus, you are more likely to use it for something special.

Apple's TLS handshake bug was trivially easy to exploit, entirely silently, in a fully automated fashion and with zero chance of detection and without leaving a trace. That's why it's a big deal.

If you are in the business of collecting massive amounts of data, this is exactly the type of bug that you would be using. SSL clients leak a large amount of data in the handshake (supported protocol versions, cipher suites, extensions, etc), allowing you to fingerprint them and detect the vulnerable ones on per connection basis. You then attack, safe in knowing that you won't be detected.

Once you get the victim's password (e.g., if you attack their email client), you can suck out all the data you want. Because you are the MITM, you can do it from their own IP address, too. All fully automated.

I think you're assuming that a platform can detect a buffer overflow. How exactly is an iPhone or iPad going to detect a buffer overflow?

I think we can all agree that owning the entire platform on a device is far worse than a MITM attack on a session. Even if it's all the sessions on that platform using TLS/SSL. executing code on the box means you can do many other things on that box, such as setting up a pivot point and keeping access to the platform as long as you want. If you can root the platform, then all bets are off.

Yes, we can agree that owning the entire platform is worse. But my point is that you can't ever achieve that without someone noticing. Think millions of devices. The malware would have to work flawlessly across a number of versions, interacting with many components installed, and even on jailbroken devices. It is very risky and intrusive, which makes it much less likely. An attack against TLS connection authentication is so _clean_ in comparison, that I can easily imagine various parties jumping to the opportunity.

You seem to be downplaying the seriousness of the bug to the point of making it seem trivial, and while I agree that there are many, much more serious types of bugs that 1. don't require active MITM, and 2. the press don't give proportionate attention, it is only "trivial" in the same way that Yahoo not using SSL for their webmail is "trivial," or the way a website leaking an authentication cookie over HTTP is. Only here it's not just Yahoo's email; it's (almost) all applications. (Not to mention that you can use these "secure" channels to deliver e.g. memory corruption exploits.)

If you don't have authenticity, the "secure" channel isn't worth very much.

If you're going to inexplicably speak directly to the author, and then do so by their first name, at least get the spelling right.

That's a perfectly understandable, and common, style of writing; it's not necessarily literally directed at the author. And surely the spelling is important whether it be the first name, or any other.

You're arguing that memory corruption is a worse vulnerability, which is true, but doesn't contradict the claim that the NSA could exploit this vulnerability to intercept traffic and add Apple devices to PRISM. So ultimately you fail to demonstrate that Gruber's post is "silly".

TLS validation bypass + automatic software update using TLS as authentication mechanism = RCE

I'm not sure why this is being classed differently.

Not quite. iphone enforces sig verification on write to the flash. A modern iPhone does not allow arbitrary base flash images to be written. Modern Jailbreak relies on kernel and user mode exploitation. Server auth bypass on SecureTransport TLS cannot lead to RCE via trojanized updates.

I think he's referring to app updates, not iOS updates. Also, desktop Mac OS X.

App installs and updates are still subject to code signing, and unless you jailbroke the device by some other means, any code that would be able to mess with the OS' public keys used to to authenticate the signatures would have to be signed itself and presumably would have to come in the form of a signed OS update.

Though ... if anywhere uses Safari on a regular basis and thus trusts their SSL to it, that would be Apple, yes? ;-)

Sparkle Framework.

Yep, you're right. PRISM is a process through which NSA obtains legal orders to demand information from companies through a federal law known as FAA 702. I tried to set the record straight on this last summer: http://news.cnet.com/8301-13578_3-57588337-38/

Since then, we've learned that FAA 702 orders add up to a tiny fraction of user accounts. The order of magnitude is 1,000 accounts per company per year. I really can't get too worked up about this; there are bad guys out there and that figure seems not immediately unreasonable. What we should get worked up about is bulk AT&T etc. fiber taps that vacuum up everything without any accountability, but, well, PRISM is a nice sexy name and everyone's attention spans are short and, yikes, isn't this confusing and BULLRUN and EDGEHILL are too hard to remember. Right?

No companies "joined" PRISM. The slides are likely referring to when NSA managed to write the conversion utilities to import FAA 702-obtained records into the PRISM database. But critical thinking is hard and life is short. Right?

Obviously, I trust you to know about these things, and you point to other bad examples which I'm not familiar with. In this context though, code execution isn't necessarily important (though I'm sure that's a dream scenario). The ability to snoop on assumed secure traffic is a pretty sweet starting point.

I take it you're familiar with threat trees? Any vulnerability is serious. It's just another vector that either realises a threat, or brings the attacker closer to realising a threat.

> it requires a specific (common, but not universal) set of circumstances to escalate it to code execution.

Man in the middle attack does not require code execution because the data is already available.

In a sense, this bug is like having part of a keyboard sniffer already installed for you.

So even though this bug is easy to exploit and deployed on hundreds of millions of devices, it's not a big deal because it doesn't give you root?

Except this bug is passively exploitable at scale.

It's the NSA, of course they knew about the bug.

It probably wasn't the only trick in their book, but it was surely one of them (and a quite useful one at that).

I appreciate the technical insight into the impact of this bug versus other commonly found vulnerabilities. I don't disagree, but it seems to me this insight is tangential to the larger story. It's a (bit) of a blow to lay men's trust of Apple if they can easily understand why a bug occurred without having to dive into the arcane details known to researchers such as yourself.

To me, the interesting question brought up by this story is how soon the vulnerability was known to potential exploiters. Did they just have to have an appropriate set of black box tests rigorously applied to every major software release to spot this? That would, sadly, diminish my trust in the quality of Apple's engineering, though I don't see a reason to attribute it to an inside job without any evidence.

> And in terms of feasibility, this bug is less exploitable than the kinds of bugs that are disclosed in every platform every month; it requires a specific (common, but not universal) set of circumstances to escalate it to code execution.

What I'm afraid of is an attacker stealing my credentials. Code execution bugs obviously is game over for me. But now with everything in the cloud, an attacker could get all I'm afraid of losing by fucking with my SSL connections. If they get access to everything I have on the web the fact that my machine executed no malicious code locally is no consolation.

I could be misunderstanding, but are you responding to the point made by the timing of the NSA's "Announcement" of their abilities with Apple products? Jon himself admits his evidence leads him to believe #3 on his list. Is that silly?

Why is it every time I read a post by you tptacek, my troll detector goes off.

Kind of makes me want to investigate what work you and your company has done for the NSA.

He was also drinking

Nice ad hominem, dude.