Hacker News Reader

> NSS has more institutional constraints; random people in Germany can't, as a general rule, add support for new TLS extensions to it.

Why not? NSS is handled by Mozilla. Mozilla prides itself on being open-source. People can submit patches to NSS by filing in Bugzilla.

> The most important code in both NSS and OpenSSL is the SSL state machine. The code for that state machine is incontrovertibly clearer and better expressed in NSS.

I agree with this to some extent (although I don't think it's quite as much better as you suggest), and cleanups in that code are needed. Other parts of NSS can certainly rival OpenSSL on the "dear god" scale however. I'll take the low-level parts of OpenSSL over NSS any day.

> NSS has had better battle-testing as a clientside browser TLS library than OpenSSL, which, apart from Android Chrome, isn't a big factor in TLS clientsides.

This is true. We will have to rack our brains and our bug database for some of the unwritten rules of TLS to port over to OpenSSL. I'm not too worried about this in the long term however.

Ultimately the proof will be in the pudding. Ask again in 12 months :)

>> NSS has more institutional constraints; random people in Germany can't, as a general rule, add support for new TLS extensions to it.

> Why not? NSS is handled by Mozilla. Mozilla prides itself on being open-source. People can submit patches to NSS by filing in Bugzilla.

Maybe he meant Mozilla as an organization is much better equipped to handle patching of NSS better than the OpenSSL team. I have seen Firefox bug reports number of times and there seems to be a proper process in place to review patches before accepting. I am not aware of anything that good on the OpenSSL side.

Am I wrong that it's easier to get code into OpenSSL, and thus into pretty much every product that uses OpenSSL, than it is to get code into the branch of NSS that Firefox and Chromium use?

I've submitted code to both. I don't feel that there's a huge difference. (Although NSS has Wan-Teh reviewing, who is a great code reviewer.)

Ok, wait: the code you've submitted to both has also probably been more obviously applicable than, say, the Heartbeat extension.

"Chromium developer has elliptic curve code accepted by NSS: film at 11."

Part of the reason it's so hard to get code into NSS is because no one really knows who owns those bits, and so they linger in obscurity. I think my TLS Channel Bindings patch is lingering on 3 or 4 years now.

However, I think you're doing a disservice if you only look at how easy it is to add code. As important is how easy it is to remove code. With NSS, it's far, far harder to remove code - and that makes it far, far harder to reason about any new code you might want to add, for who knows what grues are lurking in those shadowy depths.

Not being able to easily add or remove code is good from a security standpoint though. I would take a bugfix only, stable-API NSS over anything that is in constant flux. So it's not how fast you can add features that should be in focus for a mature product - it's how safely you can add bug-fixes, including security fixes. Unless you are saying you have a security fix lingering for 3-4 years, I think Mozilla is doing it right.

Edit: Not to ignore the valid point - having a designated code owner that responds would of course be good.

With NSS, it's far, far harder to remove code

I'm not sure how to interpret that statement. It could mean any of the following, or something I'm not even thinking of:

The code is inscrutable and/or the documentation is poor: That's obviously a problem.

Commits removing code are delayed/ignored: Probably also a problem, but maybe there's valid reasons for this, such as more thorough testing?

There's little cruft to remove: This would be a positive. Not having much of a mess to clean is hard to consider a problem.

Am I close with any of those?

It could also be that the public API is too large and exposes unnecessary bits that now make it hard to change even incidental, conceptually irrelevant behavior simply because some user probably relies on it. Even a clean, well documented implementation with active, capable maintainers is hard to refactor when the API is large and intrusive.

(I have no reason to believe this applies to NSS.)

> Am I wrong that it's easier to get code into OpenSSL

I might be wrong but the idea I have is that currently Google more or less owns OpenSSL and thus it is maybe more easy, makes more sense for them to get patches in OpenSSL and try to migrate exclusively toward OpenSSL.

Another question I have is since when the HeartBleed flaw was discovered by the Google's employee? Was it formally reported to the OpenSSL team or only to OpenSSL members working for Google? And if the other independant group of engineers had not independantly discovered the flaw would have it been disclosed as quickly as it has? Bottom line it seems to me (and I'd like to be wrong) that the current opacity of OpenSSL disserves everyone but Google.

Another question I have is since when the HeartBleed flaw was discovered by the Google's employee? Was it formally reported to the OpenSSL team or only to OpenSSL members working for Google?

I think it is generally regarded as sensible practice to disclose major problems like heartbleed that have wide ranging impact directly to the people responsible instead of a catchall "security team" address whenever possible.

I know nothing about crypto, but I found comparing both codebases at a high level very instructive from a pure software engineering / project management standpoint:

ftp://ftp.mozilla.org/pub/mozilla.org/security/nss/releases/NSS_3_16_RTM/src/nss-3.16-with-nspr-4.10.4.tar.gz

https://github.com/openssl/openssl

https://www.openssl.org/source/openssl-1.0.1g.tar.gz

In terms of sheer architecture, those two are almost epitomes of a very well maintained project (NSS) and a very poorly maintained one (OpenSSL). The OpenSSL codebase looks like it evolved for years with seldom any cleanups being done, with many people doing contributions, each person in its own unique style without much unification going on. The directory structure is chaotic, documentation is sparse, commented out code is laying around without any documented purpose, and browsing code at random you find things like this:

  /* This should be using (bs-1) and bs instead of 7 and 8, but
   * what the hell. */
   if (bs == 8)
       l=(l+7)/8*8;

It's pretty much what I would expect to happen if you simply gave everyone interested commit access, reviewed and refactored nothing, and only protested if something gets broken functionally. It's like an orchestra playing without a conductor, just a collection of "stuff" loosely connected together and corrected only as immediate needs surface. It might contain incredible individual pearls of crypto-wisdom of course, and tons of work has been put into it, but the technical debt is huge and paying it now all at once would be hard. I looked at GNU Emacs code a while ago, and had pretty similar overall impression, they for example have a test directory where every individual test suite (file) is done in a completely different style, being by a different author.

If you look at NSS on the other hand, you can tell there is someone who actually cares about the health of the codebase on a day to day basis. The directory structure is much more thought out, there are clear separate modules, there are way more tests, you can generate test coverage statistics, most function signatures are documented, there doesn't seem to be much redundant stuff in it, and so forth.

> NSS has more institutional constraints; random people in Germany can't, as a general rule, add support for new TLS extensions to it.

This is horrible for a large scale open-source project. Glibc, gcc, as examples. Trying to get them to fix their broken crap or reverse their idiotic decisions takes decades. If OpenSSL is less opposed to change, it's probably a good thing, considering the purpose of it.

> NSS has a clearer identity, as the TLS core for Firefox and Chromium. OpenSSL is a grab bag with hundreds of different stakeholders.

Do you not remember the myriad bugs that came out of Microsoft's, Netscape's and others' independently-developed SSL implementations when their only relevant benchmark was their own tools? They didn't give a shit if it broke someone else's tool because they didn't make it to support someone else's tool.

> The most important code in both NSS and OpenSSL is the SSL state machine. The code for that state machine is incontrovertibly clearer and better expressed in NSS.

So clean up the OpenSSL code! If your previous claim that it's easy to get code into OpenSSL is true, fix it! Don't throw the baby out with the bathwater!

> NSS has had better battle-testing as a clientside browser TLS library than OpenSSL, which, apart from Android Chrome, isn't a big factor in TLS clientsides.

OpenSSL is used in hundreds of TLS clients in all kinds of environments around the world. It's the de facto client library for 90% of open source tools. As opposed to whatever environments run NSS applications which is far fewer.

It seems like these are arguments for OpenSSL, not against it.

It's the de facto client library for everything but the most important TLS clients.

Ryan's argument that it's too hard to remove NSS code is compelling, but making it to easy to add code is exactly the problem we just had with OpenSSL. Heartbeat wasn't just bad code; it was code that was inappropriate for inclusion by default.

Would clean up of the OpenSSL code be accepted? I mean just from a quick glance at the fragments that have been posted, the first thing that would need to be done would be to get rid of all the ifdefs, i.e. decide on a canonical value for each of the flags in question; would this be accepted, or would the existing stakeholders respond with a list of reasons why they are committed to keeping the status quo?

> Maybe a fork of NSS would be a better plan.

Is forking it actually needed? I was under the impression that Mozilla has many of the same complaints about the current state of NSS as the Chrome team and merely don't want to dedicate the resources to cleaning it up.

"Don't want to" is a bit uncharitable. We want to, but have limited resources and a lot to do. We actually recently rewrote the certificate validation library and it is currently on Nightly. Check it out!

Of course, it is open source, so patches are always welcome. (Firefox dev here)

> My bet is that the answers to those questions will be more favorable for NSS, even with the crappy certificate handling.

Here's the thing, the SSL libraries are so much more than "just" SSL libraries, and products that use them have to weigh a lot of things in when comparing.

- Platform abstraction layers

* NSS builds atop NSPR, which is the Netscape Portable Runtime from ye-old

* OpenSSL doesn't have much of an abstraction layer - short of perhaps BIO* hiding the nature of POSIX FDs

- Scope of what they try to do

* OpenSSL attempts to provide basic BER/DER parsing, some mid-level abstractions (eg: X509* and the related structures), a "high-level" protocol (SSL), but without any 'productization'

* NSS attempts to provide a hardware abstraction layer (PKCS#11), a full certificate verification library (for both SSL/TLS and S/MIME/CMS - multiple ones, I should add), abstract notions of 'trust domains' within multi-process and single-process servers, etc

- How much flexibility they give developers

* OpenSSL gives you enough rope to hang yourself, and many developers often do. However, at the same time, it's popularity is in part due to it's generous supply of rope

* NSS is wildly inconsistent, with public APIs being adhoc (eg: for any given object, you're lucky if you find all four functions for CRUD, and even luckier if they're internally consistent, and you hit the jackpot if they're externally consistent).

As I mentioned on Twitter, there is no doubt that the SSL state machine of NSS is probably cleaner than OpenSSL's, having worked with both. However, there's a ton of hidden magic that NSS tries to handle, by design, that makes it impossible to reason about how well something will behave, what side-effects it might have, and what the security impact might be.

For an application like Chrome, I will highlight that some of the biggest challenges we've had is in ensuring proper privilege separation when validating certificates with different trust scopes. Some connections only wish to trust certain authorities, others may be ambivalent, and yet other connections may have a mixture of 'default' authorities plus administratively configured settings. Actually ensuring the separation of these authorities is a monumental task within NSS.

That's not because NSS doesn't want to deal with that problem - no, there have been multiple, half-completed attempts to do so, which have simply introduced layer upon layer of abstraction.

NSS is an excellent piece of software for some tasks. But for security software, the last thing you want is "magic surprises", which NSS is full of. Even though the SSL state machine is possible to reason about, if you can't reason about whether or not a given certificate will be marked as "trusted", then you're still in the danger zone.

This makes me wonder what will happen with the level of security in NSS, now that Google is quitting it. Can Mozilla keep maintaining it properly on its own?

They don't maintain it properly today. Go look at bugzilla, there are quite a few new feature patches from Google that just never get reviewed, let alone merged.

Many of the review peers work at Google, so eh...