About the security content of iOS 7.0.6

FWIW, the first time I went to https://notinthecert.imperialviolet.org in Mobile Safari on iOS 5.1.1, I got a "Cannot Verify Server Identity" error with Cancel, Details, and Continue buttons. After pressing Continue, it remembered my choice and did not alert me the second time.

Lock screen has been vulnerable with a bypass exploit on several occasions so caution is probably a good idea. Way too many times to give me any sort of confidence.

Indeed. My first thought upon seeing this: "how in the FUCK did automated testing not catch this IMMEDIATELY!?"

Absolutely terrifying that crypto safety is such a low QA priority that something like this could ever leave the building.

As it happens, Secure Transport (edit: on OS X) is open source. I just spent the last hour rummaging through source code ... It was later renamed to "libsecurity_ssl" when it landed on iOS. It's been around since OS X launched. They also have a folder with about four dozen regression tests and a test app, not that I've had the chance to inspect either that closely. The tarballs are spread across two folders in the tarballs index: Security (look for multi-MB file sizes) and libsecurity_ssl. The 5-digit numbers appear to be sequential revision IDs, any single-digit numbers seem historical. (E.g. libsecurity_ssl-6 is before libsecurity_ssl-10 rather than after libsecurity_ssl-55471.)

It appears they haven't posted newer source than this. The most recent timestamp I could find was Oct 11, 2013 in 55471, which corresponds to 7.0 and my 10.9 system has the same version number for Security.framework -- same bundle version of 55471 for 10.9.1 aka 13A581. Previous version numbers don't appear to be as well-maintained. I don't expect a newer release to be posted until the next OS X release, as the source was only published under 10.9, not iOS. Additionally, there's no mention of iOS 7 on http://www.opensource.apple.com/

I couldn't easily find the bug without more to go on, because the code is spread across a few components and really, I'm not an expert in TLS. It appears to have been largely unchanged from 2000-2006 or so. TLS 1.2 brought quite a few changes, but it was neat to browse through the lines of "FIXME" and "TODO" comments, as well as various diffs between releases. And neat to see how much code today still goes back to 1999-2001, sometimes all they did was add a 'k' in front of a few variable names or delete the line in the first README saying the server code wasn't tested against Windows ;-)

It sounds like when 10.9.2 is released, or at worst when 10.10 comes along, you'll see a new push of code to the opensource site. We can all diff 55471 against what comes next to see the changes. (If someone's already running 10.9.2 and its unaffected by the bug exhibited via curl, open /System/Library/Frameworks/Security.framework/Versions/Current/Resources/Info.plist and post the Bundle version.)

Gruber, is that you?