You Have to See the Two Little Words That Cost Apple Millions

Decoding the tech world.
Feb. 25 2014 4:24 PM

An Extraordinary Kind of Stupid

The weirdest thing about the Apple security bug is how simple it was.

Some iPads are among the devices affected by Apple's security bug.
Some iPads are among the devices affected by Apple's security bug.

Photo by Jessica Rinaldi/Reuters

So for the past 18 months, there has been a horrific security hole in many of Apple’s products that has allowed “man in the middle” attacks on supposedly secure Internet communications. Most iPhone, iPad, and iPod Touch devices were affected, as well as tremendous numbers of Macs running current and recent versions of OS X (any version of the 10.9 Mavericks release). (You can go to gotofail.com to see if your device is affected.) This vulnerability is exceptionally bad and ubiquitous, but it’s still the same sort of bug that gets patched constantly in various pieces of software and for which hacker groups are constantly on the lookout.

David Auerbach David Auerbach

David Auerbach is a writer and software engineer based in New York. His website is http://davidauerba.ch.

Aside from its severity, though, this bug has another extraordinary quality: It’s extremely simple. (Simple enough that the bug is already on a T-shirt.) Stupid, even. Ninety-nine percent of the time, these sorts of stupid mistakes aren’t that damaging. But that 1 percent of the time, the gods won’t save you.

Because the code in question was open source, some folks on YCombinator quickly located it; they pegged it as popping up first in the 10.9 release of OS X code. Google Web security guru Adam Langley posted a good technical analysis of the bug. But noncoders should know something about it too, because this bug is an object lesson in just how fragile the code that increasingly controls our lives can be. The simplicity with which a single mistaken line of code snowballed into one of the biggest security holes ever strikes fear into the hearts of engineers. It’s good to peek under the hood.

Advertisement

Below is the C code containing the bug, which occurs deep down in a security function called SSLVerifySignedServerKeyExchange. This function makes sure that the site your computer is talking to over an encrypted line (like google.com or chase.com) is really that site, rather than some “man in the middle” pretending to be that site. The bug causes the function to claim that the site is legit, even if it’s not.

        OSStatus err;

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

        fail:
            return err;

Even if you’ve never seen code before, you might pick up on a glaring structural anomaly here, which is that one “if” statement is followed by two “goto fail”s instead of one. If that jumped out at you, congratulations! You found the bug.

Let me simplify the code a bit to get at the essence of the bug:

        OSStatus status = EVERYTHING_IS_GREAT;

            if ((status = DoSomeSecurityStuff) is DANGER)
                goto fail;
            if ((status = KeepDoingSecurityStuff) is DANGER)
                goto fail;
                goto fail;
            if ((status = DoTheMostImportantSecurityStuff) is DANGER)
                goto fail;

        fail:
            return status;

These lines perform some calculations that test the validity of the authenticating data that the server (real or fake) has sent back to you, the client. On each “if” line, a variable labeled “status” is set to either EVERYTHING_IS_GREAT or DANGER. If status is still EVERYTHING_IS_GREAT by the end of the code, the function tells the rest of the program that everything is indeed great and authentication took place. If something went wrong, then it returns DANGER to the rest of the program. Whenever any part of the check returns DANGER, the code doesn’t bother finishing the security check—it’s already failed, so why bother? It just jumps to the end of the code via the “goto fail” statement, which causes the computer to jump to the end of the code with the “fail:” label right above it.

  Slate Plus
Slate Picks
Nov. 21 2014 1:38 PM What Happened at Slate This Week? See if you can keep pace with the copy desk, Slate’s most comprehensive reading team.