Since the usage of zip in that case will only compare the shorter of the strings to the start of the longer string, you can still achieve the timing attack, since the only inconsistency happens on the final comparison.

A better solution would be to check for length equality before doing the zip, or to pad the shorter of the strings compared to the length of the longer. Maybe also to just make the return statement:

return (len(a) == len(b)) and (result == 0)

which may be sufficient.

Anyway, rambling thoughts aside: thanks for the article.

The implementation of memcmp bails out as soon as it finds a difference:

  int
  memcmp(const void *s1, const void *s2, size_t len)
  {
    register unsigned char *a = (unsigned char*)s1;
    register unsigned char *b = (unsigned char*)s2;
    register int tmp;

    for (; len; --len) {
        if (tmp = *a++ - *b++)
            return tmp;
    }
    return 0;
  }

Seems like other, more traditional problems (weak / poor crypto, social engineering) are much more important problems.

I kind of wonder if people just really like timing attacks because they're so easy to understand, and, once you spot them, so easy to fix.

Crosby et al. conclude: We have shown that, even though the Internet induces significant timing jitter, we can reliably distinguish remote timing differences as low as 20µs. A LAN environment has lower timing jitter, allowing us to reliably distinguish remote timing differences as small as 100ns (possibly even smaller). These precise timing differences can be distinguished with only hundreds or possibly thousands of measurements.

If you'd like to bet that this sort of statistical analysis will get harder to perform over time, I'll take your wager. I'm not willing to wait until someone is publicly compromised before I fix security vulnerabilities—I do not work for an employer who's willing to tolerate a few break-ins before fixing the locks.

I'd also argue that a) a timing attack vulnerability is "weak/poor crypto," b) I don't need to write about the importance of other attack vectors (XSS, CRSF, social engineering, telepathy, bats, etc.) while writing about timing attacks, and c) if timing attacks are easy to understand and fix there would be less of them.

(Edit: fixed the B&B paper link. Derp.)

(edited to have the link i meant).

This is a mathematical problem like poker. On one hand, it exposes the issue for attack by hackers who didn't know about it before. But it also makes systems in the future more secure. What is the probability that a malicious hacker has found this issue before you and is using it to attack java based systems? What is the probability that after you expose the issue, all the vulnerable systems will be patched with your suggestion?

I'm curious what other hackers think of this. I've found XSS vulnerabilities in sites and other security issues. What do you do when you find these? Do you notify the sites? Sometimes I notify the website of the issue, sometimes I don't. I struggle with the decision myself, but I don't think I'd ever go open with such a vulnerability without first being more determined about getting a response from those who can fix the issue before the hackers find it.

http://www.google.com/codesearch?hl=en&q=digest\.equals+...

For example, it looks like org.apache.ws.security.processor.UsernameTokenProcessor.handleUsernameToken might be vulnerable (although WS-Security's nonces and timestamps make things more complicated, since you have to generate a timestamp like a week in the future, and then hope that the server does the digest comparison before rejecting you based on the nonce or timestamp):

http://www.google.com/codesearch/p?hl=en&sa=N&cd=31&...

http://www.cs.rice.edu/~dwallach/pub/crosby-timing2009.pdf

Adding jitter will not defeat timing attacks unless the jitter is dependent on the input, like blinded RSA.

You keep track of the time while doing the comparison, then before spitting back the result you feed the amount of time to a separate function that will sleep (or whatever) a certain amount of predetermined time so that all results process in the same time.

I realize this is a bit excessive, and will slow down your code quite a bit, but if you are really that worried about this type of attack, why not?

(In some languages—hi Ruby!—you might not even have access to reliable millisecond-resolution timing.)

The downside is that while the solution given in the OP will use a constant amount of time, that constant amount of time itself is not fixed. When over the years processors get faster, the code can execute faster, but in the above case it cannot.

Sure it can. You can drop the limit whenever you change processors. In fact, you can tune the limit for a given call to be the max time taken by that call. As the implementation or resources change, the limit changes.

That's easier said that done. Also, you have to come up with one for every security-relevant operation.

And, "algorithm" isn't the issue, implementation is. Given caching and the like, that's really hard to do.

An easy way to implement the "sleep to worst case" idea is to simply time each call and keep track of the max. If a given call instance returns in less time, sleep the difference. (It's probably good to add a decay mechanism.)

Although nanosleep() takes a timespec struct that allows you to give an accuracy with nanosecond precision, nanosleep doesn't really give anywhere near that level of accuracy. Under Linux, the kernel only checks on timers once every "jiffie," which defaults to 1 ms under the 2.6 kernel (on older kernels it was 10 ms, and 10 ms is still a compile option in the 2.6 kernel). Once upon a time, nanosleep would busy-wait if the sleep time given was less than 2 ms, but not anymore. On the other hand, don't be too discouraged—as long as you don't need submillisecond accuracy, nanosleep() is very good…

http://www.ddj.com/linux-open-source/184402031

Note that caching and branch prediction and the like make "constant time" implementations almost impossible.

Caching is equally unlikely to play a role. You're probably thinking of DJB's timing attacks on AES (http://cr.yp.to/antiforgery/cachetiming-20050414.pdf), but that attack resolves around the timing of S-box lookups in which the index of the lookup is key-dependent. In the constant-time algorithm proposed by Nate Lawson and others, the only array lookup is the incremented value of a for loop (i.e., constant-time). The other operations are XOR and OR, both of which are constant-time.

Branch prediction doesn't come into play because a constant-time comparison doesn't involve conditionals until all the data has been compared.

A constant-time implementation of AES? Hard. A constant-time implementation of an array comparison? Dead simple.

Not sure why you're so resistant to the idea.