5 ms·
The Heartbleed Aftermath: all CloudFlare certificates revoked and reissued
- dfc 13y agoI love the reaction from ISC's handler on duty. Globalsign's CRLs were temporarily removed from the dataset until they could verify it was "real": > Update: We temporarily stopped reading > “http://crl.globalsign.com/gs/gsorganizationvalg2.crl” . It had > 30,000 > revocations today, drowning the rest of the data. Trying to figure out if > this is real. (4/16/2014 7:30pm)x > > Update 2: I used the online chat to talk to GlobalSign tech support and > they confirmed that it is real and related to the revocations due to > Heartbleed. Adding them to the graph again. (4/16 8pm ET)
- filmgirlcw 13y agoYes! I laughed at that too. Though it makes sense that a mass-increase in data like that would seem off, especially when it is so much higher.
- 3JPLW 13y agoReally great writeup. Here's the most astounding part to me: "[Rubin Xu] decided to use Coppersmith's Attack[1] to derive the private exponent. In this attack, only pieces of the private key are needed and the private key is reconstructed mathematically. This attack is so efficient that he managed to extract the private key from only the first 50 dumps!" That's no more than ~3MB of data. Some of the other winners hit the server with hundreds of thousands of requests which might have gotten noticed. Doing this with only 50 heartbeats (which could be spread across many days) makes the "undetectable attack" even less detectable. [1]. Coppersmith's Attack: http://en.wikipedia.org/wiki/Coppersmith's_Attack http://en.wikipedia.org/wiki/Coppersmith's_Attack
- vizzah 13y agoWould love to see more details on the practical implementation of this attack / source code. Whilst detecting primes in random data is feasible, can't imagine how chunks of returned data were "unrandomized" for this type of attack.
- tptacek 13y agoCoppersmith is complex enough that I'd be impressed to see working implementations even in lab conditions; it's one of the lattice reduction crypto attacks.
- pbsd 13y agoIt's really not that bad with today's computer algebra packages. Here's a simple example in Sage that works most of the time: p = random_prime(2**512) q = random_prime(2**512) n = p*q P.<x> = PolynomialRing(Zmod(n)) partialp = p - p % 2**224 # most significant 288 bits f = x + partialp root = f.small_roots(X=2**(224), beta=0.5, epsilon=2**-5) assert(root[0] + partialp == p) partialp = p % 2**288 # least significant 288 bits f = x + partialp * inverse_mod(2**288, n) root = f.small_roots(X=2**(224), beta=0.5, epsilon=2**-5) assert( root[0]*2**288 + partialp == p )
- tptacek 13y agoMan I hate you. What's especially funny is if you search for "SAGE" and "small_roots", you get this: http://trac.sagemath.org/ticket/2424 http://trac.sagemath.org/ticket/2424 Please at least tell me small_roots was hard to implement.
- vizzah 13y agoCool. Would this approach work if 'partialp' is not coming from the starting or ending bits of the targeted prime number?
- pbsd 13y agoYes. But it makes for a worse didactic example, since Sage does not support it out of the box. Imagine you know the middle 288 bits of p. Then you have to find small roots of the equation x1*2**400 + partialp + x0, which now has 2 variables x0, x1 < 2^112. This requires building a different lattice than in the single variable case, but the complexity of the attack is not very different. Also note that it doesn't have to be p or q. Bits of d, d mod (p-1), and d mod (q-1), and possibly other intermediate values of RSA computations are also usable for this.
- jchrisa 13y agoCan someone write it up in plain language? I grok that reducing entropy is bad, is this just one in a set of attacks for attacking entropy leaks?
- j_s 13y agoSo anyone stuck using vulnerable SSL is basically using telnet; ouch! Hopefully it will be a while before this implementation makes it to the various frameworks.
- filmgirlcw 13y agoAs much shit as CloudFlare took for the way they humble-bragged about finding out about Heartbleed early, I have to give kudos to the team for doing the right thing and reissuing and revoking all the certs, as well as detailing the exploit and offering up the test site. I'll admit, when the CloudFlare Challenge launched, the cynical part of me saw it as merely a way for the company to convince itself that it didn't need to reissue and revoke all of its certs (and I maintain that is exactly what it was), but the way the company handled things after the challenge was won is what really impressed me. They didn't have to be that transparent. Moreover, the test server has provided a way for Filippo Valsorda to improve his excellent Heartbleed checker, so well played.
- gatehouse 13y agoYeah, it sounds like the reason that they came off as glib and triumphant is because they didn't anticipate how serious the vulnerability would end up being.
- eastdakota 13y agoYes that and the fact that we didn't appreciate how few people had been notified. It was a big surprise to learn we were one of only a small handful of companies.
- j_s 13y agoThis explains a lot; hopefully you can add "confirm extent of pre-disclosure distribution" to the list of things to do when this type of info arrives in the future!
- tptacek 13y agoI'm a little irritated by the challenge not because of any delay in rotating certs, but because its whole premise was invalid; the challenge was accompanied by a long blog post that presumed a model for how RSA keys hit heap memory that appears to have been self-evidently false (again: I might be wrong here). Or, to put it differently: Akamai, which attempted to write a secure heap to contain memory corruption flaws in OpenSSL, conceded immediately Willem Pinckaers observation that they weren't protecting the CRT intermediates. Akamai seemed to believe ahead of time that keys weren't simply loaded once and held. There are lots of smart people working at Cloudflare, but I think the "challenge site" approach was a waste of time. The stuff John Graham-Cumming posted today, showing bitmap diagrams of the memory space of OpenSSL processes, would have been much more productive as a starting point. Diagrams like that are a standard part of reverse-engineering practice. They should have started there, and not with the challenge site. We all make mistakes (some of mine are easy to find). I'd just say that a lesson to learn from this is, "err on the side of greater severity, not lesser." (Is what I think.)