62

Robert Graham detailed on the Errata Security blog how he was able to get the private key of the Superfish certificate. I understand that attackers can now use this key to generate certificates of their own which will be signed by the Superfish CA.

Won’t the same attack work on other root certificates already on a computer? Why was the private key on the computer in the first place?

TRiG
  • 610
  • 5
  • 14
bobby
  • 887
  • 7
  • 14
  • I guess an underlying question is why is the private key present on the client at all? Does encryption not happen server-side? – Calimo Feb 20 '15 at 08:32
  • 6
    @Calimo It is there because it is needed to the proxy can generate certificate on the fly to perform it's MITM function and inspect traffic without any warning to the user. – Stephane Feb 20 '15 at 08:48
  • As a side note, the source for the SuperFish software is Komodia, a company formed by a former Israeli IDF Intelligence Core programmer. Who at the moment is claiming they are under attack per Forbes Magazine. – Fiasco Labs Feb 20 '15 at 16:30
  • 2
    @FiascoLabs It should be noted that pretty much every Israeli serves in the IDF, and a good programmer would frequently find themselves in a high-tech IDF position rather than on the ground in Gaza. – ceejayoz Feb 20 '15 at 16:46
  • 1
    @ceejayoz - In a country that small with as many threats as they have to face, they don't have the luxury of running a volunteer army. BTW, the password to access the private cert is `komodia`. Expect exploits soon. Anything that uses Komodia software is vulnerable to the attack (Qustodio, Keep my Child Secure, etc.) – Fiasco Labs Feb 20 '15 at 16:57
  • 1
    For a hands on experiment, try playing with [mitmproxy](https://mitmproxy.org/). It's a proxy server that makes it easy to man-in-the-middle your own TLS connections. Like Superfish, it generates certificates on the fly to match the site you're looking at, and it signs each cert with a key that is stored local to the proxy, e.g. ~/.mitmproxy/mitmproxy-ca.pem. If you trust MITM proxy's CA certificate, then all of the on-the-fly certificates that it signs will also be trusted. – Mark E. Haase Feb 21 '15 at 01:33
  • 3
    Because they are idiots, mostly. Sane programs that _need_ to perform a MITM for legitimate purposes (e.g. `mitmproxy`) will generate a key pair which is unique to the host. – Stefano Sanfilippo Feb 21 '15 at 10:09
  • Adding to what mehasse and Stefano said, legitimate programs that do this will also [warn you about the security risks beforehand](http://docs.telerik.com/fiddler/images/ModelDoc/WarningDialogTrustFiddlerRootCert.png). – Brendan Long Feb 22 '15 at 18:46

4 Answers4

60

Unless the Superfish malware has been installed on your system, (which it might if you bought a Lenovo machine,) you don't have to worry. This attack worked because the secret it revealed was necessary for the malware to hijack the data; it is not a part of how legitimate certificates are authenticated.

It helps to understand the relationships between a certificate, a public key, and a private key. A private key is a secret number used to sign messages with digital signatures (or to encrypt web traffic), and it has a matching public key that can be used to verify those signatures. A certificate is a public document that contains a public key; web site owners put their public keys onto certificates and send them to companies called Certificate Authorities (CAs) who digitally sign them to prove their certificates are genuine. The digital signatures ensure the document has not been changed, assuring you the public key it contains is the genuine key of the site you're visiting.

CAs are companies everyone agrees to trust to only sign certificates from legitimate sources. They also have a private and public key pair. They keep the private key very secret, locked in a secure cryptographic device called a Hardware Security Module (HSM) and they restrict access to it so it can only be used to sign a customer's certificate when the customer generates a new key. But in order to be useful, everyone on the web needs to know their public keys. So these CAs put their own public key on a special certificate and sign it with their own private key ("self-signing"). They then send these self-signed "root certificates" to the browser vendors and OS vendors, who include them with their products. A real CA would never, ever, send out their private keys!

The trusted authority root certificates are the documents that validate all the certificates of all the connections your computer makes. Thus, your computer has to trust them. This malware is installing an untrustworthy certificate in a position of ultimate trust, compromising the security of the machine by allowing anyone who knows this key to forge a certificate for any site, hiding evidence of their tampering.

The malware abuses its position by generating phony public and private keys for every site you visit; after you connect it injects its payload into the web site's page. In order for your browser to trust these phony keys and not give you warnings, the malware generates a forged certificate that tricks your browser into believing the keys are legitimate. But like any certificate, the forgery needs to be signed by a trusted CA. To sign, the malware needs a public and private key, just like a real CA. Because the phony CA is forging these certificates right inside your computer, the private key needs to be inside your computer as well. It's impossible to keep such things secret from the owner of the computer, but they tried by taking some rudimentary steps to hide it. The blog you linked to described how he uncovered the secret.

No legitimate certificate authority would ever allow their private key to be leaked, much less send it out to a bunch of random computer owners. There was a case where a certificate authority had their secret key leaked; their reputation was ruined and they went bankrupt in a month. Since your computer doesn't contain the private keys of the legitimate certificate authorities, there is no secret for an attacker to crack.

John Deters
  • 33,897
  • 3
  • 58
  • 112
  • 7
    It should be *impossible* for a legitimate CA's private keys to be leaked. They're stored in [HSMs](https://en.wikipedia.org/wiki/Hardware_security_module) and the roots are offline. When CAs are compromised, usually some of their servers are broken into and convinced to issue fraudulent certificates; the private keys themselves aren't exposed. – Matt Nordhoff Feb 20 '15 at 11:19
  • 7
    @MattNordhoff, yes, it certainly should be impossible, however, DigiNotar is the poster child example of a company that did not take such care, and it was leaked, and every browser maker in the world and every OS in the world pulled the DigiNotar certificates from their systems immediately; DigiNotar collapsed within the month. See http://en.wikipedia.org/wiki/DigiNotar for more on the story. – John Deters Feb 20 '15 at 15:36
  • Has the certificate involved been revoked yet? – Loren Pechtel Feb 22 '15 at 04:41
  • 1
    @LorenPechtel, CRLs are typically hosted by the root CA, and signed by the root CA that issued it. The Superfish certificate is a self signed root certificate, and is itself the trusted issuer. It has no host OU in the issuer field. It has no CRL distribution point OID. It doesn't give your browser enough of the information it needs to even find out if it should be revoked. I think a viable defense will be for browser makers to support some kind of blacklist of bad root CAs, and to eradicate them through normal application updates, but that's a really slow process. – John Deters Feb 22 '15 at 06:39
  • "Unless you have a Lenovo with the Superfish malware, you don't have to worry." I came to know about this only because I randomly check HN. On the other hand there are atleast thousands of users who have no idea that their security is somewhat compromised as they haven't seen that news ( including my parents , who have no idea what I was saying or doing when I was removing that certificate via Lenovo removal tool). – Abhinav Gauniyal Feb 22 '15 at 07:42
  • They also need the private key on the computer to actually perform the MITM, even if you had some super wildcard `*.*` public key certificate on the host you would still need to have the private key available to complete the handshake. – jhoyla Feb 22 '15 at 14:45
  • There are other common local MITMs besides superfish. E.g. for anti-virus scanning. This should be able to be done reasonably securely by generating a unique key pair for this purpose for each system where the software is installed, which never leaves that system. – mc0e Feb 22 '15 at 16:38
31

Because of how Superfish works, the certificate and its private key must be easily extractable.

Superfish creates signed certificates on the fly for the network connections it intercepts, without contacting a central server. In order to do this, the private key for the Superfish CA must be embedded in the software in an easy-to-use form. Now, they didn't need to make it a plaintext .pem file, but even the best security wouldn't have slowed down a serious reverse-engineering effort by more than a day or two.

You can't share a secret with the whole world and not expect anyone to find it. Real certificate authorities do a much better job of protecting their private key, by ensuring that it never leaves their certificate-signing computer.

Mark
  • 34,513
  • 9
  • 86
  • 135
  • 1
    Avast does something similar, but I think they use a different key for every installation – bobby Feb 20 '15 at 05:18
  • 1
    Certainly the _certificate_ must be easily extractable. However, does the _private key_ need to be easily extractable? –  Feb 20 '15 at 05:21
  • 8
    @RickyDemer, yes. You can't sign newly-generated site certificates without the CA private key. – Mark Feb 20 '15 at 05:24
  • 1
    You may want to put that into your answer's initial sentence. –  Feb 20 '15 at 05:25
  • 21
    The main issue with private key extraction is the use of the same root certificate on each install. It should be autogenerated (Fiddler does it) so the private key is specific to your computer. – Guillaume Feb 20 '15 at 08:36
  • "Real certificate authorities [ensure] that it never leaves their certificate-signing computer." Superfish also does this. It's just that they have a lot more certificate-signing computers. ;-) – David Richerby Feb 22 '15 at 14:32
  • @David: well, Superfish *tried* to do that. It failed in the sense that people who've never installed Superfish, and whose computers therefore aren't certificate-signing for Superfish, now have the private key. Of course, their next action might be to become a cert-signing computer, either for proof of concept or for abuse ;-) – Steve Jessop Feb 23 '15 at 13:59
2

If you follow best practices when it comes to protecting private keys of root certificates it most certainly would not have been that easy.

If I understand Robert's blog correctly the password that protected the private key was embedded as a string in the adware binaries that shipped with the laptops. This is like shipping your safe together with your key obscurely taped to the bottom of the safe. Not a good idea! The password was also not very strong, meaning that even if it wasn't embedded in the adware for anyone to see, a determined person with knowledge and resources could be able to crack it in a reasonable amount of time.

The password to extract the private key was most likely embedded in the adware to enable the adware to decrypt traffic signed by the Superfish root cert on the fly to inject the advertising markup that the user would finally see on the web page.

This would never be the case with responsible Certificate Authorities and I'm fairly sure a CA needs to comply to certain standards, or even laws, that would prohibit them from doing such a thing.

If it was that easy to get the private keys of all root certificates PKI would have served no purpose and protect absolutely noting.

ilikebeets
  • 2,746
  • 16
  • 22
  • Embedded as a string, hmm, so doing a strings search on the binaries would show it? I think the key was taped to the front of the safe in plain sight then. – Fiasco Labs Feb 20 '15 at 06:19
  • Indeed, although in "their" defense, the binaries were obfuscated (fairly poorly) by packing them... – ilikebeets Feb 20 '15 at 06:28
  • @FiascoLabs, yes, exactly. Reverse engineers unpacked the binary, ran *strings* on it, and used that output as the dictionary to attack the private key's password. – Foo Bar Feb 20 '15 at 12:42
  • "Best practices" of protecting CA root private keys includes locking them in an HSM and requiring three-part smart card secret sharing schemes to access or alter them. This was a just a secret key baked into the malware in order to forge CA signatures on the fly; malware certainly couldn't install a hardware module to protect the secret bits! – John Deters Feb 20 '15 at 19:44
  • @JohnDeters: The root private keys are protected that well, perhaps, but the private keys for intermediate certificates, which are actually used for signing new SSL certificates, are only as secure as the CA server software, and due to trust delegation, are just about as powerful as the root keys. – Ben Voigt Feb 20 '15 at 20:50
  • @BenVoigt, security conscious CAs often don't have the signing key on the server. They delegate the task of signing to a dedicated HSM. The root key may be offline most of the time, but the key never leaves the secure tamper-resistant cryptographic environment. For example, IBM mainframes can manage certificate signing requests utilizing the Cryptographic Coprocessor for secure operations. – John Deters Feb 22 '15 at 06:52
  • @John yeah something along the lines of: root private key gets created as a set of (N-2) required smart cards, used to sign some intermediate certificates, and then powered off with the smart cards separated and all locked away. Then the intermediate, which has certificate signing purpose, but not CA signing purpose, is loaded in another HSM which is kept online and used for day to day operations. If anyone compromises the server they can create some SSL or code signing certs, but they can't steal the private key and can't sign a new intermediate cert, so only for as long as their access lasts – Ben Voigt Feb 22 '15 at 07:04
2

Superfish acts as a Man In The Middle. It dynamically generates keys which your browser trusts for domains you visit. To generate those keys it needs the private certificate, and while more obfuscation is possible, in the end it can't conceal that key.

What could be done though is to generate a unique key pair for the Certificate pair on each computer where the software is installed. This would mean that extracting the key used for the Root CA installed on one computer would not provide access to the communications of another computer with the software installed. This is not exactly rocket science, and it's inexcusable that Komodia don't do it.

I'd like to know whether other companies doing this sort of MITM are issuing unique keys. E.g. locally I've found certificates for "Skype Click to Call" and "avast! WEb/Mail Shield", which presumably function in a similar way. I've verified that if I visit https://www.google.com/ my browser sees a certificate signed by the avast key.

mc0e
  • 491
  • 2
  • 14
  • Actually Avast isn't as bad as it both uses unique keys and verifies the validity of the server cert from the real HTTPS connection: http://security.stackexchange.com/questions/82285/are-the-certificates-from-skype-click-to-call-and-avast-web-mail-shield-any – huyz Feb 23 '15 at 06:54
  • Any info on Skype Click to Call? – mc0e Feb 23 '15 at 19:49