Superfish has been getting a lot of attention – the Forbes article is one of the better overviews.
Instead of jumping in and covering the details of Superfish, let’s look at how it might work in the real world.
Let’s say that you are looking for a watch and you visit Fred’s Fine Watches. Every time you want to look at a watch, someone grabs the key to the cabinet from Fred, uses a magic key creator to create a new key, opens the cabinet, grabs the watch from Fred, studies the watch, looks for “similar” watches, and jams advertising fliers for these other watches in your face – right in the middle of Fred’s Fine Watches! Even worse, they leave the key in the lock, raising the possibility that others could use it. Further, if you decide to buy a watch from Fred, they grab your credit card, read it, and then hand it to Fred.
After leaving Fred’s Fine Watches you visit your bank. You stop by your doctor’s office. You visit the DMV for a drivers license renewal. And, since this article is written in February, you visit your accountant about taxes. Someone now has all this information. They claim they aren’t doing anything with it, but there is no particular reason to trust them.
How does all this work? Superfish is a man-in-the-middle attack that destroys the protection offered by SSL (Secure Sockets Layer). It consists of three basic components: the Superfish adware program, a new SSL Root Certificate inserted into the Windows Certificate Store, and a Certificate Authority program that can issue new certificates.
SSL serves two purposes: encryption and authentication. SSL works by using a certificate that includes a public encryption key that is used to negotiate a unique encryption key for each session. This encryption key is then use to uniquely encrypt all traffic for that session. There are two types of SSL certificates: public and private.
Public certificates are signed. This means that they can be verified by your browser as having been created from another certificate – you have at least some assurance of where the certificate came from. That certificate can then be verified as having been created from another certificate. This can continue indefinitely until you reach the top of the certificate tree, where you have a master or root certificate. These root certificates can’t be directly verified and must be trusted.
Root certificates are connected to Internet domains. For example, Google has the google.com root certificate, and is the only one who can create a signed certificate for mail.google.com, maps.google.com, etc.
Bills Browser Certificates, Inc., can only create signed certificates for billsbrowsercertificates.com. The details are a bit more complex, but this is the general idea – signed certificates can be traced back to a root certificate. If the owner of that root certificate is cautious, you can have a reasonable level of trust that the certificate is what it claims to be.
Your browser or OS comes with a (relatively small) list of root certificates that are considered trusted. Considerable effort goes into managing these root certificates and ensuring that they are good. Creation of new signed certificates based on these root certificates is tightly controlled by whoever owns the root certificate.
Certificate signing is a rather advanced topic. Let’s summarize it by saying that the mathematics behind certificate signing is sound, that implementations may be strong or weak, and that there are ways of over-riding the implementations.
Private certificates are unsigned. They are the same as public certificates, work in exactly the same way, but can’t be verified like public certificates can. Private certificates are widely used and are a vital part of communications infrastructure.
According to reports, Lenovo added a new Superfish root certificate to the Microsoft Certificate Store on certain systems. This means that Superfish is trusted by the system. Since Superfish created this certificate, they had all the information that they needed to create new signed certificates. Which they did by including a certificate authority program which creates new certificates signed by the Superfish root certificate – on your system while you are browsing. These certificates are completely normal, and there is nothing unusual about them – except the way they were created.
Again, according to reports, Superfish hijacked web sessions. Marc Rogers shows an example where Superfish has created a new SSL certificate for Bank of America. The way it works is that Superfish uses this certificate to communicate with the browser and the user. The user sees an https connection to Bank of America, with no warnings – there is, in fact, a secure encrypted session in place. Unfortunately, this connection is to Superfish. Superfish then uses the real Bank of America SSL certificate to communicate with Bank of America. This is a perfectly normal session, and BOA has no idea that anything is going on.
To recap, the user enters their bank id and password to login to the BOA site. This information is encrypted – and sent to Superfish. Superfish decrypts the information and then re-encypts it to send to BOA using the real BOA SSL certificate. Going the other way, Superfish receives information from BOA, decrypts it, reads it, re-encrypts it with the Superfish BOA certificate, and sends it back to you.
Superfish apparently creates a new SSL certificate for each site you visit. The only reason that all this works is that they were able to add a new root certificate to the certificate store – without this master certificate in the trusted certificate store they would not be able to create new trusted certificates.
Superfish can also change the web page you receive – this is the real purpose of of Superfish. In normal operation Superfish will modify the web page coming back from the web site you are visiting by inserting new ads. Think about it – you have no idea of what the original web site sent, only what Superfish has decided to show you!
Superfish is sitting in the middle of all your web sessions. It reads everything you send, sends arbitrary information to an external server (necessary for the image analysis it claims to perform, but can be used for anything), forges encryption, and changes the results you get back.
The real threat of Superfish is that it contains multiple attack vectors and, by virtue of the root certificate, has been granted high privileges. Further, the private key Superfish is using for their root certificate has been discovered, meaning that other third parties can create new signed certificates using the Superfish root certificate. There is no way to do secure browsing on a system with Superfish installed. And no way to trust the results of any browsing you do, secure or not.
Pingback: Links 21/2/2015: GNOME 3.15.90, Google Wins Android Lawsuit | Techrights
Pingback: Slashdot comments on Superfish from Lenovo – Blog of andreyv
It should be noted that there are “certificate pinning” techniques whereby websites can identify which certificates should be present on their websites. This is done in addition to the standard certificate checks.
Operating Systems should *not* assume that all trusted certificates are allowed to authorize any server. In fact, it is extremely common for a number of certificates to be trusted, and the OS should assume that there each one us for a different purpose. For example, my internal corporate PKI should not be authenticating my bank.
FWIW – RFC 6698 (DANE) allows you to place your certificate SHA256 has into your DNS record (having already, of course, secured your DNS entries with DNSSEC), and would have gone a long way to mitigate this and various other “whoops – I have more than one trust anchor installed, which one should I use for this connection” vulnerabilities.
Is there a strategy to implement RFC 6698 or some other form of pinning technology at RedHat?
As I understand it, certificate pinning is based on known certificates. If you start with a compromised system with Superfish installed you already have a bogus root certificate and the first time you see each new certificate it is the Superfish generated certificate. In this case it isn’t clear certificate pinning would solve the problem.
Using dnssec is a great idea – we should really be moving toward using dnssec by default as it avoids a range of dns issues and really helps with man-in-the-middle attacks. Unfortunately, dnssec breaks a number of applications and web sites. This makes it difficult for OS vendors to enable it by default.
DANE is a good idea but requires large numbers of people to add information to their dns records. Perhaps it can be done for critical sites, like banking, email, and Google. The challenge there would be how to present the information to users in a non-confusing way.
Pingback: Lessons from the Lenovo Malware Incident | {workplace_tech_law}