I had a job interview today, and one of the "challenges" with which I was presented was this: "We own several sites. We would like our user to be able to log into the central site as a subscriber, and then all the other sites will know what permissions that user has."

The sites are cooperative, so getting content onto each one to support this idea isn't difficult. Also, all of the sites belong to the same user, so getting a regular framework that you can deploy to all of them isn't difficult either.

The scenario we want to support is this. We have two websites, the remote authentication server framework. Let's call it "rasf.com" (which, sadly, does not go to rec.arts.sf.written), and let's call the site that wants authentication "brandx.com." (Sadly, this is a parked site that leads to a stock photography reseller.) The idea is that you're a subscriber to rasf.com. You visit brandx.com, and Brand-X initially has no idea who you are. We want Brand-X to be able to say, "Hey, Rasf, who is this guy?" and have Rasf come back "Oh, he's John Smith, he's a legit user, and here are some details."

The trick involves public key cryptography.  Both Rasf and Brand-X have public and private keys.  To understand the scenario, you must appreciate that much of the heavy lifting is done away from both sites on the browser.  The problem is that, on the browser, any windows opened between Rasf and Brand-X cannot communicate with each other; this is known as the sandbox, and it keeps malicious sites from using iframes or Ajax to either inject malware Javascript into the victim page or allow the nefarious iframe from ripping data (like username and password keystrokes) from the victim page.  We want to violate the sandbox, but how?

Assumption: You've visited Rasf recently, and have a cookie from Rasf saying "Yes, I, Joe Smith, and a valid user of Rasf and affiliated sites!"

You access a page from Brand-X.    Secretly, Brand-X creates a one-pixel-wide iframe and set the iframe's src attribute to the "authenticate this user" page on Rasf, including in the request its public key as an argument.   After the iframe loads, the loaded page from Rasf now has the session information on the browser, and it has Brand-X's public key.  The session information includes the Rasf cookie.    So now, Rasf knows two things: it knows who you are, and it knows that Brand-X wants to know who you are.

How does that who you are information get back to Brand-X?  Here's where the cooperation comes from.  The iframe from Rasf, using the onload() event, creates yet another iframe, this time calling back to a specified page (the cross-domain receiver page) on Brand-X's site, and that URL is loaded with your user ID, a cross-domain session key, and other information, all signed with the Rasf private key (so the Brand-X site can unpack with Rasf's known public key).

Now, because both the containing page and the iframe two layers in are in the same domain, they're in the same sandbox, and can communicate with one another via javascript.  The innermost iframe communicates with the cookies of the outermost page, the calls reload: all this information has now been pushed back to the Brand-X server, which can now use the signed cross-domain session key to make back-end requests of Rasf's web services API and say, "Okay, now that I know he's user 12345, and I have a session key validating this conversation, what else can you tell me about him?"

There are lots of other details here.  What if he's not logged in to Rasf at all?  Well, enlarge the 1-pixel iframe to a size big enough to show a log-in within the Rasf domain, get his username and password, authenticate and proceed as before.

This is a generic description of what Facebook Connect does, and it's how you can do it as well.