Authentication for Web Applications

Authentication is a common requirement for web applications, we want to know who is making requests and so that we can manage multi-stage transactions and protect confidential or private information. The HTTP protocol includes some provision for authentication but in modern applications this is not often used; it is more common to use application level authentication combined with HTTP cookies to identify users. More recently, we've seen the rise of federated identity management with services like Facebook and Google offering to manage user identities for web appliciations. This chapter will review the different methods that are available for authentication in web applications.

HTTP Authentication

The HTTP protocol provides two mechanisms for authentication: Basic and Digest authentication. To trigger authentication, the server returns a response code 401 Unauthorized along with a WWW-Authenticate header that provides details of the kind of authentication that is required. The two options here are to ask for Basic or Digest authentication. The browser will respond in either case by prompting the user for a username and password which are then encoded in the header and returned to the server in a new request for the same resource.

Basic Authentication

Basic authentication is the simplest and least secure of the two methods. The username and password are encoded with the Base64 algorithm and sent back in an Authorization header with the new request. Base64 makes the username and password unreadable but it can be trivially decoded, so the password is really being sent un-encrypted over the HTTP connection. Anyone able to listen in on the request (eg. a proxy server between the client and server) will be able to see the username and password and could make use of it themselves. For this reason, Basic authentication should really only be used over a secure (HTTPS) connection. In practice it is often used to provide a low level of protection for resources because it is easy to configure to password protect a directory of files in a web server like Apache.

Here's an example HTTP conversation:

GET / HTTP/1.1
Host: example.org

The server responds with a 401 code and requests authentication, the realm is used to indicate to the user what they are providing a password for.

HTTP/1.1 401 Authorization Required
WWW-Authenticate: Basic realm="Secure Area"
Content-Type: text/html
Content-Length: 311

The user is prompted for a username and password and the browser then re-sends the original request with a new Authorization header containing the Base64 encoded username and password:

GET / HTTP/1.1
Host: example.org
Authorization: Basic c3RldmU6c2VjcmV0

The server will now respond with the resource as with a normal HTTP request. Note that from now on (until the browser session closes) the Authorization header will be sent back to the server with every request, so the user only needs to provide a username and password once. There is no provision for logging out using this mechanism.

Digest Authentication

Basic authentication is not secure unless we are able to use an encrypted (HTTPS) channel to send HTTP requests. HTTP provides a more secure authentication mechanism that is safe to use over plain HTTP channels since it never sends the plaintext password. This is Digest Authentication.

This form of authentication is initiated in the same way (with a 401 response) but this time, the browser is asked to use digest authentication and is given a nonce value - a unique string that will be used as part of the conversation with the server. Instead of sending back the username and password, the client browser generates a one-way hash (using the md5 algorithm) based on the username, password, nonce value and the request being sent. This is sent as part of the next request along with the username. Since the browser has the same information as the client, it can compute the same one-way hash and compare the value it receives. If they are the same, then the user checks out and the response can be returned.

The details of the algorithm used to generate the hash that is returned is detailed in the Wikipedia page on Digest Authentication. The take-home message is that the hash value encodes information known only to the client (username, password, request URL) and server (nonce) and that this information is sent in encrypted form. Because it contains information about the requested URL, the value can't be used by an attacker in a different request.

Like Basic Authentication, Digest does not provide a mechanism for logout so the session lasts until the browser is closed. The browser uses the same nonce value for every request - the nonce is only sent in the first 401 response.

Implementation

The most common implementation of both Basic and Digest authentication is that is managed by the server. The server is configured to require a password for a given resource or collection of resources and intercepts any requests, requiring authentication before allowing access. The server takes charge of managing usernames and passwords on behalf of the application. While this is the most common approach, it is of course possible to manage HTTP based authentication within an application; there are a number of WSGI middleware packages that will provide this implementation.

Application Based Authentication

The most common approach to authentication in a web application is to manage it at the application level. This means that the application manages usernames and passwords in a database and decides when it needs to prompt for a login.

In this approach to authentication, the application must take care of managing all information about users, checking their passwords and maintaining session information. The main problem to solve is to maintain the session state between different HTTP requests. Since HTTP is a stateless protocol, every request is independent of the last and no identifying information is included in the request by default. The methods described above provide one mechanism for sending authentication information with each request, but as mentioned above, they don't provide a mechanism for logout and are generally managed by the server rather than the application. The method we describe here makes use of cookies to maintain state between requests.

It is common for a web application to present a login form that asks for a username and password from the user. This form is submitted to the application as a POST request to a URL like /login where user authentication is handled. To respond to this request, the application must first check that the username and password correspond to the stored details of a real user.

To verify the password of a user it should be a simple matter of comparing the password that is sent in the POST form submission with the password that is stored in the database for the username sent in the form. The naive implementation of this method would just store the password in the database; however, this is a significant security vulnerability. If an attacker were able to gain access to the database of the application, they would have easy access to all passwords for all users. Since many people use the same password on multiple accounts, the attacker might now be able to access Facebook or even the bank account of a user. Foro this reason, it is best practice not to store the raw password of a user but to store a one-way hash derived from the password. A one-way hash is a transformation of the input string that generates a seemingly random string of characters; the algorithm used to do this ensures that while it is easy to compute the hash from the password, going the other way is close to impossible. Instead of comparing the password with the stored password, we compare the hash of the password with the stored hash. Should an attacker gain access to the database, they will only have a list of hashes, not the raw passwords.

If an attacker gets access to the list of password hashes they can try to discover the original passwords using a brute force attack if they know the hash function that was used. Using a list of common words, they generate the hash value and compare to each database entry. If a user has used a common word (or combination) as their password then it will be exposed. Since compute power is now readily available, attackers can try not only common words but variations - pairs of words, words with numbers added to the end etc. In order to make this harder, modern systems use deliberately slow password hashing algorithms and run them a number of times over the input string. The idea is to make sure that any brute force attack takes a very long time to run. In Python you can use the hashlib module which has an implementation of the PKCS#5 password-based key derivation function 2. At the time of writing this is the recommended method of generating a secure password hash.

Having compared the submitted and stored passwords, the application is now sure that the login request is good and the user should be authenticated. In order to keep track of future requests from the same user, the application will return a cookie that will be used to identify this user. The cookie value needs to be hard to guess and assocate with a user - if we just used the username, anyone could pretend to be that user. A common option is to use a UUID (Universally Unique Identifier) which is a random string of characters; this is known as the session key. In order to keep track of who this session key was sent to, the application makes an entry in a session table as described in the earlier chapter. Once the entry is recorded in the session table, the cookie is returned to the user in the response.

Each time the user now re-visits the site, the cookie can be used to look up the session key in the database and find out the username of the authenticated user. The application can control how long a login session is valid for by setting the expiration date on the cookie or by removing the session table entry after a given amount of time. In this way a highly sensitive site like a bank, can force login to last only a few minutes while a less sensitive site (Facebook) can allow login to last for months.

Logout is supported in this scenario by removing the entry from the sesion table. Once the entry is removed, even if a valid cookie is received, the session key will not be found in the database and so the user will not be identified.

Third Party Authentication

A recent development in web applications has been the provision of authentication services by third parties. The idea here is that a new web application would not manage its own user credentials, instead it would defer to another web application which will check the username and password and tell the client application about the user. There are advantages for both the web application developer and the user in this arrangement. The developer doesn't need to be concerned about managing user details, in particular they don't need to worry about the security of passwords since they don't store them. The user doesn't need to remember yet another password, instead they have one central identity that they can manage more easily.

There are two main standards for third party authentication. OpenID is a widely used standard that allows for a range of identity providers. Shibboleth is another standard that is mainly used in academic networks and concentrates more on a trust network between identity providers.

OpenID

With OpenID, the identity of a user is a web address (URL) that can be used by a web application to validate the user's identity. The three parties involved in the transaction are the user, the OpenID provider and the relying party (the web application that will rely on the result of the authorisation). The sequence of operations is (roughly) as follows:

  • The user provides a URL to the relying party as their identity, eg. http://openid.example.org/johnsmith
  • The relying party retrieves this URL to discover the OpenID providers URL. This will either be in a <link> tag in the HTML page or in a special XML document, depending on the version of the standard being used.
  • The relying party makes a request to the OpenID provider and is given a shared secret that can be used later as part of the authentication process. This allows the relying party to know that later messages really do come from the OpenID provider.
  • The relying party returns an HTTP redirect response to the user's browser so that they are redirected to the OpenID provider. The redirect includes some information about the relying party.
  • The user is authenticated by the OpenID provider, usually this means they enter their password but there could be other methods. The OpenID provider then asks the user if they trust the relying party and want their details sent back to it.
  • If the user agrees, they are redirected back to the relying party, the redirect includes information about the user and the shared secret that was established earlier so that the relying party knows that this is genuine.
  • The relying party accepts the authentication of the user and allows them to view restricted content.

There are further options and complexities that I've glossed over here, some more details are given in the Wikipedia OpenID page and in the OpenID 2.0 Specification.

The design of this protocol means that the end user is able to manage a single identity and gets a more uniform experience when authenticating with different web applications. The web application developer gets the benefit of a large number of potential users who already have OpenID identities and doesn't need to manage their credentials. OpenID is implemented in a number of open source libraries for most of the common web development languages and toolkits. The Python OpenID module provides support for running an OpenID identity provider and for consuming OpenIDs as a relying party.

While you may not realise it, you probably already have an OpenID identity by virtue of one of the major web applications that you have an account on. The Get an OpenID page lists the major providers such as Google and Yahoo that provide an OpenID endpoint. You can also establish an OpenID identity at a dedicated provider such as MyOpenID or if you have access to a server, run your own OpenID provider using something like the Python OpenID module.

You might ask whether OpenID is in any way secure if anyone can set up an OpenID provider and use it to log in to your site. The thing to realise is that what you are getting with an OpenID identity is an assurance only that you are talking to the person who owns that identity. If that identity asserts that I am Britney Spears, you have no way of knowing whether I'm the real Britney or not, in fact you should assume that I'm not. All that OpenID does is remove the need for me to store passwords and allow the user to have the same identity on different websites.

There are a number of potential and actual security issues with OpenID which come from the way that messages are exchanged between the relying party and the identity provider. If these messages can be intercepted there is a danger of someone assuming the identity of the user and doing things on their behalf. New features have been added to version 2.0 of the standard to counter some of these problems.

OpenID is a good way to provide an identity service for a web application where all you need to know is the identity of the user, rather than any authoritative information about them. The only thing you can trust from OpenID is that this person is the same person you saw last time with this identity. For many applications, that is enough and so OpenID has distinct advantages. When you want to be able to trust the identity more, or when you want to know more about the person, other standards provide a better choice.

OAuth - access to third party services

http://softwareas.com/oauth-openid-youre-barking-up-the-wrong-tree-if-you-think-theyre-the-same-thing

https://developers.facebook.com/docs/guides/web/

https://developers.google.com/accounts/ OAuth - access to third party services on behalf of a user (as well as authentication)