Prevent account enumeration on login, reset password and registration pages

Account enumeration is a potential security risk whereby a web site gives out information about what accounts are already in the system. This may a) leave them susceptible to a brute force-esque attack and b) may violate their users privacy which may be very important for certain types of sites. There are a couple of prominent Irish websites covered below which I’m familiar with and I just happened to notice they allow account enumeration but it is important to note that the vast majority of ecommerce sites online do this too. Certainly there are greater security risks out there such as SQL Injection, badly stored passwords, not using https when you should etc. but when this exploit is found alongside other exploits… well let’s just say that’s when bad things can happen.

keep login failed messages generic

I think most developers have the possibility of account enumeration in their minds when considering the messages returned from a login page for failed logins. When logging into most websites with a non existing username you will commonly see ‘Invalid username/password combination‘ type messages rather than ‘That username does not exist‘ type messages. An attacker now cannot just keep trying usernames on the login page until they find one that does exist, at which point they may (if possible) launch a dictionary password attack against that specific account. A few examples of Irish sites doing a good job here include easons.com and water.ie:

Generic Login messages

waterie-login

Another prominent Irish site; ticketmaster.ie does present different messages to the user if the account exists or not on their login page. The following image shows the login for non existing accounts:

ticketmaster-non-existing-account

It all looks good right? No mention that the specific email address does not exist, just that the address and password as a whole did not match any accounts. If however, you know of one account that does exist (the one the hacker just created for himself) and try to login with that you’ll notice that the message is different:

ticketmaster-existing-account

Ticketmaster do appear however to have an account lock out facility which locks accounts after 5 failed accounts. What this means is that although an attacker can find out if an account exists due to the slightly different error messages returned (which would be really bad for a very private site.. think ashleymadison.com) they will only get five password tries thus a full on brute force attack against a specific account with a password dictionary is not feasible.

Although locking accounts help mitigate against brute force attacks there are a number of potential problems with this approach. It can for example leave ticketmaster.ie open to an account lockout attack, whereby the attacker can deliberately lock out a large number of accounts in a denial of service-esque attack (this is possible as ticketmaster sends a reset password rather than URL to reset a password). It also means an attacker could build a large list of valid accounts and try the same password on all of them (ticketmaster1234 perhaps) which could return a few matches. Ideally ticketmaster would not reveal if an account existed or not. 

preventing account enumeration on the forgot password page?

The same idea should of course apply to the forgot/reset password pages. Unfortunately what is very common is that although a website’s login fail message is generic, the websites goes and undoes all this good work on their forgot/reset password page. Remember easons.com who displayed ‘If you have an account but can’t remember your password, click on “Password Problems?” below‘ regardless if the account existed or not? Well here’s what happens on their reset password page:

easons-forgotpassword1 easons-forgotpassword2

Remember water.ie who upon failed login just presented ‘Please enter a valid username and password’ to the user? The first image below is what the forgot your password screen shows when the account doesn’t exist, while on the right, we see what the screen shows for an account that does exist.

water-forgot-nonexistingwater-forgot-existing

 

 

 

 

 

 

So previously undiscoverable accounts are now discoverable on both easons.com and water.ie via their reset password pages. This is probably not a huge deal from a privacy perspective (does your neighbour like books? / has your neighbour paid his water charges?) for these particular sites. If however these sites don’t implement some kind of brute force mitigation individual accounts could be vulnerable. Ticketmaster.ie, as noted does appear to have brute force protection in place, this is just as well because their password reset page gives up existing accounts too:

ticketmaster-forgot
What should happen in reset password scenarios is that the website should display the same generic message regardless if the username/email address exists. A message such as ‘Further instructions have been sent to your email address‘ or similar is often used in both cases. If the email address does exist the email can contain a password reset link (not a reset password in email – search for ‘Sending a reset password versus sending a reset URL’) which the email account holder can click on if it was them that asked for the reset or ignore if it was not. If the email account doesn’t exist in the user DB, just don’t send any email.

…but what about the registration page? I have to tell a potential new user that their chosen username/email is already taken on screen!

Not necessarily. Preventing account enumeration is most problematic on the registration page that’s for sure, at least from a usability perspective. The goal here is the same as with the login and password reset pages and that is to keep the outward (what’s on the screen) process the same for usernames/emails that already exist and those that don’t.

If the site uses the email as the username, it should be possible to do this pretty nicely and in a manner similar to before. It might require some re-jigging of your registration pages, but the idea would be to start the process by requiring the user to enter their email address. When they submit this notify them that they have been sent an email with further instructions regardless if the email is in the DB or not. If the email address is already in the system the email will advise them that they have already registered and perhaps provide a link to the password reset page. If the email address is unused the email will contain a time limited URL link to verify email and continue the registration process. There is a very small usability hit on this approach as the notification about email addresses already existing now happens via email rather than on screen, but this will affect only a tiny percentage of registrations.

Both easons.com and ticketmaster.ie use the users email as their username so could use the above approach. Their current registration processes let us all know if accounts already exist or not:

easons-createaccount ticketmaster-createaccont

Note – water.ie is not really applicable in this case as their registration process begins with the entering of a application ID and associated pin which has been sent out in the post.

What if a website uses arbitrarily chosen usernames rather than email addresses to login? Well, for sure preventing account enumeration in this case is a bigger problem than above. In this instance, the first step of the process could capture both of these and in the follow up email notify if the email entered is already taken and advise of the username already associated with the email. Not a major problem so far… but what if the email is available but the username is already being used by a different user.

On popular sites, millions of usernames could already be taken, so it would be impractical to send a ‘username already taken’ email everytime someone tries to use an existing username as it could take them god knows how tries to find a free one. In this case, the best balance between security and usability may be to notify the user trying to register that their desired username is already taken on the registration page itself rather than via email but also to reCAPTCHA enable the page so scripts won’t work. This means the task of trying to find if a username already exists requires manual intervention by the attacker which significantly reduces their ability to build lists of existing accounts. Of course reCAPTCHA from Google and similar plugins by others are not unbreakable so there is no 100% perfect solution here. Your specific implementation is likely to be yet another trade off between security and usability.

suggestions to prevent account enumeration:

  1. Use email addresses as usernames as this makes accounts easier to protect on password reset and registration pages.
  2. Show generic ‘Invalid username/password combination’ message on failed login.
  3. Show generic ‘Further instructions have been sent to your email address’ message on password reset. If email is already in DB, send a time limited reset password URL
  4. First registration step is user enters email address and submits, then the site shows generic ‘Further instructions have been sent to your email address’ message. If email is already in DB, email says already registered etc. If email not in DB, email should contain a continue registration URL.

related Links

Troy Hunt’s (my favourite microsoft MVP by a mile) amazing article ‘Everything you ever wanted to know about building a secure password reset feature‘ from 2012 which was how I was originally introduced to many of the ideas above.

Troy Hunt’s Secure Account Management Fundamentals PluralSight course. Another amazing resource from Troy.

OWASP page on preventing brute force attacks. Note in particular the section on account lockouts and the potential exploits that remain even when doing this after X bad attempts. The surface space of many of these exploits can be reduced by preventing account enumeration on login, reset and registration pages as per above.

please can I have your comments and/or questions?

I’m very interested to know your thoughts on the whole topic of account enumeration so please feel free to leave a comment. If you’re a developer or IT professional and would like to connect with me on LinkedIn please check out my profile and send me a connection request.

Google recaptcha firewall exception options

Google reCAPTCHA firewall options articleWe’ve implemented Google recaptcha in one of our web apps which is nice and works fine on our local machines. When we deploy to our dev server however we were hit by outbound firewall rules so .Net threw an exception along the lines of ‘Unable to connect to the remote server

Google reCAPTCHA firewall options

Upon investigating what firewall exceptions we would need for our various environments, we noted that Google recommends opening up the relevant port (80 if http is fine, 443 if the web app will run with a secure certificate/https) to outbound connections. This is highly insecure and most likely unacceptable to most network administrators. That option was therefore not a runner for us.

The best practice and more secure approach would be to create rules which are restrictive as possible using only specific IP addresses. We looked into this, however from researching online it seems that Google’s IP addresses change regularly which obviously creates a problem for maintainability and for the reliability of the application. One comment from April 2015 on the ‘How to access reCAPTCHA servers through a firewall‘ recaptcha wiki page in particular worried us:

Hi guys, we are using the Recaptcha solution for almost half a year now, and in this time we had to change the firewalls four or five time already, just to find out today that they have changed again..

We also noted similar comments such as the below:

Yesterday it worked fine once we configured the firewall to allow requests for the IP address that was causing the problem.  But today it’s requesting using a DIFFERENT IP address, which isn’t configured.  Not only is this a problem for us right now, but it makes me uneasy.  What happens if it changes IP addresses again?  More broadly, what is the issue, here?  Why did it use one IP yesterday and another today?

Additionally the ‘How to set up reCAPTCHA‘ wiki page also mentioned that IP addresses can change. One way to mitigate the risk of IP addresses changing might be to catch the exception that is thrown when the app attempts to communicate with http(s)://www.google.com/recaptcha/api when the firewall is blocking such communication. The catch block could then perhaps bypass reCATPCHA, considering the request to come from a human and also email IT support who can then verify if in fact IP addresses have changed. Of course this still isn’t a runner for most enterprise applications.

In the end it seems the best Google reCAPTCHA firewall approach for most will be to allow outbound requests on port (80 or 443) based on the hostname google.com. Using DNS allows us to abstract out any changes to the underlying IP addresses. Unfortunately however I believe this approach may not work for everyone as some firewalls will not support this configuration, furthermore I believe configuration of such rules is more complicated than the IP address based approach.  That being said it does provide reliability and is certainly more secure than just allowing all outbound connections on port 80 or port 443 so this would be the approach I would recommend.

What Google reCAPTCHA firewall rules are you guys using to enable use of this tool in your applications?