Share
Embed
Esoteric Web Application Vulnerabilities
A summary of the strangest vulnerabilities I've found during last year
andresriancho
April 13, 2017
Transcript
51 slides
-
/me • Application security expert (web|API) • Developer (Python!) •
Open Source evangelist • w3af project leader • Independent consultant -
ORM killed the pentest star • All modern web development
frameworks provide abstractions to interact with (no) SQL databases. Developers don’t write raw SQL queries anymore. • SQL injections are rare nowadays, this requires us testers to dig deeper into the application to find high risk vulnerabilities. -
MVC, templates and default HTML encode killed XSS • Most
modern web development frameworks use a model view controller architecture, which uses templates to render the HTML shown to users. • Templating engines, such as Jinja2, HTML encode the context data by default. • Developers need to write more code to make the template vulnerable to Cross-Site Scripting, which leads to less vulnerabilities. <ul> {% for user in user_list %} <li><a href="{{ user.url }}">{{ user.username }}</a></li> {% endfor %} </ul> -
Aggressive input decoding Ruby on Rails, Sinatra and other (ruby)
web frameworks perform aggressive input decoding: http://www.phrack.org/papers/attacking_ruby_on_rails.html post '/hello' do name = params[:name] render_response 200, name POST /hello HTTP/1.1 Host: example.com Content-Type: application/x-www-form-urlencoded name=andres POST /hello HTTP/1.1 Host: example.com Content-Type: application/json {"name": "andres"} -
Decode to a Ruby Hash POST /hello HTTP/1.1 Host: example.com
Content-Type: application/json {"name": {"foo": 1}} In all previous cases the type of the name variable was a String, but we can force it to be a Hash: -
noSQL ODM introduction When MongoId ODM (Object Document Mapper) and
similar frameworks are in use developers can write code similar to: Which will query the Mongo database and return the first registration flow where the user_id and confirmation_token match. post '/registration/complete' do registration = Registration.where({ user_id: params[:user_id], confirmation_token: params[:token] }).first ... POST /registration/complete HTTP/1.1 Host: vulnerable.com Content-Type: application/json {"token": "dee1...bb8e", "user_id": 3578} -
noSQL ODM complex queries Developers can write “complex” ODM queries
using Ruby Hashes as parameters: user = Users.where({user_id: params[:user_id], country: {"$ne": "Argentina"}}).first users = Users.where({user_id: {"$in": [123, 456, 789]}}) -
Decode to Hash leads to noSQL injection It’s possible to
bypass the token validation! post '/registration/complete' do registration = Registration.where({ user_id: params[:user_id], confirmation_token: params[:token] }).first ... POST /registration/complete HTTP/1.1 Host: vulnerable.com Content-Type: application/json {"token": {"$ne": "nomatch"}, "user_id": 3578} -
“User controlled input”.to_s Fixing this vulnerability is quick and easy:
Most developers will forget to add the .to_s and it’s easy to miss in a source code review. Recommend Sinatra param or similar. get '/registration/complete' do @registration = Registration.where({ user_id: params[:user_id].to_s, confirmation_token: params[:token].to_s }).first ... -
Call me to verify my identity #1 The application requires
users to provide a cellphone to verify their identity. A phone call is initiated by the application using a service like Twilio, the call audio contains a verification code which needs to be input into the application to verify phone ownership. HTTP request Verify my phone +1 (541) 754-3010 -
Call me to verify my identity #2 Call +1 (541)
754-3010 Send code 357896 in audio HTTP request Please call +1 (541) 754-3010 Audio for the call is available at https://vulnerable.com/audio/<uuid-4> HTTP request https://vulnerable.com/audio/<uuid-4> -
Call me to verify my identity #3 HTTP request Code
is 357896 HTTP response Welcome admin! -
Bypass phone verification Hacker wants to bypass phone verification, ideas:
◦ Hack admin’s smartphone ◦ Hack vulnerable.com ◦ Create a raw cellphone tower and sniff admin’s phone call ◦ Hack Twilio Hacking vulnerable.com seems to be the easiest path to follow. But… what do we need? -
UUID4 Version 4 UUIDs use a scheme relying only on
random numbers, thus the audio URLs can’t be brute forced: https://vulnerable.com/audio/f47ac10b-58cc-4372-a567-0e02b2c3d479 -
Zoom into HTTP request to Twilio HTTP request Please call
+1 (541) 754-3010 Audio for the call is available at https://vulnerable.com/audio/<uuid-4> POST /call/new HTTP/1.1 Host: api.twilio.com Content-Type: application/json X-Authentication-Api-Key: 2bc67a5... {"phone_number": "+1 (541) 754-3010"}, "audio_callback": "https://vulnerable.com/f47ac10b-5..."} -
Insecure Twilio API call HTTP request Please call +1 (541)
754-3010 Audio for the call is available at https://vulnerable.com/audio/<uuid-4> import requests def start_call(phone, callback_url): requests.post('https://api.twilio.com/call', data={'phone_number': phone, 'audio_callback': callback_url}) … audio_id = generate_audio(request.user_id) callback_url = 'https://%s/%s' % (request.host, audio_id) start_call(request['phone'], callback_url) -
Change Host header to exploit HTTP request Verify my phone
+1 (541) 754-3010 POST /verify-my-phone HTTP/1.1 Host: vulnerable.com Content-Type: application/json {"phone_number": "+1 (541) 754-3010"}} POST /verify-my-phone HTTP/1.1 Host: evil.com Content-Type: application/json {"phone_number": "+1 (541) 754-3010"}} -
Exploit results in modified callback_url HTTP request Please call +1
(541) 754-3010 Audio for the call is available at https://evil.com/audio/<uuid-4> HTTP request https://evil.com/audio/<uuid-4> HTTP request https://vulnerable.com/audio/<uuid-4> -
MUST-HAVE: Strict validation for Host header • Make sure that
your nginx, apache, and web frameworks validate the host header before any further code is run. • Django has strict host header validation built in using ALLOWED_HOSTS configuration setting. -
Password reset • Password resets are very sensitive and, in
some cases, insecure. The most wanted vulnerability is to be able to reset the password for a user for which we don’t have the password reset token. • Usually password resets are implemented as follows: ◦ User starts a new password reset flow ◦ An email is sent by the application containing a randomly generated token ◦ The token is used to prove that the user has access to the email address and the password is reset. -
Implementation details class AddPasswordResetTokenToUser < ActiveRecord::Migration def change add_column :users,
:pwd_reset_token, :string, default: nil end end post '/start-password-reset' do: user = Users.where({"email": params["email"]}).first token = generate_random_token() user.pwd_reset_token = token user.save! send_email(user.email, token) post '/complete-password-reset' do: user = Users.where({"pwd_reset_token": params["token"]}).first user.password = params["new_password"] user.pwd_reset_token = nil user.save! -
Token defaults to NULL in the database POST /complete-password-reset HTTP/1.1
Host: vulnerable.com Content-Type: application/json {"token": null, "new_password": "l3tm31n"} • Each time a new user is created his pwd_reset_token field is set to NULL in the database. • When the user starts a new password reset flow a randomly generated token is assigned to pwd_reset_token • What if... -
Safe defaults and strict type validation post '/complete-password-reset' do: user
= Users.where({"pwd_reset_token": params["token"].to_s}).first user.password = params["new_password"] user.pwd_reset_token = nil user.save! class AddPasswordResetTokenToUser < ActiveRecord::Migration def change add_column :users, :pwd_reset_token, :string, default: generate_random_token() end end -
Paypal’s Instant Payment Notification • I love payment gateways! See
my previous talk on this subject. • Paypal uses IPN to notify a site that a new payment has been processed and further action, such as increasing the user funds in the application, should be performed. • The developer sets the IPN URL in the merchant account settings at Paypal: https://www.example.com/paypal-handler -
Zoom into Paypal’s IPN HTTP request POST /paypal-handler HTTP/1.1 Host:
www.example.com Content-Type: application/x-www-form-urlencoded mc_gross=19.95&protection_eligibility=Eligible&address_status=confirmed&p ayer_id=LPLWNMTBWMFAY&tax=0.00&address_street=1+Main+St&payment_date=20%3 A12%3A59+Jan+13%2C+2009+PST&payment_status=Completed&charset=windows-1252 &address_zip=95131&first_name=Test&mc_fee=0.88&address_country_code=US&ad dress_name=Test+User¬ify_version=2.6&custom=665588975&payer_status=ver ified&address_country=United+States&address_city=San+Jose&quantity=1&veri fy_sign=AtkOfCXbDm2hu0ZELryHFjY-Vb7PAUvS6nMXgysbElEn9v-1XcmSoGtf&payer_em ail=gpmac_1231902590_per%40paypal.com&txn_id=61E67681CH3238416&payment_ty pe=instant&last_name=User&address_state=CA&receiver_email=gpmac_123190268 6_biz%40paypal.com&payment_fee=0.88&receiver_id=S8XGHLYDW9T3S&txn_type=ex press_checkout&item_name=&mc_currency=USD&item_number=&residence_country= US&handling_amount=0.00&transaction_subject=&payment_gross=19.95&shipping =0.00 -
Zoom into Paypal’s IPN HTTP request There are a few
important parameters that we need to understand: • mc_gross=19.95 is the amount paid by the user • custom=665588975 is the user’s ID at the merchant application, which is sent to Paypal when the user clicks the “Pay with Paypal” button in the merchant’s site • receiver_email=gpmac_1231902686_biz%40paypal.com is the merchant’s email address • payment_status=Completed is the payment status -
Insecure IPN handler import requests PAYPAL_URL = 'https://www.paypal.com/cgi-bin/webscr?cmd=_notify-validate' def handle_paypal_ipn(params):
# params contains all parameters sent by Paypal response = requests.post(PAYPAL_URL, data=params).text if response == 'VERIFIED': # The payment is valid at Paypal, mark the cart instance as paid cart = Cart.get_by_id(params['custom']) cart.record_user_payment(params['mc_gross']) cart.user.send_thanks_email else: return 'Error' -
• Attacker needs to perform a special Paypal payment using
a target specific custom_id parameter which will associate the spoofed payment with his account. • The payment is made from the attacker’s credit card to his paypal account. Money is still under his control, but the attacker will lose Paypal’s commission for each transaction. • Many example IPN implementations in github.com are vulnerable. I wonder how many were used to create applications which are currently live in production? -
Secure IPN handler import requests PAYPAL_URL = 'https://www.paypal.com/cgi-bin/webscr?cmd=_notify-validate' MERCHANT_PAYPAL_USER =
'[email protected]' def handle_paypal_ipn(params): if params['receiver_email'] == MERCHANT_PAYPAL_USER: return 'Error' # params contains all parameters sent by Paypal response = requests.post(PAYPAL_URL, data=params).text if response == 'VERIFIED': # The payment is valid at Paypal, mark the cart instance as paid cart = Cart.get_by_id(params['custom']) cart.record_user_payment(params['mc_gross']) cart.user.send_thanks_email else: return 'Error' -
Is this Paypal’s fault? • Are all payment gateways vulnerable?
• MercadoPago implemented a different communication protocol for their IPN. Their protocol is much better than Paypal’s since it doesn’t rely on the developer’s IPN handler implementation to provide security. • MercadoPago sends a GET request with the purchase ID to the IPN URL, then the developer needs to perform a GET request to https://api.mercadopago.com/ in order to retrieve the transaction details. This request is authenticated, and any attempts to access transactions from other merchants is denied. -
ActiveSupport::MessageVerifier Marshal RCE • ActiveSupport::MessageVerifier uses Ruby’s Marshal to serialize
arbitrary information, which is then signed using a developer provided secret. A verified message looks like: • The message can be decoded: BAhJIhphb...XNlYy5jb20GOgZFVA==--8bacd5cb3e72ed7c457aae1875a61d668438b616 1.9.3-p551 :006 > Base64.decode64('BAhJIhphbmRyZXNAYm9uc2FpLXNlYy5jb20GOgZFVA==') => "\x04\bI\"\[email protected]\x06:\x06ET" 1.9.3-p551 :007 > Marshal.load(Base64.decode64('BAhJIhphbmRyZXNAYm9uc2FpLXNlYy5jb20GOgZFVA==')) => "[email protected]" 1.9.3-p551 :008 > -
ActiveMessages are signed • When the application receives the signed
message, it will take the base64 encoded data and calculate HMAC SHA1 for it using using the developer controlled secret. • The calculated signature must match the one provided with the message: • Once the signature is verified the data is base64 decoded and Unmarshaled. BAhJIh...--8bacd5cb3e72ed7c457aae1875a61d668438b616 -
Guessable signing secret leads to RCE Ruby’s documentation clearly states
that unmarshaling arbitrary data is insecure and will lead to arbitrary code execution. ActiveSupport::MessageVerifier is protected against this vulnerability by a developer controlled secret. Poorly chosen secrets allow: Brute-force attack to discover the secret Specially crafted gadget/object is created, serialized and encoded. Secret is used to sign gadget Signed message is sent to the application, where it will be unmarshalled and remote code execution is achieved -
Secure ActiveSupport::MessageVerifier usage • Choose randomly generated, long, secrets to
sign your messages. • Use a different serialization method: @verifier = ActiveSupport::MessageVerifier.new(long_secret, serializer: json) -
Vulnerabilities are always there • You’re smarter than your tools.
Let the automation do the grunt work and focus your time on source code review, application logic flaws, issues specific to the target application, etc. • You’re smarter than your client. Convince them that with the source code you’ll be able to identify more vulnerabilities and provide greater ROI. • You’re smarter (well, actually more trained in security, vulnerabilities and risks) than most developers. They will make mistakes, no matter how good they are. -
For hire Does your company need these services? • Application
Penetration Test • Secure Coding Training for Developers • Source Code Review • Cloud Security Assessment Let's get in touch, I can help your company deliver secure web applications.