Share
Embed
Shiny, Let's Be Bad Guys: Exploiting and Mitigating the Top 10 Web App Vulnerabilities -- OSCON 2016 edition
Slides from the OSCON 2016 presentation of our tutorial that walks through the OWASP Top 10 web application vulnerabilities.
Mike Pirnat
May 16, 2016
Transcript
135 slides
-
Shiny Let’s Be Bad Guys! Exploiting and Mitigating the Top
1 0 Web App Vulnerabilities Mike Pirnat - @mpirnat OSCON 2016 Setup: https://github.com/mpirnat/lets-be-bad-guys -
Schedule • This session will run 1:30 PM – 5:00
PM • 30-minute break 3:00 – 3:30 PM • Here on Level 4 -
Shiny Let’s Be Bad Guys! Exploiting and Mitigating the Top
1 0 Web App Vulnerabilities Mike Pirnat - @mpirnat OSCON 2016 -
Why it Matters • Your users • Your data •
Your business -
OWASP • http://www.owasp.org • Open Web Application Security Project •
Non-profit focused on improving software security • Documentation and tools to help learn about security and protect your apps -
OWASP Top Ten • Based on risk data from 8
firms • Over 500,000 vulnerabilities, hundreds of orgs, thousands of apps • Selected & prioritized by prevalence data combined with estimates of exploitability, detectability, and impact • Updated in 2013 -
Today • Background on a type of vulnerability • Exploit
it! • Discuss prevention • Django & Flask specific advice where possible • Light examples where we have time -
Setup: 1 Make & activate a virtualenv: # Python 3...
$ pyvenv badguys # Python 2... $ virtualenv badguys $ cd badguys $ source bin/activate -
Setup: 2 Clone our repository: $ git clone https://github.com/ mpirnat/lets-be-bad-guys
src Or pull the latest changes: $ cd src $ git pull -
Setup: 3 Install dependencies: $ cd src $ pip install
-r requirements.txt -
Injection Attacks • When an application sends untrusted data to
an interpreter • Can result in data loss/corruption, lack of accountability, denial of access • Can lead to complete host takeover -
Trust No One • External users • Internal users •
Administrators -
Attack Vectors • GET parameters • POST parameters • PATH_INFO
• Some HTTP headers: Cookie, Host • Uploaded Files -
Possible Consequences • Creation of malicious SQL (or other queries)
• Accessing private files on disk • Arbitrary code execution -
Real-World Examples • Sony Playstation Network • Ruby on Rails
• HBGary • MySQL • Many, many others… -
SQL Injection • Unescaped user input causes the premature end
of a SQL query and allows a malicious query to be executed... """ select * from users where username='%s'; """ • http://localhost:8000/injection/sql -
Accessing Private Files • File system access + unvalidated user
input allows attackers to navigate the file system • http://localhost:8000/injection/file- access -
Arbitrary Code Execution • Unsafe input is dynamically evaluated or
executed • http://localhost:8000/injection/code- execution -
Prevention • Validate ALL user input • Sign cookies, don’t
accept if signature is bogus/missing • Use ORMs or bind variables when talking to the database • Don’t use eval or exec, beware of pickle, user-supplied YAML, etc. -
Django Advice • Make sure data types for your model
are tight • Use Forms instead of ModelForms for stronger validation • Make new validators as needed for your application • Make sure your URL regexes for dynamic URLs are tight -
Django Advice • Use the ORM when you can •
When you can’t, use extreme caution! • Use bind variables/parameters • No string concatenation/formatting of anything that came from the client -
Without the ORM # If it's a basic select: MyModel.objects.raw("SELECT
... WHERE foo = %s", params={'foo': ...}) # If it's more complicated: from django.db import connection cursor = connection.cursor() cursor.execute("UPDATE bar set bar = 1 WHERE foo < %s", [foo]) row = cursor.fetchall() -
Flask Advice • Use Flask-WTF to validate form input •
Use SQLAlchemy • Bind variables if you don’t -
Broken Auth & Session Management • Attacker uses leaks or
flaws in authentication or session management to impersonate users • Roll-your-own solutions contribute to the difficulty of finding these flaws -
Possible Consequences • Compromised user accounts • Compromised administrative accounts
• Unauthorized use of privileged functionality -
Prevention • Hash or encrypt passwords • Don’t let credentials
be easily overwritten • Don’t put session IDs in URLs • Allow session IDs to timeout/log out • Rotate session IDs after successful login • TLS connections for passwords, session IDs -
Django Advice • Use django.contrib.auth • Consider https://github.com/yourlabs/ django-session-security middleware
for timing out sessions • We’ll talk about transport layer security later on... -
Flask Advice • Use flask-security • Read https://pythonhosted.org/Flask- Security/ •
Avoid overuse of cookies for storing session state -
XSS Attacks • Cross-Site Scripting (XSS) • The most prevalent
web app security flaw • App includes user-supplied data in content sent to the browser without properly validating or sanitizing it -
XSS Attacks • Stored: injected code permanently stored in database,
message forum, comment, etc. • Reflected: injected code in live request to server, reflected back in error message or search result • DOM: injected code in browser DOM environment that causes scripts to run in unexpected ways (eg, reading from URL) -
Possible Consequences • Execute scripts in a victim’s browser •
Hijack sessions • Deface sites • Insert hostile content • Redirect users • Hijack browser (install malware) -
Most Often Seen... • Places where user-created text is displayed
to other users (comments, messages) • Form inputs where value is populated with user-supplied data • Script tags where user-supplied data is populated into script variables -
XSS in Dynamic URLs • Part of the URL path
is variable, isn’t validated, and gets included into the page: http://example.com/stuff/{$CATEGORY} • http://localhost:8000/cross-site- scripting/path-matching/your-path- here -
XSS in Query String Parameters • Unvalidated user input from
a query string parameter is included in the page: http://example.com/?search={$TEXT} • http://localhost:8000/cross-site- scripting/query-params?qs=awesome -
XSS in Form Fields • The value part of an
input is prematurely terminated, allowing Javascript to be injected into or adjacent to the element: <input … value="{$TEXT}"> • http://localhost:8000/cross-site- scripting/form-field -
Prevention • Escape all untrusted data based on the HTML
context the data will be placed into • Whitelist input validation • Consider auto-sanitization libraries for rich content (eg, OWASP’s AntiSamy) • Update your parents’/in-laws’ browsers! -
Django Advice • Be careful with the safe filter, django.utils.safestring,
etc. • Use form.as_p, form.as_table, form.as_ul when displaying a form in a template • Be careful with your own template tags; django.utils.html.escape is your friend! -
django.utils.html cleaned = escape(unsafe_value) cleaned = escapejs(unsafe_value) cleaned = strip_tags(unsafe_value)
cleaned = remove_tags(unsafe_value, ['script', ...]) -
Flask Advice • Don’t disable autoescaping • Be very careful
with anything you bless via Markup objects, the safe filter, or blocks where autoescape is disabled -
Insecure Direct Object Reference • Expose a reference to an
internal implementation object without verifying authorization • Attacker changes URL, GET/POST parameters, cookies -
Possible Consequences • Compromise of all data that can be
referenced by the vulnerable parameter • Unless the namespace is sparse, an attacker can easily access all available data of that type -
Exercises • Manipulate parameters in the URL to access data
that doesn’t belong to you • http://localhost:8000/direct-object- references -
Prevention • Implement access controls on any direct references to
restricted resources • Implement per-user or per-session indirect object references • This can be as much about URL design as about access control! -
Django + Flask • Lock down views: • Use Django’s
permissions architecture • Use Flask-Security or Flask-Login • Customize Django queryset for looking up objects that involve user ownership -
Custom Queryset # In models.py... class ThingyManager(models.Manager): def for_user(self, user):
return self.get_query_set().filter(user=user) class Thingy(models.Model): objects = ThingyManager() # In views.py... class ThingyUpdateView(UpdateView): def get_queryset(self): return Thingy.objects.for_user( self.request.user) -
Possible Consequences • Unauthorized access to some system data or
functionality • Potential complete system compromise -
Text This Really Happens! • Patreon compromised via exposed Werkzeug
Debugger • https://labs.detectify.com/2015/10/02/ how-patreon-got-hacked-publicly- exposed-werkzeug-debugger/ -
Prevention • Have a repeatable hardening process • Have a
process for keeping on top of updates and patches • Architecture that provides secure separation between components • Periodic scans and audits -
Django Advice • Don’t run in debug mode in production
• Keep your SECRET_KEY secret! • Keep Python code out of webserver’s root • Don’t run admin publicly • Don’t use the built-in admin for normal user admin tasks -
Flask Advice • Read http://flask.pocoo.org/docs/ security • Set app.debug =
False to turn off debugging -
Sensitive Data Exposure • Failure to properly protect credit cards,
tax IDs, authentication credentials, etc. • Sensitive data deserves extra protection such as encryption at rest or in transit, special precautions when exchanged with the browser -
Insecure Cryptographic Storage • Not encrypting worthy data • Unsafe
key generation & storage, failure to rotate keys • Weak algorithms • Weak or unsalted hashes -
Insufficient Transport Layer Protection • May not authenticate, encrypt, and
protect the confidentiality and integrity of sensitive network traffic • May use weak algorithms • May use expired or invalid certificates • May use certificates incorrectly -
Possible Consequences • Compromise of all data that should have
been encrypted • This can be highly sensitive information: credentials, credit cards, personal data, health records, etc. -
Possible Consequences • Expose individual users’ data • Account theft
• Compromise an admin account?! • Poor SSL setup can facilitate phishing and man-in-the-middle attacks -
Attack Vectors • Attacker monitors network traffic of your users
• Maybe in public places (Starbucks, conference wi-fi, etc.) • Maybe back end connections • Maybe inside your network (!!!) -
Prevention • Encrypt sensitive data at rest • Encrypt offsite
backups; manage keys separately • Use strong standard algorithms, strong keys • Hash passwords with strong standard algorithm & use appropriate salt • Protect passwords & keys from unauthorized access -
Prevention • Require SSL for all sensitive pages; redirect non-SSL
requests to SSL • Set the “secure” flag on sensitive cookies • Use only strong SSL algorithms • Ensure your cert is valid, not expired, not revoked, and matches your domain • SSL/encryption on the back end too -
Django Advice • Use django.contrib.auth for proper password salting and
hashing • Use bcrypt; see https://docs.djangoproject.com/en/ 1.8/topics/auth/passwords/ • Require SSL in Apache or Nginx • Require SSL using middleware: • Configure Django SecurityMiddleware–new in Django 1.8! • http://django-secure.readthedocs.org/en/v0.1.2/ • https://github.com/rdegges/django-sslify -
Flask Advice • Salt passwords using werkzeug.security: • http://flask.pocoo.org/snippets/54/ •
Require SSL: • https://github.com/kennethreitz/flask-sslify • https://github.com/jacobian/wsgi-sslify • Use bcrypt: https://pythonhosted.org/ passlib/ -
Missing Function Level Access Control • Application doesn’t protect its
functions properly • Misconfiguration • Forgot proper code checks -
Attack Vectors • Authorized user changes a URL or parameter
to a privileged function • Anonymous users could access private functions that aren’t protected -
Possible Consequences • Compromised user accounts • Compromised administrative accounts
• Unauthorized use of privileged functionality -
Exercises • Manipulate the URL to access privileged functionality •
http://localhost:8000/missing-access- control -
Prevention • Consider every page; public or private? • If
authentication is required, make sure that checks are in place • If additional authorization is required, make sure that checks are in place • Deny all by default; explicitly grant access to users or roles -
Django + Flask • Lock down views: • Use Django’s
permissions architecture • Use Flask-Security or Flask-Login • Don’t use Django’s built-in admin for normal user admin tasks -
CSRF Attacks • Cross-Site Request Forgery (CSRF) • Attacker tricks
victim into submitting forged HTTP requests • Attack succeeds if user is authorized/ authenticated -
Attack Vectors • Image tags • Cross-Site Scripting (XSS) •
Fake buttons • Phishing forms • Other techniques -
Possible Consequences • Cause victim to change any data the
victim is allowed to change • Cause victim to perform any function the victim is authorized to use • Impact varies based on victim’s role • Think of some possibilities... -
CSRF via Image • Craft an “image” link that triggers
some site functionality • http://localhost:8000/csrf/image -
CSRF via Form Post • Create an innocuous-looking form that
POSTs to a vulnerable location • http://localhost:8000/csrf/third-party- site -
Prevention • Don’t “do” things on a GET • Include
a unique token in a hidden field (often used in concert with a cookie) • Validate token to make sure the request is from on-site • Avoid putting the token into a query string -
Django Advice • Don’t change the built-in settings! • Do
use the CSRF middleware and template tag in forms • Be VERY CAREFUL about deactivating it (csrf_exempt decorator) • Be careful about APIs (Tastypie, oauth); http://codrspace.com/vote539/csrf- protection-in-django-tastypie/ -
Flask Advice Various CSRF solutions... • DIY/naïve: http://flask.pocoo.org/snippets/3/ • Flask-WTF:
http://flask-wtf.readthedocs.org/ en/latest/csrf.html • Flask-SeaSurf: https://flask- seasurf.readthedocs.org/en/latest/ • Flask-csrf: http://sjl.bitbucket.org/flask-csrf/ -
Components with Known Vulnerabilities • Libraries, frameworks, and other
modules almost always run with full privilege • Hard to stay up to date on everything • Do you even know all the components in use, let alone their versions? • Components with known problems can be identified & exploited with automated tools -
Attack Vectors • Attacker identifies a weak component through scanning
or manual analysis • Customize exploit as needed • More difficult the deeper the component is in the application -
Possible Consequences • Full range of weaknesses are possible •
Impact could be minimal, or... • Complete host takeover! • Data compromise! -
Prevention • Don’t use components you don’t write (unrealistic) •
Keep components up to date • Identify all components and versions • Monitor security of these components -
Python Advice • Use pip to find old packages: pip
list --outdated • Use requires.io to monitor dependencies -
Django + Flask • Keep an eye on the mailing
lists: • https://groups.google.com/group/django- announce • http://flask.pocoo.org/mailinglist • Follow @djangoproject on Twitter -
Redirection Abuse • Attacker tricks user into visiting a URL
that redirects or forwards the request without validating the redirect location • Users prone to click because the link is to a legitimate site -
External Redirection • Use a redirection URL to redirect to
an external location • http://localhost:8000/redirects-and- forwards/redirects -
Forwards • Manipulate a forward parameter to gain access to
privileged functionality • http://localhost:8000/redirects-and- forwards/forwards -
Prevention • Don’t use redirects or forwards • Don’t involve
user-supplied data to build the redirect location • Ensure the supplied value is valid and authorized for the user -
Django Advice • Use django.utils.http.is_safe_url to check redirect URLs •
Used by django.contrib.auth internally • Consider wrapping is_safe_url if you have to allow other off-domain URLs -
Checking Redirects from django.shortcuts import redirect from django.utils.http import is_safe_url
def my_view(request): ... url = 'http://www.example.com/foo/bar' if is_safe_url(url, host='www.example.com'): return redirect(url) else: ... -
Allowing Multiple Safe Hosts from django.utils.http import is_safe_url def is_whitelisted_url(url,
hosts): for host in hosts: if is_safe_url(url, host=host): return True return False >>> url = 'http://us.pycon.org' >>> whitelist = ['us.pycon.org', 'pycon.org', ...] >>> is_whitelisted_url(url, whitelist) True -
Flask Advice • Redirect back: http://flask.pocoo.org/ snippets/62/ • Redirect back,
more selectively with whitelist, using cookies for previous URL: http://flask.pocoo.org/snippets/120/ • Can modify these approaches to suit, as with the Django example -
Links • http://www.owasp.org • https://www.owasp.org/index.php/ Category:OWASP_Top_Ten_Project • https://docs.djangoproject.com/en/dev/ topics/security/ •
http://flask.pocoo.org/docs/security • https://github.com/mpirnat/lets-be-bad-guys