How to Build a Secure RPC Interface for AJAX Apps With Google Web Toolkit
Why use GWT?
Most modern web applications utilize an AJAX functionality of some sort to make them highly interactive and to have a user interface that works very much like that of a traditional desktop application. If you are looking to build a web application that takes advantage of AJAX functionality, provides a very modern look-and-feel and is very easy to maintain, I highly recommend using Google Web Toolkit (GWT).
These are some of the advantages that I believe makes GWT a good choice for building a web application:
![11 courses, 8+ hours of training](/contentassets/268131c75b0249f5bd1b3b2d7d3b3a02/hackable-book-by-ted-harrington.jpg)
11 courses, 8+ hours of training
11 courses, 8+ hours of training
- GWT is freely available from Google and there is a large developer community for it.
- GWT apps are coded entirely in Java and there is no need to understand JavaScript unless you want to add very specialized customization. The Java source is compiled into JavaScript optimized for each possible client browser, so you can take advantage of all browsers equally with a single set of source code.
- GWT apps are well supported across most browsers.
- All of the development tools needed are available for free, including Eclipse as an IDE and plenty of plugins to enhance it.
- Numerous third party toolkits that provide additional widgets, functionality, and support.
- GWT utilizes a sophisticated and powerful, yet simple-to-use, Remote Proceedure Call (RPC) mechanism to provide AJAX-style communication between the client and server.
The last bullet is a key feature of GWT that makes building AJAX functionality quite simple. But if the RPC interface you build is not designed correctly, you could be opening up the server-side of your application to vulnerability. The rest of this article will cover some design principles that should be considered if you want to be sure that the communication between the client and server is secure.
A fictional web application example
First I'll be providing a brief overview of how to build a GWT RPC interface.
It is important to note that this article is not meant to be a guide on how to build a GWT RPC interface. The point of this article is to explain how to design your interface in a secure way once you are already familiar with the implementation details. Please refer to the GWT RPC documentation for details on how to build, use, and deploy an interface if you are not already familiar with it.
The infrastructure that GWT RPC provides gives you the ability to have your server offer Java functions that can be called from the client as if making a regular Java function call, with the only caveat being that all function returns are done asynchronously and never synchronously. You can pass entire objects as input parameters and as return parameters -- GWT RPC will take care of the marshalling of these complex data types for you.
Apache Tomcat is typcially used to host the servlet or servlets that provide the server-side code for the RPC interface. The client code is JaveScript, compiled from the Java source by the GWT compiler, running in the browser.
Our web application example will be access to a list of contacts that can be viewed or edited. On the server we will have a database that stores the contacts. The client UI will give the end user the ability to view the current set of contacts, add, edit, or delete contacts from the list, and send an email to anyone on the contact list. To further illustrate security principles in the design of the RPC interface for this app, end users will need to log in and certain users will have access to only certain subsets of the entire list of contacts.
The RPC service will be called ContactsService and the service interface would be defined in ContactsService.java like this:
public interface ContactsService extends RemoteService {
// LoginStatus contains information on whether
// the login succeeded or failed and possible
// failure information
LoginStatus login(String username, String password);
void logout();
// Contact contains name, email address, phone number, etc.
// addContact() returns the contactId of the added contact
int addContact(Contact contactObject);
void deleteContact(int contactId);
Contact[] getContacts();
void sendEmailToContact(int contactId, String subject, String body);
}The service async interface would then be defined in ContactsServiceAsync.java like this:
public interface ContactsServiceAsync {
void login(String username, String password, AsyncCallback<LoginStatus> callback);
void logout(AsyncCallback<Void> callback);
void addContact(Contact contactObject, AsyncCallback<int> callback);
void deleteContact(int contactId, AsyncCallback<Void> callback);
void getContact(AsyncCallback<Contact[]> callback);
void sendEmailToContact(int contactId, String subject, String body, AsyncCallback<Void> callback);
}So in the client code, when you need to call a function in the RPC interface, the client would invoke it like this:
public int addContact() {
Contact newContact = new Contact();
newContact.name = "Fred Flintstone";
newContact.emailAddress = "fred@slatesquary.com";
newContact.phoneNumber = "555-1212";
GWT.create(ContactsService.class).addContact(newContact, new AsyncCallback<int>() {
@Override
public void onFailure(Throwable caught) {
}
@Override
public void onSuccess(int result) {
int newContactId = result;
}
});
}
Remember what is on the client and what is on the server
The reason for having an RPC interface is so that you can implement your web application in a way that utilizes code running both in the client browser and on the server. Code running on the server would typically be used to read and write a database or to invoke functionality that cannot be done from a sandboxed web browser, i.e., sending an email or connecting to servers other than the server that served up the client code (most browsers will block this to prevent cross-site scripting attacks).
So in general: code running on the client provides the user interface and direct interaction with the user, and code running on the server provides the actual data manipulation that the client code is driving. For this reason, the server is the side you need to guard carefully.
Code that executes on the client should not be responsible for security in any way. Client code is served up to the end user and once given to them they have the ability to potentially modify it in any way. There are plenty of browser plugins that allow access to and modification of code served for a web application. And the end user may not even be using a standard browser. HTTP is an open protocol and therefore anyone who has access permission (as administered on the web server) to your web application has the ability to get your application's code.
You may think that using an obfuscator on your web application's client code would make the client code secure from end-user tampering but most obfuscators are able to be foiled - although at this point you have raised the bar a little bit higher.
But note that the server-side of your RPC service is still open to be called even by someone who has access to your web application but may not be able to access in any usable way the client code meant to invoke that interface.
The functions on your server are callable by anyone that has access to your web application
A very important point to be made here (and the crux of this article) is that all of the function calls hosted on your server can be called by anyone that has access to your web application regardless of whatever you put in your client code. As illustrated in the above section you cannot rely on your client code to guard what can be done through your RPC interface.
You need to design your RPC interface in a way that only authorized end users can do what you have authorized them to do.
Use the session to track the end user
The server used to host your RPC interface keeps track of all client sessions with the server. Once a browser instance requests anything from the server, the server will track that particular browser instance and any subsequent request from the same browser instance - even from a different tab or window, they will be tracked as being in the same session. The session will be remembered even if the network connection between client and server is severed for some time period.
The session is tracked on the server with the HttpSession object. In your server code, this can be accessed from the HttpServletRequest object which is available in any RPC function implementation. Note that the RPC implementation class extends RemoteServiceServlet which extends HttpServlet. [The HttpServlet class is a powerful class that provides access to everything about the connection between the client and the server. The javadoc for HttpServlet is a valuable reference.] All of the functions available in HttpServlet are therefore available within the implementation of any of your RPC functions.
The following code gives an example of how to use the session to track if the user is considered by your server implementation to be logged in or not:
public class ContactsServiceImpl extends RemoteServiceServlet implements ContactsService {
private static final String LOGGEDINUSERNAME_ATTR = "ContactsLoggedinUserName";
@Override
public LoginStatus login(String username, String password) {
HttpServletRequest request = getThreadLocalRequest();
HttpSession session = request.getSession(true);
if (session.getAttribute(LOGGEDINUSERNAME_ATTR) != null) {
return new LoginStatus("Already logged in");
}
else {
LoginStatus loginStatus = verifyLogin(username, password);
if (loginStatus.isValid)
session.setAttribute(LOGGEDINUSERNAME_ATTR, username);
return loginStatus;
}
}
@Override
public void logout() {
HttpServletRequest request = getThreadLocalRequest();
HttpSession session = request.getSession(true);
session.removeAttribute(LOGGEDINUSERNAME_ATTR);
}
// NOTE: Not part of the RPC interface so not
// callable from the client
private boolean isLoggedIn() {
HttpServletRequest request = getThreadLocalRequest();
HttpSession session = request.getSession(true);
return session.getAttribute(LOGGEDINUSERNAME_ATTR) != null;
}
}For simplicity the code example uses the username to track the currently logged in user. But you probably want to use something like the database ID of the login entry that is likely to be more persistent over time.
Note that the session attributes appear to work like a cookie. But setting a session attribute is much different than using a cookie. Cookies are tracked on the client side by the browser. The end user has full access to the store of cookies and can easily tamper with them. Session attributes are stored only on the server and the client has no way of accessing them. If you choose to store any state about the server locally on the client, make sure that it is only in addition to and not instead of storing the same state on the server.
The bottom line is that the server alone should hold the authoritative state of the ability to access any element of the server. If the client wants to ask if a particular user is currently logged in, it should ask the server via an RPC call -- and that RPC function should verify that the caller has the right to ask the question before returning an answer.
Check access rights on every RPC call
Remember that a malicious attacker will not respect your idea of which RPC functions to call in which order. They can potentially call any function in any order and pass in any data of the types that the functions accept as parameters.
Every single function in your RPC interface must verify that the current session is authorized to get back any information the function may return (e.g. any of the contacts in the list of contacts) or is authorized to carry out any activity the function may do (e.g. send an email, modify the database, etc.).
Building on our example, every RPC function must at the very least check to see if the current session is logged in before continuing to carry out the function. And it should additionally verify that the currently logged in user has permissions to carry out the function. For example, when querying the list of contacts, certain users should only see a subset of the list of all contacts. Your database design should contain the information used to track who has access to what.
The function getContacts() will do nothing if the current session is not logged in and if a user is properly logged in, it will filter out to only the Contacts that the current user has the rights to view:
public class ContactsServiceImpl extends RemoteServiceServlet implements ContactsService {
private static final String LOGGEDINUSERNAME_ATTR = "ContactsLoggedinUserName";
@Override
public LoginStatus login(String username, String password) {
HttpServletRequest request = getThreadLocalRequest();
HttpSession session = request.getSession(true);
if (session.getAttribute(LOGGEDINUSERNAME_ATTR) != null) {
return new LoginStatus("Already logged in");
}
else {
LoginStatus loginStatus = verifyLogin(username, password);
if (loginStatus.isValid)
session.setAttribute(LOGGEDINUSERNAME_ATTR, username);
return loginStatus;
}
}
@Override
public void logout() {
HttpServletRequest request = getThreadLocalRequest();
HttpSession session = request.getSession(true);
session.removeAttribute(LOGGEDINUSERNAME_ATTR);
}
@Override
public Contact[] getContacts() {
if (!isLoggedIn())
return null;
String currentlyLoggedinUser = getCurrentLoggedinUser();
ArrayList<Contact> contactList = new ArrayList<Contact>();
for (Contact contact : getAllContactsInDatabase()) {
if (userCanSeeContact(contact, currentlyLoggedinUser))
contactList.add(contact);
}
return contactList.toArray(new Contact[contactList.size()]);
}
// NOTE: Not part of the RPC interface so not
// callable from the client
private boolean isLoggedIn() {
HttpServletRequest request = getThreadLocalRequest();
HttpSession session = request.getSession(true);
return session.getAttribute(LOGGEDINUSERNAME_ATTR) != null;
}
private String getCurrentLoggedinUser() {
HttpServletRequest request = getThreadLocalRequest();
HttpSession session = request.getSession(true);
return (String) session.getAttribute(LOGGEDINUSERNAME_ATTR);
}
}Summary
This method is one of the safest I know for building an RPC interface. However, there are other methods and you may find another scheme that works better for your particular situation, especially depending on the level of security you need for your server.
I would love to hear from folks if they have different methods that have worked well for them in the comments below.
![11 courses, 8+ hours of training](/contentassets/a657be75f19f4620a977a3534aa70a42/hackable-book-by-ted-harrington.jpg)
11 courses, 8+ hours of training
11 courses, 8+ hours of training
Useful Links
- GWT Home Page -- The top level of all information Google provides on GWT
- GWT Developer's Guide -- The best place to get started with GWT
- GWT Server Communication -- How to create an RPC interface