Monday, 26 November 2012

Puzzle 6: Multicast

int i = -1;
byte b = (byte)i;
char c = (char)b;
int i2 = (int)c;

The value of the variable in binary format is:

i = 1111 1111 1111 1111 1111 1111 1111 1111
b = 1111 1111
c = 1111 1111 1111 1111
i2 = 0000 0000 0000 0000 1111 1111 1111 1111

So we can see, from char to int, the sign is not considered. Simply prefixing 0s will do it.

If you wish to keep the sign, you need to cast char to short

short s = (short)c;
int i3 = (int)s;

result is:

s = 1111 1111 1111 1111
i3 = 1111 1111 1111 1111 1111 1111 1111 1111

If you cast byte to char, and you don't want to keep the sign. e.g. you want to achieve the following effect.

b = 1111 1111
c = 0000 0000 1111 1111

You can use bit mask:

char c = (char)(b & 0xff);

b & 0xff is of type int, so effectively

b & 0xff = 0000 0000 0000 0000 0000 0000 1111 1111

Puzzle 5: What does it mean by Hex and Octal literals are negative if their high-order bit is set?

In the book, "Java Puzzle", Puzzle 5: The Joy of Hex, there is a bold line Hex and Octal literals are negative if their high-order bit is set to explain the number 0xcafebabe is equivalent to the decimal value -889275714.

So what does "high-order bit is set" mean?

"high-order bit" is the left most bit of a given type. For example, if type is integer, which has 32 bits, then the high-order bit is the 32nd bit counting from right to left.

The 32nd bit counting from right to left is 0 in the following case, so the number is a positive number

int max = Integer.valueOf("01111111111111111111111111111111", 2);
System.out.println(max);

The result is

2147483647

which happens to be the maximum integer number.

Now let's convert the Hex format number 0xcafebabe to binary format

String s = Integer.toBinaryString(0xcafebabe);

The result is

11001010111111101011101010111110

The high-order bit is 1, therefore, it is a negative number.

Monday, 22 October 2012

Request, Flash, View Scope in Spring webflow


What is the difference among these 3 scopes?

According to the Java doc:

Request: Attributes placed in request scope exist for the life of the current request into the flow execution. When the request ends any attributes in request scope go out of scope.

Flash: Attributes placed in flash scope exist through the life of the current request and until the next view rendering. After the view renders, flash scope is cleared. Flash scope is typically used to store messages that should be preserved until after the next view renders.

View: Attributes placed in view scope exist through the life of the current view state and until the view state exits in a subsequent request. View scope is typically used to store view model objects manipulated over a series of Ajax requests.

I don't think I can see their distinction clearly from the definitions above. So I am going to run some experiments to find out myself.

In the spring-web-flow.xml, I create a <view-state>.

<view-state id="dummy">
    <on-entry>
        <set name="viewScope.viewScopeAttribute" value="'v'" />
        <set name="flashScope.flashScopeAttribute" value="'f'" />
        <set name="requestScope.requestScopeAttribute" value="'r'" />
    </on-entry>
</view-state>

the dummy.xhtml JSF page is very simple:

request scope: #{requestScopeAttribute}
flash scope: #{flashScopeAttribute}
view scope: #{viewScopeAttribute}

I was expecting to see all 3 attributes displayed, but the request scope attribute is missing.

Let's change the <view-state> to

<view-state id="dummy">
    <on-render>
        <set name="viewScope.viewScopeAttribute" value="'v'" />
        <set name="flashScope.flashScopeAttribute" value="'f'" />
        <set name="requestScope.requestScopeAttribute" value="'r'" />
    </on-render>
</view-state>

All 3 attributes display this time. why?

This is because every time Spring webflow needs to render a view, it will issue a redirect causing the view to be rendered in the subsequent GET request. This is useful because when the user hit Refresh or Back button, the browser won't give any warning.

The actions in <on-entry> occur during the first request and any attributes in this request's scope will have been blown off by the time the view is rendered. Because the view is rendered in the second request.

The actions in <on-render> occur during the second request, so the attributes in this request's scope will be kept when the view is rendered.

As for the difference between viewScope and flashScope, I really cannot tell any as long as they are in <view-state>.  I think they can be used interchangeably in <view-state>. (I could be very wrong here).

However, viewScope cannot be used in <action-state> or <decision-state>.

flashScope will retain in memory until after the view is rendered. For example:

<action-state id="dummy0">
    <on-entry>
        <set name="flashScope.flashScopeAttribute" value="'f'" />
    </on-entry>
    <evaluate expression="new java.lang.String()" />
    <transition to="dummy"/>
</action-state>
 
<view-state id="dummy" /> 

The flash scope attribute still displays on the page.

Sunday, 21 October 2012

Spring Security form-login behind the scene

In Spring in Action 3rd Edition, chapter 9, Securing Spring, page 228, here is a summary of the main points on this page.

The filter is defined in web.xml as:

<filter>
    <filter-name>springSecurityFilterChain</filter-name>
    <filter-class>org.springframework.web.filter.DelegatingFilterProxy</filter-class>
</filter>
  
<filter-mapping>
    <filter-name>springSecurityFilterChain</filter-name>
    <url-pattern>/*</url-pattern>
</filter-mapping>

The name of the filter springSecurityFilterChain is significant, meaning that you cannot give it an arbitrary name or the Spring security framework won't be able to find this filter.

The DelegatingFilterProxy doesn't do much and delegate the work to a special filter known as FilterChainProxy.

The FilterChainProxy is a single filter that chains together multiple additional filters. These filters, along with the FilterChainProxy, are created by Spring based on the security configuration. We will never need to explicitly declare the FilterChainProxy bean, so we don't need to know the details.

Ok, this is the gist of the page. As much as it says we don't need to know the details, I still have a few questions in mind.

  • How does FilterChainProxy chain together multiple other filters?
  • Which filter checks whether the provided username and password match the true credential?
  • There is an implicit object known as 'currentUser' in Spring Webflow. At what scope (request, session, flow) is this object stored?


Part 1.  How does FilterChainProxy chain together multiple other filters?


I am going to use the booking-face Spring Webflow sample project.

Let's start deployment and put a break point at DelegatingFilterProxy.java @Line 226. 

this.delegate = initDelegate(wac);

Step into it. DelegatingFilterProxy.java @Line 326. 

Filter delegate = wac.getBean(getTargetBeanName(), Filter.class);

After this line is executed, we see delegate is an instance of FilterChainProxy, and it contains lots of filters already.

FilterChainProxy[ UrlMatcher = org.springframework.security.web.util.AntUrlPathMatcher[requiresLowerCase='true']; Filter Chains: {/**=[org.springframework.security.web.context.SecurityContextPersistenceFilter@10896d0, org.springframework.security.web.authentication.logout.LogoutFilter@e542a1, org.springframework.security.web.authentication.UsernamePasswordAuthenticationFilter@179763c, org.springframework.security.web.authentication.www.BasicAuthenticationFilter@1416e4f, org.springframework.security.web.savedrequest.RequestCacheAwareFilter@b11bbf, org.springframework.security.web.servletapi.SecurityContextHolderAwareRequestFilter@194aa64, org.springframework.security.web.authentication.AnonymousAuthenticationFilter@13e985a, org.springframework.security.web.session.SessionManagementFilter@8475c4, org.springframework.security.web.access.ExceptionTranslationFilter@4fd3a5, 
org.springframework.security.web.access.intercept.FilterSecurityInterceptor@d03044]}]

It is easily seen that this FilterChainProxy bean has long been created and stored in the web application context. The ContextLoaderListener was in play and if interested, please refer to my article What does ContextLoaderListener do in Spring?

Now we put a break point at DelegatingFilterProxy.java @Line 259 and launch the application (by go to the url http://localhost:8080/booking-faces/spring/main).

invokeDelegate(delegateToUse, request, response, filterChain);

Step into it. DelegatingFilterProxy.java @Line 346

delegate.doFilter(request, response, filterChain);

Step into it. FilterChainProxy.java @134-149

FilterInvocation fi = new FilterInvocation(request, response, chain);
List<Filter> filters = getFilters(fi.getRequestUrl());

if (filters == null || filters.size() == 0) {
    if (logger.isDebugEnabled()) {
        logger.debug(fi.getRequestUrl() +
        filters == null ? " has no matching filters" : " has an empty filter list");
    }

    chain.doFilter(request, response);

    return;
}

VirtualFilterChain virtualFilterChain = new VirtualFilterChain(fi, filters);
virtualFilterChain.doFilter(fi.getRequest(), fi.getResponse());

This is the central part. It differs from an everyday Filter in that for an everyday filter, after you do some processing, you invoke chain.doFilter() to give other filters chance to do their work. A typical example is this CharacterEncodingFilter

protected void doFilterInternal(
        HttpServletRequest request, HttpServletResponse response, FilterChain filterChain)
        throws ServletException, IOException {

    if (this.encoding != null && (this.forceEncoding || request.getCharacterEncoding() == null)) {
        request.setCharacterEncoding(this.encoding);
        if (this.forceEncoding) {
            response.setCharacterEncoding(this.encoding);
        }
    }
    filterChain.doFilter(request, response);
}

The FilterChainProxy also has this line @Line 143, am I too blind to notice that? Actually this line will not be executed. To be precisely, not until all the spring created filters have done their job.

We can press F6 to verify. Line 137-146 are skipped, and we are at Line 148. We are going to step into Line 149. But before we do, I want to take note of the chain object in the method argument list. It is org.apache.catalina.core.ApplicationFilterChain@9a731a

virtualFilterChain.doFilter(fi.getRequest(), fi.getResponse());

FilterChainProxy.java @Line 355

nextFilter.doFilter(request, response, this);

Step into it. Now we are at the very first of the so called "additional filters". SecurityContextPersistenceFilter.java @Line 50.

public void doFilter(ServletRequest req, ServletResponse res, FilterChain chain)

Inspect the chain object, it is org.springframework.security.web.FilterChainProxy$VirtualFilterChain@1d418d4. The chain object is no longer the one created by the Application server (Glassfish, in this case). It now becomes the VirtualFilterChain created by Spring.

This is how, even though not defined in web.xml, these "additional filters" got chained together to do their job. Had we defined them in web.xml, we wouldn't have needed VirtualFilterChain. But the web.xml would have exploded and that would be the last thing we want to see. (Configuration is really pain in the ass)


Part 2.  Which filter checks whether the provided username and password match the true credential?


It is the UsernamePasswordAuthenticationFilter.

Let's put a break point @Line 97.

return this.getAuthenticationManager().authenticate(authRequest);

Just press F7 to jump out of this method and we are at AbstractAuthenticationProcessingFilter.java @Line 205

authResult = attemptAuthentication(request, response);

Step down till @Line 219

successfulAuthentication(request, response, authResult);

Step into it. AbstractAuthenticationProcessingFilter.java @Line 293

SecurityContextHolder.getContext().setAuthentication(authResult);

Put the authentication information into security context. The security context is saved in a thread local object.

AbstractAuthenticationProcessingFilter.java @Line 302

successHandler.onAuthenticationSuccess(request, response, authResult);

Step into it...eventually it hits SaveContextOnUpdateOrErrorResponseWrapper.java @Line 73

doSaveContext();

This line saves the security context into session.

Here is some  pseudocode to describe the login process.

String username = getUsername(request);
String password = getPassword(request);
boolean loginSuccessful = authenticationManger.authenticate(username, password);
if (loginSuccessful){
     SecurityContext sc = new SecuirtyContext(username);
     SecurityContextHolder.set(sc);
     session.setAttribute("SecurityContextKey", sc);
     redirectToLoginSuccessPage();
}else{
     redirectToLoginFailurePage();
}

Since we already save the security context into the session,  why put it into SecurityContextHolder object again?

Because SecurityContextHolder put the security context into a thread local object. Any plain java class can easily access the security context by calling the static method SecurityContextHolder.get(); However, it is not so easy to grab a session for a plain java class.


Part 3.  There is an implicit object known as 'currentUser'. At what scope (request, session, flow) is this object stored?



The implicit object 'currentUser' can be used at Spring Expression Language in web flow xml file or JSP/JSF page.

e.g. main-flow.xml

<evaluate expression="bookingService.findBookings(currentUser?.name)" result="viewScope.bookings" result-type="dataModel" />

e.g. enterSearchCriteria.xhtml

<p:panel id="bookings" header="Your Hotel Bookings" rendered="#{currentUser!=null}" ...>


We can have a look at ImplicitFlowVariableELResolver.java and FlowVariablePropertyAccessor.java. Both put 'currentUser' as a key into a static map, and the value corresponding to 'currentUser' is an object that is connected to the request context. The request context holds the external context, which holds the security context.

So the attribute 'currentUser' is never explicitly put into any scope.

When the EL resolver sees 'currentUser', it will find it in the security context attached to the request context. There is an easy way to obtain the request context by calling RequestContextHolder.getRequestContext(). The RequestContextHolder, working similarly as the way SecurityContextHolder does, put the request context into a thread local object.

We learn from part 2 and part 3 that there are two ways to obtain the current user programmatically. Thanks to thread local, they are both very straightforward:

RequestContextHolder.getRequestContext().getExternalContext().getCurrentUser();

SecurityContextHolder.getContext().getAuthentication().getPrincipal();




Wednesday, 17 October 2012

Using security namespace as the default namespace

In Spring in Action, 3rd edition, page 227, it says since the security-specific configuration is separated into a separate Spring configuration file, we can change the security namespace to be the primary namespace.

So old version:

<beans xmlns="http://www.springframework.org/schema/beans"
    xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
    xmlns:security="http://www.springframework.org/schema/security"
    xsi:schemaLocation="
        http://www.springframework.org/schema/beans 
        http://www.springframework.org/schema/beans/spring-beans-3.0.xsd
        http://www.springframework.org/schema/security 
        http://www.springframework.org/schema/security/spring-security-3.0.xsd">

    <security:http auto-config="true" use-expressions="true">
        ....
    </security:http>
</beans>

And new version:

<beans:beans xmlns:beans="http://www.springframework.org/schema/beans"
    xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
    xmlns="http://www.springframework.org/schema/security"
    xsi:schemaLocation="
        http://www.springframework.org/schema/beans 
        http://www.springframework.org/schema/beans/spring-beans-3.0.xsd
        http://www.springframework.org/schema/security
        http://www.springframework.org/schema/security/spring-security-3.0.xsd">

    <http auto-config="true" use-expressions="true">
        ....
    </http>

</beans:beans>

One thing I don't understand is why change beans to beans:beans. Why does the top element need a namespace?

Then I try to remove the namespace, the XML complains "Cannot find the declaration of element 'beans'".

Well, after reading the article XML Schema: Understanding Namespaces, I come to understand that the scope of a namespace begins at the element where it is declared. Therefore, in the old version, the element beans is associated with the default namespace (http://www.springframework.org/schema/beans),  in the new version, if we remove the namespace for the beans element, the element will be associated with the security namespace. But of course, the security schema doesn't define a beans element. Hence the complaint of the XML.

Let me rename the prefix of the beans namespace so that it looks more clear.

<bean123:beans xmlns:bean123="http://www.springframework.org/schema/beans"
    xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
    xmlns="http://www.springframework.org/schema/security"
    xsi:schemaLocation="
        http://www.springframework.org/schema/beans 
        http://www.springframework.org/schema/beans/spring-beans-3.0.xsd
        http://www.springframework.org/schema/security
        http://www.springframework.org/schema/security/spring-security-3.0.xsd">

    <http auto-config="true" use-expressions="true">
        ....
    </http>

</bean123:beans>


Tuesday, 16 October 2012

on-entry vs. on-render in Spring Webflow

The action defined in <on-entry> is executed upon entering the state.

The action defined in <on-render> is executed before the view is rendered.

The distinction seems very clear but in practice, say you need to load some information from the database to be displayed, will you use <on-entry> or <on-render>?

Both seem to be legitimate choices. Does it really matter if the data is loaded upon entry or right before the view is rendered?

Well, the answer depends on whether you want the data to be reloaded if the page gets refreshed (including partially refreshed).

Let's see an example.

In web-flow.xml, we have a view-state.

<view-state id="dummy">
    <on-entry>
        <evaluate expression="dummy.onEntry()"></evaluate>
    </on-entry>
    <on-render>
        <evaluate expression="dummy.onRender()"></evaluate>
    </on-render>
</view-state>

Dummy.java

public class Dummy implements Serializable{
    private static final long serialVersionUID = 1L;
 
    public void onRender(){
        System.out.println("On Render");
    }
 
    public void onEntry(){
        System.out.println("On Entry");
    }
}

After the view is rendered on the browser, we will see in the console

On Entry
On Render

Now we refresh the page (i.e. press F5), we will see in the console

On Entry
On Render
On Render

So the on-entry action wouldn't be executed but on-render action still would.

Monday, 15 October 2012

What does ContextLoaderListener do in Spring?

In web.xml, there is usually the following line

<listener>
    <listener-class>org.springframework.web.context.ContextLoaderListener</listener-class>
</listener>

So what does ContextLoaderListener do exactly?

As the Spring API puts it:

Bootstrap listener to start up Spring's root WebApplicationContext. Simply delegates to ContextLoader.

I believe this sentence is too vague to visualize the responsibility this class takes on. So this article tries to break down the duties performed by this class.

In summary, ContextLoaderListener does three main tasks:
  • Create a web application context
  • Read bean definitions from application context xml
  • Instantiate and initialize beans based on the bean definitions and save the beans into the web application context.
I will give detailed explaination one by one.

1. Create a web application context


Upon the start of the deployment, callback method ContextLoaderListener.contextInitialized() will be invoked by the container or application server.

ContextLoaderListener.contextInitialized() will delegate the task of creating web application context to this.contextLoader.

ContextLoaderListener @Line 111

this.contextLoader.initWebApplicationContext(event.getServletContext());

ContextLoader.java @Line 281

this.context = createWebApplicationContext(servletContext);

After this line, the web application context is created. Pretty straightforward, isn't it? If we put the break point at the next line, we will see the class of this.context is XmlWebApplicationConext.

One things needs probing here:

Who gets to pick this particular implementation of the WebApplicationConext?

Spring has provided more than one implementations, including AnnotationConfigWebApplicationContext, GenericWebApplicationContext and StaticWebApplicationContext. How does XmlWebApplicationConext distinguish itself among all the candidates? Can we use our own implementation of WebApplicationContext should the need arises?

We need to step into the createWebApplicationContext() method.

ContextLoader.java @Line 333

Class<?> contextClass = determineContextClass(sc);

Step into this method, ContextLoader.java @Line 398

String contextClassName = servletContext.getInitParameter(CONTEXT_CLASS_PARAM);

contextClassName is null because there is no such init parameter defined in web.xml

Then we have to fall back to the default value, ContextLoader.java @Line 411

contextClassName = defaultStrategies.getProperty(WebApplicationContext.class.getName());

The variable defaultStrategies is a Properties object, which holds a name-value pair:

{org.springframework.web.context.WebApplicationContext=org.springframework.web.context.support.XmlWebApplicationContext}

When is the defaultStrategies  object loaded with this name-value pair and where is this information stored initially?

ContextLoader.java @Line 164-165

ClassPathResource resource = new ClassPathResource(DEFAULT_STRATEGIES_PATH, ContextLoader.class);
defaultStrategies = PropertiesLoaderUtils.loadProperties(resource);

The value of the constant DEFAULT_STRATEGIES_PATH is ContextLoader.properties, which resides at the same package of the same jar file (spring-web.jar) as ContextLoader does.

We can write our own implementation of WebApplicationContext and add a context-param entry to web.xml.

<context-param>
    <param-name>contextClass</param-name>
    <param-value>
        org.springframework.web.context.support.CustomerWebApplicationContext
    </param-value>
</context-param>

2. Read bean definitions from application context xml


In this part, 3 questions will be addressed.

1. Where in ContextLoader does reading bean definitions take place?
2. We know that the context-param entry for contextConfigLocation is optional, so where is the default context config location defined?
3. If we do add a context-param entry for contextConfigLocation in web.xml, where does the default value gets overwritten?

ContextLoader.class @Line 284

configureAndRefreshWebApplicationContext((ConfigurableWebApplicationContext)this.context, servletContext);

Reading bean definitions takes place here. OK, to be fair, lots of things take place here, also including instantiating and initializing beans.

We have to be a bit more specific about the location.

Let's step into this method. ContextLoader.class @Line 385

wac.refresh();

Further step into this method. The actual method body of refresh(); is in the super class of XmlWebApplicationContext: AbstractApplicationConext. Let's jump to Line 437.

ConfigurableListableBeanFactory beanFactory = obtainFreshBeanFactory();

This line does more than merely obtaining a bean factory as the method name suggests. It is also precisely where loading the bean definitions from the application context xml occurs. However, it won't instantiate the beans and put them to web application context yet.

Let's step into this method. AbstractApplicationContext.java @Line 526.

refreshBeanFactory();

Further step into this method. AbstractRefreshableApplicationContext.java @Line 128.

DefaultListableBeanFactory beanFactory = createBeanFactory();

This line creates a bean factory. It's empty yet. If you watch the variable in the debugger, you will find the beanDefinitionMap and beanDefinitionNames are empty. Step down to Line 131.

loadBeanDefinitions(beanFactory);

This line, unsurprisingly,  loads all the bean definitions.

Assume we don't have a context-param entry for contextConfigLocation in web.xml, we will eventually hit AbstractRefreshableConfigApplication.java @Line 100

return (this.configLocations != null ? this.configLocations : getDefaultConfigLocations());

this.configLocation = null because it's not defined in web.xml. So it falls back to default location, which is defined at XmlWebApplicationConext @Line 65.

public static final String DEFAULT_CONFIG_LOCATION = "/WEB-INF/applicationContext.xml";

This tells us if you don't want to specify the contextConfigLocation context parameter in web.xml, you need to place the application context file under WEB-INF, and name it exactly -- applicationContext.xml

Now let's add a context-param entry for contextConfigLocation in web.xml, 

<context-param>
    <param-name>contextConfigLocation</param-name>
    <param-value>
        /WEB-INF/config/web-application-config.xml
    </param-value>
</context-param>

Restart deployment,  ContextLoader.java @Line 380-383

String initParameter = sc.getInitParameter(CONFIG_LOCATION_PARAM);
if (initParameter != null) {
    wac.setConfigLocation(initParameter);
}

The above line sets the path of the application context xml file. 

Let's jump out of the method and go back to AbstractApplicationContext.java @Line 440, and put a break point there.

prepareBeanFactory(beanFactory);

f you expand object beanFactory in the debugger, and inspect the instance variable beanDefinitionMap and beanDefinitionNames, you will find they are populated with values.

3. Instantiate and initialize beans based on the bean definitions and save the beans into the web application context


Now it is time we threw some bean definitions into the application context xml file and see how Spring instantiates them.

Let's start with something very simple. In the applicationContext.xml, add

<bean class="org.springframework.web.servlet.handler.BeanNameUrlHandlerMapping"/>

Now restart the deployment process, put a break point at AbstractApplicationContext.java @Line 465.

finishBeanFactoryInitialization(beanFactory);

All the beans are created here.

Step into the method, and pause at AbstractApplicationConext.java @Line 917

beanFactory.preInstantiateSingletons();

Step into it, put a break point at DefaultListableBeanFactory.java @Line 564

RootBeanDefinition bd = getMergedLocalBeanDefinition(beanName);

Inspect the beanName variable, we are only interested in it when it becomes 'BeanNameUrlHandlerMapping'

Then it hits DefaultListableBeanFactory.java @Line 585

getBean(beanName);

Given the bean name, the bean is instantiated through reflection. I was once puzzled by the fact that the returned bean is not assigned to any variable for further processing (like storing it to web application context). Would the bean be garbage collected?

If you debug into the getBean() method, you will reach AbstractAutowireCapableBeanFactory @Line 507-510.

addSingletonFactory(beanName, new ObjectFactory() {
    public Object getObject() throws BeansException {
        return getEarlyBeanReference(beanName, mbd, bean);
    }
});

Here, the created bean is put into the web application context. Therefore, the bean won't be garbage collected.