通过这一系列的前三部分看到了一次客户端连接在 Tomcat 内部被转换成了请求对象(org.apache.catalina.connector.Request
类的实例),并在该请求对象内部将与本次请求相关的 Host、Context、Wrapper 对象的引用。本文主要分析该请求对象在容器内部流转的经过。
再来看一下 Tomcat 7 内部的组件结构图:

如果你不了解上面这段文字描述中所谓的管道( Pipeline )和阀( Valve )的概念,别急,下面会讲到这个。先从源码层面看下这段文字描述的经过。上面提到的org.apache.catalina.connector.CoyoteAdapter
类的 service 方法:
1 public void service(org.apache.coyote.Request req,
2 org.apache.coyote.Response res)
3 throws Exception {
4
5 Request request = (Request) req.getNote(ADAPTER_NOTES);
6 Response response = (Response) res.getNote(ADAPTER_NOTES);
7
8 if (request == null) {
9
10 // Create objects
11 request = connector.createRequest();
12 request.setCoyoteRequest(req);
13 response = connector.createResponse();
14 response.setCoyoteResponse(res);
15
16 // Link objects
17 request.setResponse(response);
18 response.setRequest(request);
19
20 // Set as notes
21 req.setNote(ADAPTER_NOTES, request);
22 res.setNote(ADAPTER_NOTES, response);
23
24 // Set query string encoding
25 req.getParameters().setQueryStringEncoding
26 (connector.getURIEncoding());
27
28 }
29
30 if (connector.getXpoweredBy()) {
31 response.addHeader("X-Powered-By", POWERED_BY);
32 }
33
34 boolean comet = false;
35 boolean async = false;
36
37 try {
38
39 // Parse and set Catalina and configuration specific
40 // request parameters
41 req.getRequestProcessor().setWorkerThreadName(Thread.currentThread().getName());
42 boolean postParseSuccess = postParseRequest(req, request, res, response);
43 if (postParseSuccess) {
44 //check valves if we support async
45 request.setAsyncSupported(connector.getService().getContainer().getPipeline().isAsyncSupported());
46 // Calling the container
47 connector.getService().getContainer().getPipeline().getFirst().invoke(request, response);
48
49 if (request.isComet()) {
50 if (!response.isClosed() && !response.isError()) {
51 if (request.getAvailable() || (request.getContentLength() > 0 && (!request.isParametersParsed()))) {
52 // Invoke a read event right away if there are available bytes
53 if (event(req, res, SocketStatus.OPEN)) {
54 comet = true;
55 res.action(ActionCode.COMET_BEGIN, null);
56 }
57 } else {
58 comet = true;
59 res.action(ActionCode.COMET_BEGIN, null);
60 }
61 } else {
62 // Clear the filter chain, as otherwise it will not be reset elsewhere
63 // since this is a Comet request
64 request.setFilterChain(null);
65 }
66 }
67
68 }
69 AsyncContextImpl asyncConImpl = (AsyncContextImpl)request.getAsyncContext();
70 if (asyncConImpl != null) {
71 async = true;
72 } else if (!comet) {
73 request.finishRequest();
74 response.finishResponse();
75 if (postParseSuccess &&
76 request.getMappingData().context != null) {
77 // Log only if processing was invoked.
78 // If postParseRequest() failed, it has already logged it.
79 // If context is null this was the start of a comet request
80 // that failed and has already been logged.
81 ((Context) request.getMappingData().context).logAccess(
82 request, response,
83 System.currentTimeMillis() - req.getStartTime(),
84 false);
85 }
86 req.action(ActionCode.POST_REQUEST , null);
87 }
88
89 } catch (IOException e) {
90 // Ignore
91 } finally {
92 req.getRequestProcessor().setWorkerThreadName(null);
93 // Recycle the wrapper request and response
94 if (!comet && !async) {
95 request.recycle();
96 response.recycle();
97 } else {
98 // Clear converters so that the minimum amount of memory
99 // is used by this processor
100 request.clearEncoders();
101 response.clearEncoders();
102 }
103 }
104
105 }
之前主要分析了第 42 行的代码,通过 postParseRequest 方法的调用请求对象内保存了关于本次请求的具体要执行的 Host、Context、Wrapper 组件的引用。
看下第 47 行:
connector.getService().getContainer().getPipeline().getFirst().invoke(request, response);
虽然只有一行,但调用了一堆方法,这里对这些方法逐个分析一下:
connector.getService() 获取的是当前 connector 关联的 Service 组件,默认情况下获得的就是org.apache.catalina.core.StandardService
的对象。其 getContainer 方法获得的是org.apache.catalina.core.StandardEngine
的对象,这段的由来在前面讲 Digester 的解析文章时,createStartDigester 方法中的这段代码:
digester.addRuleSet(new EngineRuleSet("Server/Service/"));
在 EngineRuleSet 类的 addRuleInstances 方法中的这一段代码:
public void addRuleInstances(Digester digester) {
digester.addObjectCreate(prefix + "Engine",
"org.apache.catalina.core.StandardEngine",
"className");
digester.addSetProperties(prefix + "Engine");
digester.addRule(prefix + "Engine",
new LifecycleListenerRule
("org.apache.catalina.startup.EngineConfig",
"engineConfigClass"));
digester.addSetNext(prefix + "Engine",
"setContainer",
"org.apache.catalina.Container");
结合上一段代码可以看出 Tomcat 启动时,如果碰到 server.xml 里的 Server/Service/Engine 节点,先实例化一个org.apache.catalina.core.StandardEngine
对象,在第 11 到 13 行,会以 StandardEngine 对象为入参调用org.apache.catalina.core.StandardService
的 setContainer 方法。
所以上面 connector.getService().getContainer() 方法得到的实际上是 StandardEngine 对象。紧接着的 getPipeline 方法返回的是 StandardEngine 类的父类org.apache.catalina.core.ContainerBase
类的成员变量 pipeline ,看下该类中这个变量的声明代码:
/**
* The Pipeline object with which this Container is associated.
*/
protected Pipeline pipeline = new StandardPipeline(this);
所以 connector.getService().getContainer().getPipeline() 方法返回的是org.apache.catalina.core.StandardPipeline
类的对象,该对象就是本部分开头部分提到的管道( Pipeline )。
下面讲一下 Tomcat 7 中的管道和阀的概念和实现:
所有的管道类都会实现org.apache.catalina.Pipeline
这个接口,看下这个接口中定义的方法:

所有的阀类都会实现org.apache.catalina.Valve
这个接口,看下这个接口中定义的方法:

Tomcat 7 里 Pipeline 的默认实现类是org.apache.catalina.core.StandardPipeline
,其内部有三个成员变量:basic、first、container 。
/**
* The basic Valve (if any) associated with this Pipeline.
*/
protected Valve basic = null;
/**
* The Container with which this Pipeline is associated.
*/
protected Container container = null;
/**
* The first valve associated with this Pipeline.
*/
protected Valve first = null;
看下该类的 addValve 方法:
1 public void addValve(Valve valve) {
2
3 // Validate that we can add this Valve
4 if (valve instanceof Contained)
5 ((Contained) valve).setContainer(this.container);
6
7 // Start the new component if necessary
8 if (getState().isAvailable()) {
9 if (valve instanceof Lifecycle) {
10 try {
11 ((Lifecycle) valve).start();
12 } catch (LifecycleException e) {
13 log.error("StandardPipeline.addValve: start: ", e);
14 }
15 }
16 }
17
18 // Add this Valve to the set associated with this Pipeline
19 if (first == null) {
20 first = valve;
21 valve.setNext(basic);
22 } else {
23 Valve current = first;
24 while (current != null) {
25 if (current.getNext() == basic) {
26 current.setNext(valve);
27 valve.setNext(basic);
28 break;
29 }
30 current = current.getNext();
31 }
32 }
33
34 container.fireContainerEvent(Container.ADD_VALVE_EVENT, valve);
35 }
在第 18 到 32 行,每次给管道添加一个普通阀的时候如果管道内原来没有普通阀则将新添加的阀作为该管道的成员变量 first 的引用,如果管道内已有普通阀,则把新加的阀加到所有普通阀链条末端,并且将该阀的下一个阀的引用设置为管道的基础阀。这样管道内的阀结构如下图所示:

看下 getFirst 方法的实现:
public Valve getFirst() {
if (first != null) {
return first;
}
return basic;
}
如果管道中有普通阀则返回普通阀链条最开始的那个,否则就返回基础阀。
在 Tomcat 7 中所有作为普通阀的类的 invoke 方法实现中都会有这段代码:
getNext().invoke(request, response);
通过这种机制来保证调用管道最开头一端的阀的 invoke 方法,最终会执行完该管道相关的所有阀的 invoke 方法,并且最后执行的必定是该管道基础阀的 invoke 方法。
再回到connector.getService().getContainer().getPipeline().getFirst().invoke(request, response)
这段代码的解释,这里将会执行 StandardEngine 类的管道中的所有阀(包括普通阀和基础阀)的 invoke 方法,并且最后会执行基础阀的 invoke 方法。
Tomcat 7 在默认情况下 Engine 节点没有普通阀,如果想要添加普通阀的话,可以通过在 server.xml 文件的 engine 节点下添加 Valve 节点,参加该文件中的普通阀配置的示例:
<Valve className="org.apache.catalina.authenticator.SingleSignOn" />
那么就来看看 StandardEngine 类的管道中的基础阀的代码实现。先看下该基础阀设置的代码,在org.apache.catalina.core.StandardEngine
对象的构造函数中:
1 public StandardEngine() {
2
3 super();
4 pipeline.setBasic(new StandardEngineValve());
5 /* Set the jmvRoute using the system property jvmRoute */
6 try {
7 setJvmRoute(System.getProperty("jvmRoute"));
8 } catch(Exception ex) {
9 log.warn(sm.getString("standardEngine.jvmRouteFail"));
10 }
11 // By default, the engine will hold the reloading thread
12 backgroundProcessorDelay = 10;
13
14 }
第 4 行即设置基础阀。所以connector.getService().getContainer().getPipeline().getFirst().invoke(request, response)
会执行到 org.apache.catalina.core.StandardEngineValve 类的 invoke 方法:
1 public final void invoke(Request request, Response response)
2 throws IOException, ServletException {
3
4 // Select the Host to be used for this Request
5 Host host = request.getHost();
6 if (host == null) {
7 response.sendError
8 (HttpServletResponse.SC_BAD_REQUEST,
9 sm.getString("standardEngine.noHost",
10 request.getServerName()));
11 return;
12 }
13 if (request.isAsyncSupported()) {
14 request.setAsyncSupported(host.getPipeline().isAsyncSupported());
15 }
16
17 // Ask this Host to process this request
18 host.getPipeline().getFirst().invoke(request, response);
19
20 }
第 5 行,从请求对象中取出该请求关联的 Host(默认情况下是org.apache.catalina.core.StandardHost
对象),请求是如何找到关联的 Host 的请本文之前的部分。经过上述代码分析应该可以看出第 18 行会执行 StandardHost 对象的管道内所有的阀的 invoke 方法。
看下 StandardHost 的构造方法的实现:
public StandardHost() {
super();
pipeline.setBasic(new StandardHostValve());
}
所以看下org.apache.catalina.core.StandardHostValve
类的 invoke 方法:
1 public final void invoke(Request request, Response response)
2 throws IOException, ServletException {
3
4 // Select the Context to be used for this Request
5 Context context = request.getContext();
6 if (context == null) {
7 response.sendError
8 (HttpServletResponse.SC_INTERNAL_SERVER_ERROR,
9 sm.getString("standardHost.noContext"));
10 return;
11 }
12
13 // Bind the context CL to the current thread
14 if( context.getLoader() != null ) {
15 // Not started - it should check for availability first
16 // This should eventually move to Engine, it's generic.
17 if (Globals.IS_SECURITY_ENABLED) {
18 PrivilegedAction pa = new PrivilegedSetTccl(
19 context.getLoader().getClassLoader());
20 AccessController.doPrivileged(pa);
21 } else {
22 Thread.currentThread().setContextClassLoader
23 (context.getLoader().getClassLoader());
24 }
25 }
26 if (request.isAsyncSupported()) {
27 request.setAsyncSupported(context.getPipeline().isAsyncSupported());
28 }
29
30 // Don't fire listeners during async processing
31 // If a request init listener throws an exception, the request is
32 // aborted
33 boolean asyncAtStart = request.isAsync();
34 // An async error page may dispatch to another resource. This flag helps
35 // ensure an infinite error handling loop is not entered
36 boolean errorAtStart = response.isError();
37 if (asyncAtStart || context.fireRequestInitEvent(request)) {
38
39 // Ask this Context to process this request
40 try {
41 context.getPipeline().getFirst().invoke(request, response);
42 } catch (Throwable t) {
43 ExceptionUtils.handleThrowable(t);
44 if (errorAtStart) {
45 container.getLogger().error("Exception Processing " +
46 request.getRequestURI(), t);
47 } else {
48 request.setAttribute(RequestDispatcher.ERROR_EXCEPTION, t);
49 throwable(request, response, t);
50 }
51 }
52
53 // If the request was async at the start and an error occurred then
54 // the async error handling will kick-in and that will fire the
55 // request destroyed event *after* the error handling has taken
56 // place
57 if (!(request.isAsync() || (asyncAtStart &&
58 request.getAttribute(
59 RequestDispatcher.ERROR_EXCEPTION) != null))) {
60 // Protect against NPEs if context was destroyed during a
61 // long running request.
62 if (context.getState().isAvailable()) {
63 if (!errorAtStart) {
64 // Error page processing
65 response.setSuspended(false);
66
67 Throwable t = (Throwable) request.getAttribute(
68 RequestDispatcher.ERROR_EXCEPTION);
69
70 if (t != null) {
71 throwable(request, response, t);
72 } else {
73 status(request, response);
74 }
75 }
76
77 context.fireRequestDestroyEvent(request);
78 }
79 }
80 }
81
82 // Access a session (if present) to update last accessed time, based on a
83 // strict interpretation of the specification
84 if (ACCESS_SESSION) {
85 request.getSession(false);
86 }
87
88 // Restore the context classloader
89 if (Globals.IS_SECURITY_ENABLED) {
90 PrivilegedAction pa = new PrivilegedSetTccl(
91 StandardHostValve.class.getClassLoader());
92 AccessController.doPrivileged(pa);
93 } else {
94 Thread.currentThread().setContextClassLoader
95 (StandardHostValve.class.getClassLoader());
96 }
97 }
第 41 行,会调用该请求相关的 Context 的管道内所有的阀的 invoke 方法,默认情况下 Context 是org.apache.catalina.core.StandardContext
类的对象,其构造方法中设置了管道的基础阀:
public StandardContext() {
super();
pipeline.setBasic(new StandardContextValve());
broadcaster = new NotificationBroadcasterSupport();
// Set defaults
if (!Globals.STRICT_SERVLET_COMPLIANCE) {
// Strict servlet compliance requires all extension mapped servlets
// to be checked against welcome files
resourceOnlyServlets.add("jsp");
}
}
看下其基础阀的 invoke 方法代码:
1 public final void invoke(Request request, Response response)
2 throws IOException, ServletException {
3
4 // Disallow any direct access to resources under WEB-INF or META-INF
5 MessageBytes requestPathMB = request.getRequestPathMB();
6 if ((requestPathMB.startsWithIgnoreCase("/META-INF/", 0))
7 || (requestPathMB.equalsIgnoreCase("/META-INF"))
8 || (requestPathMB.startsWithIgnoreCase("/WEB-INF/", 0))
9 || (requestPathMB.equalsIgnoreCase("/WEB-INF"))) {
10 response.sendError(HttpServletResponse.SC_NOT_FOUND);
11 return;
12 }
13
14 // Select the Wrapper to be used for this Request
15 Wrapper wrapper = request.getWrapper();
16 if (wrapper == null || wrapper.isUnavailable()) {
17 response.sendError(HttpServletResponse.SC_NOT_FOUND);
18 return;
19 }
20
21 // Acknowledge the request
22 try {
23 response.sendAcknowledgement();
24 } catch (IOException ioe) {
25 container.getLogger().error(sm.getString(
26 "standardContextValve.acknowledgeException"), ioe);
27 request.setAttribute(RequestDispatcher.ERROR_EXCEPTION, ioe);
28 response.sendError(HttpServletResponse.SC_INTERNAL_SERVER_ERROR);
29 return;
30 }
31
32 if (request.isAsyncSupported()) {
33 request.setAsyncSupported(wrapper.getPipeline().isAsyncSupported());
34 }
35 wrapper.getPipeline().getFirst().invoke(request, response);
36 }
最后的第 35 行,从请求中取出关联的 wrapper 对象后调用其管道内所有阀的 invoke 方法。wrapper 对象默认是org.apache.catalina.core.StandardWrapper
类的实例,同样是在该类的构造方法中设置的基础阀:
public StandardWrapper() {
super();
swValve=new StandardWrapperValve();
pipeline.setBasic(swValve);
broadcaster = new NotificationBroadcasterSupport();
}
有兴趣可以看下基础阀org.apache.catalina.core.StandardWrapperValve
的 invoke 方法,在这里最终会调用请求的 url 所匹配的 Servlet 相关过滤器( filter )的 doFilter 方法及该 Servlet 的 service 方法(这段实现都是在过滤器链 ApplicationFilterChain 类的 doFilter 方法中),这里不再贴出代码分析。
这里可以看出容器内的 Engine、Host、Context、Wrapper 容器组件的实现的共通点:
-
这些组件内部都有一个成员变量 pipeline ,因为它们都是从
org.apache.catalina.core.ContainerBase
类继承来的,pipeline 就定义在这个类中。所以每一个容器内部都关联了一个管道。 -
都是在类的构造方法中设置管道内的基础阀。
-
所有的基础阀的实现最后都会调用其下一级容器(直接从请求中获取下一级容器对象的引用,在上面的分析中已经设置了与该请求相关的各级具体组件的引用)的 getPipeline().getFirst().invoke() 方法,直到 Wrapper 组件。因为 Wrapper 是对一个 Servlet 的包装,所以它的基础阀内部调用的过滤器链的 doFilter 方法和 Servlet 的 service 方法。
正是通过这种管道和阀的机制及上述的 3 点前提,使得请求可以从连接器内一步一步流转到具体 Servlet 的 service 方法中。这样,关于Tomcat 7 中一次请求的分析介绍完毕,从中可以看出在浏览器发出一次 Socket 连接请求之后 Tomcat 容器内运转处理的大致流程。