Netty中的ChannelPipeline源码分析

ChannelPipeline在Netty中是用来处理请求的责任链，默认实现是DefaultChannelPipeline，其构造方法如下：

private final Channel channel;
private final ChannelFuture succeededFuture;
private final VoidChannelPromise voidPromise;
final AbstractChannelHandlerContext head;
final AbstractChannelHandlerContext tail;

protected DefaultChannelPipeline(Channel channel) {
    this.channel = (Channel)ObjectUtil.checkNotNull(channel, "channel");
    this.succeededFuture = new SucceededChannelFuture(channel, (EventExecutor)null);
    this.voidPromise = new VoidChannelPromise(channel, true);
    this.tail = new DefaultChannelPipeline.TailContext(this);
    this.head = new DefaultChannelPipeline.HeadContext(this);
    this.head.next = this.tail;
    this.tail.prev = this.head;
}

ChannelPipeline和Channel是一一对应关系，一个Channel绑定一条ChannelPipeline责任链
succeededFuture 和voidPromise用来处理异步操作
AbstractChannelHandlerContext 是持有请求的上下文对象，其和ChannelHandler是对应关系（在使用Sharable注解的情况下，不同的AbstractChannelHandlerContext 还可以对应同一个ChannelHandler），ChannelPipeline责任链
处理的就AbstractChannelHandlerContext ，再将最后的AbstractChannelHandlerContext 交给ChannelHandler去做正真的逻辑处理

AbstractChannelHandlerContext构造方法如下：

private final String name;
private final DefaultChannelPipeline pipeline;
final EventExecutor executor;
private final boolean inbound;
private final boolean outbound;
private final boolean ordered;
volatile AbstractChannelHandlerContext next;
volatile AbstractChannelHandlerContext prev;

AbstractChannelHandlerContext(DefaultChannelPipeline pipeline, EventExecutor executor, String name, boolean inbound, boolean outbound) {
    this.name = (String)ObjectUtil.checkNotNull(name, "name");
    this.pipeline = pipeline;
    this.executor = executor;
    this.inbound = inbound;
    this.outbound = outbound;
    this.ordered = executor == null || executor instanceof OrderedEventExecutor;
}

name是AbstractChannelHandlerContext的名称，pipeline就是上面说的ChannelPipeline；executor是用来进行异步操作的，默认使用的是在前面博客中说过的NioEventLoop  （Netty中NioEventLoopGroup的创建源码分析）

inbound 和outbound 代表两种请求处理方式，对应Netty中的I/O操作，若是inbound则处理Input操作，由ChannelPipeline从head 开始向后遍历链表，并且只处理ChannelInboundHandler类型的AbstractChannelHandlerContext；若是outbound 则处理Output操作，由ChannelPipeline从tail开始向前遍历链表，并且只处理ChannelOutboundHandler类型的AbstractChannelHandlerContext；
ordered 是判断是否需要提供executor。

由next和prev成员可以知道，ChannelPipeline维护的是一条AbstractChannelHandlerContext的双向链表
其头节点head和尾节点tail分别默认初始化了HeadContext和TailContext

HeadContext的构造：

final class HeadContext extends AbstractChannelHandlerContext implements ChannelOutboundHandler, ChannelInboundHandler {
    private final Unsafe unsafe;
    
    HeadContext(DefaultChannelPipeline pipeline) {
    super(pipeline, (EventExecutor)null, DefaultChannelPipeline.HEAD_NAME, false, true);
    this.unsafe = pipeline.channel().unsafe();
    this.setAddComplete();
    }
}

其中setAddComplete是由AbstractChannelHandlerContext实现的：

final void setAddComplete() {
    int oldState;
    do {
        oldState = this.handlerState;
    } while(oldState != 3 && !HANDLER_STATE_UPDATER.compareAndSet(this, oldState, 2));

}

handlerState表示AbstractChannelHandlerContext对应的ChannelHandler的状态，有一下几种：

private static final int ADD_PENDING = 1;
private static final int ADD_COMPLETE = 2;
private static final int REMOVE_COMPLETE = 3;
private static final int INIT = 0;
private volatile int handlerState = 0;    

handlerState初始化默认是INIT状态。

HANDLER_STATE_UPDATER是一个原子更新器：

private static final AtomicIntegerFieldUpdater<AbstractChannelHandlerContext> HANDLER_STATE_UPDATER = AtomicIntegerFieldUpdater.newUpdater(AbstractChannelHandlerContext.class, "handlerState");

所以setAddComplete方法，就是通过CAS操作，将handlerState状态更新为ADD_COMPLETE

TailContext的构造：

final class TailContext extends AbstractChannelHandlerContext implements ChannelInboundHandler {
    TailContext(DefaultChannelPipeline pipeline) {
        super(pipeline, (EventExecutor)null, DefaultChannelPipeline.TAIL_NAME, true, false);
        this.setAddComplete();
    }
}

和HeadContext一样，将handlerState状态更新为ADD_COMPLETE

结合官方给出的ChannelPipeline的图示更容易理解：

                                             I/O Request
                                        via Channel or
                                    ChannelHandlerContext
                                                  |
+---------------------------------------------------+---------------+
|                           ChannelPipeline         |               |
|                                                  |/              |
|    +---------------------+            +-----------+----------+    |
|    | Inbound Handler  N  |            | Outbound Handler  1  |    |
|    +----------+----------+            +-----------+----------+    |
|              /|                                  |               |
|               |                                  |/              |
|    +----------+----------+            +-----------+----------+    |
|    | Inbound Handler N-1 |            | Outbound Handler  2  |    |
|    +----------+----------+            +-----------+----------+    |
|              /|                                  .               |
|               .                                   .               |
| ChannelHandlerContext.fireIN_EVT() ChannelHandlerContext.OUT_EVT()|
|        [ method call]                       [method call]         |
|               .                                   .               |
|               .                                  |/              |
|    +----------+----------+            +-----------+----------+    |
|    | Inbound Handler  2  |            | Outbound Handler M-1 |    |
|    +----------+----------+            +-----------+----------+    |
|              /|                                  |               |
|               |                                  |/              |
|    +----------+----------+            +-----------+----------+    |
|    | Inbound Handler  1  |            | Outbound Handler  M  |    |
|    +----------+----------+            +-----------+----------+    |
|              /|                                  |               |
+---------------+-----------------------------------+---------------+
              |                                  |/
+---------------+-----------------------------------+---------------+
|               |                                   |               |
|       [ Socket.read() ]                    [ Socket.write() ]     |
|                                                                   |
|  Netty Internal I/O Threads (Transport Implementation)            |
+-------------------------------------------------------------------+

下面对一些主要方法分析：
addFirst方法，有如下几种重载：

public final ChannelPipeline addFirst(ChannelHandler handler) {
    return this.addFirst((String)null, (ChannelHandler)handler);
}

public final ChannelPipeline addFirst(String name, ChannelHandler handler) {
    return this.addFirst((EventExecutorGroup)null, name, handler);
}

public final ChannelPipeline addFirst(ChannelHandler... handlers) {
    return this.addFirst((EventExecutorGroup)null, (ChannelHandler[])handlers);
}

public final ChannelPipeline addFirst(EventExecutorGroup executor, ChannelHandler... handlers) {
    if (handlers == null) {
        throw new NullPointerException("handlers");
    } else if (handlers.length != 0 && handlers[0] != null) {
        int size;
        for(size = 1; size < handlers.length && handlers[size] != null; ++size) {
            ;
        }

        for(int i = size - 1; i >= 0; --i) {
            ChannelHandler h = handlers[i];
            this.addFirst(executor, (String)null, h);
        }

        return this;
    } else {
        return this;
    }
}

public final ChannelPipeline addFirst(EventExecutorGroup group, String name, ChannelHandler handler) {
    final AbstractChannelHandlerContext newCtx;
    synchronized(this) {
        checkMultiplicity(handler);
        name = this.filterName(name, handler);
        newCtx = this.newContext(group, name, handler);
        this.addFirst0(newCtx);
        if (!this.registered) {
            newCtx.setAddPending();
            this.callHandlerCallbackLater(newCtx, true);
            return this;
        }

        EventExecutor executor = newCtx.executor();
        if (!executor.inEventLoop()) {
            newCtx.setAddPending();
            executor.execute(new Runnable() {
                public void run() {
                    DefaultChannelPipeline.this.callHandlerAdded0(newCtx);
                }
            });
            return this;
        }
    }

    this.callHandlerAdded0(newCtx);
    return this;
}

前面几种都是间接调用的第四种没什么好说的，直接看第四种addFirst
首先调用checkMultiplicity，检查ChannelHandlerAdapter在不共享的情况下是否重复：

private static void checkMultiplicity(ChannelHandler handler) {
    if (handler instanceof ChannelHandlerAdapter) {
        ChannelHandlerAdapter h = (ChannelHandlerAdapter)handler;
        if (!h.isSharable() && h.added) {
            throw new ChannelPipelineException(h.getClass().getName() + " is not a @Sharable handler, so can't be added or removed multiple times.");
        }

        h.added = true;
    }

}

isSharable方法：

public boolean isSharable() {
    Class<?> clazz = this.getClass();
    Map<Class<?>, Boolean> cache = InternalThreadLocalMap.get().handlerSharableCache();
    Boolean sharable = (Boolean)cache.get(clazz);
    if (sharable == null) {
        sharable = clazz.isAnnotationPresent(Sharable.class);
        cache.put(clazz, sharable);
    }

    return sharable;
}

首先尝试从当前线程的InternalThreadLocalMap中获取handlerSharableCache，（InternalThreadLocalMap是在Netty中使用高效的FastThreadLocal替代JDK的ThreadLocal使用的 Netty中FastThreadLocal源码分析）
InternalThreadLocalMap的handlerSharableCache方法：

public Map<Class<?>, Boolean> handlerSharableCache() {
    Map<Class<?>, Boolean> cache = this.handlerSharableCache;
    if (cache == null) {
        this.handlerSharableCache = (Map)(cache = new WeakHashMap(4));
    }

    return (Map)cache;
}

当当前线程的InternalThreadLocalMap中没有handlerSharableCache时，直接创建一个大小为4的WeakHashMap弱引用Map；

根据clazz从map中get，若是没有，需要检测当前clazz是否有Sharable注解，添加了Sharable注解的ChannelHandlerAdapter可以在不同Channel中共享使用一个单例，前提是确保线程安全；
之后会将该clazz以及是否实现Sharable注解的情况添加在cache缓存中；
其中ChannelHandler的added是用来标识是否添加过；

回到addFirst方法：
checkMultiplicity成功结束后，调用filterName方法，给当前要产生的AbstractChannelHandlerContext对象产生一个名称，
然后调用newContext方法，产生AbstractChannelHandlerContext对象：

private AbstractChannelHandlerContext newContext(EventExecutorGroup group, String name, ChannelHandler handler) {
    return new DefaultChannelHandlerContext(this, this.childExecutor(group), name, handler);
}

这里实际上产生了一个DefaultChannelHandlerContext对象：

final class DefaultChannelHandlerContext extends AbstractChannelHandlerContext {
    private final ChannelHandler handler;

    DefaultChannelHandlerContext(DefaultChannelPipeline pipeline, EventExecutor executor, String name, ChannelHandler handler) {
        super(pipeline, executor, name, isInbound(handler), isOutbound(handler));
        if (handler == null) {
            throw new NullPointerException("handler");
        } else {
            this.handler = handler;
        }
    }

    public ChannelHandler handler() {
        return this.handler;
    }

    private static boolean isInbound(ChannelHandler handler) {
        return handler instanceof ChannelInboundHandler;
    }

    private static boolean isOutbound(ChannelHandler handler) {
        return handler instanceof ChannelOutboundHandler;
    }
}

可以看到DefaultChannelHandlerContext 仅仅是将AbstractChannelHandlerContext和ChannelHandler封装了

在产生了DefaultChannelHandlerContext 对象后，调用addFirst0方法：

private void addFirst0(AbstractChannelHandlerContext newCtx) {
    AbstractChannelHandlerContext nextCtx = this.head.next;
    newCtx.prev = this.head;
    newCtx.next = nextCtx;
    this.head.next = newCtx;
    nextCtx.prev = newCtx;
}

这里就是一个简单的双向链表的操作，将newCtx节点插入到了head后面

然后判断registered成员的状态：

private boolean registered;

在初始化时是false

registered若是false，首先调用AbstractChannelHandlerContext的setAddPending方法：

final void setAddPending() {
   boolean updated = HANDLER_STATE_UPDATER.compareAndSet(this, 0, 1);

    assert updated;

}

和前面说过的setAddComplete方法同理，通过CAS操作，将handlerState状态设置为ADD_PENDING
接着调用callHandlerCallbackLater方法：

private void callHandlerCallbackLater(AbstractChannelHandlerContext ctx, boolean added) {
    assert !this.registered;

    DefaultChannelPipeline.PendingHandlerCallback task = added ? new DefaultChannelPipeline.PendingHandlerAddedTask(ctx) : new DefaultChannelPipeline.PendingHandlerRemovedTask(ctx);
    DefaultChannelPipeline.PendingHandlerCallback pending = this.pendingHandlerCallbackHead;
    if (pending == null) {
        this.pendingHandlerCallbackHead = (DefaultChannelPipeline.PendingHandlerCallback)task;
    } else {
        while(pending.next != null) {
            pending = pending.next;
        }

        pending.next = (DefaultChannelPipeline.PendingHandlerCallback)task;
    }

}

首先断言判断registered可能存在的多线程改变，然后根据added判断产生何种类型的PendingHandlerCallback
PendingHandlerCallback是用来处理ChannelHandler的两种回调，定义如下：

private abstract static class PendingHandlerCallback implements Runnable {
    final AbstractChannelHandlerContext ctx;
    DefaultChannelPipeline.PendingHandlerCallback next;

    PendingHandlerCallback(AbstractChannelHandlerContext ctx) {
        this.ctx = ctx;
    }

    abstract void execute();
}

PendingHandlerAddedTask定义如下：

private final class PendingHandlerAddedTask extends DefaultChannelPipeline.PendingHandlerCallback {
    PendingHandlerAddedTask(AbstractChannelHandlerContext ctx) {
        super(ctx);
    }

    public void run() {
        DefaultChannelPipeline.this.callHandlerAdded0(this.ctx);
    }

    void execute() {
        EventExecutor executor = this.ctx.executor();
        if (executor.inEventLoop()) {
            DefaultChannelPipeline.this.callHandlerAdded0(this.ctx);
        } else {
            try {
                executor.execute(this);
            } catch (RejectedExecutionException var3) {
                if (DefaultChannelPipeline.logger.isWarnEnabled()) {
                    DefaultChannelPipeline.logger.warn("Can't invoke handlerAdded() as the EventExecutor {} rejected it, removing handler {}.", new Object[]{executor, this.ctx.name(), var3});
                }

                DefaultChannelPipeline.remove0(this.ctx);
                this.ctx.setRemoved();
            }
        }

    }
}

除去异常处理，无论是在execute方法还是在run方法中，主要核心是异步执行callHandlerAdded0方法：

private void callHandlerAdded0(AbstractChannelHandlerContext ctx) {
    try {
        ctx.setAddComplete();
        ctx.handler().handlerAdded(ctx);
    } catch (Throwable var10) {
        boolean removed = false;

        try {
            remove0(ctx);

            try {
                ctx.handler().handlerRemoved(ctx);
            } finally {
                ctx.setRemoved();
            }

            removed = true;
        } catch (Throwable var9) {
            if (logger.isWarnEnabled()) {
                logger.warn("Failed to remove a handler: " + ctx.name(), var9);
            }
        }

        if (removed) {
            this.fireExceptionCaught(new ChannelPipelineException(ctx.handler().getClass().getName() + ".handlerAdded() has thrown an exception; removed.", var10));
        } else {
            this.fireExceptionCaught(new ChannelPipelineException(ctx.handler().getClass().getName() + ".handlerAdded() has thrown an exception; also failed to remove.", var10));
        }
    }

}

除去异常处理，主要核心就两行代码，首先通过setAddComplete方法，设置handlerState状态为ADD_COMPLETE，然后回调ChannelHandler的handlerAdded方法，这个handlerAdded方法就很熟悉了，在使用Netty处理业务逻辑时，会覆盖这个方法。

PendingHandlerRemovedTask定义如下：

private final class PendingHandlerRemovedTask extends DefaultChannelPipeline.PendingHandlerCallback {
    PendingHandlerRemovedTask(AbstractChannelHandlerContext ctx) {
        super(ctx);
    }

    public void run() {
        DefaultChannelPipeline.this.callHandlerRemoved0(this.ctx);
    }

    void execute() {
        EventExecutor executor = this.ctx.executor();
        if (executor.inEventLoop()) {
            DefaultChannelPipeline.this.callHandlerRemoved0(this.ctx);
        } else {
            try {
                executor.execute(this);
            } catch (RejectedExecutionException var3) {
                if (DefaultChannelPipeline.logger.isWarnEnabled()) {
                    DefaultChannelPipeline.logger.warn("Can't invoke handlerRemoved() as the EventExecutor {} rejected it, removing handler {}.", new Object[]{executor, this.ctx.name(), var3});
                }

                this.ctx.setRemoved();
            }
        }

    }
}

和PendingHandlerAddedTask一样，主要还是异步调用callHandlerRemoved0方法：

private void callHandlerRemoved0(AbstractChannelHandlerContext ctx) {
    try {
        try {
            ctx.handler().handlerRemoved(ctx);
        } finally {
            ctx.setRemoved();
        }
    } catch (Throwable var6) {
        this.fireExceptionCaught(new ChannelPipelineException(ctx.handler().getClass().getName() + ".handlerRemoved() has thrown an exception.", var6));
    }

}

首先直接回调ChannelHandler的handlerRemoved方法，然后通过setRemoved方法将handlerState状态设置为REMOVE_COMPLETE

回到callHandlerCallbackLater，其中成员pendingHandlerCallbackHead定义：

private DefaultChannelPipeline.PendingHandlerCallback pendingHandlerCallbackHead;

结合PendingHandlerCallback 可知，这个pendingHandlerCallbackHead是 DefaultChannelPipeline存储的一条PendingHandlerCallback单链表，用来处理ChannelHandler的handlerAdded和handlerRemoved的回调，在add的这些方法里调用callHandlerCallbackLater时，added参数都为true，所以add的ChannelHandler只向pendingHandlerCallbackHead添加了handlerAdded的回调。

回到addFirst方法，若是registered为true，先获取EventExecutor，判断是否处于轮询中，若不是，则需要开启轮询线程直接异步执行callHandlerAdded0方法，若处于轮询，由于ChannelPipeline的调用是发生在轮询时的，所以还是直接异步执行callHandlerAdded0方法。

addFirst方法到此结束，再来看addLast方法，同样有好几种重载：

public final ChannelPipeline addLast(ChannelHandler handler) {
    return this.addLast((String)null, (ChannelHandler)handler);
}

public final ChannelPipeline addLast(String name, ChannelHandler handler) {
    return this.addLast((EventExecutorGroup)null, name, handler);
}

public final ChannelPipeline addLast(ChannelHandler... handlers) {
    return this.addLast((EventExecutorGroup)null, (ChannelHandler[])handlers);
}

public final ChannelPipeline addLast(EventExecutorGroup executor, ChannelHandler... handlers) {
    if (handlers == null) {
        throw new NullPointerException("handlers");
    } else {
        ChannelHandler[] var3 = handlers;
        int var4 = handlers.length;

        for(int var5 = 0; var5 < var4; ++var5) {
            ChannelHandler h = var3[var5];
            if (h == null) {
                break;
            }

            this.addLast(executor, (String)null, h);
        }

        return this;
    }
}

public final ChannelPipeline addLast(EventExecutorGroup group, String name, ChannelHandler handler) {
    final AbstractChannelHandlerContext newCtx;
    synchronized(this) {
        checkMultiplicity(handler);
        newCtx = this.newContext(group, this.filterName(name, handler), handler);
        this.addLast0(newCtx);
        if (!this.registered) {
            newCtx.setAddPending();
            this.callHandlerCallbackLater(newCtx, true);
            return this;
        }

        EventExecutor executor = newCtx.executor();
        if (!executor.inEventLoop()) {
            newCtx.setAddPending();
            executor.execute(new Runnable() {
                public void run() {
                    DefaultChannelPipeline.this.callHandlerAdded0(newCtx);
                }
            });
            return this;
        }
    }

    this.callHandlerAdded0(newCtx);
    return this;
}

还是间接调用最后一种：
对比addFirst来看，只有addLast0不一样：

private void addLast0(AbstractChannelHandlerContext newCtx) {
    AbstractChannelHandlerContext prev = this.tail.prev;
    newCtx.prev = prev;
    newCtx.next = this.tail;
    prev.next = newCtx;
    this.tail.prev = newCtx;
}

还是非常简单的双向链表基本操作，只不过这次，是将AbstractChannelHandlerContext插入到了tail之前
还有两个，addBefore和addAfter方法，和上述方法类似，就不再累赘

接下来看看ChannelPipeline是如何完成请求的传递的：
invokeHandlerAddedIfNeeded方法：

final void invokeHandlerAddedIfNeeded() {
    assert this.channel.eventLoop().inEventLoop();

    if (this.firstRegistration) {
        this.firstRegistration = false;
        this.callHandlerAddedForAllHandlers();
    }

}

断言判断是否处于轮询线程（ChannelPipeline处理请求都是在轮询线程中，都需要异步处理）
其中firstRegistration成员在DefaultChannelPipeline初始化时为true：

private boolean firstRegistration = true;

此时设置为false，表示第一次调用，以后都不再调用后面的callHandlerAddedForAllHandlers：

private void callHandlerAddedForAllHandlers() {
    DefaultChannelPipeline.PendingHandlerCallback pendingHandlerCallbackHead;
    synchronized(this) {
        assert !this.registered;

        this.registered = true;
        pendingHandlerCallbackHead = this.pendingHandlerCallbackHead;
        this.pendingHandlerCallbackHead = null;
    }

    for(DefaultChannelPipeline.PendingHandlerCallback task = pendingHandlerCallbackHead; task != null; task = task.next) {
        task.execute();
    }

}

刚才说过registered初始是false，在这里判断符合，之后就令其为true，然后获取处理ChannelHandler的回调链表pendingHandlerCallbackHead，并且将pendingHandlerCallbackHead置为null
然后遍历这个单链表，处理ChannelHandler的handlerAdded和handlerRemoved的回调

fireChannelRegistered方法，当Channel完成了向Selector的注册后，会由channel的Unsafe进行回调，异步处理：

public final ChannelPipeline fireChannelRegistered() {
    AbstractChannelHandlerContext.invokeChannelRegistered(this.head);
    return this;
}

实际上的处理由AbstractChannelHandlerContext的静态方法invokeChannelRegistered完成，这里传递的参数head就是DefaultChannelPipeline初始化时创建的HeadContext：

static void invokeChannelRegistered(final AbstractChannelHandlerContext next) {
    EventExecutor executor = next.executor();
    if (executor.inEventLoop()) {
        next.invokeChannelRegistered();
    } else {
        executor.execute(new Runnable() {
            public void run() {
                next.invokeChannelRegistered();
            }
        });
    }

}

可以看到实际上是异步执行head对象的invokeChannelRegistered方法：

private void invokeChannelRegistered() {
    if (this.invokeHandler()) {
        try {
            ((ChannelInboundHandler)this.handler()).channelRegistered(this);
        } catch (Throwable var2) {
            this.notifyHandlerException(var2);
        }
    } else {
        this.fireChannelRegistered();
    }

}

其中invokeHandler是用来判断当前的handlerState状态：

private boolean invokeHandler() {
    int handlerState = this.handlerState;
    return handlerState == 2 || !this.ordered && handlerState == 1;
}

若是当前handlerState状态为ADD_COMPLETE，或者不需要提供EventExecutor并且状态为ADD_PENDING时返回true，否则返回false
在成立的情况下，调用ChannelInboundHandler的channelRegistered方法，由于当前是head，所以由HeadContext实现了：

public void channelRegistered(ChannelHandlerContext ctx) throws Exception {
    DefaultChannelPipeline.this.invokeHandlerAddedIfNeeded();
    ctx.fireChannelRegistered();
}

首先调用invokeHandlerAddedIfNeeded，处理ChannelHandler的handlerAdded和handlerRemoved的回调
然后调用ctx的fireChannelRegistered方法：

public ChannelHandlerContext fireChannelRegistered() {
    invokeChannelRegistered(this.findContextInbound());
    return this;
}

findContextInbound方法，用来找出下一个ChannelInboundInvoker：

private AbstractChannelHandlerContext findContextInbound() {
    AbstractChannelHandlerContext ctx = this;

    do {
        ctx = ctx.next;
    } while(!ctx.inbound);

    return ctx;
}

从当前节点向后遍历，inbound之前说过，该方法就是找到下一个ChannelInboundInvoker的类型的AbstractChannelHandlerContext，然后调用静态方法invokeChannelRegistered，重复上述操作，若是在ChannelInboundHandler中没有重写channelRegistered方法，会一直执直到完所有ChannelHandler的channelRegistered方法。
ChannelInboundHandlerAdapter中的默认channelRegistered方法：

public void channelRegistered(ChannelHandlerContext ctx) throws Exception {
    ctx.fireChannelRegistered();
}

比HeadContext中的实现还简单，直接调用fireChannelRegistered向后传递

fireChannelRead方法，是在Selector轮循到读事件就绪，会由channel的Unsafe进行回调，异步处理：

public final ChannelPipeline fireChannelRead(Object msg) {
    AbstractChannelHandlerContext.invokeChannelRead(this.head, msg);
    return this;
}

还是从head开始调用AbstractChannelHandlerContext的静态方法invokeChannelRead：

static void invokeChannelRead(final AbstractChannelHandlerContext next, Object msg) {
    final Object m = next.pipeline.touch(ObjectUtil.checkNotNull(msg, "msg"), next);
    EventExecutor executor = next.executor();
    if (executor.inEventLoop()) {
        next.invokeChannelRead(m);
    } else {
        executor.execute(new Runnable() {
            public void run() {
                next.invokeChannelRead(m);
            }
        });
    }

}

和上面一个逻辑异步调用AbstractChannelHandlerContext对象的invokeChannelRead方法：

private void invokeChannelRead(Object msg) {
    if (this.invokeHandler()) {
        try {
            ((ChannelInboundHandler)this.handler()).channelRead(this, msg);
        } catch (Throwable var3) {
            this.notifyHandlerException(var3);
        }
    } else {
        this.fireChannelRead(msg);
    }

}

这里也和上面一样，调用了HeadContext的channelRead方法：

public void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {
    ctx.fireChannelRead(msg);
}

这里直接不处理，调用ChannelHandlerContext 的fireChannelRead方法：

public ChannelHandlerContext fireChannelRead(Object msg) {
    invokeChannelRead(this.findContextInbound(), msg);
    return this;
}

和之前注册一样，选择下一个ChannelInboundHandler，重复执行上述操作。

再来看到writeAndFlush方法，和上面的就不太一样，这个发生在轮询前，用户通过channel来间接调用，在AbstractChannel中实现：

public ChannelFuture writeAndFlush(Object msg) {
    return this.pipeline.writeAndFlush(msg);
}

实际上直接调用了DefaultChannelPipeline的writeAndFlush方法：

public final ChannelFuture writeAndFlush(Object msg) {
    return this.tail.writeAndFlush(msg);
}

这里又有些不一样了，调用了tail的writeAndFlush方法，即TailContext的writeAndFlush，在AbstractChannelHandlerContext中实现：

public ChannelFuture writeAndFlush(Object msg) {
    return this.writeAndFlush(msg, this.newPromise());
}

newPromise产生了一个ChannelPromise，用来处理异步事件的；实际上调用了writeAndFlush的重载：

public ChannelFuture writeAndFlush(Object msg, ChannelPromise promise) {
    if (msg == null) {
        throw new NullPointerException("msg");
    } else if (this.isNotValidPromise(promise, true)) {
        ReferenceCountUtil.release(msg);
        return promise;
    } else {
        this.write(msg, true, promise);
        return promise;
    }
}

继续调用write方法：

private void write(Object msg, boolean flush, ChannelPromise promise) {
    AbstractChannelHandlerContext next = this.findContextOutbound();
    Object m = this.pipeline.touch(msg, next);
    EventExecutor executor = next.executor();
    if (executor.inEventLoop()) {
        if (flush) {
            next.invokeWriteAndFlush(m, promise);
        } else {
            next.invokeWrite(m, promise);
        }
    } else {
        Object task;
        if (flush) {
            task = AbstractChannelHandlerContext.WriteAndFlushTask.newInstance(next, m, promise);
        } else {
            task = AbstractChannelHandlerContext.WriteTask.newInstance(next, m, promise);
        }

        safeExecute(executor, (Runnable)task, promise, m);
    }

}

还是很相似，只不过先调用findContextOutbound找到下一个ChannelOutboundInvoker类型的ChannelHandlerContext，而且这里是从尾部往前遍历的，这样来看前面所给的图是没有任何问题的
在找到ChannelOutboundInvoker后，调用invokeWriteAndFlush或者invokeWrite方法：
invokeWriteAndFlush方法：

private void invokeWriteAndFlush(Object msg, ChannelPromise promise) {
    if (this.invokeHandler()) {
        this.invokeWrite0(msg, promise);
        this.invokeFlush0();
    } else {
        this.writeAndFlush(msg, promise);
    }

}

private void invokeWrite0(Object msg, ChannelPromise promise) {
    try {
        ((ChannelOutboundHandler)this.handler()).write(this, msg, promise);
    } catch (Throwable var4) {
        notifyOutboundHandlerException(var4, promise);
    }

}

private void invokeFlush0() {
    try {
        ((ChannelOutboundHandler)this.handler()).flush(this);
    } catch (Throwable var2) {
        this.notifyHandlerException(var2);
    }

}

可以看到invokeWriteAndFlush回调了ChannelOutboundHandler的write和flush方法

最终会调用HeadContext的write和flush方法：

public void write(ChannelHandlerContext ctx, Object msg, ChannelPromise promise) throws Exception {
    this.unsafe.write(msg, promise);
}

public void flush(ChannelHandlerContext ctx) throws Exception {
    this.unsafe.flush();
}

可以看到调用了unsafe的write和flush方法，向unsafe缓冲区写入了消息，当Selector轮询到写事件就绪时，就会通过unsafe将刚才写入的内容交由JDK的SocketChannel完成最终的write操作。

ChannelPipeline的分析到此全部结束。