Android系统进程间通信（IPC）机制Binder中的Server启动过程源代码分析

在前面一篇文章浅谈Android系统进程间通信（IPC）机制Binder中的Server和Client获得Service Manager接口之路中，介绍了在Android系统中Binder进程间通信机制中的Server角色是如何获得Service Manager远程接口的，即defaultServiceManager函数的实现。Server获得了Service Manager远程接口之后，就要把自己的Service添加到Service Manager中去，然后把自己启动起来，等待Client的请求。本文将通过分析源代码了解Server的启动过程是怎么样的。

本文通过一个具体的例子来说明Binder机制中Server的启动过程。我们知道，在Android系统中，提供了多媒体播放的功能，这个功能是以服务的形式来提供的。这里，我们就通过分析MediaPlayerService的实现来了解Media Server的启动过程。

首先，看一下MediaPlayerService的类图，以便我们理解下面要描述的内容。

我们将要介绍的主角MediaPlayerService继承于BnMediaPlayerService类，熟悉Binder机制的同学应该知道BnMediaPlayerService是一个Binder Native类，用来处理Client请求的。BnMediaPlayerService继承于BnInterface类，BnInterface是一个模板类，它定义在frameworks/base/include/binder/IInterface.h文件中：

template<typename INTERFACE>
    class BnInterface : public INTERFACE, public BBinder
    {
    public:
        virtual sp<IInterface>      queryLocalInterface(const String16& _descriptor);
        virtual const String16&     getInterfaceDescriptor() const;

    protected:
        virtual IBinder*            onAsBinder();
    };

这里可以看出，BnMediaPlayerService实际是继承了IMediaPlayerService和BBinder类。IMediaPlayerService和BBinder类又分别继承了IInterface和IBinder类，IInterface和IBinder类又同时继承了RefBase类。

实际上，BnMediaPlayerService并不是直接接收到Client处发送过来的请求，而是使用了IPCThreadState接收Client处发送过来的请求，而IPCThreadState又借助了ProcessState类来与Binder驱动程序交互。有关IPCThreadState和ProcessState的关系，可以参考上一篇文章浅谈Android系统进程间通信（IPC）机制Binder中的Server和Client获得Service Manager接口之路，接下来也会有相应的描述。IPCThreadState接收到了Client处的请求后，就会调用BBinder类的transact函数，并传入相关参数，BBinder类的transact函数最终调用BnMediaPlayerService类的onTransact函数，于是，就开始真正地处理Client的请求了。

了解了MediaPlayerService类结构之后，就要开始进入到本文的主题了。

首先，看看MediaPlayerService是如何启动的。启动MediaPlayerService的代码位于frameworks/base/media/mediaserver/main_mediaserver.cpp文件中：

int main(int argc, char** argv)
    {
        sp<ProcessState> proc(ProcessState::self());
        sp<IServiceManager> sm = defaultServiceManager();
        LOGI("ServiceManager: %p", sm.get());
        AudioFlinger::instantiate();
        MediaPlayerService::instantiate();
        CameraService::instantiate();
        AudioPolicyService::instantiate();
        ProcessState::self()->startThreadPool();
        IPCThreadState::self()->joinThreadPool();
    }

这里我们不关注AudioFlinger和CameraService相关的代码。

先看下面这句代码：

   sp<ProcessState> proc(ProcessState::self());

这句代码的作用是通过ProcessState::self()调用创建一个ProcessState实例。ProcessState::self()是ProcessState类的一个静态成员变量，定义在frameworks/base/libs/binder/ProcessState.cpp文件中：

sp<ProcessState> ProcessState::self()
    {
        if (gProcess != NULL) return gProcess;

        AutoMutex _l(gProcessMutex);
        if (gProcess == NULL) gProcess = new ProcessState;
        return gProcess;
    }

这里可以看出，这个函数作用是返回一个全局唯一的ProcessState实例gProcess。全局唯一实例变量gProcess定义在frameworks/base/libs/binder/Static.cpp文件中：

Mutex gProcessMutex;
    sp<ProcessState> gProcess;

再来看ProcessState的构造函数：

ProcessState::ProcessState()
        : mDriverFD(open_driver())
        , mVMStart(MAP_FAILED)
        , mManagesContexts(false)
        , mBinderContextCheckFunc(NULL)
        , mBinderContextUserData(NULL)
        , mThreadPoolStarted(false)
        , mThreadPoolSeq(1)
    {
        if (mDriverFD >= 0) {
            // XXX Ideally, there should be a specific define for whether we
            // have mmap (or whether we could possibly have the kernel module
            // availabla).
    #if !defined(HAVE_WIN32_IPC)
            // mmap the binder, providing a chunk of virtual address space to receive transactions.
            mVMStart = mmap(0, BINDER_VM_SIZE, PROT_READ, MAP_PRIVATE | MAP_NORESERVE, mDriverFD, 0);
            if (mVMStart == MAP_FAILED) {
                // *sigh*
                LOGE("Using /dev/binder failed: unable to mmap transaction memory.\n");
                close(mDriverFD);
                mDriverFD = -1;
            }
    #else
            mDriverFD = -1;
    #endif
        }
        if (mDriverFD < 0) {
            // Need to run without the driver, starting our own thread pool.
        }
    }

这个函数有两个关键地方，一是通过open_driver函数打开Binder设备文件/dev/binder，并将打开设备文件描述符保存在成员变量mDriverFD中；二是通过mmap来把设备文件/dev/binder映射到内存中。

先看open_driver函数的实现，这个函数同样位于frameworks/base/libs/binder/ProcessState.cpp文件中：

static int open_driver()
    {
        if (gSingleProcess) {
            return -1;
        }

        int fd = open("/dev/binder", O_RDWR);
        if (fd >= 0) {
            fcntl(fd, F_SETFD, FD_CLOEXEC);
            int vers;
    #if defined(HAVE_ANDROID_OS)
            status_t result = ioctl(fd, BINDER_VERSION, &vers);
    #else
            status_t result = -1;
            errno = EPERM;
    #endif
            if (result == -1) {
                LOGE("Binder ioctl to obtain version failed: %s", strerror(errno));
                close(fd);
                fd = -1;
            }
            if (result != 0 || vers != BINDER_CURRENT_PROTOCOL_VERSION) {
                LOGE("Binder driver protocol does not match user space protocol!");
                close(fd);
                fd = -1;
            }
    #if defined(HAVE_ANDROID_OS)
            size_t maxThreads = 15;
            result = ioctl(fd, BINDER_SET_MAX_THREADS, &maxThreads);
            if (result == -1) {
                LOGE("Binder ioctl to set max threads failed: %s", strerror(errno));
            }
    #endif

        } else {
            LOGW("Opening '/dev/binder' failed: %s\n", strerror(errno));
        }
        return fd;
    }

这个函数的作用主要是通过open文件操作函数来打开/dev/binder设备文件，然后再调用ioctl文件控制函数来分别执行BINDER_VERSION和BINDER_SET_MAX_THREADS两个命令来和Binder驱动程序进行交互，前者用于获得当前Binder驱动程序的版本号，后者用于通知Binder驱动程序，MediaPlayerService最多可同时启动15个线程来处理Client端的请求。

open在Binder驱动程序中的具体实现，请参考前面一篇文章浅谈Service Manager成为Android进程间通信（IPC）机制Binder守护进程之路，这里不再重复描述。打开/dev/binder设备文件后，Binder驱动程序就为MediaPlayerService进程创建了一个struct binder_proc结构体实例来维护MediaPlayerService进程上下文相关信息。

我们来看一下ioctl文件操作函数执行BINDER_VERSION命令的过程：

status_t result = ioctl(fd, BINDER_VERSION, &vers);

这个函数调用最终进入到Binder驱动程序的binder_ioctl函数中，我们只关注BINDER_VERSION相关的部分逻辑：

static long binder_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
    {
        int ret;
        struct binder_proc *proc = filp->private_data;
        struct binder_thread *thread;
        unsigned int size = _IOC_SIZE(cmd);
        void __user *ubuf = (void __user *)arg;

        /*printk(KERN_INFO "binder_ioctl: %d:%d %x %lx\n", proc->pid, current->pid, cmd, arg);*/

        ret = wait_event_interruptible(binder_user_error_wait, binder_stop_on_user_error < 2);
        if (ret)
            return ret;

        mutex_lock(&binder_lock);
        thread = binder_get_thread(proc);
        if (thread == NULL) {
            ret = -ENOMEM;
            goto err;
        }

        switch (cmd) {
        ......
        case BINDER_VERSION:
            if (size != sizeof(struct binder_version)) {
                ret = -EINVAL;
                goto err;
            }
            if (put_user(BINDER_CURRENT_PROTOCOL_VERSION, &((struct binder_version *)ubuf)->protocol_version)) {
                ret = -EINVAL;
                goto err;
            }
            break;
        ......
        }
        ret = 0;
    err:
            ......
        return ret;
    }

很简单，只是将BINDER_CURRENT_PROTOCOL_VERSION写入到传入的参数arg指向的用户缓冲区中去就返回了。BINDER_CURRENT_PROTOCOL_VERSION是一个宏，定义在kernel/common/drivers/staging/android/binder.h文件中：

/* This is the current protocol version. */
    #define BINDER_CURRENT_PROTOCOL_VERSION 7

这里为什么要把ubuf转换成struct binder_version之后，再通过其protocol_version成员变量再来写入呢，转了一圈，最终内容还是写入到ubuf中。我们看一下struct binder_version的定义就会明白，同样是在kernel/common/drivers/staging/android/binder.h文件中：

/* Use with BINDER_VERSION, driver fills in fields. */
    struct binder_version {
        /* driver protocol version -- increment with incompatible change */
        signed long protocol_version;
    };

从注释中可以看出来，这里是考虑到兼容性，因为以后很有可能不是用signed long来表示版本号。

这里有一个重要的地方要注意的是，由于这里是打开设备文件/dev/binder之后，第一次进入到binder_ioctl函数，因此，这里调用binder_get_thread的时候，就会为当前线程创建一个struct binder_thread结构体变量来维护线程上下文信息，具体可以参考浅谈Service Manager成为Android进程间通信（IPC）机制Binder守护进程之路一文。

接着我们再来看一下ioctl文件操作函数执行BINDER_SET_MAX_THREADS命令的过程：

result = ioctl(fd, BINDER_SET_MAX_THREADS, &maxThreads);

这个函数调用最终进入到Binder驱动程序的binder_ioctl函数中，我们只关注BINDER_SET_MAX_THREADS相关的部分逻辑：

static long binder_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
    {
        int ret;
        struct binder_proc *proc = filp->private_data;
        struct binder_thread *thread;
        unsigned int size = _IOC_SIZE(cmd);
        void __user *ubuf = (void __user *)arg;

        /*printk(KERN_INFO "binder_ioctl: %d:%d %x %lx\n", proc->pid, current->pid, cmd, arg);*/

        ret = wait_event_interruptible(binder_user_error_wait, binder_stop_on_user_error < 2);
        if (ret)
            return ret;

        mutex_lock(&binder_lock);
        thread = binder_get_thread(proc);
        if (thread == NULL) {
            ret = -ENOMEM;
            goto err;
        }

        switch (cmd) {
        ......
        case BINDER_SET_MAX_THREADS:
            if (copy_from_user(&proc->max_threads, ubuf, sizeof(proc->max_threads))) {
                ret = -EINVAL;
                goto err;
            }
            break;
        ......
        }
        ret = 0;
    err:
        ......
        return ret;
    }

这里实现也是非常简单，只是简单地把用户传进来的参数保存在proc->max_threads中就完毕了。注意，这里再调用binder_get_thread函数的时候，就可以在proc->threads中找到当前线程对应的struct binder_thread结构了，因为前面已经创建好并保存在proc->threads红黑树中。

回到ProcessState的构造函数中，这里还通过mmap函数来把设备文件/dev/binder映射到内存中，这个函数在浅谈Service Manager成为Android进程间通信（IPC）机制Binder守护进程之路一文也已经有详细介绍，这里不再重复描述。宏BINDER_VM_SIZE就定义在ProcessState.cpp文件中：

#define BINDER_VM_SIZE ((1*1024*1024) - (4096 *2))

mmap函数调用完成之后，Binder驱动程序就为当前进程预留了BINDER_VM_SIZE大小的内存空间了。

这样，ProcessState全局唯一变量gProcess就创建完毕了，回到frameworks/base/media/mediaserver/main_mediaserver.cpp文件中的main函数，下一步是调用defaultServiceManager函数来获得Service Manager的远程接口，这个已经在上一篇文章浅谈Android系统进程间通信（IPC）机制Binder中的Server和Client获得Service Manager接口之路有详细描述，读者可以回过头去参考一下。

再接下来，就进入到MediaPlayerService::instantiate函数把MediaPlayerService添加到Service Manger中去了。这个函数定义在frameworks/base/media/libmediaplayerservice/MediaPlayerService.cpp文件中：

void MediaPlayerService::instantiate() {
        defaultServiceManager()->addService(
                String16("media.player"), new MediaPlayerService());
    }

我们重点看一下IServiceManger::addService的过程，这有助于我们加深对Binder机制的理解。

在上一篇文章浅谈Android系统进程间通信（IPC）机制Binder中的Server和Client获得Service Manager接口之路中说到，defaultServiceManager返回的实际是一个BpServiceManger类实例，因此，我们看一下BpServiceManger::addService的实现，这个函数实现在frameworks/base/libs/binder/IServiceManager.cpp文件中：

class BpServiceManager : public BpInterface<IServiceManager>
    {
    public:
        BpServiceManager(const sp<IBinder>& impl)
            : BpInterface<IServiceManager>(impl)
        {
        }

        ......

        virtual status_t addService(const String16& name, const sp<IBinder>& service)
        {
            Parcel data, reply;
            data.writeInterfaceToken(IServiceManager::getInterfaceDescriptor());
            data.writeString16(name);
            data.writeStrongBinder(service);
            status_t err = remote()->transact(ADD_SERVICE_TRANSACTION, data, &reply);
            return err == NO_ERROR ? reply.readExceptionCode() 
        }

        ......

    };

这里的Parcel类是用来于序列化进程间通信数据用的。

先来看这一句的调用：

data.writeInterfaceToken(IServiceManager::getInterfaceDescriptor());

IServiceManager::getInterfaceDescriptor()返回来的是一个字符串，即"android.os.IServiceManager"，具体可以参考IServiceManger的实现。我们看一下Parcel::writeInterfaceToken的实现，位于frameworks/base/libs/binder/Parcel.cpp文件中：

// Write RPC headers.  (previously just the interface token)
    status_t Parcel::writeInterfaceToken(const String16& interface)
    {
        writeInt32(IPCThreadState::self()->getStrictModePolicy() |
                   STRICT_MODE_PENALTY_GATHER);
        // currently the interface identification token is just its name as a string
        return writeString16(interface);
    }

它的作用是写入一个整数和一个字符串到Parcel中去。

再来看下面的调用：

data.writeString16(name);

这里又是写入一个字符串到Parcel中去，这里的name即是上面传进来的"media.player"字符串。

往下看：

data.writeStrongBinder(service);

这里定入一个Binder对象到Parcel去。我们重点看一下这个函数的实现，因为它涉及到进程间传输Binder实体的问题，比较复杂，需要重点关注，同时，也是理解Binder机制的一个重点所在。注意，这里的service参数是一个MediaPlayerService对象。

status_t Parcel::writeStrongBinder(const sp<IBinder>& val)
    {
        return flatten_binder(ProcessState::self(), val, this);
    }

看到flatten_binder函数，是不是似曾相识的感觉？我们在前面一篇文章浅谈Service Manager成为Android进程间通信（IPC）机制Binder守护进程之路中，曾经提到在Binder驱动程序中，使用struct flat_binder_object来表示传输中的一个binder对象，它的定义如下所示：

/*
     * This is the flattened representation of a Binder object for transfer
     * between processes.  The 'offsets' supplied as part of a binder transaction
     * contains offsets into the data where these structures occur.  The Binder
     * driver takes care of re-writing the structure type and data as it moves
     * between processes.
     */
    struct flat_binder_object {
        /* 8 bytes for large_flat_header. */
        unsigned long       type;
        unsigned long       flags;

        /* 8 bytes of data. */
        union {
            void        *binder;    /* local object */
            signed long handle;     /* remote object */
        };

        /* extra data associated with local object */
        void            *cookie;
    };

各个成员变量的含义请参考资料Android Binder设计与实现。

我们进入到flatten_binder函数看看：

status_t flatten_binder(const sp<ProcessState>& proc,
        const sp<IBinder>& binder, Parcel* out)
    {
        flat_binder_object obj;

        obj.flags = 0x7f | FLAT_BINDER_FLAG_ACCEPTS_FDS;
        if (binder != NULL) {
            IBinder *local = binder->localBinder();
            if (!local) {
                BpBinder *proxy = binder->remoteBinder();
                if (proxy == NULL) {
                    LOGE("null proxy");
                }
                const int32_t handle = proxy ? proxy->handle() : 0;
                obj.type = BINDER_TYPE_HANDLE;
                obj.handle = handle;
                obj.cookie = NULL;
            } else {
                obj.type = BINDER_TYPE_BINDER;
                obj.binder = local->getWeakRefs();
                obj.cookie = local;
            }
        } else {
            obj.type = BINDER_TYPE_BINDER;
            obj.binder = NULL;
            obj.cookie = NULL;
        }

        return finish_flatten_binder(binder, obj, out);
    }

首先是初始化flat_binder_object的flags域：

obj.flags = 0x7f | FLAT_BINDER_FLAG_ACCEPTS_FDS;

0x7f表示处理本Binder实体请求数据包的线程的最低优先级，FLAT_BINDER_FLAG_ACCEPTS_FDS表示这个Binder实体可以接受文件描述符，Binder实体在收到文件描述符时，就会在本进程中打开这个文件。

传进来的binder即为MediaPlayerService::instantiate函数中new出来的MediaPlayerService实例，因此，不为空。又由于MediaPlayerService继承自BBinder类，它是一个本地Binder实体，因此binder->localBinder返回一个BBinder指针，而且肯定不为空，于是执行下面语句：

obj.type = BINDER_TYPE_BINDER;
    obj.binder = local->getWeakRefs();
    obj.cookie = local;

设置了flat_binder_obj的其他成员变量，注意，指向这个Binder实体地址的指针local保存在flat_binder_obj的成员变量cookie中。

函数调用finish_flatten_binder来将这个flat_binder_obj写入到Parcel中去：

inline static status_t finish_flatten_binder(
        const sp<IBinder>& binder, const flat_binder_object& flat, Parcel* out)
    {
        return out->writeObject(flat, false);
    }

Parcel::writeObject的实现如下：

status_t Parcel::writeObject(const flat_binder_object& val, bool nullMetaData)
    {
        const bool enoughData = (mDataPos+sizeof(val)) <= mDataCapacity;
        const bool enoughObjects = mObjectsSize < mObjectsCapacity;
        if (enoughData && enoughObjects) {
    restart_write:
            *reinterpret_cast<flat_binder_object*>(mData+mDataPos) = val;

            // Need to write meta-data?
            if (nullMetaData || val.binder != NULL) {
                mObjects[mObjectsSize] = mDataPos;
                acquire_object(ProcessState::self(), val, this);
                mObjectsSize++;
            }

            // remember if it's a file descriptor
            if (val.type == BINDER_TYPE_FD) {
                mHasFds = mFdsKnown = true;
            }

            return finishWrite(sizeof(flat_binder_object));
        }

        if (!enoughData) {
            const status_t err = growData(sizeof(val));
            if (err != NO_ERROR) return err;
        }
        if (!enoughObjects) {
            size_t newSize = ((mObjectsSize+2)*3)/2;
            size_t* objects = (size_t*)realloc(mObjects, newSize*sizeof(size_t));
            if (objects == NULL) return NO_MEMORY;
            mObjects = objects;
            mObjectsCapacity = newSize;
        }

        goto restart_write;
    }

这里除了把flat_binder_obj写到Parcel里面之内，还要记录这个flat_binder_obj在Parcel里面的偏移位置：

mObjects[mObjectsSize] = mDataPos;

这里因为，如果进程间传输的数据间带有Binder对象的时候，Binder驱动程序需要作进一步的处理，以维护各个Binder实体的一致性，下面我们将会看到Binder驱动程序是怎么处理这些Binder对象的。

再回到BpServiceManager::addService函数中，调用下面语句：

status_t err = remote()->transact(ADD_SERVICE_TRANSACTION, data, &reply);

回到浅谈Android系统进程间通信（IPC）机制Binder中的Server和Client获得Service Manager接口之路一文中的类图中去看一下，这里的remote成员函数来自于BpRefBase类，它返回一个BpBinder指针。因此，我们继续进入到BpBinder::transact函数中去看看：

status_t BpBinder::transact(
        uint32_t code, const Parcel& data, Parcel* reply, uint32_t flags)
    {
        // Once a binder has died, it will never come back to life.
        if (mAlive) {
            status_t status = IPCThreadState::self()->transact(
                mHandle, code, data, reply, flags);
            if (status == DEAD_OBJECT) mAlive = 0;
            return status;
        }

        return DEAD_OBJECT;
    }

这里又调用了IPCThreadState::transact进执行实际的操作。注意，这里的mHandle为0，code为ADD_SERVICE_TRANSACTION。ADD_SERVICE_TRANSACTION是上面以参数形式传进来的，那mHandle为什么是0呢？因为这里表示的是Service Manager远程接口，它的句柄值一定是0，具体请参考浅谈Android系统进程间通信（IPC）机制Binder中的Server和Client获得Service Manager接口之路一文。再进入到IPCThreadState::transact函数，看看做了些什么事情：

status_t IPCThreadState::transact(int32_t handle,
                                      uint32_t code, const Parcel& data,
                                      Parcel* reply, uint32_t flags)
    {
        status_t err = data.errorCheck();

        flags |= TF_ACCEPT_FDS;

        IF_LOG_TRANSACTIONS() {
            TextOutput::Bundle _b(alog);
            alog << "BC_TRANSACTION thr " << (void*)pthread_self() << " / hand "
                << handle << " / code " << TypeCode(code) << ": "
                << indent << data << dedent << endl;
        }

        if (err == NO_ERROR) {
            LOG_ONEWAY(">>>> SEND from pid %d uid %d %s", getpid(), getuid(),
                (flags & TF_ONE_WAY) == 0 ? "READ REPLY" : "ONE WAY");
            err = writeTransactionData(BC_TRANSACTION, flags, handle, code, data, NULL);
        }

        if (err != NO_ERROR) {
            if (reply) reply->setError(err);
            return (mLastError = err);
        }

        if ((flags & TF_ONE_WAY) == 0) {
            #if 0
            if (code == 4) { // relayout
                LOGI(">>>>>> CALLING transaction 4");
            } else {
                LOGI(">>>>>> CALLING transaction %d", code);
            }
            #endif
            if (reply) {
                err = waitForResponse(reply);
            } else {
                Parcel fakeReply;
                err = waitForResponse(&fakeReply);
            }
            #if 0
            if (code == 4) { // relayout
                LOGI("<<<<<< RETURNING transaction 4");
            } else {
                LOGI("<<<<<< RETURNING transaction %d", code);
            }
            #endif

            IF_LOG_TRANSACTIONS() {
                TextOutput::Bundle _b(alog);
                alog << "BR_REPLY thr " << (void*)pthread_self() << " / hand "
                    << handle << ": ";
                if (reply) alog << indent << *reply << dedent << endl;
                else alog << "(none requested)" << endl;
            }
        } else {
            err = waitForResponse(NULL, NULL);
        }

        return err;
    }

IPCThreadState::transact函数的参数flags是一个默认值为0的参数，上面没有传相应的实参进来，因此，这里就为0。

函数首先调用writeTransactionData函数准备好一个struct binder_transaction_data结构体变量，这个是等一下要传输给Binder驱动程序的。struct binder_transaction_data的定义我们在浅谈Service Manager成为Android进程间通信（IPC）机制Binder守护进程之路一文中有详细描述，读者不妨回过去读一下。这里为了方便描述，将struct binder_transaction_data的定义再次列出来：

struct binder_transaction_data {
        /* The first two are only used for bcTRANSACTION and brTRANSACTION,
         * identifying the target and contents of the transaction.
         */
        union {
            size_t  handle; /* target descriptor of command transaction */
            void    *ptr;   /* target descriptor of return transaction */
        } target;
        void        *cookie;    /* target object cookie */
        unsigned int    code;       /* transaction command */

        /* General information about the transaction. */
        unsigned int    flags;
        pid_t       sender_pid;
        uid_t       sender_euid;
        size_t      data_size;  /* number of bytes of data */
        size_t      offsets_size;   /* number of bytes of offsets */

        /* If this transaction is inline, the data immediately
         * follows here; otherwise, it ends with a pointer to
         * the data buffer.
         */
        union {
            struct {
                /* transaction data */
                const void  *buffer;
                /* offsets from buffer to flat_binder_object structs */
                const void  *offsets;
            } ptr;
            uint8_t buf[8];
        } data;
    };

writeTransactionData函数的实现如下：

status_t IPCThreadState::writeTransactionData(int32_t cmd, uint32_t binderFlags,
        int32_t handle, uint32_t code, const Parcel& data, status_t* statusBuffer)
    {
        binder_transaction_data tr;

        tr.target.handle = handle;
        tr.code = code;
        tr.flags = binderFlags;

        const status_t err = data.errorCheck();
        if (err == NO_ERROR) {
            tr.data_size = data.ipcDataSize();
            tr.data.ptr.buffer = data.ipcData();
            tr.offsets_size = data.ipcObjectsCount()*sizeof(size_t);
            tr.data.ptr.offsets = data.ipcObjects();
        } else if (statusBuffer) {
            tr.flags |= TF_STATUS_CODE;
            *statusBuffer = err;
            tr.data_size = sizeof(status_t);
            tr.data.ptr.buffer = statusBuffer;
            tr.offsets_size = 0;
            tr.data.ptr.offsets = NULL;
        } else {
            return (mLastError = err);
        }

        mOut.writeInt32(cmd);
        mOut.write(&tr, sizeof(tr));

        return NO_ERROR;
    }

注意，这里的cmd为BC_TRANSACTION。这个函数很简单，在这个场景下，就是执行下面语句来初始化本地变量tr：

tr.data_size = data.ipcDataSize();
    tr.data.ptr.buffer = data.ipcData();
    tr.offsets_size = data.ipcObjectsCount()*sizeof(size_t);
    tr.data.ptr.offsets = data.ipcObjects();

回忆一下上面的内容，写入到tr.data.ptr.buffer的内容相当于下面的内容：

writeInt32(IPCThreadState::self()->getStrictModePolicy() |
                   STRICT_MODE_PENALTY_GATHER);
    writeString16("android.os.IServiceManager");
    writeString16("media.player");
    writeStrongBinder(new MediaPlayerService());

其中包含了一个Binder实体MediaPlayerService，因此需要设置tr.offsets_size就为1，tr.data.ptr.offsets就指向了这个MediaPlayerService的地址在tr.data.ptr.buffer中的偏移量。最后，将tr的内容保存在IPCThreadState的成员变量mOut中。回到IPCThreadState::transact函数中，接下去看，(flags & TF_ONE_WAY) == 0为true，并且reply不为空，所以最终进入到waitForResponse(reply)这条路径来。我们看一下waitForResponse函数的实现：

status_t IPCThreadState::waitForResponse(Parcel *reply, status_t *acquireResult)
    {
        int32_t cmd;
        int32_t err;

        while (1) {
            if ((err=talkWithDriver()) < NO_ERROR) break;
            err = mIn.errorCheck();
            if (err < NO_ERROR) break;
            if (mIn.dataAvail() == 0) continue;

            cmd = mIn.readInt32();

            IF_LOG_COMMANDS() {
                alog << "Processing waitForResponse Command: "
                    << getReturnString(cmd) << endl;
            }

            switch (cmd) {
            case BR_TRANSACTION_COMPLETE:
                if (!reply && !acquireResult) goto finish;
                break;

            case BR_DEAD_REPLY:
                err = DEAD_OBJECT;
                goto finish;

            case BR_FAILED_REPLY:
                err = FAILED_TRANSACTION;
                goto finish;

            case BR_ACQUIRE_RESULT:
                {
                    LOG_ASSERT(acquireResult != NULL, "Unexpected brACQUIRE_RESULT");
                    const int32_t result = mIn.readInt32();
                    if (!acquireResult) continue;
                    *acquireResult = result ? NO_ERROR : INVALID_OPERATION;
                }
                goto finish;

            case BR_REPLY:
                {
                    binder_transaction_data tr;
                    err = mIn.read(&tr, sizeof(tr));
                    LOG_ASSERT(err == NO_ERROR, "Not enough command data for brREPLY");
                    if (err != NO_ERROR) goto finish;

                    if (reply) {
                        if ((tr.flags & TF_STATUS_CODE) == 0) {
                            reply->ipcSetDataReference(
                                reinterpret_cast<const uint8_t*>(tr.data.ptr.buffer),
                                tr.data_size,
                                reinterpret_cast<const size_t*>(tr.data.ptr.offsets),
                                tr.offsets_size/sizeof(size_t),
                                freeBuffer, this);
                        } else {
                            err = *static_cast<const status_t*>(tr.data.ptr.buffer);
                            freeBuffer(NULL,
                                reinterpret_cast<const uint8_t*>(tr.data.ptr.buffer),
                                tr.data_size,
                                reinterpret_cast<const size_t*>(tr.data.ptr.offsets),
                                tr.offsets_size/sizeof(size_t), this);
                        }
                    } else {
                        freeBuffer(NULL,
                            reinterpret_cast<const uint8_t*>(tr.data.ptr.buffer),
                            tr.data_size,
                            reinterpret_cast<const size_t*>(tr.data.ptr.offsets),
                            tr.offsets_size/sizeof(size_t), this);
                        continue;
                    }
                }
                goto finish;

            default:
                err = executeCommand(cmd);
                if (err != NO_ERROR) goto finish;
                break;
            }
        }

    finish:
        if (err != NO_ERROR) {
            if (acquireResult) *acquireResult = err;
            if (reply) reply->setError(err);
            mLastError = err;
        }

        return err;
    }

这个函数虽然很长，但是主要调用了talkWithDriver函数来与Binder驱动程序进行交互：

status_t IPCThreadState::talkWithDriver(bool doReceive)
    {
        LOG_ASSERT(mProcess->mDriverFD >= 0, "Binder driver is not opened");

        binder_write_read bwr;

        // Is the read buffer empty?
        const bool needRead = mIn.dataPosition() >= mIn.dataSize();

        // We don't want to write anything if we are still reading
        // from data left in the input buffer and the caller
        // has requested to read the next data.
        const size_t outAvail = (!doReceive || needRead) ? mOut.dataSize() : 0;

        bwr.write_size = outAvail;
        bwr.write_buffer = (long unsigned int)mOut.data();

        // This is what we'll read.
        if (doReceive && needRead) {
            bwr.read_size = mIn.dataCapacity();
            bwr.read_buffer = (long unsigned int)mIn.data();
        } else {
            bwr.read_size = 0;
        }

        IF_LOG_COMMANDS() {
            TextOutput::Bundle _b(alog);
            if (outAvail != 0) {
                alog << "Sending commands to driver: " << indent;
                const void* cmds = (const void*)bwr.write_buffer;
                const void* end = ((const uint8_t*)cmds)+bwr.write_size;
                alog << HexDump(cmds, bwr.write_size) << endl;
                while (cmds < end) cmds = printCommand(alog, cmds);
                alog << dedent;
            }
            alog << "Size of receive buffer: " << bwr.read_size
                << ", needRead: " << needRead << ", doReceive: " << doReceive << endl;
        }

        // Return immediately if there is nothing to do.
        if ((bwr.write_size == 0) && (bwr.read_size == 0)) return NO_ERROR;

        bwr.write_consumed = 0;
        bwr.read_consumed = 0;
        status_t err;
        do {
            IF_LOG_COMMANDS() {
                alog << "About to read/write, write size = " << mOut.dataSize() << endl;
            }
    #if defined(HAVE_ANDROID_OS)
            if (ioctl(mProcess->mDriverFD, BINDER_WRITE_READ, &bwr) >= 0)
                err = NO_ERROR;
            else
                err = -errno;
    #else
            err = INVALID_OPERATION;
    #endif
            IF_LOG_COMMANDS() {
                alog << "Finished read/write, write size = " << mOut.dataSize() << endl;
            }
        } while (err == -EINTR);

        IF_LOG_COMMANDS() {
            alog << "Our err: " << (void*)err << ", write consumed: "
                << bwr.write_consumed << " (of " << mOut.dataSize()
                << "), read consumed: " << bwr.read_consumed << endl;
        }

        if (err >= NO_ERROR) {
            if (bwr.write_consumed > 0) {
                if (bwr.write_consumed < (ssize_t)mOut.dataSize())
                    mOut.remove(0, bwr.write_consumed);
                else
                    mOut.setDataSize(0);
            }
            if (bwr.read_consumed > 0) {
                mIn.setDataSize(bwr.read_consumed);
                mIn.setDataPosition(0);
            }
            IF_LOG_COMMANDS() {
                TextOutput::Bundle _b(alog);
                alog << "Remaining data size: " << mOut.dataSize() << endl;
                alog << "Received commands from driver: " << indent;
                const void* cmds = mIn.data();
                const void* end = mIn.data() + mIn.dataSize();
                alog << HexDump(cmds, mIn.dataSize()) << endl;
                while (cmds < end) cmds = printReturnCommand(alog, cmds);
                alog << dedent;
            }
            return NO_ERROR;
        }

        return err;
    }

这里doReceive和needRead均为1，有兴趣的读者可以自已分析一下。因此，这里告诉Binder驱动程序，先执行write操作，再执行read操作，下面我们将会看到。

最后，通过ioctl(mProcess->mDriverFD, BINDER_WRITE_READ, &bwr)进行到Binder驱动程序的binder_ioctl函数，我们只关注cmd为BINDER_WRITE_READ的逻辑：

static long binder_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
    {
        int ret;
        struct binder_proc *proc = filp->private_data;
        struct binder_thread *thread;
        unsigned int size = _IOC_SIZE(cmd);
        void __user *ubuf = (void __user *)arg;

        /*printk(KERN_INFO "binder_ioctl: %d:%d %x %lx\n", proc->pid, current->pid, cmd, arg);*/

        ret = wait_event_interruptible(binder_user_error_wait, binder_stop_on_user_error < 2);
        if (ret)
            return ret;

        mutex_lock(&binder_lock);
        thread = binder_get_thread(proc);
        if (thread == NULL) {
            ret = -ENOMEM;
            goto err;
        }

        switch (cmd) {
        case BINDER_WRITE_READ: {
            struct binder_write_read bwr;
            if (size != sizeof(struct binder_write_read)) {
                ret = -EINVAL;
                goto err;
            }
            if (copy_from_user(&bwr, ubuf, sizeof(bwr))) {
                ret = -EFAULT;
                goto err;
            }
            if (binder_debug_mask & BINDER_DEBUG_READ_WRITE)
                printk(KERN_INFO "binder: %d:%d write %ld at %08lx, read %ld at %08lx\n",
                proc->pid, thread->pid, bwr.write_size, bwr.write_buffer, bwr.read_size, bwr.read_buffer);
            if (bwr.write_size > 0) {
                ret = binder_thread_write(proc, thread, (void __user *)bwr.write_buffer, bwr.write_size, &bwr.write_consumed);
                if (ret < 0) {
                    bwr.read_consumed = 0;
                    if (copy_to_user(ubuf, &bwr, sizeof(bwr)))
                        ret = -EFAULT;
                    goto err;
                }
            }
            if (bwr.read_size > 0) {
                ret = binder_thread_read(proc, thread, (void __user *)bwr.read_buffer, bwr.read_size, &bwr.read_consumed, filp->f_flags & O_NONBLOCK);
                if (!list_empty(&proc->todo))
                    wake_up_interruptible(&proc->wait);
                if (ret < 0) {
                    if (copy_to_user(ubuf, &bwr, sizeof(bwr)))
                        ret = -EFAULT;
                    goto err;
                }
            }
            if (binder_debug_mask & BINDER_DEBUG_READ_WRITE)
                printk(KERN_INFO "binder: %d:%d wrote %ld of %ld, read return %ld of %ld\n",
                proc->pid, thread->pid, bwr.write_consumed, bwr.write_size, bwr.read_consumed, bwr.read_size);
            if (copy_to_user(ubuf, &bwr, sizeof(bwr))) {
                ret = -EFAULT;
                goto err;
            }
            break;
        }
        ......
        }
        ret = 0;
    err:
        ......
        return ret;
    }

函数首先是将用户传进来的参数拷贝到本地变量struct binder_write_read bwr中去。这里bwr.write_size > 0为true，因此，进入到binder_thread_write函数中，我们只关注BC_TRANSACTION部分的逻辑：

binder_thread_write(struct binder_proc *proc, struct binder_thread *thread,
                        void __user *buffer, int size, signed long *consumed)
    {
        uint32_t cmd;
        void __user *ptr = buffer + *consumed;
        void __user *end = buffer + size;

        while (ptr < end && thread->return_error == BR_OK) {
            if (get_user(cmd, (uint32_t __user *)ptr))
                return -EFAULT;
            ptr += sizeof(uint32_t);
            if (_IOC_NR(cmd) < ARRAY_SIZE(binder_stats.bc)) {
                binder_stats.bc[_IOC_NR(cmd)]++;
                proc->stats.bc[_IOC_NR(cmd)]++;
                thread->stats.bc[_IOC_NR(cmd)]++;
            }
            switch (cmd) {
                .....
            case BC_TRANSACTION:
            case BC_REPLY: {
                struct binder_transaction_data tr;

                if (copy_from_user(&tr, ptr, sizeof(tr)))
                    return -EFAULT;
                ptr += sizeof(tr);
                binder_transaction(proc, thread, &tr, cmd == BC_REPLY);
                break;
            }
            ......
            }
            *consumed = ptr - buffer;
        }
        return 0;
    }

首先将用户传进来的transact参数拷贝在本地变量struct binder_transaction_data tr中去，接着调用binder_transaction函数进一步处理，这里我们忽略掉无关代码：

static void
    binder_transaction(struct binder_proc *proc, struct binder_thread *thread,
    struct binder_transaction_data *tr, int reply)
    {
        struct binder_transaction *t;
        struct binder_work *tcomplete;
        size_t *offp, *off_end;
        struct binder_proc *target_proc;
        struct binder_thread *target_thread = NULL;
        struct binder_node *target_node = NULL;
        struct list_head *target_list;
        wait_queue_head_t *target_wait;
        struct binder_transaction *in_reply_to = NULL;
        struct binder_transaction_log_entry *e;
        uint32_t return_error;

            ......

        if (reply) {
             ......
        } else {
            if (tr->target.handle) {
                ......
            } else {
                target_node = binder_context_mgr_node;
                if (target_node == NULL) {
                    return_error = BR_DEAD_REPLY;
                    goto err_no_context_mgr_node;
                }
            }
            ......
            target_proc = target_node->proc;
            if (target_proc == NULL) {
                return_error = BR_DEAD_REPLY;
                goto err_dead_binder;
            }
            ......
        }
        if (target_thread) {
            ......
        } else {
            target_list = &target_proc->todo;
            target_wait = &target_proc->wait;
        }

        ......

        /* TODO: reuse incoming transaction for reply */
        t = kzalloc(sizeof(*t), GFP_KERNEL);
        if (t == NULL) {
            return_error = BR_FAILED_REPLY;
            goto err_alloc_t_failed;
        }
        ......

        tcomplete = kzalloc(sizeof(*tcomplete), GFP_KERNEL);
        if (tcomplete == NULL) {
            return_error = BR_FAILED_REPLY;
            goto err_alloc_tcomplete_failed;
        }

        ......

        if (!reply && !(tr->flags & TF_ONE_WAY))
            t->from = thread;
        else
            t->from = NULL;
        t->sender_euid = proc->tsk->cred->euid;
        t->to_proc = target_proc;
        t->to_thread = target_thread;
        t->code = tr->code;
        t->flags = tr->flags;
        t->priority = task_nice(current);
        t->buffer = binder_alloc_buf(target_proc, tr->data_size,
            tr->offsets_size, !reply && (t->flags & TF_ONE_WAY));
        if (t->buffer == NULL) {
            return_error = BR_FAILED_REPLY;
            goto err_binder_alloc_buf_failed;
        }
        t->buffer->allow_user_free = 0;
        t->buffer->debug_id = t->debug_id;
        t->buffer->transaction = t;
        t->buffer->target_node = target_node;
        if (target_node)
            binder_inc_node(target_node, 1, 0, NULL);

        offp = (size_t *)(t->buffer->data + ALIGN(tr->data_size, sizeof(void *)));

        if (copy_from_user(t->buffer->data, tr->data.ptr.buffer, tr->data_size)) {
            ......
            return_error = BR_FAILED_REPLY;
            goto err_copy_data_failed;
        }
        if (copy_from_user(offp, tr->data.ptr.offsets, tr->offsets_size)) {
            ......
            return_error = BR_FAILED_REPLY;
            goto err_copy_data_failed;
        }
        ......

        off_end = (void *)offp + tr->offsets_size;
        for (; offp < off_end; offp++) {
            struct flat_binder_object *fp;
            ......
            fp = (struct flat_binder_object *)(t->buffer->data + *offp);
            switch (fp->type) {
            case BINDER_TYPE_BINDER:
            case BINDER_TYPE_WEAK_BINDER: {
                struct binder_ref *ref;
                struct binder_node *node = binder_get_node(proc, fp->binder);
                if (node == NULL) {
                    node = binder_new_node(proc, fp->binder, fp->cookie);
                    if (node == NULL) {
                        return_error = BR_FAILED_REPLY;
                        goto err_binder_new_node_failed;
                    }
                    node->min_priority = fp->flags & FLAT_BINDER_FLAG_PRIORITY_MASK;
                    node->accept_fds = !!(fp->flags & FLAT_BINDER_FLAG_ACCEPTS_FDS);
                }
                if (fp->cookie != node->cookie) {
                    ......
                    goto err_binder_get_ref_for_node_failed;
                }
                ref = binder_get_ref_for_node(target_proc, node);
                if (ref == NULL) {
                    return_error = BR_FAILED_REPLY;
                    goto err_binder_get_ref_for_node_failed;
                }
                if (fp->type == BINDER_TYPE_BINDER)
                    fp->type = BINDER_TYPE_HANDLE;
                else
                    fp->type = BINDER_TYPE_WEAK_HANDLE;
                fp->handle = ref->desc;
                binder_inc_ref(ref, fp->type == BINDER_TYPE_HANDLE, &thread->todo);
                ......

            } break;
            ......
            }
        }

        if (reply) {
            ......
        } else if (!(t->flags & TF_ONE_WAY)) {
            BUG_ON(t->buffer->async_transaction != 0);
            t->need_reply = 1;
            t->from_parent = thread->transaction_stack;
            thread->transaction_stack = t;
        } else {
            ......
        }
        t->work.type = BINDER_WORK_TRANSACTION;
        list_add_tail(&t->work.entry, target_list);
        tcomplete->type = BINDER_WORK_TRANSACTION_COMPLETE;
        list_add_tail(&tcomplete->entry, &thread->todo);
        if (target_wait)
            wake_up_interruptible(target_wait);
        return;
        ......
    }

注意，这里传进来的参数reply为0，tr->target.handle也为0。因此，target_proc、target_thread、target_node、target_list和target_wait的值分别为：

target_node = binder_context_mgr_node;
    target_proc = target_node->proc;
    target_list = &target_proc->todo;
    target_wait = &target_proc->wait;

接着，分配了一个待处理事务t和一个待完成工作项tcomplete，并执行初始化工作：

    /* TODO: reuse incoming transaction for reply */
        t = kzalloc(sizeof(*t), GFP_KERNEL);
        if (t == NULL) {
            return_error = BR_FAILED_REPLY;
            goto err_alloc_t_failed;
        }
        ......

        tcomplete = kzalloc(sizeof(*tcomplete), GFP_KERNEL);
        if (tcomplete == NULL) {
            return_error = BR_FAILED_REPLY;
            goto err_alloc_tcomplete_failed;
        }

        ......

        if (!reply && !(tr->flags & TF_ONE_WAY))
            t->from = thread;
        else
            t->from = NULL;
        t->sender_euid = proc->tsk->cred->euid;
        t->to_proc = target_proc;
        t->to_thread = target_thread;
        t->code = tr->code;
        t->flags = tr->flags;
        t->priority = task_nice(current);
        t->buffer = binder_alloc_buf(target_proc, tr->data_size,
            tr->offsets_size, !reply && (t->flags & TF_ONE_WAY));
        if (t->buffer == NULL) {
            return_error = BR_FAILED_REPLY;
            goto err_binder_alloc_buf_failed;
        }
        t->buffer->allow_user_free = 0;
        t->buffer->debug_id = t->debug_id;
        t->buffer->transaction = t;
        t->buffer->target_node = target_node;
        if (target_node)
            binder_inc_node(target_node, 1, 0, NULL);

        offp = (size_t *)(t->buffer->data + ALIGN(tr->data_size, sizeof(void *)));

        if (copy_from_user(t->buffer->data, tr->data.ptr.buffer, tr->data_size)) {
            ......
            return_error = BR_FAILED_REPLY;
            goto err_copy_data_failed;
        }
        if (copy_from_user(offp, tr->data.ptr.offsets, tr->offsets_size)) {
            ......
            return_error = BR_FAILED_REPLY;
            goto err_copy_data_failed;
        }

注意，这里的事务t是要交给target_proc处理的，在这个场景之下，就是Service Manager了。因此，下面的语句：

t->buffer = binder_alloc_buf(target_proc, tr->data_size,
            tr->offsets_size, !reply && (t->flags & TF_ONE_WAY));

就是在Service Manager的进程空间中分配一块内存来保存用户传进入的参数了：

    if (copy_from_user(t->buffer->data, tr->data.ptr.buffer, tr->data_size)) {
            ......
            return_error = BR_FAILED_REPLY;
            goto err_copy_data_failed;
        }
        if (copy_from_user(offp, tr->data.ptr.offsets, tr->offsets_size)) {
            ......
            return_error = BR_FAILED_REPLY;
            goto err_copy_data_failed;
        }

由于现在target_node要被使用了，增加它的引用计数：

if (target_node)
            binder_inc_node(target_node, 1, 0, NULL);

接下去的for循环，就是用来处理传输数据中的Binder对象了。在我们的场景中，有一个类型为BINDER_TYPE_BINDER的Binder实体MediaPlayerService：

    switch (fp->type) {
        case BINDER_TYPE_BINDER:
        case BINDER_TYPE_WEAK_BINDER: {
        struct binder_ref *ref;
        struct binder_node *node = binder_get_node(proc, fp->binder);
        if (node == NULL) {
            node = binder_new_node(proc, fp->binder, fp->cookie);
            if (node == NULL) {
                return_error = BR_FAILED_REPLY;
                goto err_binder_new_node_failed;
            }
            node->min_priority = fp->flags & FLAT_BINDER_FLAG_PRIORITY_MASK;
            node->accept_fds = !!(fp->flags & FLAT_BINDER_FLAG_ACCEPTS_FDS);
        }
        if (fp->cookie != node->cookie) {
            ......
            goto err_binder_get_ref_for_node_failed;
        }
        ref = binder_get_ref_for_node(target_proc, node);
        if (ref == NULL) {
            return_error = BR_FAILED_REPLY;
            goto err_binder_get_ref_for_node_failed;
        }
        if (fp->type == BINDER_TYPE_BINDER)
            fp->type = BINDER_TYPE_HANDLE;
        else
            fp->type = BINDER_TYPE_WEAK_HANDLE;
        fp->handle = ref->desc;
        binder_inc_ref(ref, fp->type == BINDER_TYPE_HANDLE, &thread->todo);
        ......

        } break;

由于是第一次在Binder驱动程序中传输这个MediaPlayerService，调用binder_get_node函数查询这个Binder实体时，会返回空，于是binder_new_node在proc中新建一个，下次就可以直接使用了。

现在，由于要把这个Binder实体MediaPlayerService交给target_proc，也就是Service Manager来管理，也就是说Service Manager要引用这个MediaPlayerService了，于是通过binder_get_ref_for_node为MediaPlayerService创建一个引用，并且通过binder_inc_ref来增加这个引用计数，防止这个引用还在使用过程当中就被销毁。注意，到了这里的时候，t->buffer中的flat_binder_obj的type已经改为BINDER_TYPE_HANDLE，handle已经改为ref->desc，跟原来不一样了，因为这个flat_binder_obj是最终是要传给Service Manager的，而Service Manager只能够通过句柄值来引用这个Binder实体。

最后，把待处理事务加入到target_list列表中去：

list_add_tail(&t->work.entry, target_list);

并且把待完成工作项加入到本线程的todo等待执行列表中去：

list_add_tail(&tcomplete->entry, &thread->todo);

现在目标进程有事情可做了，于是唤醒它：

if (target_wait)
        wake_up_interruptible(target_wait);

这里就是要唤醒Service Manager进程了。回忆一下前面浅谈Service Manager成为Android进程间通信（IPC）机制Binder守护进程之路这篇文章，此时， Service Manager正在binder_thread_read函数中调用wait_event_interruptible进入休眠状态。

这里我们先忽略一下Service Manager被唤醒之后的场景，继续MedaPlayerService的启动过程，然后再回来。

回到binder_ioctl函数，bwr.read_size > 0为true，于是进入binder_thread_read函数：

static int
    binder_thread_read(struct binder_proc *proc, struct binder_thread *thread,
                       void  __user *buffer, int size, signed long *consumed, int non_block)
    {
        void __user *ptr = buffer + *consumed;
        void __user *end = buffer + size;

        int ret = 0;
        int wait_for_proc_work;

        if (*consumed == 0) {
            if (put_user(BR_NOOP, (uint32_t __user *)ptr))
                return -EFAULT;
            ptr += sizeof(uint32_t);
        }

    retry:
        wait_for_proc_work = thread->transaction_stack == NULL && list_empty(&thread->todo);

        .......

        if (wait_for_proc_work) {
            .......
        } else {
            if (non_block) {
                if (!binder_has_thread_work(thread))
                    ret = -EAGAIN;
            } else
                ret = wait_event_interruptible(thread->wait, binder_has_thread_work(thread));
        }

        ......

        while (1) {
            uint32_t cmd;
            struct binder_transaction_data tr;
            struct binder_work *w;
            struct binder_transaction *t = NULL;

            if (!list_empty(&thread->todo))
                w = list_first_entry(&thread->todo, struct binder_work, entry);
            else if (!list_empty(&proc->todo) && wait_for_proc_work)
                w = list_first_entry(&proc->todo, struct binder_work, entry);
            else {
                if (ptr - buffer == 4 && !(thread->looper & BINDER_LOOPER_STATE_NEED_RETURN)) /* no data added */
                    goto retry;
                break;
            }

            if (end - ptr < sizeof(tr) + 4)
                break;

            switch (w->type) {
            ......
            case BINDER_WORK_TRANSACTION_COMPLETE: {
                cmd = BR_TRANSACTION_COMPLETE;
                if (put_user(cmd, (uint32_t __user *)ptr))
                    return -EFAULT;
                ptr += sizeof(uint32_t);

                binder_stat_br(proc, thread, cmd);
                if (binder_debug_mask & BINDER_DEBUG_TRANSACTION_COMPLETE)
                    printk(KERN_INFO "binder: %d:%d BR_TRANSACTION_COMPLETE\n",
                    proc->pid, thread->pid);

                list_del(&w->entry);
                kfree(w);
                binder_stats.obj_deleted[BINDER_STAT_TRANSACTION_COMPLETE]++;
                                                   } break;
            ......
            }

            if (!t)
                continue;

            ......
        }

    done:
        ......
        return 0;
    }

这里，thread->transaction_stack和thread->todo均不为空，于是wait_for_proc_work为false，由于binder_has_thread_work的时候，返回true，这里因为thread->todo不为空，因此，线程虽然调用了wait_event_interruptible，但是不会睡眠，于是继续往下执行。

由于thread->todo不为空，执行下列语句：

if (!list_empty(&thread->todo))
         w = list_first_entry(&thread->todo, struct binder_work, entry);

w->type为BINDER_WORK_TRANSACTION_COMPLETE，这是在上面的binder_transaction函数设置的，于是执行：

    switch (w->type) {
        ......
        case BINDER_WORK_TRANSACTION_COMPLETE: {
        cmd = BR_TRANSACTION_COMPLETE;
        if (put_user(cmd, (uint32_t __user *)ptr))
            return -EFAULT;
        ptr += sizeof(uint32_t);

            ......
        list_del(&w->entry);
        kfree(w);

        } break;
        ......
        }

这里就将w从thread->todo删除了。由于这里t为空，重新执行while循环，这时由于已经没有事情可做了，最后就返回到binder_ioctl函数中。注间，这里一共往用户传进来的缓冲区buffer写入了两个整数，分别是BR_NOOP和BR_TRANSACTION_COMPLETE。

binder_ioctl函数返回到用户空间之前，把数据消耗情况拷贝回用户空间中：

if (copy_to_user(ubuf, &bwr, sizeof(bwr))) {
        ret = -EFAULT;
        goto err;
    }

最后返回到IPCThreadState::talkWithDriver函数中，执行下面语句：

    if (err >= NO_ERROR) {
            if (bwr.write_consumed > 0) {
                if (bwr.write_consumed < (ssize_t)mOut.dataSize())
                    mOut.remove(0, bwr.write_consumed);
                else
                    mOut.setDataSize(0);
            }
            if (bwr.read_consumed > 0) {

mIn.setDataSize(bwr.read_consumed); mIn.setDataPosition(0);


} ...... return NO_ERROR; }`

        首先是把mOut的数据清空：

mOut.setDataSize(0);


        然后设置已经读取的内容的大小：

mIn.setDataSize(bwr.read_consumed); mIn.setDataPosition(0);


        然后返回到IPCThreadState::waitForResponse函数中。在IPCThreadState::waitForResponse函数，先是从mIn读出一个整数，这个便是BR_NOOP了，这是一个空操作，什么也不做。然后继续进入IPCThreadState::talkWithDriver函数中。  
        这时候，下面语句执行后：

const bool needRead = mIn.dataPosition() >= mIn.dataSize();


        needRead为false，因为在mIn中，尚有一个整数BR_TRANSACTION_COMPLETE未读出。

       这时候，下面语句执行后：

const size_t outAvail = (!doReceive || needRead) ? mOut.dataSize() : 0;


        outAvail等于0。因此，最后bwr.write_size和bwr.read_size均为0，IPCThreadState::talkWithDriver函数什么也不做，直接返回到IPCThreadState::waitForResponse函数中。在IPCThreadState::waitForResponse函数，又继续从mIn读出一个整数，这个便是BR_TRANSACTION_COMPLETE：

switch (cmd) { case BR_TRANSACTION_COMPLETE: if (!reply && !acquireResult) goto finish; break; ...... }


        reply不为NULL，因此，IPCThreadState::waitForResponse的循环没有结束，继续执行，又进入到IPCThreadState::talkWithDrive中。

        这次，needRead就为true了，而outAvail仍为0，所以bwr.read_size不为0，bwr.write_size为0。于是通过：

ioctl(mProcess->mDriverFD, BINDER_WRITE_READ, &bwr)


        进入到Binder驱动程序中的binder_ioctl函数中。由于bwr.write_size为0，bwr.read_size不为0，这次直接就进入到binder_thread_read函数中。这时候，thread->transaction_stack不等于0，但是thread->todo为空，于是线程就通过：

wait_event_interruptible(thread->wait, binder_has_thread_work(thread));


        进入睡眠状态，等待Service Manager来唤醒了。

        现在，我们可以回到Service Manager被唤醒的过程了。我们接着前面[浅谈Service Manager成为Android进程间通信（IPC）机制Binder守护进程之路](https://www.androidos.net.cn/redirecturl.do?url=http%3A//blog.csdn.net/luoshengyang/article/details/6621566)这篇文章的最后，继续描述。此时， Service Manager正在binder_thread_read函数中调用wait_event_interruptible_exclusive进入休眠状态。上面被MediaPlayerService启动后进程唤醒后，继续执行binder_thread_read函数：

static int binder_thread_read(struct binder_proc proc, struct binder_thread thread, void __user buffer, int size, signed long consumed, int non_block) { void user ptr = buffer + consumed; void user *end = buffer + size;

int ret = 0; int wait_for_proc_work;

if (consumed == 0) { if (put_user(BR_NOOP, (uint32_t __user )ptr)) return -EFAULT; ptr += sizeof(uint32_t); }

retry: wait_for_proc_work = thread->transaction_stack == NULL && list_empty(&thread->todo);

......

if (wait_for_proc_work) { ...... if (non_block) { if (!binder_has_proc_work(proc, thread)) ret = -EAGAIN; } else ret = wait_event_interruptible_exclusive(proc->wait, binder_has_proc_work(proc, thread)); } else { ...... }

......

while (1) { uint32_t cmd; struct binder_transaction_data tr; struct binder_work w; struct binder_transaction t = NULL;

if (!list_empty(&thread->todo)) w = list_first_entry(&thread->todo, struct binder_work, entry); else if (!list_empty(&proc->todo) && wait_for_proc_work) w = list_first_entry(&proc->todo, struct binder_work, entry); else { if (ptr - buffer == 4 && !(thread->looper & BINDER_LOOPER_STATE_NEED_RETURN)) / no data added / goto retry; break; }

if (end - ptr < sizeof(tr) + 4) break;

switch (w->type) { case BINDER_WORK_TRANSACTION: { t = container_of(w, struct binder_transaction, work); } break; ...... }

if (!t) continue;

BUG_ON(t->buffer == NULL); if (t->buffer->target_node) { struct binder_node *target_node = t->buffer->target_node; tr.target.ptr = target_node->ptr; tr.cookie = target_node->cookie; ...... cmd = BR_TRANSACTION; } else { ...... } tr.code = t->code; tr.flags = t->flags; tr.sender_euid = t->sender_euid;

if (t->from) { struct task_struct *sender = t->from->proc->tsk; tr.sender_pid = task_tgid_nr_ns(sender, current->nsproxy->pid_ns); } else { tr.sender_pid = 0; }

tr.data_size = t->buffer->data_size; tr.offsets_size = t->buffer->offsets_size; tr.data.ptr.buffer = (void )t->buffer->data + proc->user_buffer_offset; tr.data.ptr.offsets = tr.data.ptr.buffer + ALIGN(t->buffer->data_size, sizeof(void ));

if (put_user(cmd, (uint32_t __user *)ptr)) return -EFAULT; ptr += sizeof(uint32_t); if (copy_to_user(ptr, &tr, sizeof(tr))) return -EFAULT; ptr += sizeof(tr);

......

list_del(&t->work.entry); t->buffer->allow_user_free = 1; if (cmd == BR_TRANSACTION && !(t->flags & TF_ONE_WAY)) { t->to_parent = thread->transaction_stack; t->to_thread = thread; thread->transaction_stack = t; } else { t->buffer->transaction = NULL; kfree(t); binder_stats.obj_deleted[BINDER_STAT_TRANSACTION]++; } break; }

done:

...... return 0; }


        Service Manager被唤醒之后，就进入while循环开始处理事务了。这里wait_for_proc_work等于1，并且proc->todo不为空，所以从proc->todo列表中得到第一个工作项：

w = list_first_entry(&proc->todo, struct binder_work, entry);


        从上面的描述中，我们知道，这个工作项的类型为BINDER_WORK_TRANSACTION，于是通过下面语句得到事务项：

t = container_of(w, struct binder_transaction, work);


       接着就是把事务项t中的数据拷贝到本地局部变量struct binder_transaction_data tr中去了：

if (t->buffer->target_node) { struct binder_node *target_node = t->buffer->target_node; tr.target.ptr = target_node->ptr; tr.cookie = target_node->cookie; ...... cmd = BR_TRANSACTION; } else { ...... } tr.code = t->code; tr.flags = t->flags; tr.sender_euid = t->sender_euid;

if (t->from) { struct task_struct *sender = t->from->proc->tsk; tr.sender_pid = task_tgid_nr_ns(sender, current->nsproxy->pid_ns); } else { tr.sender_pid = 0; }


        这里有一个非常重要的地方，是Binder进程间通信机制的精髓所在：

tr.data.ptr.buffer = (void )t->buffer->data + proc->user_buffer_offset; tr.data.ptr.offsets = tr.data.ptr.buffer + ALIGN(t->buffer->data_size, sizeof(void ));


        t->buffer->data所指向的地址是内核空间的，现在要把数据返回给Service Manager进程的用户空间，而Service Manager进程的用户空间是不能访问内核空间的数据的，所以这里要作一下处理。怎么处理呢？我们在学面向对象语言的时候，对象的拷贝有深拷贝和浅拷贝之分，深拷贝是把另外分配一块新内存，然后把原始对象的内容搬过去，浅拷贝是并没有为新对象分配一块新空间，而只是分配一个引用，而个引用指向原始对象。Binder机制用的是类似浅拷贝的方法，通过在用户空间分配一个虚拟地址，然后让这个用户空间虚拟地址与 t->buffer->data这个内核空间虚拟地址指向同一个物理地址，这样就可以实现浅拷贝了。怎么样用户空间和内核空间的虚拟地址同时指向同一个物理地址呢？请参考前面一篇文章[浅谈Service Manager成为Android进程间通信（IPC）机制Binder守护进程之路](https://www.androidos.net.cn/redirecturl.do?url=http%3A//blog.csdn.net/luoshengyang/article/details/6621566)，那里有详细描述。这里只要将t->buffer->data加上一个偏移值proc->user_buffer_offset就可以得到t->buffer->data对应的用户空间虚拟地址了。调整了tr.data.ptr.buffer的值之后，不要忘记也要一起调整tr.data.ptr.offsets的值。  


        接着就是把tr的内容拷贝到用户传进来的缓冲区去了，指针ptr指向这个用户缓冲区的地址：

if (put_user(cmd, (uint32_t __user *)ptr)) return -EFAULT; ptr += sizeof(uint32_t); if (copy_to_user(ptr, &tr, sizeof(tr))) return -EFAULT; ptr += sizeof(tr);


         这里可以看出，这里只是对作tr.data.ptr.bufferr和tr.data.ptr.offsets的内容作了浅拷贝。

         最后，由于已经处理了这个事务，要把它从todo列表中删除：


         注意，这里的cmd == BR_TRANSACTION && !(t->flags & TF_ONE_WAY)为true，表明这个事务虽然在驱动程序中已经处理完了，但是它仍然要等待Service Manager完成之后，给驱动程序一个确认，也就是需要等待回复，于是把当前事务t放在thread->transaction_stack队列的头部：

t->to_parent = thread->transaction_stack; t->to_thread = thread; thread->transaction_stack = t;


         如果cmd == BR_TRANSACTION && !(t->flags & TF_ONE_WAY)为false，那就不需要等待回复了，直接把事务t删掉。  


         这个while最后通过一个break跳了出来，最后返回到binder_ioctl函数中：

static long binder_ioctl(struct file filp, unsigned int cmd, unsigned long arg) { int ret; struct binder_proc proc = filp->private_data; struct binder_thread thread; unsigned int size = _IOC_SIZE(cmd); void __user ubuf = (void __user *)arg;

......

switch (cmd) { case BINDER_WRITE_READ: { struct binder_write_read bwr; if (size != sizeof(struct binder_write_read)) { ret = -EINVAL; goto err; } if (copy_from_user(&bwr, ubuf, sizeof(bwr))) { ret = -EFAULT; goto err; } ...... if (bwr.read_size > 0) { ret = binder_thread_read(proc, thread, (void __user *)bwr.read_buffer, bwr.read_size, &bwr.read_consumed, filp->f_flags & O_NONBLOCK); if (!list_empty(&proc->todo)) wake_up_interruptible(&proc->wait); if (ret < 0) { if (copy_to_user(ubuf, &bwr, sizeof(bwr))) ret = -EFAULT; goto err; } } ...... if (copy_to_user(ubuf, &bwr, sizeof(bwr))) { ret = -EFAULT; goto err; } break; } ...... default: ret = -EINVAL; goto err; } ret = 0; err: ...... return ret; }


         从binder_thread_read返回来后，再看看proc->todo是否还有事务等待处理，如果是，就把睡眠在proc->wait队列的线程唤醒来处理。最后，把本地变量struct binder_write_read bwr的内容拷贝回到用户传进来的缓冲区中，就返回了。

        这里就是返回到frameworks/base/cmds/servicemanager/binder.c文件中的binder_loop函数了：

void binder_loop(struct binder_state *bs, binder_handler func) { int res; struct binder_write_read bwr; unsigned readbuf[32];

bwr.write_size = 0; bwr.write_consumed = 0; bwr.write_buffer = 0;

readbuf[0] = BC_ENTER_LOOPER; binder_write(bs, readbuf, sizeof(unsigned));

for (;;) { bwr.read_size = sizeof(readbuf); bwr.read_consumed = 0; bwr.read_buffer = (unsigned) readbuf;

res = ioctl(bs->fd, BINDER_WRITE_READ, &bwr);

if (res < 0) { LOGE("binder_loop: ioctl failed (%s)\n", strerror(errno)); break; }

res = binder_parse(bs, 0, readbuf, bwr.read_consumed, func); if (res == 0) { LOGE("binder_loop: unexpected reply?!\n"); break; } if (res < 0) { LOGE("binder_loop: io error %d %s\n", res, strerror(errno)); break; } } }


       返回来的数据都放在readbuf中，接着调用binder_parse进行解析：

int binder_parse(struct binder_state bs, struct binder_io bio, uint32_t ptr, uint32_t size, binder_handler func) { int r = 1; uint32_t end = ptr + (size / 4);

while (ptr < end) { uint32_t cmd = ptr++; ...... case BR_TRANSACTION: { struct binder_txn txn = (void ) ptr; if ((end - ptr) sizeof(uint32_t) < sizeof(struct binder_txn)) { LOGE("parse: txn too small!\n"); return -1; } binder_dump_txn(txn); if (func) { unsigned rdata[256/4]; struct binder_io msg; struct binder_io reply; int res;

bio_init(&reply, rdata, sizeof(rdata), 4); bio_init_from_txn(&msg, txn); res = func(bs, txn, &msg, &reply); binder_send_reply(bs, &reply, txn->data, res); } ptr += sizeof(*txn) / sizeof(uint32_t); break; } ...... default: LOGE("parse: OOPS %d\n", cmd); return -1; } }

return r; }


        首先把从Binder驱动程序读出来的数据转换为一个struct binder_txn结构体，保存在txn本地变量中，struct binder_txn定义在frameworks/base/cmds/servicemanager/binder.h文件中：

struct binder_txn { void target; void cookie; uint32_t code; uint32_t flags;

uint32_t sender_pid; uint32_t sender_euid;

uint32_t data_size; uint32_t offs_size; void data; void offs; };


       函数中还用到了另外一个数据结构struct binder_io，也是定义在frameworks/base/cmds/servicemanager/binder.h文件中：

struct binder_io { char data; / pointer to read/write from / uint32_t offs; / array of offsets / uint32_t data_avail; / bytes available in data buffer / uint32_t offs_avail; / entries available in offsets array /

char data0; / start of data buffer / uint32_t offs0; / start of offsets buffer / uint32_t flags; uint32_t unused; };


       接着往下看，函数调bio_init来初始化reply变量：

void bio_init(struct binder_io bio, void data, uint32_t maxdata, uint32_t maxoffs) { uint32_t n = maxoffs * sizeof(uint32_t);

if (n > maxdata) { bio->flags = BIO_F_OVERFLOW; bio->data_avail = 0; bio->offs_avail = 0; return; }

bio->data = bio->data0 = data + n; bio->offs = bio->offs0 = data; bio->data_avail = maxdata - n; bio->offs_avail = maxoffs; bio->flags = 0; }


       接着又调用bio_init_from_txn来初始化msg变量：

void bio_init_from_txn(struct binder_io bio, struct binder_txn txn) { bio->data = bio->data0 = txn->data; bio->offs = bio->offs0 = txn->offs; bio->data_avail = txn->data_size; bio->offs_avail = txn->offs_size / 4; bio->flags = BIO_F_SHARED; }


      最后，真正进行处理的函数是从参数中传进来的函数指针func，这里就是定义在frameworks/base/cmds/servicemanager/service_manager.c文件中的svcmgr_handler函数：

int svcmgr_handler(struct binder_state bs, struct binder_txn txn, struct binder_io msg, struct binder_io reply) { struct svcinfo si; uint16_t s; unsigned len; void *ptr; uint32_t strict_policy;

if (txn->target != svcmgr_handle) return -1;

// Equivalent to Parcel::enforceInterface(), reading the RPC // header with the strict mode policy mask and the interface name. // Note that we ignore the strict_policy and don't propagate it // further (since we do no outbound RPCs anyway). strict_policy = bio_get_uint32(msg); s = bio_get_string16(msg, &len); if ((len != (sizeof(svcmgr_id) / 2)) || memcmp(svcmgr_id, s, sizeof(svcmgr_id))) { fprintf(stderr,"invalid id %s\n", str8(s)); return -1; }

switch(txn->code) { ...... case SVC_MGR_ADD_SERVICE: s = bio_get_string16(msg, &len); ptr = bio_get_ref(msg); if (do_add_service(bs, s, len, ptr, txn->sender_euid)) return -1; break; ...... }

bio_put_uint32(reply, 0); return 0; }


         回忆一下，在BpServiceManager::addService时，传给Binder驱动程序的参数为：

writeInt32(IPCThreadState::self()->getStrictModePolicy() | STRICT_MODE_PENALTY_GATHER); writeString16("android.os.IServiceManager"); writeString16("media.player"); writeStrongBinder(new MediaPlayerService());


         这里的语句：

strict_policy = bio_get_uint32(msg); s = bio_get_string16(msg, &len); s = bio_get_string16(msg, &len); ptr = bio_get_ref(msg);


         就是依次把它们读取出来了，这里，我们只要看一下bio_get_ref的实现。先看一个数据结构struct binder_obj的定义：

struct binder_object { uint32_t type; uint32_t flags; void pointer; void cookie; };


        这个结构体其实就是对应struct flat_binder_obj的。

        接着看bio_get_ref实现：

void bio_get_ref(struct binder_io bio) { struct binder_object *obj;

obj = _bio_get_obj(bio); if (!obj) return 0;

if (obj->type == BINDER_TYPE_HANDLE) return obj->pointer;

return 0; }


       _bio_get_obj这个函数就不跟进去看了，它的作用就是从binder_io中取得第一个还没取获取过的binder_object。在这个场景下，就是我们最开始传过来代表MediaPlayerService的flat_binder_obj了，这个原始的flat_binder_obj的type为BINDER_TYPE_BINDER，binder为指向MediaPlayerService的弱引用的地址。在前面我们说过，在Binder驱动驱动程序里面，会把这个flat_binder_obj的type改为BINDER_TYPE_HANDLE，handle改为一个句柄值。这里的handle值就等于obj->pointer的值。

        回到svcmgr_handler函数，调用do_add_service进一步处理：

int do_add_service(struct binder_state bs, uint16_t s, unsigned len, void ptr, unsigned uid) { struct svcinfo si; // LOGI("add_service('%s',%p) uid=%d\n", str8(s), ptr, uid);

if (!ptr || (len == 0) || (len > 127)) return -1;

if (!svc_can_register(uid, s)) { LOGE("add_service('%s',%p) uid=%d - PERMISSION DENIED\n", str8(s), ptr, uid); return -1; }

si = find_svc(s, len); if (si) { if (si->ptr) { LOGE("add_service('%s',%p) uid=%d - ALREADY REGISTERED\n", str8(s), ptr, uid); return -1; } si->ptr = ptr; } else { si = malloc(sizeof(si) + (len + 1) sizeof(uint16_t)); if (!si) { LOGE("add_service('%s',%p) uid=%d - OUT OF MEMORY\n", str8(s), ptr, uid); return -1; } si->ptr = ptr; si->len = len; memcpy(si->name, s, (len + 1) * sizeof(uint16_t)); si->name[len] = '\0'; si->death.func = svcinfo_death; si->death.ptr = si; si->next = svclist; svclist = si; }

binder_acquire(bs, ptr); binder_link_to_death(bs, ptr, &si->death); return 0; }


        这个函数的实现很简单，就是把MediaPlayerService这个Binder实体的引用写到一个struct svcinfo结构体中，主要是它的名称和句柄值，然后插入到链接svclist的头部去。这样，Client来向Service Manager查询服务接口时，只要给定服务名称，Service Manger就可以返回相应的句柄值了。

        这个函数执行完成后，返回到svcmgr_handler函数，函数的最后，将一个错误码0写到reply变量中去，表示一切正常：

bio_put_uint32(reply, 0);


       svcmgr_handler函数执行完成后，返回到binder_parse函数，执行下面语句：

binder_send_reply(bs, &reply, txn->data, res);


       我们看一下binder_send_reply的实现，从函数名就可以猜到它要做什么了，告诉Binder驱动程序，它完成了Binder驱动程序交给它的任务了。

void binder_send_reply(struct binder_state bs, struct binder_io reply, void buffer_to_free, int status) { struct { uint32_t cmd_free; void buffer; uint32_t cmd_reply; struct binder_txn txn; } attribute((packed)) data;

data.cmd_free = BC_FREE_BUFFER; data.buffer = buffer_to_free; data.cmd_reply = BC_REPLY; data.txn.target = 0; data.txn.cookie = 0; data.txn.code = 0; if (status) { data.txn.flags = TF_STATUS_CODE; data.txn.data_size = sizeof(int); data.txn.offs_size = 0; data.txn.data = &status; data.txn.offs = 0; } else { data.txn.flags = 0; data.txn.data_size = reply->data - reply->data0; data.txn.offs_size = ((char) reply->offs) - ((char) reply->offs0); data.txn.data = reply->data0; data.txn.offs = reply->offs0; } binder_write(bs, &data, sizeof(data)); }


       从这里可以看出，binder_send_reply告诉Binder驱动程序执行BC_FREE_BUFFER和BC_REPLY命令，前者释放之前在binder_transaction分配的空间，地址为buffer_to_free，buffer_to_free这个地址是Binder驱动程序把自己在内核空间用的地址转换成用户空间地址再传给Service Manager的，所以Binder驱动程序拿到这个地址后，知道怎么样释放这个空间；后者告诉MediaPlayerService，它的addService操作已经完成了，错误码是0，保存在data.txn.data中。  


       再来看binder_write函数：

int binder_write(struct binder_state bs, void data, unsigned len) { struct binder_write_read bwr; int res; bwr.write_size = len; bwr.write_consumed = 0; bwr.write_buffer = (unsigned) data; bwr.read_size = 0; bwr.read_consumed = 0; bwr.read_buffer = 0; res = ioctl(bs->fd, BINDER_WRITE_READ, &bwr); if (res < 0) { fprintf(stderr,"binder_write: ioctl failed (%s)\n", strerror(errno)); } return res; }


       这里可以看出，只有写操作，没有读操作，即read_size为0。

       这里又是一个ioctl的BINDER_WRITE_READ操作。直入到驱动程序的binder_ioctl函数后，执行BINDER_WRITE_READ命令，这里就不累述了。

       最后，从binder_ioctl执行到binder_thread_write函数，我们首先看第一个命令BC_FREE_BUFFER：

int binder_thread_write(struct binder_proc proc, struct binder_thread thread, void __user buffer, int size, signed long consumed) { uint32_t cmd; void user ptr = buffer + consumed; void user *end = buffer + size;

while (ptr < end && thread->return_error == BR_OK) { if (get_user(cmd, (uint32_t user *)ptr)) return -EFAULT; ptr += sizeof(uint32_t); if (_IOC_NR(cmd) < ARRAY_SIZE(binder_stats.bc)) { binder_stats.bc[_IOC_NR(cmd)]++; proc->stats.bc[_IOC_NR(cmd)]++; thread->stats.bc[_IOC_NR(cmd)]++; } switch (cmd) { ...... case BC_FREE_BUFFER: { void user data_ptr; struct binder_buffer buffer;

if (get_user(data_ptr, (void __user )ptr)) return -EFAULT; ptr += sizeof(void *);

buffer = binder_buffer_lookup(proc, data_ptr); if (buffer == NULL) { binder_user_error("binder: %d:%d " "BC_FREE_BUFFER u%p no match\n", proc->pid, thread->pid, data_ptr); break; } if (!buffer->allow_user_free) { binder_user_error("binder: %d:%d " "BC_FREE_BUFFER u%p matched " "unreturned buffer\n", proc->pid, thread->pid, data_ptr); break; } if (binder_debug_mask & BINDER_DEBUG_FREE_BUFFER) printk(KERN_INFO "binder: %d:%d BC_FREE_BUFFER u%p found buffer %d for %s transaction\n", proc->pid, thread->pid, data_ptr, buffer->debug_id, buffer->transaction ? "active" : "finished");

if (buffer->transaction) { buffer->transaction->buffer = NULL; buffer->transaction = NULL; } if (buffer->async_transaction && buffer->target_node) { BUG_ON(!buffer->target_node->has_async_transaction); if (list_empty(&buffer->target_node->async_todo)) buffer->target_node->has_async_transaction = 0; else list_move_tail(buffer->target_node->async_todo.next, &thread->todo); } binder_transaction_buffer_release(proc, buffer, NULL); binder_free_buf(proc, buffer); break; }

...... *consumed = ptr - buffer; } return 0; }


       首先通过看这个语句：

get_user(data_ptr, (void __user )ptr)


       这个是获得要删除的Buffer的用户空间地址，接着通过下面这个语句来找到这个地址对应的struct binder_buffer信息：

buffer = binder_buffer_lookup(proc, data_ptr);


       因为这个空间是前面在binder_transaction里面分配的，所以这里一定能找到。

       最后，就可以释放这块内存了：

binder_transaction_buffer_release(proc, buffer, NULL); binder_free_buf(proc, buffer);


       再来看另外一个命令BC_REPLY：

while (ptr < end && thread->return_error == BR_OK) { if (get_user(cmd, (uint32_t __user *)ptr)) return -EFAULT; ptr += sizeof(uint32_t); if (_IOC_NR(cmd) < ARRAY_SIZE(binder_stats.bc)) { binder_stats.bc[_IOC_NR(cmd)]++; proc->stats.bc[_IOC_NR(cmd)]++; thread->stats.bc[_IOC_NR(cmd)]++; } switch (cmd) { ...... case BC_TRANSACTION: case BC_REPLY: { struct binder_transaction_data tr;

if (copy_from_user(&tr, ptr, sizeof(tr))) return -EFAULT; ptr += sizeof(tr); binder_transaction(proc, thread, &tr, cmd == BC_REPLY); break; }

...... *consumed = ptr - buffer; } return 0; }


       又再次进入到binder_transaction函数：

static void binder_transaction(struct binder_proc proc, struct binder_thread thread, struct binder_transaction_data tr, int reply) { struct binder_transaction t; struct binder_work tcomplete; size_t offp, off_end; struct binder_proc target_proc; struct binder_thread target_thread = NULL; struct binder_node target_node = NULL; struct list_head target_list; wait_queue_head_t target_wait; struct binder_transaction in_reply_to = NULL; struct binder_transaction_log_entry e; uint32_t return_error;

......

if (reply) { in_reply_to = thread->transaction_stack; if (in_reply_to == NULL) { ...... return_error = BR_FAILED_REPLY; goto err_empty_call_stack; } binder_set_nice(in_reply_to->saved_priority); if (in_reply_to->to_thread != thread) { ....... goto err_bad_call_stack; } thread->transaction_stack = in_reply_to->to_parent; target_thread = in_reply_to->from; if (target_thread == NULL) { return_error = BR_DEAD_REPLY; goto err_dead_binder; } if (target_thread->transaction_stack != in_reply_to) { ...... return_error = BR_FAILED_REPLY; in_reply_to = NULL; target_thread = NULL; goto err_dead_binder; } target_proc = target_thread->proc; } else { ...... } if (target_thread) { e->to_thread = target_thread->pid; target_list = &target_thread->todo; target_wait = &target_thread->wait; } else { ...... }

/ TODO: reuse incoming transaction for reply / t = kzalloc(sizeof(*t), GFP_KERNEL); if (t == NULL) { return_error = BR_FAILED_REPLY; goto err_alloc_t_failed; }

tcomplete = kzalloc(sizeof(*tcomplete), GFP_KERNEL); if (tcomplete == NULL) { return_error = BR_FAILED_REPLY; goto err_alloc_tcomplete_failed; }

if (!reply && !(tr->flags & TF_ONE_WAY)) t->from = thread; else t->from = NULL; t->sender_euid = proc->tsk->cred->euid; t->to_proc = target_proc; t->to_thread = target_thread; t->code = tr->code; t->flags = tr->flags; t->priority = task_nice(current); t->buffer = binder_alloc_buf(target_proc, tr->data_size, tr->offsets_size, !reply && (t->flags & TF_ONE_WAY)); if (t->buffer == NULL) { return_error = BR_FAILED_REPLY; goto err_binder_alloc_buf_failed; } t->buffer->allow_user_free = 0; t->buffer->debug_id = t->debug_id; t->buffer->transaction = t; t->buffer->target_node = target_node; if (target_node) binder_inc_node(target_node, 1, 0, NULL);

offp = (size_t )(t->buffer->data + ALIGN(tr->data_size, sizeof(void )));

if (copy_from_user(t->buffer->data, tr->data.ptr.buffer, tr->data_size)) { binder_user_error("binder: %d:%d got transaction with invalid " "data ptr\n", proc->pid, thread->pid); return_error = BR_FAILED_REPLY; goto err_copy_data_failed; } if (copy_from_user(offp, tr->data.ptr.offsets, tr->offsets_size)) { binder_user_error("binder: %d:%d got transaction with invalid " "offsets ptr\n", proc->pid, thread->pid); return_error = BR_FAILED_REPLY; goto err_copy_data_failed; }

......

if (reply) { BUG_ON(t->buffer->async_transaction != 0); binder_pop_transaction(target_thread, in_reply_to); } else if (!(t->flags & TF_ONE_WAY)) { ...... } else { ...... } t->work.type = BINDER_WORK_TRANSACTION; list_add_tail(&t->work.entry, target_list); tcomplete->type = BINDER_WORK_TRANSACTION_COMPLETE; list_add_tail(&tcomplete->entry, &thread->todo); if (target_wait) wake_up_interruptible(target_wait); return; ...... }


       注意，这里的reply为1，我们忽略掉其它无关代码。

       前面Service Manager正在binder_thread_read函数中被MediaPlayerService启动后进程唤醒后，在最后会把当前处理完的事务放在thread->transaction_stack中：

if (cmd == BR_TRANSACTION && !(t->flags & TF_ONE_WAY)) { t->to_parent = thread->transaction_stack; t->to_thread = thread; thread->transaction_stack = t; }


       所以，这里，首先是把它这个binder_transaction取回来，并且放在本地变量in_reply_to中：

in_reply_to = thread->transaction_stack;


       接着就可以通过in_reply_to得到最终发出这个事务请求的线程和进程：

target_thread = in_reply_to->from; target_proc = target_thread->proc;


        然后得到target_list和target_wait：

target_list = &target_thread->todo; target_wait = &target_thread->wait;


       下面这一段代码：

/ TODO: reuse incoming transaction for reply / t = kzalloc(sizeof(*t), GFP_KERNEL); if (t == NULL) { return_error = BR_FAILED_REPLY; goto err_alloc_t_failed; }

tcomplete = kzalloc(sizeof(*tcomplete), GFP_KERNEL); if (tcomplete == NULL) { return_error = BR_FAILED_REPLY; goto err_alloc_tcomplete_failed; }

offp = (size_t )(t->buffer->data + ALIGN(tr->data_size, sizeof(void )));


          我们在前面已经分析过了，这里不再重复。但是有一点要注意的是，这里target_node为NULL，因此，t->buffer->target_node也为NULL。

          函数本来有一个for循环，用来处理数据中的Binder对象，这里由于没有Binder对象，所以就略过了。到了下面这句代码：

binder_pop_transaction(target_thread, in_reply_to);


          我们看看做了什么事情：

static void binder_pop_transaction( struct binder_thread target_thread, struct binder_transaction t) { if (target_thread) { BUG_ON(target_thread->transaction_stack != t); BUG_ON(target_thread->transaction_stack->from != target_thread); target_thread->transaction_stack = target_thread->transaction_stack->from_parent; t->from = NULL; } t->need_reply = 0; if (t->buffer) t->buffer->transaction = NULL; kfree(t); binder_stats.obj_deleted[BINDER_STAT_TRANSACTION]++; }


        由于到了这里，已经不需要in_reply_to这个transaction了，就把它删掉。

        回到binder_transaction函数：

t->work.type = BINDER_WORK_TRANSACTION; list_add_tail(&t->work.entry, target_list); tcomplete->type = BINDER_WORK_TRANSACTION_COMPLETE; list_add_tail(&tcomplete->entry, &thread->todo);


         和前面一样，分别把t和tcomplete分别放在target_list和thread->todo队列中，这里的target_list指的就是最初调用IServiceManager::addService的MediaPlayerService的Server主线程的的thread->todo队列了，而thread->todo指的是Service Manager中用来回复IServiceManager::addService请求的线程。

        最后，唤醒等待在target_wait队列上的线程了，就是最初调用IServiceManager::addService的MediaPlayerService的Server主线程了，它最后在binder_thread_read函数中睡眠在thread->wait上，就是这里的target_wait了：

if (target_wait) wake_up_interruptible(target_wait);


        这样，Service Manger回复调用IServiceManager::addService请求就算完成了，重新回到frameworks/base/cmds/servicemanager/binder.c文件中的binder_loop函数等待下一个Client请求的到来。事实上，Service Manger回到binder_loop函数再次执行ioctl函数时候，又会再次进入到binder_thread_read函数。这时个会发现thread->todo不为空，这是因为刚才我们调用了：

list_add_tail(&tcomplete->entry, &thread->todo);


          把一个工作项tcompelete放在了在thread->todo中，这个tcompelete的type为BINDER_WORK_TRANSACTION_COMPLETE，因此，Binder驱动程序会执行下面操作：

switch (w->type) { case BINDER_WORK_TRANSACTION_COMPLETE: { cmd = BR_TRANSACTION_COMPLETE; if (put_user(cmd, (uint32_t __user *)ptr)) return -EFAULT; ptr += sizeof(uint32_t);

list_del(&w->entry); kfree(w);

} break; ...... }


        binder_loop函数执行完这个ioctl调用后，才会在下一次调用ioctl进入到Binder驱动程序进入休眠状态，等待下一次Client的请求。

        上面讲到调用IServiceManager::addService的MediaPlayerService的Server主线程被唤醒了，于是，重新执行binder_thread_read函数：

int ret = 0; int wait_for_proc_work;

if (consumed == 0) { if (put_user(BR_NOOP, (uint32_t __user )ptr)) return -EFAULT; ptr += sizeof(uint32_t); }

retry: wait_for_proc_work = thread->transaction_stack == NULL && list_empty(&thread->todo);

......

if (wait_for_proc_work) { ...... } else { if (non_block) { if (!binder_has_thread_work(thread)) ret = -EAGAIN; } else ret = wait_event_interruptible(thread->wait, binder_has_thread_work(thread)); }

......

while (1) { uint32_t cmd; struct binder_transaction_data tr; struct binder_work w; struct binder_transaction t = NULL;

......

switch (w->type) { case BINDER_WORK_TRANSACTION: { t = container_of(w, struct binder_transaction, work); } break; ...... }

if (!t) continue;

BUG_ON(t->buffer == NULL); if (t->buffer->target_node) { ...... } else { tr.target.ptr = NULL; tr.cookie = NULL; cmd = BR_REPLY; } tr.code = t->code; tr.flags = t->flags; tr.sender_euid = t->sender_euid;

if (t->from) { ...... } else { tr.sender_pid = 0; }

if (put_user(cmd, (uint32_t __user *)ptr)) return -EFAULT; ptr += sizeof(uint32_t); if (copy_to_user(ptr, &tr, sizeof(tr))) return -EFAULT; ptr += sizeof(tr);

......

list_del(&t->work.entry); t->buffer->allow_user_free = 1; if (cmd == BR_TRANSACTION && !(t->flags & TF_ONE_WAY)) { ...... } else { t->buffer->transaction = NULL; kfree(t); binder_stats.obj_deleted[BINDER_STAT_TRANSACTION]++; } break; }

done: ...... return 0; }


         在while循环中，从thread->todo得到w，w->type为BINDER_WORK_TRANSACTION，于是，得到t。从上面可以知道，Service Manager反回了一个0回来，写在t->buffer->data里面，现在把t->buffer->data加上proc->user_buffer_offset，得到用户空间地址，保存在tr.data.ptr.buffer里面，这样用户空间就可以访问这个返回码了。由于cmd不等于BR_TRANSACTION，这时就可以把t删除掉了，因为以后都不需要用了。

         执行完这个函数后，就返回到binder_ioctl函数，执行下面语句，把数据返回给用户空间：

if (copy_to_user(ubuf, &bwr, sizeof(bwr))) { ret = -EFAULT; goto err; }


         接着返回到用户空间IPCThreadState::talkWithDriver函数，最后返回到IPCThreadState::waitForResponse函数，最终执行到下面语句：

status_t IPCThreadState::waitForResponse(Parcel reply, status_t acquireResult) { int32_t cmd; int32_t err;

while (1) { if ((err=talkWithDriver()) < NO_ERROR) break;

......

cmd = mIn.readInt32();

......

switch (cmd) { ...... case BR_REPLY: { binder_transaction_data tr; err = mIn.read(&tr, sizeof(tr)); LOG_ASSERT(err == NO_ERROR, "Not enough command data for brREPLY"); if (err != NO_ERROR) goto finish;

if (reply) { if ((tr.flags & TF_STATUS_CODE) == 0) { reply->ipcSetDataReference( reinterpret_cast<const uint8_t>(tr.data.ptr.buffer), tr.data_size, reinterpret_cast<const size_t>(tr.data.ptr.offsets), tr.offsets_size/sizeof(size_t), freeBuffer, this); } else { ...... } } else { ...... } } goto finish;

...... } }

finish: ...... return err; }


        注意，这里的tr.flags等于0，这个是在上面的binder_send_reply函数里设置的。最终把结果保存在reply了：

reply->ipcSetDataReference( reinterpret_cast<const uint8_t>(tr.data.ptr.buffer), tr.data_size, reinterpret_cast<const size_t>(tr.data.ptr.offsets), tr.offsets_size/sizeof(size_t), freeBuffer, this);


       这个函数我们就不看了，有兴趣的读者可以研究一下。

       从这里层层返回，最后回到MediaPlayerService::instantiate函数中。

       至此，IServiceManager::addService终于执行完毕了。这个过程非常复杂，但是如果我们能够深刻地理解这一过程，将能很好地理解Binder机制的设计思想和实现过程。这里，对IServiceManager::addService过程中MediaPlayerService、ServiceManager和BinderDriver之间的交互作一个小结：

![](https://oss-cn-hangzhou.aliyuncs.com/codingsky/cdn/codingsky/upload/img/blog/f32af23589634d80e3bc4f6c1e164d95.gif)  


        回到frameworks/base/media/mediaserver/main_mediaserver.cpp文件中的main函数，接下去还要执行下面两个函数：

ProcessState::self()->startThreadPool(); IPCThreadState::self()->joinThreadPool();


        首先看ProcessState::startThreadPool函数的实现：

void ProcessState::startThreadPool() { AutoMutex _l(mLock); if (!mThreadPoolStarted) { mThreadPoolStarted = true; spawnPooledThread(true); } }


       这里调用spwanPooledThread：

void ProcessState::spawnPooledThread(bool isMain) { if (mThreadPoolStarted) { int32_t s = android_atomic_add(1, &mThreadPoolSeq); char buf[32]; sprintf(buf, "Binder Thread #%d", s); LOGV("Spawning new pooled thread, name=%s\n", buf); sp t = new PoolThread(isMain); t->run(buf); } }


       这里主要是创建一个线程，PoolThread继续Thread类，Thread类定义在frameworks/base/libs/utils/Threads.cpp文件中，其run函数最终调用子类的threadLoop函数，这里即为PoolThread::threadLoop函数：

virtual bool threadLoop() { IPCThreadState::self()->joinThreadPool(mIsMain); return false; }


       这里和frameworks/base/media/mediaserver/main_mediaserver.cpp文件中的main函数一样，最终都是调用了IPCThreadState::joinThreadPool函数，它们的区别是，一个参数是true，一个是默认值false。我们来看一下这个函数的实现：

void IPCThreadState::joinThreadPool(bool isMain) { LOG_THREADPOOL("*** THREAD %p (PID %d) IS JOINING THE THREAD POOL\n", (void)pthread_self(), getpid());

mOut.writeInt32(isMain ? BC_ENTER_LOOPER : BC_REGISTER_LOOPER);

......

status_t result; do { int32_t cmd;

.......

// now get the next command to be processed, waiting if necessary result = talkWithDriver(); if (result >= NO_ERROR) { size_t IN = mIn.dataAvail(); if (IN < sizeof(int32_t)) continue; cmd = mIn.readInt32(); ...... }

result = executeCommand(cmd); }

...... } while (result != -ECONNREFUSED && result != -EBADF);

.......

mOut.writeInt32(BC_EXIT_LOOPER); talkWithDriver(false); }


        这个函数最终是在一个无穷循环中，通过调用talkWithDriver函数来和Binder驱动程序进行交互，实际上就是调用talkWithDriver来等待Client的请求，然后再调用executeCommand来处理请求，而在executeCommand函数中，最终会调用BBinder::transact来真正处理Client的请求：

status_t IPCThreadState::executeCommand(int32_t cmd) { BBinder obj; RefBase::weakref_type refs; status_t result = NO_ERROR;

switch (cmd) { ......

case BR_TRANSACTION: { binder_transaction_data tr; result = mIn.read(&tr, sizeof(tr));

......

Parcel reply;

......

if (tr.target.ptr) { sp b((BBinder*)tr.cookie); const status_t error = b->transact(tr.code, buffer, &reply, tr.flags); if (error < NO_ERROR) reply.setError(error);

} else { const status_t error = the_context_object->transact(tr.code, buffer, &reply, tr.flags); if (error < NO_ERROR) reply.setError(error); }

...... } break;

....... }

if (result != NO_ERROR) { mLastError = result; }

return result; }


        接下来再看一下BBinder::transact的实现：

status_t BBinder::transact( uint32_t code, const Parcel& data, Parcel* reply, uint32_t flags) { data.setDataPosition(0);

status_t err = NO_ERROR; switch (code) { case PING_TRANSACTION: reply->writeInt32(pingBinder()); break; default: err = onTransact(code, data, reply, flags); break; }

if (reply != NULL) { reply->setDataPosition(0); }

return err; }


       最终会调用onTransact函数来处理。在这个场景中，BnMediaPlayerService继承了BBinder类，并且重载了onTransact函数，因此，这里实际上是调用了BnMediaPlayerService::onTransact函数，这个函数定义在frameworks/base/libs/media/libmedia/IMediaPlayerService.cpp文件中：

status_t BnMediaPlayerService::onTransact( uint32_t code, const Parcel& data, Parcel* reply, uint32_t flags) { switch(code) { case CREATE_URL: { ...... } break; case CREATE_FD: { ...... } break; case DECODE_URL: { ...... } break; case DECODE_FD: { ...... } break; case CREATE_MEDIA_RECORDER: { ...... } break; case CREATE_METADATA_RETRIEVER: { ...... } break; case GET_OMX: { ...... } break; default: return BBinder::onTransact(code, data, reply, flags); } }



       至此，我们就以MediaPlayerService为例，完整地介绍了Android系统进程间通信Binder机制中的Server启动过程。Server启动起来之后，就会在一个无穷循环中等待Client的请求了。在下一篇文章中，我们将介绍Client如何通过Service Manager远程接口来获得Server远程接口，进而调用Server远程接口来使用Server提供的服务，敬请关注。

**老罗的新浪微博：[http://weibo.com/shengyangluo](https://www.androidos.net.cn/redirecturl.do?url=http%3A//weibo.com/shengyangluo)，欢迎关注！**