Android IPC 机制详解:IBinder

http://www.linuxgraphics.cn/android/ipc_ibinder.html

IBinder 接口

IBinder接口是对跨进程的对象的抽象。普通对象在当前进程可以访问，如果希望对象能被其它进程访问，那就必须实现IBinder接口。IBinder接口可以指向本地对象，也可以指向远程对象，调用者不需要关心指向的对象是本地的还是远程。

transact是IBinder接口中一个比较重要的函数，它的函数原型如下：

virtual status_t transact(uint32_t code, const Parcel& data, Parcel* reply, uint32_t flags = 0) = 0;

android中的IPC的基本模型是基于客户/服务器(C/S)架构的。

客户端

请求通过内核模块中转

服务端

如果IBinder指向的是一个客户端代理，那transact只是把请求发送给服务器。服务端的IBinder的transact则提供了实际的服务。

客户端

BpBinder是远程对象在当前进程的代理，它实现了IBinder接口。它的transact 函数实现如下：

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;
}

参数说明：

code 是请求的ID号;

data 是请求的参数;

reply 是返回的结果;

flags 一些额外的标识，如FLAG_ONEWAY。通常为0。

transact只是简单的调用了IPCThreadState::self()的transact，在 IPCThreadState::transact中：

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 (reply) {
err = waitForResponse(reply);
} else {
Parcel fakeReply;
err = waitForResponse(&fakeReply);
}
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;
}
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(tr.data.ptr.buffer),
tr.data_size,
reinterpret_cast(tr.data.ptr.offsets),
tr.offsets_size/sizeof(size_t),
freeBuffer, this);
} else {
err = *static_cast(tr.data.ptr.buffer);
freeBuffer(NULL,
reinterpret_cast(tr.data.ptr.buffer),
tr.data_size,
reinterpret_cast(tr.data.ptr.offsets),
tr.offsets_size/sizeof(size_t), this);
}
} else {
freeBuffer(NULL,
reinterpret_cast(tr.data.ptr.buffer),
tr.data_size,
reinterpret_cast(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;
}

这里transact把请求经内核模块发送了给服务端，服务端处理完请求之后，沿原路返回结果给调用者。这里也可以看出请求是同步操作，它会等待直到结果返回为止。

在BpBinder之上进行简单包装，我们可以得到与服务对象相同的接口，调用者无需要关心调用的对象是远程的还是本地的。拿ServiceManager来说：
(frameworks/base/libs/utils/IServiceManager.cpp)

class BpServiceManager : public BpInterface
{
public:
BpServiceManager(const sp& impl)
: BpInterface(impl)
{
}
...
virtual status_t addService(const String16& name, const sp& 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.readInt32() : err;
}
...
};

BpServiceManager实现了
IServiceManager和IBinder两个接口，调用者可以把BpServiceManager的对象看作是一个
IServiceManager对象或者IBinder对象。当调用者把BpServiceManager对象当作IServiceManager对象使用时，所有的请求只是对BpBinder::transact的封装。这样的封装使得调用者不需要关心IServiceManager对象是本地的还是远程的了。

客户通过defaultServiceManager函数来创建BpServiceManager对象：
(frameworks/base/libs/utils/IServiceManager.cpp)

sp<IServiceManager> defaultServiceManager()
{
if (gDefaultServiceManager != NULL) return gDefaultServiceManager;
{
AutoMutex _l(gDefaultServiceManagerLock);
if (gDefaultServiceManager == NULL) {
gDefaultServiceManager = interface_cast<IServiceManager>(
ProcessState::self()->getContextObject(NULL));
}
}
return gDefaultServiceManager;
}

先通过ProcessState::self()->getContextObject(NULL)创建一个Binder对象，然后通过
interface_cast和IMPLEMENT_META_INTERFACE(ServiceManager,
“android.os.IServiceManager”)把Binder对象包装成
IServiceManager对象。原理上等同于创建了一个BpServiceManager对象。

ProcessState::self()->getContextObject调用ProcessState::getStrongProxyForHandle创建代理对象：

sp<IBinder> ProcessState::getStrongProxyForHandle(int32_t handle)
{
sp<IBinder> result;
AutoMutex _l(mLock);
handle_entry* e = lookupHandleLocked(handle);
if (e != NULL) {
// We need to create a new BpBinder if there isn't currently one, OR we
// are unable to acquire a weak reference on this current one.  See comment
// in getWeakProxyForHandle() for more info about this.
IBinder* b = e->binder;
if (b == NULL || !e->refs->attemptIncWeak(this)) {
b = new BpBinder(handle);
e->binder = b;
if (b) e->refs = b->getWeakRefs();
result = b;
} else {
// This little bit of nastyness is to allow us to add a primary
// reference to the remote proxy when this team doesn't have one
// but another team is sending the handle to us.
result.force_set(b);
e->refs->decWeak(this);
}
}
return result;
}

如果handle为空，默认为context_manager对象，context_manager实际上就是 ServiceManager。

服务端

服务端也要实现IBinder接口，BBinder类对IBinder接口提供了部分默认实现，其中transact的实现如下：

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;
}

PING_TRANSACTION请求用来检查对象是否还存在，这里简单的把
pingBinder的返回值返回给调用者。其它的请求交给onTransact处理。onTransact是BBinder里声明的一个
protected类型的虚函数，这个要求它的子类去实现。比如CameraService里的实现如下：

status_t CameraService::onTransact(
uint32_t code, const Parcel& data, Parcel* reply, uint32_t flags)
{
// permission checks...
switch (code) {
case BnCameraService::CONNECT:
IPCThreadState* ipc = IPCThreadState::self();
const int pid = ipc->getCallingPid();
const int self_pid = getpid();
if (pid != self_pid) {
// we're called from a different process, do the real check
if (!checkCallingPermission(
String16("android.permission.CAMERA")))
{
const int uid = ipc->getCallingUid();
LOGE("Permission Denial: "
"can't use the camera pid=%d, uid=%d", pid, uid);
return PERMISSION_DENIED;
}
}
break;
}
status_t err = BnCameraService::onTransact(code, data, reply, flags);
LOGD("+++ onTransact err %d code %d", err, code);
if (err == UNKNOWN_TRANSACTION || err == PERMISSION_DENIED) {
// the 'service' command interrogates this binder for its name, and then supplies it
// even for the debugging commands.  that means we need to check for it here, using
// ISurfaceComposer (since we delegated the INTERFACE_TRANSACTION handling to
// BnSurfaceComposer before falling through to this code).
LOGD("+++ onTransact code %d", code);
CHECK_INTERFACE(ICameraService, data, reply);
switch(code) {
case 1000:
{
if (gWeakHeap != 0) {
sp h = gWeakHeap.promote();
IMemoryHeap *p = gWeakHeap.unsafe_get();
LOGD("CHECKING WEAK REFERENCE %p (%p)", h.get(), p);
if (h != 0)
h->printRefs();
bool attempt_to_delete = data.readInt32() == 1;
if (attempt_to_delete) {
// NOT SAFE!
LOGD("DELETING WEAK REFERENCE %p (%p)", h.get(), p);
if (p) delete p;
}
return NO_ERROR;
}
}
break;
default:
break;
}
}
return err;
}

由此可见，服务端的onTransact是一个请求分发函数，它根据请求码(code)做相应的处理。

消息循环

服务端(任何进程都可以作为服务端)有一个线程监听来自客户端的请求，并循环处理这些请求。
如果在主线程中处理请求，可以直接调用下面的函数：

IPCThreadState::self()->joinThreadPool(mIsMain);

如果想在非主线程中处理请求，可以按下列方式：

sp
proc = ProcessState::self();
if (proc->supportsProcesses()) {
LOGV("App process: starting thread pool./n");
proc->startThreadPool();
}

startThreadPool的实现原理：

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

这里创建了PoolThread的对象，实现上就是创建了一个线程。所有的线程类都要实现threadLoop虚函数。PoolThread的threadLoop的实现如下：

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

上述代码，简而言之就是创建了一个线程，然后在线程里调用 IPCThreadState::self()->joinThreadPool函数。

下面再看joinThreadPool的实现：

do
{
...
result = talkWithDriver();
if (result >= NO_ERROR) {
size_t IN = mIn.dataAvail();
if (IN < sizeof(int32_t)) continue;
cmd = mIn.readInt32();
IF_LOG_COMMANDS() {
alog << "Processing top-level Command: "
<< getReturnString(cmd) << endl;
}
result = executeCommand(cmd);
}
...
while(...);

这个函数在循环中重复执行下列动作：

talkWithDriver 通过ioctl(mProcess->mDriverFD, BINDER_WRITE_READ,
&bwr)读取请求和写回结果。

executeCommand 执行相应的请求

在IPCThreadState::executeCommand(int32_t cmd)函数中：

对于控制对象生命周期的请求，像BR_ACQUIRE/BR_RELEASE直接做了处理。

对于BR_TRANSACTION请求，它调用被请求对象的transact函数。

按下列方式调用实际的对象：

if (tr.target.ptr) {
sp<BBinder> b((BBinder*)tr.cookie);
const status_t error = b->transact(tr.code, buffer, &reply, 0);
if (error < NO_ERROR) reply.setError(error);
} else {
const status_t error = the_context_object->transact(tr.code, buffer, &reply, 0);
if (error < NO_ERROR) reply.setError(error);
}

如果tr.target.ptr不为空，就把tr.cookie转换成一个Binder对象，并调用它的 transact函数。如果没有目标对象，就调用
the_context_object对象的 transact函数。奇怪的是，根本没有谁对the_context_object进行初始化，
the_context_object是空指针。原因是context_mgr的请求发给了 ServiceManager，所以根本不会走到else
语句里来。

内核模块

android使用了一个内核模块binder来中转各个进程之间的消息。模块源代码放在binder.c里，它是一个字符驱动程序，主要通过
binder_ioctl与用户空间的进程交换数据。其中BINDER_WRITE_READ用来读写数据，数据包中有一个cmd域用于区分不同的请求：

binder_thread_write用于发送请求或返回结果。

binder_thread_read用于读取结果。

从binder_thread_write中调用binder_transaction中转请求和返回结果，binder_transaction的实现如下：

对请求的处理：

通过对象的handle找到对象所在的进程，如果handle为空就认为对象是context_mgr，把请求发给context_mgr所在的进程。

把请求中所有的binder对象全部放到一个RB树中。

把请求放到目标进程的队列中，等待目标进程读取。

如何成为context_mgr呢？内核模块提供了BINDER_SET_CONTEXT_MGR调用:

static long binder_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
{
...
case BINDER_SET_CONTEXT_MGR:
if (binder_context_mgr_node != NULL) {
printk(KERN_ERR "binder: BINDER_SET_CONTEXT_MGR already set/n");
ret = -EBUSY;
goto err;
}
if (binder_context_mgr_uid != -1) {
if (binder_context_mgr_uid != current->euid) {
printk(KERN_ERR "binder: BINDER_SET_"
"CONTEXT_MGR bad uid %d != %d/n",
current->euid,
binder_context_mgr_uid);
ret = -EPERM;
goto err;
}
} else
binder_context_mgr_uid = current->euid;
binder_context_mgr_node = binder_new_node(proc, NULL, NULL);
if (binder_context_mgr_node == NULL) {
ret = -ENOMEM;
goto err;
}
binder_context_mgr_node->local_weak_refs++;
binder_context_mgr_node->local_strong_refs++;
binder_context_mgr_node->has_strong_ref = 1;
binder_context_mgr_node->has_weak_ref = 1;
break;

ServiceManager（frameworks/base/cmds/servicemanager)通过下列方式成为了
context_mgr进程：

int binder_become_context_manager(struct binder_state *bs)
{
return ioctl(bs->fd, BINDER_SET_CONTEXT_MGR, 0);
}
int main(int argc, char **argv)
{
struct binder_state *bs;
void *svcmgr = BINDER_SERVICE_MANAGER;
bs = binder_open(128*1024);
if (binder_become_context_manager(bs)) {
LOGE("cannot become context manager (%s)/n", strerror(errno));
return -1;
}
svcmgr_handle = svcmgr;
binder_loop(bs, svcmgr_handler);
return 0;
}

如何得到服务对象的handle

服务提供者通过defaultServiceManager得到ServiceManager对象，然后调用addService向服务管理器注册。

服务使用者通过defaultServiceManager得到ServiceManager对象，然后调用getService通过服务名称查找到服务对象的handle。

如何通过服务对象的handle找到服务所在的进程

表示服务管理器的handle，getService可以查找到系统服务的handle。这个handle只是代表了服务对象，内核模块是如何通过handle找到服务所在的进程的呢？

对于ServiceManager:
ServiceManager调用了binder_become_context_manager使用自己成为context_mgr，所有handle为0的请求都会被转发给ServiceManager。

对于系统服务和应用程序的Listener，在第一次请求内核模块时(比如调用
addService)，内核模块在一个RB树中建立了服务对象和进程的对应关系。

请求服务时，内核先通过handle找到对应的进程，然后把请求放到服务进程的队列中。

off_end = (void *)offp + tr->offsets_size;
for (; offp < off_end; offp++) {
struct flat_binder_object *fp;
if (*offp > t->buffer->data_size - sizeof(*fp)) {
binder_user_error("binder: %d:%d got transaction with "
"invalid offset, %d/n",
proc->pid, thread->pid, *offp);
return_error = BR_FAILED_REPLY;
goto err_bad_offset;
}
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) {
binder_user_error("binder: %d:%d sending u%p "
"node %d, cookie mismatch %p != %p/n",
proc->pid, thread->pid,
fp->binder, node->debug_id,
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);
if (binder_debug_mask & BINDER_DEBUG_TRANSACTION)
printk(KERN_INFO "        node %d u%p -> ref %d desc %d/n",
node->debug_id, node->ptr, ref->debug_id, ref->desc);
} break;

C调用JAVA

前面我们分析的是C代码的处理。对于JAVA代码，JAVA调用C的函数通过JNI调用即可。从内核时读取请求是在C代码(executeCommand)里进行了，那如何在C代码中调用那些用JAVA实现的服务呢？
android_os_Binder_init里的JavaBBinder对Java里的Binder对象进行包装。

JavaBBinder::onTransact调用Java里的execTransact函数：

jboolean res = env->CallBooleanMethod(mObject, gBinderOffsets.mExecTransact,
code, (int32_t)&data, (int32_t)reply, flags);
jthrowable excep = env->ExceptionOccurred();
if (excep) {
report_exception(env, excep,
"*** Uncaught remote exception!  "
"(Exceptions are not yet supported across processes.)");
res = JNI_FALSE;
/* clean up JNI local ref -- we don't return to Java code */
env->DeleteLocalRef(excep);
}

广播消息

binder不提供广播消息，不过可以ActivityManagerService服务来实现广播。
(frameworks/base/core/java/android/app/ActivityManagerNative.java)

接收广播消息需要实现接口BroadcastReceiver，然后调用ActivityManagerProxy::registerReceiver注册。

触发广播调用ActivityManagerProxy::broadcastIntent。(应用程序并不直接调用它，而是调用Context对它的包装)

Reference

文章出处：http://www.limodev.cn/blog
作者联系方式：李先静 <xianjimli at hotmail dot com>