Miracast技术详解(四):Sink源码解析

Sink源码概述

Miracast Sink端源码最早出现在Android 4.2.2上,通过googlesource可以很方便的查看:
https://android.googlesource.com/platform/frameworks/av/+/android-4.2.2_r1.2/media/libstagefright/wifi-display/sink/

但是在Android 4.3以后,Google却移除掉了这部分源码,详细的commit记录在:
https://android.googlesource.com/platform/frameworks/av/+/c4bd06130e4c3068ab58a0be88a4f765c2267563

1
2
3
4
Remove all traces of wifi display sink implementation and supporting code.

Change-Id: I64b681b7e3df1ef0dd80c0d261cacae293d5e684
related-to-bug: 8698812

虽然移除了Sink端代码,但是Source端源码是还在的,我们可以通过Android手机的投射功能实现Miracast投屏发送端。

导入源码

这里推荐使用Android Studio进行源码查看,为了方便使用IDE的代码提示及类/方法跳转等相关功能,我们需要搭建好源码环境。首先新建一个Native Project,然后把整个libstagefright相关的源码拷贝到cpp目录中,最好把相关的include头文件也一起导入(因为涉及到很多依赖),然后在CMakeLists.txt中添加这部分源码。重新sync一次,这样就能引用到相关的类与头文件,并且支持代码提示,提高我们查看源码的效率。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
include_directories(include)

file(GLOB_RECURSE MIRACASTSRC
"media/libstagefright/*.h"
"media/libstagefright/*.cpp"
)

add_library( # Sets the name of the library.
native-lib

# Sets the library as a shared library.
SHARED

# Provides a relative path to your source file(s).
native-lib.cpp ${MIRACASTSRC})

RTFSC

Sink端源码主要的核心类就这3个:WifiDisplaySink.cppRTPSink.cppTunnelRenderer.cpp。我们可以从wfd.cpp这个可执行程序的main()方法看起,看Sink端是如何进行初始化的。

最重要的几行初始化代码在main()函数的最后,几个需要关注的点:

  • 创建了ANetworkSession对象,并启动了内部的NetworkThread,此类主要用来管理通信相关的TCP和UDP连接。
  • 创建并启动了ALooper,并将WifiDisplaySink通过registerHandler()方法关联起来。其中WifiDisplaySinkRTPSinkTunnelRenderer都继承了AHandlr,都实现了onMessageReceived(),可以用来处理异步消息。
  • 整个程序采用了ALooperAHandlerAMessage这几个类配合完成的异步消息机制,它是Android Native层实现的一个异步消息机制,跟应用层的Handler那一套有点类似。
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
sp<ANetworkSession> session = new ANetworkSession;
session->start();

sp<ALooper> looper = new ALooper;

sp<WifiDisplaySink> sink = new WifiDisplaySink(session);
looper->registerHandler(sink);

if (connectToPort >= 0) {
sink->start(connectToHost.c_str(), connectToPort);
} else {
sink->start(uri.c_str());
}

looper->start(true /* runOnCallingThread */);

最后通过关键的sink->start()方法启动WifiDisplaySink,我们来看下里面的操作,以ip和端口这个start()方法为例,post了一个类型为kWhatStartAMessage对象:

1
2
3
4
5
6
void WifiDisplaySink::start(const char *sourceHost, int32_t sourcePort) {
sp<AMessage> msg = new AMessage(kWhatStart, id());
msg->setString("sourceHost", sourceHost);
msg->setInt32("sourcePort", sourcePort);
msg->post();
}

RTSP通讯

前面我们说到了WifiDisplaySink继承了AHandler,因此消息最终会回调到onMessageReceived()方法中去处理。这里通过createRTSPClient()方法创建了RTSP的TCP连接,并传入了一个AMessage对象,以用作后面的连接状态与数据的异步通知:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
void WifiDisplaySink::onMessageReceived(const sp<AMessage> &msg) {
switch (msg->what()) {
case kWhatStart:
{
...
sp<AMessage> notify = new AMessage(kWhatRTSPNotify, id());

status_t err = mNetSession->createRTSPClient(
mRTSPHost.c_str(), sourcePort, notify, &mSessionID);
CHECK_EQ(err, (status_t)OK);

mState = CONNECTING;
break;
}
...
}
}

当RTSP连接失败或成功后,会通过kWhatErrorkWhatConnected进行通知回调。若连接成功建立,则后续会通过kWhatData收到RTSP相关的数据包。这个时候就正式开始RTSP协商与会话建立,涉及到RTSP的M1-M7指令处理。具体的Rqeuest及Response流程可以看前面的RTSP协议分析文章,这里不详细展开。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
void WifiDisplaySink::onMessageReceived(const sp<AMessage> &msg) {
switch (msg->what()) {
...
case kWhatRTSPNotify:
{
int32_t reason;
CHECK(msg->findInt32("reason", &reason));

switch (reason) {
case ANetworkSession::kWhatError:
{
// 处理网络异常状态,忽略
}

case ANetworkSession::kWhatConnected:
{
// 网络连接成功回调,并设置状态为已连接
ALOGI("We're now connected.");
mState = CONNECTED;
...
}

case ANetworkSession::kWhatData:
{
// 处理接收到的RTSP数据包
onReceiveClientData(msg);
break;
}
...
}
break;
}
...
}
}

首先判断消息类型,看看是Request还是Response,这里的做法有点暴力,判断RTSP/开头则认为是Response。详细解析看注释:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
void WifiDisplaySink::onReceiveClientData(const sp<AMessage> &msg) {
...
if (method.startsWith("RTSP/")) {
// This is a response.

ResponseID id;
id.mSessionID = sessionID;
id.mCSeq = cseq;
// mResponseHandlers是key-value结构,用来存储对应请求的ResponseHandler
// 根据RTSP的交互流程,在调用sendM2, sendSetup(M6), sendPlay(M7)的时候由Sink主动发起Request
// 这个时候会调用mResponseHandlers.add()添加对应请求的ResponseHandler
ssize_t index = mResponseHandlers.indexOfKey(id);

if (index < 0) {
ALOGW("Received unsolicited server response, cseq %d", cseq);
return;
}
// 取出对应ResponseID的HandleRTSPResponseFunc,并从映射表中移除
HandleRTSPResponseFunc func = mResponseHandlers.valueAt(index);
mResponseHandlers.removeItemsAt(index);
// 调用对应的HandleRTSPResponseFunc进行回调,这个func我们可以看做一个callback
status_t err = (this->*func)(sessionID, data);
CHECK_EQ(err, (status_t)OK);
} else {
AString version;
data->getRequestField(2, &version);
// RTSP version合法性判断
if (!(version == AString("RTSP/1.0"))) {
sendErrorResponse(sessionID, "505 RTSP Version not supported", cseq);
return;
}
if (method == "OPTIONS") {
// 对应Source端发起的OPTIONS M1请求
onOptionsRequest(sessionID, cseq, data);
} else if (method == "GET_PARAMETER") {
// 对应GET_PARAMETER M3请求
onGetParameterRequest(sessionID, cseq, data);
} else if (method == "SET_PARAMETER") {
// 对应SET_PARAMETER M4与M5请求
onSetParameterRequest(sessionID, cseq, data);
} else {
sendErrorResponse(sessionID, "405 Method Not Allowed", cseq);
}
}
}

针对Request部分,我们继续根据RTSP流程进行细化,onOptionsRequest()主要处理Source端M1请求,并且响应完后马上执行sendM2()方法,以确认Source端所支持的RTSP方法请求。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
void WifiDisplaySink::onOptionsRequest(
int32_t sessionID,
int32_t cseq,
const sp<ParsedMessage> &data) {
AString response = "RTSP/1.0 200 OK\r\n";
AppendCommonResponse(&response, cseq);
response.append("Public: org.wfa.wfd1.0, GET_PARAMETER, SET_PARAMETER\r\n");
response.append("\r\n");

status_t err = mNetSession->sendRequest(sessionID, response.c_str());
CHECK_EQ(err, (status_t)OK);

err = sendM2(sessionID);
CHECK_EQ(err, (status_t)OK);
}

onGetParameterRequest()则是处理Source端M3请求,并返回自持自身支持的属性及能力,比较重要的几个属性:RTP端口号(传输流媒体用)、所支持的audio及video编解码格式等…

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
void WifiDisplaySink::onGetParameterRequest(
int32_t sessionID,
int32_t cseq,
const sp<ParsedMessage> &data) {
AString body =
"wfd_video_formats: xxx\r\n"
"wfd_audio_codecs: xxx\r\n"
"wfd_client_rtp_ports: RTP/AVP/UDP;unicast xxx 0 mode=play\r\n";

AString response = "RTSP/1.0 200 OK\r\n";
AppendCommonResponse(&response, cseq);
response.append("Content-Type: text/parameters\r\n");
response.append(StringPrintf("Content-Length: %d\r\n", body.size()));
response.append("\r\n");
response.append(body);

status_t err = mNetSession->sendRequest(sessionID, response.c_str());
CHECK_EQ(err, (status_t)OK);
}

onSetParameterRequest()则是处理Source端M4与M5请求,对于M4请求,直接进行常规的Response即可。对于M5,除了Response之外,因为请求中带有wfd_trigger_method: SETUP,会触发Sink端向Source端发送SETUP请求。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
void WifiDisplaySink::onSetParameterRequest(
int32_t sessionID,
int32_t cseq,
const sp<ParsedMessage> &data) {
const char *content = data->getContent();

if (strstr(content, "wfd_trigger_method: SETUP\r\n") != NULL) {
status_t err =
sendSetup(
sessionID,
"rtsp://x.x.x.x:x/wfd1.0/streamid=0");

CHECK_EQ(err, (status_t)OK);
}

AString response = "RTSP/1.0 200 OK\r\n";
AppendCommonResponse(&response, cseq);
response.append("\r\n");

status_t err = mNetSession->sendRequest(sessionID, response.c_str());
CHECK_EQ(err, (status_t)OK);
}

sendSetup()方法中,有两个比较重要的点。一是我们初始化了RTPSink,主要用于后续建立UDP连接与处理RTP包。二是在发送完Setup M6请求后,注册了onReceiveSetupResponse()回调。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
status_t WifiDisplaySink::sendSetup(int32_t sessionID, const char *uri) {
mRTPSink = new RTPSink(mNetSession, mSurfaceTex);
looper()->registerHandler(mRTPSink);

status_t err = mRTPSink->init(sUseTCPInterleaving);

if (err != OK) {
looper()->unregisterHandler(mRTPSink->id());
mRTPSink.clear();
return err;
}
...
err = mNetSession->sendRequest(sessionID, request.c_str(), request.size());
...
registerResponseHandler(
sessionID, mNextCSeq, &WifiDisplaySink::onReceiveSetupResponse);
...
return OK;
}

onReceiveSetupResponse()方法的最后我们完成了RTSP中的最后一步,发送PLAY M7请求,告诉发送端可以开始发送流媒体数据了。这个时候Source端会按照Sink指定的UDP端口(SETUP指令中指定)发送RTP数据包,包含音视频数据。

1
2
3
4
5
6
7
8
9
10
11
12
13
status_t WifiDisplaySink::sendPlay(int32_t sessionID, const char *uri) {
AString request = StringPrintf("PLAY %s RTSP/1.0\r\n", uri);

AppendCommonResponse(&request, mNextCSeq);

request.append(StringPrintf("Session: %s\r\n", mPlaybackSessionID.c_str()));
request.append("\r\n");

status_t err =
mNetSession->sendRequest(sessionID, request.c_str(), request.size());
...
return OK;
}

经过以上代码,RTSP的协商与会话建立就已经完成了,并且能在指定的UDP端口中收到音视频RTP数据包。看完这部分Native代码,我们完全可以在应用层用Socket或者Netty这样的第三方网络框架实现。下面我们继续分析RTPSink中是如何处理RTP的数据包的。

RTP通讯

前面我们提到,在sendSetup()方法中,我们初始化了RTPSink,并且在onReceiveSetupResponse()回调中调用configureTransport()方法,根据对应的RTP与RTCP端口,调用mRTPSink->connect()方法建立UDP连接。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
status_t WifiDisplaySink::configureTransport(const sp<ParsedMessage> &msg) {
...
int rtpPort, rtcpPort;
if (sscanf(serverPortStr.c_str(), "%d-%d", &rtpPort, &rtcpPort) != 2
|| rtpPort <= 0 || rtpPort > 65535
|| rtcpPort <=0 || rtcpPort > 65535
|| rtcpPort != rtpPort + 1) {
ALOGE("Invalid server_port description '%s'.",
serverPortStr.c_str());

return ERROR_MALFORMED;
}

if (rtpPort & 1) {
ALOGW("Server picked an odd numbered RTP port.");
}

return mRTPSink->connect(sourceHost.c_str(), rtpPort, rtcpPort);
}

connect()方法中,通过ANetworkSession完成对应RTP与RTCP连接的connect操作。其中RTP与RTCP的Session在init()方法调用的时候已经创建完毕,这里不再展示。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
status_t RTPSink::connect(
const char *host, int32_t remoteRtpPort, int32_t remoteRtcpPort) {
ALOGI("connecting RTP/RTCP sockets to %s:{%d,%d}",
host, remoteRtpPort, remoteRtcpPort);

status_t err =
mNetSession->connectUDPSession(mRTPSessionID, host, remoteRtpPort);

if (err != OK) {
return err;
}

err = mNetSession->connectUDPSession(mRTCPSessionID, host, remoteRtcpPort);

if (err != OK) {
return err;
}
...
}

前面我们提到了RTPSink继承了AHandler,因此创建相关UDPSession传入的kWhatRTPNotifykWhatRTCPNotify消息最终会回调到onMessageReceived()方法中去处理。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
void RTPSink::onMessageReceived(const sp<AMessage> &msg) {
switch (msg->what()) {
case kWhatRTPNotify:
case kWhatRTCPNotify:
{
int32_t reason;
CHECK(msg->findInt32("reason", &reason));

switch (reason) {
case ANetworkSession::kWhatError:
{
// 处理网络异常状态,忽略
}

case ANetworkSession::kWhatDatagram:
{
int32_t sessionID;
CHECK(msg->findInt32("sessionID", &sessionID));

sp<ABuffer> data;
// 取出ABuffer中的RTP或者RTCP数据包
CHECK(msg->findBuffer("data", &data));

status_t err;
// 根据消息类型分别进行RTP与RTCP包的解析
if (msg->what() == kWhatRTPNotify) {
err = parseRTP(data);
} else {
err = parseRTCP(data);
}
break;
}
...
}
break;
}
...
}
}

在前面的文章中我们也提到,经过多台手机的抓包测试,发现抓到的UDP数据中只包含了RTP数据包,而没有发现RTCP数据包。因此我们这里的分析先不考虑RTCP数据包的处理,直接看RTP包的解析即可。详细解析看注释:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
status_t RTPSink::parseRTP(const sp<ABuffer> &buffer) {
size_t size = buffer->size();
if (size < 12) {
// Too short to be a valid RTP header.
// RTP的固定包头至少12字节,包含CSRC的时候会超过12字节
return ERROR_MALFORMED;
}

const uint8_t *data = buffer->data();

// 按照以下RTP固定包头进行解析
// 0 1 2 3
// 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
// |V=2|P|X| CC |M| PT | sequence number |
// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
// | timestamp |
// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
// | synchronization source (SSRC) identifier |
// +=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+
// | contributing source (CSRC) identifiers |
// | .... |
// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
...

// 将解析出来的ssrc、timestamp、payload type、marker等属性设置进buffer的meta域
sp<AMessage> meta = buffer->meta();
meta->setInt32("ssrc", srcId);
meta->setInt32("rtp-time", rtpTime);
meta->setInt32("PT", data[1] & 0x7f);
meta->setInt32("M", data[1] >> 7);
// 包头后剩下的字节为MPEG2-TS Payload
buffer->setRange(payloadOffset, size - payloadOffset);
// 根据SSRC获取对应的Source,在一个RTP传输周期内,相同的同步源将具有相同的SSRC
ssize_t index = mSources.indexOfKey(srcId);
// 若mSources中查找不到,则创建新的Source并加入到映射表中
if (index < 0) {
if (mRenderer == NULL) {
sp<AMessage> notifyLost = new AMessage(kWhatPacketLost, id());
notifyLost->setInt32("ssrc", srcId);
// 创建TunnelRenderer,为后续的音视频渲染做准备
mRenderer = new TunnelRenderer(notifyLost, mSurfaceTex);
looper()->registerHandler(mRenderer);
}

sp<AMessage> queueBufferMsg =
new AMessage(TunnelRenderer::kWhatQueueBuffer, mRenderer->id());

sp<Source> source = new Source(seqNo, buffer, queueBufferMsg);
mSources.add(srcId, source);
} else {
// 若已经创建好了Source,直接将buffer入队
mSources.valueAt(index)->updateSeq(seqNo, buffer);
}

return OK;
}

updateSeq()方法中,会对sequence包序列号进行一些处理,如:丢包间隙过大、重复包、64K周期循环处理等,最终通过queuePacket()方法将MPEG2-TS数据包入队。其中mQueueBufferMsgTunnelRenderer::kWhatQueueBuffer类型的消息,最终会在TunnelRenderer中进行处理。

1
2
3
4
5
void RTPSink::Source::queuePacket(const sp<ABuffer> &buffer) {
sp<AMessage> msg = mQueueBufferMsg->dup();
msg->setBuffer("buffer", buffer);
msg->post();
}

播放阶段

数据包来到了TunnelRenderer,它代表一个呈现通道,会接收queuePacket()解包出来的MPEG2-TS音视频流,然后进行播放。其中TunnelRenderer继承了AHandler,因此消息最终会回调到onMessageReceived()方法中去处理。里面的处理也比较简单,主要是buffer入队操作及环境初始化。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
void TunnelRenderer::onMessageReceived(const sp<AMessage> &msg) {
switch (msg->what()) {
case kWhatQueueBuffer:
{
sp<ABuffer> buffer;
CHECK(msg->findBuffer("buffer", &buffer));
// 将buffer数据入队
queueBuffer(buffer);
// 判断StreamSource为空则初始化MediaPlayerService等渲染环境
if (mStreamSource == NULL) {
if (mTotalBytesQueued > 0ll) {
initPlayer();
} else {
ALOGI("Have %lld bytes queued...", mTotalBytesQueued);
}
} else {
// 主要是将TS数据出队,并解析成音视频裸流进行渲染
mStreamSource->doSomeWork();
}
break;
}
...
}
}

其中queueBuffer()操作主要将TS包添加到mPackets列表中,其中还对包进行了一些去重与重排序,保证包序列递增。而initPlayer()则对播放环境进行了初始化。详细注释如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
void TunnelRenderer::initPlayer() {
// 若mSurfaceTex为空(初始化WifiDisplaySink的时候没有传递mSurfaceTex),则由TunnelRenderer自行创建
// 这里也意味着想要实现自己的Sink端,我们可以在应用层创建好SurfaceTexture,并一步步传递下来
if (mSurfaceTex == NULL) {
mComposerClient = new SurfaceComposerClient;
CHECK_EQ(mComposerClient->initCheck(), (status_t)OK);

DisplayInfo info;
SurfaceComposerClient::getDisplayInfo(0, &info);
ssize_t displayWidth = info.w;
ssize_t displayHeight = info.h;
// 创建SurfaceControl,并从中获取Surface实例
mSurfaceControl =
mComposerClient->createSurface(
String8("A Surface"),
displayWidth,
displayHeight,
PIXEL_FORMAT_RGB_565,
0);
...
mSurface = mSurfaceControl->getSurface();
CHECK(mSurface != NULL);
}
...
// 通过系统MediaPlayerService创建MediaPlayer实例
mPlayer = service->create(getpid(), mPlayerClient, 0);
CHECK(mPlayer != NULL);
// 设置DataSource数据流
CHECK_EQ(mPlayer->setDataSource(mStreamSource), (status_t)OK);
// 对MediaPlayer设置SurfaceTexture,这样就能在对应的Surface上播放视频了
mPlayer->setVideoSurfaceTexture(
mSurfaceTex != NULL ? mSurfaceTex : mSurface->getSurfaceTexture());
// 启动MediaPlayer,开始播放
mPlayer->start();
}

TS包入队之后,会通过mStreamSource->doSomeWork()方法将TS包出队,并解析成音视频裸流进行渲染。详细解释如下注释:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
void TunnelRenderer::StreamSource::doSomeWork() {
Mutex::Autolock autoLock(mLock);

while (!mIndicesAvailable.empty()) {
// 将TS包出队
sp<ABuffer> srcBuffer = mOwner->dequeueBuffer();
if (srcBuffer == NULL) {
break;
}
...
ALOGV("dequeue TS packet of size %d", srcBuffer->size());
// 获取可用buffer的index
size_t index = *mIndicesAvailable.begin();
mIndicesAvailable.erase(mIndicesAvailable.begin());
// 获取buffer对应的IMemory对象
sp<IMemory> mem = mBuffers.itemAt(index);
CHECK_LE(srcBuffer->size(), mem->size());
// 检查Buffer大小,必须是188整数倍,因为单个TS包大小为188B
CHECK_EQ((srcBuffer->size() % 188), 0u);
// 通过内存拷贝的形式,将buffer发送到MediaPlayer进行解码与播放
memcpy(mem->pointer(), srcBuffer->data(), srcBuffer->size());
mListener->queueBuffer(index, srcBuffer->size());
}
}

MPEG2-TS解析

Native Sink源码中,最终会通过ATSParser.cpp对TS包进行解析,拿出最终的音视频裸流。首先我们来看下调用的入口,在MPEG2TSExtractor::feedMore()方法中,会把数据包切割成一个个188B的标准TS包,并通过feedTSPacket()送入ATSParser开始解析。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
status_t MPEG2TSExtractor::feedMore() {
Mutex::Autolock autoLock(mLock);

uint8_t packet[kTSPacketSize];
ssize_t n = mDataSource->readAt(mOffset, packet, kTSPacketSize);

if (n < (ssize_t)kTSPacketSize) {
return (n < 0) ? (status_t)n : ERROR_END_OF_STREAM;
}

mOffset += n;
return mParser->feedTSPacket(packet, kTSPacketSize);
}

status_t ATSParser::feedTSPacket(const void *data, size_t size) {
CHECK_EQ(size, kTSPacketSize);

ABitReader br((const uint8_t *)data, kTSPacketSize);
return parseTS(&br);
}

开始解析TS包(对TS包格式不熟悉的,可以查看之前MPEG2-TS流解析文章),详细解析过程如下注释:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
status_t ATSParser::parseTS(ABitReader *br) {
ALOGV("---");
// 检查TS包头的8位同步字节,固定为0x47
unsigned sync_byte = br->getBits(8);
CHECK_EQ(sync_byte, 0x47u);

MY_LOGV("transport_error_indicator = %u", br->getBits(1));
// 获取payload_unit_start_indicator,标记是否为一帧的起始
unsigned payload_unit_start_indicator = br->getBits(1);
ALOGV("payload_unit_start_indicator = %u", payload_unit_start_indicator);

MY_LOGV("transport_priority = %u", br->getBits(1));
// 获取TS包的PID
unsigned PID = br->getBits(13);
ALOGV("PID = 0x%04x", PID);

MY_LOGV("transport_scrambling_control = %u", br->getBits(2));
// 获取适配域
unsigned adaptation_field_control = br->getBits(2);
ALOGV("adaptation_field_control = %u", adaptation_field_control);

unsigned continuity_counter = br->getBits(4);
ALOGV("PID = 0x%04x, continuity_counter = %u", PID, continuity_counter);
// 判断适配域,'10'表示仅有适配域,'11'表示适配域和Payload都存在
if (adaptation_field_control == 2 || adaptation_field_control == 3) {
// 开始解析适配域
parseAdaptationField(br, PID);
}

status_t err = OK;
// 判断适配域,'01'表示仅有Payload,'11'表示适配域和Payload都存在
if (adaptation_field_control == 1 || adaptation_field_control == 3) {
// 开始根据PID解析Payload
err = parsePID(
br, PID, continuity_counter, payload_unit_start_indicator);
}

++mNumTSPacketsParsed;

return err;
}

其中对适配域的解析parseAdaptationField()中比较关键的就是对PCR时钟的解析:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
void ATSParser::parseAdaptationField(ABitReader *br, unsigned PID) {
unsigned adaptation_field_length = br->getBits(8);

if (adaptation_field_length > 0) {
...
// 解析PCR flag,判断适配域头后是否有PCR时钟
unsigned PCR_flag = br->getBits(1);
...
if (PCR_flag) {
br->skipBits(4);
// 解析33比特的低精度部分
uint64_t PCR_base = br->getBits(32);
PCR_base = (PCR_base << 1) | br->getBits(1);

br->skipBits(6);
// 解析9比特的高精度部分
unsigned PCR_ext = br->getBits(9);
...
// 计算出最终的PCR时钟
uint64_t PCR = PCR_base * 300 + PCR_ext;
...
// 更新本地PCR时钟统计
for (size_t i = 0; i < mPrograms.size(); ++i) {
updatePCR(PID, PCR, byteOffsetFromStart);
}
...
}
...
}
}

ATSParser::parsePID()方法开始根据PID对PAT与PMT进行解析,详细解析如下注释:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
status_t ATSParser::parsePID(
ABitReader *br, unsigned PID,
unsigned continuity_counter,
unsigned payload_unit_start_indicator) {
// 首先根据PID找出PSI,目前只用到了PAT与PMT这两类PSI
// 其中PAT的PID固定为0,在ATSParser初始化初期已经add进去
ssize_t sectionIndex = mPSISections.indexOfKey(PID);

if (sectionIndex >= 0) {
const sp<PSISection> &section = mPSISections.valueAt(sectionIndex);
...
ABitReader sectionBits(section->data(), section->size());
// 若PID为0,则以PAT的格式进行解析
if (PID == 0) {
parseProgramAssociationTable(&sectionBits);
} else {
// 否则将以PMT的格式进行解析
bool handled = false;
for (size_t i = 0; i < mPrograms.size(); ++i) {
status_t err;
// parsePSISection()会调用parseProgramMap()解析PMT
if (!mPrograms.editItemAt(i)->parsePSISection(
PID, &sectionBits, &err)) {
continue;
}
...
}
// 处理成功则移除PID对应的PSISections
if (!handled) {
mPSISections.removeItem(PID);
}
}
...
}
// 遍历所有节目,对音视频的ES流进行解析
bool handled = false;
for (size_t i = 0; i < mPrograms.size(); ++i) {
status_t err;
if (mPrograms.editItemAt(i)->parsePID(
PID, continuity_counter, payload_unit_start_indicator,
br, &err)) {
if (err != OK) {
return err;
}

handled = true;
break;
}
}
...
}

我们先来分析下PAT的解析过程,主要是完成[节目编号->PID]的映射解析,详见注释:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
void ATSParser::parseProgramAssociationTable(ABitReader *br) {
... // 此处省略PAT常规字段的解析
// 遍历节目列表进行[节目编号->PID]的映射解析
for (size_t i = 0; i < numProgramBytes / 4; ++i) {
unsigned program_number = br->getBits(16);
ALOGV(" program_number = %u", program_number);

MY_LOGV(" reserved = %u", br->getBits(3));
// 节目编号0的时候为network_PID,这里没有用到,忽略
if (program_number == 0) {
MY_LOGV(" network_PID = 0x%04x", br->getBits(13));
} else {
// 开始对应[节目编号->PID]的映射解析
unsigned programMapPID = br->getBits(13);

ALOGV(" program_map_PID = 0x%04x", programMapPID);

bool found = false;
for (size_t index = 0; index < mPrograms.size(); ++index) {
const sp<Program> &program = mPrograms.itemAt(index);

if (program->number() == program_number) {
program->updateProgramMapPID(programMapPID);
found = true;
break;
}
}
// 添加节目列表
if (!found) {
mPrograms.push(
new Program(this, program_number, programMapPID));
}
// 添加节目对应的PSISection
if (mPSISections.indexOfKey(programMapPID) < 0) {
mPSISections.add(programMapPID, new PSISection);
}
}
}
...
}

紧接着是PMT的解析,主要是完成[ES流->PID]的映射解析,并创建对应的Stream对象,详见注释:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
status_t ATSParser::Program::parseProgramMap(ABitReader *br) {
... // 此处省略PMT常规字段的解析
// infoBytesRemaining is the number of bytes that make up the
// variable length section of ES_infos. It does not include the
// final CRC.
size_t infoBytesRemaining = section_length - 9 - program_info_length - 4;
// 遍历所有Stream,完成[ES流->PID]的映射解析
while (infoBytesRemaining > 0) {
CHECK_GE(infoBytesRemaining, 5u);
// 这里最重要的就是streamType与PID的解析
unsigned streamType = br->getBits(8);
ALOGV(" stream_type = 0x%02x", streamType);
...
unsigned elementaryPID = br->getBits(13);
ALOGV(" elementary_PID = 0x%04x", elementaryPID);
...
// 添加StreamInfo到列表中
StreamInfo info;
info.mType = streamType;
info.mPID = elementaryPID;
infos.push(info);

infoBytesRemaining -= 5 + ES_info_length;
}
... // 此处省略PID改变导致的异常的处理,基本可以忽略

// 遍历StreamInfo,根据PID添加对应的音视频Stream
for (size_t i = 0; i < infos.size(); ++i) {
StreamInfo &info = infos.editItemAt(i);

ssize_t index = mStreams.indexOfKey(info.mPID);

if (index < 0) {
sp<Stream> stream = new Stream(
this, info.mPID, info.mType, PCR_PID);

mStreams.add(info.mPID, stream);
}
}

return OK;
}

然后就到了音视频流的解析,调用对应Streamparse()方法开始进行解析:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
bool ATSParser::Program::parsePID(
unsigned pid, unsigned continuity_counter,
unsigned payload_unit_start_indicator,
ABitReader *br, status_t *err) {
*err = OK;

ssize_t index = mStreams.indexOfKey(pid);
if (index < 0) {
return false;
}
// 根据PID查找出对应的Stream,开始解析
*err = mStreams.editValueAt(index)->parse(
continuity_counter, payload_unit_start_indicator, br);

return true;
}

Stream解析的过程中,最重要的是payload_unit_start_indicator这个参数。前面我们说到,要获得一帧完整的数据,就需要把连续几个TS包里的Payload数据全部取出来,才能组合成一个PES包。那么payload_unit_start_indicator值为1时代表一个完整的音视频数据包的开始。那么从这里开始,直到下一个值为1的包为止(相同PID的ES流),把所有的这些TS包组合起来就是一个完整的PES包。了解了这个要点,看下面的代码就比较简单了:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
status_t ATSParser::Stream::parse(
unsigned continuity_counter,
unsigned payload_unit_start_indicator, ABitReader *br) {
...
if (payload_unit_start_indicator) {
// payload_unit_start_indicator值为1并且之前已经started了,则触发flush()开始解析PES
// 因为这个时候已经组装完一个完成的PES包了
if (mPayloadStarted) {
// Otherwise we run the danger of receiving the trailing bytes
// of a PES packet that we never saw the start of and assuming
// we have a a complete PES packet.

status_t err = flush();

if (err != OK) {
return err;
}
}

mPayloadStarted = true;
}

if (!mPayloadStarted) {
return OK;
}
... // 这部分省略的代码主要是对PES的buffer进行组装

return OK;
}

flush()操作调用了parsePES()开始对PES包进行解析,然后解析出PES data,详细解析如下注释:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
status_t ATSParser::Stream::parsePES(ABitReader *br) {
...
// 解析Stream ID
unsigned stream_id = br->getBits(8);
ALOGV("stream_id = 0x%02x", stream_id);
// 获取PES包的长度,注意这里的长度为后续Payload长度,不包含自身,单位为字节
unsigned PES_packet_length = br->getBits(16);
ALOGV("PES_packet_length = %u", PES_packet_length);
// 过滤掉非音视频PES,开始解析
if (stream_id != 0xbc // program_stream_map
&& stream_id != 0xbe // padding_stream
&& stream_id != 0xbf // private_stream_2
&& stream_id != 0xf0 // ECM
&& stream_id != 0xf1 // EMM
&& stream_id != 0xff // program_stream_directory
&& stream_id != 0xf2 // DSMCC
&& stream_id != 0xf8) { // H.222.1 type E
... // 省略部分Header字段解析
if (PTS_DTS_flags == 2 || PTS_DTS_flags == 3) {
... // 解析PTS,其中2代表'10',只有PTS;3代表'11',PTS与DTS都有
ALOGV("PTS = 0x%016llx (%.2f)", PTS, PTS / 90000.0);

optional_bytes_remaining -= 5;

if (PTS_DTS_flags == 3) {
... // 解析DTS
ALOGV("DTS = %llu", DTS);

optional_bytes_remaining -= 5;
}
}
... // 省略部分Header字段解析
// ES data follows.
// 开始解析PES data
if (PES_packet_length != 0) {
CHECK_GE(PES_packet_length, PES_header_data_length + 3);
// PES data大小:包长度 - PES固定头长度3 - Header的data长度
unsigned dataLength =
PES_packet_length - 3 - PES_header_data_length;
// 对剩余字节数进行校验,确保与上面的dataLength一致
if (br->numBitsLeft() < dataLength * 8) {
ALOGE("PES packet does not carry enough data to contain "
"payload. (numBitsLeft = %d, required = %d)",
br->numBitsLeft(), dataLength * 8);

return ERROR_MALFORMED;
}

CHECK_GE(br->numBitsLeft(), dataLength * 8);
// 通过onPayloadData()回调音视频PES data
onPayloadData(
PTS_DTS_flags, PTS, DTS, br->data(), dataLength);

br->skipBits(dataLength * 8);
} else {
// PES_packet_length为0的情况下,把剩余的字节当做PES data
onPayloadData(
PTS_DTS_flags, PTS, DTS,
br->data(), br->numBitsLeft() / 8);

size_t payloadSizeBits = br->numBitsLeft();
CHECK_EQ(payloadSizeBits % 8, 0u);

ALOGV("There's %d bytes of payload.", payloadSizeBits / 8);
}
}
...
return OK;
}

总结

最终,通过onPayloadData()回调音视频裸流给MediaPlayer进行解码,进行音视频数据的播放,整个Native Sink端的流程就到此结束了。相信看完上面所有源码解析后,自己写这部分逻辑也不是难事,当然更好的办法肯定是基于Sink端的代码进行移植。

移植Native Sink的难点主要是对Native相关的依赖代码进行隔离,如:ALooperAHandler异步消息机制、ANetworkSession网络连接部分与foundation包下的相关的实现等,且移植需要一定的C/C++与NDK开发能力。这里建议可以通过Android应用层实现RTSP连接(Socket/Netty)、音视频解码(MediaCodec/FFmpeg)与渲染(SurfaceView/TextureView),然后把底层的RTP、MPEG2-TS解析部分的代码进行移植,这样可以大大减少Native相关的依赖,提高移植效率。