Misc Notes

Chromium学习笔记:程序启动入口分析(Windows)

以下笔记内容均为Windows版本。

本篇笔记跟踪记录了Chromium的启动过程,主要关注 Browser 进程和 Renderer 进程。根据 Chromium 项目的分层设计,我们把 Content API 称作为 Content 层,而把调用 Content API 实现浏览器程序的部分称作为 Embedder 层。在项目中,Embedder 层有 chromecontent_shell 等多种实现。

1、main() 函数

Chromium的main函数在 chrome\app\chrome_exe_main_win.cc,具体如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
// chrome\app\chrome_exe_main_win.cc

#if !defined(WIN_CONSOLE_APP)
int APIENTRY wWinMain(HINSTANCE instance, HINSTANCE prev, wchar_t*, int) {
#else
int main() {
HINSTANCE instance = GetModuleHandle(nullptr);
#endif
install_static::InitializeFromPrimaryModule();
SignalInitializeCrashReporting();

......

// Load and launch the chrome dll. *Everything* happens inside.
VLOG(1) << "About to load main DLL.";
MainDllLoader* loader = MakeMainDllLoader();
int rc = loader->Launch(instance, exe_entry_point_ticks);
loader->RelaunchChromeBrowserWithNewCommandLineIfNeeded();
delete loader;
return rc;
}

在main函数中,最重要的一步,就是 int rc = loader->Launch(instance, exe_entry_point_ticks); 载入 chrome.dll运行

2、载入 chrome.dll

在这里首先调用了 MakeMainDllLoader() 函数,这是一个静态函数,在chrome\app\main_dll_loader.cc 中,内容如下:

1
2
3
4
5
6
7
8
9
// chrome\app\main_dll_loader.cc

MainDllLoader* MakeMainDllLoader() {
#if defined(GOOGLE_CHROME_BUILD)
return new ChromeDllLoader();
#else
return new ChromiumDllLoader();
#endif
}

函数创建并返回一个 ChromiumDllLoader,紧接着再调用它的 Launch 函数,内容如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
// chrome\app\main_dll_loader.cc

int MainDllLoader::Launch(HINSTANCE instance,
base::TimeTicks exe_entry_point_ticks) {
const base::CommandLine& cmd_line = *base::CommandLine::ForCurrentProcess();
process_type_ = cmd_line.GetSwitchValueASCII(switches::kProcessType);

......

dll_ = Load(&file);
if (!dll_)
return chrome::RESULT_CODE_MISSING_DATA;

OnBeforeLaunch(cmd_line, process_type_, file);
DLL_MAIN chrome_main =
reinterpret_cast<DLL_MAIN>(::GetProcAddress(dll_, "ChromeMain"));
int rc = chrome_main(instance, &sandbox_info,
exe_entry_point_ticks.ToInternalValue());
OnBeforeExit(file);
return rc;
}

这里完成了 chrome.dll 的载入,并且执行里面的 ChromeMain 函数。

3、ChromeMain() 函数

ChromeMain 函数负责 Embedder 层的实现类创建,并传递给 Content 层,定义在 chrome\app\chrome_main.cc 中,内容如下:

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
// chrome\app\chrome_main.cc

extern "C" {
DLLEXPORT int __cdecl ChromeMain(HINSTANCE instance,
sandbox::SandboxInterfaceInfo* sandbox_info,
int64_t exe_entry_point_ticks);
}

......

#if defined(OS_WIN)
DLLEXPORT int __cdecl ChromeMain(HINSTANCE instance,
sandbox::SandboxInterfaceInfo* sandbox_info,
int64_t exe_entry_point_ticks) {
#elif defined(OS_POSIX)
int ChromeMain(int argc, const char** argv) {
int64_t exe_entry_point_ticks = 0;
#endif

#if defined(OS_WIN)
install_static::InitializeFromPrimaryModule();
#endif

ChromeMainDelegate chrome_main_delegate(
base::TimeTicks::FromInternalValue(exe_entry_point_ticks));
content::ContentMainParams params(&chrome_main_delegate);

......

int rv = content::ContentMain(params);

return rv;
}

在ChromeMain中,最终执行到了 content::ContentMain 这个函数。

4、content::ContentMain() 函数

代码执行到这里,进入了 Content 层,并且传入参数 content::ContentMainParams 类型的参数 params,它是由 Embedder 层传递过来的重要参数,里面包含了 Embedder 层的具体实现信息,此结构体在 content\public\app\content_main.h 中定义如下:

1
2
3
4
5
6
7
8
9
// content\public\app\content_main.h

struct ContentMainParams {
explicit ContentMainParams(ContentMainDelegate* delegate)
: delegate(delegate) {}

ContentMainDelegate* delegate;

......

其中有一个重要的成员变量 delegate,其类型为 content::ContentMainDelegate,它在 content\public\app\content_main_delegate.cc 中定义如下:

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
// content\public\app\content_main_delegate.cc

class CONTENT_EXPORT ContentMainDelegate {
public:
virtual ~ContentMainDelegate() {}

virtual bool BasicStartupComplete(int* exit_code);
virtual void PreSandboxStartup() {}
virtual void SandboxInitialized(const std::string& process_type) {}
virtual int RunProcess(
const std::string& process_type,
const MainFunctionParams& main_function_params);
virtual void ProcessExiting(const std::string& process_type) {}

......

virtual void PreCreateMainMessageLoop() {}

......

protected:
friend class ContentClientInitializer;

virtual ContentBrowserClient* CreateContentBrowserClient();
virtual ContentGpuClient* CreateContentGpuClient();
virtual ContentRendererClient* CreateContentRendererClient();
virtual ContentUtilityClient* CreateContentUtilityClient();
};

可以看到,这里定义了一系列与启动相关的操作,并且通过几个 CreateXXX 的函数,获取 ContentBrowserClientContentRendererClient 等接口具体的实现,这也是 content API 的巧妙设计,通过这种方式,将浏览器的实现放入了 content 中。

继续往下看,content::ContentMain() 中调用了 content\app\content_main.cc 中的 service_manager::Main()

1
2
3
4
5
6
7
8
9
10
11
// content\app\content_main.cc

int ContentMain(const ContentMainParams& params) {
ContentServiceManagerMainDelegate delegate(params);
service_manager::MainParams main_params(&delegate);
#if !defined(OS_WIN) && !defined(OS_ANDROID)
main_params.argc = params.argc;
main_params.argv = params.argv;
#endif
return service_manager::Main(main_params);
}

在这里,使用一个 content::ContentServiceManagerMainDelegate 对象来构建了 main_params,并传入了 service_manager::Main()

5、service_manager::Main 函数

service_manager::Main 函数位于 services\service_manager\embedder\main.cc,接收一个 MainParams 类型的参数,具体如下:

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
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
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
// services\service_manager\embedder\main.cc

int Main(const MainParams& params) {
MainDelegate* delegate = params.delegate;

......

ProcessType process_type = delegate->OverrideProcessType();

......
// A flag to indicate whether Main() has been called before. On Android, we
// may re-run Main() without restarting the browser process. This flag
// prevents initializing things more than once.
static bool is_initialized = false;
#if !defined(OS_ANDROID)
DCHECK(!is_initialized);
#endif
if (!is_initialized) {
is_initialized = true;

......

#if defined(OS_WIN)
base::win::RegisterInvalidParamHandler();
ui::win::CreateATLModuleIfNeeded();
#endif // defined(OS_WIN)

......

base::CommandLine::Init(argc, argv);

......

const auto& command_line = *base::CommandLine::ForCurrentProcess();

#if defined(OS_WIN)
base::win::SetupCRT(command_line);
#endif

MainDelegate::InitializeParams init_params;

......
mojo::core::Init(mojo_config);

......

exit_code = delegate->Initialize(init_params);

......

}

const auto& command_line = *base::CommandLine::ForCurrentProcess();
if (process_type == ProcessType::kDefault) {
std::string type_switch =
command_line.GetSwitchValueASCII(switches::kProcessType);
if (type_switch == switches::kProcessTypeServiceManager) {
process_type = ProcessType::kServiceManager;
} else if (type_switch == switches::kProcessTypeService) {
process_type = ProcessType::kService;
} else {
process_type = ProcessType::kEmbedder;
}
}
switch (process_type) {
case ProcessType::kDefault:
NOTREACHED();
break;

case ProcessType::kServiceManager:
exit_code = RunServiceManager(delegate);
break;

case ProcessType::kService:
CommonSubprocessInit();
exit_code = RunService(delegate);
break;

case ProcessType::kEmbedder:
if (delegate->IsEmbedderSubprocess())
CommonSubprocessInit();
exit_code = delegate->RunEmbedderProcess();
break;
}

......

if (process_type == ProcessType::kEmbedder)
delegate->ShutDownEmbedderProcess();

return exit_code;
}

这里截取的代码比较长,也非常重要,我们主要关注这四个部分:

  • 根据传入的 delegatecommand_line 决定进程的类型
  • 运行环境的初始化,比如 CreateATLModuleIfNeededSetupCRT 并用 is_initialized 来防止重复执行
  • 通过传入的 delegate 进行程序的初始化操作,delegate->Initialize(init_params)
  • 根据进程类型启动相应的工作

这里的 delegate 类型为 service_manager::MainDelegate*,是在 services/service_manager/embedder/main_delegate.h 中定义的抽象类,在这里我们主要关注它的 InitializeRunEmbedderProcessShutDownEmbedderProcess,其中 Initialize 为被声明为纯虚函数,RunEmbedderProcessShutDownEmbedderProcess 又是什么都不做的,代码如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
// services/service_manager/embedder/main_delegate.h

class COMPONENT_EXPORT(SERVICE_MANAGER_EMBEDDER) MainDelegate {
public:
// Perform early process initialization. Returns -1 if successful, or the exit
// code with which the process should be terminated due to initialization
// failure.
virtual int Initialize(const InitializeParams& params) = 0;

......

// Runs the embedder's own main process logic. Called exactly once after a
// successful call to Initialize(), and only if the Service Manager core does
// not know what to do otherwise -- i.e., if it is not starting a new Service
// Manager instance or launching an embedded service.
//
// Returns the exit code to use when terminating the process after
// RunEmbedderProcess() (and then ShutDown()) completes.
virtual int RunEmbedderProcess();

......

// Called just before process exit if RunEmbedderProcess() was called.
virtual void ShutDownEmbedderProcess();
1
2
3
4
5
6
7
8
9
// services/service_manager/embedder/main_delegate.cc

int MainDelegate::RunEmbedderProcess() {
return 0;
}

...

void MainDelegate::ShutDownEmbedderProcess() {}

回到 service_manager::Main(),我们看到第一句 MainDelegate* delegate = params.delegate; 中的 params.delegate 就是前面在 content::ContentMain 中构建 main_params 所使用的 content::ContentServiceManagerMainDelegate 对象,因此,上述的三个函数 InitializeRunEmbedderProcessShutDownEmbedderProcess 是由 ContentServiceManagerMainDelegate 来最终实现的,来看代码:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
// content\app\content_service_manager_main_delegate.cc

int ContentServiceManagerMainDelegate::Initialize(
const InitializeParams& params) {

......

return content_main_runner_->Initialize(content_main_params_);
}

......

int ContentServiceManagerMainDelegate::RunEmbedderProcess() {
return content_main_runner_->Run(start_service_manager_only_);
}

......

void ContentServiceManagerMainDelegate::ShutDownEmbedderProcess() {
#if !defined(OS_ANDROID)
content_main_runner_->Shutdown();
#endif
}

在这三个函数的定义中,都使用了 content_main_runner_ 这个成员变量来具体执行,它的定义为 std::unique_ptr<ContentMainRunnerImpl>

6、整个程序的Runner,content::ContentMainRunnerImpl

这个 content::ContentMainRunnerImplcontent::ContentMainRunner 接口的一个实现,先来看接口的声明:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
// content\app\content_main_runner_impl.h

class CONTENT_EXPORT ContentMainRunner {
public:
virtual ~ContentMainRunner() {}

// Create a new ContentMainRunner object.
static ContentMainRunner* Create();

// Initialize all necessary content state.
virtual int Initialize(const ContentMainParams& params) = 0;

// Perform the default run logic.
virtual int Run(bool start_service_manager_only) = 0;

// Shut down the content state.
virtual void Shutdown() = 0;
};

再来看实现类的代码:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
// content\app\content_main_runner_impl.h

class ContentMainRunnerImpl : public ContentMainRunner {
public:
static ContentMainRunnerImpl* Create();

ContentMainRunnerImpl();
~ContentMainRunnerImpl() override;

int TerminateForFatalInitializationError();

// ContentMainRunner:
int Initialize(const ContentMainParams& params) override;
int Run(bool start_service_manager_only) override;
void Shutdown() override;

......

}

7、ContentMainRunner::Initialize() 函数

先来看 Initialize 函数:

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
// content\app\content_main_runner_impl.cc

int ContentMainRunnerImpl::Initialize(const ContentMainParams& params) {
ui_task_ = params.ui_task;
created_main_parts_closure_ = params.created_main_parts_closure;

#if defined(OS_WIN)
sandbox_info_ = *params.sandbox_info;
#else // !OS_WIN

......

is_initialized_ = true;
delegate_ = params.delegate;

......

int exit_code = 0;
if (delegate_->BasicStartupComplete(&exit_code))
return exit_code;
completed_basic_startup_ = true;

......

delegate_->PreSandboxStartup();
#if defined(OS_WIN)
if (!InitializeSandbox(
service_manager::SandboxTypeFromCommandLine(command_line),
params.sandbox_info))
return TerminateForFatalInitializationError();
#elif defined(OS_MACOSX)

......

#endif

delegate_->SandboxInitialized(process_type);

......

// Return -1 to indicate no early termination.
return -1;
}

大致看一下,在这个 Initialize 中,主要是根据 command_line 启动了相应的 sandbox service,并在启动前后都触发了 delegate_->PreSandboxStartup()delegate_->SandboxInitialized(process_type),这个 delegate_ 来自于传入的 content::ContentMainParams 结构体,这个结构体是在 chrome_main.cc 中调用 content::ContentMain(params) 时所创建,所以这个 delegate_ 正是前面所提到的巧妙设计中,继承自 content::ContentMainDelegateChromeMainDelegate 对象,通过这一系列的调用,content 层就把创建 sandbox service 前后的事件触发了出来,具体实现者只要在 ChromeMainDelegate 中填充这两个时间点要做的事即可。

8、进程入口,ContentMainRunner::Run() 函数

再来看 Run 函数:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
// // content\app\content_main_runner_impl.cc

int ContentMainRunnerImpl::Run(bool start_service_manager_only) {

......

const base::CommandLine& command_line =
*base::CommandLine::ForCurrentProcess();
std::string process_type =
command_line.GetSwitchValueASCII(switches::kProcessType);

......

MainFunctionParams main_params(command_line);
main_params.ui_task = ui_task_;
main_params.created_main_parts_closure = created_main_parts_closure_;

......

if (process_type.empty())
return RunServiceManager(main_params, start_service_manager_only);

return RunOtherNamedProcessTypeMain(process_type, main_params, delegate_);
}

此处先判断 process_type 是否为空,为空则代表当前执行的是默认进程(一般情况下为 Browser 进程),则调用 RunServiceManager(),否则调用 RunOtherNamedProcessTypeMain 根据process_type 来执行相应的进程。先来看 RunServiceManager

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
// content\app\content_main_runner_impl.cc

int ContentMainRunnerImpl::RunServiceManager(MainFunctionParams& main_params,
bool start_service_manager_only) {

......

if (!service_manager_context_) {

......

delegate_->PreCreateMainMessageLoop();

......

delegate_->PostEarlyInitialization(main_params.ui_task != nullptr);

......

}

if (should_start_service_manager_only)
return -1;

is_browser_main_loop_started_ = true;
startup_data_ = std::make_unique<StartupDataImpl>();
startup_data_->thread = std::move(service_manager_thread_);
startup_data_->service_manager_context = service_manager_context_.get();
main_params.startup_data = startup_data_.get();
return RunBrowserProcessMain(main_params, delegate_);
}

同样,这里通过 delegate_ 做了一些操作之后,最后调用了 RunBrowserProcessMain() 函数,内容如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
// content\app\content_main_runner_impl.cc

int RunBrowserProcessMain(const MainFunctionParams& main_function_params,
ContentMainDelegate* delegate) {
int exit_code = delegate->RunProcess("", main_function_params);
#if defined(OS_ANDROID)
// In Android's browser process, the negative exit code doesn't mean the
// default behavior should be used as the UI message loop is managed by
// the Java and the browser process's default behavior is always
// overridden.
return exit_code;
#else
if (exit_code >= 0)
return exit_code;
return BrowserMain(main_function_params);
#endif
}

非常简单明了,首先通过 delegate->RunProcess 把执行默认进程的优先权交由 Embedder 层,如果 Embedder 层成功执行了进程并最终返回了成功标志(exit_code >= 0),那么就退出函数;如果 Embedder 层对默认进程没有定义,就继续执行 content::BrowserMain,由此,Browser 进程开始执行。

再来看 RunOtherNamedProcessTypeMain 函数:

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
// content\app\content_main_runner_impl.cc

int RunOtherNamedProcessTypeMain(const std::string& process_type,
const MainFunctionParams& main_function_params,
ContentMainDelegate* delegate) {
static const MainFunction kMainFunctions[] = {

......

{switches::kUtilityProcess, UtilityMain},
{switches::kRendererProcess, RendererMain},
{switches::kGpuProcess, GpuMain},
};

for (size_t i = 0; i < base::size(kMainFunctions); ++i) {
if (process_type == kMainFunctions[i].name) {
int exit_code = delegate->RunProcess(process_type, main_function_params);
if (exit_code >= 0)
return exit_code;
return kMainFunctions[i].function(main_function_params);
}
}

......

// If it's a process we don't know about, the embedder should know.
return delegate->RunProcess(process_type, main_function_params);
}

先建立了一个进程类型和入口函数指针的对应数组,再根据进程类型去具体执行,执行的过程与 Browser 进程一样,先通过 delegate->RunProcess 交由 Embedder 层处理,如果未处理再调用默认的进程入口函数,可以看到分别提供了 UtilityMainRendererMainGpuMain 这三个进程的入口,其中 RendererMain 则是我们关注的 Renderer 进程的入口函数,Renderer 进程从此处开始执行。最后一句,如果进程类型不在以上范围内,则交由 Embedder 去处理。

9、程序结束

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
void ContentMainRunnerImpl::Shutdown() {
DCHECK(is_initialized_);
DCHECK(!is_shutdown_);

if (completed_basic_startup_) {
const base::CommandLine& command_line =
*base::CommandLine::ForCurrentProcess();
std::string process_type =
command_line.GetSwitchValueASCII(switches::kProcessType);

delegate_->ProcessExiting(process_type);
}

#if !defined(CHROME_MULTIPLE_DLL_CHILD)
// The BrowserTaskExecutor needs to be destroyed before |exit_manager_|.
BrowserTaskExecutor::Shutdown();
#endif // !defined(CHROME_MULTIPLE_DLL_CHILD)

#if defined(OS_WIN)
#ifdef _CRTDBG_MAP_ALLOC
_CrtDumpMemoryLeaks();
#endif // _CRTDBG_MAP_ALLOC
#endif // OS_WIN

exit_manager_.reset(nullptr);

delegate_ = nullptr;
is_shutdown_ = true;
}

首先通过 delegate_->ProcessExiting(process_type) 通知 Embedder 层处理,然后做了一些善后释放的工作,最后将 is_shutdown_ 标记置为 true

10、总结

前面分析了这么多,其实结合类图来看一下还是很简单明了的,主要起到作用的就是图中标红的三个,service_manager::Main 通过 content::ContentServiceManagerMainDelegate 的实例调用了 content::ContentMainRunnerImpl 实例中的 Initialize()Run()Shutdown() 函数,而在这个Runner中,又通过 content::ContentMainDelegate 接口指针调用到了由 Embedder 层创建的 ChromeMainDelegate 实例中的函数,由此完成了程序的启动以及 Content 层对 Embedder 的交互。