From ec693e322bbd665f3e9020d36ac564d2d3734215 Mon Sep 17 00:00:00 2001 From: dqzg12300 Date: Wed, 19 Apr 2023 00:10:32 +0800 Subject: [PATCH] =?UTF-8?q?12=E7=AB=A0=E5=86=85=E5=AE=B9=E6=8F=90=E4=BA=A4?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- chapter-12/README.md | 551 ++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 544 insertions(+), 7 deletions(-) diff --git a/chapter-12/README.md b/chapter-12/README.md index 8747674..4493d43 100644 --- a/chapter-12/README.md +++ b/chapter-12/README.md @@ -606,6 +606,7 @@ extern uint32_t JniMethodStart(Thread* self) { // 当前开始函数为要监控的目标函数时,则开启输出JNI if(strstr(methodname.c_str(),runtime->GetConfigItem().jniFuncName)){ runtime->GetConfigItem().jniEnable=true; + ALOGD("mikrom enter jni %s",methodname.c_str()); } } //endadd @@ -628,6 +629,7 @@ extern void JniMethodEnd(uint32_t saved_local_ref_cookie, Thread* self) { // 当前结束函数为要监控的目标函数时,则关闭输出JNI if(strstr(methodname.c_str(),runtime->GetConfigItem().jniFuncName)){ runtime->GetConfigItem().jniEnable=false; + ALOGD("mikrom leave jni %s",methodname.c_str()); } } //endadd @@ -635,6 +637,25 @@ extern void JniMethodEnd(uint32_t saved_local_ref_cookie, Thread* self) { GoToRunnable(self); PopLocalReferences(saved_local_ref_cookie, self); } + + +static mirror::Object* JniMethodEndWithReferenceHandleResult(jobject result, + uint32_t saved_local_ref_cookie, + Thread* self) + NO_THREAD_SAFETY_ANALYSIS { + //add + Runtime* runtime=Runtime::Current(); + if(runtime->GetConfigItem().isJNIMethodPrint){ + ArtMethod* native_method = *self->GetManagedStack()->GetTopQuickFrame(); + std::string methodname=native_method->PrettyMethod(); + if(strstr(methodname.c_str(),runtime->GetConfigItem().jniFuncName)){ + runtime->GetConfigItem().jniEnable=false; + ALOGD("mikrom leave jni %s",methodname.c_str()); + } + } + //endadd + ... +} ``` ## 12.5 打桩函数分类 @@ -691,9 +712,9 @@ void ShowVarArgs(const ScopedObjectAccessAlreadyRunnable& soa,const char* funcna std::string temp; const char* className= c->GetDescriptor(&temp); ArtMethod* method = jni::DecodeArtMethod(methodID); - pthread_t threadId = pthread_self(); + pid_t pid = getpid(); // 前面加上标志是为了方便搜索日志 - ALOGD("%s /* TID %ld */","mikrom",threadId); + ALOGD("%s /* TID %d */","mikrom",pid); ALOGD("%s [+] JNIEnv->%s","mikrom",funcname); ALOGD("%s |- jclass :%s","mikrom",className); ALOGD("%s |- char* :%p","mikrom",name); @@ -751,10 +772,14 @@ void ShowVarArgs(const ScopedObjectAccessAlreadyRunnable& , if(!HasShow()){ return; } - pthread_t threadId = pthread_self(); - ALOGD("%s /* TID %ld */","mikrom",threadId); + pid_t pid = getpid(); + ALOGD("%s /* TID %d */","mikrom",pid); ALOGD("%s [+] JNIEnv->%s","mikrom",funcname); - ALOGD("%s |- jboolean* : %d","mikrom",*is_copy); + if(is_copy== nullptr){ + ALOGD("%s |- jboolean* : %d","mikrom",false); + }else{ + ALOGD("%s |- jboolean* : %d","mikrom",*is_copy); + } ALOGD("%s |= char* : %s","mikrom",data); } ``` @@ -799,8 +824,8 @@ void ShowVarArgs(const ScopedObjectAccessAlreadyRunnable& , if(!HasShow()){ return; } - pthread_t threadId = pthread_self(); - ALOGD("%s /* TID %ld */","mikrom",threadId); + pid_t pid = getpid(); + ALOGD("%s /* TID %d */","mikrom",pid); ALOGD("%s [+] JNIEnv->%s","mikrom",funcname); ALOGD("%s |- char* : %d","mikrom",data); } @@ -824,11 +849,523 @@ static jstring NewStringUTF(JNIEnv* env, const char* utf) { ​ 这个`JNI`函数不同于前面几种函数,在前几个函数中,参数是明确固定的,而`CallObjectMethodV`是通过`JNI`,调用一个`java`函数,而为此`java`函数提供的所有参数的类型,以及参数个数。都是未知的。而这些参数的信息同样是需要打桩展示出来的。 +​ 将测试样例中,被调用的`java`函数进行调整,将测试函数新增参数,并且使用`JniTrace`观察`CallObjectMethodV`的输出结果。样例函数修改如下。 + +```java +public String demo(int a,float b,long c,String d){ + return a+b+c+d; +} +``` + +​ 同时修改`native`函数中使用`JNI`调用的逻辑。 + +```c++ +extern "C" JNIEXPORT jstring JNICALL +Java_cn_mik_nativedemo_MainActivity_stringFromJNI( + JNIEnv* env, + jobject obj /* this */) { + jclass cls= env->FindClass("cn/mik/nativedemo/MainActivity"); + jmethodID mid=env->GetMethodID(cls,"demo", "(IFLjava/lang/String;)Ljava/lang/String;"); + jstring c=env->NewStringUTF("newdemo"); + jstring data= (jstring)env->CallObjectMethod(obj,mid,1,2.0f,c); + std::string datatmp= env->GetStringUTFChars(data,nullptr); + return env->NewStringUTF(datatmp.c_str()); +} +``` + +​ 再次使用`JniTrace`观察到的`CallObjectMethodV`输出如下。 + +``` + /* TID 18863 */ + 2169 ms [+] JNIEnv->CallObjectMethodV + 2169 ms |- JNIEnv* : 0x704e856090 + 2169 ms |- jobject : 0x7fe4d55d38 + 2169 ms |- jmethodID : 0x3d { demo(IFLjava/lang/String;)Ljava/lang/String; } + 2169 ms |- va_list : 0x7fe4d55b30 + 2169 ms |: jint : 1 + 2169 ms |: jfloat : 2 + 2169 ms |: jstring : 0x81 { newdemo } + 2169 ms |= jobject : 0x95 { java/lang/String } + + 2169 ms ------------------------------------------------------Backtrace------------------------------------------------------ + 2169 ms |-> 0x6f5d458ae4: _ZN7_JNIEnv16CallObjectMethodEP8_jobjectP10_jmethodIDz+0xc4 (libnativedemo.so:0x6f5d44a000) + 2169 ms |-> 0x6f5d458ae4: _ZN7_JNIEnv16CallObjectMethodEP8_jobjectP10_jmethodIDz+0xc4 (libnativedemo.so:0x6f5d44a000) +``` + +​ 从日志中能看到,参数列表被解析后将具体的值进行输出,而返回值的部分,如果是`jobject`,则将其类型进行输出。明白具体需求后,接着就可以开始根据`CallObjectMethodV`类型参数定义来准备打桩函数了。定义描述如下。 + +```c++ +static jobject CallObjectMethodV(JNIEnv* env, jobject obj, jmethodID mid, va_list args); +``` + +​ 根据参考,需要输出调用的目标函数,参数,以及其返回值。而其他的类似调用函数的情况,和该函数差不多的处理,只是返回值的输出不同。优化后的打桩函数如下。 + +```c++ +// 输出JNI的参数部分 +void ShowVarArgs(const ScopedObjectAccessAlreadyRunnable& soa, + const char* funcname, + jmethodID mid, + va_list vaList){ + if(!HasShow()){ + return; + } + + ArtMethod* method = jni::DecodeArtMethod(mid); + pid_t pid = getpid(); + ALOGD("%s /* TID %d */","mikrom",pid); + ALOGD("%s [+] JNIEnv->%s","mikrom",funcname); + ALOGD("%s |- jmethodID :0x%x {%s}","mikrom",method->GetMethodIndex(),method->PrettyMethod().c_str()); + ALOGD("%s |- va_list :%p","mikrom",&vaList); + + uint32_t shorty_len = 0; + const char* shorty = + method->GetInterfaceMethodIfProxy(kRuntimePointerSize)->GetShorty(&shorty_len); + ArgArray arg_array(shorty, shorty_len); + arg_array.VarArgsShowArg(soa, vaList); +} +void ShowVarArgs(const ScopedObjectAccessAlreadyRunnable& soa, + const char* funcname, + jmethodID mid, + va_list valist, + jobject ret){ + if(!HasShow()){ + return; + } + // 输出JNI的参数部分 + ShowVarArgs(soa,funcname,mid,valist); + // 输出JNI的返回值 + ObjPtr receiver =soa.Decode(ret); + if(receiver==nullptr){ + return; + } + ObjPtr cls=receiver->GetClass(); + if (cls->DescriptorEquals("Ljava/lang/String;")){ + ObjPtr retStr =soa.Decode(ret); + ALOGD("%s |= jstring :%s","mikrom",(const char*)retStr->GetValue()); + }else{ + std::string temp; + const char* className= cls->GetDescriptor(&temp); + ALOGD("%s |= jobject :%p {%s}","mikrom",&ret,className); + } +} +``` +​ 很多`JNI`的调用函数除了返回值的处理不同,其他调用部分都是相同的,所以将打印参数部分和打印返回值部分拆分开来,而参数`va_list`的解析,可以直接参考`AOSP`源码中,函数`BuildArgArrayFromVarArgs`对其的处理。实现如下。 + +```c++ +void VarArgsShowArg(const ScopedObjectAccessAlreadyRunnable& soa, + va_list ap) + REQUIRES_SHARED(Locks::mutator_lock_) { + std::stringstream ss; + for (size_t i = 1; i < shorty_len_; ++i) { + switch (shorty_[i]) { + case 'Z': + case 'B': + case 'C': + case 'S': + case 'I': + ss<<"mikrom"<<" |: jint : "< receiver =soa.Decode(obj); + if(receiver==nullptr){ + ss<<"mikrom"<<" |: jobject : null\n"; + break; + } + ObjPtr cls=receiver->GetClass(); + if (cls->DescriptorEquals("Ljava/lang/String;")){ + ObjPtr argStr =soa.Decode(obj); + ss<<"mikrom"<<" |: jstring : "<<(const char*)argStr->GetValue()<<"\n"; + }else{ + ss<<"mikrom"<<" |: jobject : "<<&obj<<"\n"; + } + break; + } + case 'D': + ss<<"mikrom"<<" |: jdouble : "<GetMethodID +mikrom |- jclass :Lcn/mik/nativedemo/MainActivity; +mikrom |- char* :0x7275d69f1b +mikrom |: demo +mikrom |- char* :0x7275d69eef +mikrom |: (IFLjava/lang/String;)Ljava/lang/String; +mikrom |= jmethodID :0x277 {java.lang.String cn.mik.nativedemo.MainActivity.demo(int, float, java.lang.String)} +mikrom /* TID 5465 */ +mikrom [+] JNIEnv->NewStringUTF +mikrom |- char* : newdemo +mikrom /* TID 5465 */ +mikrom [+] JNIEnv->CallObjectMethodV +mikrom |- jmethodID :0x277 {java.lang.String cn.mik.nativedemo.MainActivity.demo(int, float, java.lang.String)} +mikrom |- va_list :0x7fe0461bb0 +mikrom |: jint : 1 +mikrom |: jfloat : 2 +mikrom |: jstring : newdemo +mikrom |= jstring :3.0newdemo +mikrom /* TID 5465 */ +mikrom [+] JNIEnv->GetStringUTFChars +mikrom |- jboolean* : 0 +mikrom |= char* : 3.0newdemo +mikrom /* TID 5465 */ +mikrom [+] JNIEnv->NewStringUTF +mikrom |- char* : 3.0newdemo +mikrom leave jni java.lang.String cn.mik.nativedemo.MainActivity.stringFromJNI() +``` ## 12.6 调用栈展示 +​ 经过调整后,打桩函数已经非常接近`JniTrace`的输出效果了,但是还有最后的一点区别是在于`JNI`函数的调用堆栈,不仅仅需要看到函数的参数和返回值,还需要知道是在哪里触发了该函数。而获取调用堆栈地址,在`AOSP`源码是有相关支持的,当应用崩溃时,在`logcat`中能看到详细的堆栈信息。 + +### 12.6.1 xUnwind获取调用栈 + +​ `xUnwind`是一个开源工具,该工具将堆栈获取进行了封装,并且有简单的`demo`演示如何获取堆栈。下载地址:`https://github.com/hexhacking/xUnwind.git`。接下来将分析该工具是如何实现的获取堆栈,然后再将其内置到`AOSP`中,在`JNI`函数调用结束时获取堆栈进行输出。 + +​ `xUnwind`提供了三种调用堆栈回溯方案。 + +- `CFI (Call Frame Info)`:由安卓系统库提供。 +- `EH (Exception handling GCC extension)`:由编译器提供。 +- `FP (Frame Pointer)`:只支持 `ARM64`。 + +​ 其中`CFI`主要针对`java`层调用回溯,`FP`仅支持`ARM64`。根据功能需要,选择`EH`方案进行调用栈回溯。下面看看该工具是如何实现的,首先查看`JNI_OnLoad`的实现。 + +```c++ +JNIEXPORT jint JNICALL JNI_OnLoad(JavaVM *vm, void *reserved) { + JNIEnv *env; + jclass cls; + + (void)reserved; + + if (NULL == vm) return JNI_ERR; + if (JNI_OK != (*vm)->GetEnv(vm, (void **)&env, SAMPLE_JNI_VERSION)) return JNI_ERR; + if (NULL == env || NULL == *env) return JNI_ERR; + if (NULL == (cls = (*env)->FindClass(env, SAMPLE_JNI_CLASS_NAME))) return JNI_ERR; + // 注册对应的函数 + if (0 != (*env)->RegisterNatives(env, cls, sample_jni_methods, + sizeof(sample_jni_methods) / sizeof(sample_jni_methods[0]))) + return JNI_ERR; + // 注册信号处理 + sample_signal_register(); + + return SAMPLE_JNI_VERSION; +} +``` + +​ 然后看看`sample_signal_register`对信号进行了什么处理 + +```c++ +static void sample_signal_register(void) { + struct sigaction act; + // 清空act + memset(&act, 0, sizeof(act)); + + // 填充所有信号 + sigfillset(&act.sa_mask); + // 删除SIGSEGV信号 + sigdelset(&act.sa_mask, SIGSEGV); + // 设置信号动作的对应函数 + act.sa_sigaction = sample_sigabrt_handler; + act.sa_flags = SA_RESTART | SA_SIGINFO | SA_ONSTACK; + // 设置指定信号的动作 + sigaction(SIGABRT, &act, NULL); + // 清空act + memset(&act, 0, sizeof(act)); + // 填充所有信号 + sigfillset(&act.sa_mask); + // 设置信号动作的对应函数 + act.sa_sigaction = sample_sigsegv_handler; + act.sa_flags = SA_RESTART | SA_SIGINFO | SA_ONSTACK; + // 设置指定信号的动作 + sigaction(SIGSEGV, &act, &g_sigsegv_oldact); +} +``` + +​ 在这个函数中主要是对信号指定了处理函数,`SIGSEGV` 是一种表示段错误的信号,通常意味着进程访问了无法访问的内存地址。 + +​ `SIGABRT` 是一种表示异常终止的信号,通常由调用 `abort()` 函数或 C++ 异常处理程序显式引发。 `abort()` 函数会向进程发送 `SIGABRT` 信号,并导致进程异常终止。 + +​ 如果进程接收到 `SIGABRT` 信号,操作系统将其发送给进程,并中断进程的正常执行流程。此时,进程通常会尝试处理该信号并进行恢复或退出。如果进程没有为 `SIGABRT` 信号设置信号处理程序,则默认行为是终止进程。 + +​ 接着开始跟踪`EH`的调用栈获取函数`sample_test_eh`的实现。 + +```c++ +static void sample_test_eh(JNIEnv *env, jobject thiz, jboolean with_context, jboolean signal_interrupted) { + (void)env, (void)thiz; + + sample_test(SAMPLE_SOLUTION_EH, JNI_FALSE, with_context, signal_interrupted); +} +``` + +​ 继续查看`sample_test`的实现。 + +```c++ + +static void sample_test(int solution, jboolean remote_unwind, jboolean with_context, + jboolean signal_interrupted) { + // 将参数存储到全局变量 + g_solution = solution; + g_remote_unwind = (JNI_TRUE == remote_unwind ? true : false); + g_with_context = (JNI_TRUE == with_context ? true : false); + g_signal_interrupted = (JNI_TRUE == signal_interrupted ? true : false); + + // 原子请求,为了保证获取到的g_frames_sz没问题 + __atomic_store_n(&g_frames_sz, 0, __ATOMIC_SEQ_CST); + // 触发SIGABRT信号 + tgkill(getpid(), gettid(), SIGABRT); + + if ((solution == SAMPLE_SOLUTION_FP || solution == SAMPLE_SOLUTION_EH) && + __atomic_load_n(&g_frames_sz, __ATOMIC_SEQ_CST) > 0) + // 根据堆栈的地址信息,打印堆栈日志 + xunwind_frames_log(g_frames, g_frames_sz, SAMPLE_LOG_TAG, SAMPLE_LOG_PRIORITY, NULL); +} +``` + +​ 由于堆栈信息需要通过触发信号后,在信号处理函数中获取,所以这里的参数没办法传过去,这里就将参数放到了全局变量中。然后信号函数执行完毕后,将会填充`g_frames`调用栈的地址信息,最后使用`xunwind_frames_log`函数解析调用栈,最后输出详细的调用栈。 + +​ 接下来查看`SIGABRT`信号的处理函数`sample_sigabrt_handler`的实现。 + +```java + +static void sample_sigabrt_handler(int signum, siginfo_t *siginfo, void *context) { + (void)signum, (void)siginfo; + + if (g_solution == SAMPLE_SOLUTION_FP || g_solution == SAMPLE_SOLUTION_EH) { + if (!g_signal_interrupted) { + if (g_solution == SAMPLE_SOLUTION_FP) { + // FP local unwind + ... + } else if (g_solution == SAMPLE_SOLUTION_EH) { + // EH local unwind + size_t frames_sz = xunwind_eh_unwind(g_frames, sizeof(g_frames) / sizeof(g_frames[0]), + g_with_context ? context : NULL); + __atomic_store_n(&g_frames_sz, frames_sz, __ATOMIC_SEQ_CST); + } + } else { + // trigger a segfault, we will do "FP local unwind" in the sigsegv's signal handler + ... + } + } else if (!g_remote_unwind) { + // CFI local unwind + ... + } else { + // CFI remote unwind + ... + } +} +``` + +​ `EH`方案是通过`xunwind_eh_unwind`函数获取的调用栈信息,继续查看其具体实现。 + +```c++ +size_t xunwind_eh_unwind(uintptr_t *frames, size_t frames_cap, void *context) { + return xu_eh_unwind(frames, frames_cap, context); +} + +size_t xu_eh_unwind(uintptr_t *frames, size_t frames_cap, void *context) { + if (NULL == frames || 0 == frames_cap) return 0; + + xu_eh_info_t info; + info.frames = frames; + info.frames_cap = frames_cap; + info.frames_sz = 0; + info.prev_sp = 0; + info.uc = (ucontext_t *)context; + + _Unwind_Backtrace(xu_eh_unwind_cb, &info); + + return info.frames_sz; +} +``` + +​ `_Unwind_Backtrace`函数用于获取当前线程的调用堆栈信息。它使用`C++`异常处理机制中的`unwind`操作来实现,因此只能在支持`C++`异常处理的系统上使用。到这里就知道堆栈获取的来源了。但是这里的信息和真实看的还是有一定差距,继续看看`xunwind_frames_log`是如何对其进行转换展示的。 + +```c++ +void xunwind_frames_log(uintptr_t *frames, size_t frames_sz, const char *logtag, android_LogPriority priority, + const char *prefix) { + if (priority < ANDROID_LOG_VERBOSE || ANDROID_LOG_FATAL < priority) return; + + xu_printer_t printer; + xu_printer_init_log(&printer, logtag, priority); + + xu_formatter_print(frames, frames_sz, prefix, &printer); +} +``` + +​ 继续跟踪`xu_formatter_print`的实现。 + +```c++ + +void xu_formatter_print(uintptr_t *frames, size_t frames_sz, const char *prefix, xu_printer_t *printer) { + if (NULL == frames || 0 == frames_sz) return; + + if (NULL == prefix) prefix = ""; + + void *cache = NULL; + xdl_info_t info; + for (size_t i = 0; i < frames_sz; i++) { + memset(&info, 0, sizeof(xdl_info_t)); + int r = 0; + + if (0 != frames[i]) { + // 根据调用栈地址获取动态库的信息 + r = xdl_addr((void *)(frames[i]), &info, &cache); + + // 如果查找到的动态库的起始地址大于这条调用栈信息的地址,则从maps中重新找动态库信息 + char buf[512]; + if (0 == r || (uintptr_t)info.dli_fbase > frames[i]) + r = xu_formatter_maps_addr(frames[i], &info, buf, sizeof(buf)); + } + + // 输出打印 + if (0 == r || (uintptr_t)info.dli_fbase > frames[i]) + xu_printer_append_format(printer, XU_FORMATTER_PREFIX "\n", prefix, i, frames[i]); + else if (NULL == info.dli_fname || '\0' == info.dli_fname[0]) + xu_printer_append_format(printer, XU_FORMATTER_PREFIX "\n", prefix, i, + frames[i] - (uintptr_t)info.dli_fbase, (uintptr_t)info.dli_fbase); + else if (NULL == info.dli_sname || '\0' == info.dli_sname[0]) + xu_printer_append_format(printer, XU_FORMATTER_PREFIX "%s\n", prefix, i, + frames[i] - (uintptr_t)info.dli_fbase, info.dli_fname); + else if (0 == (uintptr_t)info.dli_saddr || (uintptr_t)info.dli_saddr > frames[i]) + xu_printer_append_format(printer, XU_FORMATTER_PREFIX "%s (%s)\n", prefix, i, + frames[i] - (uintptr_t)info.dli_fbase, info.dli_fname, info.dli_sname); + else + xu_printer_append_format(printer, XU_FORMATTER_PREFIX "%s (%s+%" PRIuPTR ")\n", prefix, i, + frames[i] - (uintptr_t)info.dli_fbase, info.dli_fname, info.dli_sname, + frames[i] - (uintptr_t)info.dli_saddr); + } + xdl_addr_clean(&cache); +} + +``` + +​ 继续查看是如何输出的。 + +```c++ +void xu_printer_append_format(xu_printer_t *self, const char *format, ...) { + va_list ap; + va_start(ap, format); + + char tmpbuf[1024]; +#pragma clang diagnostic push +#pragma clang diagnostic ignored "-Wformat-nonliteral" + vsnprintf(tmpbuf, sizeof(tmpbuf), format, ap); +#pragma clang diagnostic pop + + va_end(ap); + + xu_printer_append_string(self, tmpbuf); +} +void xu_printer_append_string(xu_printer_t *self, const char *str) { + if (XU_PRINTER_TYPE_LOG == self->type) { + __android_log_print(self->data.log.priority, self->data.log.logtag, "%s", str); + } else if (XU_PRINTER_TYPE_DUMP == self->type) { + size_t len = strlen(str); + if (len > 0) { + xu_util_write(self->data.dump.fd, str, len); + if ('\n' != str[len - 1]) xu_util_write(self->data.dump.fd, "\n", 1); + } + } else if (XU_PRINTER_TYPE_GET == self->type) { + size_t len = strlen(str); + if (len > 0) { + xu_printer_string_append_to_buf(self, str); + if ('\n' != str[len - 1]) xu_printer_string_append_to_buf(self, "\n"); + } + } +} +``` + +​ 到了最后输出的部分可以看到,支持三种方式输出,在调用时,根据不同的需求来选择合适的输出方式。 + +* `XU_PRINTER_TYPE_LOG` 直接`logcat`输出信息 +* `XU_PRINTER_TYPE_DUMP `将堆栈信息写入到指定描述符 +* `XU_PRINTER_TYPE_GET` 返回堆栈信息的字符串。 + +​ 了解完整个实现原理后,最后看看其输出效果。`EH`方案的调用栈输出如下。 + +``` +I/xunwind_tag: >>> EH: Local Process (pid: 6713), Current Thread (tid: 6713) <<< +I/xunwind_tag: #00 pc 00000000000093a4 /data/app/~~H824r8gCTUibCXd65QZfPw==/io.github.hexhacking.xunwind.sample-1gWu3tav0VA7-qGsCc7nZQ==/lib/arm64/libxunwind.so (xunwind_eh_unwind+72) +I/xunwind_tag: #01 pc 00000000000014a8 /data/app/~~H824r8gCTUibCXd65QZfPw==/io.github.hexhacking.xunwind.sample-1gWu3tav0VA7-qGsCc7nZQ==/lib/arm64/libsample.so +I/xunwind_tag: #02 pc 00000000000008b0 [vdso] (__kernel_rt_sigreturn+0) +I/xunwind_tag: #03 pc 000000000009e598 /apex/com.android.runtime/lib64/bionic/libc.so (tgkill+8) +I/xunwind_tag: #04 pc 0000000000001320 /data/app/~~H824r8gCTUibCXd65QZfPw==/io.github.hexhacking.xunwind.sample-1gWu3tav0VA7-qGsCc7nZQ==/lib/arm64/libsample.so +I/xunwind_tag: #05 pc 0000000000001254 /data/app/~~H824r8gCTUibCXd65QZfPw==/io.github.hexhacking.xunwind.sample-1gWu3tav0VA7-qGsCc7nZQ==/lib/arm64/libsample.so +I/xunwind_tag: #06 pc 0000000000222248 /apex/com.android.art/lib64/libart.so (art_quick_generic_jni_trampoline+152) +I/xunwind_tag: #07 pc 0000000000218bec /apex/com.android.art/lib64/libart.so (art_quick_invoke_static_stub+572) +I/xunwind_tag: #08 pc 0000000000290300 /apex/com.android.art/lib64/libart.so (_ZN3art9ArtMethod6InvokeEPNS_6ThreadEPjjPNS_6JValueEPKc+536) +I/xunwind_tag: #09 pc 00000000003f09e4 /apex/com.android.art/lib64/libart.so (_ZN3art11interpreter34ArtInterpreterToCompiledCodeBridgeEPNS_6ThreadEPNS_9ArtMethodEPNS_11ShadowFrameEtPNS_6JValueE+404) +I/xunwind_tag: #10 pc 00000000003eb858 /apex/com.android.art/lib64/libart.so (_ZN3art11interpreter6DoCallILb0ELb0EEEbPNS_9ArtMethodEPNS_6ThreadERNS_11ShadowFrameEPKNS_11InstructionEtPNS_6JValueE+808) +I/xunwind_tag: #11 pc 00000000007658d0 /apex/com.android.art/lib64/libart.so (MterpInvokeStatic+984) +I/xunwind_tag: #12 pc 0000000000203998 /apex/com.android.art/lib64/libart.so (mterp_op_invoke_static+24) +I/xunwind_tag: #13 pc 00000000003e3688 /apex/com.android.art/lib64/libart.so (_ZN3art11interpreterL7ExecuteEPNS_6ThreadERKNS_20CodeItemDataAccessorERNS_11ShadowFrameENS_6JValueEbb+348) +I/xunwind_tag: #14 pc 000000000074920c /apex/com.android.art/lib64/libart.so (artQuickToInterpreterBridge+780) +I/xunwind_tag: #15 pc 000000000022237c /apex/com.android.art/lib64/libart.so (art_quick_to_interpreter_bridge+92) +I/xunwind_tag: #16 pc 000000000021160c /apex/com.android.art/lib64/libart.so (nterp_helper+156) +I/xunwind_tag: #17 pc 0000000000218968 /apex/com.android.art/lib64/libart.so (art_quick_invoke_stub+552) +I/xunwind_tag: #18 pc 00000000002902d4 /apex/com.android.art/lib64/libart.so (_ZN3art9ArtMethod6InvokeEPNS_6ThreadEPjjPNS_6JValueEPKc+492) +I/xunwind_tag: #19 pc 000000000062ba5c /apex/com.android.art/lib64/libart.so (_ZN3art12InvokeMethodILNS_11PointerSizeE8EEEP8_jobjectRKNS_33ScopedObjectAccessAlreadyRunnableES3_S3_S3_m+1388) +I/xunwind_tag: #20 pc 000000000059d2f8 /apex/com.android.art/lib64/libart.so (_ZN3artL13Method_invokeEP7_JNIEnvP8_jobjectS3_P13_jobjectArray+56) +I/xunwind_tag: #21 pc 0000000000222248 /apex/com.android.art/lib64/libart.so (art_quick_generic_jni_trampoline+152) +I/xunwind_tag: #22 pc 0000000000212524 /apex/com.android.art/lib64/libart.so (nterp_helper+4020) +I/xunwind_tag: #23 pc 00000000002132e8 /apex/com.android.art/lib64/libart.so (nterp_helper+7544) +I/xunwind_tag: #24 pc 00000000002124c8 /apex/com.android.art/lib64/libart.so (nterp_helper+3928) +I/xunwind_tag: #25 pc 00000000002124c8 /apex/com.android.art/lib64/libart.so (nterp_helper+3928) +I/xunwind_tag: #26 pc 00000000002115a8 /apex/com.android.art/lib64/libart.so (nterp_helper+56) +I/xunwind_tag: #27 pc 00000000002132e8 /apex/com.android.art/lib64/libart.so (nterp_helper+7544) +I/xunwind_tag: #28 pc 00000000002115a8 /apex/com.android.art/lib64/libart.so (nterp_helper+56) +I/xunwind_tag: #29 pc 00000000002124c8 /apex/com.android.art/lib64/libart.so (nterp_helper+3928) +I/xunwind_tag: #30 pc 00000000002115a8 /apex/com.android.art/lib64/libart.so (nterp_helper+56) +I/xunwind_tag: #31 pc 00000000002115a8 /apex/com.android.art/lib64/libart.so (nterp_helper+56) +I/xunwind_tag: #32 pc 0000000000218bec /apex/com.android.art/lib64/libart.so (art_quick_invoke_static_stub+572) +I/xunwind_tag: #33 pc 0000000000290300 /apex/com.android.art/lib64/libart.so (_ZN3art9ArtMethod6InvokeEPNS_6ThreadEPjjPNS_6JValueEPKc+536) +I/xunwind_tag: #34 pc 000000000062ba5c /apex/com.android.art/lib64/libart.so (_ZN3art12InvokeMethodILNS_11PointerSizeE8EEEP8_jobjectRKNS_33ScopedObjectAccessAlreadyRunnableES3_S3_S3_m+1388) +I/xunwind_tag: #35 pc 000000000059d2f8 /apex/com.android.art/lib64/libart.so (_ZN3artL13Method_invokeEP7_JNIEnvP8_jobjectS3_P13_jobjectArray+56) +I/xunwind_tag: #36 pc 0000000000222248 /apex/com.android.art/lib64/libart.so (art_quick_generic_jni_trampoline+152) +I/xunwind_tag: #37 pc 0000000000212524 /apex/com.android.art/lib64/libart.so (nterp_helper+4020) +I/xunwind_tag: #38 pc 00000000002132e8 /apex/com.android.art/lib64/libart.so (nterp_helper+7544) +I/xunwind_tag: #39 pc 0000000000218bec /apex/com.android.art/lib64/libart.so (art_quick_invoke_static_stub+572) +I/xunwind_tag: #40 pc 0000000000290300 /apex/com.android.art/lib64/libart.so (_ZN3art9ArtMethod6InvokeEPNS_6ThreadEPjjPNS_6JValueEPKc+536) +I/xunwind_tag: #41 pc 000000000062c1b4 /apex/com.android.art/lib64/libart.so (_ZN3art17InvokeWithVarArgsIPNS_9ArtMethodEEENS_6JValueERKNS_33ScopedObjectAccessAlreadyRunnableEP8_jobjectT_St9__va_list+452) +I/xunwind_tag: #42 pc 000000000062c678 /apex/com.android.art/lib64/libart.so (_ZN3art17InvokeWithVarArgsIP10_jmethodIDEENS_6JValueERKNS_33ScopedObjectAccessAlreadyRunnableEP8_jobjectT_St9__va_list+96) +I/xunwind_tag: #43 pc 0000000000507c68 /apex/com.android.art/lib64/libart.so (_ZN3art3JNIILb1EE21CallStaticVoidMethodVEP7_JNIEnvP7_jclassP10_jmethodIDSt9__va_list+620) +I/xunwind_tag: #44 pc 00000000000aeac8 /system/lib64/libandroid_runtime.so (_ZN7_JNIEnv20CallStaticVoidMethodEP7_jclassP10_jmethodIDz+124) +I/xunwind_tag: #45 pc 00000000000ba060 /system/lib64/libandroid_runtime.so (_ZN7android14AndroidRuntime5startEPKcRKNS_6VectorINS_7String8EEEb+840) +I/xunwind_tag: #46 pc 00000000000025a8 /system/bin/app_process64 (main+1364) +I/xunwind_tag: #47 pc 00000000000498cc /apex/com.android.runtime/lib64/bionic/libc.so (__libc_init+100) +I/xunwind_tag: >>> finished <<< +``` + +### 12.6.2 优化调用栈样例 + +​ 尽管在该工具中的样例可以直接使用,但是想要将其内置到`AOSP`中还需要将其进行优化处理,简单写一个模拟注入流程的样例,对样例做出以下调整。 + +* 去掉非`EH`获取方案的代码部分 +* 优化获取调用栈信息的代码,让其结果仅输出想要关注的目标动态库的调用栈。 +* 直接输出日志调整为返回调用栈字符串。 + +​ 新建`native`的项目,将`xUnwind`的样例`apk`解压后,取出其`lib`目录中的依赖动态库,将其复制到新项目的`libs`目录中,如下图。 + +​ + +### 12.6.3 内置获取调用栈