Android debuggerd 源码分析

debuggerd 简介

Android系统自带一个实用的程序异常退出的诊断daemon debuggerd。此进程可以侦测到程序崩溃，并将崩溃时的进程状态信息输出到文件和串口中，以供开发人员分析调试使用。Debuggerd的数据被保存在/data/tombstone/目录下，共可保存10个文件，当超过10个时，会覆盖重写最早生产的文件。串口中，则直接用DEBUG的tag，输出logcat信息。 Linux kernel有自己的一套signal机制，在应用程序崩溃时，通常系统内核都会发送signal到出问题的进程，以通知进程出现什么异常，这些进程可以捕获这些signal并对其做相应的处理。通常对于程序异常信号的处理，就是退出。Android在此机制上实现了一个更实用的功能：拦截这些信号，dump进程信息以供调试。

debuggerd的运行原理

debuggerd创建一个名为 “Android:debuggerd”的socket，作为server端等待其他client端进程的连接，接收client端进程发送来的tid和action信息将由tid指定的那个进程的运行信息，按照由action指定的动作dump到文件或者控制台中可以作为debuggerd的client端的进程主要有几种：

1.       异常的C/C++程序

这种程序由bionic的linker安装异常信号的处理函数，当程序产生异常信号时，进入信号处理函数，与debuggerd建立。

2.       debuggerd程序

debuggerd可以在控制台中以命令debuggerd -b [<tid>]启动 ，然后与debuggerd daemon建立连接。这样debuggerd可以在不中断进程执行的情况下dump由tid指定的进程的信息。

3.       dumpstate

控制台中运行命令dumpstate，并指定必要的参数，命令中会调用dump_backtrace_to_file与debuggerd交互。

debuggerd的使用方法

产生异常信号的C/C++程序与debuggerd建立连接后，debuggerd将进程信息dump到tombstone_XX文件中保存到/data/tombstone/文件夹下。可通过查看tombstone_XX分析异常进程的堆栈信息。

在控制台中以命令debuggerd -b [<tid>]启动。如果加上-b参数，则由tid指定的进程的信息将dump到控制台上，否则dump到tombstone文件中。控制台中运行命令callstack/dumpstate，进程信息会写入这两个命令指定的文件中。

应用程序异常处理过程

应用程序入口属于bionic实现的一部分，则对所有android的程序有效。在应用程序入口地址__start后，__linker_init中调用debugger_init()函数来注册异常信号处理handler，以实现拦截系统异常的几个singal：SIGILL,SIGABRT, SIGBUS, SIGFPE,SIGSEGV和SIGPIPE：

linker/arch/arm/begin.S

start:
        mov     r0, sp
        mov     r1, #0
        bl      __linker_init

bionic\linker\ Linker.cpp

extern "C" Elf32_Addr __linker_init(void* raw_args) {
  …
  Elf32_Addr start_address = __linker_init_post_relocation(args, linker_addr);
  set_soinfo_pool_protection(PROT_READ);
  // Return the address that the calling assembly stub should jump to.
  return start_address;
}

static Elf32_Addr __linker_init_post_relocation(KernelArgumentBlock& args, Elf32_Addr linker_base) {
     ...
    debuggerd_init();
     ...
}

bionic\linker\Debugger.c

void debugger_init()
{
    struct sigaction act;
    memset(&act, 0, sizeof(act));
    act.sa_sigaction = debugger_signal_handler;
    act.sa_flags = SA_RESTART | SA_SIGINFO;
    sigemptyset(&act.sa_mask);
    sigaction(SIGILL, &act, NULL);
    sigaction(SIGABRT, &act, NULL);
    sigaction(SIGBUS, &act, NULL);
    sigaction(SIGFPE, &act, NULL);
    sigaction(SIGSEGV, &act, NULL);
    sigaction(SIGSTKFLT, &act, NULL);
    sigaction(SIGPIPE, &act, NULL);
}

bionic库中的链接器会对以下七种信号设置Handler(debugger_signal_handler)：

1. SIGILL(非法指令异常)
2. SIGABRT(abort退出异常)
3. SIGBUS(硬件访问异常)
4. SIGFPE(浮点运算异常)
5. SIGSEGV(内存访问异常)
6. SIGSTKFLT(协处理器栈异常)
7. SIGPIPE(管道异常)

debugger_init中act.sa_flags = SA_RESTART | SA_SIGINFO的涵义：

1)      SA_RESTART

如果指定该参数，表示若信号中断了进程的某个系统调用，则系统自动启动该系统调用。如果不指定该参数，则被中断的系统调用返回失败，错误码为EINTR。这个标志位只要用于处理慢系统调用(可能会被阻塞的系统调用)。比如调用write系统调用写某个设备被阻塞，这时进程捕获某个信号且进入相应信号处理函数返回时，该系统调用可能要返回ENINTR错误。指定这个参数后，系统调用会重启，与RETRY_ON_EINTR宏配合使用则可以保证写操作的完成

2)      SA_SIGINFO

如果指定该参数，表示信号附带的参数(siginfo_t结构体)可以被传递到信号处理函数中。

链接到bionic库上的C/C++程序崩溃时，内核会发送相应的signal，进程收到异常信号后，会转入debugger_signal_handler函数中进行处理。

void debugger_signal_handler(int n, siginfo_t* info, void* unused)
{
    char msgbuf[128];
    unsigned tid;
    int s;

    logSignalSummary(n, info);
 
    tid = gettid();
	//"android:debuggerd"
    s = socket_abstract_client(DEBUGGER_SOCKET_NAME, SOCK_STREAM);
 
    if (s >= 0) {
        /* debugger knows our pid from the credentials on the
         * local socket but we need to tell it our tid.  It
         * is paranoid and will verify that we are giving a tid
         * that's actually in our process
         */
        int  ret;
        debugger_msg_t msg;
        msg.action = DEBUGGER_ACTION_CRASH;
        msg.tid = tid;
        RETRY_ON_EINTR(ret, write(s, &msg, sizeof(msg)));
        if (ret == sizeof(msg)) {
            /* if the write failed, there is no point to read on
             * the file descriptor. */
            RETRY_ON_EINTR(ret, read(s, &tid, 1));
            int savedErrno = errno;
            notify_gdb_of_libraries();
            errno = savedErrno;
        }
 
        if (ret < 0) {
            /* read or write failed -- broken connection? */
            format_buffer(msgbuf, sizeof(msgbuf),
                "Failed while talking to debuggerd: %s", strerror(errno));
            __libc_android_log_write(ANDROID_LOG_FATAL, "libc", msgbuf);
        }
 
        close(s);
    } else {
        /* socket failed; maybe process ran out of fds */
        format_buffer(msgbuf, sizeof(msgbuf),
            "Unable to open connection to debuggerd: %s", strerror(errno));
        __libc_android_log_write(ANDROID_LOG_FATAL, "libc", msgbuf);
    }
 
    /* remove our net so we fault for real when we return */
    signal(n, SIG_DFL);
 
    /*
     * These signals are not re-thrown when we resume.  This means that
     * crashing due to (say) SIGPIPE doesn't work the way you'd expect it
     * to.  We work around this by throwing them manually.  We don't want
     * to do this for *all* signals because it'll screw up the address for
     * faults like SIGSEGV.
     */
    switch (n) {
        case SIGABRT:
        case SIGFPE:
        case SIGPIPE:
        case SIGSTKFLT:
            (void) tgkill(getpid(), gettid(), n);
            break;
        default:    // SIGILL, SIGBUS, SIGSEGV
            break;
    }
}

debugger_signal_handler函数处理流程:

1) 调用logSignalSummary将signal信息写入文件；

static void logSignalSummary(int signum, const siginfo_t* info)
{
    char buffer[128];
    char threadname[MAX_TASK_NAME_LEN + 1]; // one more for termination
    char* signame;
    switch (signum) {
        case SIGILL:    signame = "SIGILL";     break;
        case SIGABRT:   signame = "SIGABRT";    break;
        case SIGBUS:    signame = "SIGBUS";     break;
        case SIGFPE:    signame = "SIGFPE";     break;
        case SIGSEGV:   signame = "SIGSEGV";    break;
        case SIGSTKFLT: signame = "SIGSTKFLT";  break;
        case SIGPIPE:   signame = "SIGPIPE";    break;
        default:        signame = "???";        break;
    }

    if (prctl(PR_GET_NAME, (unsigned long)threadname, 0, 0, 0) != 0) {
        strcpy(threadname, "<name unknown>");
    } else {
        // short names are null terminated by prctl, but the manpage
        // implies that 16 byte names are not.
        threadname[MAX_TASK_NAME_LEN] = 0;
    }
    format_buffer(buffer, sizeof(buffer),
        "Fatal signal %d (%s) at 0x%08x (code=%d), thread %d (%s)",
        signum, signame, info->si_addr, info->si_code, gettid(), threadname);

    __libc_android_log_write(ANDROID_LOG_FATAL, "libc", buffer);
}

获取异常信号的名字和thread名字，并格式化字符串,调用函数__libc_android_log_write函数写入”/dev/log/main”中。

2) 调用socket_abstract_client函数与debuggerd建立socket连接；

s = socket_abstract_client(DEBUGGER_SOCKET_NAME, SOCK_STREAM);

3) 如果连接建立成功，则设置结构体debugger_msg_t，并发送给debuggerd；

msg.action = DEBUGGER_ACTION_CRASH;//告诉debuggerd采取何种行
msg.tid = tid;//线程号
RETRY_ON_EINTR(ret, write(s, &msg, sizeof(msg)));

4) 等待debuggerd的回复，阻塞在下面的调用中，收到回复后接着执行下面的流程；

RETRY_ON_EINTR(ret, read(s, &tid, 1));

5)  重新设置信号处理函数为SIG_DFL，即采取默认的动作；

signal(n, SIG_DFL);

6)  重新发送信号，进程从当前信号处理函数返回后，会处理这个信号，进行默认的信号处理动作，即中断进程。

debuggerd的源码分析

1.       在init进程中以deamon的方式启动，在init.rc中

service debuggerd /system/bin/debuggerd 
class main

以这种方式启动的话，进入main函数后，将调用do_server函数，作为server端为其他进程提供dump进程信息的服务。

2.       直接运行system/bin/debuggerd可执行文件，需要指定参数，用法为：

debuggerd -b [<tid>] //参数-b表示在控制台中输出backtrace

以这种方式启动的话，进入main函数后，将调用do_explicit_dump函数与debuggerd daemon通信，将指定进程的信息dump到文件或控制台。

服务启动方式

int main(int argc, char** argv) {
    if (argc == 1) {
        return do_server();
    }
}

当启动debuggerd进程传递的参数个数为1时，此时启动的debuggerd将作为一个后台服务进程，专门接收应用程序异常退出消息而产生tombstone。

static int do_server() {
    int s;
    struct sigaction act;
    int logsocket = -1;
 
    /*
     * debuggerd crashes can't be reported to debuggerd.  Reset all of the
     * crash handlers.
     */
    signal(SIGILL, SIG_DFL);
    signal(SIGABRT, SIG_DFL);
    signal(SIGBUS, SIG_DFL);
    signal(SIGFPE, SIG_DFL);
    signal(SIGSEGV, SIG_DFL);
    signal(SIGPIPE, SIG_IGN);
    signal(SIGSTKFLT, SIG_DFL);
 
    logsocket = socket_local_client("logd",
            ANDROID_SOCKET_NAMESPACE_ABSTRACT, SOCK_DGRAM);
    if(logsocket < 0) {
        logsocket = -1;
    } else {
        fcntl(logsocket, F_SETFD, FD_CLOEXEC);
    }
 
    act.sa_handler = SIG_DFL;
    sigemptyset(&act.sa_mask);
    sigaddset(&act.sa_mask,SIGCHLD);
    act.sa_flags = SA_NOCLDWAIT;
    sigaction(SIGCHLD, &act, 0);
 
    s = socket_local_server(DEBUGGER_SOCKET_NAME,
            ANDROID_SOCKET_NAMESPACE_ABSTRACT, SOCK_STREAM);
    if(s < 0) return 1;
    fcntl(s, F_SETFD, FD_CLOEXEC);
 
    LOG("debuggerd: " __DATE__ " " __TIME__ "\n");
 
    //check corefile limit.
    (void)check_corefile_limit();
 
    for(;;) {
        struct sockaddr addr;
        socklen_t alen;
        int fd;
        alen = sizeof(addr);
        XLOG("waiting for connection\n");
        fd = accept(s, &addr, &alen);
        if(fd < 0) {
            XLOG("accept failed: %s\n", strerror(errno));
            continue;
        }
 
        fcntl(fd, F_SETFD, FD_CLOEXEC);
 
        handle_request(fd);
    }
    return 0;
}

1.       忽略debuggerd自身crash的处理；

2.       建立socket通信的server端；

3.       进入无限循环中，等待并接收客户端进程连接请求，并通过handle_request()函数处理请求；

handle_request

static void handle_request(int fd) {
    XLOG("handle_request(%d)\n", fd);
 
    debugger_request_t request;
    int status = read_request(fd, &request);
    if (!status) {
        XLOG("BOOM: pid=%d uid=%d gid=%d tid=%d\n",
            request.pid, request.uid, request.gid, request.tid);
 
        /* At this point, the thread that made the request is blocked in
         * a read() call.  If the thread has crashed, then this gives us
         * time to PTRACE_ATTACH to it before it has a chance to really fault.
         *
         * The PTRACE_ATTACH sends a SIGSTOP to the target process, but it
         * won't necessarily have stopped by the time ptrace() returns.  (We
         * currently assume it does.)  We write to the file descriptor to
         * ensure that it can run as soon as we call PTRACE_CONT below.
         * See details in bionic/libc/linker/debugger.c, in function
         * debugger_signal_handler().
         */
        if (ptrace(PTRACE_ATTACH, request.tid, 0, 0)) {
            LOG("ptrace attach failed: %s\n", strerror(errno));
        } else {
            bool detach_failed = false;
            bool attach_gdb = should_attach_gdb(&request);
            if (TEMP_FAILURE_RETRY(write(fd, "\0", 1)) != 1) {
                LOG("failed responding to client: %s\n", strerror(errno));
            } else {
                char* tombstone_path = NULL;
 
                if (request.action == DEBUGGER_ACTION_CRASH) {
                    close(fd);
                    fd = -1;
                }
 
                int total_sleep_time_usec = 0;
                for (;;) {
                    int signal = wait_for_signal(request.tid, &total_sleep_time_usec);
                    if (signal < 0) {
                        break;
                    }
 
                    switch (signal) {
                    case SIGSTOP:
                        if (request.action == DEBUGGER_ACTION_DUMP_TOMBSTONE) {
                            XLOG("stopped -- dumping to tombstone\n");
                            tombstone_path = engrave_tombstone(request.pid, request.tid,
                                    signal, true, true, &detach_failed,
                                    &total_sleep_time_usec);
                        } else if (request.action == DEBUGGER_ACTION_DUMP_BACKTRACE) {
                            XLOG("stopped -- dumping to fd\n");
                            dump_backtrace(fd, request.pid, request.tid, &detach_failed,
                                    &total_sleep_time_usec);
                        } else {
                            XLOG("stopped -- continuing\n");
                            status = ptrace(PTRACE_CONT, request.tid, 0, 0);
                            if (status) {
                                LOG("ptrace continue failed: %s\n", strerror(errno));
                            }
                            continue; /* loop again */
                        }
                        break;
 
                    case SIGILL:
                    case SIGABRT:
                    case SIGBUS:
                    case SIGFPE:
                    case SIGSEGV:
                    case SIGSTKFLT: {
                        XLOG("stopped -- fatal signal\n");
                        /*
                         * Send a SIGSTOP to the process to make all of
                         * the non-signaled threads stop moving.  Without
                         * this we get a lot of "ptrace detach failed:
                         * No such process".
                         */
                        kill(request.pid, SIGSTOP);
                        /* don't dump sibling threads when attaching to GDB because it
                         * makes the process less reliable, apparently... */
                        tombstone_path = engrave_tombstone(request.pid, request.tid,
                                signal, !attach_gdb, false, &detach_failed,
                                &total_sleep_time_usec);
                        break;
                    }
 
                    case SIGPIPE:
                        LOG("socket-client process stopped due to SIGPIPE! \n");
                        break;
 
                    default:
                        XLOG("stopped -- unexpected signal\n");
                        LOG("process stopped due to unexpected signal %d\n", signal);
                        break;
                    }
                    break;
                }
 
                if (request.action == DEBUGGER_ACTION_DUMP_TOMBSTONE) {
                    if (tombstone_path) {
                        write(fd, tombstone_path, strlen(tombstone_path));
                    }
                    close(fd);
                    fd = -1;
                }
                free(tombstone_path);
            }
 
            XLOG("detaching\n");
            if (attach_gdb) {
                /* stop the process so we can debug */
                kill(request.pid, SIGSTOP);
 
                /* detach so we can attach gdbserver */
                if (ptrace(PTRACE_DETACH, request.tid, 0, 0)) {
                    LOG("ptrace detach from %d failed: %s\n", request.tid, strerror(errno));
                    detach_failed = true;
                }
 
                /*
                 * if debug.db.uid is set, its value indicates if we should wait
                 * for user action for the crashing process.
                 * in this case, we log a message and turn the debug LED on
                 * waiting for a gdb connection (for instance)
                 */
                wait_for_user_action(request.pid);
            } else {
                /* just detach */
                if (ptrace(PTRACE_DETACH, request.tid, 0, 0)) {
                    LOG("ptrace detach from %d failed: %s\n", request.tid, strerror(errno));
                    detach_failed = true;
                }
            }
 
            /* resume stopped process (so it can crash in peace). */
            kill(request.pid, SIGCONT);
 
            /* If we didn't successfully detach, we're still the parent, and the
             * actual parent won't receive a death notification via wait(2).  At this point
             * there's not much we can do about that. */
            if (detach_failed) {
                LOG("debuggerd committing suicide to free the zombie!\n");
                kill(getpid(), SIGKILL);
            }
        }
 
    }
    if (fd >= 0) {
        close(fd);
    }
}

 1)  调用read_request函数读取client端进程发送来的数据：

static int read_request(int fd, debugger_request_t* out_request) {
    struct ucred cr;
    int len = sizeof(cr);
    int status = getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &cr, &len);
    if (status != 0) {
        LOG("cannot get credentials\n");
        return -1;
    }

    XLOG("reading tid\n");
    fcntl(fd, F_SETFL, O_NONBLOCK);

    struct pollfd pollfds[1];
    pollfds[0].fd = fd;
    pollfds[0].events = POLLIN;
    pollfds[0].revents = 0;
    status = TEMP_FAILURE_RETRY(poll(pollfds, 1, 3000));
    if (status != 1) {
        LOG("timed out reading tid\n");
        return -1;
    }

    debugger_msg_t msg;
    status = TEMP_FAILURE_RETRY(read(fd, &msg, sizeof(msg)));
    if (status < 0) {
        LOG("read failure? %s\n", strerror(errno));
        return -1;
    }
    if (status != sizeof(msg)) {
        LOG("invalid crash request of size %d\n", status);
        return -1;
    }

    out_request->action = msg.action;
    out_request->tid = msg.tid;
    out_request->pid = cr.pid;
    out_request->uid = cr.uid;
    out_request->gid = cr.gid;

    if (msg.action == DEBUGGER_ACTION_CRASH) {
        /* Ensure that the tid reported by the crashing process is valid. */
        char buf[64];
        struct stat s;
        snprintf(buf, sizeof buf, "/proc/%d/task/%d", out_request->pid, out_request->tid);
        if(stat(buf, &s)) {
            LOG("tid %d does not exist in pid %d. ignoring debug request\n",
                    out_request->tid, out_request->pid);
            return -1;
        }
    } else if (cr.uid == 0
            || (cr.uid == AID_SYSTEM && msg.action == DEBUGGER_ACTION_DUMP_BACKTRACE)) {
        /* Only root or system can ask us to attach to any process and dump it explicitly.
         * However, system is only allowed to collect backtraces but cannot dump tombstones. */
        status = get_process_info(out_request->tid, &out_request->pid,
                &out_request->uid, &out_request->gid);
        if (status < 0) {
            LOG("tid %d does not exist. ignoring explicit dump request\n",
                    out_request->tid);
            return -1;
        }
    } else {
        /* No one else is not allowed to dump arbitrary processes. */
        return -1;
    }
    return 0;
}

?  从socket中读取client端进程的pid uid gid

getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &cr, &len);

?  轮询socket句柄

struct pollfd pollfds[1];
pollfds[0].fd = fd;
pollfds[0].events = POLLIN;
pollfds[0].revents = 0;
status = TEMP_FAILURE_RETRY(poll(pollfds, 1, 3000));

?  从socket上读取debugger_msg_t结构体

debugger_msg_t msg;
status = TEMP_FAILURE_RETRY(read(fd, &msg, sizeof(msg)));
if (status < 0) {
	LOG("read failure? %s\n", strerror(errno));
	return -1;
}
if (status != sizeof(msg)) {
	LOG("invalid crash request of size %d\n", status);
	return -1;
}
out_request->action = msg.action;
out_request->tid = msg.tid;
out_request->pid = cr.pid;
out_request->uid = cr.uid;
out_request->gid = cr.gid;

? 如果debugger_msg_t中设置的action为DEBUGGER_ACTION_CRASH，说明是crash的C/C++进程发来的请求，则判断传进来的tid是否有效。

if (msg.action == DEBUGGER_ACTION_CRASH) {
	/* Ensure that the tid reported by the crashing process is valid. */
	char buf[64];
	struct stat s;
	snprintf(buf, sizeof buf, "/proc/%d/task/%d", out_request->pid, out_request->tid);
	if(stat(buf, &s)) {
		LOG("tid %d does not exist in pid %d. ignoring debug request\n",
				out_request->tid, out_request->pid);
		return -1;
	}
}

?   如果debugger_msg_t中设置的action为DEBUGGER_ACTION_DUMP_BACKTRACE说明是其他方式(debuggerd)发来的请求，则要求必须为root权限或者system权限，然后再判断tid是否有效。

 2)  从read_request返回后，调用ptrace函数attach到tid指定的进程，此时debuggerd将变为被attache进程的父进程，然后ptrace函数会向子进程发送SIGSTOP信号将子进程停下来。此时，父进程有机会检查子进程核心image和寄存器的值。

ptrace(PTRACE_ATTACH, request.tid, 0, 0)

 3)  调用下面的语句给client端子进程回复消息，使clinet端的进程能从read调用中返回。

TEMP_FAILURE_RETRY(write(fd, "\0", 1)

 4)  在for循环中等待子进程停止。

int signal = wait_for_signal(request.tid, &total_sleep_time_usec);

 5)  子进程根据收到的不同信号、不同的action进行不同的处理

switch (signal) {
	case SIGSTOP:
		if (request.action == DEBUGGER_ACTION_DUMP_TOMBSTONE) {
			XLOG("stopped -- dumping to tombstone\n");
			tombstone_path = engrave_tombstone(request.pid, request.tid,
					signal, true, true, &detach_failed,
					&total_sleep_time_usec);
		} else if (request.action == DEBUGGER_ACTION_DUMP_BACKTRACE) {
			XLOG("stopped -- dumping to fd\n");
			dump_backtrace(fd, request.pid, request.tid, &detach_failed,
					&total_sleep_time_usec);
		} else {
			XLOG("stopped -- continuing\n");
			status = ptrace(PTRACE_CONT, request.tid, 0, 0);
			if (status) {
				LOG("ptrace continue failed: %s\n", strerror(errno));
			}
			continue; /* loop again */
		}
		break;
	case SIGILL:
	case SIGABRT:
	case SIGBUS:
	case SIGFPE:
	case SIGSEGV:
	case SIGSTKFLT: {
		XLOG("stopped -- fatal signal\n");
		kill(request.pid, SIGSTOP);
		tombstone_path = engrave_tombstone(request.pid, request.tid,
				signal, !attach_gdb, false, &detach_failed,
				&total_sleep_time_usec);
		break;
	}
	case SIGPIPE:
		LOG("socket-client process stopped due to SIGPIPE! \n");
		break;
	default:
		XLOG("stopped -- unexpected signal\n");
		LOG("process stopped due to unexpected signal %d\n", signal);
		break;
}

?子进程收到SIGSTOP说明进程并没有发生crash，根据action不同将进程信息写入tombstone文件。

?子进程收到七种异常信号说明是进程发生crash，调用engrave_tombstone直接将dump的信息写到tombstone。

 6)  调用ptrace(PTRACE_DETACH, request.tid, 0, 0)解除对子进程的追踪；

if (attach_gdb) {
	kill(request.pid, SIGSTOP);
	if (ptrace(PTRACE_DETACH, request.tid, 0, 0)) {
		LOG("ptrace detach from %d failed: %s\n", request.tid, strerror(errno));
		detach_failed = true;
	}
	wait_for_user_action(request.pid);
} else {
	if (ptrace(PTRACE_DETACH, request.tid, 0, 0)) {
		LOG("ptrace detach from %d failed: %s\n", request.tid, strerror(errno));
		detach_failed = true;
	}
}

如果运行了类似以下指令：adb shell setprop debug.db.uid 10000；则所有uid<10000的进程发生crash的时候attach_gdb为true，将停止crash进程，并调用ptrace(PTRACE_DETACH, request.tid, 0, 0) 解除对crash进程的追踪后，开始等待gdb的连接。

adb forward tcp:5039 tcp:5039

adb shell gdbserver :5039 --attach pid &

用户按下HOME或者VOLUME DOWN按键，可以使进程继续进行，自然crash

attach_gdb为false时，只会解除对子进程的追踪。

 7)  调用kill(request.pid, SIGCONT)恢复被停止的子进程，并让其自然终止；

engrave_tombstone

char* engrave_tombstone(pid_t pid, pid_t tid, int signal,
        bool dump_sibling_threads, bool quiet, bool* detach_failed,
        int* total_sleep_time_usec) {
    mkdir(TOMBSTONE_DIR, 0755);
    chown(TOMBSTONE_DIR, AID_SYSTEM, AID_SYSTEM);

    //dump maps & check corefile limit .
    dump_creash_maps(pid);  //creat maps file

    int fd;
    char* path = find_and_open_tombstone(&fd);
    if (!path) {
        *detach_failed = false;
        return NULL;
    }

    log_t log;
    log.tfd = fd;
    log.quiet = quiet;
    *detach_failed = dump_crash(&log, pid, tid, signal, dump_sibling_threads,
            total_sleep_time_usec);

    close(fd);
    return path;
}

对于crash的C/C++进程，主要通过这个函数dump进程信息

1.创建”/data/tombstones”文件夹并修改权限

2.调用函数find_and_open_tombstone，tombstone_XX文件最多10个，超过则覆盖最早的

3.调用dump_crash将所有信息dump到tombstone文件：

? dump_build_info(log);

? dump_thread_info(log, pid, tid, true);

? dump_fault_addr(log, tid, signal);

? dump_thread(context, log, tid, true, total_sleep_time_usec) dump进程的上下文信息

? dump_logs(log, pid, true);

? dump_sibling_thread_report(context, log, pid, tid, total_sleep_time_usec);

dump_backtrace

void dump_backtrace(int fd, pid_t pid, pid_t tid, bool* detach_failed,
        int* total_sleep_time_usec) {
    log_t log;
    log.tfd = fd;
    log.quiet = true;

    ptrace_context_t* context = load_ptrace_context(tid);
    dump_process_header(&log, pid);
    dump_thread(&log, tid, context, true, detach_failed, total_sleep_time_usec);

    char task_path[64];
    snprintf(task_path, sizeof(task_path), "/proc/%d/task", pid);
    DIR* d = opendir(task_path);
    if (d) {
        struct dirent debuf;
        struct dirent *de;
        while (!readdir_r(d, &debuf, &de) && de) {
            if (!strcmp(de->d_name, ".") || !strcmp(de->d_name, "..")) {
                continue;
            }

            char* end;
            pid_t new_tid = strtoul(de->d_name, &end, 10);
            if (*end || new_tid == tid) {
                continue;
            }

            dump_thread(&log, new_tid, context, false, detach_failed, total_sleep_time_usec);
        }
        closedir(d);
    }

    dump_process_footer(&log, pid);
    free_ptrace_context(context);
}

?  dump_process_header(&log, pid);

?  dump_thread(&log, tid, context, true, detach_failed, total_sleep_time_usec);

? dump_process_footer(&log, pid);      

调试工具方式

int main(int argc, char** argv) {
    bool dump_backtrace = false;
    bool have_tid = false;
    pid_t tid = 0;
    for (int i = 1; i < argc; i++) {
        if (!strcmp(argv[i], "-b")) {
            dump_backtrace = true;
        } else if (!have_tid) {
            tid = atoi(argv[i]);
            have_tid = true;
        } else {
            usage();
            return 1;
        }
    }
    if (!have_tid) {
        usage();
        return 1;
    }
    return do_explicit_dump(tid, dump_backtrace);
}

通过do_explicit_dump函数dump出指定进程的栈信息等

static int do_explicit_dump(pid_t tid, bool dump_backtrace) {
    fprintf(stdout, "Sending request to dump task %d.\n", tid);
 
    if (dump_backtrace) {
        fflush(stdout);
        if (dump_backtrace_to_file(tid, fileno(stdout)) < 0) {
            fputs("Error dumping backtrace.\n", stderr);
            return 1;
        }
    } else {
        char tombstone_path[PATH_MAX];
        if (dump_tombstone(tid, tombstone_path, sizeof(tombstone_path)) < 0) {
            fputs("Error dumping tombstone.\n", stderr);
            return 1;
        }
        fprintf(stderr, "Tombstone written to: %s\n", tombstone_path);
    }
    return 0;
}

?  dump_backtrace_to_file(tid, fileno(stdout))

?  dump_tombstone(tid, tombstone_path, sizeof(tombstone_path))

原文：http://blog.csdn.net/yangwen123/article/details/38397359