调试系列1:bugreport源码篇

Posted by Gityuan on June 10, 2016

基于android 6.0, 分析bugreport过程

framework/native/cmds/bugreport/bugreport.cpp
framework/native/cmds/dumpstate/dumpstate.cpp
framework/native/cmds/dumpstate/utils.c

一、概述

通过adb命令可获取bugrepport信息,并输出到文件当前路径的bugreport.txt文件:

adb bugreport > bugreport.txt

对于Android系统调试分析,bugreport信息量非常之大,几乎涵盖整个系统各个层面内容,对于分析BUG是一大利器,本文先从从源码角度来分析一下Bugreport的实现原理。

二、原理分析

Android系统源码中framework/native/cmds/bugreport目录通过Android.mk定义了bugreport项目,在系统编译完成后会生成bugreport可执行文件,位于系统/system/bin/bugreport。当执行adb bugreport时,便会调用这个可执行文件,进入bugreport.cpp中的main()方法。

2.1 bugreport.main

[-> bugreport.cpp]

int main() {
  //启动dumpstate服务
  property_set("ctl.start", "dumpstate");
  //需要多次尝试,直到dumpstate服务启动完成,才能建立socket通信
  int s;
  for (int i = 0; i < 20; i++) {
    s = socket_local_client("dumpstate", ANDROID_SOCKET_NAMESPACE_RESERVED,
                            SOCK_STREAM);
    if (s >= 0)
      break;
    //休眠1s后再次尝试连接
    sleep(1);
  }
  if (s == -1) {
    printf("Failed to connect to dumpstate service: %s\n", strerror(errno));
    return 1;
  }
  //当3分钟没有任何数据可读,则超时停止读取并退出。
  //dumpstate服务中不存在大于1分钟的timetout,因而不可预见的超时的情况下留有很大的回旋余地。
  struct timeval tv;
  tv.tv_sec = 3 * 60;
  tv.tv_usec = 0;
  if (setsockopt(s, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv)) == -1) {
    printf("WARNING: Cannot set socket timeout: %s\n", strerror(errno));
  }
  while (1) {
    char buffer[65536];
    ssize_t bytes_read = TEMP_FAILURE_RETRY(read(s, buffer, sizeof(buffer)));
    if (bytes_read == 0) {
      break;
    } else if (bytes_read == -1) {
      // EAGAIN意味着timeout,Bugreport读异常终止
      if (errno == EAGAIN) {
        errno = ETIMEDOUT;
      }
      break;
    }
    ssize_t bytes_to_send = bytes_read;
    ssize_t bytes_written;
    //不断循环得将读取数据输出到stdout
    do {
      bytes_written = TEMP_FAILURE_RETRY(write(STDOUT_FILENO,
                       buffer + bytes_read - bytes_to_send, bytes_to_send));
      if (bytes_written == -1) {
        return 1; //将数据无法写入stdout
      }
      bytes_to_send -= bytes_written;
    } while (bytes_written != 0 && bytes_to_send > 0);
  }
  close(s);
  return 0;
}

该过程先启动dumpstate服务,Bugreport再通过socket建立于dumpstate的通信,这个过程会尝试20次socket连接建立直到成功连接。 在socket通道中如果持续3分钟没有任何数据可读,则超时停止读取并退出。由于dumpstate服务中不存在大于1分钟的timetout,因而不可预见的超时的情况下留有很大的回旋余地。

当从socket读取到数据后,写入到标准时输出或者重定向到文件。可见bugreport数据的来源都是dumpstate服务,那么接下来去看看dumpstate服务的工作。

2.2 dumpstate.main

[-> dumpstate.cpp]

int main(int argc, char *argv[]) {
    struct sigaction sigact;
    int do_add_date = 0;
    int do_vibrate = 1;
    char* use_outfile = 0;
    int use_socket = 0;
    int do_fb = 0;
    int do_broadcast = 0;
    if (getuid() != 0) {
        //兼容性考虑,旧版本支持直接调用dumpstate命令,新版本通过调用/system/bin/bugreport来替代。
        //当检测到直接调用,则强制执行bugreport命令。
        return execl("/system/bin/bugreport", "/system/bin/bugreport", NULL);
    }
    ALOGI("begin\n");
    //清空句柄SIGPIPE
    memset(&sigact, 0, sizeof(sigact));
    sigact.sa_handler = sigpipe_handler;
    sigaction(SIGPIPE, &sigact, NULL);
    //提高当前进程的优先级,防止被OOM Killer杀死
    setpriority(PRIO_PROCESS, 0, -20);
    FILE *oom_adj = fopen("/proc/self/oom_adj", "we");
    if (oom_adj) {
        fputs("-17", oom_adj);
        fclose(oom_adj);
    }
    //参数解析
    int c;
    while ((c = getopt(argc, argv, "dho:svqzpB")) != -1) {
        switch (c) {
            case 'd': do_add_date = 1;       break;
            case 'o': use_outfile = optarg;  break;
            case 's': use_socket = 1;        break;
            case 'v': break;  // compatibility no-op
            case 'q': do_vibrate = 0;        break;
            case 'p': do_fb = 1;             break;
            case 'B': do_broadcast = 1;      break;
            case '?': printf("\n");
            case 'h':
                usage();
                exit(1);
        }
    }
    //建立socket
    if (use_socket) {
        redirect_to_socket(stdout, "dumpstate");
    }
    //打开vibrator
    FILE *vibrator = 0;
    if (do_vibrate) {
        vibrator = fopen("/sys/class/timed_output/vibrator/enable", "we");
        if (vibrator) {
            vibrate(vibrator, 150);
        }
    }
    //读取/proc/cmdline
    FILE *cmdline = fopen("/proc/cmdline", "re");
    if (cmdline != NULL) {
        fgets(cmdline_buf, sizeof(cmdline_buf), cmdline);
        fclose(cmdline);
    }
    //收集虚拟机和native进程的stack traces(需要root权限)
    dump_traces_path = dump_traces();
    //获取tombstone文件描述符
    get_tombstone_fds(tombstone_data);
    //确保capabilities
    if (prctl(PR_SET_KEEPCAPS, 1) < 0) {
        ALOGE("prctl(PR_SET_KEEPCAPS) failed: %s\n", strerror(errno));
        return -1;
    }
    //切换到非root用户和组,在切换之前都是处于root权限
    gid_t groups[] = { AID_LOG, AID_SDCARD_R, AID_SDCARD_RW,
            AID_MOUNT, AID_INET, AID_NET_BW_STATS };
    if (setgroups(sizeof(groups)/sizeof(groups[0]), groups) != 0) {
        ALOGE("Unable to setgroups, aborting: %s\n", strerror(errno));
        return -1;
    }
    if (setgid(AID_SHELL) != 0) {
        ALOGE("Unable to setgid, aborting: %s\n", strerror(errno));
        return -1;
    }
    if (setuid(AID_SHELL) != 0) {
        ALOGE("Unable to setuid, aborting: %s\n", strerror(errno));
        return -1;
    }
    struct __user_cap_header_struct capheader;
    struct __user_cap_data_struct capdata[2];
    memset(&capheader, 0, sizeof(capheader));
    memset(&capdata, 0, sizeof(capdata));
    capheader.version = _LINUX_CAPABILITY_VERSION_3;
    capheader.pid = 0;
    capdata[CAP_TO_INDEX(CAP_SYSLOG)].permitted = CAP_TO_MASK(CAP_SYSLOG);
    capdata[CAP_TO_INDEX(CAP_SYSLOG)].effective = CAP_TO_MASK(CAP_SYSLOG);
    capdata[0].inheritable = 0;
    capdata[1].inheritable = 0;
    if (capset(&capheader, &capdata[0]) < 0) {
        ALOGE("capset failed: %s\n", strerror(errno));
        return -1;
    }
    //如果需要,则重定向输出
    char path[PATH_MAX], tmp_path[PATH_MAX];
    pid_t gzip_pid = -1;
    if (!use_socket && use_outfile) {
        strlcpy(path, use_outfile, sizeof(path));
        if (do_add_date) {
            char date[80];
            time_t now = time(NULL);
            strftime(date, sizeof(date), "-%Y-%m-%d-%H-%M-%S", localtime(&now));
            strlcat(path, date, sizeof(path));
        }
        if (do_fb) {
            strlcpy(screenshot_path, path, sizeof(screenshot_path));
            strlcat(screenshot_path, ".png", sizeof(screenshot_path));
        }
        strlcat(path, ".txt", sizeof(path));
        strlcpy(tmp_path, path, sizeof(tmp_path));
        strlcat(tmp_path, ".tmp", sizeof(tmp_path));
        redirect_to_file(stdout, tmp_path);
    }
    //这里是真正干活的地方 【见小节 3.3】
    dumpstate();
    //通过震动提醒已完成所有dump操作
    if (vibrator) {
        for (int i = 0; i < 3; i++) {
            vibrate(vibrator, 75);
            usleep((75 + 50) * 1000);
        }
        fclose(vibrator);
    }
    //等待gzip的完成,等进程退出时则会被杀
    if (gzip_pid > 0) {
        fclose(stdout);
        waitpid(gzip_pid, NULL, 0);
    }
    //重命名.tmp文件到最终位置
    if (use_outfile && rename(tmp_path, path)) {
        fprintf(stderr, "rename(%s, %s): %s\n", tmp_path, path, strerror(errno));
    }
    //通过发送广播告知ActivityManager已完成bugreport操作
    if (do_broadcast && use_outfile && do_fb) {
        run_command(NULL, 5, "/system/bin/am", "broadcast", "--user", "0",
                "-a", "android.intent.action.BUGREPORT_FINISHED",
                "--es", "android.intent.extra.BUGREPORT", path,
                "--es", "android.intent.extra.SCREENSHOT", screenshot_path,
                "--receiver-permission", "android.permission.DUMP", NULL);
    }
    ALOGI("done\n");
    return 0;
}

整个过程的工作流程:

  1. 提高执行dumpsate所在进程的优先级,防止被OOM Killer杀死;
  2. 参数解析,可通过命令adb shell dumpstate -h查看dumpstate命令所支持的参数;
  3. 打开vibrator,用于在执行bugreport时,手机会先震动一下用于提醒开始抓取系统信息;
  4. 通过dump_traces()来完成收集虚拟机和native进程的stack traces;
  5. 通过get_tombstone_fds来获取tombstone文件描述符;
  6. 开始执行切换到非root用户和组,在这之前的执行都处于root权限;
  7. 执行dumpstate(),这里是真正干活的地方
  8. 再次通过震动以提醒dump操作执行完成;
  9. 发送广播,告知ActivityManager已完成bugreport操作。

接下来就重点说说dumpstate()功能:

2.3 dumpstate()

该方法负责整个bugreport内容输出的最为核心的功能。

[-> dumpstate.cpp ]

static void dumpstate() {
    ...
    property_get("ro.build.display.id", build, "(unknown)");
    property_get("ro.build.fingerprint", fingerprint, "(unknown)");
    property_get("ro.build.type", build_type, "(unknown)");
    property_get("ro.baseband", radio, "(unknown)");
    property_get("ro.bootloader", bootloader, "(unknown)");
    property_get("gsm.operator.alpha", network, "(unknown)");
    strftime(date, sizeof(date), "%Y-%m-%d %H:%M:%S", localtime(&now));
    //开头信息
    printf("========================================================\n");
    printf("== dumpstate: %s\n", date);
    printf("========================================================\n");
    printf("\n");
    printf("Build: %s\n", build);
    printf("Build fingerprint: '%s'\n", fingerprint);
    printf("Bootloader: %s\n", bootloader);
    printf("Radio: %s\n", radio);
    printf("Network: %s\n", network);
    printf("Kernel: "); dump_file(NULL, "/proc/version");
    printf("Command line: %s\n", strtok(cmdline_buf, "\n"));
    printf("\n");
    //记录系统运行时长和休眠时长
    run_command("UPTIME", 10, "uptime", NULL);

    //输出mmcblk0设备信息
    dump_files("UPTIME MMC PERF", mmcblk0, skip_not_stat, dump_stat_from_fd);

    dump_file("MEMORY INFO", "/proc/meminfo");
    run_command("CPU INFO", 10, "top", "-n", "1", "-d", "1", "-m", "30", "-t", NULL);
    run_command("PROCRANK", 20, "procrank", NULL);
    dump_file("VIRTUAL MEMORY STATS", "/proc/vmstat");
    dump_file("VMALLOC INFO", "/proc/vmallocinfo");
    dump_file("SLAB INFO", "/proc/slabinfo");
    dump_file("ZONEINFO", "/proc/zoneinfo");
    dump_file("PAGETYPEINFO", "/proc/pagetypeinfo");
    dump_file("BUDDYINFO", "/proc/buddyinfo");
    dump_file("FRAGMENTATION INFO", "/d/extfrag/unusable_index");
    dump_file("KERNEL WAKELOCKS", "/proc/wakelocks");
    dump_file("KERNEL WAKE SOURCES", "/d/wakeup_sources");
    dump_file("KERNEL CPUFREQ", "/sys/devices/system/cpu/cpu0/cpufreq/stats/time_in_state");
    dump_file("KERNEL SYNC", "/d/sync");
    run_command("PROCESSES", 10, "ps", "-P", NULL);
    run_command("PROCESSES AND THREADS", 10, "ps", "-t", "-p", "-P", NULL);
    run_command("PROCESSES (SELINUX LABELS)", 10, "ps", "-Z", NULL);
    run_command("LIBRANK", 10, "librank", NULL);

    //输出kernel log
    do_dmesg();

    //所有已打开文件
    run_command("LIST OF OPEN FILES", 10, SU_PATH, "root", "lsof", NULL);
    //遍历所有进程的show map
    for_each_pid(do_showmap, "SMAPS OF ALL PROCESSES");
    //显示所有线程的blocked位置
    for_each_tid(show_wchan, "BLOCKED PROCESS WAIT-CHANNELS");

    //SYSTEM LOG
    timeout = logcat_timeout("main") + logcat_timeout("system") + logcat_timeout("crash");
    if (timeout < 20000) {
        timeout = 20000;
    }
    run_command("SYSTEM LOG", timeout / 1000, "logcat", "-v", "threadtime", "-d", "*:v", NULL);

    //EVENT LOG
    timeout = logcat_timeout("events");
    if (timeout < 20000) {
        timeout = 20000;
    }
    run_command("EVENT LOG", timeout / 1000, "logcat", "-b", "events", "-v", "threadtime", "-d", "*:v", NULL);

    //RADIO LOG
    timeout = logcat_timeout("radio");
    if (timeout < 20000) {
        timeout = 20000;
    }
    run_command("RADIO LOG", timeout / 1000, "logcat", "-b", "radio", "-v", "threadtime", "-d", "*:v", NULL);

    //Log统计信息
    run_command("LOG STATISTICS", 10, "logcat", "-b", "all", "-S", NULL);

    //输出当前虚拟机和native进程的vm traces
    if (dump_traces_path != NULL) {
        dump_file("VM TRACES JUST NOW", dump_traces_path);
    }

    //输出上次发生ANR时vm traces,即路径/data/anr/traces.txt
    struct stat st;
    char anr_traces_path[PATH_MAX];
    property_get("dalvik.vm.stack-trace-file", anr_traces_path, "");
    if (!anr_traces_path[0]) {
        printf("*** NO VM TRACES FILE DEFINED (dalvik.vm.stack-trace-file)\n\n");
    } else {
      int fd = TEMP_FAILURE_RETRY(open(anr_traces_path,
                                   O_RDONLY | O_CLOEXEC | O_NOFOLLOW | O_NONBLOCK));
      if (fd < 0) {
          printf("*** NO ANR VM TRACES FILE (%s): %s\n\n", anr_traces_path, strerror(errno));
      } else {
          dump_file_from_fd("VM TRACES AT LAST ANR", anr_traces_path, fd);
      }
    }

    //输出慢操作的vm traces,例如/data/anr/slow1.txt
    if (anr_traces_path[0] != 0) {
        int tail = strlen(anr_traces_path)-1;
        while (tail > 0 && anr_traces_path[tail] != '/') {
            tail--;
        }
        int i = 0;
        while (1) {
            //例如trace文件为/data/anr/slow1.txt
            sprintf(anr_traces_path+tail+1, "slow%02d.txt", i);
            if (stat(anr_traces_path, &st)) {
                break;
            }
            dump_file("VM TRACES WHEN SLOW", anr_traces_path);
            i++;
        }
    }

    //输出tombstone信息,NUM_TOMBSTONES=10,例如/data/tombstones/tombstone_1
    int dumped = 0;
    for (size_t i = 0; i < NUM_TOMBSTONES; i++) {
        if (tombstone_data[i].fd != -1) {
            dumped = 1;
            dump_file_from_fd("TOMBSTONE", tombstone_data[i].name, tombstone_data[i].fd);
            tombstone_data[i].fd = -1;
        }
    }
    if (!dumped) {
        printf("*** NO TOMBSTONES to dump in %s\n\n", TOMBSTONE_DIR);
    }

    dump_file("NETWORK DEV INFO", "/proc/net/dev");
    dump_file("QTAGUID NETWORK INTERFACES INFO", "/proc/net/xt_qtaguid/iface_stat_all");
    dump_file("QTAGUID NETWORK INTERFACES INFO (xt)", "/proc/net/xt_qtaguid/iface_stat_fmt");
    dump_file("QTAGUID CTRL INFO", "/proc/net/xt_qtaguid/ctrl");
    dump_file("QTAGUID STATS INFO", "/proc/net/xt_qtaguid/stats");

    //输出上次的kernel log
    if (!stat(PSTORE_LAST_KMSG, &st)) {
        //文件为/sys/fs/pstore/console-ramoops
        dump_file("LAST KMSG", PSTORE_LAST_KMSG);
    } else {
        //文件为/proc/last_kmsg
        dump_file("LAST KMSG", "/proc/last_kmsg");
    }

    //输出上次 logcat,内核必须设置CONFIG_PSTORE_PMSG
    run_command("LAST LOGCAT", 10, "logcat", "-L", "-v", "threadtime",
                                             "-b", "all", "-d", "*:v", NULL);

    //wifi驱动/固件 以及ip相关信息
    run_command("NETWORK INTERFACES", 10, "ip", "link", NULL);
    run_command("IPv4 ADDRESSES", 10, "ip", "-4", "addr", "show", NULL);
    run_command("IPv6 ADDRESSES", 10, "ip", "-6", "addr", "show", NULL);
    run_command("IP RULES", 10, "ip", "rule", "show", NULL);
    run_command("IP RULES v6", 10, "ip", "-6", "rule", "show", NULL);
    dump_route_tables();
    run_command("ARP CACHE", 10, "ip", "-4", "neigh", "show", NULL);
    run_command("IPv6 ND CACHE", 10, "ip", "-6", "neigh", "show", NULL);
    run_command("IPTABLES", 10, SU_PATH, "root", "iptables", "-L", "-nvx", NULL);
    run_command("IP6TABLES", 10, SU_PATH, "root", "ip6tables", "-L", "-nvx", NULL);
    run_command("IPTABLE NAT", 10, SU_PATH, "root", "iptables", "-t", "nat", "-L", "-nvx", NULL);
    run_command("IPTABLE RAW", 10, SU_PATH, "root", "iptables", "-t", "raw", "-L", "-nvx", NULL);
    run_command("IP6TABLE RAW", 10, SU_PATH, "root", "ip6tables", "-t", "raw", "-L", "-nvx", NULL);
    run_command("WIFI NETWORKS", 20, SU_PATH, "root", "wpa_cli", "IFNAME=wlan0", "list_networks", NULL);

    //中断向量表
    dump_file("INTERRUPTS (1)", "/proc/interrupts");
    run_command("NETWORK DIAGNOSTICS", 10, "dumpsys", "connectivity", "--diag", NULL);
    //中断向量表(二次输出)
    dump_file("INTERRUPTS (2)", "/proc/interrupts");

    //获取properties属性值
    print_properties();
    run_command("VOLD DUMP", 10, "vdc", "dump", NULL);
    run_command("SECURE CONTAINERS", 10, "vdc", "asec", "list", NULL);
    //可用空间
    run_command("FILESYSTEMS & FREE SPACE", 10, "df", NULL);
    run_command("LAST RADIO LOG", 10, "parse_radio_log", "/proc/last_radio_log", NULL);

    //背光信息
    printf("------ BACKLIGHTS ------\n");
    printf("LCD brightness="); dump_file(NULL, "/sys/class/leds/lcd-backlight/brightness");
    printf("Button brightness="); dump_file(NULL, "/sys/class/leds/button-backlight/brightness");
    printf("Keyboard brightness="); dump_file(NULL, "/sys/class/leds/keyboard-backlight/brightness");
    printf("ALS mode="); dump_file(NULL, "/sys/class/leds/lcd-backlight/als");
    printf("LCD driver registers:\n"); dump_file(NULL, "/sys/class/leds/lcd-backlight/registers");
    printf("\n");

    //Binder相关
    dump_file("BINDER FAILED TRANSACTION LOG", "/sys/kernel/debug/binder/failed_transaction_log");
    dump_file("BINDER TRANSACTION LOG", "/sys/kernel/debug/binder/transaction_log");
    dump_file("BINDER TRANSACTIONS", "/sys/kernel/debug/binder/transactions");
    dump_file("BINDER STATS", "/sys/kernel/debug/binder/stats");
    dump_file("BINDER STATE", "/sys/kernel/debug/binder/state");

    printf("========================================================\n");
    printf("== Board\n");
    printf("========================================================\n");
    dumpstate_board(); printf("\n");

    //输出framework各种服务的dumpsys信息
    printf("========================================================\n");
    printf("== Android Framework Services\n");
    printf("========================================================\n");
    run_command("DUMPSYS", 60, "dumpsys", NULL); //很耗时则timeout=60s

    printf("========================================================\n");
    printf("== Checkins\n");
    printf("========================================================\n");
    run_command("CHECKIN BATTERYSTATS", 30, "dumpsys", "batterystats", "-c", NULL);
    run_command("CHECKIN MEMINFO", 30, "dumpsys", "meminfo", "--checkin", NULL);
    run_command("CHECKIN NETSTATS", 30, "dumpsys", "netstats", "--checkin", NULL);
    run_command("CHECKIN PROCSTATS", 30, "dumpsys", "procstats", "-c", NULL);
    run_command("CHECKIN USAGESTATS", 30, "dumpsys", "usagestats", "-c", NULL);
    run_command("CHECKIN PACKAGE", 30, "dumpsys", "package", "--checkin", NULL);

    //输出当前 运行中activity/service/provider信息
    printf("========================================================\n");
    printf("== Running Application Activities\n");
    printf("========================================================\n");
    run_command("APP ACTIVITIES", 30, "dumpsys", "activity", "all", NULL);
    printf("========================================================\n");
    printf("== Running Application Services\n");
    printf("========================================================\n");
    run_command("APP SERVICES", 30, "dumpsys", "activity", "service", "all", NULL);
    printf("========================================================\n");
    printf("== Running Application Providers\n");
    printf("========================================================\n");
    run_command("APP SERVICES", 30, "dumpsys", "activity", "provider", "all", NULL);
    printf("========================================================\n");
    printf("== dumpstate: done\n");
    printf("========================================================\n");
}

该方法涉及run_command其他几个方法见下方:

2.3.1 run_command()

[-> utils.c]

int run_command(const char *title, int timeout_seconds, const char *command, ...) {
    fflush(stdout);
    uint64_t start = nanotime();
    //通过fork创建子进程
    pid_t pid = fork();
    if (pid < 0) {
        printf("*** fork: %s\n", strerror(errno));
        return pid;
    }

    //子进程执行
    if (pid == 0) {
        const char *args[1024] = {command};
        size_t arg;
        //确保dumpstate结束后能关闭子进程
        prctl(PR_SET_PDEATHSIG, SIGKILL);
        struct sigaction sigact;
        memset(&sigact, 0, sizeof(sigact));
        sigact.sa_handler = SIG_IGN;
        //忽略SIGPIPE
        sigaction(SIGPIPE, &sigact, NULL);
        va_list ap;
        va_start(ap, command);

        if (title) printf("------ %s (%s", title, command);
        for (arg = 1; arg < sizeof(args) / sizeof(args[0]); ++arg) {
            args[arg] = va_arg(ap, const char *);
            if (args[arg] == NULL) break;
            if (title) printf(" %s", args[arg]);
        }
        if (title) printf(") ------\n");
        fflush(stdout);
        //执行命令
        execvp(command, (char**) args);
        printf("*** exec(%s): %s\n", command, strerror(errno));
        fflush(stdout);
        _exit(-1); //进程退出
    }
    //父进程执行,主要处理子进程退出
    int status;
    bool ret = waitpid_with_timeout(pid, timeout_seconds, &status);
    uint64_t elapsed = nanotime() - start;
    if (!ret) {
        if (errno == ETIMEDOUT) {
            printf("*** %s: Timed out after %.3fs (killing pid %d)\n", command,
                   (float) elapsed / NANOS_PER_SEC, pid);
        } else {
            printf("*** %s: Error after %.4fs (killing pid %d)\n", command,
                   (float) elapsed / NANOS_PER_SEC, pid);
        }
        kill(pid, SIGTERM);
        if (!waitpid_with_timeout(pid, 5, NULL)) {
            kill(pid, SIGKILL);
            if (!waitpid_with_timeout(pid, 5, NULL)) {
                printf("*** %s: Cannot kill %d even with SIGKILL.\n", command, pid);
            }
        }
        return -1;
    }
    if (WIFSIGNALED(status)) {
        printf("*** %s: Killed by signal %d\n", command, WTERMSIG(status));
    } else if (WIFEXITED(status) && WEXITSTATUS(status) > 0) {
        printf("*** %s: Exit code %d\n", command, WEXITSTATUS(status));
    }
    if (title) printf("[%s: %.3fs elapsed]\n\n", command, (float)elapsed / NANOS_PER_SEC);
    return status;
}

功能是fork子进程并等待它执行完成,或者超时退出。当命令title不为空时,每次输出结果,都分别以下面作为开头和结尾:

------ <title> (<command>) ------
[<command>: <执行时长> elapsed]

2.3.2 dump_file()

[-> utils.c]

int dump_file(const char *title, const char *path) {
    //尝试打开文件
    int fd = TEMP_FAILURE_RETRY(open(path, O_RDONLY | O_NONBLOCK | O_CLOEXEC));
    if (fd < 0) {
        //无法打开文件时,则输出如下信息
        int err = errno;
        if (title) printf("------ %s (%s) ------\n", title, path);
        printf("*** %s: %s\n", path, strerror(err));
        if (title) printf("\n");
        return -1;
    }
    //输出文件内容
    return _dump_file_from_fd(title, path, fd);
}

当可以正确打开文件时,则执行_dump_file_from_fd,输出文件内容

static int _dump_file_from_fd(const char *title, const char *path, int fd) {
    if (title) printf("------ %s (%s", title, path);
    if (title) {
        struct stat st;
        //文件路径为/proc/或者/sys/
        if (memcmp(path, "/proc/", 6) && memcmp(path, "/sys/", 5) && !fstat(fd, &st)) {
            char stamp[80];
            time_t mtime = st.st_mtime; //文件上次修改时间
            strftime(stamp, sizeof(stamp), "%Y-%m-%d %H:%M:%S", localtime(&mtime));
            printf(": %s", stamp);
        }
        printf(") ------\n");
    }
    bool newline = false;
    fd_set read_set;
    struct timeval tm;
    while (1) {
        FD_ZERO(&read_set);
        FD_SET(fd, &read_set);
        //30s无数据可读则超时
        tm.tv_sec = 30;
        tm.tv_usec = 0;
        uint64_t elapsed = nanotime();
        int ret = TEMP_FAILURE_RETRY(select(fd + 1, &read_set, NULL, NULL, &tm));
        if (ret == -1) {
            printf("*** %s: select failed: %s\n", path, strerror(errno));
            newline = true;
            break;
        } else if (ret == 0) {
            elapsed = nanotime() - elapsed;
            printf("*** %s: Timed out after %.3fs\n", path,
                   (float) elapsed / NANOS_PER_SEC);
            newline = true;
            break;
        } else {
            char buffer[65536];
            // 读取数据
            ssize_t bytes_read = TEMP_FAILURE_RETRY(read(fd, buffer, sizeof(buffer)));
            if (bytes_read > 0) {
                fwrite(buffer, bytes_read, 1, stdout);
                newline = (buffer[bytes_read-1] == '\n');
            } else {
                if (bytes_read == -1) {
                    printf("*** %s: Failed to read from fd: %s", path, strerror(errno));
                    newline = true;
                }
                break;
            }
        }
    }
    close(fd);
    if (!newline) printf("\n");
    if (title) printf("\n");
    return 0;
}

当打不开文件或者出错则输出:

------ <title> (<path>) ------
*** <path>: <err>

当文件路径为/proc/或者/sys/,则输出时间/文件上次修改时间:

------ <title> (<path>: <文件修改时间>) ------

2.3.3 dump_files()

dump_files(“UPTIME MMC PERF”, mmcblk0, skip_not_stat, dump_stat_from_fd);

其中skip_not_stat是指忽略mmcblk0目录下的非stat文件,dump_files该方法遍历输出mmcblk0(即”/sys/block/mmcblk0/”)目录下所有stat文件,具体的输出调用dump_stat_from_fd方法来完成,该方法输出每个分区的读写速度:

static int dump_stat_from_fd(const char *title __unused, const char *path, int fd) {
    unsigned long fields[11], read_perf, write_perf;
    bool z;
    char *cp, *buffer = NULL;
    size_t i = 0;
    FILE *fp = fdopen(fd, "rb"); //打开文件
    getline(&buffer, &i, fp);
    fclose(fp);
    if (!buffer) {
        return -errno;
    }
    i = strlen(buffer);
    while ((i > 0) && (buffer[i - 1] == '\n')) {
        buffer[--i] = '\0';
    }
    if (!*buffer) {
        free(buffer);
        return 0;
    }
    z = true;
    for (cp = buffer, i = 0; i < (sizeof(fields) / sizeof(fields[0])); ++i) {
        fields[i] = strtol(cp, &cp, 0);
        if (fields[i] != 0) {
            z = false;
        }
    }
    if (z) { /* never accessed */
        free(buffer);
        return 0;
    }
    if (!strncmp(path, mmcblk0, sizeof(mmcblk0) - 1)) {
        path += sizeof(mmcblk0) - 1;
    }
    //例如输出/sys/block/mmcblk0/mmcblk0p13/stat内容
    printf("%s: %s\n", path, buffer);
    free(buffer);
    read_perf = 0;
    if (fields[3]) {
        //计算读的性能
        read_perf = 512 * fields[2] / fields[3];
    }
    write_perf = 0;
    if (fields[7]) {
        //计算写的性能
        write_perf = 512 * fields[6] / fields[7];
    }
    printf("%s: read: %luKB/s write: %luKB/s\n", path, read_perf, write_perf);
    //worst_write_perf默认值为20000kb/s
    if ((write_perf > 1) && (write_perf < worst_write_perf)) {
        worst_write_perf = write_perf;
    }
    return 0;
}

例如:stat文件共有11个数据:

mmcblk0p13/stat:  15  369  100  10  57  7239  5000  250  0  900  2610

则mmcblk0p13/stat的read_perf = 512* 100/10 = 5120KB/s, write_perf= 512* 5000/250 = 10240KB/s

2.3.4 dump_traces()

dump虚拟机和native的stack traces,并返回trace文件位置

const char *dump_traces() {
    const char* result = NULL;
    char traces_path[PROPERTY_VALUE_MAX] = "";

    //traces_path等于/data/anr/traces.txt
    property_get("dalvik.vm.stack-trace-file", traces_path, "");
    if (!traces_path[0]) return NULL;

    char anr_traces_path[PATH_MAX];
    strlcpy(anr_traces_path, traces_path, sizeof(anr_traces_path));
    strlcat(anr_traces_path, ".anr", sizeof(anr_traces_path));
    //文件重命名
    if (rename(traces_path, anr_traces_path) && errno != ENOENT) {
        fprintf(stderr, "rename(%s, %s): %s\n", traces_path, anr_traces_path, strerror(errno));
        return NULL; //没有权限重命令
    }

    char anr_traces_dir[PATH_MAX];
    strlcpy(anr_traces_dir, traces_path, sizeof(anr_traces_dir));
    char *slash = strrchr(anr_traces_dir, '/');
    if (slash != NULL) {
        *slash = '\0';
        //创建文件夹
        if (!mkdir(anr_traces_dir, 0775)) {
            chown(anr_traces_dir, AID_SYSTEM, AID_SYSTEM);
            chmod(anr_traces_dir, 0775);
            if (selinux_android_restorecon(anr_traces_dir, 0) == -1) {
                fprintf(stderr, "restorecon failed for %s: %s\n", anr_traces_dir, strerror(errno));
            }
        } else if (errno != EEXIST) {
            fprintf(stderr, "mkdir(%s): %s\n", anr_traces_dir, strerror(errno));
            return NULL;
        }
    }

    //创建一个新的空文件traces.txt
    int fd = TEMP_FAILURE_RETRY(open(traces_path, O_CREAT | O_WRONLY | O_TRUNC | O_NOFOLLOW | O_CLOEXEC,
                                     0666));  /* -rw-rw-rw- */
    if (fd < 0) {
        fprintf(stderr, "%s: %s\n", traces_path, strerror(errno));
        return NULL;
    }
    int chmod_ret = fchmod(fd, 0666);
    if (chmod_ret < 0) {
        fprintf(stderr, "fchmod on %s failed: %s\n", traces_path, strerror(errno));
        close(fd);
        return NULL;
    }

    // * walk /proc and kill -QUIT all Dalvik processes */
    DIR *proc = opendir("/proc");
    if (proc == NULL) {
        fprintf(stderr, "/proc: %s\n", strerror(errno));
        goto error_close_fd;
    }

    //当进程完成dump操作时,通过inotify来通知
    int ifd = inotify_init();
    if (ifd < 0) {
        fprintf(stderr, "inotify_init: %s\n", strerror(errno));
        goto error_close_fd;
    }

    int wfd = inotify_add_watch(ifd, traces_path, IN_CLOSE_WRITE);
    if (wfd < 0) {
        fprintf(stderr, "inotify_add_watch(%s): %s\n", traces_path, strerror(errno));
        goto error_close_ifd;
    }

    struct dirent *d;
    int dalvik_found = 0;
    while ((d = readdir(proc))) {
        int pid = atoi(d->d_name);
        if (pid <= 0) continue;

        char path[PATH_MAX];
        char data[PATH_MAX];
        snprintf(path, sizeof(path), "/proc/%d/exe", pid);
        ssize_t len = readlink(path, data, sizeof(data) - 1);
        if (len <= 0) {
            continue;
        }
        data[len] = '\0';

        if (!strncmp(data, "/system/bin/app_process", strlen("/system/bin/app_process"))) {
            snprintf(path, sizeof(path), "/proc/%d/cmdline", pid);
            int cfd = TEMP_FAILURE_RETRY(open(path, O_RDONLY | O_CLOEXEC));
            len = read(cfd, data, sizeof(data) - 1);
            close(cfd);
            if (len <= 0) {
                continue;
            }
            data[len] = '\0';
            //略过zygote,并不输出它的栈信息
            if (!strncmp(data, "zygote", strlen("zygote"))) {
                continue;
            }

            ++dalvik_found;
            uint64_t start = nanotime();
            if (kill(pid, SIGQUIT)) {
                fprintf(stderr, "kill(%d, SIGQUIT): %s\n", pid, strerror(errno));
                continue;
            }

            /* wait for the writable-close notification from inotify */
            struct pollfd pfd = { ifd, POLLIN, 0 };
            int ret = poll(&pfd, 1, 5000);  /* 5s超时*/
            if (ret < 0) {
                fprintf(stderr, "poll: %s\n", strerror(errno));
            } else if (ret == 0) {
                fprintf(stderr, "warning: timed out dumping pid %d\n", pid);
            } else {
                struct inotify_event ie;
                read(ifd, &ie, sizeof(ie));
            }

            if (lseek(fd, 0, SEEK_END) < 0) {
                fprintf(stderr, "lseek: %s\n", strerror(errno));
            } else {
                dprintf(fd, "[dump dalvik stack %d: %.3fs elapsed]\n",
                        pid, (float)(nanotime() - start) / NANOS_PER_SEC);
            }
        } else if (should_dump_native_traces(data)) {
            //native进程trace
            if (lseek(fd, 0, SEEK_END) < 0) {
                fprintf(stderr, "lseek: %s\n", strerror(errno));
            } else {
                static uint16_t timeout_failures = 0;
                uint64_t start = nanotime();

                /* If 3 backtrace dumps fail in a row, consider debuggerd dead. */
                if (timeout_failures == 3) {
                    dprintf(fd, "too many stack dump failures, skipping...\n");
                // 超时时长为20s
                } else if (dump_backtrace_to_file_timeout(pid, fd, 20) == -1) {
                    dprintf(fd, "dumping failed, likely due to a timeout\n");
                    timeout_failures++;
                } else {
                    timeout_failures = 0;
                }
                dprintf(fd, "[dump native stack %d: %.3fs elapsed]\n",
                        pid, (float)(nanotime() - start) / NANOS_PER_SEC);
            }
        }
    }

    if (dalvik_found == 0) {
        fprintf(stderr, "Warning: no Dalvik processes found to dump stacks\n");
    }

    static char dump_traces_path[PATH_MAX];
    strlcpy(dump_traces_path, traces_path, sizeof(dump_traces_path));
    strlcat(dump_traces_path, ".bugreport", sizeof(dump_traces_path));
    if (rename(traces_path, dump_traces_path)) {
        fprintf(stderr, "rename(%s, %s): %s\n", traces_path, dump_traces_path, strerror(errno));
        goto error_close_ifd;
    }
    result = dump_traces_path;

    /* replace the saved [ANR] traces.txt file */
    rename(anr_traces_path, traces_path);

error_close_ifd:
    close(ifd);
error_close_fd:
    close(fd);
    return result;
}

该方法其中两个重要的步骤:

  • 输出Java进程的trace是通过发送signal 3来dump相应信息。
  • 输出native进程的trace是通过dump_backtrace_to_file_timeout,并且超时时长为20s;

2.3.5 do_dmesg()

void do_dmesg() {
    printf("------ KERNEL LOG (dmesg) ------\n");
    //获取kernel buffer的大小
    int size = klogctl(KLOG_SIZE_BUFFER, NULL, 0);
    if (size <= 0) {
        printf("Unexpected klogctl return value: %d\n\n", size);
        return;
    }
    char *buf = (char *) malloc(size + 1);
    if (buf == NULL) {
        printf("memory allocation failed\n\n");
        return;
    }
    //获取kernel log
    int retval = klogctl(KLOG_READ_ALL, buf, size);
    if (retval < 0) {
        printf("klogctl failure\n\n");
        free(buf);
        return;
    }
    buf[retval] = '\0';
    printf("%s\n\n", buf);
    free(buf);
    return;
}

2.4 总结

bugreport通过socket与dumpstate服务建立通信,在dumpstate.cpp中的dumpstate()方法完成核心功能,该功能依次输出内容项, 主要分为5大类:

  1. current log: kernel,system, event, radio;
  2. last log: kernel, system, radio;
  3. vm traces: just now, last ANR, tombstones
  4. dumpsys: all, checkin, app
  5. system info:cpu, memory, io等

从bugreport内容的输出顺序的角度,再详细列举其内容:

  1. 系统build以及运行时长等相关信息;
  2. 内存/CPU/进程等信息;
  3. kernel log
  4. lsof、map及Wait-Channels;
  5. system log
  6. event log
  7. radio log;
  8. vm traces
    1. VM TRACES JUST NOW (/data/anr/traces.txt.bugreport) (抓bugreport时主动触发)
    2. VM TRACES AT LAST ANR (/data/anr/traces.txt) (存在则输出)
    3. TOMBSTONE (/data/tombstones/tombstone_xx) (存在这输出)
  9. network相关信息;
  10. last kernel log;
  11. last system log;
  12. ip相关信息;
  13. 中断向量表
  14. property以及fs等信息
  15. last radio log;
  16. Binder相关信息;
  17. dumpsys all:
  18. dumpsys checkin相关:
    • dumpsys batterystats电池统计;
    • dumpsys meminfo内存
    • dumpsys netstats网络统计;
    • dumpsys procstats进程统计;
    • dumpsys usagestats使用情况;
    • dumpsys package.
  19. dumpsys app相关
    • dumpsys activity;
    • dumpsys activity service all;
    • dumpsys activity provider all.

Tips: bugreport几乎涵盖整个系统信息,内容非常长,每一个子项都以------ xxx ------开头。 例如APP ACTIVITIES的开头便是 ------ APP ACTIVITIES (dumpsys activity all) ------,其中括号内的便是输出该信息指令,即dumpsys activity all,还有可能是内容所在节点,各个子项目类似的规律,看完前面的源码分析过程,相信你肯定能明白。下面一篇文章再进一步从bugreport内容的角度来说明其寓意。


欢迎关注我的微博: weibo.com/gityuan !请随意 ¥打赏支持 将激励创作更多技术干货!