当前位置:   article > 正文

compile error怎么解决_一个看似是系统问题的应用问题的解决过程

compile error: libcompiler: error: unable to execute command: executable

作者:王小二C  2019/09/04

前言

今天遇到一个问题,应用工程师分析是系统层的问题,然后就把这个锅给了我。最后我又把锅甩回给了应用工程师。

异常log如下:

I [2019-08-18 10:11:08 GMT+8] binder: 1433:1561 transaction failed 29201/-28, size 828-8 line 3135W BroadcastQueue: Can't deliver broadcast to com.xxxx.xxxxx (pid 4712). Crashing it.W BroadcastQueue: Failure sending broadcast Intent { act=android.intent.action.BATTERY_CHANGED flg=0x60000010 (has extras) }W BroadcastQueue: android.os.DeadObjectException: Transaction failed on small parcel; remote process probably diedW BroadcastQueue:     at android.os.BinderProxy.transactNative(Native Method)W BroadcastQueue:     at android.os.BinderProxy.transact(Binder.java:1127)W BroadcastQueue:     at android.app.IApplicationThread$Stub$Proxy.scheduleRegisteredReceiver(IApplicationThread.java:1237)W BroadcastQueue:     at com.android.server.am.BroadcastQueue.performReceiveLocked(BroadcastQueue.java:496)W BroadcastQueue:     at com.android.server.am.BroadcastQueue.deliverToRegisteredReceiverLocked(BroadcastQueue.java:715)W BroadcastQueue:     at com.android.server.am.BroadcastQueue.processNextBroadcastLocked(BroadcastQueue.java:875)W BroadcastQueue:     at com.android.server.am.BroadcastQueue.processNextBroadcast(BroadcastQueue.java:834)W BroadcastQueue:     at com.android.server.am.BroadcastQueue$BroadcastHandler.handleMessage(BroadcastQueue.java:172)W BroadcastQueue:     at android.os.Handler.dispatchMessage(Handler.java:106)W BroadcastQueue:     at android.os.Looper.loop(Looper.java:193)W BroadcastQueue:     at android.os.HandlerThread.run(HandlerThread.java:65)W BroadcastQueue:     at com.android.server.ServiceThread.run(ServiceThread.java:44)

初步分析

从log来看的确好像是binder驱动第3135行出现问题(line 3135)

binder: 1433:1561 transaction failed 29201/-28, size 828-8 line 3135

一看Binder.c的代码发现对不上3135行,估计是编译版本不一样,因为测试是user版本,代码优化了,这怎么办?我需要在userdebug的版本上复现此问题。

Binder.c3129    if (target_node && target_node->txn_security_ctx) {3130              u32 secid;3131           size_t added_size;31323133            security_task_getsecid(proc->tsk, &secid);3134         ret = security_secid_to_secctx(secid, &secctx, &secctx_sz);3135           if (ret) {3136                  return_error = BR_FAILED_REPLY;3137                    return_error_param = ret;3138                    return_error_line = __LINE__;3139                        goto err_get_secctx_failed;3140            }

聪明机智的我瞬间判断应该是对应到3164行,应该是binder server无法申请足够的buffer,别问我怎么想到的,有时候解决问题就得靠猜。

3154    t->buffer = binder_alloc_new_buf(&target_proc->alloc, tr->data_size,3155           tr->offsets_size, extra_buffers_size,3156             !reply && (t->flags & TF_ONE_WAY));3157       if (IS_ERR(t->buffer)) {3158                /*3159            * -ESRCH indicates VMA cleared. The target is dying.3160              */3161         return_error_param = PTR_ERR(t->buffer);3162          return_error = return_error_param == -ESRCH ?3163                    BR_DEAD_REPLY : BR_FAILED_REPLY;3164             return_error_line = __LINE__;3165         t->buffer = NULL;3166         goto err_binder_alloc_buf_failed;3167       }

为了证明我判断是对的,我写了如下的一个demo安装到userdebug来制造binder server无法申请足够的buffer的情况

public class MainActivity extends Activity implements View.OnClickListener {    private MyLinear mRoot;    private IMyAidlInterface myAidlInterface;    @Override    protected void onCreate(Bundle savedInstanceState) {        super.onCreate(savedInstanceState);        setContentView(R.layout.activity_main);        mRoot = (MyLinear) findViewById(R.id.root);        mRoot.setOnClickListener(this);        Intent intent = new Intent(this, MyService.class);        bindService(intent, new ServiceConnection() {            @Override            public void onServiceConnected(ComponentName name, IBinder service) {                myAidlInterface = IMyAidlInterface.Stub.asInterface(service);            }            @Override            public void onServiceDisconnected(ComponentName name) {            }        }, Context.BIND_AUTO_CREATE);    }    @Override    public void onClick(View v) {        try {            while (true) {                myAidlInterface.send("dfdafsdfafdasdfadfadsfafd");            }        } catch (Exception e) {        }    }}public class MyService extends Service {    @Override    public IBinder onBind(Intent intent) {        return new MyBinder();    }    public class MyBinder extends IMyAidlInterface.Stub {        @Override        public void send(String e) throws RemoteException {            //后面发现这个4秒的卡顿,不加也可以复现此问题            try {                Thread.sleep(4000);            } catch (Exception ee) {            }        }    }}// IMyAidlInterface.aidlpackage com.tct.activitydemo;// Declare any non-default types here with import statementsinterface IMyAidlInterface {    /**     * Demonstrates some basic types that you can use as parameters     * and return values in AIDL.     */    oneway void send(String e);}

出现的异常log,和之前log相比,除了行数不对,error code是一样的都是29201/-28,而且行数果然是3164行,所以我的推测是对的。

//binder: 1433:1561 transaction failed 29201/-28, size 828-8 line 3135binder: 30286:30286 transaction failed 29201/-28, size 140-0 line 3164

#初步分析结论 广播的发送失败是原因,在一次binder通信中,无法向广播注册的App的binder驱动中映射的共享内存申请足够buffer。

#重大发现 我发现出问题的应用注册了300多个广播,都是监听android.intent.action.BATTERY_CHANGED,具体log就不贴了,在bugreport中会有当前系统所有广播的dumpsys的信息。我贴出meminfo的信息,发现三百多个activity没有被GC,因为那个广播是在activity中被动态注册的,所以变相可以证明注册了300多个广播

** MEMINFO in pid 4712 [com.xxxxxxxxxxx] **.....省略没用的信息,看下面activity的数量是325  Objects               Views:    13600         ViewRootImpl:        2         AppContexts:      329           Activities:      325              Assets:        3        AssetManagers:        0       Local Binders:      352        Proxy Binders:      366       Parcel memory:      887         Parcel count:     2235    Death Recipients:        2      OpenSSL Sockets:        0            WebViews:        0

推测

当这个广播发送的时候,由于他的接受者有300多个,每一次接收都会在申请一次buffer,如果短时间一下子申请,非常有可能超过binder驱动的(1mb-8kb)/2的限制,有人会问为什么是(1mb-8kb)/2而不是1mb-8kb,因为scheduleRegisteredReceiver是oneway的,对这个有疑问的,可以看一下我的另外一个文章:[[007]一次Binder通信最大可以传输多大的数据?]

进一步分析

其实一般分析到这里,对于我来说已经可以把锅甩回给了应用层,但是这个问题的好奇心促使我继续分析下去,一定要找到广播重复注册的原因。

涉及保密,我把应用层代码的精简成自己的代码

public class MyActivity extends Activity {    private MyReceiver receiver;    @Override    protected void onCreate(Bundle savedInstanceState) {        super.onCreate(savedInstanceState);        setContentView(R.layout.activity_main2);        IntentFilter intentFilter = new IntentFilter();        intentFilter.addAction("fdafdafsdafaffasdfad");        receiver = new MyReceiver(this);        getApplication().registerReceiver(receiver, intentFilter);    }    @Override    protected void onDestroy() {        try {            //因为上下文不同,会导致unregisterReceiver失败,从而导致MyActivity和MyReceiver,无法被GC            unregisterReceiver(receiver);        } catch (Exception e) {        }        super.onDestroy();    }    public static class MyReceiver extends BroadcastReceiver {        private Context mContext;        public MyReceiver(Context context) {            mContext = context;        }        @Override        public void onReceive(Context context, Intent intent) {        }    }}

水落石出

原来应用开发工程师,在registerReceiver和unregisterReceiver使用了不同的context,导致了unregisterReceiver的失败,从而导致MyReceiver的无法被释放,而且这个代码还会导致MyActivity的内存泄露。

举个例子来还原一下这个现场

小明寄了一份投诉信到信访局门口的信箱,然后信访局的1号工作人员拿了这份投诉信,进行处理。这个就是一次完整的oneway的binder通信。但是有一天小明一下子拿了300封投诉信,一封封的塞到信箱里,然后信访局的1~16号工作人员同时拿了16封信进行处理,但是还是架不住小明的塞信的速度,很快信箱就爆了,小明说了一句:垃圾信访局,我信还没有塞完呢。

其实大家可能有点难以理解一次binder通信,内核发生了什么,请看下面图,看懂下图,才能知道我在说什么:

b0bf19f465a909ee4f9c49d8b6f5f850.png

References

[1] [007]一次Binder通信最大可以传输多大的数据? 简书地址:https://www.jianshu.com/p/ea4fc6aefaa8

ac2530edefa4430c2f6524d832fd5390.png

声明:本文内容由网友自发贡献,不代表【wpsshop博客】立场,版权归原作者所有,本站不承担相应法律责任。如您发现有侵权的内容,请联系我们。转载请注明出处:https://www.wpsshop.cn/w/不正经/article/detail/224398
推荐阅读
相关标签
  

闽ICP备14008679号