pthread clean up
https://www.ibm.com/developerworks/cn/linux/thread/posix_threadapi/part4/
http://www.cnblogs.com/xfiver/archive/2013/01/23/2873725.html
线程终止方式
一般来说,Posix的线程终止有两种情况:正常终止和非正常终止。线程主动调用pthread_exit()或者从线程函数中return都将使线程正常退出,这是可预见的退出方式;非正常终止是线程在其他线程的干预下,或者由于自身运行出错(比如访问非法地址)而退出,这种退出方式是不可预见的。
线程终止时的清理
不论是可预见的线程终止还是异常终止,都会存在资源释放的问题,在不考虑因运行出错而退出的前提下,如何保证线程终止时能顺利的释放掉自己所占用的资源,特别是锁资源,就是一个必须考虑解决的问题。
最经常出现的情形是资源独占锁的使用:线程为了访问临界资源而为其加上锁,但在访问过程中被外界取消,如果线程处于响应取消状态,且采用异步方式响应,或者在打开独占锁以前的运行路径上存在取消点,则该临界资源将永远处于锁定状态得不到释放。外界取消操作是不可预见的,因此的确需要一个机制来简化用于资源释放的编程。
在POSIX线程API中提供了一个pthread_cleanup_push()/pthread_cleanup_pop()函数对用于自动释放资源--从pthread_cleanup_push()的调用点到pthread_cleanup_pop()之间的程序段中的终止动作(包括调用pthread_exit()和取消点终止)都将执行pthread_cleanup_push()所指定的清理函数。API定义如下:
void pthread_cleanup_push(void (*routine) (void *), void *arg) void pthread_cleanup_pop(int execute)
pthread_cleanup_push()/pthread_cleanup_pop()采用先入后出的栈结构管理,void routine(void *arg)函数在调用pthread_cleanup_push()时压入清理函数栈,多次对pthread_cleanup_push()的调用将在清理函数栈中形成一个函数链,在执行该函数链时按照压栈的相反顺序弹出。
execute参数表示执行到pthread_cleanup_pop()时是否在弹出清理函数的同时执行该函数,为0表示不执行,非0为执行;这个参数并不影响异常终止时清理函数的执行。
补充说明: 1. 线程创建 pthread_create()函数返回值0,表示创建成功,线程id保存载tidp中;失败则返回非零,需自行处理,不会修改errno值 2. 线程终止 a. 任一线程调用exit, _Exit, _exit都将导致整个进程终止; b. 单个线程退出方式有三种: 1> 线程执行函数start_rtn()中使用return返回,返回值为线程退出码; 2> 被同一个进程的其他线程使用pthread_cancel()取消; 3> 线程自身调用了pthread_exit(); 说明:pthread_join(pthread_t tid, void **rval_ptr)函数会阻塞调用线程,直到tid线程通过上述三种方式终止退出,且return/pthread_exit()方式会设置相应线程退出码rval_ptr,而pthread_cancel()取消的线程,将退出码设置为PTHREAD_CANCELED. 3. 线程清理处理程序(thread cleanup handler) 3.a> pthread_cleanup_push()与pthread_cleanup_pop()均为<pthread.h>中实现的宏定义,具体实现如下: ? pthread_cleanup_push and pthread_cleanup_pop are macros and must always be used in matching pairs at the same nesting level of braces. */ # define pthread_cleanup_push(routine, arg) \ do { \ __pthread_cleanup_class __clframe (routine, arg) /* Remove a cleanup handler installed by the matching pthread_cleanup_push. If EXECUTE is non-zero, the handler function is called. */ # define pthread_cleanup_pop(execute) \ __clframe.__setdoit (execute); \ } while (0) 可见push/pop中的{/}是一一对应的,因此pthread_cleanup_push/pop()也应一一对应出现,否则编译出错。 3.b> 当线程执行下列之一操作时调用清理函数,thread_cleanup_push由栈结构实现,注意清理程序调用的顺序,先入后出。 1: 调用pthread_exit()时,而直接return不会出发清理函数; 2: 相应取消请求pthread_cancel()时; 3: 使用非零execute参数调用pthread_cleanup_pop()时; 尤其需注意pthread_cleanup_pop()参数不同及此语句所处位置不同而有不同效果。 看此代码实例,注意return或pthread_exit()位置不同导致pthread_cleanup_pop()不同参数的效果变化。
include <pthread.h> void testPointerSize() { void *tret; printf("size of pointer in x86-64:%d\n",sizeof(tret)); //result is 8 in x86-64. //which is 4 in x86-32. printf("size of int in x86-64:%d\n",sizeof(int)); //result is 4 in x86-64. //which is also 4 in x86-32. } void cleanup(void *arg) { printf("cleanup:%s\n",(char *)arg); } void * thr_fn1(void *arg) { printf("thread 1 start\n"); pthread_cleanup_push(cleanup, "thread 1 first handler"); pthread_cleanup_push(cleanup, "thread 1 second handler"); if(arg) return ((void *)1);//arg !=0 ,return here. // return here will not triger any cleanup. pthread_cleanup_pop(0); pthread_cleanup_pop(1); return ((void *)2);//will not run this } void * thr_fn2(void *arg) { printf("thread 2 start\n"); pthread_cleanup_push(cleanup, "thread 2 first handler"); pthread_cleanup_push(cleanup, "thread 2 second handler"); pthread_cleanup_pop(0); pthread_cleanup_pop(1); return ((void *)2); // return here can triger cleanup second handler; } void * thr_fn3(void *arg) { printf("thread 3 start\n"); pthread_cleanup_push(cleanup, "thread 3 first handler"); pthread_cleanup_push(cleanup, "thread 3 second handler"); if(arg) pthread_exit((void *)3); //pthread_exit() here will triger both cleanup first&second handler. pthread_cleanup_pop(1); pthread_cleanup_pop(0); pthread_exit((void *)3);//wont run this } void * thr_fn4(void *arg) { printf("thread 4 start\n"); pthread_cleanup_push(cleanup, "thread 4 first handler"); pthread_cleanup_push(cleanup, "thread 4 second handler"); pthread_cleanup_pop(1); pthread_cleanup_pop(0); pthread_exit((void *)4); //pthread_exit() here will triger cleanup second handler. } int main(void) { testPointerSize(); int err; pthread_t tid1, tid2, tid3, tid4; void *tret; err = pthread_create(&tid1, NULL, thr_fn1, (void *)1); err = pthread_join(tid1,&tret); printf("thread 1 exit code %d\n",(int)tret); err = pthread_create(&tid2, NULL, thr_fn2, (void *)2); err = pthread_join(tid2, &tret); printf("thread 2 exit code %d\n",(int)tret); err = pthread_create(&tid3, NULL, thr_fn3, (void *)3); err = pthread_join(tid3,&tret); printf("thread 3 exit code %d\n",(int)tret); err = pthread_create(&tid4, NULL, thr_fn4, (void *)4); err = pthread_join(tid4, &tret); printf("thread 4 exit code %d\n",(int)tret); }
运行结果: ? [root@hello testData]# ./test size of pointer in x86-64:8 size of int in x86-64:4 thread 1 start thread 1 exit code 1 thread 2 start cleanup:thread 2 first handler thread 2 exit code 2 thread 3 start cleanup:thread 3 second handler cleanup:thread 3 first handler thread 3 exit code 3 thread 4 start cleanup:thread 4 second handler thread 4 exit code 4
由上述测试程序总结如下:
1> push与pop间的return,将导致清理程序不被触发;
2> 位于pop之后return,由pop的参数确定是否触发清理程序,非零参数触发,零参数不触发;
3> push/pop间的pthread_exit(),将触发所有清理函数;
4>位于pop之后的pthread_exit()时,pop参数决定是否触发清理程序;
其实,上述四种情况只是测试验证了前文3.b所说三个条件,加深理解。
参考文献:
2. <UNIX环境高级编程(第2版)> P295-296程序
3. pthread_cleanup_push()/pthread_cleanup_pop()的详解
4. Linux中vim的列编辑实例 (Mark记录)
pthread_cleanup_push()/pthread_cleanup_pop()是以宏方式实现的,这是pthread.h中的宏定义:
#define pthread_cleanup_push(routine,arg) \ { struct _pthread_cleanup_buffer _buffer; \ _pthread_cleanup_push (&_buffer, (routine), (arg)); #define pthread_cleanup_pop(execute) \ _pthread_cleanup_pop (&_buffer, (execute)); }
可见,pthread_cleanup_push()带有一个"{",而pthread_cleanup_pop()带有一个"}",因此这两个函数必须成对出现,且必须位于程序的同一级别的代码段中才能通过编译。在下面的例子里,当线程在"do some work"中终止时,将主动调用pthread_mutex_unlock(mut),以完成解锁动作。
pthread_cleanup_push(pthread_mutex_unlock, (void *) &mut); pthread_mutex_lock(&mut); /* do some work */ pthread_mutex_unlock(&mut); pthread_cleanup_pop(0);
必须要注意的是,如果线程处于PTHREAD_CANCEL_ASYNCHRONOUS状态,上述代码段就有可能出错,因为CANCEL事件有可能在pthread_cleanup_push()和pthread_mutex_lock()之间发生,或者在pthread_mutex_unlock()和pthread_cleanup_pop()之间发生,从而导致清理函数unlock一个并没有加锁的mutex变量,造成错误。因此,在使用清理函数的时候,都应该暂时设置成PTHREAD_CANCEL_DEFERRED模式。为此,POSIX的Linux实现中还提供了一对不保证可移植的pthread_cleanup_push_defer_np()/pthread_cleanup_pop_defer_np()扩展函数,功能与以下代码段相当:
{ int oldtype; pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, &oldtype); pthread_cleanup_push(routine, arg); ... pthread_cleanup_pop(execute); pthread_setcanceltype(oldtype, NULL); }