关于sem_unlink什么时候删除信号量

sem_unlink在man手册里有这么一段话：

sem_unlink() removes the named semaphore referred to by name.  
The semaphore name is removed immediately.  The semaphore  is
destroyed  once all other processes that have the semaphore open close it.

这段话的意思是说，sem_unlink会马上删除指定的信号量名，但要等到所有打开该信号量的进程关闭该信号量后才删除该信号。详细地说，当进程创建一个有名信号量，会在/dev/shm下生成一个sem.xxx的文件，所有打开该信号量的进程（包括创建它的进程）都会增加该文件的引用计数，并且这个计数由内核管理。当调用sem_unlink时，/dev/shm下的sem.xxx文件会马上被删除，但是信号量本身并没有被删除，所有已打开该信号量的进程仍能正常使用它。直到所有打开该信号量的进程关闭该信号量后，内核才真正删除信号量。

　　因此，信号量跟sem.xxx并不是共存亡的。下面来测试下：

/* sem_create.cpp */

#include <iostream>#include <fcntl.h>           /* For O_* constants */#include <sys/stat.h>        /* For mode constants */#include <semaphore.h>#include <unistd.h>#include <cerrno>#include <cstring>int main(){    sem_t *psem = sem_open( "/sem_test",O_CREAT,S_IRWXU,1 );    if ( SEM_FAILED == psem )    {        std::cout << "sem_open fail:" << strerror(errno) << std::endl;        return 0;    }        int sval = 0;    sem_getvalue( psem,&sval );        std::cout << "sleep now ....value is " << sval << std::endl;    sleep( 30 );        sem_getvalue( psem,&sval );    std::cout << "wake up ....value is " << sval << std::endl;        sem_post( psem );    sem_getvalue( psem,&sval );    std::cout << "raise value and try again ....value is " << sval << std::endl;        sem_close( psem );}

/** * * sem_destruct.cpp */#include <iostream>#include <fcntl.h>           /* For O_* constants */#include <sys/stat.h>        /* For mode constants */#include <semaphore.h>#include <cerrno>#include <cstring>int main(){    sem_t *psem = sem_open( "/sem_test",O_CREAT,S_IRWXU,1 );    if ( SEM_FAILED == psem )    {        std::cout << "sem_open fail:" << strerror(errno) << std::endl;        return 0;    }        std::cout << "decrease value and remove sem now..." << std::endl;    sem_wait ( psem );        sem_close( psem );    sem_unlink( "/sem_test" );        std::cout << "sem should remove from /dev/shm,but not really remove" << std::endl;}

分别编译上面的文件：

g++ -pthread -o sem_create sem_create.cppg++ -pthread -o sem_destruct sem_destruct.cpp

先运行sem_create，再运行sem_destruct这时个可以看到输出为：

xzc@xzc-HP-ProBook-4446s:~/code/sem$ ./sem_createsleep now ....value is 1xzc@xzc-HP-ProBook-4446s:~/code/sem$ ./sem_destructdecrease value and remove sem now...sem should remove from /dev/shm,but not really removewake up ....value is 0raise value and try again ....value is 1

可以看到，即使sem_destruct从/dev/shm把信号量删除了，ls /dev/shm也确认对应的文件删除了。但是经过sleep的sem_create进程在wake up后仍能对信号量进行操作。

　　可以看到，有名信号量一旦打开后，跟名字并没有多大关系。下面再来测试有名信号量调用sem_unlink但仍有进程在使用，这时另外一个进程又要创建同名信号量的情况

/**  *  * sem_again.cpp  */#include <iostream>#include <fcntl.h>           /* For O_* constants */#include <sys/stat.h>        /* For mode constants */#include <semaphore.h>#include <unistd.h>#include <cerrno>#include <cstring>int main(){    sem_t *psem = sem_open( "/sem_test",O_CREAT|O_EXCL,S_IRWXU,1 );    if ( SEM_FAILED == psem )    {        std::cout << "sem_open fail:" << strerror(errno) << std::endl;        return 0;    }        int sval = 0;    sem_getvalue( psem,&sval );        //std::cout << "sleep now ....value is " << sval << std::endl;    //sleep( 30 );        sem_getvalue( psem,&sval );    std::cout << "value is " << sval << std::endl;        sem_post( psem );    sem_getvalue( psem,&sval );    std::cout << "raise value and try again ....value is " << sval << std::endl;        sem_close( psem );    sem_unlink( "/sem_test" );}

编译一下：g++ -pthread -o sem_again sem_again.cpp，然后依次运行sem_create、sem_destruct、sem_again

xzc@xzc-HP-ProBook-4446s:~/code/sem$ ./sem_createsleep now ....value is 1xzc@xzc-HP-ProBook-4446s:~/code/sem$ ./sem_destructdecrease value and remove sem now...sem should remove from /dev/shm,but not really removexzc@xzc-HP-ProBook-4446s:~/code/sem$ ./sem_againvalue is 1raise value and try again ....value is 2wake up ....value is 0raise value and try again ....value is 1

在sem_again中特意使用了O_CREATE|O_EXCL标志，如果已存在同名信号量，则创建失败。但从上面的结果来看，创建成功并且互不干扰。也就是说，一个有名信号量，在打开并调用sem_unlink后可以当作无名信号量来使用(但这时新的进程还想使用该信号量，不可能了)。

　　注：上面只做了简单测试，并未测试sem_unlink后各进程的竞争情况，在生产环境请自行测试。

PS：在debian 6上，不用unistd.h也能调用sleep函数，在ubuntu 14.04上不行。在debian 6上是链接的库是-lrt，在ubuntu 14.04上是-pthread，-lrt不行。linux这是要闹哪样。

转自：http://www.cnblogs.com/coding-my-life/p/4215862.html?utm_source=tuicool