理解dup、dup和fcntl

/dup函数的作用：复制一个现有的句柄，产生一个与“源句柄特性”完全一样的新句柄（也即生成一个新的句柄号，并关联到同一个设备）
//dup2函数的作用：复制一个现有的句柄到另一个句柄上，目标句柄的特性与“源句柄特性”完全一样（也即首先关闭目标句柄，与设备断连，接着从源句柄完全拷贝复制到目标句柄）

//dup和dup2都是系统服务，window平台对应DuplicateHandle函数

/* DUP.C: This program uses the variable old to save
 * the original stdout. It then opens a new file named
 * new and forces stdout to refer to it. Finally, it
 * restores stdout to its original state.
 */

＃i nclude <io.h>
＃i nclude <stdlib.h>
＃i nclude <stdio.h>

void main( void )
{
   int old;
   FILE *new;

   old = _dup( 1 );   /* "old" now refers to "stdout" */
                      /* Note:  file handle 1 == "stdout" */
   if( old == -1 )
   {
      perror( "_dup( 1 ) failure" );
      exit( 1 );
   }
   write( old, "This goes to stdout first/r/n", 27 );
   if( ( new = fopen( "data", "w" ) ) == NULL )
   {
      puts( "Can't open file 'data'/n" );
      exit( 1 );
   }

   /* stdout now refers to file "data" */
   if( -1 == _dup2( _fileno( new ), 1 ) )
   {
      perror( "Can't _dup2 stdout" );
      exit( 1 );
   }
   puts( "This goes to file 'data'/r/n" );

   /* Flush stdout stream buffer so it goes to correct file */
   fflush( stdout );
   fclose( new );

   /* Restore original stdout */
   _dup2( old, 1 );
   puts( "This goes to stdout/n" );
   puts( "The file 'data' contains:" );
   system( "type data" );
}

Output

This goes to stdout first
This goes to file 'data'

This goes to stdout

The file 'data' contains:

This goes to file 'data'

关于fcntl(fd, F_SETFD, FD_CLOEXEC)设置exec时close的属性

snd_ctl_hw_open
#define SNDRV_FILE_CONTROL    ALSA_DEVICE_DIRECTORY "controlC%i"
sprintf(filename, SNDRV_FILE_CONTROL, card); // 路径/dev/snd/controlC0
fd = snd_open_device(filename, fmode);
fcntl(fd, F_SETFD, FD_CLOEXEC); // 这里设置为FD_CLOEXEC表示当程序执行exec函数时本fd将被系统自动关闭,表示不传递给exec创建的新进程, 如果设置为fcntl(fd, F_SETFD, 0);那么本fd将保持打开状态复制到exec创建的新进程中[luther.gliethttp].
进入内核系统调用
sys_fcntl
do_fcntl
    case F_SETFD:
        err = 0;
        set_close_on_exec(fd, arg & FD_CLOEXEC);

void fastcall set_close_on_exec(unsigned int fd, int flag)
{
    struct files_struct *files = current->files;
    struct fdtable *fdt;
    spin_lock(&files->file_lock);
    fdt = files_fdtable(files);
    if (flag)
        FD_SET(fd, fdt->close_on_exec);
    else
        FD_CLR(fd, fdt->close_on_exec);
    spin_unlock(&files->file_lock);
}
下面是man fcntl看到的对FD_CLOEXEC解释

File descriptor flags
   The  following  commands manipulate the flags associated with a file descriptor.  Currently, only one such flag is
   defined: FD_CLOEXEC, the close-on-exec flag.  If the FD_CLOEXEC bit is 0, the file  descriptor  will  remain  open
   across an execve(2), otherwise it will be closed.

   F_GETFD (void)
          Read the file descriptor flags; arg is ignored.
   F_SETFD (long)
          Set the file descriptor flags to the value specified by arg.

int CWatchDogDaemon::Fork(void)
{
	// an error detection pipe
	int err, fd[2];
	pipe(fd);
 
	// fork child process
	pid = fork();
	if (pid == -1)
		return pid;
 
	if (pid == 0)
	{
		// close pipe if exec succ
		close(fd[0]);
		fcntl(fd[1], F_SETFD, FD_CLOEXEC);
 
		Exec();
 
		err = errno;
		log_error("%s: exec(): %m", name);
		write(fd[1], &err, sizeof(err));
		exit(-1);
	}
 
	close(fd[1]);
	if(read(fd[0], &err, sizeof(err))==sizeof(err))
	{
		errno = err;
		return -1;
	}
	close(fd[0]);
 
	AttachWatchDog();
	return pid;
}
初一看会以为有问题，觉得由于read阻塞读写，父进程无论怎么样都会读到一个字节导致父进程会退出。
最后发现是下面这行代码在起作用：
fcntl(fd[1], F_SETFD, FD_CLOEXEC);
设置该标志会使得进程调用exec相关接口时如果成功会关闭管道，关闭管道时会向管理写一个字节的结束标志。这样read就会返回1导致条件不成立，反之则不关闭管道父进程就能正确得到子进程执行失败的消息。