封闭文件描述符发生了什么？

Question

在下面的示例代码中,我们打开一个文件描述符到sandbox.log,将它作为stdout提供给子进程,然后关闭文件描述符,但子进程仍然可以写入该文件.是subprocess.Popen在内部复制文件描述符吗？将文件描述符传递给子进程后关闭它是否安全？

import subprocess
import os
import time


print 'create or clear sandbox.log'
subprocess.call('touch sandbox.log', shell=True)
subprocess.call('echo "" > sandbox.log', shell=True)

print 'open the file descriptor'
fd = os.open('sandbox.log', os.O_WRONLY)

command = 'sleep 10 && echo "hello world"'
print 'run the command'
p = subprocess.Popen(command, stdout=fd, stderr=subprocess.STDOUT, shell=True)

os.close(fd)

try:
    os.close(fd)
except OSError:
    print 'fd is already closed'
else:
    print 'fd takes some time to close'

if p.poll() is None:
    print 'p isnt finished, but fd is closed'

p.wait()
print 'p just finished'

with open('sandbox.log') as f:
    if any('hello world' in line for line in f):
        raise Exception("There's text in sandbox.log. Whats going on?")

作为参考,我得到以下输出作为脚本运行上面的代码:

% python test_close_fd.py
create or clear sandbox.log
open the file descriptor
run the command
fd is already closed
p isnt finished, but fd is closed
p just finished
Traceback (most recent call last):
  File "test_close_fd.py", line 34, in <module>
    raise Exception("There's text in sandbox.log. Whats going on?")
Exception: There's text in sandbox.log. Whats going on?

Answer 1

每个进程都有自己的一组文件描述符.在一个程序中关闭fd不会影响另一个程序.这就是为什么每个程序都可以使用相同的FD号码标准输入(0),标准输出(1),和stderr(2),为什么Shell脚本通常只开放FD 3,而不必检查它是否可用.

文件描述符由子进程继承,除非您通过设置close-on-exec标志明确地阻止它.默认情况下,如果没有该标志,子进程将获取父文件描述符的副本.