How do you cleanly exit after enabling seccomp in

2019-05-04 15:34发布

问题:

I have enabled seccomp via python-prctl in a project. I can't quite figure out how to exit cleanly - the result is always a kill.

I saw some examples that use ctypes or ffi to try to reference libc, but if I expect them with WIFEXITED they also seem to have the same issue.

Example code below. The result is always "We were killed to death".

def main():
  pid = os.fork()
  if not pid:
    prctl.set_seccomp(True)
    os.write(0, 'Hi\n')

#    os._exit(0)
#    _exit(0)
#    sys._exit(0)
#    return
#    ?!@#(*!  What do?

  endpid, status = os.waitpid(pid, 0)
  print 'Child forked as %d and returned with %d' % (endpid, status)
  if not os.WIFEXITED(status):
    print 'Exitted abnormally'
    if os.WIFSIGNALED:
      if os.WTERMSIG(status) == signal.SIGKILL:
        print 'We were killed to death'
  else:
    print 'Returned with %d' % (os.WEXITSTATUS(status))

Quick update since I forgot the libc stuff:

Defining _exit() above with either of these still resulted in a kill.

# FFI Method
ffi = cffi.FFI()
# Use _exit, which avoids atexit(), etc
ffi.cdef('void _exit(int);')
libc = ffi.dlopen(None)
_exit = libc._exit

.... or ....

# ctypes method
libc = cdll.LoadLibrary('libc-2.18.so')
_exit = libc._exit

回答1:

Someone helped me find this answer:

In glibc up to version 2.3, the _exit() wrapper function invoked the kernel system
call of the same name.  Since glibc 2.3, the wrapper function invokes
exit_group(2), in order to terminate all of the threads in a process.

Since the _exit wraps exit_group and exit_group is not in the seccomp filter.... it gets killed. strace of the python execution shows the exit_group call.