Activity

1. asyd commented on Sep 1, 2017

   

   asyd

   on Sep 1, 2017

   More actions

   Any news?
2. socketpair commented on Sep 2, 2017

   

   socketpair

   on Sep 2, 2017

   Author

   More actions

   Hm. I did not work on this. Someone should choose a way of solving this issue.
3. dohoangkhiem commented on Sep 4, 2017

   

   dohoangkhiem

   on Sep 4, 2017

   More actions

   @socketpair Do you have any suggestion on this? Changing the socket permission from outside can make it works but just temporarily, as long as the application is restarted the permission is lost and process like nginx (with www-data user) cannot connect anymore.
4. socketpair commented on Sep 4, 2017

   

   socketpair

   on Sep 4, 2017

   Author

   More actions

   Well.

   I do not understand what you mean as "outside". Anyway, too broad permission is a security bug. If appliation is restarted, UNIX-socket is recreated (re-bound) and this code will fire again.

   Variant 1 will looks like:

   def do_bind(sk, addr, mode):
       pid = os.fork()
       if pid == 0:
           try:
               os.umask(~mode & 0o777)
               sk.bind(addr)
           except:
               os._exit(1)
           else:
               os._exit(0)
       (pid, status) = os.waitpid(pid, 0)
       if status:
           raise RuntimeError('Failed to bind the socket.')

   Variant 2 is not atomic.

   Variant 3 (since asyncio now is a port of CPython): Do that under GIL locked in C. But anyway this will affect third-party threads.