Re: [PATCH -next 00/10] Fixes to controlling tty handling

On Thu, 16 Oct 2014 14:59:40 -0400
Peter Hurley [off-list ref] wrote:

Hi Greg,

This patch series:
1. removes stale code from the controlling tty handling functions
2. relocates the ctty functions to eliminate forward declarations
3. fixes several unsafe races when setting the controlling tty
4. eliminates holding tty_mutex as a necessary condition of
   setting the controlling terminal

#4 is part of an overall effort to reduce the tty_mutex footprint.

Unfortunately, this series does not fix two other race conditions:
1. disassociate_ctty()/no_tty() does not teardown the tty<->process
associations atomically wrt job control, so it is possible to
observe spurious error conditions from job control (tty_check_change()
and job_control()). I'm looking into inverting the lock order of
tty->ctrl_lock and tsk->sighand->siglock() to see if holding ctrl_lock
is a suitable solution for atomic teardown. Especially now that
ctrl_lock is not used for flow control anymore :)
2. task_pgrp() and task_session() are used unsafely. These fixes
will be clearer after #1 is fixed.

Reviewed-by: Alan Cox <redacted>

I can't prove entirely to my satisfaction that the claim in #9 is true in
the presence of simultaenous hangups opens and setsid but the locking
appears to be correct for the cases I was trying to figure out anyway.

Makes my head hurt just reviewing bits of this so thanks for doing all
this work !