Message-ID: <87mu4bjlqm.fsf@x220.int.ebiederm.org>
Date:   Tue, 07 Jul 2020 12:09:05 -0500
From:   ebiederm@...ssion.com (Eric W. Biederman)
To:     Christian Brauner <christian.brauner@...ntu.com>
Cc:     Alexei Starovoitov <alexei.starovoitov@...il.com>,
        linux-kernel@...r.kernel.org, David Miller <davem@...emloft.net>,
        Greg Kroah-Hartman <greg@...ah.com>,
        Tetsuo Handa <penguin-kernel@...ove.sakura.ne.jp>,
        Kees Cook <keescook@...omium.org>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Alexei Starovoitov <ast@...nel.org>,
        Al Viro <viro@...iv.linux.org.uk>, bpf <bpf@...r.kernel.org>,
        linux-fsdevel <linux-fsdevel@...r.kernel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Masahiro Yamada <yamada.masahiro@...ionext.com>,
        Gary Lin <GLin@...e.com>, Bruno Meneguele <bmeneg@...hat.com>,
        LSM List <linux-security-module@...r.kernel.org>,
        Casey Schaufler <casey@...aufler-ca.com>,
        Luis Chamberlain <mcgrof@...nel.org>,
        Linus Torvalds <torvalds@...ux-foundation.org>
Subject: Re: [PATCH v3 13/16] exit: Factor thread_group_exited out of pidfd_poll

Christian Brauner <christian.brauner@...ntu.com> writes:

> On Fri, Jul 03, 2020 at 04:37:47PM -0500, Eric W. Biederman wrote:
>> Alexei Starovoitov <alexei.starovoitov@...il.com> writes:
>> 
>> > On Thu, Jul 02, 2020 at 11:41:37AM -0500, Eric W. Biederman wrote:
>> >> Create an independent helper thread_group_exited report return true
>> >> when all threads have passed exit_notify in do_exit.  AKA all of the
>> >> threads are at least zombies and might be dead or completely gone.
>> >> 
>> >> Create this helper by taking the logic out of pidfd_poll where
>> >> it is already tested, and adding a missing READ_ONCE on
>> >> the read of task->exit_state.
>> >> 
>> >> I will be changing the user mode driver code to use this same logic
>> >> to know when a user mode driver needs to be restarted.
>> >> 
>> >> Place the new helper thread_group_exited in kernel/exit.c and
>> >> EXPORT it so it can be used by modules.
>> >> 
>> >> Signed-off-by: "Eric W. Biederman" <ebiederm@...ssion.com>
>> >> ---
>> >>  include/linux/sched/signal.h |  2 ++
>> >>  kernel/exit.c                | 24 ++++++++++++++++++++++++
>> >>  kernel/fork.c                |  6 +-----
>> >>  3 files changed, 27 insertions(+), 5 deletions(-)
>> >> 
>> >> diff --git a/include/linux/sched/signal.h b/include/linux/sched/signal.h
>> >> index 0ee5e696c5d8..1bad18a1d8ba 100644
>> >> --- a/include/linux/sched/signal.h
>> >> +++ b/include/linux/sched/signal.h
>> >> @@ -674,6 +674,8 @@ static inline int thread_group_empty(struct task_struct *p)
>> >>  #define delay_group_leader(p) \
>> >>  		(thread_group_leader(p) && !thread_group_empty(p))
>> >>  
>> >> +extern bool thread_group_exited(struct pid *pid);
>> >> +
>> >>  extern struct sighand_struct *__lock_task_sighand(struct task_struct *task,
>> >>  							unsigned long *flags);
>> >>  
>> >> diff --git a/kernel/exit.c b/kernel/exit.c
>> >> index d3294b611df1..a7f112feb0f6 100644
>> >> --- a/kernel/exit.c
>> >> +++ b/kernel/exit.c
>> >> @@ -1713,6 +1713,30 @@ COMPAT_SYSCALL_DEFINE5(waitid,
>> >>  }
>> >>  #endif
>> >>  
>> >> +/**
>> >> + * thread_group_exited - check that a thread group has exited
>> >> + * @pid: tgid of thread group to be checked.
>> >> + *
>> >> + * Test if thread group is has exited (all threads are zombies, dead
>> >> + * or completely gone).
>> >> + *
>> >> + * Return: true if the thread group has exited. false otherwise.
>> >> + */
>> >> +bool thread_group_exited(struct pid *pid)
>> >> +{
>> >> +	struct task_struct *task;
>> >> +	bool exited;
>> >> +
>> >> +	rcu_read_lock();
>> >> +	task = pid_task(pid, PIDTYPE_PID);
>> >> +	exited = !task ||
>> >> +		(READ_ONCE(task->exit_state) && thread_group_empty(task));
>> >> +	rcu_read_unlock();
>> >> +
>> >> +	return exited;
>> >> +}
>> >
>> > I'm not sure why you think READ_ONCE was missing.
>> > It's different in wait_consider_task() where READ_ONCE is needed because
>> > of multiple checks. Here it's done once.
>> 
>> In practice it probably has no effect on the generated code.  But
>> READ_ONCE is about telling the compiler not to be clever.  Don't use
>> tearing loads or stores etc.  When all of the other readers are using
>> READ_ONCE I just get nervous if we have a case that doesn't.
>
> That's not true. The only place where READ_ONCE(->exit_state) is used is
> in wait_consider_task() and nowhere else. We had that discussion a while
> ago where I or someone proposed to simply place a READ_ONCE() around all
> accesses to exit_state for the sake of kcsan and we agreed that it's
> unnecessary and not to do this.
> But it obviously doesn't hurt to have it.

There is a larger discussion to be had around the proper handling of
exit_state.

In this particular case because we are accessing exit_state with
only rcu_read_lock protection, because the outcome of the read
is about correctness, and because the compiler has nothing else
telling it not to re-read exit_state, I believe we actually need
the READ_ONCE.

At the same time it would take a pretty special compiler to want to
reaccess that field in thread_group_exited.

I have looked through and I don't find any of the other access of
exit_state where the result is about correctness (so that we care)
and we don't hold tasklist_lock.

But I have removed the necessary wording from the commit comment.

There is a much larger discussion to be had about what to do with
exit_state, because I think I found about half the accesses were
slightly buggy in one form or another.

Eric

Powered by blists - more mailing lists