Re: [PATCH] signalfd: don't dequeue the forced fatal signals

[Oleg Nesterov <oleg@redhat.com>]

On 04/08, kernel test robot wrote:
>
> kernel test robot noticed the following build warnings:

...

> sparse warnings: (new ones prefixed by >>)
> >> fs/signalfd.c:53:40: sparse: sparse: incorrect type in initializer (different address spaces) @@     expected struct k_sigaction *k @@     got struct k_sigaction [noderef] __rcu * @@

...

> vim +53 fs/signalfd.c
>
>     50	
>     51	static void mk_sigmask(struct signalfd_ctx *ctx, sigset_t *sigmask)
>     52	{
>   > 53		struct k_sigaction *k = current->sighand->action;

I am going to ignore this new warning...

Yes, task_struct->sighand is __rcu. Not sure this annotation makes a lot of sense.

In any case. current->sighand is always stable. Plus task->sighand is stable under siglock.

We have a lot of (correct) non-rcu deferences of ->sighand.

I think that only lock_task_sighand() needs rcu_dereference(tsk->sighand).

Say, __exit_signal() does

	sighand = rcu_dereference_check(tsk->sighand,
					lockdep_tasklist_lock_is_held());

To me this just adds the unnecessary noise. I do not want to add another precedent.

Oleg.