Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Mon, 22 Mar 2021 17:05:59 +0100
From:   Oleg Nesterov <oleg@redhat.com>
To:     Jens Axboe <axboe@kernel.dk>
Cc:     io-uring@vger.kernel.org, torvalds@linux-foundation.org,
        ebiederm@xmission.com, linux-kernel@vger.kernel.org
Subject: Re: [PATCHSET 0/2] PF_IO_WORKER signal tweaks
Message-ID: <20210322160558.GA20390@redhat.com>
References: <20210320153832.1033687-1-axboe@kernel.dk>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20210320153832.1033687-1-axboe@kernel.dk>
User-Agent: Mutt/1.5.24 (2015-08-30)
Precedence: bulk

On 03/20, Jens Axboe wrote:
>
> Hi,
>
> Been trying to ensure that we do the right thing wrt signals and
> PF_IO_WORKER threads

OMG. Am I understand correctly? create_io_thread() can create a sub-
thread of userspace process which acts as a kernel thread?

Looks like this is the recent feature I wasn't aware... Can't really
comment right now, just some random and possibly wrong notes.

> 1) Just don't allow signals to them in general. We do mask everything
>    as blocked, outside of SIGKILL, so things like wants_signal() will
>    never return true for them.

This only means that signal_wake_up() won't be called. But the signal
will be queued if sent via tkill/etc, I don't think this is what we want?

A PF_IO_WORKER thread should ignore the signals. But it seems that the
PF_IO_WORKER check in sig_task_ignored() makes no sense and can't help.
I don't think PF_IO_WORKER && SIG_KTHREAD_KERNEL is possible.

Not to mention that sig_ignored() won't even call sig_task_ignored(),
it will return false exactly because the signal is blocked.

Confused.

Plus the the setting of tsk->blocked in create_io_thread() looks racy,
signal_pending() can be already true. And in fact it can't really help,
calculate_sigpending() can set TIF_SIGPENDING after wake_up_new_task()
anyway.

And why does create_io_thread() use lower_32_bits() ? This looks very
confusing. This

	.exit_signal    = (lower_32_bits(flags) & CSIGNAL);

too. Firstly, the rhs is always zero, secondly it is ignored because
of CLONE_THREAD.


ptrace_attach() checks PF_IO_WORKER too. Yes, but 'gdb -p' will try
to attach to every thread /proc/pid/tasks, so it will probably just
hang?

Oleg.