Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Mon, 15 Apr 2019 15:24:16 +0200
From:   Oleg Nesterov <oleg@redhat.com>
To:     Christian Brauner <christian@brauner.io>
Cc:     torvalds@linux-foundation.org, viro@zeniv.linux.org.uk,
        jannh@google.com, dhowells@redhat.com, linux-api@vger.kernel.org,
        linux-kernel@vger.kernel.org, serge@hallyn.com, luto@kernel.org,
        arnd@arndb.de, ebiederm@xmission.com, keescook@chromium.org,
        tglx@linutronix.de, mtk.manpages@gmail.com,
        akpm@linux-foundation.org, cyphar@cyphar.com,
        joel@joelfernandes.org, dancol@google.com
Subject: Re: [PATCH 2/4] clone: add CLONE_PIDFD
Message-ID: <20190415132416.GB22204@redhat.com>
References: <20190414201436.19502-1-christian@brauner.io>
 <20190414201436.19502-3-christian@brauner.io>
 <20190415105209.GA22204@redhat.com>
 <20190415114204.ydczeuwmi74wfsuv@brauner.io>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190415114204.ydczeuwmi74wfsuv@brauner.io>
User-Agent: Mutt/1.5.24 (2015-08-30)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 04/15, Christian Brauner wrote:
>
> > CLONE_PARENT_SETTID doesn't look very usefule, so what if we add
> >
> > 	if ((clone_flags & (CLONE_PIDFD|CLONE_PARENT_SETTID)) ==
> > 	                   (CLONE_PIDFD|CLONE_PARENT_SETTID))
> > 		return ERR_PTR(-EINVAL);
> >
> > at the start of copy_process() ?
> >
> > Then it can do
> >
> > 	if (clone_flags & CLONE_PIDFD) {
> > 		retval = pidfd_create(pid, &pidfdf);
> > 		if (retval < 0)
> > 			goto bad_fork_free_pid;
> > 		retval = put_user(retval, parent_tidptr)
> > 		if (retval < 0)
> > 			goto bad_fork_free_pid;
> > 	}
>
> Uhhh Oleg, that is nifty. I have to say I like that a lot. This would
> let us return the pid and the pidfd in one go and we can also start
> pidfd numbering at 0.

Christian, sorry if it was already discussed, but I can't force myself to
read all the previous discussions ;)

If we forget about CONFIG_PROC_FS, why do we really want to create a file?


Suppose we add a global u64 counter incremented by copy_process and reported
in /proc/$pid/status. Suppose that clone(CLONE_PIDFD) writes this counter to
*parent_tidptr. Let's denote this counter as UNIQ_PID.

Now, if you want to (say) safely kill a task and you have its UNIQ_PID, you
can do

	kill_by_pid_uniq(int pid, u64 uniq_pid)
	{
		pidfd = open("/proc/$pid", O_DIRECTORY);

		status = openat(pidfd, "status");
		u64 this_uniq_pid = ... read UNIQ_PID from status ...;

		if (uniq_pid != this_uniq_pid)
			return;

		pidfd_send_signal(pidfd);
	}

Why else do we want pidfd?

Oleg.