Received-SPF: pass (google.com: domain of linux-kernel+bounces-46612-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
Date: Wed, 31 Jan 2024 15:39:00 +0100
From: Christian Brauner <brauner@kernel.org>
To: Oleg Nesterov <oleg@redhat.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>, 
	Tycho Andersen <tycho@tycho.pizza>, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v3 0/1] pidfd: implement PIDFD_THREAD flag for
 pidfd_open()
Message-ID: <20240131-engel-entern-9b5c96659948@brauner>
References: <20240131132541.GA23632@redhat.com>
 <20240131141204.GA24130@redhat.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <20240131141204.GA24130@redhat.com>

On Wed, Jan 31, 2024 at 03:12:04PM +0100, Oleg Nesterov wrote:
> Before I forget this...
> 
> After this patch we can easily add another feature, pidfd_poll()
> can add, say, POLLHUP to poll_flags if the pid is "dead".
> 
> So the user can do
> 
> 	poll(pidfd, { .revents = POLLHUP });
> 
> and it will block until release_task() is called and this pid is
> no longer in use (pid_task() == NULL).
> 
> Do you think this can be useful?

Yeah, I think this is something that people would find useful. IIUC, it
would essentially allow them to do things like wait until a task has
been waited upon which for service managers is very useful information.

---

Btw, bigger picture for a second. You probably haven't kept track of
this but the fact that we got pidfds - thanks a lot to your review and
help - has made quite a huge difference in userspace. Since we last did
any meaningful work we got:

* systemd completely relying on pidfds to manage services to guard
  against any pid races.
* Extended dbus to allow authentication via pidfds.
* Extended policy kit to enable secure authentication of processes via pidfds.
* Language support for pidfds: Go, Rust etc.
* An endless number of tools that added support for them.
* glibc support for pidfd apis.

There's a bunch more. That literally obliterated whole bug classes.

I think with PIDFD_THREAD support it might make it interesting to use it
for some pthread management as well.