Re: [PATCH v1 1/2] Add polling support to pidfd

From: Christian Brauner <christian@brauner.io>
Date: 2019-04-26 15:31:57
Also in: linux-kselftest, lkml

On April 26, 2019 5:21:40 PM GMT+02:00, Christian Brauner [off-list ref] wrote:

On Fri, Apr 26, 2019 at 10:23:37AM -0400, Joel Fernandes wrote:

quoted

On Fri, Apr 26, 2019 at 12:24:04AM +0200, Christian Brauner wrote:

quoted

On Thu, Apr 25, 2019 at 03:00:09PM -0400, Joel Fernandes (Google)

wrote:

quoted

pidfd are file descriptors referring to a process created with

the

quoted

CLONE_PIDFD clone(2) flag. Android low memory killer (LMK) needs

pidfd

quoted

polling support to replace code that currently checks for

existence of

quoted

/proc/pid for knowing that a process that is signalled to be

killed has

quoted

died, which is both racy and slow. The pidfd poll approach is

race-free,

quoted

and also allows the LMK to do other things (such as by polling on

other

quoted

fds) while awaiting the process being killed to die.

Thanks for the patch!

Ok, let me be a little bit anal.
Please start the commit message with what this patch does and then

add

quoted

The subject title is "Add polling support to pidfd", but ok I should

write a

quoted

better commit message.

Yeah, it's really just that we should really just have a simple
paragraph that expresses this makes the codebase do X.

quoted

the justification why. You just say the "pidfd-poll" approach. You

can

quoted

probably assume that CLONE_PIDFD is available for this patch. So
something like:

"This patch makes pidfds pollable. Specifically, it allows

listeners to

quoted

be informed when the process the pidfd referes to exits. This patch

only

quoted

introduces the ability to poll thread-group leaders since pidfds
currently can only reference those..."

Then justify the use-case and then go into implementation details.
That's usually how I would think about this:
- Change the codebase to do X
- Why do we need X
- Are there any technical details worth mentioning in the commit

message

quoted

(- Are there any controversial points that people stumbled upon but

that

quoted

  have been settled sufficiently.)

Generally the "how" in the patch should be in the code, but ok.

That's why I said: technical details that are worth mentioning.
Sometimes you have controversial bits that are obviously to be
understood in the code but it still might be worth explaining why one
had to do it this way. Like say what we did for the pidfd_send_signal()
thing where we explained why O_PATH is disallowed.

quoted

I changed the first 3 paragraphs of the changelog to the following,

is that

quoted

better? :

Android low memory killer (LMK) needs to know when a process dies

once

quoted

it is sent the kill signal. It does so by checking for the existence

of

quoted

/proc/pid which is both racy and slow. For example, if a PID is

reused

quoted

between when LMK sends a kill signal and checks for existence of the
PID, since the wrong PID is now possibly checked for existence.

This patch adds polling support to pidfd. Using the polling support,

LMK

quoted

will be able to get notified when a process exists in race-free and

fast

quoted

way, and allows the LMK to do other things (such as by polling on

other

quoted

fds) while awaiting the process being killed to die.

For notification to polling processes, we follow the same existing
mechanism in the kernel used when the parent of the task group is to

be

quoted

notified of a child's death (do_notify_parent).  This is precisely

when

quoted

the tasks waiting on a poll of pidfd are also awakened in this patch.

quoted

pidfd are file descriptors referring to a process created with

the

quoted

CLONE_PIDFD clone(2) flag. Android low memory killer (LMK) needs

pidfd

quoted

polling support to replace code that currently checks for

existence of

quoted

/proc/pid for knowing that a process that is signalled to be

killed has

quoted

died, which is both racy and slow. The pidfd poll approach is

race-free,

quoted

and also allows the LMK to do other things (such as by polling on

other

quoted

fds) while awaiting the process being killed to die.

quoted

It prevents a situation where a PID is reused between when LMK

sends a

quoted

kill signal and checks for existence of the PID, since the wrong

PID is

quoted

now possibly checked for existence.

In this patch, we follow the same existing mechanism in the

kernel used

quoted

when the parent of the task group is to be notified

(do_notify_parent).

quoted

This is when the tasks waiting on a poll of pidfd are also

awakened.

quoted

We have decided to include the waitqueue in struct pid for the

following

quoted

reasons:
1. The wait queue has to survive for the lifetime of the poll.

Including

quoted

it in task_struct would not be option in this case because the

task can

quoted

be reaped and destroyed before the poll returns.

2. By including the struct pid for the waitqueue means that

during

quoted

de_thread(), the new thread group leader automatically gets the

new

quoted

waitqueue/pid even though its task_struct is different.

Appropriate test cases are added in the second patch to provide

coverage

quoted

of all the cases the patch is handling.

Andy had a similar patch [1] in the past which was a good

reference

quoted

however this patch tries to handle different situations properly

related

quoted

to thread group existence, and how/where it notifies. And also

solves

quoted

other bugs (waitqueue lifetime).  Daniel had a similar patch [2]
recently which this patch supercedes.

[1] https://lore.kernel.org/patchwork/patch/345098/
[2]

https://lore.kernel.org/lkml/20181029175322.189042-1-dancol@google.com/ (local)

quoted

Cc: luto@amacapital.net
Cc: rostedt@goodmis.org
Cc: dancol@google.com
Cc: sspatil@google.com
Cc: christian@brauner.io
Cc: jannh@google.com
Cc: surenb@google.com
Cc: timmurray@google.com
Cc: Jonathan Kowalski <redacted>
Cc: torvalds@linux-foundation.org
Cc: kernel-team@android.com

These should all be in the form:

Cc: Firstname Lastname <redacted>

If this bothers you too much, I can also just remove the CC list from

the

quoted

changelog here, and include it in my invocation of git-send-email

instead..

quoted

but I have seen commits in the tree that don't follow this rule.

Yeah, but they should. There are people with multiple emails over the
years and they might not necessarily contain their first and last
name. And I don't want to have to mailmap them or sm. Having their
names
in there just makes it easier. Also, every single other DCO-*related*
line follows:

Random J Developer [off-list ref]

This should too. If others are sloppy and allow this, fine. No reason
we
should.

quoted

There are people missing from the Cc that really should be there...

If you look at the CC list of the email, people in the

get_maintainer.pl

quoted

script were also added. I did run get_maintainer.pl and checkpatch.

But ok, I

quoted

will add the folks you are suggesting as well. Thanks.

get_maintainer.pl is not the last word.

quoted

Even though he usually doesn't respond that often, please Cc Al on

this.

quoted

If he responds it's usually rather important.

No issues on that, but I am wondering if he should also be in

MAINTAINERS

quoted

file somewhere such that get_maintainer.pl does pick him up. I added

him.

It's often not about someone being a maintainer but whether or not they
have valuable input.

"[...] This tag documents that potentially interested parties have been
included in the discussion."

quoted

Oleg has reviewed your RFC patch quite substantially and given

valuable

quoted

feedback and has an opinion on this thing and is best acquainted

with

quoted

the exit code. So please add him to the Cc of the commit message in

the

quoted

appropriate form and also add him to the Cc of the thread.

Done.

Thanks!

quoted

Probably also want linux-api for good measure since a lot of people

are

quoted

subscribed that would care about pollable pidfds. I'd also add Kees
since he had some interest in this work and David (Howells).

Done, I added all of them and CC will go out to them next time.

Thanks.

Cool. That really wasn't a "you've done this wrong". It's rather really
just to make sure that everyone who might catch a big f*ck up on our
part has had a chance to tell us so. :)

quoted

Co-developed-by: Daniel Colascione <redacted>

Every CDB needs to give a SOB as well.

Ok, done. thanks.

Fwiw, I only learned this recently too.

quoted

Signed-off-by: Joel Fernandes (Google) <redacted>

---

RFC -> v1:
* Based on CLONE_PIDFD patches: https://lwn.net/Articles/786244/
* Updated selftests.
* Renamed poll wake function to do_notify_pidfd.
* Removed depending on EXIT flags
* Removed POLLERR flag since semantics are controversial and
  we don't have usecases for it right now (later we can add if

there's

quoted

  a need for it).

 include/linux/pid.h |  3 +++
 kernel/fork.c       | 33 +++++++++++++++++++++++++++++++++
 kernel/pid.c        |  2 ++
 kernel/signal.c     | 14 ++++++++++++++
 4 files changed, 52 insertions(+)

diff --git a/include/linux/pid.h b/include/linux/pid.h
index 3c8ef5a199ca..1484db6ca8d1 100644
--- a/include/linux/pid.h
+++ b/include/linux/pid.h

@@ -3,6 +3,7 @@
 #define _LINUX_PID_H
 
 #include <linux/rculist.h>
+#include <linux/wait.h>
 
 enum pid_type
 {

@@ -60,6 +61,8 @@ struct pid
 	unsigned int level;
 	/* lists of tasks that use this pid */
 	struct hlist_head tasks[PIDTYPE_MAX];
+	/* wait queue for pidfd notifications */
+	wait_queue_head_t wait_pidfd;
 	struct rcu_head rcu;
 	struct upid numbers[1];
 };

diff --git a/kernel/fork.c b/kernel/fork.c
index 5525837ed80e..fb3b614f6456 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c

@@ -1685,8 +1685,41 @@ static void pidfd_show_fdinfo(struct

seq_file *m, struct file *f)

quoted

 }
 #endif
 
+static unsigned int pidfd_poll(struct file *file, struct

poll_table_struct *pts)

quoted

+{
+	struct task_struct *task;
+	struct pid *pid;
+	int poll_flags = 0;
+
+	/*
+	 * tasklist_lock must be held because to avoid racing with
+	 * changes in exit_state and wake up. Basically to avoid:
+	 *
+	 * P0: read exit_state = 0
+	 * P1: write exit_state = EXIT_DEAD
+	 * P1: Do a wake up - wq is empty, so do nothing
+	 * P0: Queue for polling - wait forever.
+	 */
+	read_lock(&tasklist_lock);
+	pid = file->private_data;
+	task = pid_task(pid, PIDTYPE_PID);
+	WARN_ON_ONCE(task && !thread_group_leader(task));
+
+	if (!task || (task->exit_state && thread_group_empty(task)))
+		poll_flags = POLLIN | POLLRDNORM;

So we block until the thread-group is empty? Hm, the thread-group

leader

quoted

remains in zombie state until all threads are gone. Should probably

just

quoted

be a short comment somewhere that callers are only informed about a
whole thread-group exit and not about when the thread-group leader

has

quoted

actually exited.

Ok, I'll add a comment.

quoted

I would like the ability to extend this interface in the future to

allow

quoted

for actually reading data from the pidfd on EPOLLIN.
POSIX specifies that POLLIN and POLLRDNORM are set even if the
message to be read is zero. So one cheap way of doing this would
probably be to do a 0 read/ioctl. That wouldn't hurt your very

limited

quoted

usecase and people could test whether the read returned non-0 data

and

quoted

if so they know this interface got extended. If we never extend it

here

quoted

it won't matter.

I am a bit confused. What specific changes to this patch are you

proposing?

quoted

This patch makes poll block until the process exits. In the future,

we can

quoted

make it unblock for a other reasons as well, that's fine with me. I

don't see

quoted

how this patch prevents such extensions.

I guess I should've asked the following:
What happens right now, when you get EPOLLIN on the pidfd and you and
out of ignorance you do:

read(pidfd, ...)

I guess it returns EINVAL which is fine.
So you can ignore that comment.

quoted

+	if (!poll_flags)
+		poll_wait(file, &pid->wait_pidfd, pts);
+
+	read_unlock(&tasklist_lock);
+
+	return poll_flags;
+}

quoted

+
+
 const struct file_operations pidfd_fops = {
 	.release = pidfd_release,
+	.poll = pidfd_poll,
 #ifdef CONFIG_PROC_FS
 	.show_fdinfo = pidfd_show_fdinfo,
 #endif

diff --git a/kernel/pid.c b/kernel/pid.c
index 20881598bdfa..5c90c239242f 100644
--- a/kernel/pid.c
+++ b/kernel/pid.c

@@ -214,6 +214,8 @@ struct pid *alloc_pid(struct pid_namespace

*ns)

quoted

 	for (type = 0; type < PIDTYPE_MAX; ++type)
 		INIT_HLIST_HEAD(&pid->tasks[type]);
 
+	init_waitqueue_head(&pid->wait_pidfd);
+
 	upid = pid->numbers + ns->level;
 	spin_lock_irq(&pidmap_lock);
 	if (!(ns->pid_allocated & PIDNS_ADDING))

diff --git a/kernel/signal.c b/kernel/signal.c
index 1581140f2d99..16e7718316e5 100644
--- a/kernel/signal.c
+++ b/kernel/signal.c

@@ -1800,6 +1800,17 @@ int send_sigqueue(struct sigqueue *q,

struct pid *pid, enum pid_type type)

quoted

 	return ret;
 }
 
+static void do_notify_pidfd(struct task_struct *task)

Maybe a short command that this helper can only be called when we

know

quoted

that task is a thread-group leader wouldn't hurt so there's no

confusion

quoted

later.

Ok, will do.

thanks,

 - Joel

`h`	back out one level
`j`	next message in thread
`k`	previous message in thread
`l`	drill in
`Esc`	close help / fold thread tree
`?`	toggle this help