Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   "Eric W. Biederman" <ebiederm@xmission.com>
To:     michael.christie@oracle.com
Cc:     Oleg Nesterov <oleg@redhat.com>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        linux@leemhuis.info, nicolas.dichtel@6wind.com, axboe@kernel.dk,
        linux-kernel@vger.kernel.org,
        virtualization@lists.linux-foundation.org, mst@redhat.com,
        sgarzare@redhat.com, jasowang@redhat.com, stefanha@redhat.com,
        brauner@kernel.org
References: <20230524141022.GA19091@redhat.com>
        <87ttw1zt4i.fsf@email.froward.int.ebiederm.org>
        <20230525115512.GA9229@redhat.com>
        <87y1lcxwcj.fsf@email.froward.int.ebiederm.org>
        <CAHk-=wj4DS=2F5mW+K2P7cVqrsuGd3rKE_2k2BqnnPeeYhUCvg@mail.gmail.com>
        <87cz2mrtnk.fsf@email.froward.int.ebiederm.org>
        <CAHk-=whsi9JFP-okH3jXHrA8rh8bMuuSt6ZgkmPwiDMAn437qA@mail.gmail.com>
        <87mt1pmezu.fsf@email.froward.int.ebiederm.org>
        <20230529111859.GA15193@redhat.com>
        <022f4de6-9eae-0a94-0f55-b84be4982fc3@oracle.com>
        <20230529174646.GB15193@redhat.com>
        <ff1bce1a-62d8-1b2e-4560-1ce1ffc209bc@oracle.com>
        <723ac1ba-31d2-92fe-4010-42d8cd70d5df@oracle.com>
Date:   Mon, 29 May 2023 21:48:34 -0500
In-Reply-To: <723ac1ba-31d2-92fe-4010-42d8cd70d5df@oracle.com> (michael
        christie's message of "Mon, 29 May 2023 14:46:57 -0500")
Message-ID: <87bki2h6v1.fsf@email.froward.int.ebiederm.org>
User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/27.1 (gnu/linux)
MIME-Version: 1.0
Content-Type: text/plain
Subject: Re: [PATCH 3/3] fork, vhost: Use CLONE_THREAD to fix freezer/ps
 regression
Precedence: bulk

michael.christie@oracle.com writes:

> On 5/29/23 2:35 PM, Mike Christie wrote:
>>> Hmm... If we you CLONE_THREAD the exiting vhost_worker() will auto-reap itself,
>> Oh wait, are you saying that when we get auto-reaped then we would do the last
>> fput and call the file_operations->release function right? We actually set
>> task_struct->files = NULL for the vhost_task task_struct, so I think we call
>> release a little sooner than you think.
>> 
>> vhost_task_create() sets kernel_clone_args->no_files, so the vhost_task task_struc
>> that gets created works like kthreads where it doesn't do a CLONE_FILES and it
>> doesn't do a dup_fd.
>> 
>> So when we do de_thread() -> zap_other_threads(), that will kill all the threads
>> in the group right? So when they exit, it will call our release function since
>> we don't have refcount on ourself.
>> 
>
> Just to make sure I'm on the same page now.
>
> In the past thread when were discussing the patch below and you guys were saying
> that it doesn't really ignore SIGKILL because we will hit the
> SIGNAL_GROUP_EXIT/group_exec_task checks and the parent is going to exit, it was
> on purpose.
>
> Instead of a signal to tell me when do exit, I was using the parent exiting and doing
> the last fput on the vhost device's file which calls our release function. That then
> allowed the vhost code to use the vhost_task to handle the outstanding IOs and then
> do vhost_task_should_stop to tell the vhost_task to exit when the oustanding IO
> had completed.
>
> On the vhost side of things it's really nice, because all the shutdown paths work
> the same and we don't need rcu/locking in the main IO path to handle the vhost_task
> exiting while we are using it.

The code below does nothing for exec.

You really need to call get_signal to handle SIGSTOP/freeze/process exit.

A variant on my coredump proposal looks like it can handle exec as well.
It isn't pretty but it looks good enough for right now.

If you could test the patch I posted up thread I think that is something
imperfect that is good enough for now.

Eric