Received-SPF: pass (google.com: domain of linux-nfs+bounces-270-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Date: Mon, 4 Dec 2023 02:40:31 +0000
From: Al Viro <viro@zeniv.linux.org.uk>
To: NeilBrown <neilb@suse.de>
Cc: Christian Brauner <brauner@kernel.org>, Jens Axboe <axboe@kernel.dk>,
	Oleg Nesterov <oleg@redhat.com>,
	Chuck Lever <chuck.lever@oracle.com>,
	Jeff Layton <jlayton@kernel.org>, Ingo Molnar <mingo@redhat.com>,
	Peter Zijlstra <peterz@infradead.org>,
	Juri Lelli <juri.lelli@redhat.com>,
	Vincent Guittot <vincent.guittot@linaro.org>,
	linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
	linux-nfs@vger.kernel.org
Subject: Re: [PATCH 1/2] Allow a kthread to declare that it calls
 task_work_run()
Message-ID: <20231204024031.GV38156@ZenIV>
References: <20231204014042.6754-1-neilb@suse.de>
 <20231204014042.6754-2-neilb@suse.de>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20231204014042.6754-2-neilb@suse.de>
Sender: Al Viro <viro@ftp.linux.org.uk>

On Mon, Dec 04, 2023 at 12:36:41PM +1100, NeilBrown wrote:

> This means that any cost for doing the work is not imposed on the kernel
> thread, and importantly excessive amounts of work cannot apply
> back-pressure to reduce the amount of new work queued.

It also means that a stuck ->release() won't end up with stuck
kernel thread...

> earlier than would be ideal.  When __dput (from the workqueue) calls

WTF is that __dput thing?  __fput, perhaps?

> This patch adds a new process flag PF_RUNS_TASK_WORK which is now used
> instead of PF_KTHREAD to determine whether it is sensible to queue
> something to task_works.  This flag is always set for non-kernel threads.

*ugh*

What's that flag for?  task_work_add() always can fail; any caller must
have a fallback to cope with that possibility; fput() certainly does.

Just have the kernel threads born with ->task_works set to &work_exited
and provide a primitive that would flip it from that to NULL.

> @@ -1328,7 +1328,7 @@ static void mntput_no_expire(struct mount *mnt)
>  
>  	if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
>  		struct task_struct *task = current;
> -		if (likely(!(task->flags & PF_KTHREAD))) {
> +		if (likely((task->flags & PF_RUNS_TASK_WORK))) {
>  			init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
>  			if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
>  				return;

Now, *that* is something I have much stronger objections to.
Stuck filesystem shutdown is far more likely than stuck
->release().  You are seriously asking for trouble here.

Why would you want to have nfsd block on that?