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:   Benjamin Segall <bsegall@google.com>
To:     Christian Brauner <brauner@kernel.org>
Cc:     Eric Biggers <ebiggers@kernel.org>,
        Alexander Viro <viro@zeniv.linux.org.uk>,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
        stable@vger.kernel.org, Andrew Morton <akpm@linux-foundation.org>
Subject: Re: [PATCH RESEND] epoll: ep_autoremove_wake_function should use
 list_del_init_careful
References: <xm26pm6hvfer.fsf@google.com>
        <20230531015748.GB1648@quark.localdomain>
        <20230531-zupacken-laute-22564cd952f7@brauner>
Date:   Wed, 31 May 2023 15:15:41 -0700
In-Reply-To: <20230531-zupacken-laute-22564cd952f7@brauner> (Christian
        Brauner's message of "Wed, 31 May 2023 09:53:01 +0200")
Message-ID: <xm26ilc8uoz6.fsf@google.com>
User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/28.2 (gnu/linux)
MIME-Version: 1.0
Content-Type: text/plain
Precedence: bulk

Christian Brauner <brauner@kernel.org> writes:

> On Tue, May 30, 2023 at 06:57:48PM -0700, Eric Biggers wrote:
>> On Tue, May 30, 2023 at 11:32:28AM -0700, Benjamin Segall wrote:
>> > autoremove_wake_function uses list_del_init_careful, so should epoll's
>> > more aggressive variant. It only doesn't because it was copied from an
>> > older wait.c rather than the most recent.
>> > 
>> > Fixes: a16ceb139610 ("epoll: autoremove wakers even more aggressively")
>> > Signed-off-by: Ben Segall <bsegall@google.com>
>> > Cc: stable@vger.kernel.org
>> > ---
>> >  fs/eventpoll.c | 2 +-
>> >  1 file changed, 1 insertion(+), 1 deletion(-)
>> > 
>> > diff --git a/fs/eventpoll.c b/fs/eventpoll.c
>> > index 52954d4637b5..081df056398a 100644
>> > --- a/fs/eventpoll.c
>> > +++ b/fs/eventpoll.c
>> > @@ -1756,11 +1756,11 @@ static struct timespec64 *ep_timeout_to_timespec(struct timespec64 *to, long ms)
>> >  static int ep_autoremove_wake_function(struct wait_queue_entry *wq_entry,
>> >  				       unsigned int mode, int sync, void *key)
>> >  {
>> >  	int ret = default_wake_function(wq_entry, mode, sync, key);
>> >  
>> > -	list_del_init(&wq_entry->entry);
>> > +	list_del_init_careful(&wq_entry->entry);
>> >  	return ret;
>> >  }
>> 
>> Can you please provide a more detailed explanation about why
>> list_del_init_careful() is needed here?
>
> Yeah, this needs more explanation... Next time someone looks at this
> code and there's a *_careful() added they'll want to know why.

So the general reason is the same as with autoremove_wake_function, it
pairs with the list_entry_careful in ep_poll (which is epoll's modified
copy of finish_wait).

I think the original actual _problem_ was a -stable issue that was fixed
instead by doing additional backports, so this may just avoid potential
extra loops and avoid potential compiler shenanigans from the data race.