Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Organization: Red Hat UK Ltd. Registered Address: Red Hat UK Ltd, Amberley
        Place, 107-111 Peascod Street, Windsor, Berkshire, SI4 1TE, United
        Kingdom.
        Registered in England and Wales under Company Registration No. 3798903
From:   David Howells <dhowells@redhat.com>
In-Reply-To: <877emb2740.fsf@notabene.neil.brown.name>
References: <877emb2740.fsf@notabene.neil.brown.name> <20180222073330.36259-1-carmark.dlut@gmail.com> <afbfd186-1f06-a03c-79b5-c98a12d35c75@metrics.net>
To:     NeilBrown <neilb@suse.com>
Cc:     dhowells@redhat.com, Andrew Morton <akpm@linux-foundation.org>,
        Anthony DeRobertis <aderobertis@metrics.net>,
        linux-cachefs@redhat.com, linux-kernel@vger.kernel.org,
        Lei Xue <carmark.dlut@gmail.com>,
        Vegard Nossum <vegard.nossum@gmail.com>,
        Daniel Axtens <dja@axtens.net>,
        KiranKumar Modukuri <kiran.modukuri@gmail.com>
Subject: Re: [PATCH] cachefiles: fix multiple-put race.
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <24654.1530695695.1@warthog.procyon.org.uk>
Date:   Wed, 04 Jul 2018 10:14:55 +0100
Message-ID: <24655.1530695695@warthog.procyon.org.uk>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

NeilBrown <neilb@suse.com> wrote:

> +	fscache_enqueue_retrieval(monitor->op);
> +
>  	spin_lock(&object->work_lock);
>  	list_add_tail(&monitor->op_link, &monitor->op->to_do);
>  	spin_unlock(&object->work_lock);
>  
> -	fscache_enqueue_retrieval(monitor->op);

That won't necessarily work because the work processor can then happen before
you've added the work to the to_do list.

I'm thinking that KiranKumar's solution might be the best one.

The problem is that cachefiles_read_waiter() doesn't have a ref on the monitor
object but is entirely dependent on the waitqueue lock for safety.  I think
KiranKumar's patch is correct to take a ref before doing the queuing.  It
might be possible to then pass this along to the work processor, but that
might be too fiddly.

Actually, I want to get rid of the page monitoring stuff entirely as it's
quite fragile and use an iterator and direct-IO instead, but we have to fix
this for now.

David