Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
Date:   Thu, 2 Jun 2022 11:19:42 +0800
From:   Ming Lei <ming.lei@redhat.com>
To:     Pavel Machek <pavel@ucw.cz>
Cc:     Jens Axboe <axboe@kernel.dk>, linux-block@vger.kernel.org,
        linux-kernel@vger.kernel.org, io-uring@vger.kernel.org,
        Gabriel Krisman Bertazi <krisman@collabora.com>,
        ZiyangZhang <ZiyangZhang@linux.alibaba.com>,
        Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com>
Subject: Re: [RFC PATCH] ubd: add io_uring based userspace block driver
Message-ID: <YpgsTojc4mVKghZA@T590>
References: <20220509092312.254354-1-ming.lei@redhat.com>
 <20220530070700.GF1363@bug>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20220530070700.GF1363@bug>
Precedence: bulk

Hello Pavel,

On Mon, May 30, 2022 at 09:07:00AM +0200, Pavel Machek wrote:
> Hi!
> 
> > This is the driver part of userspace block driver(ubd driver), the other
> > part is userspace daemon part(ubdsrv)[1].
> 
> > @@ -0,0 +1,1193 @@
> > +// SPDX-License-Identifier: GPL-2.0-or-later
> > +/*
> > + * Userspace block device - block device which IO is handled from userspace
> > + *
> > + * Take full use of io_uring passthrough command for communicating with
> > + * ubd userspace daemon(ubdsrvd) for handling basic IO request.
> 
> > +
> > +static inline unsigned int ubd_req_build_flags(struct request *req)
> > +{
> ...
> > +	if (req->cmd_flags & REQ_SWAP)
> > +		flags |= UBD_IO_F_SWAP;
> > +
> > +	return flags;
> > +}
> 
> Does it work? How do you guarantee operation will be deadlock-free with swapping and
> writebacks going on?

The above is just for providing command flags to user side, so that the
user side can understand/handle the request better.

prtrl(PR_SET_IO_FLUSHER) has been merged for avoiding the deadlock.

> 
> What are restriction on ubdsrv? What happens when it needs to allocate memory, or is
> swapped out?

Yeah, ubd_copy_pages() needs to pin pages for copying data between
user VM and block request pages, and get_user_pages may run out of pages.
But I think forward progress can still be provided by reserving one VM buffer
with single page locked.

> Have mm people seen this?

I remembered that the early RFC with related discussion is CCed to mm
list, and all follow-up are CC to linux-kernel.

Not one big deal, will Cc mm list in the future post.


thanks,
Ming