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;
MIME-Version: 1.0
References: <26aba1b5-8393-a20a-3ce9-f82425673f4d@kernel.dk>
In-Reply-To: <26aba1b5-8393-a20a-3ce9-f82425673f4d@kernel.dk>
From:   Linus Torvalds <torvalds@linux-foundation.org>
Date:   Sat, 6 May 2023 08:27:39 -0700
Message-ID: <CAHk-=wj=LwLDYrjHpMM+QnE2T+u4P9-UXhXGkAMXiyfGjGnNEA@mail.gmail.com>
Subject: Re: [GIT PULL] Pipe FMODE_NOWAIT support
To:     Jens Axboe <axboe@kernel.dk>
Cc:     "linux-fsdevel@vger.kernel.org" <linux-fsdevel@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Sat, May 6, 2023 at 3:33=E2=80=AFAM Jens Axboe <axboe@kernel.dk> wrote:
>
> Here's the revised edition of the FMODE_NOWAIT support for pipes, in
> which we just flag it as such supporting FMODE_NOWAIT unconditionally,
> but clear it if we ever end up using splice/vmsplice on the pipe. The
> pipe read/write side is perfectly fine for nonblocking IO, however
> splice and vmsplice can potentially wait for IO with the pipe lock held.

Ok, pulled.

It does strike me that one of the (few) users is the io_uring
__io_file_supports_nowait() thing, and that thing is *disgusting*.

Wouldn't it be much nicer if FMODE_NOWAIT ended up working
automatically on a block file descriptor too? You did all this "make
pipes set FMODE_NOWAIT", but then that io_uring code does

        if (S_ISBLK(mode)) {
                if (IS_ENABLED(CONFIG_BLOCK) &&
                    io_bdev_nowait(I_BDEV(file->f_mapping->host)))
                        return true;
                return false;
        }

rather than just rely on FMODE_NOWAIT for block devices too...

And it's a bit odd in other ways, because the kiocb code for
RWF_NOWAIT - which is the other user of FMODE_NOWAIT - does *not* seem
to do any of those bdev games. So that other user does seem to just
rely on the file mode bit having been set up by open.

In fact, I see 'blkdev_open()' doing

        filp->f_mode |=3D FMODE_NOWAIT | FMODE_BUF_RASYNC;

so I really don't see why __io_file_supports_nowait() then does that
special check for S_ISBLK().

Something is very rotten in the state of Denmark.

But that's all independent of this pipe side, which now looks fine to me.

              Linus