Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:4 as permitted sender) client-ip=2620:137:e000::3:4;
MIME-Version: 1.0
References: <20231130141624.3338942-4-amir73il@gmail.com> <20231205001620.4566-1-spasswolf@web.de>
In-Reply-To: <20231205001620.4566-1-spasswolf@web.de>
From:   Amir Goldstein <amir73il@gmail.com>
Date:   Tue, 5 Dec 2023 05:45:57 +0200
Message-ID: <CAOQ4uxiw8a+zh-x2a+A+EEZOFj1KYrBQucCvDv6s9w0XeDW-ZA@mail.gmail.com>
Subject: Re: [PATCH] fs: read_write: make default in vfs_copy_file_range() reachable
To:     Bert Karwatzki <spasswolf@web.de>,
        Christian Brauner <brauner@kernel.org>
Cc:     Jan Kara <jack@suse.cz>, axboe@kernel.dk, dhowells@redhat.com,
        hch@lst.de, jlayton@kernel.org, josef@toxicpanda.com,
        linux-fsdevel@vger.kernel.org, miklos@szeredi.hu,
        viro@zeniv.linux.org.uk, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Tue, Dec 5, 2023 at 2:16=E2=80=AFAM Bert Karwatzki <spasswolf@web.de> wr=
ote:
>
> If vfs_copy_file_range() is called with (flags & COPY_FILE_SPLICE =3D=3D =
0)
> and both file_out->f_op->copy_file_range and file_in->f_op->remap_file_ra=
nge
> are NULL, too, the default call to do_splice_direct() cannot be reached.
> This patch adds an else clause to make the default reachable in all
> cases.
>
> Signed-off-by: Bert Karwatzki <spasswolf@web.de>

Hi Bert,

Thank you for testing and reporting this so early!!

I would edit the commit message differently, but anyway, I think that
the fix should be folded into commit 05ee2d85cd4a ("fs: use
do_splice_direct() for nfsd/ksmbd server-side-copy").

Since I end up making a mistake every time I touch this code,
I also added a small edit to your patch below, that should make the logic
more clear to readers. Hopefully, that will help me avoid making a mistake
the next time I touch this code...

Would you mind testing my revised fix, so we can add:
  Tested-by: Bert Karwatzki <spasswolf@web.de>
when folding it into the original patch?

Thanks,
Amir.

> ---
>  fs/read_write.c | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/fs/read_write.c b/fs/read_write.c
> index e0c2c1b5962b..3599c54bd26d 100644
> --- a/fs/read_write.c
> +++ b/fs/read_write.c
> @@ -1554,6 +1554,8 @@ ssize_t vfs_copy_file_range(struct file *file_in, l=
off_t pos_in,
>                 /* fallback to splice */
>                 if (ret <=3D 0)
>                         splice =3D true;
> +       } else {

This is logically correct because of the earlier "same sb" check in
generic_copy_file_checks(), but we better spell out the logic here as well:

+            } else if (file_inode(file_in)->i_sb =3D=3D
file_inode(file_out)->i_sb) {
+                    /* Fallback to splice for same sb copy for
backward compat */

> +               splice =3D true;
>         }
>
>         file_end_write(file_out);
> --
> 2.39.2
>
> Since linux-next-20231204 I noticed that it was impossible to start the
> game Path of Exile (using the steam client). I bisected the error to
> commit 05ee2d85cd4ace5cd37dc24132e3fd7f5142ebef. Reverting this commit
> in linux-next-20231204 made the game start again and after inserting
> printks into vfs_copy_file_range() I found that steam (via python3)
> calls this function with (flags & COPY_FILE_SPLICE =3D=3D 0),
> file_out->f_op->copy_file_range =3D=3D NULL and
> file_in->f_op->remap_file_range =3D=3D NULL so the default is never reach=
ed.
> This patch adds a catch all else clause so the default is reached in
> all cases. This patch fixes the describe issue with steam and Path of
> Exile.
>
> Bert Karwatzki