Re: [PATCH v2 1/3] fs: fork splice_file_range() from do_splice_direct()

From: Amir Goldstein <amir73il@gmail.com>
To: Christoph Hellwig <hch@lst.de>, Jeff Layton <jlayton@kernel.org>
Cc: Christian Brauner <brauner@kernel.org>,
	Josef Bacik <josef@toxicpanda.com>, Jan Kara <jack@suse.cz>,
	 David Howells <dhowells@redhat.com>,
	Jens Axboe <axboe@kernel.dk>,  Miklos Szeredi <miklos@szeredi.hu>,
	Al Viro <viro@zeniv.linux.org.uk>,
	 linux-fsdevel@vger.kernel.org
Subject: Re: [PATCH v2 1/3] fs: fork splice_file_range() from do_splice_direct()
Date: Mon, 4 Dec 2023 15:29:43 +0200	[thread overview]
Message-ID: <CAOQ4uxjZAjJSR-AUH+UQM3AX9Ota3DVxygFSVkpEQdxK15n_qQ@mail.gmail.com> (raw)
In-Reply-To: <20231204083849.GC32438@lst.de>

On Mon, Dec 4, 2023 at 10:38 AM Christoph Hellwig <hch@lst.de> wrote:
>
> On Mon, Dec 04, 2023 at 09:37:33AM +0100, Christoph Hellwig wrote:
> > >             put_rd_wr_caps(src_ci, src_got, dst_ci, dst_got);
> > > -           ret = do_splice_direct(src_file, &src_off, dst_file,
> > > -                                  &dst_off, src_objlen, flags);
> > > +           ret = splice_file_range(src_file, &src_off, dst_file, &dst_off,
> > > +                                   src_objlen);
> >
> > Shouldn't ceph be switched to use generic_copy_file_range?
> > That does the capping of the size which we want, and doesn't update
> > the file offsets, which would require recalculation in the ceph code.
> >

IDK. I did not want to change the logic of the ceph code.
I am not sure that we must impose MAX_RW_COUNT limit on ceph,
although, i_layout.object_size may already be limited? Jeff?

> > But this could avoid another exported API as splice_file_range could
> > simply be folded into generic_copy_file_range which should reduce
> > confusion.  And splice really is a mess for so many different layers
> > of the onion being exposed.  I've been wanting to reduce some of that
> > for a while but haven't found a really nice way yet.
>
> (and generic_copy_file_range really should be renamed to
> splice_copy_file_range and moved to splice.c)

That depends if we are keeping two helpers.
One with a cap of MAX_RW_COUNT and one without.
If we are going to keep two helpers, I'd rather keep things as they are.
If one helper, then I personally prefer splice_file_range() over
splice_copy_file_range() and other reviewers (Jan) liked this
name as well.

Thanks,
Amir.