Date:   Wed, 8 Feb 2023 08:19:56 -0800
From:   Christoph Hellwig <hch@infradead.org>
To:     David Howells <dhowells@redhat.com>
Cc:     Christoph Hellwig <hch@infradead.org>,
        Jens Axboe <axboe@kernel.dk>,
        Al Viro <viro@zeniv.linux.org.uk>,
        Matthew Wilcox <willy@infradead.org>, Jan Kara <jack@suse.cz>,
        Jeff Layton <jlayton@kernel.org>,
        David Hildenbrand <david@redhat.com>,
        Jason Gunthorpe <jgg@nvidia.com>,
        Logan Gunthorpe <logang@deltatee.com>,
        Hillf Danton <hdanton@sina.com>, linux-fsdevel@vger.kernel.org,
        linux-block@vger.kernel.org, linux-kernel@vger.kernel.org,
        linux-mm@kvack.org,
        syzbot+a440341a59e3b7142895@syzkaller.appspotmail.com,
        Christoph Hellwig <hch@lst.de>,
        John Hubbard <jhubbard@nvidia.com>
Subject: Re: [PATCH v12 01/10] vfs, iomap: Fix generic_file_splice_read() to
 avoid reversion of ITER_PIPE
Message-ID: <Y+PLrOM05FMCiTIg@infradead.org>
References: <Y+MydH2HZ7ihITli@infradead.org>
 <20230207171305.3716974-1-dhowells@redhat.com>
 <20230207171305.3716974-2-dhowells@redhat.com>
 <176199.1675872591@warthog.procyon.org.uk>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <176199.1675872591@warthog.procyon.org.uk>
Precedence: bulk

On Wed, Feb 08, 2023 at 04:09:51PM +0000, David Howells wrote:
> How about one of two different solutions?
> 
>  (1) Repurpose the function I proposed for generic_file_splice_read() but only
>      for splicing from O_DIRECT files; reading from non-O_DIRECT files would
>      use an ITER_PIPE as upstream.

Given the amounts of problems we had with O_DIRECT vs splice, and the
fact that even doing this is a bit pointless that seems sensible to me.

>      for splicing from O_DIRECT files, as (1), but also replace the splice
>      from a buffered file with something like the patch below.  This uses
>      filemap_get_pages() to do the reading and to get a bunch of folios from
>      the pagecache that we can then splice into the pipe directly.

I defintively like the idea of killing ITER_PIPE.  Isn't the 16
folios in a folio tree often much less than what we could fit into
a single pipe buf?  Unless you have a file system that can use
huge folios for buffered I/O and actually does this might significantly
limit performance.

With that in mind I'll try to find some time to review your actual
patch, but I'm a little busy at the moment.