Date:   Tue, 7 Feb 2023 21:26:12 -0800
From:   Christoph Hellwig <hch@infradead.org>
To:     David Howells <dhowells@redhat.com>
Cc:     Jens Axboe <axboe@kernel.dk>, Al Viro <viro@zeniv.linux.org.uk>,
        Christoph Hellwig <hch@infradead.org>,
        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+MydH2HZ7ihITli@infradead.org>
References: <20230207171305.3716974-1-dhowells@redhat.com>
 <20230207171305.3716974-2-dhowells@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230207171305.3716974-2-dhowells@redhat.com>
Precedence: bulk

Subject nitpick:  this does not ouch iomap at all.

> Fix this by replacing the use of an ITER_PIPE iterator with an ITER_BVEC
> iterator for which reversion won't free the buffers.  Bulk allocate all the
> buffers we think we're going to use in advance, do the read synchronously
> and only then trim the buffer down.  The pages we did use get pushed into
> the pipe.
> 
> This is more efficient by virtue of doing a bulk page allocation, but
> slightly less efficient by ignoring any partial page in the pipe.

For the usual case of a buffered read into the iter, this completely
changes the semantics:

 - before the pagecache pages just grew a new reference and were
   added to the pipe buffer, and I/O was done without an extra
   copy
 - with this patch you always allocate an extra set of pages for
   the pipe and copy to it

So I very much doubt this makes anything more efficient, and I don't
think we can just do it.

We'll have to fix reverting of pipe buffers, just as I already pointed
out in your cifs series that tries to play the same game.