Re: [PATCH v4 1/3] random: convert to using fops->read_iter()

From: "Jason A. Donenfeld" <Jason@zx2c4.com>
To: Jens Axboe <axboe@kernel.dk>
Cc: Theodore Ts'o <tytso@mit.edu>, Christoph Hellwig <hch@lst.de>,
	LKML <linux-kernel@vger.kernel.org>,
	Al Viro <viro@zeniv.linux.org.uk>
Subject: Re: [PATCH v4 1/3] random: convert to using fops->read_iter()
Date: Fri, 20 May 2022 16:39:41 +0200	[thread overview]
Message-ID: <YoeoLWTQ29bZCOFI@zx2c4.com> (raw)
In-Reply-To: <35a5ff9f-79d7-feb5-dfa7-a775b9352416@kernel.dk>

Hi Jens,

On Fri, May 20, 2022 at 08:36:17AM -0600, Jens Axboe wrote:
> On 5/20/22 7:37 AM, Jason A. Donenfeld wrote:
> > On Fri, May 20, 2022 at 11:44:57AM +0200, Jason A. Donenfeld wrote:
> >>  const struct file_operations urandom_fops = {
> >> -	.read = urandom_read,
> >> +	.read_iter = urandom_read_iter,
> > 
> > One thing I noticed is that drivers/char/mem.c has both the .read and
> > the .read_iter functions for /dev/zero and /dev/null and such. I wonder
> > if the .read ones can be removed?
> 
> I'm not sure if we have a clear "always use this if available" set of
> rules for this. Ideally we'd want it to be:
> 
> 1) Use ->read_iter() if available
> 2) If not, use ->read()
> 
> Might require a bit of auditing to ensure that's the case, and if we
> can say that it is, then we could clean that up too.

The only case I found where it wasn't in that order was:
https://lore.kernel.org/lkml/20220520135103.166972-1-Jason@zx2c4.com/

Jason