Re: [PATCH v4 0/3] random: convert to using iters, for Al Viro

[Al Viro <viro@zeniv.linux.org.uk>]

On Fri, May 20, 2022 at 09:53:30AM -0600, Jens Axboe wrote:
> On 5/20/22 9:47 AM, Al Viro wrote:
> > On Fri, May 20, 2022 at 09:34:46AM -0600, Jens Axboe wrote:
> > 
> >> I'm very sure, otherwise we're just accepting that we're breaking real
> >> world applications.
> > 
> > "Breaking" as in "it used to work with earlier kernels, doesn't work with
> > recent ones"?  Details, please...
> 
> Yes, as in exactly that. This is what drove this addition of
> ->read_iter() for urandom. See commit:
> 
> ommit 36e2c7421f02a22f71c9283e55fdb672a9eb58e7
> Author: Christoph Hellwig <hch@lst.de>
> Date:   Thu Sep 3 16:22:34 2020 +0200
> 
>     fs: don't allow splice read/write without explicit ops
> 
> related to the set_fs() changes, and now go look for any commit that
> has:
> 
> Fixes: 36e2c7421f02 ("fs: don't allow splice read/write without explicit ops")
> 
> in it and see that this isn't an isolated incident at all.
> 
> tldr - splice from /dev/urandom used to work, and I recently got a
> report internally on an application that broke on upgrade from 5.6 to
> 5.12 exactly because it now just just -EINVAL's instead.

IIRC, Linus' position at the time had been along the lines of
"splice is not so good ABI anyway, so let's do it and fix up
the places that do get real-world complaints once such appear".
So /dev/urandom is one such place...