Re: splice() from /dev/zero to a pipe does not work (5.9+)

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

On Fri, May 07, 2021 at 12:29:44PM -0700, Linus Torvalds wrote:
> On Fri, May 7, 2021 at 12:17 PM Al Viro <viro@zeniv.linux.org.uk> wrote:
> >
> > Umm...  That would do wonders to anything that used to do
> > copy_to_user()/clear_user()/copy_to_user() and got converted
> > to copy_to_iter()/iov_iter_zero()/copy_to_iter()...
> 
> I didn't mean for iov_iter_zero doing this - only splice_read_zero().
> 
> > Are you sure we can shove zero page into pipe, anyway?
> > IIRC, get_page()/put_page() on that is not allowed,
> 
> That's what the
> 
>     buf->ops = &zero_pipe_buf_ops;
> 
> is for. The zero_pipe_buf_ops would have empty get and release
> functions, and a 'steal' function that always returns false.
> 
> That's how the pipe pages are supposed to work: there are people who
> put non-page data (ie things like skbuff allocations etc) into a
> splice pipe buffer. It's why we have those "ops" pointers.

Supposed to - sure, but I'd like to verify that they actually do work
that way before we go there.  Let me RTFS a bit...