Re: [PATCH 1/4] fs/splice: enhance direct pipe & splice for moving pages in kernel

[Ming Lei <ming.lei@redhat.com>]

On Mon, Feb 13, 2023 at 12:04:27PM -0800, Linus Torvalds wrote:
> On Sat, Feb 11, 2023 at 5:39 PM Ming Lei <ming.lei@redhat.com> wrote:
> >
> > >
> > >  (a) what's the point of MAY_READ? A non-readable page sounds insane
> > > and wrong. All sinks expect to be able to read.
> >
> > For example, it is one page which needs sink end to fill data, so
> > we needn't to zero it in source end every time, just for avoiding
> > leak kernel data if (unexpected)sink end simply tried to read from
> > the spliced page instead of writing data to page.
> 
> I still don't understand.
> 
> A sink *reads* the data. It doesn't write the data.
> 
> There's no point trying to deal with "if unexpectedly doing crazy
> things". If a sink writes the data, the sinkm is so unbelievably buggy
> that it's not even funny.
> 
> And having two flags that you then say "have to be used together" is pointless.

Actually I think it is fine to use the pipe buffer flags separately,
if MAY_READ/MAY_WRITE is set in source end, the sink side need to respect
it. All current in-tree source end actually implies both MAY_READ & MAY_WRITE.

> It's not two different flags if you can't use them separately!
> 
> So I think your explanations are anything *but* explaining what you
> want. They are just strange and not sensible.
> 
> Please explain to me in small words and simple sentences what it is
> you want. And no, if the explanation is "the sink wants to write to
> the buffer", then that's not an explanation, it's just insanity.
> 
> We *used* to have the concept of "gifting" the buffer explicitly to
> the sink, so that the sink could - instead of reading from it - decide
> to just use the whole buffer as-is long term. The idea was that tthe
> buffer woudl literally be moved from the source to the destination,
> ownership and all.
> 
> But if that's what you want, then it's not about "sink writes". It's
> literally about the splice() wanting to move not just the data, but
> the whole ownership of the buffer.

Yeah, it is actually transferring the buffer ownership, and looks
SPLICE_F_GIFT is exactly the case, but the driver side needs to set
QUEUE_FLAG_STABLE_WRITES for avoiding writeback to touch these pages.

Follows the idea:

file(devices(such as, fuse, ublk), produce pipe buffer) -> direct pipe -> file(consume the pipe buffer)

The 'consume' could be READ or WRITE.

So once SPLICE_F_GIFT is set from source side, the two buffer flags
aren't needed any more, right?

Please see the detailed explanation & use case in following link:

https://lore.kernel.org/linux-block/409656a0-7db5-d87c-3bb2-c05ff7af89af@kernel.dk/T/#m237e5973571b3d85df9fa519cf2c9762440009ba



Thanks,
Ming