On Sun, Apr 29, 2012 at 11:29 AM, Linus Torvalds
<torvalds@linux-foundation.org> wrote:
>
> But I think I have another approach. We can make the *writer* be the
> only one that cares about the packetized nature, and if it's a packet
> write, it will set a PIPE_BUF_PACKET flag in the pipe buffer. The
> reader can react to that. I think that actually has the potential to
> make the code a bit prettier too..

Patch attached.

This keeps Alan's idea of using O_DIRECT, but only makes it matter for
the writer - because now the "packet" thing is a per-pipe-buffer state
(we already had per-pipe-buffer flags, so this is not anything new).
So when you *write* using a O_DIRECT pipe, it will create packetized
buffers, and the reader just sees that directly.

So this should "just work", and doesn't need any extra flags in the
pipe_inode_info. And the user space interface remains the same: you
can create these packet pipes with "pipe2(fd, O_DIRECT)" if you want
to, and my test-program gives exactly the same output.

                      Linus