Re: [PATCH 0/5] bio: Direct IO: convert to pin_user_pages_fast()

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

On Mon, Aug 24, 2020 at 07:07:02PM -0700, John Hubbard wrote:
> On 8/24/20 6:54 PM, Al Viro wrote:
> > On Fri, Aug 21, 2020 at 09:20:54PM -0700, John Hubbard wrote:
> > 
> > > Direct IO behavior:
> > > 
> > >      ITER_IOVEC:
> > >          pin_user_pages_fast();
> > >          break;
> > > 
> > >      ITER_KVEC:    // already elevated page refcount, leave alone
> > >      ITER_BVEC:    // already elevated page refcount, leave alone
> > >      ITER_PIPE:    // just, no :)
> > 
> > Why?  What's wrong with splice to O_DIRECT file?
> > 
> 
> Oh! I'll take a look. Is this the fs/splice.c stuff?  I ruled this out early
> mainly based on Christoph's comment in [1] ("ITER_PIPE is rejected іn the
> direct I/O path"), but if it's supportable then I'll hook it up.

; cat >a.c <<'EOF'
#define _GNU_SOURCE
#include <fcntl.h>
#include <unistd.h>
#include <stdlib.h>

int main()
{
        int fd = open("./a.out", O_DIRECT);
        splice(fd, NULL, 1, NULL, 4096, 0);
	return 0;
}
EOF
; cc a.c
; ./a.out | wc -c
4096

and you just had ->read_iter() called with ITER_PIPE destination.