On Wed, May 10, 2017 at 03:11:18AM +0100, Al Viro wrote:

On Tue, May 09, 2017 at 09:03:22AM -0700, Christoph Hellwig wrote:

quoted

On Tue, May 09, 2017 at 06:02:50AM -0700, Christoph Hellwig wrote:

quoted

On Tue, May 09, 2017 at 06:00:01AM -0700, Andy Lutomirski wrote:

quoted

fs/splice.c has some, ahem, interesting uses that have been the source
of nasty exploits in the past.  Converting them to use iov_iter
properly would be really, really nice.  Christoph, I don't suppose
you'd like to do that?

I can take care of all the fs code including this one.

Oh?

quoted

I spent the afternoon hacking up where I'd like this to head.  It's
completely untested as of now:

	http://git.infradead.org/users/hch/vfs.git/shortlog/refs/heads/setfs-elimination

And just what happens to driver that has no ->read_iter()?  Unless I'm
seriously misreading that, NAK with extreme prejudice.

FWIW, some parts of that queue are obviously sane; it's the conversions of
kernel_write() and friends to ->read_iter/->write_iter() that are non-starters.
That stuff is used in too many situations; we can't guarantee that all of
them will be for files that have those.

As for default_file_splice_read(), I seriously suspect that with your change
we could as well just make it return -EINVAL and be done with that; places
that have ->read_iter() tend to have explicit ->splice_read() and it looks
like the ones that do not should simply use generic_file_read_iter().
I hadn't checked that, but there's not a lot of those:

arch/s390/hypfs/inode.c:437:    .read_iter      = hypfs_read_iter,
drivers/char/mem.c:798: .read_iter      = read_iter_null,
drivers/char/mem.c:813: .read_iter      = read_iter_zero,
drivers/char/mem.c:824: .read_iter      = read_iter_zero,
drivers/char/raw.c:286: .read_iter      = blkdev_read_iter,
drivers/net/tap.c:1134: .read_iter      = tap_read_iter,
drivers/net/tun.c:2423: .read_iter  = tun_chr_read_iter,
drivers/usb/gadget/function/f_fs.c:1255:        .read_iter =    ffs_epfile_read_iter,
drivers/usb/gadget/legacy/inode.c:703:  .read_iter =    ep_read_iter,
drivers/vhost/net.c:1252:       .read_iter      = vhost_net_chr_read_iter,
fs/9p/vfs_file.c:641:   .read_iter = generic_file_read_iter,
fs/9p/vfs_file.c:652:   .read_iter = generic_file_read_iter,
fs/9p/vfs_file.c:664:   .read_iter = v9fs_file_read_iter,
fs/9p/vfs_file.c:675:   .read_iter = v9fs_file_read_iter,
fs/9p/vfs_file.c:687:   .read_iter = v9fs_mmap_file_read_iter,
fs/9p/vfs_file.c:698:   .read_iter = v9fs_mmap_file_read_iter,
fs/fuse/cuse.c:180:     .read_iter              = cuse_read_iter,
fs/fuse/file.c:3015:    .read_iter      = fuse_direct_read_iter,
fs/hugetlbfs/inode.c:980:       .read_iter              = hugetlbfs_read_iter,
fs/ncpfs/file.c:248:    .read_iter      = ncp_file_read_iter,
fs/orangefs/file.c:742: .read_iter      = orangefs_file_read_iter,
fs/pipe.c:1011: .read_iter      = pipe_read,
sound/core/pcm_native.c:3696:           .read_iter =            snd_pcm_readv,

is the full list and I'm fairly certain that most of them will work with
generic_file_splice_read() just fine.  drivers/char definitely will, so
will ncpfs/orangefs/hugetlbfs/most of 9p ones (two of the latter might
need some care in p9_client_read(), but that should be doable easily enough).
pipe is irrelevant (->splice_read() won't be called for those).  fuse ones
should be doable, but that might take a bit more infrastructure work in
lib/iov_iter.c.  vhost, gadgetfs, tun/tap - no idea at the moment.