Re: [PATCH v3 8/9] x86: use __uaccess_begin_nospec and ASM_IFENCE in get_user paths

[Christoph Hellwig]

On Wed, Jan 17, 2018 at 11:26:08AM -0800, Linus Torvalds wrote:
> But there are about ~100 set_fs() calls in generic code, and some of
> those really are pretty fundamental. Doing things like "kernel_read()"
> without set_fs() is basically impossible.

Not if we move to iov_iter or iov_iter-like behavior for all reads
and writes.  There is an issue with how vectored writes are handles
in plain read/write vs read_iter/write_iter inherited from readv/writev,
but that's nothing a flag, or a second set of methods with the
same signature.

But there are more annoying things, most notable in-kernel ioctls
calls.  We have quite a few of them, and while many are just utterly
stupid and can be replaced with direct function calls or new methods
(I've done quite a few conversions of those) some might be left.
Something like iov_iter might be the answer again.

Then we have things like probe_kernel_read/probe_kernel_write which
abuse the exception handling in get/put user.  But with a little
arch helper we don't strictly need get_fs/set_fs for that.