On Fri, 22 Aug 2025 09:46:37 -0400
Linus Torvalds [off-list ref] wrote:

On Fri, 22 Aug 2025 at 05:58, Christophe Leroy
[off-list ref] wrote:

quoted

The results of "access_ok()" can be mis-speculated.  The result is that
you can end speculatively:

        if (access_ok(from, size))
                // Right here

I actually think that we should probably just make access_ok() itself do this.

You'd need to re-introduce the read/write parameter.
And you'd want it to be compile time.
Although going through the code changing them to read_access_ok()
and write_access_ok() would probably leave you with a lot fewer calls.

We don't have *that* many users since we have been de-emphasizing the
"check ahead of time" model, and any that are performance-critical can
these days be turned into masked addresses.

Or aim to allocate a guard page on all archs, support 'masked' access
on all of them, and then just delete access_ok().
That'll make it look less ugly.
Perhaps not this week though :-)

	David

As it is, now we're in the situation that careful places - like
_inline_copy_from_user(), and with your patch  copy_from_user_iter() -
do maybe wethis by hand and are ugly as a result, and lazy and
probably incorrect places don't do it at all.

That said, I don't object to this patch and maybe we should do that
access_ok() change later and independently of any powerpc work.

                 Linus