On Tue, Oct 15, 2019 at 11:08 AM Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
>
> Another question: right now we have
> if (!access_ok(uaddr, sizeof(u32)))
> return -EFAULT;
>
> ret = arch_futex_atomic_op_inuser(op, oparg, &oldval, uaddr);
> if (ret)
> return ret;
> in kernel/futex.c. Would there be any objections to moving access_ok()
> inside the instances and moving pagefault_disable()/pagefault_enable() outside?
I think we should remove all the "atomic" versions, and just make the
rule be that if you want atomic, you surround it with
pagefault_disable()/pagefault_enable().
That covers not just the futex ops (where "atomic" is actually
somewhat ambiguous - the ops themselves are atomic too, so the naming
might stay, although arguably the "futex" part makes that pointless
too), but also copy_to_user_inatomic() and the powerpc version of
__get_user_inatomic().
So we'd aim to get rid of all the "inatomic" ones entirely.
Same ultimately probably goes for the NMI versions. We should just
make it be a rule that we can use all of the user access functions
with pagefault_{dis,en}able() around them, and they'll be "safe" to
use in atomic context.
One issue with the NMI versions is that they actually want to avoid
the current value of set_fs(). So copy_from_user_nmi() (at least on
x86) is special in that it does
if (__range_not_ok(from, n, TASK_SIZE))
return n;
instead of access_ok() because of that issue.
NMI also has some other issues (nmi_uaccess_okay() on x86, at least),
but those *probably* could be handled at page fault time instead.
Anyway, NMI is so special that I'd suggest leaving it for later, but
the non-NMI atomic accesses I would suggest you clean up at the same
time.
I think the *only* reason we have the "inatomic()" versions is that
the regular ones do that "might_fault()" testing unconditionally, and
might_fault() _used_ to be just a might_sleep() - so it's not about
functionality per se, it's about "we have this sanity check that we
need to undo".
We've already made "might_fault()" look at pagefault_disabled(), so I
think a lot of the reasons for inatomic are entirely historical.
Linus
