Re: [PATCH] Convert filldir[64]() from __put_user() to unsafe_put_user()

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

On Sun, Oct 06, 2019 at 07:06:19PM -0700, Linus Torvalds wrote:
> On Sun, Oct 6, 2019 at 6:24 PM Al Viro <viro@zeniv.linux.org.uk> wrote:
> >
> > Ugh...  I wonder if it would be better to lift STAC/CLAC out of
> > raw_copy_to_user(), rather than trying to reinvent its guts
> > in readdir.c...
> 
> Yeah, I suspect that's the best option.
> 
> Do something like
> 
>  - lift STAC/CLAC out of raw_copy_to_user
> 
>  - rename it to unsafe_copy_to_user
> 
>  - create a new raw_copy_to_user that is just unsafe_copy_to_user()
> with the STAC/CLAC around it.
> 
> and the end result would actually be cleanert than what we have now
> (which duplicates that STAC/CLAC for each size case etc).
> 
> And then for the "architecture doesn't have user_access_begin/end()"
> fallback case, we just do
> 
>    #define unsafe_copy_to_user raw_copy_to_user

Callers of raw_copy_to_user():
arch/hexagon/mm/uaccess.c:27:           uncleared = raw_copy_to_user(dest, &empty_zero_page, PAGE_SIZE);
arch/hexagon/mm/uaccess.c:34:           count = raw_copy_to_user(dest, &empty_zero_page, count);
arch/powerpc/kvm/book3s_64_mmu_radix.c:68:              ret = raw_copy_to_user(to, from, n);
arch/s390/include/asm/uaccess.h:150:    size = raw_copy_to_user(ptr, x, size);
include/asm-generic/uaccess.h:145:      return unlikely(raw_copy_to_user(ptr, x, size)) ? -EFAULT : 0;
include/linux/uaccess.h:93:     return raw_copy_to_user(to, from, n);
include/linux/uaccess.h:102:    return raw_copy_to_user(to, from, n);
include/linux/uaccess.h:131:            n = raw_copy_to_user(to, from, n);
lib/iov_iter.c:142:             n = raw_copy_to_user(to, from, n);
lib/usercopy.c:28:              n = raw_copy_to_user(to, from, n);


Out of those, only __copy_to_user_inatomic(), __copy_to_user(),
_copy_to_user() and iov_iter.c:copyout() can be called on
any architecture.

The last two should just do user_access_begin()/user_access_end()
instead of access_ok().  __copy_to_user_inatomic() has very few callers as well:

arch/mips/kernel/unaligned.c:1307:                      res = __copy_to_user_inatomic(addr, fpr, sizeof(*fpr));
drivers/gpu/drm/i915/i915_gem.c:345:    unwritten = __copy_to_user_inatomic(user_data,
lib/test_kasan.c:471:   unused = __copy_to_user_inatomic(usermem, kmem, size + 1);
mm/maccess.c:98:        ret = __copy_to_user_inatomic((__force void __user *)dst, src, size);

So few, in fact, that I wonder if we want to keep it at all; the only
thing stopping me from "let's remove it" is that I don't understand
the i915 side of things.  Where does it do an equivalent of access_ok()?

And mm/maccess.c one is __probe_kernel_write(), so presumably we don't
want stac/clac there at all...

So do we want to bother with separation between raw_copy_to_user() and
unsafe_copy_to_user()?  After all, __copy_to_user() also has only few
callers, most of them in arch/*

I'll take a look into that tomorrow - half-asleep right now...