Re: [PATCH v1 1/4] lib: introduce copy_struct_from_user() helper

[Al Viro <viro at zeniv dot linux dot org dot uk>]

On Wed, Sep 25, 2019 at 10:48:31AM -0700, Linus Torvalds wrote:
> On Wed, Sep 25, 2019 at 10:21 AM Aleksa Sarai <cyphar@cyphar.com> wrote:
> >
> > Just to make sure I understand, the following diff would this solve the
> > problem? If so, I'll apply it, and re-send in a few hours.
> 
> Actually, looking at it more, it's still buggy.
> 
> That final "size smaller than unsigned long" doesn't correctly handle
> the case of (say) a single byte in the middle of a 8-byte word.
> 
> So you need to do something like this:
> 
>     int is_zeroed_user(const void __user *from, size_t size)
>   {
>         unsigned long val, mask, align;
> 
>         if (unlikely(!size))
>                 return true;
> 
>         if (!user_access_begin(from, size))
>                 return -EFAULT;
> 
>         align = (uintptr_t) from % sizeof(unsigned long);
>         from -= align;
>         size += align;
> 
>         mask = ~aligned_byte_mask(align);
> 
>         while (size >= sizeof(unsigned long)) {
>                 unsafe_get_user(val, (unsigned long __user *) from, err_fault);
>                 val &= mask;
>                 if (unlikely(val))
>                         goto done;
>                 mask = ~0ul;
>                 from += sizeof(unsigned long);
>                 size -= sizeof(unsigned long);
>         }
> 
>         if (size) {
>                 /* (@from + @size) is unaligned. */
>                 unsafe_get_user(val, (unsigned long __user *) from, err_fault);
>                 mask &= aligned_byte_mask(size);
>                 val &= mask;
>         }

IMO it's better to lift reading the first word out of the loop, like this:

	align = (uintptr_t) from % sizeof(unsigned long);
	from -= align;

	unsafe_get_user(val, (unsigned long __user *) from, err_fault);
	if (align) {
		size += align;
		val &= ~aligned_byte_mask(align);
	}

	while (size > sizeof(unsigned long)) {
		if (unlikely(val))
			goto done;
		from += sizeof(unsigned long);
		size -= sizeof(unsigned long);
		unsafe_get_user(val, (unsigned long __user *) from, err_fault);
	}

	if (size != size(unsigned long))
		val &= aligned_byte_mask(size);
done:

Do you see any problems with that variant?