On Sun, Jul 26, 2020 at 06:34:22PM +0200, Christoph Hellwig wrote:
> On Sun, Jul 26, 2020 at 05:32:14PM +0100, Al Viro wrote:
> > > +static int compat_copy_fs_qfilestat(struct compat_fs_qfilestat __user
> > > *to,
> > > + struct fs_qfilestat *from)
> > > +{
> > > + if (copy_to_user(to, from, sizeof(*to)) ||
> > > + put_user(from->qfs_nextents, &to->qfs_nextents))
> > > + return -EFAULT;
> > > + return 0;
> > > +}
> >
> > do we have any need of that put_user()? Note that you don't even call
> > that thing unless compat_need_64bit_alignment_fixup() is true. And AFAICS
> > all such cases are little-endian...
>
> The main reason it is there is to preserve the previous semantics.
> And no, I don't think we actually need it on x86. But what if some
> poor souls adds a BE version that needs this? E.g. arm oabi has similar
> weird alignment, and now imagine someone adding arm64 compat code for
> that..
I'd probably add /* just in case some poor sod fucks up the same way for
big-endian biarch */
next to that put_user(), then ;-)
