Re: [PATCH 32/32] aio: implement io_pgetevents

[Christoph Hellwig]

On Wed, Jan 17, 2018 at 04:27:21AM +0000, Al Viro wrote:
> On Tue, Jan 16, 2018 at 07:41:24PM -0500, Jeff Moyer wrote:
> >  	if (sigmask) {
> > -		if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
> > +		if (!access_ok(VERIFY_READ, sigmask,
> > +			       sizeof(void *) + sizeof(size_t)) ||
> > +		    __get_user(up, (sigset_t __user * __user *)sigmask) ||
> > +		    __get_user(sigsetsize,
> > +			       (size_t __user *)(sigmask + sizeof(void *))))
> >  			return -EFAULT;
> 
> How about copy_from_user() on a struct?  Making eyes bleed is fun, but
> people tend to get annoyed when you do it to them...

Above is the copy & paste version from pselect.  I've got both copy_from_user
and that horrible version in my tree, and if we really need this awfull
calling convention copy_from_user certainly is much better.  pselect
also should be switched to explicit struct + copy_from_user while
we're at it.  In fact glibc defines a struct for the userland version
to start with.