Re: [RFC PATCH for-next 4/4] epoll: epoll() syscall definition

[Nathaniel Yazdani <n1ght.4nd.d4y@gmail.com>]

On Tue, 2014-02-25 at 10:21 +0000, Eric Wong wrote:
> Nathaniel Yazdani <n1ght.4nd.d4y@gmail.com> wrote:
> > + * stores triggered eventpoll entries in the 'out' array. The input array is
> > + * _not_ read-only, because the resulting event mask gets written back to each
> > + * entry's ->ep_events field. When successful, this will be the same as before
> > + * (plus EPOLLERR & EPOLLHUP). If ->ep_events gets cleared, then it is reasonable
> > + * to infer that the entry's ->ep_fildes was a bad file descriptor.
> > + */
> 
> > +	if (!access_ok(VERIFY_WRITE, in, inc * sizeof(struct epoll)))
> > +		goto out;
> > +	for (i = 0; i < inc; ++i) {
> > +		int fd, io;
> > +		long long id;
> > +
> > +		ret = -EFAULT;
> > +		if (__get_user(fd, &in[i].ep_fildes) ||
> > +		    __get_user(io, &in[i].ep_events) ||
> > +		    __get_user(id, &in[i].ep_ident))
> > +			goto out;
> > +
> > +		ep_control(file->private_data, fd, &io, id, 0);
> > +		ret = -EFAULT;
> > +		if (__put_user(io, &in[i].ep_events))
> > +			goto out;
> 
> I don't think we should waste cycles writing to 'in' on success.

Fair enough, my thought process was mainly too add some consistency to
the system call, but removing that constraint would clean up
ep_control() anyway.