Re: [PATCH 1/4] pidfd: support PIDFD_NONBLOCK in pidfd_open()

From: Oleg Nesterov
Date: Tue Sep 01 2020 - 12:53:34 EST


On 09/01, Christian Brauner wrote:
>
> On Tue, Sep 01, 2020 at 06:23:10PM +0200, Oleg Nesterov wrote:
> > On 08/31, Christian Brauner wrote:
> > >
> > > --- /dev/null
> > > +++ b/include/uapi/linux/pidfd.h
> > > @@ -0,0 +1,12 @@
> > > +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
> > > +
> > > +#ifndef _UAPI_LINUX_PIDFD_H
> > > +#define _UAPI_LINUX_PIDFD_H
> > > +
> > > +#include <linux/types.h>
> > > +#include <linux/fcntl.h>
> > > +
> > > +/* Flags for pidfd_open(). */
> > > +#define PIDFD_NONBLOCK O_NONBLOCK
> > > +
> > > +#endif /* _UAPI_LINUX_PIDFD_H */
> >
> > Why? Can't we simply use O_NONBLOCK ?
>
> It's the same thing we seem to do for any other (anon inode) fds:
>
> include/linux/eventfd.h:#define EFD_NONBLOCK O_NONBLOCK
> include/uapi/linux/inotify.h:#define IN_NONBLOCK O_NONBLOCK
> include/uapi/linux/signalfd.h:#define SFD_NONBLOCK O_NONBLOCK
> include/uapi/linux/timerfd.h:#define TFD_NONBLOCK O_NONBLOCK
>
> also for O_CLOEXEC:
>
> include/linux/eventfd.h:#define EFD_CLOEXEC O_CLOEXEC
> include/linux/userfaultfd_k.h:#define UFFD_CLOEXEC O_CLOEXEC
> include/uapi/linux/eventpoll.h:#define EPOLL_CLOEXEC O_CLOEXEC
> include/uapi/linux/mount.h:#define OPEN_TREE_CLOEXEC O_CLOEXEC
> include/uapi/linux/perf_event.h:#define PERF_FLAG_FD_CLOEXEC (1UL << 3) /* O_CLOEXEC */
> include/uapi/linux/signalfd.h:#define SFD_CLOEXEC O_CLOEXEC
> include/uapi/linux/timerfd.h:#define TFD_CLOEXEC O_CLOEXEC
>
> So I think we should just do the same.

Hmm, OK, then I have to agree.

> A clean flag namespace seems
> nicer to me too tbh.

Disagree but this doesn't matter ;)

Oleg.