Re: [musl] [PATCH] sys/epoll.h: add epoll ioctls

[Rich Felker <dalias@libc.org>]

On Sun, Jun 02, 2024 at 04:05:25PM -0700, Joe Damato wrote:
> On Wed, May 29, 2024 at 08:11:13AM -0700, Joe Damato wrote:
> > On Wed, May 29, 2024 at 09:17:07AM -0400, Rich Felker wrote:
> > > On Wed, May 29, 2024 at 06:49:59AM +0000, Joe Damato wrote:
> > > > add two ioctls to get and set struct epoll_params to allow users to
> > > > control epoll based busy polling of network sockets.
> > > > 
> > > > added to uapi in commit 18e2bf0edf4dd88d9656ec92395aa47392e85b61 (Linux
> > > > kernel 6.9 and newer).
> > > > ---
> > > >  include/sys/epoll.h | 12 ++++++++++++
> > > >  1 file changed, 12 insertions(+)
> > > > 
> > > > diff --git a/include/sys/epoll.h b/include/sys/epoll.h
> > > > index ac81a841..5f975c4a 100644
> > > > --- a/include/sys/epoll.h
> > > > +++ b/include/sys/epoll.h
> > > > @@ -7,6 +7,7 @@ extern "C" {
> > > >  
> > > >  #include <stdint.h>
> > > >  #include <sys/types.h>
> > > > +#include <sys/ioctl.h>
> > > >  #include <fcntl.h>
> > > >  
> > > >  #define __NEED_sigset_t
> > > > @@ -54,6 +55,17 @@ __attribute__ ((__packed__))
> > > >  #endif
> > > >  ;
> > > >  
> > > > +struct epoll_params {
> > > > +	uint32_t busy_poll_usecs;
> > > > +	uint16_t busy_poll_budget;
> > > > +	uint8_t prefer_busy_poll;
> > > > +
> > > > +	uint8_t __pad;
> > > > +};
> > > > +
> > > > +#define EPOLL_IOC_TYPE 0x8A
> > > > +#define EPIOCSPARAMS _IOW(EPOLL_IOC_TYPE, 0x01, struct epoll_params)
> > > > +#define EPIOCGPARAMS _IOR(EPOLL_IOC_TYPE, 0x02, struct epoll_params)
> > > >  
> > > >  int epoll_create(int);
> > > >  int epoll_create1(int);
> > > > -- 
> > > > 2.34.1
> > > 
> > > This is probably okay, but we should at least ask if sys/ioctl.h is
> > > going to be a namespace mess. Is the intent to bring all of it in, or
> > > just to get the EPIOC* macros which depend on _IOW and _IOR?
> > 
> > Yes, sys/ioctl.h is pulled in for the _IOW and _IOR macros.
> > Similar to, for example, sys/mtio.h in musl, which also pulls in
> > sys/ioctl.h.
> > 
> > > On glibc, does it pull in sys/ioctl.h?
> > 
> > Yes, the code I've submit for glibc does pull in sys/ioctl.h.
> > 
> > That code has been approved by a glibc committer, but not yet merged
> > to the tree (I assume that will happen in a few days):
> > 
> > https://sourceware.org/pipermail/libc-alpha/2024-May/157166.html
> 
> Just wanted to follow up on the above.
> 
> Were you expecting me to make any changes or did you want to wait
> until libc takes the code before accepting it?
> 
> FWIW:
> 
> uclibc has taken the patch here: 
>   https://cgit.uclibc-ng.org/cgi/cgit/uclibc-ng.git/commit/?id=8bb33a2e1f2baec2078581d77e181f1ead5f51aa
> 
> And musl has similar code in include/sys/mount.h:
>   https://git.musl-libc.org/cgit/musl/tree/include/sys/mount.h#n8

I think it's okay as-is if this is what everyone else is doing too.
This is not a standard header so there aren't strong constraints on
what it can do; I just didn't want to be gratuitously more
namespace-invasive than on other systems with the same header.

Rich