svn commit: r316874 - head/sys/kern

[Maxim Sobolev]

Peter, It is actually the other way around. If you take syslogd code out of
11-stable and earlier that would work just fine with or without r316874.
But since r285910 syslogd in head had been refactored a lot and I think
that particular bug has introduced that has been masked by the shutdown()
on datagram sockets becoming a NOP after r285910. Then r316874 restored our
historical behavior for the shutdown(2) and bingo, bug in the new syslogd
code is now causing it to spin when shutdown() != NOP.

-Max

On Fri, Apr 14, 2017 at 12:46 PM, Peter Wemm <peter at wemm.org> wrote:

> On Friday, April 14, 2017 12:41:52 PM Maxim Sobolev wrote:
> > Thanks, Peter. I will try to look into this asap.
>
> I don't understand what is going on yet. Presumably there must be other
> changes in play that affect udp/select sometime between the original 2015
> change and this. The syslogd -s code is Old(TM).  I'm also wondering
> whether
> the -s code even works at all since the 2015 / r285910 change...
>
> > -Max
> >
> > On Apr 14, 2017 12:32 PM, "Peter Wemm" <peter at wemm.org> wrote:
> > > On Friday, April 14, 2017 11:49:26 AM Peter Wemm wrote:
> > > > On Friday, April 14, 2017 05:23:28 PM Maxim Sobolev wrote:
> > > > > Author: sobomax
> > > > > Date: Fri Apr 14 17:23:28 2017
> > > > > New Revision: 316874
> > > > > URL: https://svnweb.freebsd.org/changeset/base/316874
> > > > >
> > > > > Log:
> > > > >   Restore ability to shutdown DGRAM sockets, still forcing
> ENOTCONN to
> > >
> > > be
> > >
> > > > > returned by the shutdown(2) system call. This ability has been
> lost as
> > > > > part
> > > > > of the svn revision 285910.
> > > > >
> > > > >   Reviewed by:      ed, rwatson, glebius, hiren
> > > > >   MFC after:        2 weeks
> > > > >   Differential Revision:    https://reviews.freebsd.org/D10351
> > > >
> > > > This appears to have broken syslogd and had a major change in
> behavior
> > >
> > > with
> > >
> > > > regards to select(2).
> > > >
> > > > If you run syslogd with the -s flag (which is default), it now spins
> at
> > >
> > > 100%
> > >
> > > > cpu as all the shutdown sockets now return readable from select.
> > > >
> > > > Old releases / jails also manifest this behavor.  If it wasn't for
> > > > losing
> > > > the ability to run old branch binaries I'd suggest changing syslogd
> > > > instead, but we depend on this in the cluster and I expect others do
> > > > too.
> > > >
> > > > I'm not 100% certain that this change is the culprit but a heads-up
> > > > can't
> > > > hurt. I'll try reverting it on the freebsd cluster next, after fixing
> > > > the
> > > > broken auditing changes.
> > > >
> > > > -Peter
> > >
> > > I can confirm that reverting r316874 fixes syslogd and backwards
> > > compatability
> > > with old branches.
> > >
> > > --
> > > Peter Wemm - peter at wemm.org; peter at FreeBSD.org; peter at yahoo-inc.com;
> > > KI6FJV
> > > UTF-8: for when a ' or ... just won\342\200\231t do\342\200\246
>
> --
> Peter Wemm - peter at wemm.org; peter at FreeBSD.org; peter at yahoo-inc.com;
> KI6FJV
> UTF-8: for when a ' or ... just won\342\200\231t do\342\200\246
>