Date: Sat, 1 Apr 2006 12:06:39 -0800 From: John-Mark Gurney <gurney_j@resnet.uoregon.edu> To: =?iso-8859-1?Q?V=E1clav?= Haisman <V.Haisman@sh.cvut.cz> Cc: hackers@freebsd.org, Dan Nelson <dnelson@allantgroup.com> Subject: Re: kqueue oddity Message-ID: <20060401200639.GD53811@funkthat.com> In-Reply-To: <442EC7F7.3000608@sh.cvut.cz> References: <442E9354.5010700@sh.cvut.cz> <20060401180311.GA83967@dan.emsphone.com> <442EC7F7.3000608@sh.cvut.cz>
next in thread | previous in thread | raw e-mail | index | archive | help
Vclav Haisman wrote this message on Sat, Apr 01, 2006 at 20:35 +0200:
> Dan Nelson wrote:
> > In the last episode (Apr 01), Vclav Haisman said:
> >> Hi,
> >> I have a little problem with kqueue/kevent. I am getting somewhat
> >> strange events from kevent(). The struct kevent's contents looks like this:
> >>
> >> kevent {.ident = 1, .filter = 0xfffffffe, .flags = 0x0001,
> >> .fflags = 0x00000000, .data = 0x000084c6}
> >>
> >> .ident == 1 is stdout.
> >> .filter == 0x fffffffe is -2 is EVFILT_WRITE which makes sense for stdout.
> >> .fflags is empty, nothing out of ordinary for EVFILT_WRITE.
> >> And here comes the oddity: .flags == 0x0001 means EV_ADD which totally
> >> doesn't make sense to me and neither does according to man kqueue. The
> >> code that exhibits this behaviour is io_loop_handler_run() in
> >> ioloop-kqueue.c of Dovecot IMAP server.
> >
> > It's a kqueue bug, but a minor one. The problem is that the same
> > "flags" field is used to pass "actions" from the client, and return
> > status from the kernel. When you call kqueue with EV_ADD, the kernel
> > never clears the flags when creating the internal object, so it's
> > always visible when events fire. This can get annoying when trussing
> > kqueue-using programs because very single event has the EV_ADD flag :)
> > Apply the following diff to have the kernel strip EV_ADD:
> >
> > Index: kern_event.c
> > ===================================================================
> > RCS file: /home/ncvs/src/sys/kern/kern_event.c,v
> > retrieving revision 1.79.2.9
> > diff -u -p -r1.79.2.9 kern_event.c
> > --- kern_event.c 29 Jun 2005 18:08:31 -0000 1.79.2.9
> > +++ kern_event.c 7 Jul 2005 18:51:29 -0000
> > @@ -864,6 +864,7 @@ findkn:
> > kev->fflags = 0;
> > kev->data = 0;
> > kn->kn_kevent = *kev;
> > + kn->kn_kevent.flags &= ~EV_ADD;
> > kn->kn_status = KN_INFLUX|KN_DETACHED;
> >
> > error = knote_attach(kn, kq);
> >
> >
> Thanks. Why isn't this fix applied? From the patch's time stamp it seems
> like it is known for quite some time now.
Because I was never informed of the patch, so had no clue that it was
necessary... I'll commit a similar patch shortly...
--
John-Mark Gurney Voice: +1 415 225 5579
"All that I will do, has been done, All that I have, has not."
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20060401200639.GD53811>