Re: kqueue EV_DISPATCH and EV_EOF interaction

[Mike Belopuhov]

On Fri, Mar 30, 2018 at 01:21 +0200, Mike Belopuhov wrote:
> 
> Hi,
> 
> This appears to be an issue with reactivating disabled event sources
> in kqueue_register.  Something along the lines of FreeBSD commits:
> 
> https://svnweb.freebsd.org/base?view=revision&revision=274560 and
> https://reviews.freebsd.org/rS295786 where parent differential review
> https://reviews.freebsd.org/D5307 has some additional comments.
> 
> In any case, by either porting their code (#else branch) or slightly
> adjusting our own (I think that should be enough), I can no longer
> reproduce the issue you've reported.  Please test and report back if
> that solves your original issue.  Either variants will require
> rigorous testing and a thorough review.
> 
> Cheers,
> Mike
> 

After a bit of tinkering, I think I can minimize the change even
further.  Basically we just need to call the filter once and if
there's some data available, it'll return true and we'll mark the
knote as active.

diff --git sys/kern/kern_event.c sys/kern/kern_event.c
index fb9cad360b1..4e0949645cb 100644
--- sys/kern/kern_event.c
+++ sys/kern/kern_event.c
@@ -671,10 +671,12 @@ kqueue_register(struct kqueue *kq, struct kevent *kev, 
struct proc *p)
        }
 
        if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
                s = splhigh();
                kn->kn_status &= ~KN_DISABLED;
+               if (kn->kn_fop->f_event(kn, 0))
+                       kn->kn_status |= KN_ACTIVE;
                if ((kn->kn_status & KN_ACTIVE) &&
                    ((kn->kn_status & KN_QUEUED) == 0))
                        knote_enqueue(kn);
                splx(s);
        }