Avoid unsynchronized updates to kn_status.

kn_status is protected by the kqueue's lock, but we were updating it
without the kqueue lock held.  For EVFILT_TIMER knotes, there is no
knlist lock, so the knote activation could occur during the kn_status
update and result in KN_QUEUED being lost, in which case we'd enqueue
an already-enqueued knote, corrupting the queue.

Fix the problem by setting or clearing KN_DISABLED before dropping the
kqueue lock to call into the filter.  KN_DISABLED is used only by the
core kevent code, so there is no side effect from setting it earlier.

Reported and tested by:	Sylvain GALLIANO <sg@efficientip.com>
Reviewed by:	kib
MFC after:	3 days
Sponsored by:	The FreeBSD Foundation
Differential Revision:	https://reviews.freebsd.org/D18060

1 files changed, 13 insertions, 8 deletions

diff --git a/sys/kern/kern_event.c b/sys/kern/kern_event.c
index fa59153f77e6..bf725d4d3fb3 100644
--- a/sys/kern/kern_event.c
+++ b/sys/kern/kern_event.c
@@ -1533,6 +1533,8 @@ findkn:
 			kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
 			    EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT);
 			kn->kn_status = KN_DETACHED;
+			if ((kev->flags & EV_DISABLE) != 0)
+				kn->kn_status |= KN_DISABLED;
 			kn_enter_flux(kn);
 
 			error = knote_attach(kn, kq);
@@ -1568,6 +1570,11 @@ findkn:
 		KNOTE_ACTIVATE(kn, 1);
 	}
 
+	if ((kev->flags & EV_ENABLE) != 0)
+		kn->kn_status &= ~KN_DISABLED;
+	else if ((kev->flags & EV_DISABLE) != 0)
+		kn->kn_status |= KN_DISABLED;
+
 	/*
 	 * The user may change some filter values after the initial EV_ADD,
 	 * but doing so will not reset any filter which has already been
@@ -1585,19 +1592,17 @@ findkn:
 		kn->kn_sdata = kev->data;
 	}
 
+done_ev_add:
 	/*
 	 * We can get here with kn->kn_knlist == NULL.  This can happen when
 	 * the initial attach event decides that the event is "completed" 
-	 * already.  i.e. filt_procattach is called on a zombie process.  It
-	 * will call filt_proc which will remove it from the list, and NULL
+	 * already, e.g., filt_procattach() is called on a zombie process.  It
+	 * will call filt_proc() which will remove it from the list, and NULL
 	 * kn_knlist.
+	 *
+	 * KN_DISABLED will be stable while the knote is in flux, so the
+	 * unlocked read will not race with an update.
 	 */
-done_ev_add:
-	if ((kev->flags & EV_ENABLE) != 0)
-		kn->kn_status &= ~KN_DISABLED;
-	else if ((kev->flags & EV_DISABLE) != 0)
-		kn->kn_status |= KN_DISABLED;
-
 	if ((kn->kn_status & KN_DISABLED) == 0)
 		event = kn->kn_fop->f_event(kn, 0);
 	else