Only process EV_DISPATCH and EV_ONESHOT when we get a valid event

If the copyout callback for a filter indicates that the event should be ignored/dropped, we should not process EV_DISPATCH or EV_ONESHOT, as this would lead to missing actual events in the future.

Some filters (e.g. EVFILT_PROC) do specifically want to process EV_ONESHOT even when dropping an event, so this also introduces a special filter ID (EVFILT_DROP_POSTPROCESS) that enables the old behavior for individual events. Additionally, rather than just using `0` as a special filter ID, this commit introduces EVFILT_DROP (to be clear about its meaning).

src/common/knote.c

@@ -62,7 +62,7 @@ knote_release(struct knote *kn)
         if (/*kn->kn_flags & KNFL_KNOTE_DELETED*/ 1) {
             dbg_printf("freeing knote at %p (delayed)", kn);
             LIST_INSERT_HEAD(&kn->kn_kq->kq_tofree, kn, kn_entries2free);
-            kn->kev.filter = 0;
+            kn->kev.filter = EVFILT_DROP;
         } else {
             dbg_puts("this should never happen");
         }

src/common/private.h

@@ -75,6 +75,9 @@ struct eventfd {
 #define KNFL_PASSIVE_SOCKET  (0x01)  /* Socket is in listen(2) mode */
 #define KNFL_REGULAR_FILE    (0x02)  /* File descriptor is a regular file */
 #define KNFL_KNOTE_DELETED   (0x10)  /* The knote object is no longer valid */
+
+#define EVFILT_DROP 0
+#define EVFILT_DROP_POSTPROCESS (EVFILT_SYSCOUNT + 1)
  
 struct knote {
     struct kevent_internal_s     kev;

src/linux/machport.c

@@ -86,7 +86,7 @@ evfilt_machport_copyout(struct kevent64_s *dst, struct knote *src, void *ptr)
 	if (reply.header.code == 0xdead) {
 		// server indicated there was actually no event available to read right now;
 		// drop the event
-		dst->filter = 0;
+		dst->filter = EVFILT_DROP;
 		return 0;
 	}
 
@@ -123,15 +123,13 @@ evfilt_machport_knote_create(struct filter *filt, struct knote *kn)
     ev.events = kn->data.events;
     ev.data.ptr = kn;
 
-    kn->kdata.kn_dupfd = eventfd(0, EFD_CLOEXEC);
-
 	int status = _dserver_rpc_kqchan_mach_port_open_4libkqueue(port, (void*)kn->kev.ext[0], kn->kev.ext[1], kn->kev.fflags, &kn->kdata.kn_dupfd);
 	if (status < 0) {
 		dbg_printf("evfilt_machport_open: %s", strerror(-status));
 		return (-1);
 	}
 
-	dbg_printf("evfilt_machport_open: listening to FD %d", kn->kdata.kn_dupfd);
+	dbg_printf("evfilt_machport_open: listening to FD %d for events %d", kn->kdata.kn_dupfd, ev.events);
 
     fcntl(kn->kdata.kn_dupfd, F_SETFD, FD_CLOEXEC);
 
@@ -209,6 +207,8 @@ evfilt_machport_knote_enable(struct filter *filt, struct knote *kn)
     ev.events = kn->data.events;
     ev.data.ptr = kn;
 
+	dbg_printf("enabling machport knote with ID=%llu for events %d", kn->kev.ident, ev.events);
+
 	if (epoll_ctl(kn->kn_epollfd, EPOLL_CTL_ADD, kn->kdata.kn_dupfd, &ev) < 0) {
 		dbg_perror("epoll_ctl(2)");
 		return (-1);
@@ -219,6 +219,7 @@ evfilt_machport_knote_enable(struct filter *filt, struct knote *kn)
 int
 evfilt_machport_knote_disable(struct filter *filt, struct knote *kn)
 {
+	dbg_printf("disable machport knote with ID=%llu", kn->kev.ident);
 	if (epoll_ctl(kn->kn_epollfd, EPOLL_CTL_DEL, kn->kdata.kn_dupfd, NULL) < 0) {
 		dbg_perror("epoll_ctl(2)");
 		return (-1);

src/linux/platform.c

@@ -193,45 +193,47 @@ linux_kevent_copyout(struct kqueue *kq, int nready,
     struct filter *filt;
     struct knote *kn;
     int i, nret, rv;
+    struct kevent64_s* event;
 
     nret = nready;
     for (i = 0; i < nready; i++) {
         ev = &epevt_get()[i];
         kn = (struct knote *) ev->data.ptr;
+        event = eventlist;
         if (kn->kev.filter == 0) {
             dbg_puts("kevent copyout for zero filter, discarding!");
             nret--;
             continue;
         }
         filt = &kq->kq_filt[~(kn->kev.filter)];
-        rv = filt->kf_copyout(eventlist, kn, ev);
+        rv = filt->kf_copyout(event, kn, ev);
         if (slowpath(rv < 0)) {
             dbg_puts("knote_copyout failed");
             /* XXX-FIXME: hard to handle this without losing events */
             abort();
         }
 
-        /*
-         * Certain flags cause the associated knote to be deleted
-         * or disabled.
-         */
-        if (eventlist->flags & EV_DISPATCH) 
-            knote_disable(filt, kn); //FIXME: Error checking
-        if (eventlist->flags & EV_ONESHOT) {
-            knote_delete(filt, kn); //FIXME: Error checking
-        }
-
         /* If an empty kevent structure is returned, the event is discarded. */
         /* TODO: add these semantics to windows + solaris platform.c */
 
-		/* EV_DELETE is an internal trick to not have this event passed to the app.
-		 * Typically used along with EV_ONESHOT - see linux/proc.c
-		 */
-        if (fastpath(eventlist->filter != 0) && !(eventlist->flags & EV_DELETE)) {
+        if (fastpath(event->filter != EVFILT_DROP && event->filter != EVFILT_DROP_POSTPROCESS)) {
             eventlist++;
         } else {
             dbg_puts("spurious wakeup, discarding event");
             nret--;
+            if (event->filter != EVFILT_DROP_POSTPROCESS) {
+                continue;
+            }
+        }
+
+        /*
+         * Certain flags cause the associated knote to be deleted
+         * or disabled.
+         */
+        if (event->flags & EV_DISPATCH) 
+            knote_disable(filt, kn); //FIXME: Error checking
+        if (event->flags & EV_ONESHOT) {
+            knote_delete(filt, kn); //FIXME: Error checking
         }
     }
 

src/linux/proc.c

@@ -94,7 +94,7 @@ evfilt_proc_copyout(struct kevent64_s *dst, struct knote *src, void *ptr)
 	if (reply.header.code == 0xdead) {
 		// server indicated there was actually no event available to read right now;
 		// drop the event
-		dst->filter = 0;
+		dst->filter = EVFILT_DROP;
 		return 0;
 	}
 
@@ -155,7 +155,7 @@ evfilt_proc_copyout(struct kevent64_s *dst, struct knote *src, void *ptr)
 		// don't deliver it to the program.
 
 		if (!(src->kev.fflags & NOTE_FORK))
-			dst->filter = 0; // drop event
+			dst->filter = EVFILT_DROP; // drop event
 	}
 	else if (dst->fflags & NOTE_EXIT)
 	{
@@ -163,11 +163,12 @@ evfilt_proc_copyout(struct kevent64_s *dst, struct knote *src, void *ptr)
 		dst->flags |= EV_EOF | EV_ONESHOT;
 		
 		// NOTE_EXIT is always announced so that we can
-		// remove the knote. Add EV_DELETE to avoid passing
-		// the event to the application if it is not interested
-		// in this event
+		// remove the knote. Avoid passing the event to
+		// the application if it is not interested in
+		// this event.
+		// We use EVFILT_DROP_POSTPROCESS to force the EV_ONESHOT to be processed.
 		if (!(src->kev.fflags & NOTE_EXIT))
-			dst->flags |= EV_DELETE;
+			dst->filter = EVFILT_DROP_POSTPROCESS; // drop event
 	}
 
     return (0);

src/linux/read.c

@@ -64,7 +64,7 @@ evfilt_read_copyout(struct kevent64_s *dst, struct knote *src, void *ptr)
         dst->data = get_eof_offset(src->kev.ident);
 
         if (dst->data == 0) {
-            dst->filter = 0;    /* Will cause the kevent to be discarded */
+            dst->filter = EVFILT_DROP;    /* Will cause the kevent to be discarded */
             if (epoll_ctl(src->kn_epollfd, EPOLL_CTL_DEL, src->kdata.kn_eventfd, NULL) < 0) {
                 dbg_perror("epoll_ctl(2)");
                 return (-1);

src/linux/vnode.c

@@ -174,7 +174,7 @@ evfilt_vnode_copyout(struct kevent64_s *dst, struct knote *src, void *ptr UNUSED
     dbg_printf("inotify event: %s", inotify_event_dump(evt));
     if (evt->mask & IN_IGNORED) {
         /* TODO: possibly return error when fs is unmounted */
-        dst->filter = 0;
+        dst->filter = EVFILT_DROP;
         return (0);
     }
 
@@ -184,7 +184,7 @@ scriptors reference the same file.
     */
     if (evt->mask & IN_CLOSE_WRITE || evt->mask & IN_CLOSE_NOWRITE) {
         src->kn_flags |= EV_ONESHOT; /* KLUDGE: causes the knote to be deleted */
-        dst->filter = 0; /* KLUDGE: causes the event to be discarded */
+        dst->filter = EVFILT_DROP_POSTPROCESS; /* KLUDGE: causes the event to be discarded */
         return (0);
     }