bhyve: add backend rx backpressure to virtio-net

If a VM is flooded with more ingress packets than the guest OS
can handle, the current virtio-net code will keep reading those
packets and drop most of them as no space is available in the
receive queue. This is an undesirable receive livelock, which
is a waste of CPU and memory resources and potentially opens to
DoS attacks.
With this change, virtio-net uses the new netbe_rx_disable()
function to disable ingress operation in the backend while the
guest is short on RX buffers. Once the guest makes more buffers
available to the RX virtqueue, ingress operation is enabled again
by calling netbe_rx_enable().

Reviewed by:	bryanv, jhb
MFC after:	2 weeks
Differential Revision:	https://reviews.freebsd.org/D20987

usr.sbin/bhyve/mevent.c

@@ -321,6 +321,14 @@ mevent_add(int tfd, enum ev_type type,
 	return mevent_add_state(tfd, type, func, param, MEV_ADD);
 }
 
+struct mevent *
+mevent_add_disabled(int tfd, enum ev_type type,
+		    void (*func)(int, enum ev_type, void *), void *param)
+{
+
+	return mevent_add_state(tfd, type, func, param, MEV_ADD_DISABLED);
+}
+
 static int
 mevent_update(struct mevent *evp, int newstate)
 {

usr.sbin/bhyve/mevent.h

@@ -43,6 +43,9 @@ struct mevent;
 struct mevent *mevent_add(int fd, enum ev_type type, 
 			  void (*func)(int, enum ev_type, void *),
 			  void *param);
+struct mevent *mevent_add_disabled(int fd, enum ev_type type,
+			  void (*func)(int, enum ev_type, void *),
+			  void *param);
 int	mevent_enable(struct mevent *evp);
 int	mevent_disable(struct mevent *evp);
 int	mevent_delete(struct mevent *evp);

usr.sbin/bhyve/net_backends.c

@@ -220,7 +220,7 @@ tap_init(struct net_backend *be, const char *devname,
 		errx(EX_OSERR, "Unable to apply rights for sandbox");
 #endif
 
-	priv->mevp = mevent_add(be->fd, EVF_READ, cb, param);
+	priv->mevp = mevent_add_disabled(be->fd, EVF_READ, cb, param);
 	if (priv->mevp == NULL) {
 		WPRINTF(("Could not register event\n"));
 		goto error;
@@ -432,7 +432,7 @@ netmap_init(struct net_backend *be, const char *devname,
 	priv->cb_param = param;
 	be->fd = priv->nmd->fd;
 
-	priv->mevp = mevent_add(be->fd, EVF_READ, cb, param);
+	priv->mevp = mevent_add_disabled(be->fd, EVF_READ, cb, param);
 	if (priv->mevp == NULL) {
 		WPRINTF(("Could not register event\n"));
 		return (-1);

usr.sbin/bhyve/pci_e82545.c

@@ -2351,6 +2351,8 @@ e82545_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
 		net_genmac(pi, sc->esc_mac.octet);
 	}
 
+	netbe_rx_enable(sc->esc_be);
+
 	/* H/w initiated reset */
 	e82545_reset(sc, 0);
 

usr.sbin/bhyve/pci_virtio_net.c

@@ -101,7 +101,6 @@ struct pci_vtnet_softc {
 
 	net_backend_t	*vsc_be;
 
-	int		vsc_rx_ready;
 	int		resetting;	/* protected by tx_mtx */
 
 	uint64_t	vsc_features;	/* negotiated features */
@@ -156,7 +155,6 @@ pci_vtnet_reset(void *vsc)
 		pthread_mutex_lock(&sc->tx_mtx);
 	}
 
-	sc->vsc_rx_ready = 0;
 	sc->rx_merge = 1;
 	sc->rx_vhdrlen = sizeof(struct virtio_net_rxhdr);
 
@@ -180,30 +178,29 @@ pci_vtnet_rx(struct pci_vtnet_softc *sc)
 	int len, n;
 	uint16_t idx;
 
-	if (!sc->vsc_rx_ready) {
-		/*
-		 * The rx ring has not yet been set up.
-		 * Drop the packet and try later.
-		 */
-		netbe_rx_discard(sc->vsc_be);
-		return;
-	}
-
-	/*
-	 * Check for available rx buffers
-	 */
 	vq = &sc->vsc_queues[VTNET_RXQ];
-	if (!vq_has_descs(vq)) {
+	for (;;) {
 		/*
-		 * No available rx buffers. Drop the packet and try later.
-		 * Interrupt on empty, if that's negotiated.
+		 * Check for available rx buffers.
 		 */
-		netbe_rx_discard(sc->vsc_be);
-		vq_endchains(vq, /*used_all_avail=*/1);
-		return;
-	}
+		if (!vq_has_descs(vq)) {
+			/* No rx buffers. Enable RX kicks and double check. */
+			vq_kick_enable(vq);
+			if (!vq_has_descs(vq)) {
+				/*
+				 * Still no buffers. Interrupt if needed
+				 * (including for NOTIFY_ON_EMPTY), and
+				 * disable the backend until the next kick.
+				 */
+				vq_endchains(vq, /*used_all_avail=*/1);
+				netbe_rx_disable(sc->vsc_be);
+				return;
+			}
+
+			/* More rx buffers found, so keep going. */
+			vq_kick_disable(vq);
+		}
 
-	do {
 		/*
 		 * Get descriptor chain.
 		 */
@@ -215,7 +212,8 @@ pci_vtnet_rx(struct pci_vtnet_softc *sc)
 		if (len <= 0) {
 			/*
 			 * No more packets (len == 0), or backend errored
-			 * (err < 0). Return unused available buffers.
+			 * (err < 0). Return unused available buffers
+			 * and stop.
 			 */
 			vq_retchain(vq);
 			/* Interrupt if needed/appropriate and stop. */
@@ -225,10 +223,8 @@ pci_vtnet_rx(struct pci_vtnet_softc *sc)
 
 		/* Publish the info to the guest */
 		vq_relchain(vq, idx, (uint32_t)len);
-	} while (vq_has_descs(vq));
+	}
 
-	/* Interrupt if needed, including for NOTIFY_ON_EMPTY. */
-	vq_endchains(vq, /*used_all_avail=*/1);
 }
 
 /*
@@ -254,13 +250,11 @@ pci_vtnet_ping_rxq(void *vsc, struct vqueue_info *vq)
 	struct pci_vtnet_softc *sc = vsc;
 
 	/*
-	 * A qnotify means that the rx process can now begin
+	 * A qnotify means that the rx process can now begin.
 	 */
 	pthread_mutex_lock(&sc->rx_mtx);
-	if (sc->vsc_rx_ready == 0) {
-		sc->vsc_rx_ready = 1;
-		vq_kick_disable(vq);
-	}
+	vq_kick_disable(vq);
+	netbe_rx_enable(sc->vsc_be);
 	pthread_mutex_unlock(&sc->rx_mtx);
 }