diff --git a/sys/dev/ic/hd64570.c b/sys/dev/ic/hd64570.c
index 359b03f..9df2225 100644
--- a/sys/dev/ic/hd64570.c
+++ b/sys/dev/ic/hd64570.c
@@ -171,6 +171,9 @@ static struct mbuf *sca_mbuf_alloc(struct sca_softc *, void *, u_int);
 static	void sca_frame_print(sca_port_t *, sca_desc_t *, u_int8_t *);
 #endif
 
+static IFNET_WRAP_IF_OUTPUT_DECL(sca_output);
+static IFNET_WRAP_IF_OUTPUT_IMPL(sca_output);
+#define SCA_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(sca_output)
 
 #define	sca_read_1(sc, reg)		(sc)->sc_read_1(sc, reg)
 #define	sca_read_2(sc, reg)		(sc)->sc_read_2(sc, reg)
@@ -446,7 +449,7 @@ sca_port_attach(struct sca_softc *sc, u_int port)
 	ifp->if_type = IFT_PTPSERIAL;
 	ifp->if_hdrlen = HDLC_HDRLEN;
 	ifp->if_ioctl = sca_ioctl;
-	ifp->if_output = sca_output;
+	ifp->if_output = SCA_OUTPUT_LOCK;
 	ifp->if_watchdog = sca_watchdog;
 	ifp->if_snd.ifq_maxlen = IFQ_MAXLEN;
 	scp->linkq.ifq_maxlen = 5; /* if we exceed this we are hosed already */
@@ -799,6 +802,8 @@ sca_output(
 	int s, error, len;
 	short mflags;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	error = 0;
 
 	if ((ifp->if_flags & IFF_UP) != IFF_UP) {
diff --git a/sys/dev/ic/midway.c b/sys/dev/ic/midway.c
index 83bf0f1..df057e7 100644
--- a/sys/dev/ic/midway.c
+++ b/sys/dev/ic/midway.c
@@ -765,7 +765,7 @@ done_probe:
 #endif
   ifp->if_flags = IFF_SIMPLEX|IFF_NOTRAILERS;
   ifp->if_ioctl = en_ioctl;
-  ifp->if_output = atm_output;
+  ifp->if_output = ATM_OUTPUT_LOCK;
   ifp->if_start = en_start;
   IFQ_SET_READY(&ifp->if_snd);
 
diff --git a/sys/dev/ic/pdq_ifsubr.c b/sys/dev/ic/pdq_ifsubr.c
index f117850..3b81413 100644
--- a/sys/dev/ic/pdq_ifsubr.c
+++ b/sys/dev/ic/pdq_ifsubr.c
@@ -496,7 +496,7 @@ pdq_ifattach(
 
     ifp->if_ioctl = pdq_ifioctl;
 #if !defined(__NetBSD__)
-    ifp->if_output = fddi_output;
+    ifp->if_output = FDDI_OUTPUT_LOCK;
 #endif
     ifp->if_start = pdq_ifstart;
     IFQ_SET_READY(&ifp->if_snd);
diff --git a/sys/dev/pci/if_lmc.c b/sys/dev/pci/if_lmc.c
index a40fb80..5afa064 100644
--- a/sys/dev/pci/if_lmc.c
+++ b/sys/dev/pci/if_lmc.c
@@ -3360,6 +3360,8 @@ ifnet_output(struct ifnet *ifp, struct mbuf *m,
   softc_t *sc = IFP2SC(ifp);
   int error = 0;
 
+  KASSERT(KERNEL_LOCKED_P());
+
   /* Fail if the link is down. */
   if (sc->status.link_state != STATE_UP)
     {
@@ -3393,6 +3395,10 @@ ifnet_output(struct ifnet *ifp, struct mbuf *m,
   return error;
   }
 
+static IFNET_WRAP_IF_OUTPUT_DECL(ifnet_output);
+static IFNET_WRAP_IF_OUTPUT_IMPL(ifnet_output);
+#define IFNET_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(ifnet_output)
+
 static int  /* context: process */
 ifnet_ioctl(struct ifnet *ifp, u_long cmd, void *data)
   {
@@ -3511,7 +3517,7 @@ ifnet_setup(struct ifnet *ifp)
   ifp->if_flags   |= IFF_SIMPLEX;
   ifp->if_flags   |= IFF_NOARP;
   ifp->_if_input    = ifnet_input;
-  ifp->if_output   = ifnet_output;
+  ifp->if_output   = IFNET_OUTPUT_LOCK;
   ifp->if_start    = ifnet_start;
   ifp->if_ioctl    = ifnet_ioctl;
   ifp->if_watchdog = ifnet_watchdog;
diff --git a/sys/dev/pci/if_wm.c b/sys/dev/pci/if_wm.c
index 71d4104..6778cc8 100644
--- a/sys/dev/pci/if_wm.c
+++ b/sys/dev/pci/if_wm.c
@@ -2414,6 +2414,7 @@ alloc_retry:
 	strlcpy(ifp->if_xname, xname, IFNAMSIZ);
 	ifp->if_softc = sc;
 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
+	ifp->if_extflags = IFEF_MPSAFE;
 	ifp->if_ioctl = wm_ioctl;
 	if ((sc->sc_flags & WM_F_NEWQUEUE) != 0) {
 		ifp->if_start = wm_nq_start;
diff --git a/sys/dev/ppbus/if_plip.c b/sys/dev/ppbus/if_plip.c
index 96f2201..a3f4897 100644
--- a/sys/dev/ppbus/if_plip.c
+++ b/sys/dev/ppbus/if_plip.c
@@ -198,6 +198,9 @@ static int lpoutput(struct ifnet *, struct mbuf *, const struct sockaddr *,
 static void lpstart(struct ifnet *);
 static void lp_intr(void *);
 
+static IFNET_WRAP_IF_OUTPUT_DECL(lpoutput);
+static IFNET_WRAP_IF_OUTPUT_IMPL(lpoutput);
+#define LPOUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(lpoutput)
 
 static int
 lp_probe(device_t parent, cfdata_t match, void *aux)
@@ -230,7 +233,7 @@ lp_attach(device_t parent, device_t self, void *aux)
 	ifp->if_mtu = LPMTU;
 	ifp->if_flags = IFF_SIMPLEX | IFF_POINTOPOINT | IFF_MULTICAST;
 	ifp->if_ioctl = lpioctl;
-	ifp->if_output = lpoutput;
+	ifp->if_output = LPOUTOUT_LOCK;
 	ifp->if_start = lpstart;
 	ifp->if_type = IFT_PARA;
 	ifp->if_hdrlen = 0;
@@ -707,6 +710,8 @@ lpoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
 	int err;
 	int s;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	s = splnet();
 
 	if(sc->sc_dev_ok) {
diff --git a/sys/dev/qbus/if_dmc.c b/sys/dev/qbus/if_dmc.c
index ba3c987..3d969cc 100644
--- a/sys/dev/qbus/if_dmc.c
+++ b/sys/dev/qbus/if_dmc.c
@@ -182,6 +182,11 @@ static  int dmcoutput(struct ifnet *, struct mbuf *, struct sockaddr *,
 	struct rtentry *);
 static  void dmcreset(device_t);
 
+static IFNET_WRAP_IF_OUTPUT_DECL(dmcoutput);
+static IFNET_WRAP_IF_OUTPUT_IMPL(dmcoutput);
+#define DMCOUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(dmcoutput)
+
+
 CFATTACH_DECL_NEW(dmc, sizeof(struct dmc_softc),
     dmcmatch, dmcattach, NULL, NULL);
 
@@ -259,7 +264,7 @@ dmcattach(device_t parent, device_t self, void *aux)
 	strlcpy(sc->sc_if.if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
 	sc->sc_if.if_mtu = DMCMTU;
 	sc->sc_if.if_init = dmcinit;
-	sc->sc_if.if_output = dmcoutput;
+	sc->sc_if.if_output = DMCOUTPUT_LOCK;
 	sc->sc_if.if_ioctl = dmcioctl;
 	sc->sc_if.if_watchdog = dmctimeout;
 	sc->sc_if.if_flags = IFF_POINTOPOINT;
@@ -426,6 +431,8 @@ dmcstart(struct ifnet *ifp)
 	struct dmcbufs *rp;
 	int n;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	/*
 	 * Dequeue up to NXMT requests and map them to the UNIBUS.
 	 * If no more requests, or no dmc buffers available, just return.
diff --git a/sys/dev/usb/if_upl.c b/sys/dev/usb/if_upl.c
index 8e9dcc0..6b2817b 100644
--- a/sys/dev/usb/if_upl.c
+++ b/sys/dev/usb/if_upl.c
@@ -196,6 +196,11 @@ Static int upl_output(struct ifnet *, struct mbuf *, const struct sockaddr *,
 		      const struct rtentry *);
 Static void upl_input(struct ifnet *, struct mbuf *);
 
+Static IFNET_WRAP_IF_OUTPUT_DECL(upl_output);
+Static IFNET_WRAP_IF_OUTPUT_IMPL(upl_output);
+#define UPL_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(upl_output)
+
+
 /*
  * Probe for a Prolific chip.
  */
@@ -298,7 +303,7 @@ upl_attach(device_t parent, device_t self, void *aux)
 	ifp->if_type = IFT_OTHER;
 	ifp->if_addrlen = 0;
 	ifp->if_hdrlen = 0;
-	ifp->if_output = upl_output;
+	ifp->if_output = UPL_OUTPUT_LOCK;
 	ifp->_if_input = upl_input;
 	ifp->if_baudrate = 12000000;
 	ifp->if_dlt = DLT_RAW;
@@ -989,6 +994,8 @@ upl_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
 {
 	int error;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	DPRINTFN(10,("%s: %s: enter\n",
 		     device_xname(((struct upl_softc *)ifp->if_softc)->sc_dev),
 		     __func__));
diff --git a/sys/dev/usb/uhso.c b/sys/dev/usb/uhso.c
index cf153f0..83207be 100644
--- a/sys/dev/usb/uhso.c
+++ b/sys/dev/usb/uhso.c
@@ -424,6 +424,10 @@ Static void uhso_ifnet_start(struct ifnet *);
 Static int  uhso_ifnet_output(struct ifnet *, struct mbuf *,
     const struct sockaddr *, const struct rtentry *);
 
+Static IFNET_WRAP_IF_OUTPUT_DECL(uhso_ifnet_output);
+Static IFNET_WRAP_IF_OUTPUT_IMPL(uhso_ifnet_output);
+#define UHSO_IFNET_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(uhso_ifnet_output)
+
 
 /*******************************************************************************
  *
@@ -1937,7 +1941,7 @@ uhso_ifnet_attach(struct uhso_softc *sc, struct usbd_interface *ifh, int index)
 	ifp->if_flags = IFF_NOARP | IFF_SIMPLEX;
 	ifp->if_ioctl = uhso_ifnet_ioctl;
 	ifp->if_start = uhso_ifnet_start;
-	ifp->if_output = uhso_ifnet_output;
+	ifp->if_output = UHSO_IFNET_OUTPUT_LOCK;
 	IFQ_SET_READY(&ifp->if_snd);
 
 	hp->hp_sc = sc;
@@ -2344,6 +2348,8 @@ uhso_ifnet_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
 {
 	int error;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	if (!ISSET(ifp->if_flags, IFF_RUNNING))
 		return EIO;
 
diff --git a/sys/net/if.c b/sys/net/if.c
index 8b076e9..277ae2e 100644
--- a/sys/net/if.c
+++ b/sys/net/if.c
@@ -2788,7 +2788,7 @@ if_transmit(struct ifnet *ifp, struct mbuf *m)
 		ifp->if_omcasts++;
 
 	if ((ifp->if_flags & IFF_OACTIVE) == 0)
-		(*ifp->if_start)(ifp);
+		if_start_lock(ifp);
 out:
 	splx(s);
 
diff --git a/sys/net/if.h b/sys/net/if.h
index 8c10b21..046c2b8 100644
--- a/sys/net/if.h
+++ b/sys/net/if.h
@@ -242,7 +242,8 @@ typedef struct ifnet {
 	u_short	if_index;		/* numeric abbreviation for this if */
 	short	if_timer;		/* time 'til if_slowtimo called */
 	short	if_flags;		/* up/down, broadcast, etc. */
-	short	if__pad1;		/* be nice to m68k ports */
+	short	if_extflags;		/* device driver MP-safe, etc. */
+					/* be nice to m68k ports */
 	struct	if_data if_data;	/* statistics and other data about if */
 	/*
 	 * Procedure handles.  If you add more of these, don't forget the
@@ -379,6 +380,28 @@ typedef struct ifnet {
 #define	IFF_LINK2	0x4000		/* per link layer defined bit */
 #define	IFF_MULTICAST	0x8000		/* supports multicast */
 
+#define	IFEF_MPSAFE	0x0001		/* if_start() can run parallel */
+
+static inline bool
+if_is_mpsafe(struct ifnet *ifp)
+{
+
+	return ((ifp->if_extflags & IFEF_MPSAFE) != 0);
+}
+
+static inline void
+if_start_lock(struct ifnet *ifp)
+{
+
+	if (if_is_mpsafe(ifp)) {
+		(*ifp->if_start)(ifp);
+	} else {
+		KERNEL_LOCK(1, NULL);
+		(*ifp->if_start)(ifp);
+		KERNEL_UNLOCK_ONE(NULL);
+	}
+}
+
 #define	IFFBITS \
     "\020\1UP\2BROADCAST\3DEBUG\4LOOPBACK\5POINTOPOINT\6NOTRAILERS" \
     "\7RUNNING\10NOARP\11PROMISC\12ALLMULTI\13OACTIVE\14SIMPLEX" \
@@ -1028,6 +1051,25 @@ __END_DECLS
 		}							\
 	} while (0)
 
+#define	IFNET_WRAP_IF_OUTPUT_DECL(name)					\
+	int								\
+	name##_wrapper(struct ifnet *, struct mbuf *,			\
+	    const struct sockaddr *, const struct rtentry *)
+
+#define	IFNET_WRAP_IF_OUTPUT_IMPL(name)					\
+	int								\
+	name##_wrapper(struct ifnet *__ifp, struct mbuf *__m,		\
+	    const struct sockaddr *__sa, const struct rtentry *__rt)	\
+	{								\
+		int __ret;						\
+		KERNEL_LOCK(1, NULL);					\
+		__ret = name(__ifp, __m, __sa, __rt);			\
+		KERNEL_UNLOCK_ONE(NULL);				\
+		return __ret;						\
+	}
+
+#define	IFNET_WRAP_IF_OUTPUT_CALL(name) name##_wrapper
+
 extern struct pslist_head ifnet_pslist;
 extern struct psref_class *ifnet_psref_class;
 extern kmutex_t ifnet_mtx;
diff --git a/sys/net/if_arcsubr.c b/sys/net/if_arcsubr.c
index 953a35a..46c5334 100644
--- a/sys/net/if_arcsubr.c
+++ b/sys/net/if_arcsubr.c
@@ -105,6 +105,10 @@ static	int arc_output(struct ifnet *, struct mbuf *,
 	    const struct sockaddr *, const struct rtentry *);
 static	void arc_input(struct ifnet *, struct mbuf *);
 
+static IFNET_WRAP_IF_OUTPUT_DECL(arc_output);
+static IFNET_WRAP_IF_OUTPUT_IMPL(arc_output);
+#define ARC_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(arc_output)
+
 /*
  * ARCnet output routine.
  * Encapsulate a packet of type family for the local net.
@@ -123,6 +127,8 @@ arc_output(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
 	uint8_t			atype, adst, myself;
 	int			tfrags, sflag, fsflag, rsflag;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) != (IFF_UP|IFF_RUNNING))
 		return (ENETDOWN); /* m, m1 aren't initialized yet */
 
@@ -620,7 +626,7 @@ arc_ifattach(struct ifnet *ifp, uint8_t lla)
 		    "%s: arc_ipmtu is %d, but must not exceed %d\n",
 		    ifp->if_xname, arc_ipmtu, ARC_PHDS_MAXMTU);
 
-	ifp->if_output = arc_output;
+	ifp->if_output = ARC_OUTPUT_LOCK;
 	ifp->_if_input = arc_input;
 	ac = (struct arccom *)ifp;
 	ac->ac_seqid = (time_second) & 0xFFFF; /* try to make seqid unique */
diff --git a/sys/net/if_atm.h b/sys/net/if_atm.h
index 78daeb3..3f82860 100644
--- a/sys/net/if_atm.h
+++ b/sys/net/if_atm.h
@@ -137,6 +137,10 @@ void	atm_input(struct ifnet *, struct atm_pseudohdr *,
 		struct mbuf *, void *);
 int	atm_output(struct ifnet *, struct mbuf *, const struct sockaddr *,
 		const struct rtentry *);
+
+IFNET_WRAP_IF_OUTPUT_DECL(atm_output);
+#define ATM_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(atm_output)
+
 #endif
 #ifdef ATM_PVCEXT
 #ifdef _KERNEL
diff --git a/sys/net/if_atmsubr.c b/sys/net/if_atmsubr.c
index 2c54589..8a40528 100644
--- a/sys/net/if_atmsubr.c
+++ b/sys/net/if_atmsubr.c
@@ -71,6 +71,8 @@ __KERNEL_RCSID(0, "$NetBSD: if_atmsubr.c,v 1.59 2016/06/10 13:27:15 ozaki-r Exp
 #include <netnatm/natm.h>
 #endif
 
+IFNET_WRAP_IF_OUTPUT_IMPL(atm_output);
+
 #define senderr(e) { error = (e); goto bad;}
 
 /*
@@ -99,6 +101,8 @@ atm_output(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
 	struct atmllc *atmllc;
 	uint32_t atm_flags;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) != (IFF_UP|IFF_RUNNING))
 		senderr(ENETDOWN);
 
@@ -294,7 +298,7 @@ atm_ifattach(struct ifnet *ifp)
 	ifp->if_hdrlen = 0;
 	ifp->if_dlt = DLT_ATM_RFC1483;
 	ifp->if_mtu = ATMMTU;
-	ifp->if_output = atm_output;
+	ifp->if_output = ATM_OUTPUT_LOCK;
 #if 0 /* XXX XXX XXX */
 	ifp->if_input = atm_input;
 #endif
diff --git a/sys/net/if_bridge.c b/sys/net/if_bridge.c
index 64354d6..5e421e0 100644
--- a/sys/net/if_bridge.c
+++ b/sys/net/if_bridge.c
@@ -1378,6 +1378,7 @@ bridge_enqueue(struct bridge_softc *sc, struct ifnet *dst_ifp, struct mbuf *m,
 	}
 
 #ifdef ALTQ
+	KERNEL_LOCK(1, NULL);
 	/*
 	 * If ALTQ is enabled on the member interface, do
 	 * classification; the queueing discipline might
@@ -1388,6 +1389,7 @@ bridge_enqueue(struct bridge_softc *sc, struct ifnet *dst_ifp, struct mbuf *m,
 		/* XXX IFT_ETHER */
 		altq_etherclassify(&dst_ifp->if_snd, m);
 	}
+	KERNEL_UNLOCK_ONE(NULL);
 #endif /* ALTQ */
 
 	len = m->m_pkthdr.len;
diff --git a/sys/net/if_ecosubr.c b/sys/net/if_ecosubr.c
index 0987661..b0c403c 100644
--- a/sys/net/if_ecosubr.c
+++ b/sys/net/if_ecosubr.c
@@ -111,6 +111,10 @@ static void eco_defer(struct ifnet *, struct mbuf *, int);
 static void eco_retry_free(struct eco_retry *er);
 static void eco_retry(void *);
 
+static IFNET_WRAP_IF_OUTPUT_DECL(eco_output);
+static IFNET_WRAP_IF_OUTPUT_IMPL(eco_output);
+#define ECO_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(eco_output)
+
 void
 eco_ifattach(struct ifnet *ifp, const uint8_t *lla)
 {
@@ -122,7 +126,7 @@ eco_ifattach(struct ifnet *ifp, const uint8_t *lla)
 	ifp->if_dlt = DLT_ECONET;
 	ifp->if_mtu = ECO_MTU;
 
-	ifp->if_output	 = eco_output;
+	ifp->if_output	 = ECO_OUTPUT_LOCK;
 	ifp->if_input	 = eco_input;
 	ifp->if_start	 = eco_start;
 	ifp->if_ioctl	 = eco_ioctl;
@@ -179,6 +183,8 @@ eco_output(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
 	struct eco_arp *ecah;
 #endif
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) != (IFF_UP|IFF_RUNNING))
 		senderr(ENETDOWN);
 	/*
diff --git a/sys/net/if_ethersubr.c b/sys/net/if_ethersubr.c
index 95a248f..20a812d 100644
--- a/sys/net/if_ethersubr.c
+++ b/sys/net/if_ethersubr.c
@@ -204,10 +204,6 @@ ether_output(struct ifnet * const ifp0, struct mbuf * const m0,
 	struct at_ifaddr *aa;
 #endif /* NETATALK */
 
-#ifndef NET_MPSAFE
-	KASSERT(KERNEL_LOCKED_P());
-#endif
-
 #ifdef MBUFTRACE
 	m_claimm(m, ifp->if_mowner);
 #endif
@@ -238,13 +234,17 @@ ether_output(struct ifnet * const ifp0, struct mbuf * const m0,
 
 #ifdef INET
 	case AF_INET:
+		KERNEL_LOCK(1, NULL);
 		if (m->m_flags & M_BCAST)
 			(void)memcpy(edst, etherbroadcastaddr, sizeof(edst));
 		else if (m->m_flags & M_MCAST)
 			ETHER_MAP_IP_MULTICAST(&satocsin(dst)->sin_addr, edst);
 		else if ((error = arpresolve(ifp, rt, m, dst, edst,
-		    sizeof(edst))) != 0)
+		    sizeof(edst))) != 0) {
+			KERNEL_UNLOCK_ONE(NULL);
 			return error == EWOULDBLOCK ? 0 : error;
+		}
+		KERNEL_UNLOCK_ONE(NULL);
 		/* If broadcasting on a simplex interface, loopback a copy */
 		if ((m->m_flags & M_BCAST) && (ifp->if_flags & IFF_SIMPLEX))
 			mcopy = m_copy(m, 0, (int)M_COPYALL);
@@ -292,10 +292,12 @@ ether_output(struct ifnet * const ifp0, struct mbuf * const m0,
 #endif
 #ifdef NETATALK
     case AF_APPLETALK:
+		KERNEL_LOCK(1, NULL);
 		if (aarpresolve(ifp, m, (const struct sockaddr_at *)dst, edst)) {
 #ifdef NETATALKDEBUG
 			printf("aarpresolv failed\n");
 #endif /* NETATALKDEBUG */
+			KERNEL_UNLOCK_ONE(NULL);
 			return (0);
 		}
 		/*
@@ -303,8 +305,10 @@ ether_output(struct ifnet * const ifp0, struct mbuf * const m0,
 		 */
 		aa = (struct at_ifaddr *) at_ifawithnet(
 		    (const struct sockaddr_at *)dst, ifp);
-		if (aa == NULL)
+		if (aa == NULL) {
+		    KERNEL_UNLOCK_ONE(NULL);
 		    goto bad;
+		}
 
 		/*
 		 * In the phase 2 case, we need to prepend an mbuf for the
@@ -325,6 +329,7 @@ ether_output(struct ifnet * const ifp0, struct mbuf * const m0,
 		} else {
 			etype = htons(ETHERTYPE_ATALK);
 		}
+		KERNEL_UNLOCK_ONE(NULL);
 		break;
 #endif /* NETATALK */
 	case pseudo_AF_HDRCMPLT:
@@ -349,6 +354,7 @@ ether_output(struct ifnet * const ifp0, struct mbuf * const m0,
 	}
 
 #ifdef MPLS
+	KERNEL_LOCK(1, NULL);
 	{
 		struct m_tag *mtag;
 		mtag = m_tag_find(m, PACKET_TAG_MPLS, NULL);
@@ -358,6 +364,7 @@ ether_output(struct ifnet * const ifp0, struct mbuf * const m0,
 			m_tag_delete(m, mtag);
 		}
 	}
+	KERNEL_UNLOCK_ONE(NULL);
 #endif
 
 	if (mcopy)
@@ -410,6 +417,7 @@ ether_output(struct ifnet * const ifp0, struct mbuf * const m0,
 #endif /* NCARP > 0 */
 
 #ifdef ALTQ
+	KERNEL_LOCK(1, NULL);
 	/*
 	 * If ALTQ is enabled on the parent interface, do
 	 * classification; the queueing discipline might not
@@ -418,6 +426,7 @@ ether_output(struct ifnet * const ifp0, struct mbuf * const m0,
 	 */
 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
 		altq_etherclassify(&ifp->if_snd, m);
+	KERNEL_UNLOCK_ONE(NULL);
 #endif
 	return ifq_enqueue(ifp, m);
 
diff --git a/sys/net/if_faith.c b/sys/net/if_faith.c
index 7c5e191..3f8f781 100644
--- a/sys/net/if_faith.c
+++ b/sys/net/if_faith.c
@@ -98,6 +98,11 @@ static struct if_clone faith_cloner =
 
 #define	FAITHMTU	1500
 
+static IFNET_WRAP_IF_OUTPUT_DECL(faithoutput);
+static IFNET_WRAP_IF_OUTPUT_IMPL(faithoutput);
+#define FAITHOUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(faithoutput)
+
+
 /* ARGSUSED */
 void
 faithattach(int count)
@@ -119,7 +124,7 @@ faith_clone_create(struct if_clone *ifc, int unit)
 	/* Change to BROADCAST experimentaly to announce its prefix. */
 	ifp->if_flags = /* IFF_LOOPBACK */ IFF_BROADCAST | IFF_MULTICAST;
 	ifp->if_ioctl = faithioctl;
-	ifp->if_output = faithoutput;
+	ifp->if_output = FAITHOUTPUT_LOCK;
 	ifp->if_type = IFT_FAITH;
 	ifp->if_hdrlen = 0;
 	ifp->if_addrlen = 0;
@@ -150,6 +155,8 @@ faithoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
 	int s, error;
 	uint32_t af;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	if ((m->m_flags & M_PKTHDR) == 0)
 		panic("faithoutput no HDR");
 	af = dst->sa_family;
diff --git a/sys/net/if_fddisubr.c b/sys/net/if_fddisubr.c
index 4ce90f2..86571bc 100644
--- a/sys/net/if_fddisubr.c
+++ b/sys/net/if_fddisubr.c
@@ -179,6 +179,10 @@ static	int fddi_output(struct ifnet *, struct mbuf *,
 	    const struct sockaddr *, const struct rtentry *);
 static	void fddi_input(struct ifnet *, struct mbuf *);
 
+static IFNET_WRAP_IF_OUTPUT_DECL(fddi_output);
+static IFNET_WRAP_IF_OUTPUT_IMPL(fddi_output);
+#define FDDI_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(fddi_output)
+
 /*
  * FDDI output routine.
  * Encapsulate a packet of type family for the local net.
@@ -196,6 +200,8 @@ fddi_output(struct ifnet *ifp0, struct mbuf *m0, const struct sockaddr *dst,
 	struct mbuf *mcopy = NULL;
 	struct ifnet *ifp = ifp0;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	MCLAIM(m, ifp->if_mowner);
 
 #if NCARP > 0
@@ -596,7 +602,7 @@ fddi_ifattach(struct ifnet *ifp, void *lla)
 	ifp->if_hdrlen = 21;
 	ifp->if_dlt = DLT_FDDI;
 	ifp->if_mtu = FDDIMTU;
-	ifp->if_output = fddi_output;
+	ifp->if_output = FDDI_OUTPUT_LOCK;
 	ifp->_if_input = fddi_input;
 	ifp->if_baudrate = IF_Mbps(100);
 #ifdef IFF_NOTRAILERS
diff --git a/sys/net/if_gif.c b/sys/net/if_gif.c
index 83b97c8..5d1e380 100644
--- a/sys/net/if_gif.c
+++ b/sys/net/if_gif.c
@@ -35,6 +35,7 @@ __KERNEL_RCSID(0, "$NetBSD: if_gif.c,v 1.110 2016/06/10 13:27:16 ozaki-r Exp $")
 
 #ifdef _KERNEL_OPT
 #include "opt_inet.h"
+#include "opt_net_mpsafe.h"
 #endif
 
 #include <sys/param.h>
@@ -85,6 +86,12 @@ __KERNEL_RCSID(0, "$NetBSD: if_gif.c,v 1.110 2016/06/10 13:27:16 ozaki-r Exp $")
 
 #include "ioconf.h"
 
+#if 0 /* NOTYET */
+#ifdef NET_MPSAFE
+#define GIF_MPSAFE 1
+#endif
+#endif
+
 static void	gifintr(void *);
 
 /*
@@ -125,6 +132,14 @@ static struct if_clone gif_cloner =
 #endif
 static int max_gif_nesting = MAX_GIF_NEST;
 
+#ifndef GIF_MPSAFE
+static IFNET_WRAP_IF_OUTPUT_DECL(gif_output);
+static IFNET_WRAP_IF_OUTPUT_IMPL(gif_output);
+#define GIF_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(gif_output)
+#else
+#define GIF_OUTPUT_LOCK gif_output
+#endif
+
 static void
 gif_sysctl_setup(struct sysctllog **clog)
 {
@@ -187,7 +202,7 @@ gifattach0(struct gif_softc *sc)
 	sc->gif_if.if_mtu    = GIF_MTU;
 	sc->gif_if.if_flags  = IFF_POINTOPOINT | IFF_MULTICAST;
 	sc->gif_if.if_ioctl  = gif_ioctl;
-	sc->gif_if.if_output = gif_output;
+	sc->gif_if.if_output = GIF_OUTPUT_LOCK;
 	sc->gif_if.if_type   = IFT_GIF;
 	sc->gif_if.if_dlt    = DLT_NULL;
 	sc->gif_if.if_softc  = sc;
@@ -320,6 +335,10 @@ gif_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
 	int error = 0;
 	int s;
 
+#ifndef GIF_MPSAFE
+	KASSERT(KERNEL_LOCKED_P());
+#endif
+
 	IFQ_CLASSIFY(&ifp->if_snd, m, dst->sa_family);
 
 	if ((error = gif_check_nesting(ifp, m)) != 0) {
diff --git a/sys/net/if_gre.c b/sys/net/if_gre.c
index 9391bbb..a74212f 100644
--- a/sys/net/if_gre.c
+++ b/sys/net/if_gre.c
@@ -167,6 +167,10 @@ static bool gre_fp_send(struct gre_softc *, enum gre_msg, file_t *);
 static bool gre_fp_recv(struct gre_softc *);
 static void gre_fp_recvloop(void *);
 
+static IFNET_WRAP_IF_OUTPUT_DECL(gre_output);
+static IFNET_WRAP_IF_OUTPUT_IMPL(gre_output);
+#define GRE_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(gre_output)
+
 static void
 gre_bufq_init(struct gre_bufq *bq, size_t len0)
 {
@@ -295,7 +299,7 @@ gre_clone_create(struct if_clone *ifc, int unit)
 	sc->sc_if.if_dlt = DLT_NULL;
 	sc->sc_if.if_mtu = GREMTU;
 	sc->sc_if.if_flags = IFF_POINTOPOINT|IFF_MULTICAST;
-	sc->sc_if.if_output = gre_output;
+	sc->sc_if.if_output = GRE_OUTPUT_LOCK;
 	sc->sc_if.if_ioctl = gre_ioctl;
 	sp = &sc->sc_soparm;
 	sockaddr_copy(sstosa(&sp->sp_dst), sizeof(sp->sp_dst), any);
@@ -885,6 +889,8 @@ gre_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
 	struct gre_h *gh;
 	uint16_t etype = 0;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) != (IFF_UP|IFF_RUNNING)) {
 		m_freem(m);
 		error = ENETDOWN;
diff --git a/sys/net/if_hippisubr.c b/sys/net/if_hippisubr.c
index 8ec5c69..5b15320 100644
--- a/sys/net/if_hippisubr.c
+++ b/sys/net/if_hippisubr.c
@@ -76,6 +76,10 @@ static int	hippi_output(struct ifnet *, struct mbuf *,
 			     const struct sockaddr *, const struct rtentry *);
 static void	hippi_input(struct ifnet *, struct mbuf *);
 
+static IFNET_WRAP_IF_OUTPUT_DECL(hippi_output);
+static IFNET_WRAP_IF_OUTPUT_IMPL(hippi_output);
+#define HIPPI_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(hippi_output)
+
 /*
  * HIPPI output routine.
  * Encapsulate a packet of type family for the local net.
@@ -95,6 +99,8 @@ hippi_output(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
 	uint32_t *cci;
 	uint32_t d2_len;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) != (IFF_UP|IFF_RUNNING))
 		senderr(ENETDOWN);
 
@@ -286,7 +292,7 @@ hippi_ifattach(struct ifnet *ifp, void *lla)
 	ifp->if_hdrlen = sizeof(struct hippi_header) + 8; /* add CCI */
 	ifp->if_dlt = DLT_HIPPI;
 	ifp->if_mtu = HIPPIMTU;
-	ifp->if_output = hippi_output;
+	ifp->if_output = HIPPI_OUTPUT_LOCK;
 	ifp->_if_input = hippi_input;
 	ifp->if_baudrate = IF_Mbps(800);	/* XXX double-check */
 
diff --git a/sys/net/if_ieee1394subr.c b/sys/net/if_ieee1394subr.c
index 58ca37b..f4de8a7 100644
--- a/sys/net/if_ieee1394subr.c
+++ b/sys/net/if_ieee1394subr.c
@@ -82,6 +82,10 @@ static int  ieee1394_output(struct ifnet *, struct mbuf *,
 		const struct sockaddr *, const struct rtentry *);
 static struct mbuf *ieee1394_reass(struct ifnet *, struct mbuf *, uint16_t);
 
+static IFNET_WRAP_IF_OUTPUT_DECL(ieee1394_output);
+static IFNET_WRAP_IF_OUTPUT_IMPL(ieee1394_output);
+#define IEEE1394_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(ieee1394_output)
+
 static int
 ieee1394_output(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
     const struct rtentry *rt)
@@ -98,6 +102,8 @@ ieee1394_output(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
 	struct m_tag *mtag;
 	int unicast;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) != (IFF_UP|IFF_RUNNING))
 		senderr(ENETDOWN);
 
@@ -637,7 +643,7 @@ ieee1394_ifattach(struct ifnet *ifp, const struct ieee1394_hwaddr *hwaddr)
 	ifp->if_hdrlen = sizeof(struct ieee1394_header);
 	ifp->if_dlt = DLT_EN10MB;	/* XXX */
 	ifp->if_mtu = IEEE1394MTU;
-	ifp->if_output = ieee1394_output;
+	ifp->if_output = IEEE1394_OUTPUT_LOCK;
 	ifp->if_drain = ieee1394_drain;
 	ifp->if_watchdog = ieee1394_watchdog;
 	ifp->if_timer = 1;
diff --git a/sys/net/if_loop.c b/sys/net/if_loop.c
index a62c058..9f824d1 100644
--- a/sys/net/if_loop.c
+++ b/sys/net/if_loop.c
@@ -211,12 +211,12 @@ looutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
 	struct ifqueue *ifq = NULL;
 	int s, isr = -1;
 	int csum_flags;
+	int error = 0;
 	size_t pktlen;
 
 	MCLAIM(m, ifp->if_mowner);
-#ifndef NET_MPSAFE
-	KASSERT(KERNEL_LOCKED_P());
-#endif
+
+	KERNEL_LOCK(1, NULL);
 
 	if ((m->m_flags & M_PKTHDR) == 0)
 		panic("looutput: no header mbuf");
@@ -226,8 +226,9 @@ looutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
 
 	if (rt && rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
 		m_freem(m);
-		return (rt->rt_flags & RTF_BLACKHOLE ? 0 :
+		error = (rt->rt_flags & RTF_BLACKHOLE ? 0 :
 			rt->rt_flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
+		goto out;
 	}
 
 	pktlen = m->m_pkthdr.len;
@@ -241,8 +242,6 @@ looutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
 	 */
 	if ((ALTQ_IS_ENABLED(&ifp->if_snd) || TBR_IS_ENABLED(&ifp->if_snd)) &&
 	    ifp->if_start == lostart) {
-		int error;
-
 		/*
 		 * If the queueing discipline needs packet classification,
 		 * do it before prepending the link headers.
@@ -250,12 +249,14 @@ looutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
 		IFQ_CLASSIFY(&ifp->if_snd, m, dst->sa_family);
 
 		M_PREPEND(m, sizeof(uint32_t), M_DONTWAIT);
-		if (m == NULL)
-			return (ENOBUFS);
+		if (m == NULL) {
+			error = ENOBUFS;
+			goto out;
+		}
 		*(mtod(m, uint32_t *)) = dst->sa_family;
 
 		error = ifp->if_transmit(ifp, m);
-		return (error);
+		goto out;
 	}
 #endif /* ALTQ */
 
@@ -310,12 +311,13 @@ looutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
 		printf("%s: can't handle af%d\n", ifp->if_xname,
 		    dst->sa_family);
 		m_freem(m);
-		return (EAFNOSUPPORT);
+		error = EAFNOSUPPORT;
+		goto out;
 	}
 
 	s = splnet();
 	if (__predict_true(pktq)) {
-		int error = 0;
+		error = 0;
 
 		if (__predict_true(pktq_enqueue(pktq, m, 0))) {
 			ifp->if_ipackets++;
@@ -325,20 +327,23 @@ looutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
 			error = ENOBUFS;
 		}
 		splx(s);
-		return error;
+		goto out;;
 	}
 	if (IF_QFULL(ifq)) {
 		IF_DROP(ifq);
 		m_freem(m);
 		splx(s);
-		return (ENOBUFS);
+		error = ENOBUFS;
+		goto out;
 	}
 	IF_ENQUEUE(ifq, m);
 	schednetisr(isr);
 	ifp->if_ipackets++;
 	ifp->if_ibytes += m->m_pkthdr.len;
 	splx(s);
-	return (0);
+out:
+	KERNEL_UNLOCK_ONE(NULL);
+	return error;
 }
 
 #ifdef ALTQ
diff --git a/sys/net/if_mpls.c b/sys/net/if_mpls.c
index 96dd7f2..5111dcb 100644
--- a/sys/net/if_mpls.c
+++ b/sys/net/if_mpls.c
@@ -111,6 +111,10 @@ extern int mpls_defttl, mpls_mapttl_inet, mpls_mapttl_inet6, mpls_icmp_respond,
 	mpls_forwarding, mpls_frame_accept, mpls_mapprec_inet, mpls_mapclass_inet6,
 	mpls_rfc4182;
 
+static IFNET_WRAP_IF_OUTPUT_DECL(mpls_output);
+static IFNET_WRAP_IF_OUTPUT_IMPL(mpls_output);
+#define MPLS_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(mpls_output)
+
 /* ARGSUSED */
 void
 ifmplsattach(int count)
@@ -134,7 +138,7 @@ mpls_clone_create(struct if_clone *ifc, int unit)
 	sc->sc_if.if_mtu = 1500;
 	sc->sc_if.if_flags = 0;
 	sc->sc_if._if_input = mpls_input;
-	sc->sc_if.if_output = mpls_output;
+	sc->sc_if.if_output = MPLS_OUTPUT_LOCK;
 	sc->sc_if.if_ioctl = mpls_ioctl;
 
 	if_attach(&sc->sc_if);
diff --git a/sys/net/if_ppp.c b/sys/net/if_ppp.c
index 30e3aa9..89b021f 100644
--- a/sys/net/if_ppp.c
+++ b/sys/net/if_ppp.c
@@ -223,6 +223,7 @@ static struct compressor *ppp_get_compressor(uint8_t);
 static void ppp_compressor_rele(struct compressor *);
 #endif /* PPP_COMPRESS */
 
+IFNET_WRAP_IF_OUTPUT_IMPL(pppoutput);
 
 /*
  * Called from boot code to establish ppp interfaces.
@@ -295,7 +296,7 @@ ppp_create(const char *name, int unit)
 	sc->sc_if.if_hdrlen = PPP_HDRLEN;
 	sc->sc_if.if_dlt = DLT_NULL;
 	sc->sc_if.if_ioctl = pppsioctl;
-	sc->sc_if.if_output = pppoutput;
+	sc->sc_if.if_output = PPPOUTPUT_LOCK;
 #ifdef ALTQ
 	sc->sc_if.if_start = ppp_ifstart;
 #endif
@@ -852,6 +853,8 @@ pppoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
 	enum NPmode mode;
 	int len;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	    if (sc->sc_devp == NULL || (ifp->if_flags & IFF_RUNNING) == 0
 		|| ((ifp->if_flags & IFF_UP) == 0 && dst->sa_family != AF_UNSPEC)) {
 		    error = ENETDOWN;	/* sort of */
diff --git a/sys/net/if_pppvar.h b/sys/net/if_pppvar.h
index e8e2ebd..0addcff 100644
--- a/sys/net/if_pppvar.h
+++ b/sys/net/if_pppvar.h
@@ -160,6 +160,11 @@ void	ppppktin(struct ppp_softc *, struct mbuf *, int);
 struct	mbuf *ppp_dequeue(struct ppp_softc *);
 int	pppoutput(struct ifnet *, struct mbuf *, const struct sockaddr *,
 	    const struct rtentry *);
+
+IFNET_WRAP_IF_OUTPUT_DECL(pppoutput);
+#define PPPOUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(pppoutput)
+
+
 #endif /* _KERNEL */
 
 #endif /* !_NET_IF_PPPVAR_H_ */
diff --git a/sys/net/if_sl.c b/sys/net/if_sl.c
index 460d3ab..5875e86 100644
--- a/sys/net/if_sl.c
+++ b/sys/net/if_sl.c
@@ -202,6 +202,10 @@ static int	sloutput(struct ifnet *, struct mbuf *, const struct sockaddr *,
 static int	slstart(struct tty *);
 static int	sltioctl(struct tty *, u_long, void *, int, struct lwp *);
 
+static IFNET_WRAP_IF_OUTPUT_DECL(sloutput);
+static IFNET_WRAP_IF_OUTPUT_IMPL(sloutput);
+#define SLOUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(sloutput)
+
 static struct linesw slip_disc = {
 	.l_name = "slip",
 	.l_open = slopen,
@@ -238,7 +242,7 @@ sl_clone_create(struct if_clone *ifc, int unit)
 	sc->sc_if.if_flags = IFF_POINTOPOINT | SC_AUTOCOMP | IFF_MULTICAST;
 	sc->sc_if.if_type = IFT_SLIP;
 	sc->sc_if.if_ioctl = slioctl;
-	sc->sc_if.if_output = sloutput;
+	sc->sc_if.if_output = SLOUTPUT_LOCK;
 	sc->sc_if.if_dlt = DLT_SLIP;
 	sc->sc_fastq.ifq_maxlen = 32;
 	IFQ_SET_READY(&sc->sc_if.if_snd);
@@ -440,6 +444,8 @@ sloutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
 	struct ifqueue *ifq = NULL;
 	int s, error;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	IFQ_CLASSIFY(&ifp->if_snd, m, dst->sa_family);
 
 	/*
diff --git a/sys/net/if_spppsubr.c b/sys/net/if_spppsubr.c
index 41afd19..ed4d0f5 100644
--- a/sys/net/if_spppsubr.c
+++ b/sys/net/if_spppsubr.c
@@ -380,6 +380,10 @@ static void sppp_gen_ip6_addr(struct sppp *sp, const struct in6_addr *src);
 static void sppp_suggest_ip6_addr(struct sppp *sp, struct in6_addr *src);
 #endif
 
+static IFNET_WRAP_IF_OUTPUT_DECL(sppp_output);
+static IFNET_WRAP_IF_OUTPUT_IMPL(sppp_output);
+#define SPPP_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(sppp_output)
+
 /* our control protocol descriptors */
 static const struct cp lcp = {
 	PPP_LCP, IDX_LCP, CP_LCP, "lcp",
@@ -648,6 +652,8 @@ sppp_output(struct ifnet *ifp, struct mbuf *m,
 	int s, error = 0;
 	uint16_t protocol;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	s = splnet();
 
 	sp->pp_last_activity = time_uptime;
@@ -855,7 +861,7 @@ sppp_attach(struct ifnet *ifp)
 	spppq = sp;
 
 	sp->pp_if.if_type = IFT_PPP;
-	sp->pp_if.if_output = sppp_output;
+	sp->pp_if.if_output = SPPP_OUTPUT_LOCK;
 	sp->pp_fastq.ifq_maxlen = 32;
 	sp->pp_cpq.ifq_maxlen = 20;
 	sp->pp_loopcnt = 0;
diff --git a/sys/net/if_srt.c b/sys/net/if_srt.c
index 117b1c9..ef2f2f9 100644
--- a/sys/net/if_srt.c
+++ b/sys/net/if_srt.c
@@ -235,6 +235,9 @@ srt_if_output(
 	return (*r->u.dstifp->if_output)(r->u.dstifp,m,&r->dst.sa,0);
 }
 
+static IFNET_WRAP_IF_OUTPUT_IMPL(srt_if_output);
+#define SRT_IF_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(srf_if_output)
+
 static int
 srt_clone_create(struct if_clone *cl, int unit)
 {
@@ -256,7 +259,7 @@ srt_clone_create(struct if_clone *cl, int unit)
 	sc->intf.if_flags = IFF_POINTOPOINT;
 	sc->intf.if_type = IFT_OTHER;
 	sc->intf.if_ioctl = &srt_if_ioctl;
-	sc->intf.if_output = &srt_if_output;
+	sc->intf.if_output = &SRT_IF_OUTPUT_LOCK;
 	sc->intf.if_dlt = DLT_RAW;
 	if_attach(&sc->intf);
 	if_alloc_sadl(&sc->intf);
diff --git a/sys/net/if_stf.c b/sys/net/if_stf.c
index 25c76f4..35d796c 100644
--- a/sys/net/if_stf.c
+++ b/sys/net/if_stf.c
@@ -177,6 +177,10 @@ static int stf_checkaddr6(struct stf_softc *, const struct in6_addr *,
 static void stf_rtrequest(int, struct rtentry *, const struct rt_addrinfo *);
 static int stf_ioctl(struct ifnet *, u_long, void *);
 
+static IFNET_WRAP_IF_OUTPUT_DECL(stf_output);
+static IFNET_WRAP_IF_OUTPUT_IMPL(stf_output);
+#define STF_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(stf_output)
+
 /* ARGSUSED */
 void
 stfattach(int count)
@@ -211,7 +215,7 @@ stf_clone_create(struct if_clone *ifc, int unit)
 	sc->sc_if.if_mtu    = STF_MTU;
 	sc->sc_if.if_flags  = 0;
 	sc->sc_if.if_ioctl  = stf_ioctl;
-	sc->sc_if.if_output = stf_output;
+	sc->sc_if.if_output = STF_OUTPUT_LOCK;
 	sc->sc_if.if_type   = IFT_STF;
 	sc->sc_if.if_dlt    = DLT_NULL;
 	if_attach(&sc->sc_if);
@@ -340,6 +344,8 @@ stf_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
 		struct sockaddr_in	dst4;
 	} u;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	sc = (struct stf_softc*)ifp;
 	dst6 = (const struct sockaddr_in6 *)dst;
 
diff --git a/sys/net/if_strip.c b/sys/net/if_strip.c
index bc36188b..ba28196 100644
--- a/sys/net/if_strip.c
+++ b/sys/net/if_strip.c
@@ -337,6 +337,10 @@ static int	stripoutput(struct ifnet *,
 static int	stripstart(struct tty *);
 static int	striptioctl(struct tty *, u_long, void *, int, struct lwp *);
 
+static IFNET_WRAP_IF_OUTPUT_DECL(stripoutput);
+static IFNET_WRAP_IF_OUTPUT_IMPL(stripoutput);
+#define STRIPOUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(stripoutput)
+
 static struct linesw strip_disc = {
 	.l_name = "strip",
 	.l_open = stripopen,
@@ -376,7 +380,7 @@ strip_clone_create(struct if_clone *ifc, int unit)
 #endif
 	sc->sc_if.if_type = IFT_SLIP;
 	sc->sc_if.if_ioctl = stripioctl;
-	sc->sc_if.if_output = stripoutput;
+	sc->sc_if.if_output = STRIPOUTPUT_LOCK;
 	sc->sc_if.if_dlt = DLT_SLIP;
 	sc->sc_fastq.ifq_maxlen = 32;
 	IFQ_SET_READY(&sc->sc_if.if_snd);
@@ -732,6 +736,8 @@ stripoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
 	int s, error;
 	u_char dl_addrbuf[STARMODE_ADDR_LEN+1];
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	/*
 	 * Verify tty line is up and alive.
 	 */
diff --git a/sys/net/if_tokensubr.c b/sys/net/if_tokensubr.c
index 8ddd016..d75739b 100644
--- a/sys/net/if_tokensubr.c
+++ b/sys/net/if_tokensubr.c
@@ -146,6 +146,10 @@ static int	token_output(struct ifnet *, struct mbuf *,
 			     const struct sockaddr *, const struct rtentry *);
 static void	token_input(struct ifnet *, struct mbuf *);
 
+static IFNET_WRAP_IF_OUTPUT_DECL(token_output);
+static IFNET_WRAP_IF_OUTPUT_IMPL(token_output);
+#define TOKEN_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(token_output)
+
 /*
  * Token Ring output routine.
  * Encapsulate a packet of type family for the local net.
@@ -170,6 +174,8 @@ token_output(struct ifnet *ifp0, struct mbuf *m0, const struct sockaddr *dst,
 	struct ifnet *ifp = ifp0;
 	size_t riflen = 0;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 #if NCARP > 0
 	if (ifp->if_type == IFT_CARP) {
 		struct ifaddr *ifa;
@@ -499,7 +505,7 @@ token_ifattach(struct ifnet *ifp, void *lla)
 	ifp->if_hdrlen = 14;
 	ifp->if_dlt = DLT_IEEE802;
 	ifp->if_mtu = ISO88025_MTU;
-	ifp->if_output = token_output;
+	ifp->if_output = TOKEN_OUTPUT_LOCK;
 	ifp->_if_input = token_input;
 	ifp->if_broadcastaddr = tokenbroadcastaddr;
 #ifdef IFF_NOTRAILERS
diff --git a/sys/net/if_tun.c b/sys/net/if_tun.c
index dacf4e7..c9ddd40 100644
--- a/sys/net/if_tun.c
+++ b/sys/net/if_tun.c
@@ -90,6 +90,10 @@ static void tunstart(struct ifnet *);
 static struct tun_softc *tun_find_unit(dev_t);
 static struct tun_softc *tun_find_zunit(int);
 
+static IFNET_WRAP_IF_OUTPUT_DECL(tun_output);
+static IFNET_WRAP_IF_OUTPUT_IMPL(tun_output);
+#define TUN_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(tun_output)
+
 static dev_type_open(tunopen);
 static dev_type_close(tunclose);
 static dev_type_read(tunread);
@@ -208,7 +212,7 @@ tunattach0(struct tun_softc *tp)
 	ifp->if_softc = tp;
 	ifp->if_mtu = TUNMTU;
 	ifp->if_ioctl = tun_ioctl;
-	ifp->if_output = tun_output;
+	ifp->if_output = TUN_OUTPUT_LOCK;
 #ifdef ALTQ
 	ifp->if_start = tunstart;
 #endif
@@ -504,6 +508,8 @@ tun_output(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
 	uint32_t	*af;
 #endif
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	s = splnet();
 	mutex_enter(&tp->tun_lock);
 	TUNDEBUG ("%s: tun_output\n", ifp->if_xname);
diff --git a/sys/net80211/ieee80211_output.c b/sys/net80211/ieee80211_output.c
index 1ce0782..04b9d3a 100644
--- a/sys/net80211/ieee80211_output.c
+++ b/sys/net80211/ieee80211_output.c
@@ -221,7 +221,7 @@ ieee80211_mgmt_output(struct ieee80211com *ic, struct ieee80211_node *ni,
 		ic->ic_mgt_timer = timer;
 		ifp->if_timer = 1;
 	}
-	(*ifp->if_start)(ifp);
+	if_start_lock(ifp);
 	return 0;
 }
 
@@ -268,7 +268,7 @@ ieee80211_send_nulldata(struct ieee80211_node *ni)
 	    wh->i_fc[1] & IEEE80211_FC1_PWR_MGT ? "ena" : "dis");
 
 	IF_ENQUEUE(&ic->ic_mgtq, m);		/* cheat */
-	(*ifp->if_start)(ifp);
+	if_start_lock(ifp);
 
 	return 0;
 }
@@ -1362,7 +1362,7 @@ ieee80211_send_probereq(struct ieee80211_node *ni,
 	    ieee80211_chan2ieee(ic, ic->ic_curchan));
 
 	IF_ENQUEUE(&ic->ic_mgtq, m);
-	(*ic->ic_ifp->if_start)(ic->ic_ifp);
+	if_start_lock(ic->ic_ifp);
 	return 0;
 }
 
diff --git a/sys/net80211/ieee80211_proto.c b/sys/net80211/ieee80211_proto.c
index 9a48ea8..a028411 100644
--- a/sys/net80211/ieee80211_proto.c
+++ b/sys/net80211/ieee80211_proto.c
@@ -1121,7 +1121,7 @@ ieee80211_newstate(struct ieee80211com *ic, enum ieee80211_state nstate, int arg
 			if (ic->ic_opmode == IEEE80211_M_STA)
 				ieee80211_notify_node_join(ic, ni, 
 					arg == IEEE80211_FC0_SUBTYPE_ASSOC_RESP);
-			(*ifp->if_start)(ifp);	/* XXX not authorized yet */
+			if_start_lock(ifp);	/* XXX not authorized yet */
 			break;
 		}
 		/*
diff --git a/sys/netinet/if_arp.c b/sys/netinet/if_arp.c
index 0070f65..79ebd54 100644
--- a/sys/netinet/if_arp.c
+++ b/sys/netinet/if_arp.c
@@ -1888,9 +1888,7 @@ revarprequest(struct ifnet *ifp)
 	sa.sa_len = 2;
 	m->m_flags |= M_BCAST;
 
-	KERNEL_LOCK(1, NULL);
 	(*ifp->if_output)(ifp, m, &sa, NULL);
-	KERNEL_UNLOCK_ONE(NULL);
 }
 
 /*
diff --git a/sys/netinet/ip_carp.c b/sys/netinet/ip_carp.c
index 7b9adad3..1d20cf5 100644
--- a/sys/netinet/ip_carp.c
+++ b/sys/netinet/ip_carp.c
@@ -236,6 +236,8 @@ static void	sysctl_net_inet_carp_setup(struct sysctllog **);
 struct if_clone carp_cloner =
     IF_CLONE_INITIALIZER("carp", carp_clone_create, carp_clone_destroy);
 
+IFNET_WRAP_IF_OUTPUT_IMPL(carp_output);
+
 static __inline u_int16_t
 carp_cksum(struct mbuf *m, int len)
 {
@@ -809,7 +811,7 @@ carp_clone_create(struct if_clone *ifc, int unit)
 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
 	ifp->if_ioctl = carp_ioctl;
 	ifp->if_start = carp_start;
-	ifp->if_output = carp_output;
+	ifp->if_output = CARP_OUTPUT_LOCK;
 	ifp->if_type = IFT_CARP;
 	ifp->if_addrlen = ETHER_ADDR_LEN;
 	ifp->if_hdrlen = ETHER_HDR_LEN;
diff --git a/sys/netinet/ip_carp.h b/sys/netinet/ip_carp.h
index 8996e79..eac7a91 100644
--- a/sys/netinet/ip_carp.h
+++ b/sys/netinet/ip_carp.h
@@ -164,5 +164,9 @@ struct ifnet	*carp_ourether(void *, struct ether_header *, u_char, int);
 int		 carp_input(struct mbuf *, u_int8_t *, u_int8_t *, u_int16_t);
 int		 carp_output(struct ifnet *, struct mbuf *,
 		     const struct sockaddr *, const struct rtentry *);
+
+IFNET_WRAP_IF_OUTPUT_DECL(carp_output);
+#define CARP_OUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(carp_output)
+
 #endif /* _KERNEL */
 #endif /* _NETINET_IP_CARP_H_ */
diff --git a/sys/netinet/ip_output.c b/sys/netinet/ip_output.c
index 466e2e5..f360d2e 100644
--- a/sys/netinet/ip_output.c
+++ b/sys/netinet/ip_output.c
@@ -212,15 +212,8 @@ ip_if_output(struct ifnet * const ifp, struct mbuf * const m,
 		return error;
 	}
 
-#ifndef NET_MPSAFE
-	KERNEL_LOCK(1, NULL);
-#endif
-
 	error = (*ifp->if_output)(ifp, m, dst, rt);
 
-#ifndef NET_MPSAFE
-	KERNEL_UNLOCK_ONE(NULL);
-#endif
 	return error;
 }
 
diff --git a/sys/netinet6/nd6.c b/sys/netinet6/nd6.c
index 8f5b246..291c7e7 100644
--- a/sys/netinet6/nd6.c
+++ b/sys/netinet6/nd6.c
@@ -2291,16 +2291,10 @@ nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
 	if (ln != NULL)
 		LLE_WUNLOCK(ln);
 
-#ifndef NET_MPSAFE
-	KERNEL_LOCK(1, NULL);
-#endif
 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
 		error = (*ifp->if_output)(origifp, m, sin6tocsa(dst), rt);
 	else
 		error = (*ifp->if_output)(ifp, m, sin6tocsa(dst), rt);
-#ifndef NET_MPSAFE
-	KERNEL_UNLOCK_ONE(NULL);
-#endif
 	goto exit;
 
   bad:
diff --git a/sys/netisdn/i4b_ipr.c b/sys/netisdn/i4b_ipr.c
index 9d9d0da..6a6a0c4 100644
--- a/sys/netisdn/i4b_ipr.c
+++ b/sys/netisdn/i4b_ipr.c
@@ -269,6 +269,10 @@ static void ipr_dialresponse(void *softc, int status, cause_t cause);
 static void ipr_updown(void *softc, int updown);
 static void* ipr_get_softc(int unit);
 
+static IFNET_WRAP_IF_OUTPUT_DECL(iripoutput);
+static IFNET_WRAP_IF_OUTPUT_IMPL(iripoutput);
+#define IRIPOUTPUT_LOCK IFNET_WRAP_IF_OUTPUT_CALL(iripoutput)
+
 static const struct isdn_l4_driver_functions
 ipr_l4_functions = {
 	ipr_rx_data_rdy,
@@ -339,7 +343,7 @@ iripattach(void)
 		sc->sc_if.if_mtu = I4BIPRMTU;
 		sc->sc_if.if_type = IFT_ISDNBASIC;
 		sc->sc_if.if_ioctl = iripioctl;
-		sc->sc_if.if_output = iripoutput;
+		sc->sc_if.if_output = IRIPOUTPUT_LOCK;
 
 		IFQ_SET_MAXLEN(&sc->sc_if.if_snd, I4BIPRMAXQLEN);
 		sc->sc_fastq.ifq_maxlen = I4BIPRMAXQLEN;
@@ -423,6 +427,8 @@ iripoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
 	struct ifqueue *ifq = NULL;
 	struct ip *ip;
 
+	KASSERT(KERNEL_LOCKED_P());
+
 	s = splnet();
 
 #if defined(__FreeBSD__) || defined(__bsdi__)
diff --git a/sys/netnatm/natm.c b/sys/netnatm/natm.c
index ca1e329..f14f592 100644
--- a/sys/netnatm/natm.c
+++ b/sys/netnatm/natm.c
@@ -155,7 +155,7 @@ natm_connect(struct socket *so, struct sockaddr *nam, struct lwp *l)
 	if (ifp == NULL || (ifp->if_flags & IFF_RUNNING) == 0) {
 		return ENXIO;
 	}
-	if (ifp->if_output != atm_output) {
+	if (ifp->if_output != ATM_OUTPUT_LOCK) {
 		return EAFNOSUPPORT;
 	}
 
@@ -368,7 +368,7 @@ natm_send(struct socket *so, struct mbuf *m, struct sockaddr *nam,
 	ATM_PH_VPI(aph) = npcb->npcb_vpi;
 	ATM_PH_SETVCI(aph, npcb->npcb_vci);
 	ATM_PH_FLAGS(aph) = (proto == PROTO_NATMAAL5) ? ATM_PH_AAL5 : 0;
-	error = atm_output(npcb->npcb_ifp, m, NULL, NULL);
+	error = ATM_OUTPUT_LOCK(npcb->npcb_ifp, m, NULL, NULL);
 	splx(s);
 
 	return error;
@@ -404,7 +404,7 @@ natm_send(struct socket *so, struct mbuf *m, struct mbuf *nam,
 	ATM_PH_VPI(aph) = npcb->npcb_vpi;
 	ATM_PH_SETVCI(aph, npcb->npcb_vci);
 	ATM_PH_FLAGS(aph) = (proto == PROTO_NATMAAL5) ? ATM_PH_AAL5 : 0;
-	error = atm_output(npcb->npcb_ifp, m, NULL, NULL);
+	error = ATM_OUTPUT_LOCK(npcb->npcb_ifp, m, NULL, NULL);
 	splx(s);
 
 	return error;