split the interface send queue (struct ifqueue) implementation out.

the intention is to make it more clear what belongs to a transmit
queue and what belongs to an interface.

suggested by and ok mpi@

5 files changed, 511 insertions, 441 deletions

diff --git a/sys/conf/files b/sys/conf/files
index 212a2923f3e..199aabe5e3b 100644
--- a/sys/conf/files
+++ b/sys/conf/files
@@ -1,4 +1,4 @@
-#	$OpenBSD: files,v 1.609 2015/12/03 12:42:03 goda Exp $
+#	$OpenBSD: files,v 1.610 2015/12/08 10:06:11 dlg Exp $
 #	$NetBSD: files,v 1.87 1996/05/19 17:17:50 jonathan Exp $
 
 #	@(#)files.newconf	7.5 (Berkeley) 5/10/93
@@ -759,6 +759,7 @@ file net/art.c				art
 file net/bpf.c				bpfilter		needs-count
 file net/bpf_filter.c			bpfilter
 file net/if.c
+file net/ifq.c
 file net/if_ethersubr.c			ether			needs-flag
 file net/if_etherip.c			etherip			needs-flag
 file net/if_spppsubr.c			sppp
diff --git a/sys/net/if.c b/sys/net/if.c
index bc2d21f6bb9..ff469a17b61 100644
--- a/sys/net/if.c
+++ b/sys/net/if.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: if.c,v 1.422 2015/12/05 10:07:55 tedu Exp $	*/
+/*	$OpenBSD: if.c,v 1.423 2015/12/08 10:06:12 dlg Exp $	*/
 /*	$NetBSD: if.c,v 1.35 1996/05/07 05:26:04 thorpej Exp $	*/
 
 /*
@@ -557,68 +557,6 @@ if_start_locked(struct ifnet *ifp)
 	KERNEL_UNLOCK();
 }
 
-static inline unsigned int
-ifq_enter(struct ifqueue *ifq)
-{
-	return (atomic_inc_int_nv(&ifq->ifq_serializer) == 1);
-}
-
-static inline unsigned int
-ifq_leave(struct ifqueue *ifq)
-{
-	if (atomic_cas_uint(&ifq->ifq_serializer, 1, 0) == 1)
-		return (1);
-
-	ifq->ifq_serializer = 1;
-
-	return (0);
-}
-
-void
-if_start_mpsafe(struct ifnet *ifp)
-{
-	struct ifqueue *ifq = &ifp->if_snd;
-
-	if (!ifq_enter(ifq))
-		return;
-
-	do {
-		if (__predict_false(!ISSET(ifp->if_flags, IFF_RUNNING))) {
-			ifq->ifq_serializer = 0;
-			wakeup_one(&ifq->ifq_serializer);
-			return;
-		}
-
-		if (ifq_empty(ifq) || ifq_is_oactive(ifq))
-			continue;
-
-		ifp->if_start(ifp);
-
-	} while (!ifq_leave(ifq));
-}
-
-void
-if_start_barrier(struct ifnet *ifp)
-{
-	struct sleep_state sls;
-	struct ifqueue *ifq = &ifp->if_snd;
-
-	/* this should only be called from converted drivers */
-	KASSERT(ISSET(ifp->if_xflags, IFXF_MPSAFE));
-
-	/* drivers should only call this on the way down */
-	KASSERT(!ISSET(ifp->if_flags, IFF_RUNNING));
-
-	if (ifq->ifq_serializer == 0)
-		return;
-
-	if_start_mpsafe(ifp); /* spin the wheel to guarantee a wakeup */
-	do {
-		sleep_setup(&sls, &ifq->ifq_serializer, PWAIT, "ifbar");
-		sleep_finish(&sls, ifq->ifq_serializer != 0);
-	} while (ifq->ifq_serializer != 0);
-}
-
 int
 if_enqueue(struct ifnet *ifp, struct mbuf *m)
 {
@@ -2768,311 +2706,6 @@ niq_enlist(struct niqueue *niq, struct mbuf_list *ml)
 	return (rv);
 }
 
-/*
- * send queues.
- */
-
-void		*priq_alloc(void *);
-void		 priq_free(void *);
-int		 priq_enq(struct ifqueue *, struct mbuf *);
-struct mbuf	*priq_deq_begin(struct ifqueue *, void **);
-void		 priq_deq_commit(struct ifqueue *, struct mbuf *, void *);
-void		 priq_purge(struct ifqueue *, struct mbuf_list *);
-
-const struct ifq_ops priq_ops = {
-	priq_alloc,
-	priq_free,
-	priq_enq,
-	priq_deq_begin,
-	priq_deq_commit,
-	priq_purge,
-};
-
-const struct ifq_ops * const ifq_priq_ops = &priq_ops;
-
-struct priq_list {
-	struct mbuf		*head;
-	struct mbuf		*tail;
-};
-
-struct priq {
-	struct priq_list	 pq_lists[IFQ_NQUEUES];
-};
-
-void *
-priq_alloc(void *null)
-{
-	return (malloc(sizeof(struct priq), M_DEVBUF, M_WAITOK | M_ZERO));
-}
-
-void
-priq_free(void *pq)
-{
-	free(pq, M_DEVBUF, sizeof(struct priq));
-}
-
-int
-priq_enq(struct ifqueue *ifq, struct mbuf *m)
-{
-	struct priq *pq;
-	struct priq_list *pl;
-
-	if (ifq_len(ifq) >= ifq->ifq_maxlen)
-		return (ENOBUFS);
-
-	pq = ifq->ifq_q;
-	KASSERT(m->m_pkthdr.pf.prio <= IFQ_MAXPRIO);
-	pl = &pq->pq_lists[m->m_pkthdr.pf.prio];
-
-	m->m_nextpkt = NULL;
-	if (pl->tail == NULL)
-		pl->head = m;
-	else
-		pl->tail->m_nextpkt = m;
-	pl->tail = m;
-
-	return (0);
-}
-
-struct mbuf *
-priq_deq_begin(struct ifqueue *ifq, void **cookiep)
-{
-	struct priq *pq = ifq->ifq_q;
-	struct priq_list *pl;
-	unsigned int prio = nitems(pq->pq_lists);
-	struct mbuf *m;
-
-	do {
-		pl = &pq->pq_lists[--prio];
-		m = pl->head;
-		if (m != NULL) {
-			*cookiep = pl;
-			return (m);
-		}
-	} while (prio > 0);
-
-	return (NULL);
-}
-
-void
-priq_deq_commit(struct ifqueue *ifq, struct mbuf *m, void *cookie)
-{
-	struct priq_list *pl = cookie;
-
-	KASSERT(pl->head == m);
-
-	pl->head = m->m_nextpkt;
-	m->m_nextpkt = NULL;
-
-	if (pl->head == NULL)
-		pl->tail = NULL;
-}
-
-void
-priq_purge(struct ifqueue *ifq, struct mbuf_list *ml)
-{
-	struct priq *pq = ifq->ifq_q;
-	struct priq_list *pl;
-	unsigned int prio = nitems(pq->pq_lists);
-	struct mbuf *m, *n;
-
-	do {
-		pl = &pq->pq_lists[--prio];
-
-		for (m = pl->head; m != NULL; m = n) {
-			n = m->m_nextpkt;
-			ml_enqueue(ml, m);
-		}
-
-		pl->head = pl->tail = NULL;
-	} while (prio > 0);
-}
-
-int
-ifq_enqueue_try(struct ifqueue *ifq, struct mbuf *m)
-{
-	int rv;
-
-	mtx_enter(&ifq->ifq_mtx);
-	rv = ifq->ifq_ops->ifqop_enq(ifq, m);
-	if (rv == 0)
-		ifq->ifq_len++;
-	else
-		ifq->ifq_drops++;
-	mtx_leave(&ifq->ifq_mtx);
-
-	return (rv);
-}
-
-int
-ifq_enqueue(struct ifqueue *ifq, struct mbuf *m)
-{
-	int err;
-
-	err = ifq_enqueue_try(ifq, m);
-	if (err != 0)
-		m_freem(m);
-
-	return (err);
-}
-
-struct mbuf *
-ifq_deq_begin(struct ifqueue *ifq)
-{
-	struct mbuf *m = NULL;
-	void *cookie;
-
-	mtx_enter(&ifq->ifq_mtx);
-	if (ifq->ifq_len == 0 ||
-	    (m = ifq->ifq_ops->ifqop_deq_begin(ifq, &cookie)) == NULL) {
-		mtx_leave(&ifq->ifq_mtx);
-		return (NULL);
-	}
-
-	m->m_pkthdr.ph_cookie = cookie;
-
-	return (m);
-}
-
-void
-ifq_deq_commit(struct ifqueue *ifq, struct mbuf *m)
-{
-	void *cookie;
-
-	KASSERT(m != NULL);
-	cookie = m->m_pkthdr.ph_cookie;
-
-	ifq->ifq_ops->ifqop_deq_commit(ifq, m, cookie);
-	ifq->ifq_len--;
-	mtx_leave(&ifq->ifq_mtx);
-}
-
-void
-ifq_deq_rollback(struct ifqueue *ifq, struct mbuf *m)
-{
-	KASSERT(m != NULL);
-
-	mtx_leave(&ifq->ifq_mtx);
-}
-
-struct mbuf *
-ifq_dequeue(struct ifqueue *ifq)
-{
-	struct mbuf *m;
-
-	m = ifq_deq_begin(ifq);
-	if (m == NULL)
-		return (NULL);
-
-	ifq_deq_commit(ifq, m);
-
-	return (m);
-}
-
-unsigned int
-ifq_purge(struct ifqueue *ifq)
-{
-	struct mbuf_list ml = MBUF_LIST_INITIALIZER();
-	unsigned int rv;
-
-	mtx_enter(&ifq->ifq_mtx);
-	ifq->ifq_ops->ifqop_purge(ifq, &ml);
-	rv = ifq->ifq_len;
-	ifq->ifq_len = 0;
-	ifq->ifq_drops += rv;
-	mtx_leave(&ifq->ifq_mtx);
-
-	KASSERT(rv == ml_len(&ml));
-
-	ml_purge(&ml);
-
-	return (rv);
-}
-
-void
-ifq_init(struct ifqueue *ifq)
-{
-	mtx_init(&ifq->ifq_mtx, IPL_NET);
-	ifq->ifq_drops = 0;
-
-	/* default to priq */
-	ifq->ifq_ops = &priq_ops;
-	ifq->ifq_q = priq_ops.ifqop_alloc(NULL);
-
-	ifq->ifq_serializer = 0;
-	ifq->ifq_len = 0;
-
-	if (ifq->ifq_maxlen == 0)
-		ifq_set_maxlen(ifq, IFQ_MAXLEN);
-}
-
-void
-ifq_attach(struct ifqueue *ifq, const struct ifq_ops *newops, void *opsarg)
-{
-	struct mbuf_list ml = MBUF_LIST_INITIALIZER();
-	struct mbuf_list free_ml = MBUF_LIST_INITIALIZER();
-	struct mbuf *m;
-	const struct ifq_ops *oldops;
-	void *newq, *oldq;
-
-	newq = newops->ifqop_alloc(opsarg);
-
-	mtx_enter(&ifq->ifq_mtx);
-	ifq->ifq_ops->ifqop_purge(ifq, &ml);
-	ifq->ifq_len = 0;
-
-	oldops = ifq->ifq_ops;
-	oldq = ifq->ifq_q;
-
-	ifq->ifq_ops = newops;
-	ifq->ifq_q = newq;
-
-	while ((m = ml_dequeue(&ml)) != NULL) {
-		if (ifq->ifq_ops->ifqop_enq(ifq, m) != 0) {
-			ifq->ifq_drops++;
-			ml_enqueue(&free_ml, m);
-		} else
-			ifq->ifq_len++;
-	}
-	mtx_leave(&ifq->ifq_mtx);
-
-	oldops->ifqop_free(oldq);
-
-	ml_purge(&free_ml);
-}
-
-void *
-ifq_q_enter(struct ifqueue *ifq, const struct ifq_ops *ops)
-{
-	mtx_enter(&ifq->ifq_mtx);
-	if (ifq->ifq_ops == ops)
-		return (ifq->ifq_q);
-
-	mtx_leave(&ifq->ifq_mtx);
-
-	return (NULL);
-}
-
-void
-ifq_q_leave(struct ifqueue *ifq, void *q)
-{
-	KASSERT(q == ifq->ifq_q);
-	mtx_leave(&ifq->ifq_mtx);
-}
-
-void
-ifq_destroy(struct ifqueue *ifq)
-{
-	struct mbuf_list ml = MBUF_LIST_INITIALIZER();
-
-	/* don't need to lock because this is the last use of the ifq */
-
-	ifq->ifq_ops->ifqop_purge(ifq, &ml);
-	ifq->ifq_ops->ifqop_free(ifq->ifq_q);
-
-	ml_purge(&ml);
-}
-
 __dead void
 unhandled_af(int af)
 {
diff --git a/sys/net/if_var.h b/sys/net/if_var.h
index 5f57c6920bc..631e4cd9f21 100644
--- a/sys/net/if_var.h
+++ b/sys/net/if_var.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: if_var.h,v 1.64 2015/12/05 16:24:59 mpi Exp $	*/
+/*	$OpenBSD: if_var.h,v 1.65 2015/12/08 10:06:12 dlg Exp $	*/
 /*	$NetBSD: if.h,v 1.23 1996/05/07 02:40:27 thorpej Exp $	*/
 
 /*
@@ -44,6 +44,8 @@
 #include <sys/refcnt.h>
 #include <sys/time.h>
 
+#include <net/ifq.h>
+
 /*
  * Structures defining a network interface, providing a packet
  * transport mechanism (ala level 0 of the PUP protocols).
@@ -91,35 +93,6 @@ struct if_clone {
 	{ { 0 }, name, sizeof(name) - 1, create, destroy }
 
 /*
- * Structure defining the send queue for a network interface.
- */
-
-struct ifqueue;
-
-struct ifq_ops {
-	void			*(*ifqop_alloc)(void *);
-	void			 (*ifqop_free)(void *);
-	int			 (*ifqop_enq)(struct ifqueue *, struct mbuf *);
-	struct mbuf 		*(*ifqop_deq_begin)(struct ifqueue *, void **);
-	void			 (*ifqop_deq_commit)(struct ifqueue *,
-				    struct mbuf *, void *);
-	void	 		 (*ifqop_purge)(struct ifqueue *,
-				    struct mbuf_list *);
-};
-
-struct ifqueue {
-	struct mutex		 ifq_mtx;
-	uint64_t		 ifq_drops;
-	const struct ifq_ops	*ifq_ops;
-	void			*ifq_q;
-	unsigned int		 ifq_len;
-	unsigned int		 ifq_serializer;
-	unsigned int		 ifq_oactive;
-
-	unsigned int		 ifq_maxlen;
-};
-
-/*
  * Structure defining a queue for a network interface.
  *
  * (Would like to call this struct ``if'', but C isn't PL/1.)
@@ -262,48 +235,6 @@ struct ifg_list {
 	TAILQ_ENTRY(ifg_list)	 ifgl_next;
 };
 
-/*
- * Interface send queues.
- */
-
-void		 ifq_init(struct ifqueue *);
-void		 ifq_attach(struct ifqueue *, const struct ifq_ops *, void *);
-void		 ifq_destroy(struct ifqueue *);
-int		 ifq_enqueue_try(struct ifqueue *, struct mbuf *);
-int		 ifq_enqueue(struct ifqueue *, struct mbuf *);
-struct mbuf	*ifq_deq_begin(struct ifqueue *);
-void		 ifq_deq_commit(struct ifqueue *, struct mbuf *);
-void		 ifq_deq_rollback(struct ifqueue *, struct mbuf *);
-struct mbuf	*ifq_dequeue(struct ifqueue *);
-unsigned int	 ifq_purge(struct ifqueue *);
-void		*ifq_q_enter(struct ifqueue *, const struct ifq_ops *);
-void		 ifq_q_leave(struct ifqueue *, void *);
-
-#define	ifq_len(_ifq)			((_ifq)->ifq_len)
-#define	ifq_empty(_ifq)			(ifq_len(_ifq) == 0)
-#define	ifq_set_maxlen(_ifq, _l)	((_ifq)->ifq_maxlen = (_l))
-
-static inline void
-ifq_set_oactive(struct ifqueue *ifq)
-{
-	ifq->ifq_oactive = 1;
-}
-
-static inline void
-ifq_clr_oactive(struct ifqueue *ifq)
-{
-	ifq->ifq_oactive = 0;
-}
-
-static inline unsigned int
-ifq_is_oactive(struct ifqueue *ifq)
-{
-	return (ifq->ifq_oactive);
-}
-
-extern const struct ifq_ops * const ifq_priq_ops;
-
-#define	IFQ_MAXLEN	256
 #define	IFNET_SLOWHZ	1		/* granularity is 1 second */
 
 /*
diff --git a/sys/net/ifq.c b/sys/net/ifq.c
new file mode 100644
index 00000000000..51d41dac4d9
--- /dev/null
+++ b/sys/net/ifq.c
@@ -0,0 +1,409 @@
+/*	$OpenBSD: ifq.c,v 1.1 2015/12/08 10:06:12 dlg Exp $ */
+
+/*
+ * Copyright (c) 2015 David Gwynne <dlg@openbsd.org>
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/socket.h>
+#include <sys/mbuf.h>
+#include <sys/proc.h>
+#include <sys/atomic.h>
+
+#include <net/if.h>
+#include <net/if_var.h>
+
+/*
+ * priq glue
+ */
+void		*priq_alloc(void *);
+void		 priq_free(void *);
+int		 priq_enq(struct ifqueue *, struct mbuf *);
+struct mbuf	*priq_deq_begin(struct ifqueue *, void **);
+void		 priq_deq_commit(struct ifqueue *, struct mbuf *, void *);
+void		 priq_purge(struct ifqueue *, struct mbuf_list *);
+
+const struct ifq_ops priq_ops = {
+	priq_alloc,
+	priq_free,
+	priq_enq,
+	priq_deq_begin,
+	priq_deq_commit,
+	priq_purge,
+};
+
+const struct ifq_ops * const ifq_priq_ops = &priq_ops;
+
+/*
+ * priq internal structures
+ */
+
+struct priq_list {
+	struct mbuf		*head;
+	struct mbuf		*tail;
+};
+
+struct priq {
+	struct priq_list	 pq_lists[IFQ_NQUEUES];
+};
+
+/*
+ * ifqueue serialiser
+ */
+
+static inline unsigned int
+ifq_enter(struct ifqueue *ifq)
+{
+	return (atomic_inc_int_nv(&ifq->ifq_serializer) == 1);
+}
+
+static inline unsigned int
+ifq_leave(struct ifqueue *ifq)
+{
+	if (atomic_cas_uint(&ifq->ifq_serializer, 1, 0) == 1)
+		return (1);
+
+	ifq->ifq_serializer = 1;
+
+	return (0);
+}
+
+void
+if_start_mpsafe(struct ifnet *ifp)
+{
+	struct ifqueue *ifq = &ifp->if_snd;
+
+	if (!ifq_enter(ifq))
+		return;
+
+	do {
+		if (__predict_false(!ISSET(ifp->if_flags, IFF_RUNNING))) {
+			ifq->ifq_serializer = 0;
+			wakeup_one(&ifq->ifq_serializer);
+			return;
+		}
+
+		if (ifq_empty(ifq) || ifq_is_oactive(ifq))
+			continue;
+
+		ifp->if_start(ifp);
+
+	} while (!ifq_leave(ifq));
+}
+
+void
+if_start_barrier(struct ifnet *ifp)
+{
+	struct sleep_state sls;
+	struct ifqueue *ifq = &ifp->if_snd;
+
+	/* this should only be called from converted drivers */
+	KASSERT(ISSET(ifp->if_xflags, IFXF_MPSAFE));
+
+	/* drivers should only call this on the way down */
+	KASSERT(!ISSET(ifp->if_flags, IFF_RUNNING));
+
+	if (ifq->ifq_serializer == 0)
+		return;
+
+	if_start_mpsafe(ifp); /* spin the wheel to guarantee a wakeup */
+	do {
+		sleep_setup(&sls, &ifq->ifq_serializer, PWAIT, "ifbar");
+		sleep_finish(&sls, ifq->ifq_serializer != 0);
+	} while (ifq->ifq_serializer != 0);
+}
+
+/*
+ * ifqueue mbuf queue API
+ */
+
+void
+ifq_init(struct ifqueue *ifq)
+{
+	mtx_init(&ifq->ifq_mtx, IPL_NET);
+	ifq->ifq_drops = 0;
+
+	/* default to priq */
+	ifq->ifq_ops = &priq_ops;
+	ifq->ifq_q = priq_ops.ifqop_alloc(NULL);
+
+	ifq->ifq_serializer = 0;
+	ifq->ifq_len = 0;
+
+	if (ifq->ifq_maxlen == 0)
+		ifq_set_maxlen(ifq, IFQ_MAXLEN);
+}
+
+void
+ifq_attach(struct ifqueue *ifq, const struct ifq_ops *newops, void *opsarg)
+{
+	struct mbuf_list ml = MBUF_LIST_INITIALIZER();
+	struct mbuf_list free_ml = MBUF_LIST_INITIALIZER();
+	struct mbuf *m;
+	const struct ifq_ops *oldops;
+	void *newq, *oldq;
+
+	newq = newops->ifqop_alloc(opsarg);
+
+	mtx_enter(&ifq->ifq_mtx);
+	ifq->ifq_ops->ifqop_purge(ifq, &ml);
+	ifq->ifq_len = 0;
+
+	oldops = ifq->ifq_ops;
+	oldq = ifq->ifq_q;
+
+	ifq->ifq_ops = newops;
+	ifq->ifq_q = newq;
+
+	while ((m = ml_dequeue(&ml)) != NULL) {
+		if (ifq->ifq_ops->ifqop_enq(ifq, m) != 0) {
+			ifq->ifq_drops++;
+			ml_enqueue(&free_ml, m);
+		} else
+			ifq->ifq_len++;
+	}
+	mtx_leave(&ifq->ifq_mtx);
+
+	oldops->ifqop_free(oldq);
+
+	ml_purge(&free_ml);
+}
+
+void
+ifq_destroy(struct ifqueue *ifq)
+{
+	struct mbuf_list ml = MBUF_LIST_INITIALIZER();
+
+	/* don't need to lock because this is the last use of the ifq */
+
+	ifq->ifq_ops->ifqop_purge(ifq, &ml);
+	ifq->ifq_ops->ifqop_free(ifq->ifq_q);
+
+	ml_purge(&ml);
+}
+
+int
+ifq_enqueue_try(struct ifqueue *ifq, struct mbuf *m)
+{
+	int rv;
+
+	mtx_enter(&ifq->ifq_mtx);
+	rv = ifq->ifq_ops->ifqop_enq(ifq, m);
+	if (rv == 0)
+		ifq->ifq_len++;
+	else
+		ifq->ifq_drops++;
+	mtx_leave(&ifq->ifq_mtx);
+
+	return (rv);
+}
+
+int
+ifq_enqueue(struct ifqueue *ifq, struct mbuf *m)
+{
+	int err;
+
+	err = ifq_enqueue_try(ifq, m);
+	if (err != 0)
+		m_freem(m);
+
+	return (err);
+}
+
+struct mbuf *
+ifq_deq_begin(struct ifqueue *ifq)
+{
+	struct mbuf *m = NULL;
+	void *cookie;
+
+	mtx_enter(&ifq->ifq_mtx);
+	if (ifq->ifq_len == 0 ||
+	    (m = ifq->ifq_ops->ifqop_deq_begin(ifq, &cookie)) == NULL) {
+		mtx_leave(&ifq->ifq_mtx);
+		return (NULL);
+	}
+
+	m->m_pkthdr.ph_cookie = cookie;
+
+	return (m);
+}
+
+void
+ifq_deq_commit(struct ifqueue *ifq, struct mbuf *m)
+{
+	void *cookie;
+
+	KASSERT(m != NULL);
+	cookie = m->m_pkthdr.ph_cookie;
+
+	ifq->ifq_ops->ifqop_deq_commit(ifq, m, cookie);
+	ifq->ifq_len--;
+	mtx_leave(&ifq->ifq_mtx);
+}
+
+void
+ifq_deq_rollback(struct ifqueue *ifq, struct mbuf *m)
+{
+	KASSERT(m != NULL);
+
+	mtx_leave(&ifq->ifq_mtx);
+}
+
+struct mbuf *
+ifq_dequeue(struct ifqueue *ifq)
+{
+	struct mbuf *m;
+
+	m = ifq_deq_begin(ifq);
+	if (m == NULL)
+		return (NULL);
+
+	ifq_deq_commit(ifq, m);
+
+	return (m);
+}
+
+unsigned int
+ifq_purge(struct ifqueue *ifq)
+{
+	struct mbuf_list ml = MBUF_LIST_INITIALIZER();
+	unsigned int rv;
+
+	mtx_enter(&ifq->ifq_mtx);
+	ifq->ifq_ops->ifqop_purge(ifq, &ml);
+	rv = ifq->ifq_len;
+	ifq->ifq_len = 0;
+	ifq->ifq_drops += rv;
+	mtx_leave(&ifq->ifq_mtx);
+
+	KASSERT(rv == ml_len(&ml));
+
+	ml_purge(&ml);
+
+	return (rv);
+}
+
+void *
+ifq_q_enter(struct ifqueue *ifq, const struct ifq_ops *ops)
+{
+	mtx_enter(&ifq->ifq_mtx);
+	if (ifq->ifq_ops == ops)
+		return (ifq->ifq_q);
+
+	mtx_leave(&ifq->ifq_mtx);
+
+	return (NULL);
+}
+
+void
+ifq_q_leave(struct ifqueue *ifq, void *q)
+{
+	KASSERT(q == ifq->ifq_q);
+	mtx_leave(&ifq->ifq_mtx);
+}
+
+/*
+ * priq implementation
+ */
+
+void *
+priq_alloc(void *null)
+{
+	return (malloc(sizeof(struct priq), M_DEVBUF, M_WAITOK | M_ZERO));
+}
+
+void
+priq_free(void *pq)
+{
+	free(pq, M_DEVBUF, sizeof(struct priq));
+}
+
+int
+priq_enq(struct ifqueue *ifq, struct mbuf *m)
+{
+	struct priq *pq;
+	struct priq_list *pl;
+
+	if (ifq_len(ifq) >= ifq->ifq_maxlen)
+		return (ENOBUFS);
+
+	pq = ifq->ifq_q;
+	KASSERT(m->m_pkthdr.pf.prio <= IFQ_MAXPRIO);
+	pl = &pq->pq_lists[m->m_pkthdr.pf.prio];
+
+	m->m_nextpkt = NULL;
+	if (pl->tail == NULL)
+		pl->head = m;
+	else
+		pl->tail->m_nextpkt = m;
+	pl->tail = m;
+
+	return (0);
+}
+
+struct mbuf *
+priq_deq_begin(struct ifqueue *ifq, void **cookiep)
+{
+	struct priq *pq = ifq->ifq_q;
+	struct priq_list *pl;
+	unsigned int prio = nitems(pq->pq_lists);
+	struct mbuf *m;
+
+	do {
+		pl = &pq->pq_lists[--prio];
+		m = pl->head;
+		if (m != NULL) {
+			*cookiep = pl;
+			return (m);
+		}
+	} while (prio > 0);
+
+	return (NULL);
+}
+
+void
+priq_deq_commit(struct ifqueue *ifq, struct mbuf *m, void *cookie)
+{
+	struct priq_list *pl = cookie;
+
+	KASSERT(pl->head == m);
+
+	pl->head = m->m_nextpkt;
+	m->m_nextpkt = NULL;
+
+	if (pl->head == NULL)
+		pl->tail = NULL;
+}
+
+void
+priq_purge(struct ifqueue *ifq, struct mbuf_list *ml)
+{
+	struct priq *pq = ifq->ifq_q;
+	struct priq_list *pl;
+	unsigned int prio = nitems(pq->pq_lists);
+	struct mbuf *m, *n;
+
+	do {
+		pl = &pq->pq_lists[--prio];
+
+		for (m = pl->head; m != NULL; m = n) {
+			n = m->m_nextpkt;
+			ml_enqueue(ml, m);
+		}
+
+		pl->head = pl->tail = NULL;
+	} while (prio > 0);
+}
diff --git a/sys/net/ifq.h b/sys/net/ifq.h
new file mode 100644
index 00000000000..e16dbf4f399
--- /dev/null
+++ b/sys/net/ifq.h
@@ -0,0 +1,96 @@
+/*	$OpenBSD: ifq.h,v 1.1 2015/12/08 10:06:12 dlg Exp $ */
+
+/*
+ * Copyright (c) 2015 David Gwynne <dlg@openbsd.org>
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#ifndef _NET_IFQ_H_
+#define _NET_IFQ_H_
+
+struct ifnet;
+
+struct ifq_ops;
+
+struct ifqueue {
+	struct mutex		 ifq_mtx;
+	uint64_t		 ifq_drops;
+	const struct ifq_ops	*ifq_ops;
+	void			*ifq_q;
+	unsigned int		 ifq_len;
+	unsigned int		 ifq_serializer;
+	unsigned int		 ifq_oactive;
+
+	unsigned int		 ifq_maxlen;
+};
+
+#ifdef _KERNEL
+
+#define IFQ_MAXLEN		256
+
+struct ifq_ops {
+	void			*(*ifqop_alloc)(void *);
+	void			 (*ifqop_free)(void *);
+	int			 (*ifqop_enq)(struct ifqueue *, struct mbuf *);
+	struct mbuf 		*(*ifqop_deq_begin)(struct ifqueue *, void **);
+	void			 (*ifqop_deq_commit)(struct ifqueue *,
+				    struct mbuf *, void *);
+	void	 		 (*ifqop_purge)(struct ifqueue *,
+				    struct mbuf_list *);
+};
+
+/*
+ * Interface send queues.
+ */
+
+void		 ifq_init(struct ifqueue *);
+void		 ifq_attach(struct ifqueue *, const struct ifq_ops *, void *);
+void		 ifq_destroy(struct ifqueue *);
+int		 ifq_enqueue_try(struct ifqueue *, struct mbuf *);
+int		 ifq_enqueue(struct ifqueue *, struct mbuf *);
+struct mbuf	*ifq_deq_begin(struct ifqueue *);
+void		 ifq_deq_commit(struct ifqueue *, struct mbuf *);
+void		 ifq_deq_rollback(struct ifqueue *, struct mbuf *);
+struct mbuf	*ifq_dequeue(struct ifqueue *);
+unsigned int	 ifq_purge(struct ifqueue *);
+void		*ifq_q_enter(struct ifqueue *, const struct ifq_ops *);
+void		 ifq_q_leave(struct ifqueue *, void *);
+
+#define	ifq_len(_ifq)			((_ifq)->ifq_len)
+#define	ifq_empty(_ifq)			(ifq_len(_ifq) == 0)
+#define	ifq_set_maxlen(_ifq, _l)	((_ifq)->ifq_maxlen = (_l))
+
+static inline void
+ifq_set_oactive(struct ifqueue *ifq)
+{
+	ifq->ifq_oactive = 1;
+}
+
+static inline void
+ifq_clr_oactive(struct ifqueue *ifq)
+{
+	ifq->ifq_oactive = 0;
+}
+
+static inline unsigned int
+ifq_is_oactive(struct ifqueue *ifq)
+{
+	return (ifq->ifq_oactive);
+}
+
+extern const struct ifq_ops * const ifq_priq_ops;
+
+#endif /* _KERNEL */
+
+#endif /* _NET_IFQ_H_ */