2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2008 The FreeBSD Foundation
5 * Copyright (c) 2009-2010 Bjoern A. Zeeb <bz@FreeBSD.org>
8 * This software was developed by CK Software GmbH under sponsorship
9 * from the FreeBSD Foundation.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * A pair of virtual back-to-back connected ethernet like interfaces
35 * (``two interfaces with a virtual cross-over cable'').
37 * This is mostly intended to be used to provide connectivity between
38 * different virtual network stack instances.
41 * Things to re-think once we have more experience:
42 * - ifp->if_reassign function once we can test with vimage. Depending on
43 * how if_vmove() is going to be improved.
44 * - Real random etheraddrs that are checked to be uniquish; we would need
45 * to re-do them in case we move the interface between network stacks
46 * in a private if_reassign function.
47 * In case we bridge to a real interface/network or between indepedent
48 * epairs on multiple stacks/machines, we may need this.
49 * For now let the user handle that case.
52 #include <sys/cdefs.h>
53 __FBSDID("$FreeBSD$");
55 #include <sys/param.h>
58 #include <sys/kernel.h>
59 #include <sys/libkern.h>
60 #include <sys/malloc.h>
62 #include <sys/module.h>
64 #include <sys/refcount.h>
65 #include <sys/queue.h>
67 #include <sys/socket.h>
68 #include <sys/sockio.h>
69 #include <sys/sysctl.h>
70 #include <sys/types.h>
73 #include <net/ethernet.h>
75 #include <net/if_var.h>
76 #include <net/if_clone.h>
77 #include <net/if_media.h>
78 #include <net/if_var.h>
79 #include <net/if_types.h>
80 #include <net/netisr.h>
83 SYSCTL_DECL(_net_link);
84 static SYSCTL_NODE(_net_link, OID_AUTO, epair, CTLFLAG_RW, 0, "epair sysctl");
87 static int epair_debug = 0;
88 SYSCTL_INT(_net_link_epair, OID_AUTO, epair_debug, CTLFLAG_RW,
89 &epair_debug, 0, "if_epair(4) debugging.");
90 #define DPRINTF(fmt, arg...) \
92 printf("[%s:%d] " fmt, __func__, __LINE__, ##arg)
94 #define DPRINTF(fmt, arg...)
97 static void epair_nh_sintr(struct mbuf *);
98 static struct mbuf *epair_nh_m2cpuid(struct mbuf *, uintptr_t, u_int *);
99 static void epair_nh_drainedcpu(u_int);
101 static void epair_start_locked(struct ifnet *);
102 static int epair_media_change(struct ifnet *);
103 static void epair_media_status(struct ifnet *, struct ifmediareq *);
105 static int epair_clone_match(struct if_clone *, const char *);
106 static int epair_clone_create(struct if_clone *, char *, size_t, caddr_t);
107 static int epair_clone_destroy(struct if_clone *, struct ifnet *);
109 static const char epairname[] = "epair";
111 /* Netisr related definitions and sysctl. */
112 static struct netisr_handler epair_nh = {
113 .nh_name = epairname,
114 .nh_proto = NETISR_EPAIR,
115 .nh_policy = NETISR_POLICY_CPU,
116 .nh_handler = epair_nh_sintr,
117 .nh_m2cpuid = epair_nh_m2cpuid,
118 .nh_drainedcpu = epair_nh_drainedcpu,
122 sysctl_epair_netisr_maxqlen(SYSCTL_HANDLER_ARGS)
126 netisr_getqlimit(&epair_nh, &qlimit);
127 error = sysctl_handle_int(oidp, &qlimit, 0, req);
128 if (error || !req->newptr)
132 return (netisr_setqlimit(&epair_nh, qlimit));
134 SYSCTL_PROC(_net_link_epair, OID_AUTO, netisr_maxqlen, CTLTYPE_INT|CTLFLAG_RW,
135 0, 0, sysctl_epair_netisr_maxqlen, "I",
136 "Maximum if_epair(4) netisr \"hw\" queue length");
139 struct ifnet *ifp; /* This ifp. */
140 struct ifnet *oifp; /* other ifp of pair. */
141 struct ifmedia media; /* Media config (fake). */
142 u_int refcount; /* # of mbufs in flight. */
143 u_int cpuid; /* CPU ID assigned upon creation. */
144 void (*if_qflush)(struct ifnet *);
145 /* Original if_qflush routine. */
149 * Per-CPU list of ifps with data in the ifq that needs to be flushed
150 * to the netisr ``hw'' queue before we allow any further direct queuing
151 * to the ``hw'' queue.
153 struct epair_ifp_drain {
154 STAILQ_ENTRY(epair_ifp_drain) ifp_next;
157 STAILQ_HEAD(eid_list, epair_ifp_drain);
159 #define EPAIR_LOCK_INIT(dpcpu) mtx_init(&(dpcpu)->if_epair_mtx, \
160 "if_epair", NULL, MTX_DEF)
161 #define EPAIR_LOCK_DESTROY(dpcpu) mtx_destroy(&(dpcpu)->if_epair_mtx)
162 #define EPAIR_LOCK_ASSERT(dpcpu) mtx_assert(&(dpcpu)->if_epair_mtx, \
164 #define EPAIR_LOCK(dpcpu) mtx_lock(&(dpcpu)->if_epair_mtx)
165 #define EPAIR_UNLOCK(dpcpu) mtx_unlock(&(dpcpu)->if_epair_mtx)
168 #define EPAIR_REFCOUNT_INIT(r, v) refcount_init((r), (v))
169 #define EPAIR_REFCOUNT_AQUIRE(r) refcount_acquire((r))
170 #define EPAIR_REFCOUNT_RELEASE(r) refcount_release((r))
171 #define EPAIR_REFCOUNT_ASSERT(a, p) KASSERT(a, p)
173 #define EPAIR_REFCOUNT_INIT(r, v)
174 #define EPAIR_REFCOUNT_AQUIRE(r)
175 #define EPAIR_REFCOUNT_RELEASE(r)
176 #define EPAIR_REFCOUNT_ASSERT(a, p)
179 static MALLOC_DEFINE(M_EPAIR, epairname,
180 "Pair of virtual cross-over connected Ethernet-like interfaces");
182 static VNET_DEFINE(struct if_clone *, epair_cloner);
183 #define V_epair_cloner VNET(epair_cloner)
186 * DPCPU area and functions.
189 struct mtx if_epair_mtx; /* Per-CPU locking. */
190 int epair_drv_flags; /* Per-CPU ``hw'' drv flags. */
191 struct eid_list epair_ifp_drain_list; /* Per-CPU list of ifps with
192 * data in the ifq. */
194 DPCPU_DEFINE(struct epair_dpcpu, epair_dpcpu);
197 epair_dpcpu_init(void)
199 struct epair_dpcpu *epair_dpcpu;
204 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu);
206 /* Initialize per-cpu lock. */
207 EPAIR_LOCK_INIT(epair_dpcpu);
209 /* Driver flags are per-cpu as are our netisr "hw" queues. */
210 epair_dpcpu->epair_drv_flags = 0;
213 * Initialize per-cpu drain list.
214 * Manually do what STAILQ_HEAD_INITIALIZER would do.
216 s = &epair_dpcpu->epair_ifp_drain_list;
217 s->stqh_first = NULL;
218 s->stqh_last = &s->stqh_first;
223 epair_dpcpu_detach(void)
225 struct epair_dpcpu *epair_dpcpu;
229 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu);
231 /* Destroy per-cpu lock. */
232 EPAIR_LOCK_DESTROY(epair_dpcpu);
240 cpuid_from_ifp(struct ifnet *ifp)
242 struct epair_softc *sc;
252 * Netisr handler functions.
255 epair_nh_sintr(struct mbuf *m)
258 struct epair_softc *sc __unused;
260 ifp = m->m_pkthdr.rcvif;
261 (*ifp->if_input)(ifp, m);
263 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
264 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1,
265 ("%s: ifp=%p sc->refcount not >= 1: %d",
266 __func__, ifp, sc->refcount));
267 DPRINTF("ifp=%p refcount=%u\n", ifp, sc->refcount);
271 epair_nh_m2cpuid(struct mbuf *m, uintptr_t source, u_int *cpuid)
274 *cpuid = cpuid_from_ifp(m->m_pkthdr.rcvif);
280 epair_nh_drainedcpu(u_int cpuid)
282 struct epair_dpcpu *epair_dpcpu;
283 struct epair_ifp_drain *elm, *tvar;
286 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu);
287 EPAIR_LOCK(epair_dpcpu);
289 * Assume our "hw" queue and possibly ifq will be emptied
290 * again. In case we will overflow the "hw" queue while
291 * draining, epair_start_locked will set IFF_DRV_OACTIVE
292 * again and we will stop and return.
294 STAILQ_FOREACH_SAFE(elm, &epair_dpcpu->epair_ifp_drain_list,
297 epair_dpcpu->epair_drv_flags &= ~IFF_DRV_OACTIVE;
298 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
299 epair_start_locked(ifp);
301 IFQ_LOCK(&ifp->if_snd);
302 if (IFQ_IS_EMPTY(&ifp->if_snd)) {
303 struct epair_softc *sc __unused;
305 STAILQ_REMOVE(&epair_dpcpu->epair_ifp_drain_list,
306 elm, epair_ifp_drain, ifp_next);
307 /* The cached ifp goes off the list. */
309 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
310 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1,
311 ("%s: ifp=%p sc->refcount not >= 1: %d",
312 __func__, ifp, sc->refcount));
315 IFQ_UNLOCK(&ifp->if_snd);
317 if ((ifp->if_drv_flags & IFF_DRV_OACTIVE) != 0) {
318 /* Our "hw"q overflew again. */
319 epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE;
320 DPRINTF("hw queue length overflow at %u\n",
325 EPAIR_UNLOCK(epair_dpcpu);
329 * Network interface (`if') related functions.
332 epair_remove_ifp_from_draining(struct ifnet *ifp)
334 struct epair_dpcpu *epair_dpcpu;
335 struct epair_ifp_drain *elm, *tvar;
339 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu);
340 EPAIR_LOCK(epair_dpcpu);
341 STAILQ_FOREACH_SAFE(elm, &epair_dpcpu->epair_ifp_drain_list,
343 if (ifp == elm->ifp) {
344 struct epair_softc *sc __unused;
347 &epair_dpcpu->epair_ifp_drain_list, elm,
348 epair_ifp_drain, ifp_next);
349 /* The cached ifp goes off the list. */
351 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
352 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1,
353 ("%s: ifp=%p sc->refcount not >= 1: %d",
354 __func__, ifp, sc->refcount));
358 EPAIR_UNLOCK(epair_dpcpu);
363 epair_add_ifp_for_draining(struct ifnet *ifp)
365 struct epair_dpcpu *epair_dpcpu;
366 struct epair_softc *sc;
367 struct epair_ifp_drain *elm = NULL;
370 epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu);
371 EPAIR_LOCK_ASSERT(epair_dpcpu);
372 STAILQ_FOREACH(elm, &epair_dpcpu->epair_ifp_drain_list, ifp_next)
375 /* If the ifp is there already, return success. */
379 elm = malloc(sizeof(struct epair_ifp_drain), M_EPAIR, M_NOWAIT|M_ZERO);
384 /* Add a reference for the ifp pointer on the list. */
385 EPAIR_REFCOUNT_AQUIRE(&sc->refcount);
386 STAILQ_INSERT_TAIL(&epair_dpcpu->epair_ifp_drain_list, elm, ifp_next);
392 epair_start_locked(struct ifnet *ifp)
394 struct epair_dpcpu *epair_dpcpu;
396 struct epair_softc *sc;
400 DPRINTF("ifp=%p\n", ifp);
402 epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu);
403 EPAIR_LOCK_ASSERT(epair_dpcpu);
405 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
407 if ((ifp->if_flags & IFF_UP) == 0)
411 * We get packets here from ether_output via if_handoff()
412 * and need to put them into the input queue of the oifp
413 * and call oifp->if_input() via netisr/epair_sintr().
418 IFQ_DEQUEUE(&ifp->if_snd, m);
424 * In case the outgoing interface is not usable,
427 if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
428 (oifp->if_flags & IFF_UP) ==0) {
429 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
433 DPRINTF("packet %s -> %s\n", ifp->if_xname, oifp->if_xname);
436 * Add a reference so the interface cannot go while the
437 * packet is in transit as we rely on rcvif to stay valid.
439 EPAIR_REFCOUNT_AQUIRE(&sc->refcount);
440 m->m_pkthdr.rcvif = oifp;
441 CURVNET_SET_QUIET(oifp->if_vnet);
442 error = netisr_queue(NETISR_EPAIR, m);
445 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
446 /* Someone else received the packet. */
447 if_inc_counter(oifp, IFCOUNTER_IPACKETS, 1);
449 /* The packet was freed already. */
450 epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE;
451 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
452 (void) epair_add_ifp_for_draining(ifp);
453 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
454 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
455 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1,
456 ("%s: ifp=%p sc->refcount not >= 1: %d",
457 __func__, oifp, sc->refcount));
463 epair_start(struct ifnet *ifp)
465 struct epair_dpcpu *epair_dpcpu;
467 epair_dpcpu = DPCPU_ID_PTR(cpuid_from_ifp(ifp), epair_dpcpu);
468 EPAIR_LOCK(epair_dpcpu);
469 epair_start_locked(ifp);
470 EPAIR_UNLOCK(epair_dpcpu);
474 epair_transmit_locked(struct ifnet *ifp, struct mbuf *m)
476 struct epair_dpcpu *epair_dpcpu;
477 struct epair_softc *sc;
482 DPRINTF("ifp=%p m=%p\n", ifp, m);
484 epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu);
485 EPAIR_LOCK_ASSERT(epair_dpcpu);
491 * We are not going to use the interface en/dequeue mechanism
492 * on the TX side. We are called from ether_output_frame()
493 * and will put the packet into the incoming queue of the
494 * other interface of our pair via the netsir.
496 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
500 if ((ifp->if_flags & IFF_UP) == 0) {
508 * In case the outgoing interface is not usable,
512 if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
513 (oifp->if_flags & IFF_UP) ==0) {
514 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
518 len = m->m_pkthdr.len;
520 DPRINTF("packet %s -> %s\n", ifp->if_xname, oifp->if_xname);
523 /* Support ALTQ via the classic if_start() path. */
524 IF_LOCK(&ifp->if_snd);
525 if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
526 ALTQ_ENQUEUE(&ifp->if_snd, m, NULL, error);
528 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
529 IF_UNLOCK(&ifp->if_snd);
531 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
532 if (mflags & (M_BCAST|M_MCAST))
533 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
535 if ((ifp->if_drv_flags & IFF_DRV_OACTIVE) == 0)
536 epair_start_locked(ifp);
538 (void)epair_add_ifp_for_draining(ifp);
542 IF_UNLOCK(&ifp->if_snd);
545 if ((epair_dpcpu->epair_drv_flags & IFF_DRV_OACTIVE) != 0) {
547 * Our hardware queue is full, try to fall back
548 * queuing to the ifq but do not call ifp->if_start.
549 * Either we are lucky or the packet is gone.
551 IFQ_ENQUEUE(&ifp->if_snd, m, error);
553 (void)epair_add_ifp_for_draining(ifp);
558 * Add a reference so the interface cannot go while the
559 * packet is in transit as we rely on rcvif to stay valid.
561 EPAIR_REFCOUNT_AQUIRE(&sc->refcount);
562 m->m_pkthdr.rcvif = oifp;
563 CURVNET_SET_QUIET(oifp->if_vnet);
564 error = netisr_queue(NETISR_EPAIR, m);
567 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
569 * IFQ_HANDOFF_ADJ/ip_handoff() update statistics,
570 * but as we bypass all this we have to duplicate
571 * the logic another time.
573 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
574 if (mflags & (M_BCAST|M_MCAST))
575 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
576 /* Someone else received the packet. */
577 if_inc_counter(oifp, IFCOUNTER_IPACKETS, 1);
579 /* The packet was freed already. */
580 epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE;
581 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
582 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
583 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
584 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1,
585 ("%s: ifp=%p sc->refcount not >= 1: %d",
586 __func__, oifp, sc->refcount));
593 epair_transmit(struct ifnet *ifp, struct mbuf *m)
595 struct epair_dpcpu *epair_dpcpu;
598 epair_dpcpu = DPCPU_ID_PTR(cpuid_from_ifp(ifp), epair_dpcpu);
599 EPAIR_LOCK(epair_dpcpu);
600 error = epair_transmit_locked(ifp, m);
601 EPAIR_UNLOCK(epair_dpcpu);
606 epair_qflush(struct ifnet *ifp)
608 struct epair_softc *sc;
611 KASSERT(sc != NULL, ("%s: ifp=%p, epair_softc gone? sc=%p\n",
614 * Remove this ifp from all backpointer lists. The interface will not
615 * usable for flushing anyway nor should it have anything to flush
618 epair_remove_ifp_from_draining(ifp);
625 epair_media_change(struct ifnet *ifp __unused)
633 epair_media_status(struct ifnet *ifp __unused, struct ifmediareq *imr)
636 imr->ifm_status = IFM_AVALID | IFM_ACTIVE;
637 imr->ifm_active = IFM_ETHER | IFM_10G_T | IFM_FDX;
641 epair_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
643 struct epair_softc *sc;
647 ifr = (struct ifreq *)data;
658 error = ifmedia_ioctl(ifp, ifr, &sc->media, cmd);
662 /* We basically allow all kinds of MTUs. */
663 ifp->if_mtu = ifr->ifr_mtu;
668 /* Let the common ethernet handler process this. */
669 error = ether_ioctl(ifp, cmd, data);
677 epair_init(void *dummy __unused)
683 * Interface cloning functions.
684 * We use our private ones so that we can create/destroy our secondary
685 * device along with the primary one.
688 epair_clone_match(struct if_clone *ifc, const char *name)
692 DPRINTF("name='%s'\n", name);
695 * Our base name is epair.
696 * Our interfaces will be named epair<n>[ab].
697 * So accept anything of the following list:
700 * but not the epair<n>[ab] versions.
702 if (strncmp(epairname, name, sizeof(epairname)-1) != 0)
705 for (cp = name + sizeof(epairname) - 1; *cp != '\0'; cp++) {
706 if (*cp < '0' || *cp > '9')
714 epair_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
716 struct epair_softc *sca, *scb;
719 int error, unit, wildcard;
723 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
726 * We are abusing params to create our second interface.
727 * Actually we already created it and called if_clone_create()
728 * for it to do the official insertion procedure the moment we knew
729 * it cannot fail anymore. So just do attach it here.
732 scb = (struct epair_softc *)params;
734 /* Copy epairNa etheraddr and change the last byte. */
735 memcpy(eaddr, scb->oifp->if_hw_addr, ETHER_ADDR_LEN);
737 ether_ifattach(ifp, eaddr);
738 /* Correctly set the name for the cloner list. */
739 strlcpy(name, ifp->if_xname, len);
743 /* Try to see if a special unit was requested. */
744 error = ifc_name2unit(name, &unit);
747 wildcard = (unit < 0);
749 error = ifc_alloc_unit(ifc, &unit);
754 * If no unit had been given, we need to adjust the ifName.
755 * Also make sure there is space for our extra [ab] suffix.
757 for (dp = name; *dp != '\0'; dp++);
759 error = snprintf(dp, len - (dp - name), "%d", unit);
760 if (error > len - (dp - name) - 1) {
761 /* ifName too long. */
762 ifc_free_unit(ifc, unit);
767 if (len - (dp - name) - 1 < 1) {
768 /* No space left for our [ab] suffix. */
769 ifc_free_unit(ifc, unit);
773 /* Must not change dp so we can replace 'a' by 'b' later. */
776 /* Check if 'a' and 'b' interfaces already exist. */
777 if (ifunit(name) != NULL)
780 if (ifunit(name) != NULL)
783 /* Allocate memory for both [ab] interfaces */
784 sca = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
785 EPAIR_REFCOUNT_INIT(&sca->refcount, 1);
786 sca->ifp = if_alloc(IFT_ETHER);
787 if (sca->ifp == NULL) {
789 ifc_free_unit(ifc, unit);
793 scb = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
794 EPAIR_REFCOUNT_INIT(&scb->refcount, 1);
795 scb->ifp = if_alloc(IFT_ETHER);
796 if (scb->ifp == NULL) {
800 ifc_free_unit(ifc, unit);
805 * Cross-reference the interfaces so we will be able to free both.
807 sca->oifp = scb->ifp;
808 scb->oifp = sca->ifp;
811 * Calculate the cpuid for netisr queueing based on the
812 * ifIndex of the interfaces. As long as we cannot configure
813 * this or use cpuset information easily we cannot guarantee
814 * cache locality but we can at least allow parallelism.
817 netisr_get_cpuid(sca->ifp->if_index);
819 netisr_get_cpuid(scb->ifp->if_index);
821 /* Initialise pseudo media types. */
822 ifmedia_init(&sca->media, 0, epair_media_change, epair_media_status);
823 ifmedia_add(&sca->media, IFM_ETHER | IFM_10G_T, 0, NULL);
824 ifmedia_set(&sca->media, IFM_ETHER | IFM_10G_T);
825 ifmedia_init(&scb->media, 0, epair_media_change, epair_media_status);
826 ifmedia_add(&scb->media, IFM_ETHER | IFM_10G_T, 0, NULL);
827 ifmedia_set(&scb->media, IFM_ETHER | IFM_10G_T);
829 /* Finish initialization of interface <n>a. */
832 strlcpy(ifp->if_xname, name, IFNAMSIZ);
833 ifp->if_dname = epairname;
834 ifp->if_dunit = unit;
835 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
836 ifp->if_capabilities = IFCAP_VLAN_MTU;
837 ifp->if_capenable = IFCAP_VLAN_MTU;
838 ifp->if_start = epair_start;
839 ifp->if_ioctl = epair_ioctl;
840 ifp->if_init = epair_init;
841 if_setsendqlen(ifp, ifqmaxlen);
842 if_setsendqready(ifp);
845 * Calculate the etheraddr hashing the hostid and the
846 * interface index. The result would be hopefully unique
848 getcredhostid(curthread->td_ucred, (unsigned long *)&hostid);
850 arc4rand(&hostid, sizeof(hostid), 0);
851 key[0] = (uint32_t)ifp->if_index;
852 key[1] = (uint32_t)(hostid & 0xffffffff);
853 key[2] = (uint32_t)((hostid >> 32) & 0xfffffffff);
854 hash = jenkins_hash32(key, 3, 0);
857 memcpy(&eaddr[1], &hash, 4);
859 ether_ifattach(ifp, eaddr);
860 sca->if_qflush = ifp->if_qflush;
861 ifp->if_qflush = epair_qflush;
862 ifp->if_transmit = epair_transmit;
863 ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */
865 /* Swap the name and finish initialization of interface <n>b. */
870 strlcpy(ifp->if_xname, name, IFNAMSIZ);
871 ifp->if_dname = epairname;
872 ifp->if_dunit = unit;
873 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
874 ifp->if_capabilities = IFCAP_VLAN_MTU;
875 ifp->if_capenable = IFCAP_VLAN_MTU;
876 ifp->if_start = epair_start;
877 ifp->if_ioctl = epair_ioctl;
878 ifp->if_init = epair_init;
879 if_setsendqlen(ifp, ifqmaxlen);
880 if_setsendqready(ifp);
881 /* We need to play some tricks here for the second interface. */
882 strlcpy(name, epairname, len);
883 error = if_clone_create(name, len, (caddr_t)scb);
885 panic("%s: if_clone_create() for our 2nd iface failed: %d",
887 scb->if_qflush = ifp->if_qflush;
888 ifp->if_qflush = epair_qflush;
889 ifp->if_transmit = epair_transmit;
890 ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */
893 * Restore name to <n>a as the ifp for this will go into the
894 * cloner list for the initial call.
896 strlcpy(name, sca->ifp->if_xname, len);
897 DPRINTF("name='%s/%db' created sca=%p scb=%p\n", name, unit, sca, scb);
899 /* Tell the world, that we are ready to rock. */
900 sca->ifp->if_drv_flags |= IFF_DRV_RUNNING;
901 scb->ifp->if_drv_flags |= IFF_DRV_RUNNING;
902 if_link_state_change(sca->ifp, LINK_STATE_UP);
903 if_link_state_change(scb->ifp, LINK_STATE_UP);
909 epair_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
912 struct epair_softc *sca, *scb;
915 DPRINTF("ifp=%p\n", ifp);
918 * In case we called into if_clone_destroyif() ourselves
919 * again to remove the second interface, the softc will be
920 * NULL. In that case so not do anything but return success.
922 if (ifp->if_softc == NULL)
925 unit = ifp->if_dunit;
928 scb = oifp->if_softc;
930 DPRINTF("ifp=%p oifp=%p\n", ifp, oifp);
931 if_link_state_change(ifp, LINK_STATE_DOWN);
932 if_link_state_change(oifp, LINK_STATE_DOWN);
933 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
934 oifp->if_drv_flags &= ~IFF_DRV_RUNNING;
937 * Get rid of our second half. As the other of the two
938 * interfaces may reside in a different vnet, we need to
939 * switch before freeing them.
941 CURVNET_SET_QUIET(oifp->if_vnet);
942 ether_ifdetach(oifp);
944 * Wait for all packets to be dispatched to if_input.
945 * The numbers can only go down as the interface is
946 * detached so there is no need to use atomics.
948 DPRINTF("scb refcnt=%u\n", scb->refcount);
949 EPAIR_REFCOUNT_ASSERT(scb->refcount == 1,
950 ("%s: ifp=%p scb->refcount!=1: %d", __func__, oifp, scb->refcount));
951 oifp->if_softc = NULL;
952 error = if_clone_destroyif(ifc, oifp);
954 panic("%s: if_clone_destroyif() for our 2nd iface failed: %d",
957 ifmedia_removeall(&scb->media);
963 * Wait for all packets to be dispatched to if_input.
965 DPRINTF("sca refcnt=%u\n", sca->refcount);
966 EPAIR_REFCOUNT_ASSERT(sca->refcount == 1,
967 ("%s: ifp=%p sca->refcount!=1: %d", __func__, ifp, sca->refcount));
969 ifmedia_removeall(&sca->media);
971 ifc_free_unit(ifc, unit);
977 vnet_epair_init(const void *unused __unused)
980 V_epair_cloner = if_clone_advanced(epairname, 0,
981 epair_clone_match, epair_clone_create, epair_clone_destroy);
983 netisr_register_vnet(&epair_nh);
986 VNET_SYSINIT(vnet_epair_init, SI_SUB_PSEUDO, SI_ORDER_ANY,
987 vnet_epair_init, NULL);
990 vnet_epair_uninit(const void *unused __unused)
994 netisr_unregister_vnet(&epair_nh);
996 if_clone_detach(V_epair_cloner);
998 VNET_SYSUNINIT(vnet_epair_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
999 vnet_epair_uninit, NULL);
1002 epair_uninit(const void *unused __unused)
1004 netisr_unregister(&epair_nh);
1005 epair_dpcpu_detach();
1007 printf("%s unloaded.\n", epairname);
1009 SYSUNINIT(epair_uninit, SI_SUB_INIT_IF, SI_ORDER_MIDDLE,
1010 epair_uninit, NULL);
1013 epair_modevent(module_t mod, int type, void *data)
1019 /* For now limit us to one global mutex and one inq. */
1021 epair_nh.nh_qlimit = 42 * ifqmaxlen; /* 42 shall be the number. */
1022 if (TUNABLE_INT_FETCH("net.link.epair.netisr_maxqlen", &qlimit))
1023 epair_nh.nh_qlimit = qlimit;
1024 netisr_register(&epair_nh);
1026 printf("%s initialized.\n", epairname);
1029 /* Handled in epair_uninit() */
1032 return (EOPNOTSUPP);
1037 static moduledata_t epair_mod = {
1043 DECLARE_MODULE(if_epair, epair_mod, SI_SUB_PSEUDO, SI_ORDER_MIDDLE);
1044 MODULE_VERSION(if_epair, 1);