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 | CTLFLAG_MPSAFE, 0,
88 static int epair_debug = 0;
89 SYSCTL_INT(_net_link_epair, OID_AUTO, epair_debug, CTLFLAG_RW,
90 &epair_debug, 0, "if_epair(4) debugging.");
91 #define DPRINTF(fmt, arg...) \
93 printf("[%s:%d] " fmt, __func__, __LINE__, ##arg)
95 #define DPRINTF(fmt, arg...)
98 static void epair_nh_sintr(struct mbuf *);
99 static struct mbuf *epair_nh_m2cpuid(struct mbuf *, uintptr_t, u_int *);
100 static void epair_nh_drainedcpu(u_int);
102 static void epair_start_locked(struct ifnet *);
103 static int epair_media_change(struct ifnet *);
104 static void epair_media_status(struct ifnet *, struct ifmediareq *);
106 static int epair_clone_match(struct if_clone *, const char *);
107 static int epair_clone_create(struct if_clone *, char *, size_t, caddr_t);
108 static int epair_clone_destroy(struct if_clone *, struct ifnet *);
110 static const char epairname[] = "epair";
111 static unsigned int next_index = 0;
113 /* Netisr related definitions and sysctl. */
114 static struct netisr_handler epair_nh = {
115 .nh_name = epairname,
116 .nh_proto = NETISR_EPAIR,
117 .nh_policy = NETISR_POLICY_CPU,
118 .nh_handler = epair_nh_sintr,
119 .nh_m2cpuid = epair_nh_m2cpuid,
120 .nh_drainedcpu = epair_nh_drainedcpu,
124 sysctl_epair_netisr_maxqlen(SYSCTL_HANDLER_ARGS)
128 netisr_getqlimit(&epair_nh, &qlimit);
129 error = sysctl_handle_int(oidp, &qlimit, 0, req);
130 if (error || !req->newptr)
134 return (netisr_setqlimit(&epair_nh, qlimit));
136 SYSCTL_PROC(_net_link_epair, OID_AUTO, netisr_maxqlen,
137 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0,
138 sysctl_epair_netisr_maxqlen, "I",
139 "Maximum if_epair(4) netisr \"hw\" queue length");
142 struct ifnet *ifp; /* This ifp. */
143 struct ifnet *oifp; /* other ifp of pair. */
144 struct ifmedia media; /* Media config (fake). */
145 u_int refcount; /* # of mbufs in flight. */
146 u_int cpuid; /* CPU ID assigned upon creation. */
147 void (*if_qflush)(struct ifnet *);
148 /* Original if_qflush routine. */
152 * Per-CPU list of ifps with data in the ifq that needs to be flushed
153 * to the netisr ``hw'' queue before we allow any further direct queuing
154 * to the ``hw'' queue.
156 struct epair_ifp_drain {
157 STAILQ_ENTRY(epair_ifp_drain) ifp_next;
160 STAILQ_HEAD(eid_list, epair_ifp_drain);
162 #define EPAIR_LOCK_INIT(dpcpu) mtx_init(&(dpcpu)->if_epair_mtx, \
163 "if_epair", NULL, MTX_DEF)
164 #define EPAIR_LOCK_DESTROY(dpcpu) mtx_destroy(&(dpcpu)->if_epair_mtx)
165 #define EPAIR_LOCK_ASSERT(dpcpu) mtx_assert(&(dpcpu)->if_epair_mtx, \
167 #define EPAIR_LOCK(dpcpu) mtx_lock(&(dpcpu)->if_epair_mtx)
168 #define EPAIR_UNLOCK(dpcpu) mtx_unlock(&(dpcpu)->if_epair_mtx)
171 #define EPAIR_REFCOUNT_INIT(r, v) refcount_init((r), (v))
172 #define EPAIR_REFCOUNT_AQUIRE(r) refcount_acquire((r))
173 #define EPAIR_REFCOUNT_RELEASE(r) refcount_release((r))
174 #define EPAIR_REFCOUNT_ASSERT(a, p) KASSERT(a, p)
176 #define EPAIR_REFCOUNT_INIT(r, v)
177 #define EPAIR_REFCOUNT_AQUIRE(r)
178 #define EPAIR_REFCOUNT_RELEASE(r)
179 #define EPAIR_REFCOUNT_ASSERT(a, p)
182 static MALLOC_DEFINE(M_EPAIR, epairname,
183 "Pair of virtual cross-over connected Ethernet-like interfaces");
185 VNET_DEFINE_STATIC(struct if_clone *, epair_cloner);
186 #define V_epair_cloner VNET(epair_cloner)
189 * DPCPU area and functions.
192 struct mtx if_epair_mtx; /* Per-CPU locking. */
193 int epair_drv_flags; /* Per-CPU ``hw'' drv flags. */
194 struct eid_list epair_ifp_drain_list; /* Per-CPU list of ifps with
195 * data in the ifq. */
197 DPCPU_DEFINE(struct epair_dpcpu, epair_dpcpu);
200 epair_dpcpu_init(void)
202 struct epair_dpcpu *epair_dpcpu;
207 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu);
209 /* Initialize per-cpu lock. */
210 EPAIR_LOCK_INIT(epair_dpcpu);
212 /* Driver flags are per-cpu as are our netisr "hw" queues. */
213 epair_dpcpu->epair_drv_flags = 0;
216 * Initialize per-cpu drain list.
217 * Manually do what STAILQ_HEAD_INITIALIZER would do.
219 s = &epair_dpcpu->epair_ifp_drain_list;
220 s->stqh_first = NULL;
221 s->stqh_last = &s->stqh_first;
226 epair_dpcpu_detach(void)
228 struct epair_dpcpu *epair_dpcpu;
232 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu);
234 /* Destroy per-cpu lock. */
235 EPAIR_LOCK_DESTROY(epair_dpcpu);
243 cpuid_from_ifp(struct ifnet *ifp)
245 struct epair_softc *sc;
255 * Netisr handler functions.
258 epair_nh_sintr(struct mbuf *m)
261 struct epair_softc *sc __unused;
263 ifp = m->m_pkthdr.rcvif;
264 (*ifp->if_input)(ifp, m);
266 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
267 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1,
268 ("%s: ifp=%p sc->refcount not >= 1: %d",
269 __func__, ifp, sc->refcount));
270 DPRINTF("ifp=%p refcount=%u\n", ifp, sc->refcount);
274 epair_nh_m2cpuid(struct mbuf *m, uintptr_t source, u_int *cpuid)
277 *cpuid = cpuid_from_ifp(m->m_pkthdr.rcvif);
283 epair_nh_drainedcpu(u_int cpuid)
285 struct epair_dpcpu *epair_dpcpu;
286 struct epair_ifp_drain *elm, *tvar;
289 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu);
290 EPAIR_LOCK(epair_dpcpu);
292 * Assume our "hw" queue and possibly ifq will be emptied
293 * again. In case we will overflow the "hw" queue while
294 * draining, epair_start_locked will set IFF_DRV_OACTIVE
295 * again and we will stop and return.
297 STAILQ_FOREACH_SAFE(elm, &epair_dpcpu->epair_ifp_drain_list,
300 epair_dpcpu->epair_drv_flags &= ~IFF_DRV_OACTIVE;
301 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
302 epair_start_locked(ifp);
304 IFQ_LOCK(&ifp->if_snd);
305 if (IFQ_IS_EMPTY(&ifp->if_snd)) {
306 struct epair_softc *sc __unused;
308 STAILQ_REMOVE(&epair_dpcpu->epair_ifp_drain_list,
309 elm, epair_ifp_drain, ifp_next);
310 /* The cached ifp goes off the list. */
312 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
313 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1,
314 ("%s: ifp=%p sc->refcount not >= 1: %d",
315 __func__, ifp, sc->refcount));
318 IFQ_UNLOCK(&ifp->if_snd);
320 if ((ifp->if_drv_flags & IFF_DRV_OACTIVE) != 0) {
321 /* Our "hw"q overflew again. */
322 epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE;
323 DPRINTF("hw queue length overflow at %u\n",
328 EPAIR_UNLOCK(epair_dpcpu);
332 * Network interface (`if') related functions.
335 epair_remove_ifp_from_draining(struct ifnet *ifp)
337 struct epair_dpcpu *epair_dpcpu;
338 struct epair_ifp_drain *elm, *tvar;
342 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu);
343 EPAIR_LOCK(epair_dpcpu);
344 STAILQ_FOREACH_SAFE(elm, &epair_dpcpu->epair_ifp_drain_list,
346 if (ifp == elm->ifp) {
347 struct epair_softc *sc __unused;
350 &epair_dpcpu->epair_ifp_drain_list, elm,
351 epair_ifp_drain, ifp_next);
352 /* The cached ifp goes off the list. */
354 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
355 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1,
356 ("%s: ifp=%p sc->refcount not >= 1: %d",
357 __func__, ifp, sc->refcount));
361 EPAIR_UNLOCK(epair_dpcpu);
366 epair_add_ifp_for_draining(struct ifnet *ifp)
368 struct epair_dpcpu *epair_dpcpu;
369 struct epair_softc *sc;
370 struct epair_ifp_drain *elm = NULL;
373 epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu);
374 EPAIR_LOCK_ASSERT(epair_dpcpu);
375 STAILQ_FOREACH(elm, &epair_dpcpu->epair_ifp_drain_list, ifp_next)
378 /* If the ifp is there already, return success. */
382 elm = malloc(sizeof(struct epair_ifp_drain), M_EPAIR, M_NOWAIT|M_ZERO);
387 /* Add a reference for the ifp pointer on the list. */
388 EPAIR_REFCOUNT_AQUIRE(&sc->refcount);
389 STAILQ_INSERT_TAIL(&epair_dpcpu->epair_ifp_drain_list, elm, ifp_next);
395 epair_start_locked(struct ifnet *ifp)
397 struct epair_dpcpu *epair_dpcpu;
399 struct epair_softc *sc;
403 DPRINTF("ifp=%p\n", ifp);
405 epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu);
406 EPAIR_LOCK_ASSERT(epair_dpcpu);
408 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
410 if ((ifp->if_flags & IFF_UP) == 0)
414 * We get packets here from ether_output via if_handoff()
415 * and need to put them into the input queue of the oifp
416 * and call oifp->if_input() via netisr/epair_sintr().
421 IFQ_DEQUEUE(&ifp->if_snd, m);
427 * In case the outgoing interface is not usable,
430 if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
431 (oifp->if_flags & IFF_UP) ==0) {
432 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
436 DPRINTF("packet %s -> %s\n", ifp->if_xname, oifp->if_xname);
439 * Add a reference so the interface cannot go while the
440 * packet is in transit as we rely on rcvif to stay valid.
442 EPAIR_REFCOUNT_AQUIRE(&sc->refcount);
443 m->m_pkthdr.rcvif = oifp;
444 CURVNET_SET_QUIET(oifp->if_vnet);
445 error = netisr_queue(NETISR_EPAIR, m);
448 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
449 /* Someone else received the packet. */
450 if_inc_counter(oifp, IFCOUNTER_IPACKETS, 1);
452 /* The packet was freed already. */
453 epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE;
454 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
455 (void) epair_add_ifp_for_draining(ifp);
456 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
457 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
458 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1,
459 ("%s: ifp=%p sc->refcount not >= 1: %d",
460 __func__, oifp, sc->refcount));
466 epair_start(struct ifnet *ifp)
468 struct epair_dpcpu *epair_dpcpu;
470 epair_dpcpu = DPCPU_ID_PTR(cpuid_from_ifp(ifp), epair_dpcpu);
471 EPAIR_LOCK(epair_dpcpu);
472 epair_start_locked(ifp);
473 EPAIR_UNLOCK(epair_dpcpu);
477 epair_transmit_locked(struct ifnet *ifp, struct mbuf *m)
479 struct epair_dpcpu *epair_dpcpu;
480 struct epair_softc *sc;
485 DPRINTF("ifp=%p m=%p\n", ifp, m);
487 epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu);
488 EPAIR_LOCK_ASSERT(epair_dpcpu);
494 * We are not going to use the interface en/dequeue mechanism
495 * on the TX side. We are called from ether_output_frame()
496 * and will put the packet into the incoming queue of the
497 * other interface of our pair via the netsir.
499 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
503 if ((ifp->if_flags & IFF_UP) == 0) {
511 * In case the outgoing interface is not usable,
515 if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
516 (oifp->if_flags & IFF_UP) ==0) {
517 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
521 len = m->m_pkthdr.len;
523 DPRINTF("packet %s -> %s\n", ifp->if_xname, oifp->if_xname);
526 /* Support ALTQ via the classic if_start() path. */
527 IF_LOCK(&ifp->if_snd);
528 if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
529 ALTQ_ENQUEUE(&ifp->if_snd, m, NULL, error);
531 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
532 IF_UNLOCK(&ifp->if_snd);
534 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
535 if (mflags & (M_BCAST|M_MCAST))
536 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
538 if ((ifp->if_drv_flags & IFF_DRV_OACTIVE) == 0)
539 epair_start_locked(ifp);
541 (void)epair_add_ifp_for_draining(ifp);
545 IF_UNLOCK(&ifp->if_snd);
548 if ((epair_dpcpu->epair_drv_flags & IFF_DRV_OACTIVE) != 0) {
550 * Our hardware queue is full, try to fall back
551 * queuing to the ifq but do not call ifp->if_start.
552 * Either we are lucky or the packet is gone.
554 IFQ_ENQUEUE(&ifp->if_snd, m, error);
556 (void)epair_add_ifp_for_draining(ifp);
561 * Add a reference so the interface cannot go while the
562 * packet is in transit as we rely on rcvif to stay valid.
564 EPAIR_REFCOUNT_AQUIRE(&sc->refcount);
565 m->m_pkthdr.rcvif = oifp;
566 CURVNET_SET_QUIET(oifp->if_vnet);
567 error = netisr_queue(NETISR_EPAIR, m);
570 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
572 * IFQ_HANDOFF_ADJ/ip_handoff() update statistics,
573 * but as we bypass all this we have to duplicate
574 * the logic another time.
576 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
577 if (mflags & (M_BCAST|M_MCAST))
578 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
579 /* Someone else received the packet. */
580 if_inc_counter(oifp, IFCOUNTER_IPACKETS, 1);
582 /* The packet was freed already. */
583 epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE;
584 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
585 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
586 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
587 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1,
588 ("%s: ifp=%p sc->refcount not >= 1: %d",
589 __func__, oifp, sc->refcount));
596 epair_transmit(struct ifnet *ifp, struct mbuf *m)
598 struct epair_dpcpu *epair_dpcpu;
601 epair_dpcpu = DPCPU_ID_PTR(cpuid_from_ifp(ifp), epair_dpcpu);
602 EPAIR_LOCK(epair_dpcpu);
603 error = epair_transmit_locked(ifp, m);
604 EPAIR_UNLOCK(epair_dpcpu);
609 epair_qflush(struct ifnet *ifp)
611 struct epair_softc *sc;
616 * See epair_clone_destroy(), we can end up getting called twice.
617 * Don't do anything on the second call.
623 * Remove this ifp from all backpointer lists. The interface will not
624 * usable for flushing anyway nor should it have anything to flush
627 epair_remove_ifp_from_draining(ifp);
634 epair_media_change(struct ifnet *ifp __unused)
642 epair_media_status(struct ifnet *ifp __unused, struct ifmediareq *imr)
645 imr->ifm_status = IFM_AVALID | IFM_ACTIVE;
646 imr->ifm_active = IFM_ETHER | IFM_10G_T | IFM_FDX;
650 epair_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
652 struct epair_softc *sc;
656 ifr = (struct ifreq *)data;
667 error = ifmedia_ioctl(ifp, ifr, &sc->media, cmd);
671 /* We basically allow all kinds of MTUs. */
672 ifp->if_mtu = ifr->ifr_mtu;
677 /* Let the common ethernet handler process this. */
678 error = ether_ioctl(ifp, cmd, data);
686 epair_init(void *dummy __unused)
691 * Interface cloning functions.
692 * We use our private ones so that we can create/destroy our secondary
693 * device along with the primary one.
696 epair_clone_match(struct if_clone *ifc, const char *name)
700 DPRINTF("name='%s'\n", name);
703 * Our base name is epair.
704 * Our interfaces will be named epair<n>[ab].
705 * So accept anything of the following list:
708 * but not the epair<n>[ab] versions.
710 if (strncmp(epairname, name, sizeof(epairname)-1) != 0)
713 for (cp = name + sizeof(epairname) - 1; *cp != '\0'; cp++) {
714 if (*cp < '0' || *cp > '9')
722 epair_clone_add(struct if_clone *ifc, struct epair_softc *scb)
725 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
728 /* Copy epairNa etheraddr and change the last byte. */
729 memcpy(eaddr, scb->oifp->if_hw_addr, ETHER_ADDR_LEN);
731 ether_ifattach(ifp, eaddr);
733 if_clone_addif(ifc, ifp);
737 epair_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
739 struct epair_softc *sca, *scb;
742 int error, unit, wildcard;
746 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
748 /* Try to see if a special unit was requested. */
749 error = ifc_name2unit(name, &unit);
752 wildcard = (unit < 0);
754 error = ifc_alloc_unit(ifc, &unit);
759 * If no unit had been given, we need to adjust the ifName.
760 * Also make sure there is space for our extra [ab] suffix.
762 for (dp = name; *dp != '\0'; dp++);
764 error = snprintf(dp, len - (dp - name), "%d", unit);
765 if (error > len - (dp - name) - 1) {
766 /* ifName too long. */
767 ifc_free_unit(ifc, unit);
772 if (len - (dp - name) - 1 < 1) {
773 /* No space left for our [ab] suffix. */
774 ifc_free_unit(ifc, unit);
778 /* Must not change dp so we can replace 'a' by 'b' later. */
781 /* Check if 'a' and 'b' interfaces already exist. */
782 if (ifunit(name) != NULL)
785 if (ifunit(name) != NULL)
788 /* Allocate memory for both [ab] interfaces */
789 sca = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
790 EPAIR_REFCOUNT_INIT(&sca->refcount, 1);
791 sca->ifp = if_alloc(IFT_ETHER);
792 if (sca->ifp == NULL) {
794 ifc_free_unit(ifc, unit);
798 scb = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
799 EPAIR_REFCOUNT_INIT(&scb->refcount, 1);
800 scb->ifp = if_alloc(IFT_ETHER);
801 if (scb->ifp == NULL) {
805 ifc_free_unit(ifc, unit);
810 * Cross-reference the interfaces so we will be able to free both.
812 sca->oifp = scb->ifp;
813 scb->oifp = sca->ifp;
816 * Calculate the cpuid for netisr queueing based on the
817 * ifIndex of the interfaces. As long as we cannot configure
818 * this or use cpuset information easily we cannot guarantee
819 * cache locality but we can at least allow parallelism.
822 netisr_get_cpuid(sca->ifp->if_index);
824 netisr_get_cpuid(scb->ifp->if_index);
826 /* Initialise pseudo media types. */
827 ifmedia_init(&sca->media, 0, epair_media_change, epair_media_status);
828 ifmedia_add(&sca->media, IFM_ETHER | IFM_10G_T, 0, NULL);
829 ifmedia_set(&sca->media, IFM_ETHER | IFM_10G_T);
830 ifmedia_init(&scb->media, 0, epair_media_change, epair_media_status);
831 ifmedia_add(&scb->media, IFM_ETHER | IFM_10G_T, 0, NULL);
832 ifmedia_set(&scb->media, IFM_ETHER | IFM_10G_T);
834 /* Finish initialization of interface <n>a. */
837 strlcpy(ifp->if_xname, name, IFNAMSIZ);
838 ifp->if_dname = epairname;
839 ifp->if_dunit = unit;
840 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
841 ifp->if_capabilities = IFCAP_VLAN_MTU;
842 ifp->if_capenable = IFCAP_VLAN_MTU;
843 ifp->if_start = epair_start;
844 ifp->if_ioctl = epair_ioctl;
845 ifp->if_init = epair_init;
846 if_setsendqlen(ifp, ifqmaxlen);
847 if_setsendqready(ifp);
850 * Calculate the etheraddr hashing the hostid and the
851 * interface index. The result would be hopefully unique.
852 * Note that the "a" component of an epair instance may get moved
853 * to a different VNET after creation. In that case its index
854 * will be freed and the index can get reused by new epair instance.
855 * Make sure we do not create same etheraddr again.
857 getcredhostid(curthread->td_ucred, (unsigned long *)&hostid);
859 arc4rand(&hostid, sizeof(hostid), 0);
861 if (ifp->if_index > next_index)
862 next_index = ifp->if_index;
866 key[0] = (uint32_t)next_index;
867 key[1] = (uint32_t)(hostid & 0xffffffff);
868 key[2] = (uint32_t)((hostid >> 32) & 0xfffffffff);
869 hash = jenkins_hash32(key, 3, 0);
872 memcpy(&eaddr[1], &hash, 4);
874 ether_ifattach(ifp, eaddr);
875 sca->if_qflush = ifp->if_qflush;
876 ifp->if_qflush = epair_qflush;
877 ifp->if_transmit = epair_transmit;
878 ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */
880 /* Swap the name and finish initialization of interface <n>b. */
885 strlcpy(ifp->if_xname, name, IFNAMSIZ);
886 ifp->if_dname = epairname;
887 ifp->if_dunit = unit;
888 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
889 ifp->if_capabilities = IFCAP_VLAN_MTU;
890 ifp->if_capenable = IFCAP_VLAN_MTU;
891 ifp->if_start = epair_start;
892 ifp->if_ioctl = epair_ioctl;
893 ifp->if_init = epair_init;
894 if_setsendqlen(ifp, ifqmaxlen);
895 if_setsendqready(ifp);
896 /* We need to play some tricks here for the second interface. */
897 strlcpy(name, epairname, len);
899 /* Correctly set the name for the cloner list. */
900 strlcpy(name, scb->ifp->if_xname, len);
901 epair_clone_add(ifc, scb);
903 scb->if_qflush = ifp->if_qflush;
904 ifp->if_qflush = epair_qflush;
905 ifp->if_transmit = epair_transmit;
906 ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */
909 * Restore name to <n>a as the ifp for this will go into the
910 * cloner list for the initial call.
912 strlcpy(name, sca->ifp->if_xname, len);
913 DPRINTF("name='%s/%db' created sca=%p scb=%p\n", name, unit, sca, scb);
915 /* Tell the world, that we are ready to rock. */
916 sca->ifp->if_drv_flags |= IFF_DRV_RUNNING;
917 scb->ifp->if_drv_flags |= IFF_DRV_RUNNING;
918 if_link_state_change(sca->ifp, LINK_STATE_UP);
919 if_link_state_change(scb->ifp, LINK_STATE_UP);
925 epair_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
928 struct epair_softc *sca, *scb;
931 DPRINTF("ifp=%p\n", ifp);
934 * In case we called into if_clone_destroyif() ourselves
935 * again to remove the second interface, the softc will be
936 * NULL. In that case so not do anything but return success.
938 if (ifp->if_softc == NULL)
941 unit = ifp->if_dunit;
944 scb = oifp->if_softc;
946 DPRINTF("ifp=%p oifp=%p\n", ifp, oifp);
947 if_link_state_change(ifp, LINK_STATE_DOWN);
948 if_link_state_change(oifp, LINK_STATE_DOWN);
949 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
950 oifp->if_drv_flags &= ~IFF_DRV_RUNNING;
953 * Get rid of our second half. As the other of the two
954 * interfaces may reside in a different vnet, we need to
955 * switch before freeing them.
957 CURVNET_SET_QUIET(oifp->if_vnet);
958 ether_ifdetach(oifp);
960 * Wait for all packets to be dispatched to if_input.
961 * The numbers can only go down as the interface is
962 * detached so there is no need to use atomics.
964 DPRINTF("scb refcnt=%u\n", scb->refcount);
965 EPAIR_REFCOUNT_ASSERT(scb->refcount == 1,
966 ("%s: ifp=%p scb->refcount!=1: %d", __func__, oifp, scb->refcount));
967 oifp->if_softc = NULL;
968 error = if_clone_destroyif(ifc, oifp);
970 panic("%s: if_clone_destroyif() for our 2nd iface failed: %d",
973 ifmedia_removeall(&scb->media);
979 * Wait for all packets to be dispatched to if_input.
981 DPRINTF("sca refcnt=%u\n", sca->refcount);
982 EPAIR_REFCOUNT_ASSERT(sca->refcount == 1,
983 ("%s: ifp=%p sca->refcount!=1: %d", __func__, ifp, sca->refcount));
985 ifmedia_removeall(&sca->media);
987 ifc_free_unit(ifc, unit);
993 vnet_epair_init(const void *unused __unused)
996 V_epair_cloner = if_clone_advanced(epairname, 0,
997 epair_clone_match, epair_clone_create, epair_clone_destroy);
999 netisr_register_vnet(&epair_nh);
1002 VNET_SYSINIT(vnet_epair_init, SI_SUB_PSEUDO, SI_ORDER_ANY,
1003 vnet_epair_init, NULL);
1006 vnet_epair_uninit(const void *unused __unused)
1010 netisr_unregister_vnet(&epair_nh);
1012 if_clone_detach(V_epair_cloner);
1014 VNET_SYSUNINIT(vnet_epair_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
1015 vnet_epair_uninit, NULL);
1018 epair_uninit(const void *unused __unused)
1020 netisr_unregister(&epair_nh);
1021 epair_dpcpu_detach();
1023 printf("%s unloaded.\n", epairname);
1025 SYSUNINIT(epair_uninit, SI_SUB_INIT_IF, SI_ORDER_MIDDLE,
1026 epair_uninit, NULL);
1029 epair_modevent(module_t mod, int type, void *data)
1035 /* For now limit us to one global mutex and one inq. */
1037 epair_nh.nh_qlimit = 42 * ifqmaxlen; /* 42 shall be the number. */
1038 if (TUNABLE_INT_FETCH("net.link.epair.netisr_maxqlen", &qlimit))
1039 epair_nh.nh_qlimit = qlimit;
1040 netisr_register(&epair_nh);
1042 printf("%s initialized.\n", epairname);
1045 /* Handled in epair_uninit() */
1048 return (EOPNOTSUPP);
1053 static moduledata_t epair_mod = {
1059 DECLARE_MODULE(if_epair, epair_mod, SI_SUB_PSEUDO, SI_ORDER_MIDDLE);
1060 MODULE_VERSION(if_epair, 1);