xref: /freebsd/sys/net/if_epair.c (revision 5bb3134a8c21cb87b30e135ef168483f0333dabb)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2008 The FreeBSD Foundation
5  * Copyright (c) 2009-2021 Bjoern A. Zeeb <bz@FreeBSD.org>
6  *
7  * This software was developed by CK Software GmbH under sponsorship
8  * from the FreeBSD Foundation.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  * notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  * notice, this list of conditions and the following disclaimer in the
17  * documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*
33  * A pair of virtual back-to-back connected ethernet like interfaces
34  * (``two interfaces with a virtual cross-over cable'').
35  *
36  * This is mostly intended to be used to provide connectivity between
37  * different virtual network stack instances.
38  */
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include <sys/param.h>
44 #include <sys/hash.h>
45 #include <sys/jail.h>
46 #include <sys/kernel.h>
47 #include <sys/libkern.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/module.h>
51 #include <sys/proc.h>
52 #include <sys/queue.h>
53 #include <sys/smp.h>
54 #include <sys/socket.h>
55 #include <sys/sockio.h>
56 #include <sys/sysctl.h>
57 #include <sys/types.h>
58 #include <sys/buf_ring.h>
59 #include <sys/bus.h>
60 #include <sys/interrupt.h>
61 
62 #include <net/bpf.h>
63 #include <net/ethernet.h>
64 #include <net/if.h>
65 #include <net/if_var.h>
66 #include <net/if_clone.h>
67 #include <net/if_media.h>
68 #include <net/if_var.h>
69 #include <net/if_types.h>
70 #include <net/netisr.h>
71 #include <net/vnet.h>
72 
73 static int epair_clone_match(struct if_clone *, const char *);
74 static int epair_clone_create(struct if_clone *, char *, size_t, caddr_t);
75 static int epair_clone_destroy(struct if_clone *, struct ifnet *);
76 
77 static const char epairname[] = "epair";
78 #define	RXRSIZE	4096	/* Probably overkill by 4-8x. */
79 
80 static MALLOC_DEFINE(M_EPAIR, epairname,
81     "Pair of virtual cross-over connected Ethernet-like interfaces");
82 
83 VNET_DEFINE_STATIC(struct if_clone *, epair_cloner);
84 #define	V_epair_cloner	VNET(epair_cloner)
85 
86 static unsigned int next_index = 0;
87 #define	EPAIR_LOCK_INIT()		mtx_init(&epair_n_index_mtx, "epairidx", \
88 					    NULL, MTX_DEF)
89 #define	EPAIR_LOCK_DESTROY()		mtx_destroy(&epair_n_index_mtx)
90 #define	EPAIR_LOCK()			mtx_lock(&epair_n_index_mtx)
91 #define	EPAIR_UNLOCK()			mtx_unlock(&epair_n_index_mtx)
92 
93 static void				*swi_cookie[MAXCPU];	/* swi(9). */
94 static STAILQ_HEAD(, epair_softc)	swi_sc[MAXCPU];
95 
96 static struct mtx epair_n_index_mtx;
97 struct epair_softc {
98 	struct ifnet	*ifp;		/* This ifp. */
99 	struct ifnet	*oifp;		/* other ifp of pair. */
100 	void		*swi_cookie;	/* swi(9). */
101 	struct buf_ring	*rxring[2];
102 	volatile int	ridx;		/* 0 || 1 */
103 	struct ifmedia	media;		/* Media config (fake). */
104 	uint32_t	cpuidx;
105 	STAILQ_ENTRY(epair_softc) entry;
106 };
107 
108 static void
109 epair_clear_mbuf(struct mbuf *m)
110 {
111 	/* Remove any CSUM_SND_TAG as ether_input will barf. */
112 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
113 		m_snd_tag_rele(m->m_pkthdr.snd_tag);
114 		m->m_pkthdr.snd_tag = NULL;
115 		m->m_pkthdr.csum_flags &= ~CSUM_SND_TAG;
116 	}
117 
118 	m_tag_delete_nonpersistent(m);
119 }
120 
121 static void
122 epair_if_input(struct epair_softc *sc, int ridx)
123 {
124 	struct epoch_tracker et;
125 	struct ifnet *ifp;
126 	struct mbuf *m;
127 
128 	ifp = sc->ifp;
129 	NET_EPOCH_ENTER(et);
130 	do {
131 		m = buf_ring_dequeue_sc(sc->rxring[ridx]);
132 		if (m == NULL)
133 			break;
134 
135 		MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0);
136 		(*ifp->if_input)(ifp, m);
137 
138 	} while (1);
139 	NET_EPOCH_EXIT(et);
140 }
141 
142 static void
143 epair_sintr(struct epair_softc *sc)
144 {
145 	int ridx, nidx;
146 
147 	if_ref(sc->ifp);
148 	do {
149 		ridx = sc->ridx;
150 		nidx = (ridx == 0) ? 1 : 0;
151 	} while (!atomic_cmpset_int(&sc->ridx, ridx, nidx));
152 	epair_if_input(sc, ridx);
153 
154 	if_rele(sc->ifp);
155 }
156 
157 static void
158 epair_intr(void *arg)
159 {
160 	struct epair_softc *sc;
161 	uint32_t cpuidx;
162 
163 	cpuidx = (uintptr_t)arg;
164 	/* If this is a problem, this is a read-mostly situation. */
165 	EPAIR_LOCK();
166 	STAILQ_FOREACH(sc, &swi_sc[cpuidx], entry) {
167 		/* Do this lockless. */
168 		if (buf_ring_empty(sc->rxring[sc->ridx]))
169 			continue;
170 		epair_sintr(sc);
171 	}
172 	EPAIR_UNLOCK();
173 
174 	return;
175 }
176 
177 static int
178 epair_menq(struct mbuf *m, struct epair_softc *osc)
179 {
180 	struct ifnet *ifp, *oifp;
181 	int len, ret;
182 	int ridx;
183 	short mflags;
184 	bool was_empty;
185 
186 	/*
187 	 * I know this looks weird. We pass the "other sc" as we need that one
188 	 * and can get both ifps from it as well.
189 	 */
190 	oifp = osc->ifp;
191 	ifp = osc->oifp;
192 
193 	M_ASSERTPKTHDR(m);
194 	epair_clear_mbuf(m);
195 	if_setrcvif(m, oifp);
196 	M_SETFIB(m, oifp->if_fib);
197 
198 	/* Save values as once the mbuf is queued, it's not ours anymore. */
199 	len = m->m_pkthdr.len;
200 	mflags = m->m_flags;
201 
202 	MPASS(m->m_nextpkt == NULL);
203 	MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0);
204 
205 	ridx = atomic_load_int(&osc->ridx);
206 	was_empty = buf_ring_empty(osc->rxring[ridx]);
207 	ret = buf_ring_enqueue(osc->rxring[ridx], m);
208 	if (ret != 0) {
209 		/* Ring is full. */
210 		m_freem(m);
211 		return (0);
212 	}
213 
214 	if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
215 	/*
216 	 * IFQ_HANDOFF_ADJ/ip_handoff() update statistics,
217 	 * but as we bypass all this we have to duplicate
218 	 * the logic another time.
219 	 */
220 	if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
221 	if (mflags & (M_BCAST|M_MCAST))
222 		if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
223 	/* Someone else received the packet. */
224 	if_inc_counter(oifp, IFCOUNTER_IPACKETS, 1);
225 
226 	/* Kick the interrupt handler for the first packet. */
227 	if (was_empty && osc->swi_cookie != NULL)
228 		swi_sched(osc->swi_cookie, 0);
229 
230 	return (0);
231 }
232 
233 static void
234 epair_start(struct ifnet *ifp)
235 {
236 	struct mbuf *m;
237 	struct epair_softc *sc;
238 	struct ifnet *oifp;
239 
240 	/*
241 	 * We get packets here from ether_output via if_handoff()
242 	 * and need to put them into the input queue of the oifp
243 	 * and will put the packet into the receive-queue (rxq) of the
244 	 * other interface (oifp) of our pair.
245 	 */
246 	sc = ifp->if_softc;
247 	oifp = sc->oifp;
248 	sc = oifp->if_softc;
249 	for (;;) {
250 		IFQ_DEQUEUE(&ifp->if_snd, m);
251 		if (m == NULL)
252 			break;
253 		M_ASSERTPKTHDR(m);
254 		BPF_MTAP(ifp, m);
255 
256 		/* In case either interface is not usable drop the packet. */
257 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
258 		    (ifp->if_flags & IFF_UP) == 0 ||
259 		    (oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
260 		    (oifp->if_flags & IFF_UP) == 0) {
261 			m_freem(m);
262 			continue;
263 		}
264 
265 		(void) epair_menq(m, sc);
266 	}
267 }
268 
269 static int
270 epair_transmit(struct ifnet *ifp, struct mbuf *m)
271 {
272 	struct epair_softc *sc;
273 	struct ifnet *oifp;
274 	int error, len;
275 	short mflags;
276 
277 	if (m == NULL)
278 		return (0);
279 	M_ASSERTPKTHDR(m);
280 
281 	/*
282 	 * We are not going to use the interface en/dequeue mechanism
283 	 * on the TX side. We are called from ether_output_frame()
284 	 * and will put the packet into the receive-queue (rxq) of the
285 	 * other interface (oifp) of our pair.
286 	 */
287 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
288 		m_freem(m);
289 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
290 		return (ENXIO);
291 	}
292 	if ((ifp->if_flags & IFF_UP) == 0) {
293 		m_freem(m);
294 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
295 		return (ENETDOWN);
296 	}
297 
298 	BPF_MTAP(ifp, m);
299 
300 	/*
301 	 * In case the outgoing interface is not usable,
302 	 * drop the packet.
303 	 */
304 	sc = ifp->if_softc;
305 	oifp = sc->oifp;
306 	if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
307 	    (oifp->if_flags & IFF_UP) == 0) {
308 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
309 		m_freem(m);
310 		return (0);
311 	}
312 	len = m->m_pkthdr.len;
313 	mflags = m->m_flags;
314 
315 #ifdef ALTQ
316 	/* Support ALTQ via the classic if_start() path. */
317 	IF_LOCK(&ifp->if_snd);
318 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
319 		ALTQ_ENQUEUE(&ifp->if_snd, m, NULL, error);
320 		if (error)
321 			if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
322 		IF_UNLOCK(&ifp->if_snd);
323 		if (!error) {
324 			if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
325 			if (mflags & (M_BCAST|M_MCAST))
326 				if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
327 			epair_start(ifp);
328 		}
329 		return (error);
330 	}
331 	IF_UNLOCK(&ifp->if_snd);
332 #endif
333 
334 	error = epair_menq(m, oifp->if_softc);
335 	return (error);
336 }
337 
338 static int
339 epair_media_change(struct ifnet *ifp __unused)
340 {
341 
342 	/* Do nothing. */
343 	return (0);
344 }
345 
346 static void
347 epair_media_status(struct ifnet *ifp __unused, struct ifmediareq *imr)
348 {
349 
350 	imr->ifm_status = IFM_AVALID | IFM_ACTIVE;
351 	imr->ifm_active = IFM_ETHER | IFM_10G_T | IFM_FDX;
352 }
353 
354 static int
355 epair_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
356 {
357 	struct epair_softc *sc;
358 	struct ifreq *ifr;
359 	int error;
360 
361 	ifr = (struct ifreq *)data;
362 	switch (cmd) {
363 	case SIOCSIFFLAGS:
364 	case SIOCADDMULTI:
365 	case SIOCDELMULTI:
366 		error = 0;
367 		break;
368 
369 	case SIOCSIFMEDIA:
370 	case SIOCGIFMEDIA:
371 		sc = ifp->if_softc;
372 		error = ifmedia_ioctl(ifp, ifr, &sc->media, cmd);
373 		break;
374 
375 	case SIOCSIFMTU:
376 		/* We basically allow all kinds of MTUs. */
377 		ifp->if_mtu = ifr->ifr_mtu;
378 		error = 0;
379 		break;
380 
381 	default:
382 		/* Let the common ethernet handler process this. */
383 		error = ether_ioctl(ifp, cmd, data);
384 		break;
385 	}
386 
387 	return (error);
388 }
389 
390 static void
391 epair_init(void *dummy __unused)
392 {
393 }
394 
395 /*
396  * Interface cloning functions.
397  * We use our private ones so that we can create/destroy our secondary
398  * device along with the primary one.
399  */
400 static int
401 epair_clone_match(struct if_clone *ifc, const char *name)
402 {
403 	const char *cp;
404 
405 	/*
406 	 * Our base name is epair.
407 	 * Our interfaces will be named epair<n>[ab].
408 	 * So accept anything of the following list:
409 	 * - epair
410 	 * - epair<n>
411 	 * but not the epair<n>[ab] versions.
412 	 */
413 	if (strncmp(epairname, name, sizeof(epairname)-1) != 0)
414 		return (0);
415 
416 	for (cp = name + sizeof(epairname) - 1; *cp != '\0'; cp++) {
417 		if (*cp < '0' || *cp > '9')
418 			return (0);
419 	}
420 
421 	return (1);
422 }
423 
424 static void
425 epair_clone_add(struct if_clone *ifc, struct epair_softc *scb)
426 {
427 	struct ifnet *ifp;
428 	uint8_t eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
429 
430 	ifp = scb->ifp;
431 	/* Copy epairNa etheraddr and change the last byte. */
432 	memcpy(eaddr, scb->oifp->if_hw_addr, ETHER_ADDR_LEN);
433 	eaddr[5] = 0x0b;
434 	ether_ifattach(ifp, eaddr);
435 
436 	if_clone_addif(ifc, ifp);
437 }
438 
439 static int
440 epair_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
441 {
442 	struct epair_softc *sca, *scb;
443 	struct ifnet *ifp;
444 	char *dp;
445 	int error, unit, wildcard;
446 	uint64_t hostid;
447 	uint32_t key[3];
448 	uint32_t hash;
449 	uint8_t eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
450 
451 	/* Try to see if a special unit was requested. */
452 	error = ifc_name2unit(name, &unit);
453 	if (error != 0)
454 		return (error);
455 	wildcard = (unit < 0);
456 
457 	error = ifc_alloc_unit(ifc, &unit);
458 	if (error != 0)
459 		return (error);
460 
461 	/*
462 	 * If no unit had been given, we need to adjust the ifName.
463 	 * Also make sure there is space for our extra [ab] suffix.
464 	 */
465 	for (dp = name; *dp != '\0'; dp++);
466 	if (wildcard) {
467 		error = snprintf(dp, len - (dp - name), "%d", unit);
468 		if (error > len - (dp - name) - 1) {
469 			/* ifName too long. */
470 			ifc_free_unit(ifc, unit);
471 			return (ENOSPC);
472 		}
473 		dp += error;
474 	}
475 	if (len - (dp - name) - 1 < 1) {
476 		/* No space left for our [ab] suffix. */
477 		ifc_free_unit(ifc, unit);
478 		return (ENOSPC);
479 	}
480 	*dp = 'b';
481 	/* Must not change dp so we can replace 'a' by 'b' later. */
482 	*(dp+1) = '\0';
483 
484 	/* Check if 'a' and 'b' interfaces already exist. */
485 	if (ifunit(name) != NULL)
486 		return (EEXIST);
487 	*dp = 'a';
488 	if (ifunit(name) != NULL)
489 		return (EEXIST);
490 
491 	/* Allocate memory for both [ab] interfaces */
492 	sca = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
493 	sca->ifp = if_alloc(IFT_ETHER);
494 	if (sca->ifp == NULL) {
495 		free(sca, M_EPAIR);
496 		ifc_free_unit(ifc, unit);
497 		return (ENOSPC);
498 	}
499 	sca->rxring[0] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK,NULL);
500 	sca->rxring[1] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL);
501 
502 	scb = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
503 	scb->ifp = if_alloc(IFT_ETHER);
504 	if (scb->ifp == NULL) {
505 		free(scb, M_EPAIR);
506 		if_free(sca->ifp);
507 		free(sca, M_EPAIR);
508 		ifc_free_unit(ifc, unit);
509 		return (ENOSPC);
510 	}
511 	scb->rxring[0] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL);
512 	scb->rxring[1] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL);
513 
514 	/*
515 	 * Cross-reference the interfaces so we will be able to free both.
516 	 */
517 	sca->oifp = scb->ifp;
518 	scb->oifp = sca->ifp;
519 
520 	EPAIR_LOCK();
521 #ifdef SMP
522 	/* Get an approximate distribution. */
523 	hash = next_index % mp_ncpus;
524 #else
525 	hash = 0;
526 #endif
527 	if (swi_cookie[hash] == NULL) {
528 		void *cookie;
529 
530 		EPAIR_UNLOCK();
531 		error = swi_add(NULL, epairname,
532 		    epair_intr, (void *)(uintptr_t)hash,
533 		    SWI_NET, INTR_MPSAFE, &cookie);
534 		if (error) {
535 			buf_ring_free(scb->rxring[0], M_EPAIR);
536 			buf_ring_free(scb->rxring[1], M_EPAIR);
537 			if_free(scb->ifp);
538 			free(scb, M_EPAIR);
539 			buf_ring_free(sca->rxring[0], M_EPAIR);
540 			buf_ring_free(sca->rxring[1], M_EPAIR);
541 			if_free(sca->ifp);
542 			free(sca, M_EPAIR);
543 			ifc_free_unit(ifc, unit);
544 			return (ENOSPC);
545 		}
546 		EPAIR_LOCK();
547 		/* Recheck under lock even though a race is very unlikely. */
548 		if (swi_cookie[hash] == NULL) {
549 			swi_cookie[hash] = cookie;
550 		} else {
551 			EPAIR_UNLOCK();
552 			(void) swi_remove(cookie);
553 			EPAIR_LOCK();
554 		}
555 	}
556 	sca->cpuidx = hash;
557 	STAILQ_INSERT_TAIL(&swi_sc[hash], sca, entry);
558 	sca->swi_cookie = swi_cookie[hash];
559 	scb->cpuidx = hash;
560 	STAILQ_INSERT_TAIL(&swi_sc[hash], scb, entry);
561 	scb->swi_cookie = swi_cookie[hash];
562 	EPAIR_UNLOCK();
563 
564 	/* Initialise pseudo media types. */
565 	ifmedia_init(&sca->media, 0, epair_media_change, epair_media_status);
566 	ifmedia_add(&sca->media, IFM_ETHER | IFM_10G_T, 0, NULL);
567 	ifmedia_set(&sca->media, IFM_ETHER | IFM_10G_T);
568 	ifmedia_init(&scb->media, 0, epair_media_change, epair_media_status);
569 	ifmedia_add(&scb->media, IFM_ETHER | IFM_10G_T, 0, NULL);
570 	ifmedia_set(&scb->media, IFM_ETHER | IFM_10G_T);
571 
572 	/* Finish initialization of interface <n>a. */
573 	ifp = sca->ifp;
574 	ifp->if_softc = sca;
575 	strlcpy(ifp->if_xname, name, IFNAMSIZ);
576 	ifp->if_dname = epairname;
577 	ifp->if_dunit = unit;
578 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
579 	ifp->if_flags |= IFF_KNOWSEPOCH;
580 	ifp->if_capabilities = IFCAP_VLAN_MTU;
581 	ifp->if_capenable = IFCAP_VLAN_MTU;
582 	ifp->if_start = epair_start;
583 	ifp->if_ioctl = epair_ioctl;
584 	ifp->if_init  = epair_init;
585 	if_setsendqlen(ifp, ifqmaxlen);
586 	if_setsendqready(ifp);
587 
588 	/*
589 	 * Calculate the etheraddr hashing the hostid and the
590 	 * interface index. The result would be hopefully unique.
591 	 * Note that the "a" component of an epair instance may get moved
592 	 * to a different VNET after creation. In that case its index
593 	 * will be freed and the index can get reused by new epair instance.
594 	 * Make sure we do not create same etheraddr again.
595 	 */
596 	getcredhostid(curthread->td_ucred, (unsigned long *)&hostid);
597 	if (hostid == 0)
598 		arc4rand(&hostid, sizeof(hostid), 0);
599 
600 	EPAIR_LOCK();
601 	if (ifp->if_index > next_index)
602 		next_index = ifp->if_index;
603 	else
604 		next_index++;
605 
606 	key[0] = (uint32_t)next_index;
607 	EPAIR_UNLOCK();
608 	key[1] = (uint32_t)(hostid & 0xffffffff);
609 	key[2] = (uint32_t)((hostid >> 32) & 0xfffffffff);
610 	hash = jenkins_hash32(key, 3, 0);
611 
612 	eaddr[0] = 0x02;
613 	memcpy(&eaddr[1], &hash, 4);
614 	eaddr[5] = 0x0a;
615 	ether_ifattach(ifp, eaddr);
616 	ifp->if_baudrate = IF_Gbps(10);	/* arbitrary maximum */
617 	ifp->if_transmit = epair_transmit;
618 
619 	/* Swap the name and finish initialization of interface <n>b. */
620 	*dp = 'b';
621 
622 	ifp = scb->ifp;
623 	ifp->if_softc = scb;
624 	strlcpy(ifp->if_xname, name, IFNAMSIZ);
625 	ifp->if_dname = epairname;
626 	ifp->if_dunit = unit;
627 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
628 	ifp->if_capabilities = IFCAP_VLAN_MTU;
629 	ifp->if_capenable = IFCAP_VLAN_MTU;
630 	ifp->if_start = epair_start;
631 	ifp->if_ioctl = epair_ioctl;
632 	ifp->if_init  = epair_init;
633 	if_setsendqlen(ifp, ifqmaxlen);
634 	if_setsendqready(ifp);
635 	/* We need to play some tricks here for the second interface. */
636 	strlcpy(name, epairname, len);
637 
638 	/* Correctly set the name for the cloner list. */
639 	strlcpy(name, scb->ifp->if_xname, len);
640 	epair_clone_add(ifc, scb);
641 
642 	ifp->if_baudrate = IF_Gbps(10);	/* arbitrary maximum */
643 	ifp->if_transmit = epair_transmit;
644 
645 	/*
646 	 * Restore name to <n>a as the ifp for this will go into the
647 	 * cloner list for the initial call.
648 	 */
649 	strlcpy(name, sca->ifp->if_xname, len);
650 
651 	/* Tell the world, that we are ready to rock. */
652 	sca->ifp->if_drv_flags |= IFF_DRV_RUNNING;
653 	if_link_state_change(sca->ifp, LINK_STATE_UP);
654 	scb->ifp->if_drv_flags |= IFF_DRV_RUNNING;
655 	if_link_state_change(scb->ifp, LINK_STATE_UP);
656 
657 	return (0);
658 }
659 
660 static void
661 epair_drain_rings(struct epair_softc *sc)
662 {
663 	int ridx;
664 	struct mbuf *m;
665 
666 	for (ridx = 0; ridx < 2; ridx++) {
667 		do {
668 			m = buf_ring_dequeue_sc(sc->rxring[ridx]);
669 			if (m == NULL)
670 				break;
671 			m_freem(m);
672 		} while (1);
673 	}
674 }
675 
676 static int
677 epair_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
678 {
679 	struct ifnet *oifp;
680 	struct epair_softc *sca, *scb;
681 	int unit, error;
682 
683 	/*
684 	 * In case we called into if_clone_destroyif() ourselves
685 	 * again to remove the second interface, the softc will be
686 	 * NULL. In that case so not do anything but return success.
687 	 */
688 	if (ifp->if_softc == NULL)
689 		return (0);
690 
691 	unit = ifp->if_dunit;
692 	sca = ifp->if_softc;
693 	oifp = sca->oifp;
694 	scb = oifp->if_softc;
695 
696 	/* Frist get the interfaces down and detached. */
697 	if_link_state_change(ifp, LINK_STATE_DOWN);
698 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
699 	if_link_state_change(oifp, LINK_STATE_DOWN);
700 	oifp->if_drv_flags &= ~IFF_DRV_RUNNING;
701 
702 	ether_ifdetach(ifp);
703 	ether_ifdetach(oifp);
704 
705 	/* Second stop interrupt handler. */
706 	EPAIR_LOCK();
707 	STAILQ_REMOVE(&swi_sc[sca->cpuidx], sca, epair_softc, entry);
708 	STAILQ_REMOVE(&swi_sc[scb->cpuidx], scb, epair_softc, entry);
709 	EPAIR_UNLOCK();
710 	sca->swi_cookie = NULL;
711 	scb->swi_cookie = NULL;
712 
713 	/* Third free any queued packets and all the resources. */
714 	CURVNET_SET_QUIET(oifp->if_vnet);
715 	epair_drain_rings(scb);
716 	oifp->if_softc = NULL;
717 	error = if_clone_destroyif(ifc, oifp);
718 	if (error)
719 		panic("%s: if_clone_destroyif() for our 2nd iface failed: %d",
720 		    __func__, error);
721 	if_free(oifp);
722 	ifmedia_removeall(&scb->media);
723 	buf_ring_free(scb->rxring[0], M_EPAIR);
724 	buf_ring_free(scb->rxring[1], M_EPAIR);
725 	free(scb, M_EPAIR);
726 	CURVNET_RESTORE();
727 
728 	epair_drain_rings(sca);
729 	if_free(ifp);
730 	ifmedia_removeall(&sca->media);
731 	buf_ring_free(sca->rxring[0], M_EPAIR);
732 	buf_ring_free(sca->rxring[1], M_EPAIR);
733 	free(sca, M_EPAIR);
734 
735 	/* Last free the cloner unit. */
736 	ifc_free_unit(ifc, unit);
737 
738 	return (0);
739 }
740 
741 static void
742 vnet_epair_init(const void *unused __unused)
743 {
744 
745 	V_epair_cloner = if_clone_advanced(epairname, 0,
746 	    epair_clone_match, epair_clone_create, epair_clone_destroy);
747 }
748 VNET_SYSINIT(vnet_epair_init, SI_SUB_PSEUDO, SI_ORDER_ANY,
749     vnet_epair_init, NULL);
750 
751 static void
752 vnet_epair_uninit(const void *unused __unused)
753 {
754 
755 	if_clone_detach(V_epair_cloner);
756 }
757 VNET_SYSUNINIT(vnet_epair_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
758     vnet_epair_uninit, NULL);
759 
760 static int
761 epair_modevent(module_t mod, int type, void *data)
762 {
763 	int i;
764 
765 	switch (type) {
766 	case MOD_LOAD:
767 		for (i = 0; i < MAXCPU; i++) {
768 			swi_cookie[i] = NULL;
769 			STAILQ_INIT(&swi_sc[i]);
770 		}
771 		EPAIR_LOCK_INIT();
772 		if (bootverbose)
773 			printf("%s: %s initialized.\n", __func__, epairname);
774 		break;
775 	case MOD_UNLOAD:
776 		EPAIR_LOCK();
777 		for (i = 0; i < MAXCPU; i++) {
778 			if (!STAILQ_EMPTY(&swi_sc[i])) {
779 				printf("%s: swi_sc[%d] active\n", __func__, i);
780 				EPAIR_UNLOCK();
781 				return (EBUSY);
782 			}
783 		}
784 		EPAIR_UNLOCK();
785 		for (i = 0; i < MAXCPU; i++)
786 			if (swi_cookie[i] != NULL)
787 				(void) swi_remove(swi_cookie[i]);
788 		EPAIR_LOCK_DESTROY();
789 		if (bootverbose)
790 			printf("%s: %s unloaded.\n", __func__, epairname);
791 		break;
792 	default:
793 		return (EOPNOTSUPP);
794 	}
795 	return (0);
796 }
797 
798 static moduledata_t epair_mod = {
799 	"if_epair",
800 	epair_modevent,
801 	0
802 };
803 
804 DECLARE_MODULE(if_epair, epair_mod, SI_SUB_PSEUDO, SI_ORDER_MIDDLE);
805 MODULE_VERSION(if_epair, 3);
806