xref: /src/sys/net/if_epair.c (revision 574d46b25ab6f42ad97017bb8b72520e9539e36f)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 #include "opt_rss.h"
42 #include "opt_inet.h"
43 #include "opt_inet6.h"
44 
45 #include <sys/param.h>
46 #include <sys/bus.h>
47 #include <sys/hash.h>
48 #include <sys/interrupt.h>
49 #include <sys/jail.h>
50 #include <sys/kernel.h>
51 #include <sys/libkern.h>
52 #include <sys/malloc.h>
53 #include <sys/mbuf.h>
54 #include <sys/module.h>
55 #include <sys/proc.h>
56 #include <sys/queue.h>
57 #include <sys/sched.h>
58 #include <sys/smp.h>
59 #include <sys/socket.h>
60 #include <sys/sockio.h>
61 #include <sys/sysctl.h>
62 #include <sys/taskqueue.h>
63 
64 #include <net/bpf.h>
65 #include <net/ethernet.h>
66 #include <net/if.h>
67 #include <net/if_var.h>
68 #include <net/if_clone.h>
69 #include <net/if_media.h>
70 #include <net/if_private.h>
71 #include <net/if_types.h>
72 #include <net/if_vlan_var.h>
73 #include <net/netisr.h>
74 #ifdef RSS
75 #include <net/rss_config.h>
76 #ifdef INET
77 #include <netinet/in_rss.h>
78 #endif
79 #ifdef INET6
80 #include <netinet6/in6_rss.h>
81 #endif
82 #endif
83 #include <net/vnet.h>
84 
85 static const char epairname[] = "epair";
86 #define	RXRSIZE	4096	/* Probably overkill by 4-8x. */
87 
88 static MALLOC_DEFINE(M_EPAIR, epairname,
89     "Pair of virtual cross-over connected Ethernet-like interfaces");
90 
91 VNET_DEFINE_STATIC(struct if_clone *, epair_cloner);
92 #define	V_epair_cloner	VNET(epair_cloner)
93 
94 static unsigned int next_index = 0;
95 #define	EPAIR_LOCK_INIT()		mtx_init(&epair_n_index_mtx, "epairidx", \
96 					    NULL, MTX_DEF)
97 #define	EPAIR_LOCK_DESTROY()		mtx_destroy(&epair_n_index_mtx)
98 #define	EPAIR_LOCK()			mtx_lock(&epair_n_index_mtx)
99 #define	EPAIR_UNLOCK()			mtx_unlock(&epair_n_index_mtx)
100 
101 SYSCTL_DECL(_net_link);
102 static SYSCTL_NODE(_net_link, OID_AUTO, epair, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
103 	"Pair of virtual cross-over connected Ethernet-like interfaces");
104 
105 static bool use_ether_gen_addr = true;
106 SYSCTL_BOOL(_net_link_epair, OID_AUTO, ether_gen_addr, CTLFLAG_RWTUN,
107 	&use_ether_gen_addr, false,
108 	"Generate MAC with FreeBSD OUI using ether_gen_addr(9)");
109 
110 struct epair_softc;
111 struct epair_queue {
112 	struct mtx		 mtx;
113 	struct mbufq		 q;
114 	int			 id;
115 	enum {
116 		EPAIR_QUEUE_IDLE,
117 		EPAIR_QUEUE_WAKING,
118 		EPAIR_QUEUE_RUNNING,
119 	}			 state;
120 	struct task		 tx_task;
121 	struct epair_softc	*sc;
122 };
123 
124 static struct mtx epair_n_index_mtx;
125 struct epair_softc {
126 	struct ifnet		*ifp;		/* This ifp. */
127 	struct ifnet		*oifp;		/* other ifp of pair. */
128 	int			 num_queues;
129 	struct epair_queue	*queues;
130 	struct ifmedia		 media;		/* Media config (fake). */
131 	STAILQ_ENTRY(epair_softc) entry;
132 };
133 
134 struct epair_tasks_t {
135 	int			 tasks;
136 	struct taskqueue	 *tq[MAXCPU];
137 };
138 
139 static struct epair_tasks_t epair_tasks;
140 
141 static void
epair_clear_mbuf(struct mbuf * m)142 epair_clear_mbuf(struct mbuf *m)
143 {
144 	M_ASSERTPKTHDR(m);
145 
146 	/* Remove any CSUM_SND_TAG as ether_input will barf. */
147 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
148 		m_snd_tag_rele(m->m_pkthdr.snd_tag);
149 		m->m_pkthdr.snd_tag = NULL;
150 		m->m_pkthdr.csum_flags &= ~CSUM_SND_TAG;
151 	}
152 
153 	m_tag_delete_nonpersistent(m);
154 }
155 
156 static void
epair_tx_start_deferred(void * arg,int pending)157 epair_tx_start_deferred(void *arg, int pending)
158 {
159 	struct epair_queue *q = (struct epair_queue *)arg;
160 	if_t ifp;
161 	struct mbuf *m, *n;
162 	bool resched;
163 
164 	ifp = q->sc->ifp;
165 
166 	if_ref(ifp);
167 	CURVNET_SET(ifp->if_vnet);
168 
169 	mtx_lock(&q->mtx);
170 	m = mbufq_flush(&q->q);
171 	q->state = EPAIR_QUEUE_RUNNING;
172 	mtx_unlock(&q->mtx);
173 
174 	while (m != NULL) {
175 		n = STAILQ_NEXT(m, m_stailqpkt);
176 		m->m_nextpkt = NULL;
177 		if_input(ifp, m);
178 		m = n;
179 	}
180 
181 	/*
182 	 * Avoid flushing the queue more than once per task.  We can otherwise
183 	 * end up starving ourselves in a multi-epair routing configuration.
184 	 */
185 	mtx_lock(&q->mtx);
186 	if (!mbufq_empty(&q->q)) {
187 		resched = true;
188 		q->state = EPAIR_QUEUE_WAKING;
189 	} else {
190 		resched = false;
191 		q->state = EPAIR_QUEUE_IDLE;
192 	}
193 	mtx_unlock(&q->mtx);
194 
195 	if (resched)
196 		taskqueue_enqueue(epair_tasks.tq[q->id], &q->tx_task);
197 
198 	CURVNET_RESTORE();
199 	if_rele(ifp);
200 }
201 
202 static struct epair_queue *
epair_select_queue(struct epair_softc * sc,struct mbuf * m)203 epair_select_queue(struct epair_softc *sc, struct mbuf *m)
204 {
205 	uint32_t bucket;
206 #ifdef RSS
207 	struct ether_header *eh;
208 	int ret;
209 
210 	ret = rss_m2bucket(m, &bucket);
211 	if (ret) {
212 		/* Actually hash the packet. */
213 		eh = mtod(m, struct ether_header *);
214 
215 		switch (ntohs(eh->ether_type)) {
216 #ifdef INET
217 		case ETHERTYPE_IP:
218 			rss_soft_m2cpuid_v4(m, 0, &bucket);
219 			break;
220 #endif
221 #ifdef INET6
222 		case ETHERTYPE_IPV6:
223 			rss_soft_m2cpuid_v6(m, 0, &bucket);
224 			break;
225 #endif
226 		default:
227 			bucket = 0;
228 			break;
229 		}
230 	}
231 	bucket %= sc->num_queues;
232 #else
233 	bucket = 0;
234 #endif
235 	return (&sc->queues[bucket]);
236 }
237 
238 static void
epair_prepare_mbuf(struct mbuf * m,struct ifnet * src_ifp)239 epair_prepare_mbuf(struct mbuf *m, struct ifnet *src_ifp)
240 {
241 	M_ASSERTPKTHDR(m);
242 	epair_clear_mbuf(m);
243 	if_setrcvif(m, src_ifp);
244 	M_SETFIB(m, src_ifp->if_fib);
245 
246 	MPASS(m->m_nextpkt == NULL);
247 	MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0);
248 }
249 
250 static void
epair_menq(struct mbuf * m,struct epair_softc * osc)251 epair_menq(struct mbuf *m, struct epair_softc *osc)
252 {
253 	struct epair_queue *q;
254 	struct ifnet *ifp, *oifp;
255 	int error, len;
256 	bool mcast;
257 
258 	/*
259 	 * I know this looks weird. We pass the "other sc" as we need that one
260 	 * and can get both ifps from it as well.
261 	 */
262 	oifp = osc->ifp;
263 	ifp = osc->oifp;
264 
265 	epair_prepare_mbuf(m, oifp);
266 
267 	/* Save values as once the mbuf is queued, it's not ours anymore. */
268 	len = m->m_pkthdr.len;
269 	mcast = (m->m_flags & (M_BCAST | M_MCAST)) != 0;
270 
271 	q = epair_select_queue(osc, m);
272 
273 	mtx_lock(&q->mtx);
274 	if (q->state == EPAIR_QUEUE_IDLE) {
275 		q->state = EPAIR_QUEUE_WAKING;
276 		taskqueue_enqueue(epair_tasks.tq[q->id], &q->tx_task);
277 	}
278 	error = mbufq_enqueue(&q->q, m);
279 	mtx_unlock(&q->mtx);
280 
281 	if (error != 0) {
282 		m_freem(m);
283 		if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
284 	} else {
285 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
286 		if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
287 		if (mcast)
288 			if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
289 		if_inc_counter(oifp, IFCOUNTER_IPACKETS, 1);
290 	}
291 }
292 
293 static void
epair_start(struct ifnet * ifp)294 epair_start(struct ifnet *ifp)
295 {
296 	struct mbuf *m;
297 	struct epair_softc *sc;
298 	struct ifnet *oifp;
299 
300 	/*
301 	 * We get packets here from ether_output via if_handoff()
302 	 * and need to put them into the input queue of the oifp
303 	 * and will put the packet into the receive-queue (rxq) of the
304 	 * other interface (oifp) of our pair.
305 	 */
306 	sc = ifp->if_softc;
307 	oifp = sc->oifp;
308 	sc = oifp->if_softc;
309 	for (;;) {
310 		IFQ_DEQUEUE(&ifp->if_snd, m);
311 		if (m == NULL)
312 			break;
313 		M_ASSERTPKTHDR(m);
314 		BPF_MTAP(ifp, m);
315 
316 		/* In case either interface is not usable drop the packet. */
317 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
318 		    (ifp->if_flags & IFF_UP) == 0 ||
319 		    (oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
320 		    (oifp->if_flags & IFF_UP) == 0) {
321 			m_freem(m);
322 			continue;
323 		}
324 
325 		epair_menq(m, sc);
326 	}
327 }
328 
329 static int
epair_transmit(struct ifnet * ifp,struct mbuf * m)330 epair_transmit(struct ifnet *ifp, struct mbuf *m)
331 {
332 	struct epair_softc *sc;
333 	struct ifnet *oifp;
334 #ifdef ALTQ
335 	int len;
336 	bool mcast;
337 #endif
338 
339 	if (m == NULL)
340 		return (0);
341 	M_ASSERTPKTHDR(m);
342 
343 	/*
344 	 * We could just transmit this, but it makes testing easier if we're a
345 	 * little bit more like real hardware.
346 	 * Allow just that little bit extra for ethernet (and vlan) headers.
347 	 */
348 	if (m->m_pkthdr.len > (ifp->if_mtu + sizeof(struct ether_vlan_header))) {
349 		m_freem(m);
350 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
351 		return (E2BIG);
352 	}
353 
354 	/*
355 	 * We are not going to use the interface en/dequeue mechanism
356 	 * on the TX side. We are called from ether_output_frame()
357 	 * and will put the packet into the receive-queue (rxq) of the
358 	 * other interface (oifp) of our pair.
359 	 */
360 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
361 		m_freem(m);
362 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
363 		return (ENXIO);
364 	}
365 	if ((ifp->if_flags & IFF_UP) == 0) {
366 		m_freem(m);
367 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
368 		return (ENETDOWN);
369 	}
370 
371 	BPF_MTAP(ifp, m);
372 
373 	/*
374 	 * In case the outgoing interface is not usable,
375 	 * drop the packet.
376 	 */
377 	sc = ifp->if_softc;
378 	oifp = sc->oifp;
379 	if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
380 	    (oifp->if_flags & IFF_UP) == 0) {
381 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
382 		m_freem(m);
383 		return (0);
384 	}
385 
386 #ifdef ALTQ
387 	len = m->m_pkthdr.len;
388 	mcast = (m->m_flags & (M_BCAST | M_MCAST)) != 0;
389 	int error = 0;
390 
391 	/* Support ALTQ via the classic if_start() path. */
392 	IF_LOCK(&ifp->if_snd);
393 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
394 		ALTQ_ENQUEUE(&ifp->if_snd, m, NULL, error);
395 		if (error)
396 			if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
397 		IF_UNLOCK(&ifp->if_snd);
398 		if (!error) {
399 			if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
400 			if (mcast)
401 				if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
402 			epair_start(ifp);
403 		}
404 		return (error);
405 	}
406 	IF_UNLOCK(&ifp->if_snd);
407 #endif
408 
409 	epair_menq(m, oifp->if_softc);
410 	return (0);
411 }
412 
413 static void
epair_qflush(struct ifnet * ifp __unused)414 epair_qflush(struct ifnet *ifp __unused)
415 {
416 }
417 
418 static int
epair_media_change(struct ifnet * ifp __unused)419 epair_media_change(struct ifnet *ifp __unused)
420 {
421 
422 	/* Do nothing. */
423 	return (0);
424 }
425 
426 static void
epair_media_status(struct ifnet * ifp __unused,struct ifmediareq * imr)427 epair_media_status(struct ifnet *ifp __unused, struct ifmediareq *imr)
428 {
429 
430 	imr->ifm_status = IFM_AVALID | IFM_ACTIVE;
431 	imr->ifm_active = IFM_ETHER | IFM_10G_T | IFM_FDX;
432 }
433 
434 /*
435  * Update ifp->if_hwassist according to the current value of ifp->if_capenable.
436  */
437 static void
epair_caps_changed(struct ifnet * ifp)438 epair_caps_changed(struct ifnet *ifp)
439 {
440 	uint64_t hwassist = 0;
441 
442 	if (ifp->if_capenable & IFCAP_TXCSUM)
443 		hwassist |= CSUM_IP_TCP | CSUM_IP_UDP;
444 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
445 		hwassist |= CSUM_IP6_TCP | CSUM_IP6_UDP;
446 	ifp->if_hwassist = hwassist;
447 }
448 
449 static int
epair_ioctl(struct ifnet * ifp,u_long cmd,caddr_t data)450 epair_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
451 {
452 	struct epair_softc *sc;
453 	struct ifreq *ifr;
454 	int error;
455 
456 	ifr = (struct ifreq *)data;
457 	switch (cmd) {
458 	case SIOCSIFFLAGS:
459 	case SIOCADDMULTI:
460 	case SIOCDELMULTI:
461 		error = 0;
462 		break;
463 
464 	case SIOCSIFMEDIA:
465 	case SIOCGIFMEDIA:
466 		sc = ifp->if_softc;
467 		error = ifmedia_ioctl(ifp, ifr, &sc->media, cmd);
468 		break;
469 
470 	case SIOCSIFMTU:
471 		/* We basically allow all kinds of MTUs. */
472 		ifp->if_mtu = ifr->ifr_mtu;
473 		error = 0;
474 		break;
475 
476 	case SIOCGIFCAP:
477 		ifr->ifr_reqcap = ifp->if_capabilities;
478 		ifr->ifr_curcap = ifp->if_capenable;
479 		error = 0;
480 		break;
481 	case SIOCSIFCAP:
482 		/*
483 		 * Enable/disable capabilities as requested, besides
484 		 * IFCAP_RXCSUM(_IPV6), which always remain enabled.
485 		 * Incoming packets may have the mbuf flag CSUM_DATA_VALID set.
486 		 * Without IFCAP_RXCSUM(_IPV6), this flag would have to be
487 		 * removed, which does not seem helpful.
488 		 */
489 		ifp->if_capenable = ifr->ifr_reqcap | IFCAP_RXCSUM |
490 		    IFCAP_RXCSUM_IPV6;
491 		epair_caps_changed(ifp);
492 		VLAN_CAPABILITIES(ifp);
493 		/*
494 		 * If IFCAP_TXCSUM(_IPV6) has been changed, change it on the
495 		 * other epair interface as well.
496 		 * A bridge disables IFCAP_TXCSUM(_IPV6) when adding one epair
497 		 * interface if another interface in the bridge has it disabled.
498 		 * In that case this capability needs to be disabled on the
499 		 * other epair interface to avoid sending packets in the bridge
500 		 * that rely on this capability.
501 		 * Do the same for IFCAP_VLAN_HWTAGGING. If the sending epair
502 		 * end has this capability enabled, the other end has to have
503 		 * it enabled too. Otherwise, epair would have to add the VLAN
504 		 * tag in the Ethernet header.
505 		 */
506 		sc = ifp->if_softc;
507 		if ((ifp->if_capenable ^ sc->oifp->if_capenable) &
508 		    (IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6 | IFCAP_VLAN_HWTAGGING)) {
509 			sc->oifp->if_capenable &=
510 			    ~(IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6 |
511 			      IFCAP_VLAN_HWTAGGING);
512 			sc->oifp->if_capenable |= ifp->if_capenable &
513 			    (IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6 |
514 			     IFCAP_VLAN_HWTAGGING);
515 			epair_caps_changed(sc->oifp);
516 			VLAN_CAPABILITIES(sc->oifp);
517 		}
518 		error = 0;
519 		break;
520 
521 	default:
522 		/* Let the common ethernet handler process this. */
523 		error = ether_ioctl(ifp, cmd, data);
524 		break;
525 	}
526 
527 	return (error);
528 }
529 
530 static void
epair_init(void * dummy __unused)531 epair_init(void *dummy __unused)
532 {
533 }
534 
535 /*
536  * Interface cloning functions.
537  * We use our private ones so that we can create/destroy our secondary
538  * device along with the primary one.
539  */
540 static int
epair_clone_match(struct if_clone * ifc,const char * name)541 epair_clone_match(struct if_clone *ifc, const char *name)
542 {
543 	const char *cp;
544 
545 	/*
546 	 * Our base name is epair.
547 	 * Our interfaces will be named epair<n>[ab].
548 	 * So accept anything of the following list:
549 	 * - epair
550 	 * - epair<n>
551 	 * but not the epair<n>[ab] versions.
552 	 */
553 	if (strncmp(epairname, name, sizeof(epairname)-1) != 0)
554 		return (0);
555 
556 	for (cp = name + sizeof(epairname) - 1; *cp != '\0'; cp++) {
557 		if (*cp < '0' || *cp > '9')
558 			return (0);
559 	}
560 
561 	return (1);
562 }
563 
564 static void
epair_generate_mac_byname(struct epair_softc * sc,uint8_t eaddr[])565 epair_generate_mac_byname(struct epair_softc *sc, uint8_t eaddr[])
566 {
567 	struct ether_addr gen_eaddr;
568 	int i;
569 
570 	ether_gen_addr_byname(if_name(sc->ifp), &gen_eaddr);
571 	for (i = 0; i < ETHER_ADDR_LEN; i++)
572 		eaddr[i] = gen_eaddr.octet[i];
573 }
574 
575 static void
epair_clone_add(struct if_clone * ifc,struct epair_softc * scb)576 epair_clone_add(struct if_clone *ifc, struct epair_softc *scb)
577 {
578 	struct ifnet *ifp;
579 	uint8_t eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
580 
581 	ifp = scb->ifp;
582 	if (!use_ether_gen_addr) {
583 		/* Copy epairNa etheraddr and change the last byte. */
584 		memcpy(eaddr, scb->oifp->if_hw_addr, ETHER_ADDR_LEN);
585 		eaddr[5] = 0x0b;
586 	} else
587 		epair_generate_mac_byname(scb, eaddr);
588 	ether_ifattach(ifp, eaddr);
589 
590 	if_clone_addif(ifc, ifp);
591 }
592 
593 static struct epair_softc *
epair_alloc_sc(struct if_clone * ifc)594 epair_alloc_sc(struct if_clone *ifc)
595 {
596 	struct epair_softc *sc;
597 
598 	struct ifnet *ifp = if_alloc(IFT_ETHER);
599 	sc = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
600 	sc->ifp = ifp;
601 	sc->num_queues = epair_tasks.tasks;
602 	sc->queues = mallocarray(sc->num_queues, sizeof(struct epair_queue),
603 	    M_EPAIR, M_WAITOK);
604 	for (int i = 0; i < sc->num_queues; i++) {
605 		struct epair_queue *q = &sc->queues[i];
606 		q->id = i;
607 		q->state = EPAIR_QUEUE_IDLE;
608 		mtx_init(&q->mtx, "epairq", NULL, MTX_DEF | MTX_NEW);
609 		mbufq_init(&q->q, RXRSIZE);
610 		q->sc = sc;
611 		NET_TASK_INIT(&q->tx_task, 0, epair_tx_start_deferred, q);
612 	}
613 
614 	/* Initialise pseudo media types. */
615 	ifmedia_init(&sc->media, 0, epair_media_change, epair_media_status);
616 	ifmedia_add(&sc->media, IFM_ETHER | IFM_10G_T, 0, NULL);
617 	ifmedia_set(&sc->media, IFM_ETHER | IFM_10G_T);
618 
619 	return (sc);
620 }
621 
622 static void
epair_setup_ifp(struct epair_softc * sc,char * name,int unit)623 epair_setup_ifp(struct epair_softc *sc, char *name, int unit)
624 {
625 	struct ifnet *ifp = sc->ifp;
626 
627 	ifp->if_softc = sc;
628 	strlcpy(ifp->if_xname, name, IFNAMSIZ);
629 	ifp->if_dname = epairname;
630 	ifp->if_dunit = unit;
631 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
632 	ifp->if_capabilities =
633 	    IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING |
634 	    IFCAP_TXCSUM | IFCAP_RXCSUM |
635 	    IFCAP_TXCSUM_IPV6 | IFCAP_RXCSUM_IPV6;
636 	ifp->if_capenable = ifp->if_capabilities;
637 	epair_caps_changed(ifp);
638 	ifp->if_transmit = epair_transmit;
639 	ifp->if_qflush = epair_qflush;
640 	ifp->if_start = epair_start;
641 	ifp->if_ioctl = epair_ioctl;
642 	ifp->if_init  = epair_init;
643 	if_setsendqlen(ifp, ifqmaxlen);
644 	if_setsendqready(ifp);
645 
646 	ifp->if_baudrate = IF_Gbps(10);	/* arbitrary maximum */
647 }
648 
649 static void
epair_generate_mac(struct epair_softc * sc,uint8_t * eaddr)650 epair_generate_mac(struct epair_softc *sc, uint8_t *eaddr)
651 {
652 	uint32_t key[3];
653 	uint32_t hash;
654 	uint64_t hostid;
655 
656 	EPAIR_LOCK();
657 #ifdef SMP
658 	/* Get an approximate distribution. */
659 	hash = next_index % mp_ncpus;
660 #else
661 	hash = 0;
662 #endif
663 	EPAIR_UNLOCK();
664 
665 	/*
666 	 * Calculate the etheraddr hashing the hostid and the
667 	 * interface index. The result would be hopefully unique.
668 	 * Note that the "a" component of an epair instance may get moved
669 	 * to a different VNET after creation. In that case its index
670 	 * will be freed and the index can get reused by new epair instance.
671 	 * Make sure we do not create same etheraddr again.
672 	 */
673 	getcredhostid(curthread->td_ucred, (unsigned long *)&hostid);
674 	if (hostid == 0)
675 		arc4rand(&hostid, sizeof(hostid), 0);
676 
677 	struct ifnet *ifp = sc->ifp;
678 	EPAIR_LOCK();
679 	if (ifp->if_index > next_index)
680 		next_index = ifp->if_index;
681 	else
682 		next_index++;
683 
684 	key[0] = (uint32_t)next_index;
685 	EPAIR_UNLOCK();
686 	key[1] = (uint32_t)(hostid & 0xffffffff);
687 	key[2] = (uint32_t)((hostid >> 32) & 0xfffffffff);
688 	hash = jenkins_hash32(key, 3, 0);
689 
690 	eaddr[0] = 0x02;
691 	memcpy(&eaddr[1], &hash, 4);
692 	eaddr[5] = 0x0a;
693 }
694 
695 static void
epair_free_sc(struct epair_softc * sc)696 epair_free_sc(struct epair_softc *sc)
697 {
698 
699 	if_free(sc->ifp);
700 	ifmedia_removeall(&sc->media);
701 	for (int i = 0; i < sc->num_queues; i++) {
702 		struct epair_queue *q = &sc->queues[i];
703 		mtx_destroy(&q->mtx);
704 	}
705 	free(sc->queues, M_EPAIR);
706 	free(sc, M_EPAIR);
707 }
708 
709 static void
epair_set_state(struct ifnet * ifp,bool running)710 epair_set_state(struct ifnet *ifp, bool running)
711 {
712 	if (running) {
713 		ifp->if_drv_flags |= IFF_DRV_RUNNING;
714 		if_link_state_change(ifp, LINK_STATE_UP);
715 	} else {
716 		if_link_state_change(ifp, LINK_STATE_DOWN);
717 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
718 	}
719 }
720 
721 static int
epair_handle_unit(struct if_clone * ifc,char * name,size_t len,int * punit)722 epair_handle_unit(struct if_clone *ifc, char *name, size_t len, int *punit)
723 {
724 	int error = 0, unit, wildcard;
725 	char *dp;
726 
727 	/* Try to see if a special unit was requested. */
728 	error = ifc_name2unit(name, &unit);
729 	if (error != 0)
730 		return (error);
731 	wildcard = (unit < 0);
732 
733 	error = ifc_alloc_unit(ifc, &unit);
734 	if (error != 0)
735 		return (error);
736 
737 	/*
738 	 * If no unit had been given, we need to adjust the ifName.
739 	 * Also make sure there is space for our extra [ab] suffix.
740 	 */
741 	for (dp = name; *dp != '\0'; dp++);
742 	if (wildcard) {
743 		int slen = snprintf(dp, len - (dp - name), "%d", unit);
744 		if (slen > len - (dp - name) - 1) {
745 			/* ifName too long. */
746 			error = ENOSPC;
747 			goto done;
748 		}
749 		dp += slen;
750 	}
751 	if (len - (dp - name) - 1 < 1) {
752 		/* No space left for our [ab] suffix. */
753 		error = ENOSPC;
754 		goto done;
755 	}
756 	*dp = 'b';
757 	/* Must not change dp so we can replace 'a' by 'b' later. */
758 	*(dp+1) = '\0';
759 
760 	/* Check if 'a' and 'b' interfaces already exist. */
761 	if (ifunit(name) != NULL) {
762 		error = EEXIST;
763 		goto done;
764 	}
765 
766 	*dp = 'a';
767 	if (ifunit(name) != NULL) {
768 		error = EEXIST;
769 		goto done;
770 	}
771 	*punit = unit;
772 done:
773 	if (error != 0)
774 		ifc_free_unit(ifc, unit);
775 
776 	return (error);
777 }
778 
779 static int
epair_clone_create(struct if_clone * ifc,char * name,size_t len,struct ifc_data * ifd,struct ifnet ** ifpp)780 epair_clone_create(struct if_clone *ifc, char *name, size_t len,
781     struct ifc_data *ifd, struct ifnet **ifpp)
782 {
783 	struct epair_softc *sca, *scb;
784 	struct ifnet *ifp;
785 	char *dp;
786 	int error, unit;
787 	uint8_t eaddr[ETHER_ADDR_LEN];	/* 00:00:00:00:00:00 */
788 
789 	error = epair_handle_unit(ifc, name, len, &unit);
790 	if (error != 0)
791 		return (error);
792 
793 	/* Allocate memory for both [ab] interfaces */
794 	sca = epair_alloc_sc(ifc);
795 	scb = epair_alloc_sc(ifc);
796 
797 	/*
798 	 * Cross-reference the interfaces so we will be able to free both.
799 	 */
800 	sca->oifp = scb->ifp;
801 	scb->oifp = sca->ifp;
802 
803 	/* Finish initialization of interface <n>a. */
804 	ifp = sca->ifp;
805 	epair_setup_ifp(sca, name, unit);
806 	if (!use_ether_gen_addr)
807 		epair_generate_mac(sca, eaddr);
808 	else
809 		epair_generate_mac_byname(sca, eaddr);
810 
811 	ether_ifattach(ifp, eaddr);
812 
813 	/* Swap the name and finish initialization of interface <n>b. */
814 	dp = name + strlen(name) - 1;
815 	*dp = 'b';
816 
817 	epair_setup_ifp(scb, name, unit);
818 
819 	ifp = scb->ifp;
820 	/* We need to play some tricks here for the second interface. */
821 	strlcpy(name, epairname, len);
822 	/* Correctly set the name for the cloner list. */
823 	strlcpy(name, scb->ifp->if_xname, len);
824 
825 	epair_clone_add(ifc, scb);
826 
827 	/*
828 	 * Restore name to <n>a as the ifp for this will go into the
829 	 * cloner list for the initial call.
830 	 */
831 	strlcpy(name, sca->ifp->if_xname, len);
832 
833 	/* Tell the world, that we are ready to rock. */
834 	epair_set_state(sca->ifp, true);
835 	epair_set_state(scb->ifp, true);
836 
837 	*ifpp = sca->ifp;
838 
839 	return (0);
840 }
841 
842 static void
epair_drain_rings(struct epair_softc * sc)843 epair_drain_rings(struct epair_softc *sc)
844 {
845 	for (int i = 0; i < sc->num_queues; i++) {
846 		struct epair_queue *q;
847 		struct mbuf *m, *n;
848 
849 		q = &sc->queues[i];
850 		mtx_lock(&q->mtx);
851 		m = mbufq_flush(&q->q);
852 		mtx_unlock(&q->mtx);
853 
854 		for (; m != NULL; m = n) {
855 			n = m->m_nextpkt;
856 			m_freem(m);
857 		}
858 	}
859 }
860 
861 static int
epair_clone_destroy(struct if_clone * ifc,struct ifnet * ifp,uint32_t flags)862 epair_clone_destroy(struct if_clone *ifc, struct ifnet *ifp, uint32_t flags)
863 {
864 	struct ifnet *oifp;
865 	struct epair_softc *sca, *scb;
866 	int unit, error;
867 
868 	/*
869 	 * In case we called into if_clone_destroyif() ourselves
870 	 * again to remove the second interface, the softc will be
871 	 * NULL. In that case so not do anything but return success.
872 	 */
873 	if (ifp->if_softc == NULL)
874 		return (0);
875 
876 	unit = ifp->if_dunit;
877 	sca = ifp->if_softc;
878 	oifp = sca->oifp;
879 	scb = oifp->if_softc;
880 
881 	/* Frist get the interfaces down and detached. */
882 	epair_set_state(ifp, false);
883 	epair_set_state(oifp, false);
884 
885 	ether_ifdetach(ifp);
886 	ether_ifdetach(oifp);
887 
888 	/* Third free any queued packets and all the resources. */
889 	CURVNET_SET_QUIET(oifp->if_vnet);
890 	epair_drain_rings(scb);
891 	oifp->if_softc = NULL;
892 	error = if_clone_destroyif(ifc, oifp);
893 	if (error)
894 		panic("%s: if_clone_destroyif() for our 2nd iface failed: %d",
895 		    __func__, error);
896 	epair_free_sc(scb);
897 	CURVNET_RESTORE();
898 
899 	epair_drain_rings(sca);
900 	epair_free_sc(sca);
901 
902 	/* Last free the cloner unit. */
903 	ifc_free_unit(ifc, unit);
904 
905 	return (0);
906 }
907 
908 static void
vnet_epair_init(const void * unused __unused)909 vnet_epair_init(const void *unused __unused)
910 {
911 	struct if_clone_addreq req = {
912 		.match_f = epair_clone_match,
913 		.create_f = epair_clone_create,
914 		.destroy_f = epair_clone_destroy,
915 	};
916 	V_epair_cloner = ifc_attach_cloner(epairname, &req);
917 }
918 VNET_SYSINIT(vnet_epair_init, SI_SUB_PSEUDO, SI_ORDER_ANY,
919     vnet_epair_init, NULL);
920 
921 static void
vnet_epair_uninit(const void * unused __unused)922 vnet_epair_uninit(const void *unused __unused)
923 {
924 
925 	ifc_detach_cloner(V_epair_cloner);
926 }
927 VNET_SYSUNINIT(vnet_epair_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
928     vnet_epair_uninit, NULL);
929 
930 static int
epair_mod_init(void)931 epair_mod_init(void)
932 {
933 	char name[32];
934 	epair_tasks.tasks = 0;
935 
936 #ifdef RSS
937 	int cpu;
938 
939 	CPU_FOREACH(cpu) {
940 		cpuset_t cpu_mask;
941 
942 		/* Pin to this CPU so we get appropriate NUMA allocations. */
943 		thread_lock(curthread);
944 		sched_bind(curthread, cpu);
945 		thread_unlock(curthread);
946 
947 		snprintf(name, sizeof(name), "epair_task_%d", cpu);
948 
949 		epair_tasks.tq[cpu] = taskqueue_create(name, M_WAITOK,
950 		    taskqueue_thread_enqueue,
951 		    &epair_tasks.tq[cpu]);
952 		CPU_SETOF(cpu, &cpu_mask);
953 		taskqueue_start_threads_cpuset(&epair_tasks.tq[cpu], 1, PI_NET,
954 		    &cpu_mask, "%s", name);
955 
956 		epair_tasks.tasks++;
957 	}
958 	thread_lock(curthread);
959 	sched_unbind(curthread);
960 	thread_unlock(curthread);
961 #else
962 	snprintf(name, sizeof(name), "epair_task");
963 
964 	epair_tasks.tq[0] = taskqueue_create(name, M_WAITOK,
965 	    taskqueue_thread_enqueue,
966 	    &epair_tasks.tq[0]);
967 	taskqueue_start_threads(&epair_tasks.tq[0], 1, PI_NET, "%s", name);
968 
969 	epair_tasks.tasks = 1;
970 #endif
971 
972 	return (0);
973 }
974 
975 static void
epair_mod_cleanup(void)976 epair_mod_cleanup(void)
977 {
978 
979 	for (int i = 0; i < epair_tasks.tasks; i++) {
980 		taskqueue_drain_all(epair_tasks.tq[i]);
981 		taskqueue_free(epair_tasks.tq[i]);
982 	}
983 }
984 
985 static int
epair_modevent(module_t mod,int type,void * data)986 epair_modevent(module_t mod, int type, void *data)
987 {
988 	int ret;
989 
990 	switch (type) {
991 	case MOD_LOAD:
992 		EPAIR_LOCK_INIT();
993 		ret = epair_mod_init();
994 		if (ret != 0)
995 			return (ret);
996 		if (bootverbose)
997 			printf("%s: %s initialized.\n", __func__, epairname);
998 		break;
999 	case MOD_UNLOAD:
1000 		epair_mod_cleanup();
1001 		EPAIR_LOCK_DESTROY();
1002 		if (bootverbose)
1003 			printf("%s: %s unloaded.\n", __func__, epairname);
1004 		break;
1005 	default:
1006 		return (EOPNOTSUPP);
1007 	}
1008 	return (0);
1009 }
1010 
1011 static moduledata_t epair_mod = {
1012 	"if_epair",
1013 	epair_modevent,
1014 	0
1015 };
1016 
1017 DECLARE_MODULE(if_epair, epair_mod, SI_SUB_PSEUDO, SI_ORDER_MIDDLE);
1018 MODULE_VERSION(if_epair, 3);
1019