xref: /netbsd/sys/net/if_ethersubr.c (revision 9eea1329)
1 /*	$NetBSD: if_ethersubr.c,v 1.323 2022/11/15 10:47:39 roy Exp $	*/
2 
3 /*
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT 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 PROJECT 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  * Copyright (c) 1982, 1989, 1993
34  *	The Regents of the University of California.  All rights reserved.
35  *
36  * Redistribution and use in source and binary forms, with or without
37  * modification, are permitted provided that the following conditions
38  * are met:
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 3. Neither the name of the University nor the names of its contributors
45  *    may be used to endorse or promote products derived from this software
46  *    without specific prior written permission.
47  *
48  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58  * SUCH DAMAGE.
59  *
60  *	@(#)if_ethersubr.c	8.2 (Berkeley) 4/4/96
61  */
62 
63 #include <sys/cdefs.h>
64 __KERNEL_RCSID(0, "$NetBSD: if_ethersubr.c,v 1.323 2022/11/15 10:47:39 roy Exp $");
65 
66 #ifdef _KERNEL_OPT
67 #include "opt_inet.h"
68 #include "opt_atalk.h"
69 #include "opt_mbuftrace.h"
70 #include "opt_mpls.h"
71 #include "opt_gateway.h"
72 #include "opt_pppoe.h"
73 #include "opt_net_mpsafe.h"
74 #endif
75 
76 #include "vlan.h"
77 #include "pppoe.h"
78 #include "bridge.h"
79 #include "arp.h"
80 #include "agr.h"
81 
82 #include <sys/sysctl.h>
83 #include <sys/mbuf.h>
84 #include <sys/mutex.h>
85 #include <sys/ioctl.h>
86 #include <sys/errno.h>
87 #include <sys/device.h>
88 #include <sys/entropy.h>
89 #include <sys/rndsource.h>
90 #include <sys/cpu.h>
91 #include <sys/kmem.h>
92 #include <sys/hook.h>
93 
94 #include <net/if.h>
95 #include <net/route.h>
96 #include <net/if_llc.h>
97 #include <net/if_dl.h>
98 #include <net/if_types.h>
99 #include <net/pktqueue.h>
100 
101 #include <net/if_media.h>
102 #include <dev/mii/mii.h>
103 #include <dev/mii/miivar.h>
104 
105 #if NARP == 0
106 /*
107  * XXX there should really be a way to issue this warning from within config(8)
108  */
109 #error You have included NETATALK or a pseudo-device in your configuration that depends on the presence of ethernet interfaces, but have no such interfaces configured. Check if you really need pseudo-device bridge, pppoe, vlan or options NETATALK.
110 #endif
111 
112 #include <net/bpf.h>
113 
114 #include <net/if_ether.h>
115 #include <net/if_vlanvar.h>
116 
117 #if NPPPOE > 0
118 #include <net/if_pppoe.h>
119 #endif
120 
121 #if NAGR > 0
122 #include <net/ether_slowprotocols.h>
123 #include <net/agr/ieee8023ad.h>
124 #include <net/agr/if_agrvar.h>
125 #endif
126 
127 #if NBRIDGE > 0
128 #include <net/if_bridgevar.h>
129 #endif
130 
131 #include <netinet/in.h>
132 #ifdef INET
133 #include <netinet/in_var.h>
134 #endif
135 #include <netinet/if_inarp.h>
136 
137 #ifdef INET6
138 #ifndef INET
139 #include <netinet/in.h>
140 #endif
141 #include <netinet6/in6_var.h>
142 #include <netinet6/nd6.h>
143 #endif
144 
145 #include "carp.h"
146 #if NCARP > 0
147 #include <netinet/ip_carp.h>
148 #endif
149 
150 #ifdef NETATALK
151 #include <netatalk/at.h>
152 #include <netatalk/at_var.h>
153 #include <netatalk/at_extern.h>
154 
155 #define llc_snap_org_code llc_un.type_snap.org_code
156 #define llc_snap_ether_type llc_un.type_snap.ether_type
157 
158 extern u_char	at_org_code[3];
159 extern u_char	aarp_org_code[3];
160 #endif /* NETATALK */
161 
162 #ifdef MPLS
163 #include <netmpls/mpls.h>
164 #include <netmpls/mpls_var.h>
165 #endif
166 
167 CTASSERT(sizeof(struct ether_addr) == 6);
168 CTASSERT(sizeof(struct ether_header) == 14);
169 
170 #ifdef DIAGNOSTIC
171 static struct timeval bigpktppslim_last;
172 static int bigpktppslim = 2;	/* XXX */
173 static int bigpktpps_count;
174 static kmutex_t bigpktpps_lock __cacheline_aligned;
175 #endif
176 
177 const uint8_t etherbroadcastaddr[ETHER_ADDR_LEN] =
178     { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
179 const uint8_t ethermulticastaddr_slowprotocols[ETHER_ADDR_LEN] =
180     { 0x01, 0x80, 0xc2, 0x00, 0x00, 0x02 };
181 #define senderr(e) { error = (e); goto bad;}
182 
183 static pktq_rps_hash_func_t ether_pktq_rps_hash_p;
184 
185 static int ether_output(struct ifnet *, struct mbuf *,
186     const struct sockaddr *, const struct rtentry *);
187 
188 /*
189  * Ethernet output routine.
190  * Encapsulate a packet of type family for the local net.
191  * Assumes that ifp is actually pointer to ethercom structure.
192  */
193 static int
ether_output(struct ifnet * const ifp0,struct mbuf * const m0,const struct sockaddr * const dst,const struct rtentry * rt)194 ether_output(struct ifnet * const ifp0, struct mbuf * const m0,
195     const struct sockaddr * const dst, const struct rtentry *rt)
196 {
197 	uint8_t esrc[ETHER_ADDR_LEN], edst[ETHER_ADDR_LEN];
198 	uint16_t etype = 0;
199 	int error = 0, hdrcmplt = 0;
200 	struct mbuf *m = m0;
201 	struct mbuf *mcopy = NULL;
202 	struct ether_header *eh;
203 	struct ifnet *ifp = ifp0;
204 #ifdef INET
205 	struct arphdr *ah;
206 #endif
207 #ifdef NETATALK
208 	struct at_ifaddr *aa;
209 #endif
210 
211 #ifdef MBUFTRACE
212 	m_claimm(m, ifp->if_mowner);
213 #endif
214 
215 #if NCARP > 0
216 	if (ifp->if_type == IFT_CARP) {
217 		struct ifaddr *ifa;
218 		int s = pserialize_read_enter();
219 
220 		/* loop back if this is going to the carp interface */
221 		if (dst != NULL && ifp0->if_link_state == LINK_STATE_UP &&
222 		    (ifa = ifa_ifwithaddr(dst)) != NULL) {
223 			if (ifa->ifa_ifp == ifp0) {
224 				pserialize_read_exit(s);
225 				return looutput(ifp0, m, dst, rt);
226 			}
227 		}
228 		pserialize_read_exit(s);
229 
230 		ifp = ifp->if_carpdev;
231 		/* ac = (struct arpcom *)ifp; */
232 
233 		if ((ifp0->if_flags & (IFF_UP | IFF_RUNNING)) !=
234 		    (IFF_UP | IFF_RUNNING))
235 			senderr(ENETDOWN);
236 	}
237 #endif
238 
239 	if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) != (IFF_UP | IFF_RUNNING))
240 		senderr(ENETDOWN);
241 
242 	switch (dst->sa_family) {
243 
244 #ifdef INET
245 	case AF_INET:
246 		if (m->m_flags & M_BCAST) {
247 			memcpy(edst, etherbroadcastaddr, sizeof(edst));
248 		} else if (m->m_flags & M_MCAST) {
249 			ETHER_MAP_IP_MULTICAST(&satocsin(dst)->sin_addr, edst);
250 		} else {
251 			error = arpresolve(ifp0, rt, m, dst, edst, sizeof(edst));
252 			if (error)
253 				return (error == EWOULDBLOCK) ? 0 : error;
254 		}
255 		/* If broadcasting on a simplex interface, loopback a copy */
256 		if ((m->m_flags & M_BCAST) && (ifp->if_flags & IFF_SIMPLEX))
257 			mcopy = m_copypacket(m, M_DONTWAIT);
258 		etype = htons(ETHERTYPE_IP);
259 		break;
260 
261 	case AF_ARP:
262 		ah = mtod(m, struct arphdr *);
263 		if (m->m_flags & M_BCAST) {
264 			memcpy(edst, etherbroadcastaddr, sizeof(edst));
265 		} else {
266 			void *tha = ar_tha(ah);
267 
268 			if (tha == NULL) {
269 				/* fake with ARPHRD_IEEE1394 */
270 				m_freem(m);
271 				return 0;
272 			}
273 			memcpy(edst, tha, sizeof(edst));
274 		}
275 
276 		ah->ar_hrd = htons(ARPHRD_ETHER);
277 
278 		switch (ntohs(ah->ar_op)) {
279 		case ARPOP_REVREQUEST:
280 		case ARPOP_REVREPLY:
281 			etype = htons(ETHERTYPE_REVARP);
282 			break;
283 
284 		case ARPOP_REQUEST:
285 		case ARPOP_REPLY:
286 		default:
287 			etype = htons(ETHERTYPE_ARP);
288 		}
289 		break;
290 #endif
291 
292 #ifdef INET6
293 	case AF_INET6:
294 		if (m->m_flags & M_BCAST) {
295 			memcpy(edst, etherbroadcastaddr, sizeof(edst));
296 		} else if (m->m_flags & M_MCAST) {
297 			ETHER_MAP_IPV6_MULTICAST(&satocsin6(dst)->sin6_addr,
298 			    edst);
299 		} else {
300 			error = nd6_resolve(ifp0, rt, m, dst, edst,
301 			    sizeof(edst));
302 			if (error)
303 				return (error == EWOULDBLOCK) ? 0 : error;
304 		}
305 		etype = htons(ETHERTYPE_IPV6);
306 		break;
307 #endif
308 
309 #ifdef NETATALK
310 	case AF_APPLETALK: {
311 		struct ifaddr *ifa;
312 		int s;
313 
314 		KERNEL_LOCK(1, NULL);
315 
316 		if (!aarpresolve(ifp, m, (const struct sockaddr_at *)dst, edst)) {
317 			KERNEL_UNLOCK_ONE(NULL);
318 			return 0;
319 		}
320 
321 		/*
322 		 * ifaddr is the first thing in at_ifaddr
323 		 */
324 		s = pserialize_read_enter();
325 		ifa = at_ifawithnet((const struct sockaddr_at *)dst, ifp);
326 		if (ifa == NULL) {
327 			pserialize_read_exit(s);
328 			KERNEL_UNLOCK_ONE(NULL);
329 			senderr(EADDRNOTAVAIL);
330 		}
331 		aa = (struct at_ifaddr *)ifa;
332 
333 		/*
334 		 * In the phase 2 case, we need to prepend an mbuf for the
335 		 * llc header.
336 		 */
337 		if (aa->aa_flags & AFA_PHASE2) {
338 			struct llc llc;
339 
340 			M_PREPEND(m, sizeof(struct llc), M_DONTWAIT);
341 			if (m == NULL) {
342 				pserialize_read_exit(s);
343 				KERNEL_UNLOCK_ONE(NULL);
344 				senderr(ENOBUFS);
345 			}
346 
347 			llc.llc_dsap = llc.llc_ssap = LLC_SNAP_LSAP;
348 			llc.llc_control = LLC_UI;
349 			memcpy(llc.llc_snap_org_code, at_org_code,
350 			    sizeof(llc.llc_snap_org_code));
351 			llc.llc_snap_ether_type = htons(ETHERTYPE_ATALK);
352 			memcpy(mtod(m, void *), &llc, sizeof(struct llc));
353 		} else {
354 			etype = htons(ETHERTYPE_ATALK);
355 		}
356 		pserialize_read_exit(s);
357 		KERNEL_UNLOCK_ONE(NULL);
358 		break;
359 	}
360 #endif /* NETATALK */
361 
362 	case pseudo_AF_HDRCMPLT:
363 		hdrcmplt = 1;
364 		memcpy(esrc,
365 		    ((const struct ether_header *)dst->sa_data)->ether_shost,
366 		    sizeof(esrc));
367 		/* FALLTHROUGH */
368 
369 	case AF_UNSPEC:
370 		memcpy(edst,
371 		    ((const struct ether_header *)dst->sa_data)->ether_dhost,
372 		    sizeof(edst));
373 		/* AF_UNSPEC doesn't swap the byte order of the ether_type. */
374 		etype = ((const struct ether_header *)dst->sa_data)->ether_type;
375 		break;
376 
377 	default:
378 		printf("%s: can't handle af%d\n", ifp->if_xname,
379 		    dst->sa_family);
380 		senderr(EAFNOSUPPORT);
381 	}
382 
383 #ifdef MPLS
384 	{
385 		struct m_tag *mtag;
386 		mtag = m_tag_find(m, PACKET_TAG_MPLS);
387 		if (mtag != NULL) {
388 			/* Having the tag itself indicates it's MPLS */
389 			etype = htons(ETHERTYPE_MPLS);
390 			m_tag_delete(m, mtag);
391 		}
392 	}
393 #endif
394 
395 	if (mcopy)
396 		(void)looutput(ifp, mcopy, dst, rt);
397 
398 	KASSERT((m->m_flags & M_PKTHDR) != 0);
399 
400 	/*
401 	 * If no ether type is set, this must be a 802.2 formatted packet.
402 	 */
403 	if (etype == 0)
404 		etype = htons(m->m_pkthdr.len);
405 
406 	/*
407 	 * Add local net header. If no space in first mbuf, allocate another.
408 	 */
409 	M_PREPEND(m, sizeof(struct ether_header), M_DONTWAIT);
410 	if (m == NULL)
411 		senderr(ENOBUFS);
412 
413 	eh = mtod(m, struct ether_header *);
414 	/* Note: etype is already in network byte order. */
415 	memcpy(&eh->ether_type, &etype, sizeof(eh->ether_type));
416 	memcpy(eh->ether_dhost, edst, sizeof(edst));
417 	if (hdrcmplt) {
418 		memcpy(eh->ether_shost, esrc, sizeof(eh->ether_shost));
419 	} else {
420 	 	memcpy(eh->ether_shost, CLLADDR(ifp->if_sadl),
421 		    sizeof(eh->ether_shost));
422 	}
423 
424 #if NCARP > 0
425 	if (ifp0 != ifp && ifp0->if_type == IFT_CARP) {
426 	 	memcpy(eh->ether_shost, CLLADDR(ifp0->if_sadl),
427 		    sizeof(eh->ether_shost));
428 	}
429 #endif
430 
431 	if ((error = pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_OUT)) != 0)
432 		return error;
433 	if (m == NULL)
434 		return 0;
435 
436 #if NBRIDGE > 0
437 	/*
438 	 * Bridges require special output handling.
439 	 */
440 	if (ifp->if_bridge)
441 		return bridge_output(ifp, m, NULL, NULL);
442 #endif
443 
444 #if NCARP > 0
445 	if (ifp != ifp0)
446 		if_statadd(ifp0, if_obytes, m->m_pkthdr.len + ETHER_HDR_LEN);
447 #endif
448 
449 #ifdef ALTQ
450 	KERNEL_LOCK(1, NULL);
451 	/*
452 	 * If ALTQ is enabled on the parent interface, do
453 	 * classification; the queueing discipline might not
454 	 * require classification, but might require the
455 	 * address family/header pointer in the pktattr.
456 	 */
457 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
458 		altq_etherclassify(&ifp->if_snd, m);
459 	KERNEL_UNLOCK_ONE(NULL);
460 #endif
461 	return ifq_enqueue(ifp, m);
462 
463 bad:
464 	if_statinc(ifp, if_oerrors);
465 	if (m)
466 		m_freem(m);
467 	return error;
468 }
469 
470 #ifdef ALTQ
471 /*
472  * This routine is a slight hack to allow a packet to be classified
473  * if the Ethernet headers are present.  It will go away when ALTQ's
474  * classification engine understands link headers.
475  *
476  * XXX: We may need to do m_pullups here. First to ensure struct ether_header
477  * is indeed contiguous, then to read the LLC and so on.
478  */
479 void
altq_etherclassify(struct ifaltq * ifq,struct mbuf * m)480 altq_etherclassify(struct ifaltq *ifq, struct mbuf *m)
481 {
482 	struct ether_header *eh;
483 	struct mbuf *mtop = m;
484 	uint16_t ether_type;
485 	int hlen, af, hdrsize;
486 	void *hdr;
487 
488 	KASSERT((mtop->m_flags & M_PKTHDR) != 0);
489 
490 	hlen = ETHER_HDR_LEN;
491 	eh = mtod(m, struct ether_header *);
492 
493 	ether_type = htons(eh->ether_type);
494 
495 	if (ether_type < ETHERMTU) {
496 		/* LLC/SNAP */
497 		struct llc *llc = (struct llc *)(eh + 1);
498 		hlen += 8;
499 
500 		if (m->m_len < hlen ||
501 		    llc->llc_dsap != LLC_SNAP_LSAP ||
502 		    llc->llc_ssap != LLC_SNAP_LSAP ||
503 		    llc->llc_control != LLC_UI) {
504 			/* Not SNAP. */
505 			goto bad;
506 		}
507 
508 		ether_type = htons(llc->llc_un.type_snap.ether_type);
509 	}
510 
511 	switch (ether_type) {
512 	case ETHERTYPE_IP:
513 		af = AF_INET;
514 		hdrsize = 20;		/* sizeof(struct ip) */
515 		break;
516 
517 	case ETHERTYPE_IPV6:
518 		af = AF_INET6;
519 		hdrsize = 40;		/* sizeof(struct ip6_hdr) */
520 		break;
521 
522 	default:
523 		af = AF_UNSPEC;
524 		hdrsize = 0;
525 		break;
526 	}
527 
528 	while (m->m_len <= hlen) {
529 		hlen -= m->m_len;
530 		m = m->m_next;
531 		if (m == NULL)
532 			goto bad;
533 	}
534 
535 	if (m->m_len < (hlen + hdrsize)) {
536 		/*
537 		 * protocol header not in a single mbuf.
538 		 * We can't cope with this situation right
539 		 * now (but it shouldn't ever happen, really, anyhow).
540 		 */
541 #ifdef DEBUG
542 		printf("altq_etherclassify: headers span multiple mbufs: "
543 		    "%d < %d\n", m->m_len, (hlen + hdrsize));
544 #endif
545 		goto bad;
546 	}
547 
548 	m->m_data += hlen;
549 	m->m_len -= hlen;
550 
551 	hdr = mtod(m, void *);
552 
553 	if (ALTQ_NEEDS_CLASSIFY(ifq)) {
554 		mtop->m_pkthdr.pattr_class =
555 		    (*ifq->altq_classify)(ifq->altq_clfier, m, af);
556 	}
557 	mtop->m_pkthdr.pattr_af = af;
558 	mtop->m_pkthdr.pattr_hdr = hdr;
559 
560 	m->m_data -= hlen;
561 	m->m_len += hlen;
562 
563 	return;
564 
565 bad:
566 	mtop->m_pkthdr.pattr_class = NULL;
567 	mtop->m_pkthdr.pattr_hdr = NULL;
568 	mtop->m_pkthdr.pattr_af = AF_UNSPEC;
569 }
570 #endif /* ALTQ */
571 
572 #if defined (LLC) || defined (NETATALK)
573 static void
ether_input_llc(struct ifnet * ifp,struct mbuf * m,struct ether_header * eh)574 ether_input_llc(struct ifnet *ifp, struct mbuf *m, struct ether_header *eh)
575 {
576 	pktqueue_t *pktq = NULL;
577 	struct llc *l;
578 
579 	if (m->m_len < sizeof(*eh) + sizeof(struct llc))
580 		goto error;
581 
582 	l = (struct llc *)(eh+1);
583 	switch (l->llc_dsap) {
584 #ifdef NETATALK
585 	case LLC_SNAP_LSAP:
586 		switch (l->llc_control) {
587 		case LLC_UI:
588 			if (l->llc_ssap != LLC_SNAP_LSAP)
589 				goto error;
590 
591 			if (memcmp(&(l->llc_snap_org_code)[0],
592 			    at_org_code, sizeof(at_org_code)) == 0 &&
593 			    ntohs(l->llc_snap_ether_type) ==
594 			    ETHERTYPE_ATALK) {
595 				pktq = at_pktq2;
596 				m_adj(m, sizeof(struct ether_header)
597 				    + sizeof(struct llc));
598 				break;
599 			}
600 
601 			if (memcmp(&(l->llc_snap_org_code)[0],
602 			    aarp_org_code,
603 			    sizeof(aarp_org_code)) == 0 &&
604 			    ntohs(l->llc_snap_ether_type) ==
605 			    ETHERTYPE_AARP) {
606 				m_adj(m, sizeof(struct ether_header)
607 				    + sizeof(struct llc));
608 				aarpinput(ifp, m); /* XXX queue? */
609 				return;
610 			}
611 
612 		default:
613 			goto error;
614 		}
615 		break;
616 #endif
617 	default:
618 		goto noproto;
619 	}
620 
621 	KASSERT(pktq != NULL);
622 	if (__predict_false(!pktq_enqueue(pktq, m, 0))) {
623 		m_freem(m);
624 	}
625 	return;
626 
627 noproto:
628 	m_freem(m);
629 	if_statinc(ifp, if_noproto);
630 	return;
631 error:
632 	m_freem(m);
633 	if_statinc(ifp, if_ierrors);
634 	return;
635 }
636 #endif /* defined (LLC) || defined (NETATALK) */
637 
638 /*
639  * Process a received Ethernet packet;
640  * the packet is in the mbuf chain m with
641  * the ether header.
642  */
643 void
ether_input(struct ifnet * ifp,struct mbuf * m)644 ether_input(struct ifnet *ifp, struct mbuf *m)
645 {
646 #if NVLAN > 0 || defined(MBUFTRACE)
647 	struct ethercom *ec = (struct ethercom *) ifp;
648 #endif
649 	pktqueue_t *pktq = NULL;
650 	uint16_t etype;
651 	struct ether_header *eh;
652 	size_t ehlen;
653 	static int earlypkts;
654 
655 	/* No RPS for not-IP. */
656 	pktq_rps_hash_func_t rps_hash = NULL;
657 
658 	KASSERT(!cpu_intr_p());
659 	KASSERT((m->m_flags & M_PKTHDR) != 0);
660 
661 	if ((ifp->if_flags & IFF_UP) == 0)
662 		goto drop;
663 
664 #ifdef MBUFTRACE
665 	m_claimm(m, &ec->ec_rx_mowner);
666 #endif
667 
668 	if (__predict_false(m->m_len < sizeof(*eh))) {
669 		if ((m = m_pullup(m, sizeof(*eh))) == NULL) {
670 			if_statinc(ifp, if_ierrors);
671 			return;
672 		}
673 	}
674 
675 	eh = mtod(m, struct ether_header *);
676 	etype = ntohs(eh->ether_type);
677 	ehlen = sizeof(*eh);
678 
679 	if (__predict_false(earlypkts < 100 ||
680 		entropy_epoch() == (unsigned)-1)) {
681 		rnd_add_data(NULL, eh, ehlen, 0);
682 		earlypkts++;
683 	}
684 
685 	/*
686 	 * Determine if the packet is within its size limits. For MPLS the
687 	 * header length is variable, so we skip the check.
688 	 */
689 	if (etype != ETHERTYPE_MPLS && m->m_pkthdr.len >
690 	    ETHER_MAX_FRAME(ifp, etype, m->m_flags & M_HASFCS)) {
691 #ifdef DIAGNOSTIC
692 		mutex_enter(&bigpktpps_lock);
693 		if (ppsratecheck(&bigpktppslim_last, &bigpktpps_count,
694 		    bigpktppslim)) {
695 			printf("%s: discarding oversize frame (len=%d)\n",
696 			    ifp->if_xname, m->m_pkthdr.len);
697 		}
698 		mutex_exit(&bigpktpps_lock);
699 #endif
700 		goto error;
701 	}
702 
703 	if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
704 		/*
705 		 * If this is not a simplex interface, drop the packet
706 		 * if it came from us.
707 		 */
708 		if ((ifp->if_flags & IFF_SIMPLEX) == 0 &&
709 		    memcmp(CLLADDR(ifp->if_sadl), eh->ether_shost,
710 		    ETHER_ADDR_LEN) == 0) {
711 			goto drop;
712 		}
713 
714 		if (memcmp(etherbroadcastaddr,
715 		    eh->ether_dhost, ETHER_ADDR_LEN) == 0)
716 			m->m_flags |= M_BCAST;
717 		else
718 			m->m_flags |= M_MCAST;
719 		if_statinc(ifp, if_imcasts);
720 	}
721 
722 	/* If the CRC is still on the packet, trim it off. */
723 	if (m->m_flags & M_HASFCS) {
724 		m_adj(m, -ETHER_CRC_LEN);
725 		m->m_flags &= ~M_HASFCS;
726 	}
727 
728 	if_statadd(ifp, if_ibytes, m->m_pkthdr.len);
729 
730 	if (!vlan_has_tag(m) && etype == ETHERTYPE_VLAN) {
731 		m = ether_strip_vlantag(m);
732 		if (m == NULL) {
733 			if_statinc(ifp, if_ierrors);
734 			return;
735 		}
736 
737 		eh = mtod(m, struct ether_header *);
738 		etype = ntohs(eh->ether_type);
739 		ehlen = sizeof(*eh);
740 	}
741 
742 	if ((m->m_flags & (M_BCAST | M_MCAST | M_PROMISC)) == 0 &&
743 	    (ifp->if_flags & IFF_PROMISC) != 0 &&
744 	    memcmp(CLLADDR(ifp->if_sadl), eh->ether_dhost,
745 	     ETHER_ADDR_LEN) != 0) {
746 		m->m_flags |= M_PROMISC;
747 	}
748 
749 	if ((m->m_flags & M_PROMISC) == 0) {
750 		if (pfil_run_hooks(ifp->if_pfil, &m, ifp, PFIL_IN) != 0)
751 			return;
752 		if (m == NULL)
753 			return;
754 
755 		eh = mtod(m, struct ether_header *);
756 		etype = ntohs(eh->ether_type);
757 	}
758 
759 	/*
760 	 * Processing a logical interfaces that are able
761 	 * to configure vlan(4).
762 	*/
763 #if NAGR > 0
764 	if (ifp->if_lagg != NULL &&
765 	    __predict_true(etype != ETHERTYPE_SLOWPROTOCOLS)) {
766 		m->m_flags &= ~M_PROMISC;
767 		agr_input(ifp, m);
768 		return;
769 	}
770 #endif
771 
772 	/*
773 	 * VLAN processing.
774 	 *
775 	 * VLAN provides service delimiting so the frames are
776 	 * processed before other handlings. If a VLAN interface
777 	 * does not exist to take those frames, they're returned
778 	 * to ether_input().
779 	 */
780 
781 	if (vlan_has_tag(m)) {
782 		if (EVL_VLANOFTAG(vlan_get_tag(m)) == 0) {
783 			if (etype == ETHERTYPE_VLAN ||
784 			     etype == ETHERTYPE_QINQ)
785 				goto drop;
786 
787 			/* XXX we should actually use the prio value? */
788 			m->m_flags &= ~M_VLANTAG;
789 		} else {
790 #if NVLAN > 0
791 			if (ec->ec_nvlans > 0) {
792 				m = vlan_input(ifp, m);
793 
794 				/* vlan_input() called ether_input() recursively */
795 				if (m == NULL)
796 					return;
797 			}
798 #endif
799 			/* drop VLAN frames not for this port. */
800 			goto noproto;
801 		}
802 	}
803 
804 #if NCARP > 0
805 	if (__predict_false(ifp->if_carp && ifp->if_type != IFT_CARP)) {
806 		/*
807 		 * Clear M_PROMISC, in case the packet comes from a
808 		 * vlan.
809 		 */
810 		m->m_flags &= ~M_PROMISC;
811 		if (carp_input(m, (uint8_t *)&eh->ether_shost,
812 		    (uint8_t *)&eh->ether_dhost, eh->ether_type) == 0)
813 			return;
814 	}
815 #endif
816 
817 	/*
818 	 * Handle protocols that expect to have the Ethernet header
819 	 * (and possibly FCS) intact.
820 	 */
821 	switch (etype) {
822 #if NPPPOE > 0
823 	case ETHERTYPE_PPPOEDISC:
824 		pppoedisc_input(ifp, m);
825 		return;
826 
827 	case ETHERTYPE_PPPOE:
828 		pppoe_input(ifp, m);
829 		return;
830 #endif
831 
832 	case ETHERTYPE_SLOWPROTOCOLS: {
833 		uint8_t subtype;
834 
835 		if (m->m_pkthdr.len < sizeof(*eh) + sizeof(subtype))
836 			goto error;
837 
838 		m_copydata(m, sizeof(*eh), sizeof(subtype), &subtype);
839 		switch (subtype) {
840 #if NAGR > 0
841 		case SLOWPROTOCOLS_SUBTYPE_LACP:
842 			if (ifp->if_lagg != NULL) {
843 				ieee8023ad_lacp_input(ifp, m);
844 				return;
845 			}
846 			break;
847 
848 		case SLOWPROTOCOLS_SUBTYPE_MARKER:
849 			if (ifp->if_lagg != NULL) {
850 				ieee8023ad_marker_input(ifp, m);
851 				return;
852 			}
853 			break;
854 #endif
855 
856 		default:
857 			if (subtype == 0 || subtype > 10) {
858 				/* illegal value */
859 				goto error;
860 			}
861 			/* unknown subtype */
862 			break;
863 		}
864 	}
865 	/* FALLTHROUGH */
866 	default:
867 		if (m->m_flags & M_PROMISC)
868 			goto drop;
869 	}
870 
871 	/* If the CRC is still on the packet, trim it off. */
872 	if (m->m_flags & M_HASFCS) {
873 		m_adj(m, -ETHER_CRC_LEN);
874 		m->m_flags &= ~M_HASFCS;
875 	}
876 
877 	/* etype represents the size of the payload in this case */
878 	if (etype <= ETHERMTU + sizeof(struct ether_header)) {
879 		KASSERT(ehlen == sizeof(*eh));
880 #if defined (LLC) || defined (NETATALK)
881 		ether_input_llc(ifp, m, eh);
882 		return;
883 #else
884 		/* ethertype of 0-1500 is regarded as noproto */
885 		goto noproto;
886 #endif
887 	}
888 
889 	/* For ARP packets, store the source address so that
890 	 * ARP DAD probes can be validated. */
891 	if (etype == ETHERTYPE_ARP) {
892 		struct m_tag *mtag;
893 
894 		mtag = m_tag_get(PACKET_TAG_ETHERNET_SRC, ETHER_ADDR_LEN,
895 		    M_NOWAIT);
896 		if (mtag != NULL) {
897 			memcpy(mtag + 1, &eh->ether_shost, ETHER_ADDR_LEN);
898 			m_tag_prepend(m, mtag);
899 		}
900 	}
901 
902 	/* Strip off the Ethernet header. */
903 	m_adj(m, ehlen);
904 
905 	switch (etype) {
906 #ifdef INET
907 	case ETHERTYPE_IP:
908 #ifdef GATEWAY
909 		if (ipflow_fastforward(m))
910 			return;
911 #endif
912 		pktq = ip_pktq;
913 		rps_hash = atomic_load_relaxed(&ether_pktq_rps_hash_p);
914 		break;
915 
916 	case ETHERTYPE_ARP:
917 		pktq = arp_pktq;
918 		break;
919 
920 	case ETHERTYPE_REVARP:
921 		revarpinput(m);	/* XXX queue? */
922 		return;
923 #endif
924 
925 #ifdef INET6
926 	case ETHERTYPE_IPV6:
927 		if (__predict_false(!in6_present))
928 			goto noproto;
929 #ifdef GATEWAY
930 		if (ip6flow_fastforward(&m))
931 			return;
932 #endif
933 		pktq = ip6_pktq;
934 		rps_hash = atomic_load_relaxed(&ether_pktq_rps_hash_p);
935 		break;
936 #endif
937 
938 #ifdef NETATALK
939 	case ETHERTYPE_ATALK:
940 		pktq = at_pktq1;
941 		break;
942 
943 	case ETHERTYPE_AARP:
944 		aarpinput(ifp, m); /* XXX queue? */
945 		return;
946 #endif
947 
948 #ifdef MPLS
949 	case ETHERTYPE_MPLS:
950 		pktq = mpls_pktq;
951 		break;
952 #endif
953 
954 	default:
955 		goto noproto;
956 	}
957 
958 	KASSERT(pktq != NULL);
959 	const uint32_t h = rps_hash ? pktq_rps_hash(&rps_hash, m) : 0;
960 	if (__predict_false(!pktq_enqueue(pktq, m, h))) {
961 		m_freem(m);
962 	}
963 	return;
964 
965 drop:
966 	m_freem(m);
967 	if_statinc(ifp, if_iqdrops);
968 	return;
969 noproto:
970 	m_freem(m);
971 	if_statinc(ifp, if_noproto);
972 	return;
973 error:
974 	m_freem(m);
975 	if_statinc(ifp, if_ierrors);
976 	return;
977 }
978 
979 static void
ether_bpf_mtap(struct bpf_if * bp,struct mbuf * m,u_int direction)980 ether_bpf_mtap(struct bpf_if *bp, struct mbuf *m, u_int direction)
981 {
982 	struct ether_vlan_header evl;
983 	struct m_hdr mh, md;
984 
985 	KASSERT(bp != NULL);
986 
987 	if (!vlan_has_tag(m)) {
988 		bpf_mtap3(bp, m, direction);
989 		return;
990 	}
991 
992 	memcpy(&evl, mtod(m, char *), ETHER_HDR_LEN);
993 	evl.evl_proto = evl.evl_encap_proto;
994 	evl.evl_encap_proto = htons(ETHERTYPE_VLAN);
995 	evl.evl_tag = htons(vlan_get_tag(m));
996 
997 	md.mh_flags = 0;
998 	md.mh_data = m->m_data + ETHER_HDR_LEN;
999 	md.mh_len = m->m_len - ETHER_HDR_LEN;
1000 	md.mh_next = m->m_next;
1001 
1002 	mh.mh_flags = 0;
1003 	mh.mh_data = (char *)&evl;
1004 	mh.mh_len = sizeof(evl);
1005 	mh.mh_next = (struct mbuf *)&md;
1006 
1007 	bpf_mtap3(bp, (struct mbuf *)&mh, direction);
1008 }
1009 
1010 /*
1011  * Convert Ethernet address to printable (loggable) representation.
1012  */
1013 char *
ether_sprintf(const u_char * ap)1014 ether_sprintf(const u_char *ap)
1015 {
1016 	static char etherbuf[3 * ETHER_ADDR_LEN];
1017 	return ether_snprintf(etherbuf, sizeof(etherbuf), ap);
1018 }
1019 
1020 char *
ether_snprintf(char * buf,size_t len,const u_char * ap)1021 ether_snprintf(char *buf, size_t len, const u_char *ap)
1022 {
1023 	char *cp = buf;
1024 	size_t i;
1025 
1026 	for (i = 0; i < len / 3; i++) {
1027 		*cp++ = hexdigits[*ap >> 4];
1028 		*cp++ = hexdigits[*ap++ & 0xf];
1029 		*cp++ = ':';
1030 	}
1031 	*--cp = '\0';
1032 	return buf;
1033 }
1034 
1035 /*
1036  * Perform common duties while attaching to interface list
1037  */
1038 void
ether_ifattach(struct ifnet * ifp,const uint8_t * lla)1039 ether_ifattach(struct ifnet *ifp, const uint8_t *lla)
1040 {
1041 	struct ethercom *ec = (struct ethercom *)ifp;
1042 	char xnamebuf[HOOKNAMSIZ];
1043 
1044 	ifp->if_type = IFT_ETHER;
1045 	ifp->if_hdrlen = ETHER_HDR_LEN;
1046 	ifp->if_dlt = DLT_EN10MB;
1047 	ifp->if_mtu = ETHERMTU;
1048 	ifp->if_output = ether_output;
1049 	ifp->_if_input = ether_input;
1050 	ifp->if_bpf_mtap = ether_bpf_mtap;
1051 	if (ifp->if_baudrate == 0)
1052 		ifp->if_baudrate = IF_Mbps(10);		/* just a default */
1053 
1054 	if (lla != NULL)
1055 		if_set_sadl(ifp, lla, ETHER_ADDR_LEN, !ETHER_IS_LOCAL(lla));
1056 
1057 	LIST_INIT(&ec->ec_multiaddrs);
1058 	SIMPLEQ_INIT(&ec->ec_vids);
1059 	ec->ec_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NET);
1060 	ec->ec_flags = 0;
1061 	ifp->if_broadcastaddr = etherbroadcastaddr;
1062 	bpf_attach(ifp, DLT_EN10MB, sizeof(struct ether_header));
1063 	snprintf(xnamebuf, sizeof(xnamebuf),
1064 	    "%s-ether_ifdetachhooks", ifp->if_xname);
1065 	ec->ec_ifdetach_hooks = simplehook_create(IPL_NET, xnamebuf);
1066 #ifdef MBUFTRACE
1067 	mowner_init_owner(&ec->ec_tx_mowner, ifp->if_xname, "tx");
1068 	mowner_init_owner(&ec->ec_rx_mowner, ifp->if_xname, "rx");
1069 	MOWNER_ATTACH(&ec->ec_tx_mowner);
1070 	MOWNER_ATTACH(&ec->ec_rx_mowner);
1071 	ifp->if_mowner = &ec->ec_tx_mowner;
1072 #endif
1073 }
1074 
1075 void
ether_ifdetach(struct ifnet * ifp)1076 ether_ifdetach(struct ifnet *ifp)
1077 {
1078 	struct ethercom *ec = (void *) ifp;
1079 	struct ether_multi *enm;
1080 
1081 	IFNET_ASSERT_UNLOCKED(ifp);
1082 	/*
1083 	 * Prevent further calls to ioctl (for example turning off
1084 	 * promiscuous mode from the bridge code), which eventually can
1085 	 * call if_init() which can cause panics because the interface
1086 	 * is in the process of being detached. Return device not configured
1087 	 * instead.
1088 	 */
1089 	ifp->if_ioctl = __FPTRCAST(int (*)(struct ifnet *, u_long, void *),
1090 	    enxio);
1091 
1092 	simplehook_dohooks(ec->ec_ifdetach_hooks);
1093 	KASSERT(!simplehook_has_hooks(ec->ec_ifdetach_hooks));
1094 	simplehook_destroy(ec->ec_ifdetach_hooks);
1095 
1096 	bpf_detach(ifp);
1097 
1098 	ETHER_LOCK(ec);
1099 	KASSERT(ec->ec_nvlans == 0);
1100 	while ((enm = LIST_FIRST(&ec->ec_multiaddrs)) != NULL) {
1101 		LIST_REMOVE(enm, enm_list);
1102 		kmem_free(enm, sizeof(*enm));
1103 		ec->ec_multicnt--;
1104 	}
1105 	ETHER_UNLOCK(ec);
1106 
1107 	mutex_obj_free(ec->ec_lock);
1108 	ec->ec_lock = NULL;
1109 
1110 	ifp->if_mowner = NULL;
1111 	MOWNER_DETACH(&ec->ec_rx_mowner);
1112 	MOWNER_DETACH(&ec->ec_tx_mowner);
1113 }
1114 
1115 void *
ether_ifdetachhook_establish(struct ifnet * ifp,void (* fn)(void *),void * arg)1116 ether_ifdetachhook_establish(struct ifnet *ifp,
1117     void (*fn)(void *), void *arg)
1118 {
1119 	struct ethercom *ec;
1120 	khook_t *hk;
1121 
1122 	if (ifp->if_type != IFT_ETHER)
1123 		return NULL;
1124 
1125 	ec = (struct ethercom *)ifp;
1126 	hk = simplehook_establish(ec->ec_ifdetach_hooks,
1127 	    fn, arg);
1128 
1129 	return (void *)hk;
1130 }
1131 
1132 void
ether_ifdetachhook_disestablish(struct ifnet * ifp,void * vhook,kmutex_t * lock)1133 ether_ifdetachhook_disestablish(struct ifnet *ifp,
1134     void *vhook, kmutex_t *lock)
1135 {
1136 	struct ethercom *ec;
1137 
1138 	if (vhook == NULL)
1139 		return;
1140 
1141 	ec = (struct ethercom *)ifp;
1142 	simplehook_disestablish(ec->ec_ifdetach_hooks, vhook, lock);
1143 }
1144 
1145 #if 0
1146 /*
1147  * This is for reference.  We have a table-driven version
1148  * of the little-endian crc32 generator, which is faster
1149  * than the double-loop.
1150  */
1151 uint32_t
1152 ether_crc32_le(const uint8_t *buf, size_t len)
1153 {
1154 	uint32_t c, crc, carry;
1155 	size_t i, j;
1156 
1157 	crc = 0xffffffffU;	/* initial value */
1158 
1159 	for (i = 0; i < len; i++) {
1160 		c = buf[i];
1161 		for (j = 0; j < 8; j++) {
1162 			carry = ((crc & 0x01) ? 1 : 0) ^ (c & 0x01);
1163 			crc >>= 1;
1164 			c >>= 1;
1165 			if (carry)
1166 				crc = (crc ^ ETHER_CRC_POLY_LE);
1167 		}
1168 	}
1169 
1170 	return (crc);
1171 }
1172 #else
1173 uint32_t
ether_crc32_le(const uint8_t * buf,size_t len)1174 ether_crc32_le(const uint8_t *buf, size_t len)
1175 {
1176 	static const uint32_t crctab[] = {
1177 		0x00000000, 0x1db71064, 0x3b6e20c8, 0x26d930ac,
1178 		0x76dc4190, 0x6b6b51f4, 0x4db26158, 0x5005713c,
1179 		0xedb88320, 0xf00f9344, 0xd6d6a3e8, 0xcb61b38c,
1180 		0x9b64c2b0, 0x86d3d2d4, 0xa00ae278, 0xbdbdf21c
1181 	};
1182 	uint32_t crc;
1183 	size_t i;
1184 
1185 	crc = 0xffffffffU;	/* initial value */
1186 
1187 	for (i = 0; i < len; i++) {
1188 		crc ^= buf[i];
1189 		crc = (crc >> 4) ^ crctab[crc & 0xf];
1190 		crc = (crc >> 4) ^ crctab[crc & 0xf];
1191 	}
1192 
1193 	return (crc);
1194 }
1195 #endif
1196 
1197 uint32_t
ether_crc32_be(const uint8_t * buf,size_t len)1198 ether_crc32_be(const uint8_t *buf, size_t len)
1199 {
1200 	uint32_t c, crc, carry;
1201 	size_t i, j;
1202 
1203 	crc = 0xffffffffU;	/* initial value */
1204 
1205 	for (i = 0; i < len; i++) {
1206 		c = buf[i];
1207 		for (j = 0; j < 8; j++) {
1208 			carry = ((crc & 0x80000000U) ? 1 : 0) ^ (c & 0x01);
1209 			crc <<= 1;
1210 			c >>= 1;
1211 			if (carry)
1212 				crc = (crc ^ ETHER_CRC_POLY_BE) | carry;
1213 		}
1214 	}
1215 
1216 	return (crc);
1217 }
1218 
1219 #ifdef INET
1220 const uint8_t ether_ipmulticast_min[ETHER_ADDR_LEN] =
1221     { 0x01, 0x00, 0x5e, 0x00, 0x00, 0x00 };
1222 const uint8_t ether_ipmulticast_max[ETHER_ADDR_LEN] =
1223     { 0x01, 0x00, 0x5e, 0x7f, 0xff, 0xff };
1224 #endif
1225 #ifdef INET6
1226 const uint8_t ether_ip6multicast_min[ETHER_ADDR_LEN] =
1227     { 0x33, 0x33, 0x00, 0x00, 0x00, 0x00 };
1228 const uint8_t ether_ip6multicast_max[ETHER_ADDR_LEN] =
1229     { 0x33, 0x33, 0xff, 0xff, 0xff, 0xff };
1230 #endif
1231 
1232 /*
1233  * ether_aton implementation, not using a static buffer.
1234  */
1235 int
ether_aton_r(u_char * dest,size_t len,const char * str)1236 ether_aton_r(u_char *dest, size_t len, const char *str)
1237 {
1238 	const u_char *cp = (const void *)str;
1239 	u_char *ep;
1240 
1241 #define atox(c)	(((c) <= '9') ? ((c) - '0') : ((toupper(c) - 'A') + 10))
1242 
1243 	if (len < ETHER_ADDR_LEN)
1244 		return ENOSPC;
1245 
1246 	ep = dest + ETHER_ADDR_LEN;
1247 
1248 	while (*cp) {
1249 		if (!isxdigit(*cp))
1250 			return EINVAL;
1251 
1252 		*dest = atox(*cp);
1253 		cp++;
1254 		if (isxdigit(*cp)) {
1255 			*dest = (*dest << 4) | atox(*cp);
1256 			cp++;
1257 		}
1258 		dest++;
1259 
1260 		if (dest == ep)
1261 			return (*cp == '\0') ? 0 : ENAMETOOLONG;
1262 
1263 		switch (*cp) {
1264 		case ':':
1265 		case '-':
1266 		case '.':
1267 			cp++;
1268 			break;
1269 		}
1270 	}
1271 	return ENOBUFS;
1272 }
1273 
1274 /*
1275  * Convert a sockaddr into an Ethernet address or range of Ethernet
1276  * addresses.
1277  */
1278 int
ether_multiaddr(const struct sockaddr * sa,uint8_t addrlo[ETHER_ADDR_LEN],uint8_t addrhi[ETHER_ADDR_LEN])1279 ether_multiaddr(const struct sockaddr *sa, uint8_t addrlo[ETHER_ADDR_LEN],
1280     uint8_t addrhi[ETHER_ADDR_LEN])
1281 {
1282 #ifdef INET
1283 	const struct sockaddr_in *sin;
1284 #endif
1285 #ifdef INET6
1286 	const struct sockaddr_in6 *sin6;
1287 #endif
1288 
1289 	switch (sa->sa_family) {
1290 
1291 	case AF_UNSPEC:
1292 		memcpy(addrlo, sa->sa_data, ETHER_ADDR_LEN);
1293 		memcpy(addrhi, addrlo, ETHER_ADDR_LEN);
1294 		break;
1295 
1296 #ifdef INET
1297 	case AF_INET:
1298 		sin = satocsin(sa);
1299 		if (sin->sin_addr.s_addr == INADDR_ANY) {
1300 			/*
1301 			 * An IP address of INADDR_ANY means listen to
1302 			 * or stop listening to all of the Ethernet
1303 			 * multicast addresses used for IP.
1304 			 * (This is for the sake of IP multicast routers.)
1305 			 */
1306 			memcpy(addrlo, ether_ipmulticast_min, ETHER_ADDR_LEN);
1307 			memcpy(addrhi, ether_ipmulticast_max, ETHER_ADDR_LEN);
1308 		} else {
1309 			ETHER_MAP_IP_MULTICAST(&sin->sin_addr, addrlo);
1310 			memcpy(addrhi, addrlo, ETHER_ADDR_LEN);
1311 		}
1312 		break;
1313 #endif
1314 #ifdef INET6
1315 	case AF_INET6:
1316 		sin6 = satocsin6(sa);
1317 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
1318 			/*
1319 			 * An IP6 address of 0 means listen to or stop
1320 			 * listening to all of the Ethernet multicast
1321 			 * address used for IP6.
1322 			 * (This is used for multicast routers.)
1323 			 */
1324 			memcpy(addrlo, ether_ip6multicast_min, ETHER_ADDR_LEN);
1325 			memcpy(addrhi, ether_ip6multicast_max, ETHER_ADDR_LEN);
1326 		} else {
1327 			ETHER_MAP_IPV6_MULTICAST(&sin6->sin6_addr, addrlo);
1328 			memcpy(addrhi, addrlo, ETHER_ADDR_LEN);
1329 		}
1330 		break;
1331 #endif
1332 
1333 	default:
1334 		return EAFNOSUPPORT;
1335 	}
1336 	return 0;
1337 }
1338 
1339 /*
1340  * Add an Ethernet multicast address or range of addresses to the list for a
1341  * given interface.
1342  */
1343 int
ether_addmulti(const struct sockaddr * sa,struct ethercom * ec)1344 ether_addmulti(const struct sockaddr *sa, struct ethercom *ec)
1345 {
1346 	struct ether_multi *enm, *_enm;
1347 	u_char addrlo[ETHER_ADDR_LEN];
1348 	u_char addrhi[ETHER_ADDR_LEN];
1349 	int error = 0;
1350 
1351 	/* Allocate out of lock */
1352 	enm = kmem_alloc(sizeof(*enm), KM_SLEEP);
1353 
1354 	ETHER_LOCK(ec);
1355 	error = ether_multiaddr(sa, addrlo, addrhi);
1356 	if (error != 0)
1357 		goto out;
1358 
1359 	/*
1360 	 * Verify that we have valid Ethernet multicast addresses.
1361 	 */
1362 	if (!ETHER_IS_MULTICAST(addrlo) || !ETHER_IS_MULTICAST(addrhi)) {
1363 		error = EINVAL;
1364 		goto out;
1365 	}
1366 
1367 	/*
1368 	 * See if the address range is already in the list.
1369 	 */
1370 	_enm = ether_lookup_multi(addrlo, addrhi, ec);
1371 	if (_enm != NULL) {
1372 		/*
1373 		 * Found it; just increment the reference count.
1374 		 */
1375 		++_enm->enm_refcount;
1376 		error = 0;
1377 		goto out;
1378 	}
1379 
1380 	/*
1381 	 * Link a new multicast record into the interface's multicast list.
1382 	 */
1383 	memcpy(enm->enm_addrlo, addrlo, ETHER_ADDR_LEN);
1384 	memcpy(enm->enm_addrhi, addrhi, ETHER_ADDR_LEN);
1385 	enm->enm_refcount = 1;
1386 	LIST_INSERT_HEAD(&ec->ec_multiaddrs, enm, enm_list);
1387 	ec->ec_multicnt++;
1388 
1389 	/*
1390 	 * Return ENETRESET to inform the driver that the list has changed
1391 	 * and its reception filter should be adjusted accordingly.
1392 	 */
1393 	error = ENETRESET;
1394 	enm = NULL;
1395 
1396 out:
1397 	ETHER_UNLOCK(ec);
1398 	if (enm != NULL)
1399 		kmem_free(enm, sizeof(*enm));
1400 	return error;
1401 }
1402 
1403 /*
1404  * Delete a multicast address record.
1405  */
1406 int
ether_delmulti(const struct sockaddr * sa,struct ethercom * ec)1407 ether_delmulti(const struct sockaddr *sa, struct ethercom *ec)
1408 {
1409 	struct ether_multi *enm;
1410 	u_char addrlo[ETHER_ADDR_LEN];
1411 	u_char addrhi[ETHER_ADDR_LEN];
1412 	int error;
1413 
1414 	ETHER_LOCK(ec);
1415 	error = ether_multiaddr(sa, addrlo, addrhi);
1416 	if (error != 0)
1417 		goto error;
1418 
1419 	/*
1420 	 * Look up the address in our list.
1421 	 */
1422 	enm = ether_lookup_multi(addrlo, addrhi, ec);
1423 	if (enm == NULL) {
1424 		error = ENXIO;
1425 		goto error;
1426 	}
1427 	if (--enm->enm_refcount != 0) {
1428 		/*
1429 		 * Still some claims to this record.
1430 		 */
1431 		error = 0;
1432 		goto error;
1433 	}
1434 
1435 	/*
1436 	 * No remaining claims to this record; unlink and free it.
1437 	 */
1438 	LIST_REMOVE(enm, enm_list);
1439 	ec->ec_multicnt--;
1440 	ETHER_UNLOCK(ec);
1441 	kmem_free(enm, sizeof(*enm));
1442 
1443 	/*
1444 	 * Return ENETRESET to inform the driver that the list has changed
1445 	 * and its reception filter should be adjusted accordingly.
1446 	 */
1447 	return ENETRESET;
1448 
1449 error:
1450 	ETHER_UNLOCK(ec);
1451 	return error;
1452 }
1453 
1454 void
ether_set_ifflags_cb(struct ethercom * ec,ether_cb_t cb)1455 ether_set_ifflags_cb(struct ethercom *ec, ether_cb_t cb)
1456 {
1457 	ec->ec_ifflags_cb = cb;
1458 }
1459 
1460 void
ether_set_vlan_cb(struct ethercom * ec,ether_vlancb_t cb)1461 ether_set_vlan_cb(struct ethercom *ec, ether_vlancb_t cb)
1462 {
1463 
1464 	ec->ec_vlan_cb = cb;
1465 }
1466 
1467 static int
ether_ioctl_reinit(struct ethercom * ec)1468 ether_ioctl_reinit(struct ethercom *ec)
1469 {
1470 	struct ifnet *ifp = &ec->ec_if;
1471 	int error;
1472 
1473 	KASSERTMSG(IFNET_LOCKED(ifp), "%s", ifp->if_xname);
1474 
1475 	switch (ifp->if_flags & (IFF_UP | IFF_RUNNING)) {
1476 	case IFF_RUNNING:
1477 		/*
1478 		 * If interface is marked down and it is running,
1479 		 * then stop and disable it.
1480 		 */
1481 		if_stop(ifp, 1);
1482 		break;
1483 	case IFF_UP:
1484 		/*
1485 		 * If interface is marked up and it is stopped, then
1486 		 * start it.
1487 		 */
1488 		return if_init(ifp);
1489 	case IFF_UP | IFF_RUNNING:
1490 		error = 0;
1491 		if (ec->ec_ifflags_cb != NULL) {
1492 			error = (*ec->ec_ifflags_cb)(ec);
1493 			if (error == ENETRESET) {
1494 				/*
1495 				 * Reset the interface to pick up
1496 				 * changes in any other flags that
1497 				 * affect the hardware state.
1498 				 */
1499 				return if_init(ifp);
1500 			}
1501 		} else
1502 			error = if_init(ifp);
1503 		return error;
1504 	case 0:
1505 		break;
1506 	}
1507 
1508 	return 0;
1509 }
1510 
1511 /*
1512  * Common ioctls for Ethernet interfaces.  Note, we must be
1513  * called at splnet().
1514  */
1515 int
ether_ioctl(struct ifnet * ifp,u_long cmd,void * data)1516 ether_ioctl(struct ifnet *ifp, u_long cmd, void *data)
1517 {
1518 	struct ethercom *ec = (void *)ifp;
1519 	struct eccapreq *eccr;
1520 	struct ifreq *ifr = (struct ifreq *)data;
1521 	struct if_laddrreq *iflr = data;
1522 	const struct sockaddr_dl *sdl;
1523 	static const uint8_t zero[ETHER_ADDR_LEN];
1524 	int error;
1525 
1526 	switch (cmd) {
1527 	case SIOCINITIFADDR:
1528 	    {
1529 		struct ifaddr *ifa = (struct ifaddr *)data;
1530 		if (ifa->ifa_addr->sa_family != AF_LINK
1531 		    && (ifp->if_flags & (IFF_UP | IFF_RUNNING)) !=
1532 		       (IFF_UP | IFF_RUNNING)) {
1533 			ifp->if_flags |= IFF_UP;
1534 			if ((error = if_init(ifp)) != 0)
1535 				return error;
1536 		}
1537 #ifdef INET
1538 		if (ifa->ifa_addr->sa_family == AF_INET)
1539 			arp_ifinit(ifp, ifa);
1540 #endif
1541 		return 0;
1542 	    }
1543 
1544 	case SIOCSIFMTU:
1545 	    {
1546 		int maxmtu;
1547 
1548 		if (ec->ec_capabilities & ETHERCAP_JUMBO_MTU)
1549 			maxmtu = ETHERMTU_JUMBO;
1550 		else
1551 			maxmtu = ETHERMTU;
1552 
1553 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > maxmtu)
1554 			return EINVAL;
1555 		else if ((error = ifioctl_common(ifp, cmd, data)) != ENETRESET)
1556 			return error;
1557 		else if (ifp->if_flags & IFF_UP) {
1558 			/* Make sure the device notices the MTU change. */
1559 			return if_init(ifp);
1560 		} else
1561 			return 0;
1562 	    }
1563 
1564 	case SIOCSIFFLAGS:
1565 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
1566 			return error;
1567 		return ether_ioctl_reinit(ec);
1568 	case SIOCGIFFLAGS:
1569 		error = ifioctl_common(ifp, cmd, data);
1570 		if (error == 0) {
1571 			/* Set IFF_ALLMULTI for backcompat */
1572 			ifr->ifr_flags |= (ec->ec_flags & ETHER_F_ALLMULTI) ?
1573 			    IFF_ALLMULTI : 0;
1574 		}
1575 		return error;
1576 	case SIOCGETHERCAP:
1577 		eccr = (struct eccapreq *)data;
1578 		eccr->eccr_capabilities = ec->ec_capabilities;
1579 		eccr->eccr_capenable = ec->ec_capenable;
1580 		return 0;
1581 	case SIOCSETHERCAP:
1582 		eccr = (struct eccapreq *)data;
1583 		if ((eccr->eccr_capenable & ~ec->ec_capabilities) != 0)
1584 			return EINVAL;
1585 		if (eccr->eccr_capenable == ec->ec_capenable)
1586 			return 0;
1587 #if 0 /* notyet */
1588 		ec->ec_capenable = (ec->ec_capenable & ETHERCAP_CANTCHANGE)
1589 		    | (eccr->eccr_capenable & ~ETHERCAP_CANTCHANGE);
1590 #else
1591 		ec->ec_capenable = eccr->eccr_capenable;
1592 #endif
1593 		return ether_ioctl_reinit(ec);
1594 	case SIOCADDMULTI:
1595 		return ether_addmulti(ifreq_getaddr(cmd, ifr), ec);
1596 	case SIOCDELMULTI:
1597 		return ether_delmulti(ifreq_getaddr(cmd, ifr), ec);
1598 	case SIOCSIFMEDIA:
1599 	case SIOCGIFMEDIA:
1600 		if (ec->ec_mii != NULL)
1601 			return ifmedia_ioctl(ifp, ifr, &ec->ec_mii->mii_media,
1602 			    cmd);
1603 		else if (ec->ec_ifmedia != NULL)
1604 			return ifmedia_ioctl(ifp, ifr, ec->ec_ifmedia, cmd);
1605 		else
1606 			return ENOTTY;
1607 		break;
1608 	case SIOCALIFADDR:
1609 		sdl = satocsdl(sstocsa(&iflr->addr));
1610 		if (sdl->sdl_family != AF_LINK)
1611 			;
1612 		else if (ETHER_IS_MULTICAST(CLLADDR(sdl)))
1613 			return EINVAL;
1614 		else if (memcmp(zero, CLLADDR(sdl), sizeof(zero)) == 0)
1615 			return EINVAL;
1616 		/*FALLTHROUGH*/
1617 	default:
1618 		return ifioctl_common(ifp, cmd, data);
1619 	}
1620 	return 0;
1621 }
1622 
1623 /*
1624  * Enable/disable passing VLAN packets if the parent interface supports it.
1625  * Return:
1626  * 	 0: Ok
1627  *	-1: Parent interface does not support vlans
1628  *	>0: Error
1629  */
1630 int
ether_enable_vlan_mtu(struct ifnet * ifp)1631 ether_enable_vlan_mtu(struct ifnet *ifp)
1632 {
1633 	int error;
1634 	struct ethercom *ec = (void *)ifp;
1635 
1636 	/* Parent does not support VLAN's */
1637 	if ((ec->ec_capabilities & ETHERCAP_VLAN_MTU) == 0)
1638 		return -1;
1639 
1640 	/*
1641 	 * Parent supports the VLAN_MTU capability,
1642 	 * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames;
1643 	 * enable it.
1644 	 */
1645 	ec->ec_capenable |= ETHERCAP_VLAN_MTU;
1646 
1647 	/* Interface is down, defer for later */
1648 	if ((ifp->if_flags & IFF_UP) == 0)
1649 		return 0;
1650 
1651 	if ((error = if_flags_set(ifp, ifp->if_flags)) == 0)
1652 		return 0;
1653 
1654 	ec->ec_capenable &= ~ETHERCAP_VLAN_MTU;
1655 	return error;
1656 }
1657 
1658 int
ether_disable_vlan_mtu(struct ifnet * ifp)1659 ether_disable_vlan_mtu(struct ifnet *ifp)
1660 {
1661 	int error;
1662 	struct ethercom *ec = (void *)ifp;
1663 
1664 	/* We still have VLAN's, defer for later */
1665 	if (ec->ec_nvlans != 0)
1666 		return 0;
1667 
1668 	/* Parent does not support VLAB's, nothing to do. */
1669 	if ((ec->ec_capenable & ETHERCAP_VLAN_MTU) == 0)
1670 		return -1;
1671 
1672 	/*
1673 	 * Disable Tx/Rx of VLAN-sized frames.
1674 	 */
1675 	ec->ec_capenable &= ~ETHERCAP_VLAN_MTU;
1676 
1677 	/* Interface is down, defer for later */
1678 	if ((ifp->if_flags & IFF_UP) == 0)
1679 		return 0;
1680 
1681 	if ((error = if_flags_set(ifp, ifp->if_flags)) == 0)
1682 		return 0;
1683 
1684 	ec->ec_capenable |= ETHERCAP_VLAN_MTU;
1685 	return error;
1686 }
1687 
1688 /*
1689  * Add and delete VLAN TAG
1690  */
1691 int
ether_add_vlantag(struct ifnet * ifp,uint16_t vtag,bool * vlanmtu_status)1692 ether_add_vlantag(struct ifnet *ifp, uint16_t vtag, bool *vlanmtu_status)
1693 {
1694 	struct ethercom *ec = (void *)ifp;
1695 	struct vlanid_list *vidp;
1696 	bool vlanmtu_enabled;
1697 	uint16_t vid = EVL_VLANOFTAG(vtag);
1698 	int error;
1699 
1700 	vlanmtu_enabled = false;
1701 
1702 	/* Add a vid to the list */
1703 	vidp = kmem_alloc(sizeof(*vidp), KM_SLEEP);
1704 	vidp->vid = vid;
1705 
1706 	ETHER_LOCK(ec);
1707 	ec->ec_nvlans++;
1708 	SIMPLEQ_INSERT_TAIL(&ec->ec_vids, vidp, vid_list);
1709 	ETHER_UNLOCK(ec);
1710 
1711 	if (ec->ec_nvlans == 1) {
1712 		IFNET_LOCK(ifp);
1713 		error = ether_enable_vlan_mtu(ifp);
1714 		IFNET_UNLOCK(ifp);
1715 
1716 		if (error == 0) {
1717 			vlanmtu_enabled = true;
1718 		} else if (error != -1) {
1719 			goto fail;
1720 		}
1721 	}
1722 
1723 	if (ec->ec_vlan_cb != NULL) {
1724 		error = (*ec->ec_vlan_cb)(ec, vid, true);
1725 		if (error != 0)
1726 			goto fail;
1727 	}
1728 
1729 	if (vlanmtu_status != NULL)
1730 		*vlanmtu_status = vlanmtu_enabled;
1731 
1732 	return 0;
1733 fail:
1734 	ETHER_LOCK(ec);
1735 	ec->ec_nvlans--;
1736 	SIMPLEQ_REMOVE(&ec->ec_vids, vidp, vlanid_list, vid_list);
1737 	ETHER_UNLOCK(ec);
1738 
1739 	if (vlanmtu_enabled) {
1740 		IFNET_LOCK(ifp);
1741 		(void)ether_disable_vlan_mtu(ifp);
1742 		IFNET_UNLOCK(ifp);
1743 	}
1744 
1745 	kmem_free(vidp, sizeof(*vidp));
1746 
1747 	return error;
1748 }
1749 
1750 int
ether_del_vlantag(struct ifnet * ifp,uint16_t vtag)1751 ether_del_vlantag(struct ifnet *ifp, uint16_t vtag)
1752 {
1753 	struct ethercom *ec = (void *)ifp;
1754 	struct vlanid_list *vidp;
1755 	uint16_t vid = EVL_VLANOFTAG(vtag);
1756 
1757 	ETHER_LOCK(ec);
1758 	SIMPLEQ_FOREACH(vidp, &ec->ec_vids, vid_list) {
1759 		if (vidp->vid == vid) {
1760 			SIMPLEQ_REMOVE(&ec->ec_vids, vidp,
1761 			    vlanid_list, vid_list);
1762 			ec->ec_nvlans--;
1763 			break;
1764 		}
1765 	}
1766 	ETHER_UNLOCK(ec);
1767 
1768 	if (vidp == NULL)
1769 		return ENOENT;
1770 
1771 	if (ec->ec_vlan_cb != NULL) {
1772 		(void)(*ec->ec_vlan_cb)(ec, vidp->vid, false);
1773 	}
1774 
1775 	if (ec->ec_nvlans == 0) {
1776 		IFNET_LOCK(ifp);
1777 		(void)ether_disable_vlan_mtu(ifp);
1778 		IFNET_UNLOCK(ifp);
1779 	}
1780 
1781 	kmem_free(vidp, sizeof(*vidp));
1782 
1783 	return 0;
1784 }
1785 
1786 int
ether_inject_vlantag(struct mbuf ** mp,uint16_t etype,uint16_t tag)1787 ether_inject_vlantag(struct mbuf **mp, uint16_t etype, uint16_t tag)
1788 {
1789 	static const size_t min_data_len =
1790 	    ETHER_MIN_LEN - ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN;
1791 	/* Used to pad ethernet frames with < ETHER_MIN_LEN bytes */
1792 	static const char vlan_zero_pad_buff[ETHER_MIN_LEN] = { 0 };
1793 
1794 	struct ether_vlan_header *evl;
1795 	struct mbuf *m = *mp;
1796 	int error;
1797 
1798 	error = 0;
1799 
1800 	M_PREPEND(m, ETHER_VLAN_ENCAP_LEN, M_DONTWAIT);
1801 	if (m == NULL) {
1802 		error = ENOBUFS;
1803 		goto out;
1804 	}
1805 
1806 	if (m->m_len < sizeof(*evl)) {
1807 		m = m_pullup(m, sizeof(*evl));
1808 		if (m == NULL) {
1809 			error = ENOBUFS;
1810 			goto out;
1811 		}
1812 	}
1813 
1814 	/*
1815 	 * Transform the Ethernet header into an
1816 	 * Ethernet header with 802.1Q encapsulation.
1817 	 */
1818 	memmove(mtod(m, void *),
1819 	    mtod(m, char *) + ETHER_VLAN_ENCAP_LEN,
1820 	    sizeof(struct ether_header));
1821 	evl = mtod(m, struct ether_vlan_header *);
1822 	evl->evl_proto = evl->evl_encap_proto;
1823 	evl->evl_encap_proto = htons(etype);
1824 	evl->evl_tag = htons(tag);
1825 
1826 	/*
1827 	 * To cater for VLAN-aware layer 2 ethernet
1828 	 * switches which may need to strip the tag
1829 	 * before forwarding the packet, make sure
1830 	 * the packet+tag is at least 68 bytes long.
1831 	 * This is necessary because our parent will
1832 	 * only pad to 64 bytes (ETHER_MIN_LEN) and
1833 	 * some switches will not pad by themselves
1834 	 * after deleting a tag.
1835 	 */
1836 	if (m->m_pkthdr.len < min_data_len) {
1837 		m_copyback(m, m->m_pkthdr.len,
1838 		    min_data_len - m->m_pkthdr.len,
1839 		    vlan_zero_pad_buff);
1840 	}
1841 
1842 	m->m_flags &= ~M_VLANTAG;
1843 
1844 out:
1845 	*mp = m;
1846 	return error;
1847 }
1848 
1849 struct mbuf *
ether_strip_vlantag(struct mbuf * m)1850 ether_strip_vlantag(struct mbuf *m)
1851 {
1852 	struct ether_vlan_header *evl;
1853 
1854 	if (m->m_len < sizeof(*evl) &&
1855 	    (m = m_pullup(m, sizeof(*evl))) == NULL) {
1856 		return NULL;
1857 	}
1858 
1859 	if (m_makewritable(&m, 0, sizeof(*evl), M_DONTWAIT)) {
1860 		m_freem(m);
1861 		return NULL;
1862 	}
1863 
1864 	evl = mtod(m, struct ether_vlan_header *);
1865 	KASSERT(ntohs(evl->evl_encap_proto) == ETHERTYPE_VLAN);
1866 
1867 	vlan_set_tag(m, ntohs(evl->evl_tag));
1868 
1869 	/*
1870 	 * Restore the original ethertype.  We'll remove
1871 	 * the encapsulation after we've found the vlan
1872 	 * interface corresponding to the tag.
1873 	 */
1874 	evl->evl_encap_proto = evl->evl_proto;
1875 
1876 	/*
1877 	 * Remove the encapsulation header and append tag.
1878 	 * The original header has already been fixed up above.
1879 	 */
1880 	vlan_set_tag(m, ntohs(evl->evl_tag));
1881 	memmove((char *)evl + ETHER_VLAN_ENCAP_LEN, evl,
1882 	    offsetof(struct ether_vlan_header, evl_encap_proto));
1883 	m_adj(m, ETHER_VLAN_ENCAP_LEN);
1884 
1885 	return m;
1886 }
1887 
1888 static int
ether_multicast_sysctl(SYSCTLFN_ARGS)1889 ether_multicast_sysctl(SYSCTLFN_ARGS)
1890 {
1891 	struct ether_multi *enm;
1892 	struct ifnet *ifp;
1893 	struct ethercom *ec;
1894 	int error = 0;
1895 	size_t written;
1896 	struct psref psref;
1897 	int bound;
1898 	unsigned int multicnt;
1899 	struct ether_multi_sysctl *addrs;
1900 	int i;
1901 
1902 	if (namelen != 1)
1903 		return EINVAL;
1904 
1905 	bound = curlwp_bind();
1906 	ifp = if_get_byindex(name[0], &psref);
1907 	if (ifp == NULL) {
1908 		error = ENODEV;
1909 		goto out;
1910 	}
1911 	if (ifp->if_type != IFT_ETHER) {
1912 		if_put(ifp, &psref);
1913 		*oldlenp = 0;
1914 		goto out;
1915 	}
1916 	ec = (struct ethercom *)ifp;
1917 
1918 	if (oldp == NULL) {
1919 		if_put(ifp, &psref);
1920 		*oldlenp = ec->ec_multicnt * sizeof(*addrs);
1921 		goto out;
1922 	}
1923 
1924 	/*
1925 	 * ec->ec_lock is a spin mutex so we cannot call sysctl_copyout, which
1926 	 * is sleepable, while holding it. Copy data to a local buffer first
1927 	 * with the lock taken and then call sysctl_copyout without holding it.
1928 	 */
1929 retry:
1930 	multicnt = ec->ec_multicnt;
1931 
1932 	if (multicnt == 0) {
1933 		if_put(ifp, &psref);
1934 		*oldlenp = 0;
1935 		goto out;
1936 	}
1937 
1938 	addrs = kmem_zalloc(sizeof(*addrs) * multicnt, KM_SLEEP);
1939 
1940 	ETHER_LOCK(ec);
1941 	if (multicnt != ec->ec_multicnt) {
1942 		/* The number of multicast addresses has changed */
1943 		ETHER_UNLOCK(ec);
1944 		kmem_free(addrs, sizeof(*addrs) * multicnt);
1945 		goto retry;
1946 	}
1947 
1948 	i = 0;
1949 	LIST_FOREACH(enm, &ec->ec_multiaddrs, enm_list) {
1950 		struct ether_multi_sysctl *addr = &addrs[i];
1951 		addr->enm_refcount = enm->enm_refcount;
1952 		memcpy(addr->enm_addrlo, enm->enm_addrlo, ETHER_ADDR_LEN);
1953 		memcpy(addr->enm_addrhi, enm->enm_addrhi, ETHER_ADDR_LEN);
1954 		i++;
1955 	}
1956 	ETHER_UNLOCK(ec);
1957 
1958 	error = 0;
1959 	written = 0;
1960 	for (i = 0; i < multicnt; i++) {
1961 		struct ether_multi_sysctl *addr = &addrs[i];
1962 
1963 		if (written + sizeof(*addr) > *oldlenp)
1964 			break;
1965 		error = sysctl_copyout(l, addr, oldp, sizeof(*addr));
1966 		if (error)
1967 			break;
1968 		written += sizeof(*addr);
1969 		oldp = (char *)oldp + sizeof(*addr);
1970 	}
1971 	kmem_free(addrs, sizeof(*addrs) * multicnt);
1972 
1973 	if_put(ifp, &psref);
1974 
1975 	*oldlenp = written;
1976 out:
1977 	curlwp_bindx(bound);
1978 	return error;
1979 }
1980 
1981 static void
ether_sysctl_setup(struct sysctllog ** clog)1982 ether_sysctl_setup(struct sysctllog **clog)
1983 {
1984 	const struct sysctlnode *rnode = NULL;
1985 
1986 	sysctl_createv(clog, 0, NULL, &rnode,
1987 		       CTLFLAG_PERMANENT,
1988 		       CTLTYPE_NODE, "ether",
1989 		       SYSCTL_DESCR("Ethernet-specific information"),
1990 		       NULL, 0, NULL, 0,
1991 		       CTL_NET, CTL_CREATE, CTL_EOL);
1992 
1993 	sysctl_createv(clog, 0, &rnode, NULL,
1994 		       CTLFLAG_PERMANENT,
1995 		       CTLTYPE_NODE, "multicast",
1996 		       SYSCTL_DESCR("multicast addresses"),
1997 		       ether_multicast_sysctl, 0, NULL, 0,
1998 		       CTL_CREATE, CTL_EOL);
1999 
2000 	sysctl_createv(clog, 0, &rnode, NULL,
2001 		       CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
2002 		       CTLTYPE_STRING, "rps_hash",
2003 		       SYSCTL_DESCR("Interface rps hash function control"),
2004 		       sysctl_pktq_rps_hash_handler, 0, (void *)&ether_pktq_rps_hash_p,
2005 		       PKTQ_RPS_HASH_NAME_LEN,
2006 		       CTL_CREATE, CTL_EOL);
2007 }
2008 
2009 void
etherinit(void)2010 etherinit(void)
2011 {
2012 
2013 #ifdef DIAGNOSTIC
2014 	mutex_init(&bigpktpps_lock, MUTEX_DEFAULT, IPL_NET);
2015 #endif
2016 	ether_pktq_rps_hash_p = pktq_rps_hash_default;
2017 	ether_sysctl_setup(NULL);
2018 }
2019