xref: /dflybsd-src/sys/net/if_ethersubr.c (revision fc962bc679e9e19b6e8b92bfd9d55faf0515eacd)
1 /*
2  * Copyright (c) 1982, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)if_ethersubr.c	8.1 (Berkeley) 6/10/93
30  * $FreeBSD: src/sys/net/if_ethersubr.c,v 1.70.2.33 2003/04/28 15:45:53 archie Exp $
31  */
32 
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #include "opt_mpls.h"
36 #include "opt_netgraph.h"
37 #include "opt_carp.h"
38 #include "opt_rss.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/globaldata.h>
43 #include <sys/kernel.h>
44 #include <sys/ktr.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/msgport.h>
49 #include <sys/socket.h>
50 #include <sys/sockio.h>
51 #include <sys/sysctl.h>
52 #include <sys/thread.h>
53 
54 #include <sys/thread2.h>
55 #include <sys/mplock2.h>
56 
57 #include <net/if.h>
58 #include <net/netisr.h>
59 #include <net/route.h>
60 #include <net/if_llc.h>
61 #include <net/if_dl.h>
62 #include <net/if_types.h>
63 #include <net/ifq_var.h>
64 #include <net/bpf.h>
65 #include <net/ethernet.h>
66 #include <net/vlan/if_vlan_ether.h>
67 #include <net/vlan/if_vlan_var.h>
68 #include <net/netmsg2.h>
69 #include <net/netisr2.h>
70 
71 #if defined(INET) || defined(INET6)
72 #include <netinet/in.h>
73 #include <netinet/ip_var.h>
74 #include <netinet/tcp_var.h>
75 #include <netinet/if_ether.h>
76 #include <netinet/ip_flow.h>
77 #include <net/ipfw/ip_fw.h>
78 #include <net/ipfw3/ip_fw.h>
79 #include <net/dummynet/ip_dummynet.h>
80 #endif
81 #ifdef INET6
82 #include <netinet6/nd6.h>
83 #endif
84 
85 #ifdef CARP
86 #include <netinet/ip_carp.h>
87 #endif
88 
89 #ifdef MPLS
90 #include <netproto/mpls/mpls.h>
91 #endif
92 
93 /* netgraph node hooks for ng_ether(4) */
94 void	(*ng_ether_input_p)(struct ifnet *ifp, struct mbuf **mp);
95 void	(*ng_ether_input_orphan_p)(struct ifnet *ifp, struct mbuf *m);
96 int	(*ng_ether_output_p)(struct ifnet *ifp, struct mbuf **mp);
97 void	(*ng_ether_attach_p)(struct ifnet *ifp);
98 void	(*ng_ether_detach_p)(struct ifnet *ifp);
99 
100 void	(*vlan_input_p)(struct mbuf *);
101 
102 static int ether_output(struct ifnet *, struct mbuf *, struct sockaddr *,
103 			struct rtentry *);
104 static void ether_restore_header(struct mbuf **, const struct ether_header *,
105 				 const struct ether_header *);
106 static int ether_characterize(struct mbuf **);
107 static void ether_dispatch(int, struct mbuf *, int);
108 
109 /*
110  * if_bridge support
111  */
112 struct mbuf *(*bridge_input_p)(struct ifnet *, struct mbuf *);
113 int (*bridge_output_p)(struct ifnet *, struct mbuf *);
114 void (*bridge_dn_p)(struct mbuf *, struct ifnet *);
115 struct ifnet *(*bridge_interface_p)(void *if_bridge);
116 
117 static int ether_resolvemulti(struct ifnet *, struct sockaddr **,
118 			      struct sockaddr *);
119 
120 /*
121  * if_lagg(4) support
122  */
123 void	(*lagg_input_p)(struct ifnet *, struct mbuf *);
124 int (*lagg_output_p)(struct ifnet *, struct mbuf *);
125 
126 const uint8_t etherbroadcastaddr[ETHER_ADDR_LEN] = {
127 	0xff, 0xff, 0xff, 0xff, 0xff, 0xff
128 };
129 
130 #define gotoerr(e) do { error = (e); goto bad; } while (0)
131 #define IFP2AC(ifp) ((struct arpcom *)(ifp))
132 
133 static boolean_t ether_ipfw_chk(struct mbuf **m0, struct ifnet *dst,
134 				struct ip_fw **rule,
135 				const struct ether_header *eh);
136 
137 static int ether_ipfw;
138 static u_long ether_restore_hdr;
139 static u_long ether_prepend_hdr;
140 static u_long ether_input_wronghash;
141 static int ether_debug;
142 
143 #ifdef RSS_DEBUG
144 static u_long ether_pktinfo_try;
145 static u_long ether_pktinfo_hit;
146 static u_long ether_rss_nopi;
147 static u_long ether_rss_nohash;
148 static u_long ether_input_requeue;
149 #endif
150 static u_long ether_input_wronghwhash;
151 static int ether_input_ckhash;
152 
153 #define ETHER_TSOLEN_DEFAULT	(4 * ETHERMTU)
154 
155 #define ETHER_NMBCLUSTERS_DEFMIN	32
156 #define ETHER_NMBCLUSTERS_DEFAULT	256
157 
158 static int ether_tsolen_default = ETHER_TSOLEN_DEFAULT;
159 TUNABLE_INT("net.link.ether.tsolen", &ether_tsolen_default);
160 
161 static int ether_nmbclusters_default = ETHER_NMBCLUSTERS_DEFAULT;
162 TUNABLE_INT("net.link.ether.nmbclusters", &ether_nmbclusters_default);
163 
164 SYSCTL_DECL(_net_link);
165 SYSCTL_NODE(_net_link, IFT_ETHER, ether, CTLFLAG_RW, 0, "Ethernet");
166 SYSCTL_INT(_net_link_ether, OID_AUTO, debug, CTLFLAG_RW,
167     &ether_debug, 0, "Ether debug");
168 SYSCTL_INT(_net_link_ether, OID_AUTO, ipfw, CTLFLAG_RW,
169     &ether_ipfw, 0, "Pass ether pkts through firewall");
170 SYSCTL_ULONG(_net_link_ether, OID_AUTO, restore_hdr, CTLFLAG_RW,
171     &ether_restore_hdr, 0, "# of ether header restoration");
172 SYSCTL_ULONG(_net_link_ether, OID_AUTO, prepend_hdr, CTLFLAG_RW,
173     &ether_prepend_hdr, 0,
174     "# of ether header restoration which prepends mbuf");
175 SYSCTL_ULONG(_net_link_ether, OID_AUTO, input_wronghash, CTLFLAG_RW,
176     &ether_input_wronghash, 0, "# of input packets with wrong hash");
177 SYSCTL_INT(_net_link_ether, OID_AUTO, tsolen, CTLFLAG_RW,
178     &ether_tsolen_default, 0, "Default max TSO length");
179 
180 #ifdef RSS_DEBUG
181 SYSCTL_ULONG(_net_link_ether, OID_AUTO, rss_nopi, CTLFLAG_RW,
182     &ether_rss_nopi, 0, "# of packets do not have pktinfo");
183 SYSCTL_ULONG(_net_link_ether, OID_AUTO, rss_nohash, CTLFLAG_RW,
184     &ether_rss_nohash, 0, "# of packets do not have hash");
185 SYSCTL_ULONG(_net_link_ether, OID_AUTO, pktinfo_try, CTLFLAG_RW,
186     &ether_pktinfo_try, 0,
187     "# of tries to find packets' msgport using pktinfo");
188 SYSCTL_ULONG(_net_link_ether, OID_AUTO, pktinfo_hit, CTLFLAG_RW,
189     &ether_pktinfo_hit, 0,
190     "# of packets whose msgport are found using pktinfo");
191 SYSCTL_ULONG(_net_link_ether, OID_AUTO, input_requeue, CTLFLAG_RW,
192     &ether_input_requeue, 0, "# of input packets gets requeued");
193 #endif
194 SYSCTL_ULONG(_net_link_ether, OID_AUTO, input_wronghwhash, CTLFLAG_RW,
195     &ether_input_wronghwhash, 0, "# of input packets with wrong hw hash");
196 SYSCTL_INT(_net_link_ether, OID_AUTO, always_ckhash, CTLFLAG_RW,
197     &ether_input_ckhash, 0, "always check hash");
198 
199 #define ETHER_KTR_STR		"ifp=%p"
200 #define ETHER_KTR_ARGS	struct ifnet *ifp
201 #ifndef KTR_ETHERNET
202 #define KTR_ETHERNET		KTR_ALL
203 #endif
204 KTR_INFO_MASTER(ether);
205 KTR_INFO(KTR_ETHERNET, ether, pkt_beg, 0, ETHER_KTR_STR, ETHER_KTR_ARGS);
206 KTR_INFO(KTR_ETHERNET, ether, pkt_end, 1, ETHER_KTR_STR, ETHER_KTR_ARGS);
207 KTR_INFO(KTR_ETHERNET, ether, disp_beg, 2, ETHER_KTR_STR, ETHER_KTR_ARGS);
208 KTR_INFO(KTR_ETHERNET, ether, disp_end, 3, ETHER_KTR_STR, ETHER_KTR_ARGS);
209 #define logether(name, arg)	KTR_LOG(ether_ ## name, arg)
210 
211 /*
212  * Ethernet output routine.
213  * Encapsulate a packet of type family for the local net.
214  * Use trailer local net encapsulation if enough data in first
215  * packet leaves a multiple of 512 bytes of data in remainder.
216  * Assumes that ifp is actually pointer to arpcom structure.
217  */
218 static int
219 ether_output(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst,
220 	     struct rtentry *rt)
221 {
222 	struct ether_header *eh, *deh;
223 	u_char *edst;
224 	int loop_copy = 0;
225 	int hlen = ETHER_HDR_LEN;	/* link layer header length */
226 	struct arpcom *ac = IFP2AC(ifp);
227 	int error;
228 
229 	ASSERT_NETISR_NCPUS(mycpuid);
230 	ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp);
231 
232 	if (ifp->if_flags & IFF_MONITOR)
233 		gotoerr(ENETDOWN);
234 	if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) != (IFF_UP | IFF_RUNNING))
235 		gotoerr(ENETDOWN);
236 
237 	M_PREPEND(m, sizeof(struct ether_header), M_NOWAIT);
238 	if (m == NULL)
239 		return (ENOBUFS);
240 	m->m_pkthdr.csum_lhlen = sizeof(struct ether_header);
241 	eh = mtod(m, struct ether_header *);
242 	edst = eh->ether_dhost;
243 
244 	/*
245 	 * Fill in the destination ethernet address and frame type.
246 	 */
247 	switch (dst->sa_family) {
248 #ifdef INET
249 	case AF_INET:
250 		if (!arpresolve(ifp, rt, m, dst, edst))
251 			return (0);	/* if not yet resolved */
252 #ifdef MPLS
253 		if (m->m_flags & M_MPLSLABELED)
254 			eh->ether_type = htons(ETHERTYPE_MPLS);
255 		else
256 #endif
257 			eh->ether_type = htons(ETHERTYPE_IP);
258 		break;
259 #endif
260 #ifdef INET6
261 	case AF_INET6:
262 		if (!nd6_storelladdr(&ac->ac_if, rt, m, dst, edst))
263 			return (0);		/* Something bad happenned. */
264 		eh->ether_type = htons(ETHERTYPE_IPV6);
265 		break;
266 #endif
267 	case pseudo_AF_HDRCMPLT:
268 	case AF_UNSPEC:
269 		loop_copy = -1; /* if this is for us, don't do it */
270 		deh = (struct ether_header *)dst->sa_data;
271 		memcpy(edst, deh->ether_dhost, ETHER_ADDR_LEN);
272 		eh->ether_type = deh->ether_type;
273 		break;
274 
275 	default:
276 		if_printf(ifp, "can't handle af%d\n", dst->sa_family);
277 		gotoerr(EAFNOSUPPORT);
278 	}
279 
280 	if (dst->sa_family == pseudo_AF_HDRCMPLT)	/* unlikely */
281 		memcpy(eh->ether_shost,
282 		       ((struct ether_header *)dst->sa_data)->ether_shost,
283 		       ETHER_ADDR_LEN);
284 	else
285 		memcpy(eh->ether_shost, ac->ac_enaddr, ETHER_ADDR_LEN);
286 
287 	/*
288 	 * Bridges require special output handling.
289 	 */
290 	if (ifp->if_bridge) {
291 		KASSERT(bridge_output_p != NULL,
292 			("%s: if_bridge not loaded!", __func__));
293 		return bridge_output_p(ifp, m);
294 	}
295 #if 0 /* XXX */
296 	if (ifp->if_lagg) {
297 		KASSERT(lagg_output_p != NULL,
298 			("%s: if_lagg not loaded!", __func__));
299 		return lagg_output_p(ifp, m);
300 	}
301 #endif
302 
303 	/*
304 	 * If a simplex interface, and the packet is being sent to our
305 	 * Ethernet address or a broadcast address, loopback a copy.
306 	 * XXX To make a simplex device behave exactly like a duplex
307 	 * device, we should copy in the case of sending to our own
308 	 * ethernet address (thus letting the original actually appear
309 	 * on the wire). However, we don't do that here for security
310 	 * reasons and compatibility with the original behavior.
311 	 */
312 	if ((ifp->if_flags & IFF_SIMPLEX) && (loop_copy != -1)) {
313 		int csum_flags = 0;
314 
315 		if (m->m_pkthdr.csum_flags & CSUM_IP)
316 			csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID);
317 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA)
318 			csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
319 		if ((m->m_flags & M_BCAST) || (loop_copy > 0)) {
320 			struct mbuf *n;
321 
322 			if ((n = m_copypacket(m, M_NOWAIT)) != NULL) {
323 				n->m_pkthdr.csum_flags |= csum_flags;
324 				if (csum_flags & CSUM_DATA_VALID)
325 					n->m_pkthdr.csum_data = 0xffff;
326 				if_simloop(ifp, n, dst->sa_family, hlen);
327 			} else
328 				IFNET_STAT_INC(ifp, iqdrops, 1);
329 		} else if (bcmp(eh->ether_dhost, eh->ether_shost,
330 				ETHER_ADDR_LEN) == 0) {
331 			m->m_pkthdr.csum_flags |= csum_flags;
332 			if (csum_flags & CSUM_DATA_VALID)
333 				m->m_pkthdr.csum_data = 0xffff;
334 			if_simloop(ifp, m, dst->sa_family, hlen);
335 			return (0);	/* XXX */
336 		}
337 	}
338 
339 #ifdef CARP
340 	if (ifp->if_type == IFT_CARP) {
341 		ifp = carp_parent(ifp);
342 		if (ifp == NULL)
343 			gotoerr(ENETUNREACH);
344 
345 		ac = IFP2AC(ifp);
346 
347 		/*
348 		 * Check precondition again
349 		 */
350 		ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp);
351 
352 		if (ifp->if_flags & IFF_MONITOR)
353 			gotoerr(ENETDOWN);
354 		if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) !=
355 		    (IFF_UP | IFF_RUNNING))
356 			gotoerr(ENETDOWN);
357 	}
358 #endif
359 
360 	/* Handle ng_ether(4) processing, if any */
361 	if (ng_ether_output_p != NULL) {
362 		/*
363 		 * Hold BGL and recheck ng_ether_output_p
364 		 */
365 		get_mplock();
366 		if (ng_ether_output_p != NULL) {
367 			if ((error = ng_ether_output_p(ifp, &m)) != 0) {
368 				rel_mplock();
369 				goto bad;
370 			}
371 			if (m == NULL) {
372 				rel_mplock();
373 				return (0);
374 			}
375 		}
376 		rel_mplock();
377 	}
378 
379 	/* Continue with link-layer output */
380 	return ether_output_frame(ifp, m);
381 
382 bad:
383 	m_freem(m);
384 	return (error);
385 }
386 
387 /*
388  * Returns the bridge interface an ifp is associated
389  * with.
390  *
391  * Only call if ifp->if_bridge != NULL.
392  */
393 struct ifnet *
394 ether_bridge_interface(struct ifnet *ifp)
395 {
396 	if (bridge_interface_p)
397 		return(bridge_interface_p(ifp->if_bridge));
398 	return (ifp);
399 }
400 
401 /*
402  * Ethernet link layer output routine to send a raw frame to the device.
403  *
404  * This assumes that the 14 byte Ethernet header is present and contiguous
405  * in the first mbuf.
406  */
407 int
408 ether_output_frame(struct ifnet *ifp, struct mbuf *m)
409 {
410 	struct ip_fw *rule = NULL;
411 	int error = 0;
412 	struct altq_pktattr pktattr;
413 
414 	ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp);
415 
416 	if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
417 		struct m_tag *mtag;
418 
419 		/* Extract info from dummynet tag */
420 		mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
421 		KKASSERT(mtag != NULL);
422 		rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
423 		KKASSERT(rule != NULL);
424 
425 		m_tag_delete(m, mtag);
426 		m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
427 	}
428 
429 	if (ifq_is_enabled(&ifp->if_snd))
430 		altq_etherclassify(&ifp->if_snd, m, &pktattr);
431 	crit_enter();
432 	if ((IPFW_LOADED || IPFW3_LOADED) && ether_ipfw != 0) {
433 		struct ether_header save_eh, *eh;
434 
435 		eh = mtod(m, struct ether_header *);
436 		save_eh = *eh;
437 		m_adj(m, ETHER_HDR_LEN);
438 		if (!ether_ipfw_chk(&m, ifp, &rule, eh)) {
439 			crit_exit();
440 			if (m != NULL) {
441 				m_freem(m);
442 				return ENOBUFS; /* pkt dropped */
443 			} else
444 				return 0;	/* consumed e.g. in a pipe */
445 		}
446 
447 		/* packet was ok, restore the ethernet header */
448 		ether_restore_header(&m, eh, &save_eh);
449 		if (m == NULL) {
450 			crit_exit();
451 			return ENOBUFS;
452 		}
453 	}
454 	crit_exit();
455 
456 	/*
457 	 * Queue message on interface, update output statistics if
458 	 * successful, and start output if interface not yet active.
459 	 */
460 	error = ifq_dispatch(ifp, m, &pktattr);
461 	return (error);
462 }
463 
464 /*
465  * ipfw processing for ethernet packets (in and out).
466  * The second parameter is NULL from ether_demux(), and ifp from
467  * ether_output_frame().
468  */
469 static boolean_t
470 ether_ipfw_chk(struct mbuf **m0, struct ifnet *dst, struct ip_fw **rule,
471 	       const struct ether_header *eh)
472 {
473 	struct ether_header save_eh = *eh;	/* might be a ptr in *m0 */
474 	struct ip_fw_args args;
475 	struct m_tag *mtag;
476 	struct mbuf *m;
477 	int i;
478 
479 	if (*rule != NULL && fw_one_pass)
480 		return TRUE; /* dummynet packet, already partially processed */
481 
482 	/*
483 	 * I need some amount of data to be contiguous.
484 	 */
485 	i = min((*m0)->m_pkthdr.len, max_protohdr);
486 	if ((*m0)->m_len < i) {
487 		*m0 = m_pullup(*m0, i);
488 		if (*m0 == NULL)
489 			return FALSE;
490 	}
491 
492 	/*
493 	 * Clean up tags
494 	 */
495 	if ((mtag = m_tag_find(*m0, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL)
496 		m_tag_delete(*m0, mtag);
497 	if ((*m0)->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
498 		mtag = m_tag_find(*m0, PACKET_TAG_IPFORWARD, NULL);
499 		KKASSERT(mtag != NULL);
500 		m_tag_delete(*m0, mtag);
501 		(*m0)->m_pkthdr.fw_flags &= ~IPFORWARD_MBUF_TAGGED;
502 	}
503 
504 	args.m = *m0;		/* the packet we are looking at		*/
505 	args.oif = dst;		/* destination, if any			*/
506 	args.rule = *rule;	/* matching rule to restart		*/
507 	args.eh = &save_eh;	/* MAC header for bridged/MAC packets	*/
508 	i = ip_fw_chk_ptr(&args);
509 	*m0 = args.m;
510 	*rule = args.rule;
511 
512 	if (*m0 == NULL)
513 		return FALSE;
514 
515 	switch (i) {
516 	case IP_FW_PASS:
517 		return TRUE;
518 
519 	case IP_FW_DIVERT:
520 	case IP_FW_TEE:
521 	case IP_FW_DENY:
522 		/*
523 		 * XXX at some point add support for divert/forward actions.
524 		 * If none of the above matches, we have to drop the pkt.
525 		 */
526 		return FALSE;
527 
528 	case IP_FW_DUMMYNET:
529 		/*
530 		 * Pass the pkt to dummynet, which consumes it.
531 		 */
532 		m = *m0;	/* pass the original to dummynet */
533 		*m0 = NULL;	/* and nothing back to the caller */
534 
535 		ether_restore_header(&m, eh, &save_eh);
536 		if (m == NULL)
537 			return FALSE;
538 
539 		m = ip_fw_dn_io_ptr(m, args.cookie,
540 		    dst ? DN_TO_ETH_OUT: DN_TO_ETH_DEMUX, &args);
541 		if (m != NULL)
542 			ip_dn_queue(m);
543 		return FALSE;
544 
545 	default:
546 		panic("unknown ipfw return value: %d", i);
547 	}
548 }
549 
550 /*
551  * Perform common duties while attaching to interface list
552  */
553 void
554 ether_ifattach(struct ifnet *ifp, const uint8_t *lla,
555     lwkt_serialize_t serializer)
556 {
557 	ether_ifattach_bpf(ifp, lla, DLT_EN10MB, sizeof(struct ether_header),
558 	    serializer);
559 }
560 
561 void
562 ether_ifattach_bpf(struct ifnet *ifp, const uint8_t *lla,
563     u_int dlt, u_int hdrlen, lwkt_serialize_t serializer)
564 {
565 	struct sockaddr_dl *sdl;
566 	char ethstr[ETHER_ADDRSTRLEN + 1];
567 	struct ifaltq *ifq;
568 	int i;
569 
570 	/*
571 	 * If driver does not configure # of mbuf clusters/jclusters
572 	 * that could sit on the device queues for quite some time,
573 	 * we then assume:
574 	 * - The device queues only consume mbuf clusters.
575 	 * - No more than ether_nmbclusters_default (by default 256)
576 	 *   mbuf clusters will sit on the device queues for quite
577 	 *   some time.
578 	 */
579 	if (ifp->if_nmbclusters <= 0 && ifp->if_nmbjclusters <= 0) {
580 		if (ether_nmbclusters_default < ETHER_NMBCLUSTERS_DEFMIN) {
581 			kprintf("ether nmbclusters %d -> %d\n",
582 			    ether_nmbclusters_default,
583 			    ETHER_NMBCLUSTERS_DEFAULT);
584 			ether_nmbclusters_default = ETHER_NMBCLUSTERS_DEFAULT;
585 		}
586 		ifp->if_nmbclusters = ether_nmbclusters_default;
587 	}
588 
589 	ifp->if_type = IFT_ETHER;
590 	ifp->if_addrlen = ETHER_ADDR_LEN;
591 	ifp->if_hdrlen = ETHER_HDR_LEN;
592 	if_attach(ifp, serializer);
593 	ifq = &ifp->if_snd;
594 	for (i = 0; i < ifq->altq_subq_cnt; ++i) {
595 		struct ifaltq_subque *ifsq = ifq_get_subq(ifq, i);
596 
597 		ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen *
598 		    (ETHER_MAX_LEN - ETHER_CRC_LEN);
599 	}
600 	ifp->if_mtu = ETHERMTU;
601 	if (ifp->if_tsolen <= 0) {
602 		if ((ether_tsolen_default / ETHERMTU) < 2) {
603 			kprintf("ether TSO maxlen %d -> %d\n",
604 			    ether_tsolen_default, ETHER_TSOLEN_DEFAULT);
605 			ether_tsolen_default = ETHER_TSOLEN_DEFAULT;
606 		}
607 		ifp->if_tsolen = ether_tsolen_default;
608 	}
609 	if (ifp->if_baudrate == 0)
610 		ifp->if_baudrate = 10000000;
611 	ifp->if_output = ether_output;
612 	ifp->if_input = ether_input;
613 	ifp->if_resolvemulti = ether_resolvemulti;
614 	ifp->if_broadcastaddr = etherbroadcastaddr;
615 	sdl = IF_LLSOCKADDR(ifp);
616 	sdl->sdl_type = IFT_ETHER;
617 	sdl->sdl_alen = ifp->if_addrlen;
618 	bcopy(lla, LLADDR(sdl), ifp->if_addrlen);
619 	/*
620 	 * XXX Keep the current drivers happy.
621 	 * XXX Remove once all drivers have been cleaned up
622 	 */
623 	if (lla != IFP2AC(ifp)->ac_enaddr)
624 		bcopy(lla, IFP2AC(ifp)->ac_enaddr, ifp->if_addrlen);
625 	bpfattach(ifp, dlt, hdrlen);
626 	if (ng_ether_attach_p != NULL)
627 		(*ng_ether_attach_p)(ifp);
628 
629 	if_printf(ifp, "MAC address: %s\n", kether_ntoa(lla, ethstr));
630 }
631 
632 /*
633  * Perform common duties while detaching an Ethernet interface
634  */
635 void
636 ether_ifdetach(struct ifnet *ifp)
637 {
638 	if_down(ifp);
639 
640 	if (ng_ether_detach_p != NULL)
641 		(*ng_ether_detach_p)(ifp);
642 	bpfdetach(ifp);
643 	if_detach(ifp);
644 }
645 
646 int
647 ether_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
648 {
649 	struct ifaddr *ifa = (struct ifaddr *) data;
650 	struct ifreq *ifr = (struct ifreq *) data;
651 	int error = 0;
652 
653 #define IF_INIT(ifp) \
654 do { \
655 	if (((ifp)->if_flags & IFF_UP) == 0) { \
656 		(ifp)->if_flags |= IFF_UP; \
657 		(ifp)->if_init((ifp)->if_softc); \
658 	} \
659 } while (0)
660 
661 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
662 
663 	switch (command) {
664 	case SIOCSIFADDR:
665 		switch (ifa->ifa_addr->sa_family) {
666 #ifdef INET
667 		case AF_INET:
668 			IF_INIT(ifp);	/* before arpwhohas */
669 			arp_ifinit(ifp, ifa);
670 			break;
671 #endif
672 		default:
673 			IF_INIT(ifp);
674 			break;
675 		}
676 		break;
677 
678 	case SIOCGIFADDR:
679 		bcopy(IFP2AC(ifp)->ac_enaddr,
680 		      ((struct sockaddr *)ifr->ifr_data)->sa_data,
681 		      ETHER_ADDR_LEN);
682 		break;
683 
684 	case SIOCSIFMTU:
685 		/*
686 		 * Set the interface MTU.
687 		 */
688 		if (ifr->ifr_mtu > ETHERMTU) {
689 			error = EINVAL;
690 		} else {
691 			ifp->if_mtu = ifr->ifr_mtu;
692 		}
693 		break;
694 	default:
695 		error = EINVAL;
696 		break;
697 	}
698 	return (error);
699 
700 #undef IF_INIT
701 }
702 
703 static int
704 ether_resolvemulti(
705 	struct ifnet *ifp,
706 	struct sockaddr **llsa,
707 	struct sockaddr *sa)
708 {
709 	struct sockaddr_dl *sdl;
710 #ifdef INET
711 	struct sockaddr_in *sin;
712 #endif
713 #ifdef INET6
714 	struct sockaddr_in6 *sin6;
715 #endif
716 	u_char *e_addr;
717 
718 	switch(sa->sa_family) {
719 	case AF_LINK:
720 		/*
721 		 * No mapping needed. Just check that it's a valid MC address.
722 		 */
723 		sdl = (struct sockaddr_dl *)sa;
724 		e_addr = LLADDR(sdl);
725 		if ((e_addr[0] & 1) != 1)
726 			return EADDRNOTAVAIL;
727 		*llsa = NULL;
728 		return 0;
729 
730 #ifdef INET
731 	case AF_INET:
732 		sin = (struct sockaddr_in *)sa;
733 		if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
734 			return EADDRNOTAVAIL;
735 		sdl = kmalloc(sizeof *sdl, M_IFMADDR, M_WAITOK | M_ZERO);
736 		sdl->sdl_len = sizeof *sdl;
737 		sdl->sdl_family = AF_LINK;
738 		sdl->sdl_index = ifp->if_index;
739 		sdl->sdl_type = IFT_ETHER;
740 		sdl->sdl_alen = ETHER_ADDR_LEN;
741 		e_addr = LLADDR(sdl);
742 		ETHER_MAP_IP_MULTICAST(&sin->sin_addr, e_addr);
743 		*llsa = (struct sockaddr *)sdl;
744 		return 0;
745 #endif
746 #ifdef INET6
747 	case AF_INET6:
748 		sin6 = (struct sockaddr_in6 *)sa;
749 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
750 			/*
751 			 * An IP6 address of 0 means listen to all
752 			 * of the Ethernet multicast address used for IP6.
753 			 * (This is used for multicast routers.)
754 			 */
755 			ifp->if_flags |= IFF_ALLMULTI;
756 			*llsa = NULL;
757 			return 0;
758 		}
759 		if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
760 			return EADDRNOTAVAIL;
761 		sdl = kmalloc(sizeof *sdl, M_IFMADDR, M_WAITOK | M_ZERO);
762 		sdl->sdl_len = sizeof *sdl;
763 		sdl->sdl_family = AF_LINK;
764 		sdl->sdl_index = ifp->if_index;
765 		sdl->sdl_type = IFT_ETHER;
766 		sdl->sdl_alen = ETHER_ADDR_LEN;
767 		e_addr = LLADDR(sdl);
768 		ETHER_MAP_IPV6_MULTICAST(&sin6->sin6_addr, e_addr);
769 		*llsa = (struct sockaddr *)sdl;
770 		return 0;
771 #endif
772 
773 	default:
774 		/*
775 		 * Well, the text isn't quite right, but it's the name
776 		 * that counts...
777 		 */
778 		return EAFNOSUPPORT;
779 	}
780 }
781 
782 #if 0
783 /*
784  * This is for reference.  We have a table-driven version
785  * of the little-endian crc32 generator, which is faster
786  * than the double-loop.
787  */
788 uint32_t
789 ether_crc32_le(const uint8_t *buf, size_t len)
790 {
791 	uint32_t c, crc, carry;
792 	size_t i, j;
793 
794 	crc = 0xffffffffU;	/* initial value */
795 
796 	for (i = 0; i < len; i++) {
797 		c = buf[i];
798 		for (j = 0; j < 8; j++) {
799 			carry = ((crc & 0x01) ? 1 : 0) ^ (c & 0x01);
800 			crc >>= 1;
801 			c >>= 1;
802 			if (carry)
803 				crc = (crc ^ ETHER_CRC_POLY_LE);
804 		}
805 	}
806 
807 	return (crc);
808 }
809 #else
810 uint32_t
811 ether_crc32_le(const uint8_t *buf, size_t len)
812 {
813 	static const uint32_t crctab[] = {
814 		0x00000000, 0x1db71064, 0x3b6e20c8, 0x26d930ac,
815 		0x76dc4190, 0x6b6b51f4, 0x4db26158, 0x5005713c,
816 		0xedb88320, 0xf00f9344, 0xd6d6a3e8, 0xcb61b38c,
817 		0x9b64c2b0, 0x86d3d2d4, 0xa00ae278, 0xbdbdf21c
818 	};
819 	uint32_t crc;
820 	size_t i;
821 
822 	crc = 0xffffffffU;	/* initial value */
823 
824 	for (i = 0; i < len; i++) {
825 		crc ^= buf[i];
826 		crc = (crc >> 4) ^ crctab[crc & 0xf];
827 		crc = (crc >> 4) ^ crctab[crc & 0xf];
828 	}
829 
830 	return (crc);
831 }
832 #endif
833 
834 uint32_t
835 ether_crc32_be(const uint8_t *buf, size_t len)
836 {
837 	uint32_t c, crc, carry;
838 	size_t i, j;
839 
840 	crc = 0xffffffffU;	/* initial value */
841 
842 	for (i = 0; i < len; i++) {
843 		c = buf[i];
844 		for (j = 0; j < 8; j++) {
845 			carry = ((crc & 0x80000000U) ? 1 : 0) ^ (c & 0x01);
846 			crc <<= 1;
847 			c >>= 1;
848 			if (carry)
849 				crc = (crc ^ ETHER_CRC_POLY_BE) | carry;
850 		}
851 	}
852 
853 	return (crc);
854 }
855 
856 /*
857  * find the size of ethernet header, and call classifier
858  */
859 void
860 altq_etherclassify(struct ifaltq *ifq, struct mbuf *m,
861 		   struct altq_pktattr *pktattr)
862 {
863 	struct ether_header *eh;
864 	uint16_t ether_type;
865 	int hlen, af, hdrsize;
866 
867 	hlen = sizeof(struct ether_header);
868 	eh = mtod(m, struct ether_header *);
869 
870 	ether_type = ntohs(eh->ether_type);
871 	if (ether_type < ETHERMTU) {
872 		/* ick! LLC/SNAP */
873 		struct llc *llc = (struct llc *)(eh + 1);
874 		hlen += 8;
875 
876 		if (m->m_len < hlen ||
877 		    llc->llc_dsap != LLC_SNAP_LSAP ||
878 		    llc->llc_ssap != LLC_SNAP_LSAP ||
879 		    llc->llc_control != LLC_UI)
880 			goto bad;  /* not snap! */
881 
882 		ether_type = ntohs(llc->llc_un.type_snap.ether_type);
883 	}
884 
885 	if (ether_type == ETHERTYPE_IP) {
886 		af = AF_INET;
887 		hdrsize = 20;  /* sizeof(struct ip) */
888 #ifdef INET6
889 	} else if (ether_type == ETHERTYPE_IPV6) {
890 		af = AF_INET6;
891 		hdrsize = 40;  /* sizeof(struct ip6_hdr) */
892 #endif
893 	} else
894 		goto bad;
895 
896 	while (m->m_len <= hlen) {
897 		hlen -= m->m_len;
898 		m = m->m_next;
899 	}
900 	if (m->m_len < hlen + hdrsize) {
901 		/*
902 		 * ip header is not in a single mbuf.  this should not
903 		 * happen in the current code.
904 		 * (todo: use m_pulldown in the future)
905 		 */
906 		goto bad;
907 	}
908 	m->m_data += hlen;
909 	m->m_len -= hlen;
910 	ifq_classify(ifq, m, af, pktattr);
911 	m->m_data -= hlen;
912 	m->m_len += hlen;
913 
914 	return;
915 
916 bad:
917 	pktattr->pattr_class = NULL;
918 	pktattr->pattr_hdr = NULL;
919 	pktattr->pattr_af = AF_UNSPEC;
920 }
921 
922 static void
923 ether_restore_header(struct mbuf **m0, const struct ether_header *eh,
924 		     const struct ether_header *save_eh)
925 {
926 	struct mbuf *m = *m0;
927 
928 	ether_restore_hdr++;
929 
930 	/*
931 	 * Prepend the header, optimize for the common case of
932 	 * eh pointing into the mbuf.
933 	 */
934 	if ((const void *)(eh + 1) == (void *)m->m_data) {
935 		m->m_data -= ETHER_HDR_LEN;
936 		m->m_len += ETHER_HDR_LEN;
937 		m->m_pkthdr.len += ETHER_HDR_LEN;
938 	} else {
939 		ether_prepend_hdr++;
940 
941 		M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT);
942 		if (m != NULL) {
943 			bcopy(save_eh, mtod(m, struct ether_header *),
944 			      ETHER_HDR_LEN);
945 		}
946 	}
947 	*m0 = m;
948 }
949 
950 /*
951  * Upper layer processing for a received Ethernet packet.
952  */
953 void
954 ether_demux_oncpu(struct ifnet *ifp, struct mbuf *m)
955 {
956 	struct ether_header *eh;
957 	int isr, discard = 0;
958 	u_short ether_type;
959 	struct ip_fw *rule = NULL;
960 
961 	M_ASSERTPKTHDR(m);
962 	KASSERT(m->m_len >= ETHER_HDR_LEN,
963 		("ether header is not contiguous!"));
964 
965 	eh = mtod(m, struct ether_header *);
966 
967 	if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
968 		struct m_tag *mtag;
969 
970 		/* Extract info from dummynet tag */
971 		mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
972 		KKASSERT(mtag != NULL);
973 		rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
974 		KKASSERT(rule != NULL);
975 
976 		m_tag_delete(m, mtag);
977 		m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
978 
979 		/* packet is passing the second time */
980 		goto post_stats;
981 	}
982 
983 	/*
984 	 * We got a packet which was unicast to a different Ethernet
985 	 * address.  If the driver is working properly, then this
986 	 * situation can only happen when the interface is in
987 	 * promiscuous mode.  We defer the packet discarding until the
988 	 * vlan processing is done, so that vlan/bridge or vlan/netgraph
989 	 * could work.
990 	 */
991 	if (((ifp->if_flags & (IFF_PROMISC | IFF_PPROMISC)) == IFF_PROMISC) &&
992 	    !ETHER_IS_MULTICAST(eh->ether_dhost) &&
993 	    bcmp(eh->ether_dhost, IFP2AC(ifp)->ac_enaddr, ETHER_ADDR_LEN)) {
994 		if (ether_debug & 1) {
995 			kprintf("%02x:%02x:%02x:%02x:%02x:%02x "
996 				"%02x:%02x:%02x:%02x:%02x:%02x "
997 				"%04x vs %02x:%02x:%02x:%02x:%02x:%02x\n",
998 				eh->ether_dhost[0],
999 				eh->ether_dhost[1],
1000 				eh->ether_dhost[2],
1001 				eh->ether_dhost[3],
1002 				eh->ether_dhost[4],
1003 				eh->ether_dhost[5],
1004 				eh->ether_shost[0],
1005 				eh->ether_shost[1],
1006 				eh->ether_shost[2],
1007 				eh->ether_shost[3],
1008 				eh->ether_shost[4],
1009 				eh->ether_shost[5],
1010 				eh->ether_type,
1011 				((u_char *)IFP2AC(ifp)->ac_enaddr)[0],
1012 				((u_char *)IFP2AC(ifp)->ac_enaddr)[1],
1013 				((u_char *)IFP2AC(ifp)->ac_enaddr)[2],
1014 				((u_char *)IFP2AC(ifp)->ac_enaddr)[3],
1015 				((u_char *)IFP2AC(ifp)->ac_enaddr)[4],
1016 				((u_char *)IFP2AC(ifp)->ac_enaddr)[5]
1017 			);
1018 		}
1019 		if ((ether_debug & 2) == 0)
1020 			discard = 1;
1021 	}
1022 
1023 post_stats:
1024 	if ((IPFW_LOADED || IPFW3_LOADED) && ether_ipfw != 0 && !discard) {
1025 		struct ether_header save_eh = *eh;
1026 
1027 		/* XXX old crufty stuff, needs to be removed */
1028 		m_adj(m, sizeof(struct ether_header));
1029 
1030 		if (!ether_ipfw_chk(&m, NULL, &rule, eh)) {
1031 			m_freem(m);
1032 			return;
1033 		}
1034 
1035 		ether_restore_header(&m, eh, &save_eh);
1036 		if (m == NULL)
1037 			return;
1038 		eh = mtod(m, struct ether_header *);
1039 	}
1040 
1041 	ether_type = ntohs(eh->ether_type);
1042 	KKASSERT(ether_type != ETHERTYPE_VLAN);
1043 
1044         /* Handle input from a lagg(4) port */
1045         if (ifp->if_type == IFT_IEEE8023ADLAG) {
1046                 KASSERT(lagg_input_p != NULL,
1047                     ("%s: if_lagg not loaded!", __func__));
1048                 (*lagg_input_p)(ifp, m);
1049 		return;
1050         }
1051 
1052 	if (m->m_flags & M_VLANTAG) {
1053 		void (*vlan_input_func)(struct mbuf *);
1054 
1055 		vlan_input_func = vlan_input_p;
1056 		/* Make sure 'vlan_input_func' is really used. */
1057 		cpu_ccfence();
1058 		if (vlan_input_func != NULL) {
1059 			vlan_input_func(m);
1060 		} else {
1061 			IFNET_STAT_INC(m->m_pkthdr.rcvif, noproto, 1);
1062 			m_freem(m);
1063 		}
1064 		return;
1065 	}
1066 
1067 	/*
1068 	 * If we have been asked to discard this packet
1069 	 * (e.g. not for us), drop it before entering
1070 	 * the upper layer.
1071 	 */
1072 	if (discard) {
1073 		m_freem(m);
1074 		return;
1075 	}
1076 
1077 	/*
1078 	 * Clear protocol specific flags,
1079 	 * before entering the upper layer.
1080 	 */
1081 	m->m_flags &= ~M_ETHER_FLAGS;
1082 
1083 	/* Strip ethernet header. */
1084 	m_adj(m, sizeof(struct ether_header));
1085 
1086 	switch (ether_type) {
1087 #ifdef INET
1088 	case ETHERTYPE_IP:
1089 		if ((m->m_flags & M_LENCHECKED) == 0) {
1090 			if (!ip_lengthcheck(&m, 0))
1091 				return;
1092 		}
1093 		if (ipflow_fastforward(m))
1094 			return;
1095 		isr = NETISR_IP;
1096 		break;
1097 
1098 	case ETHERTYPE_ARP:
1099 		if (ifp->if_flags & IFF_NOARP) {
1100 			/* Discard packet if ARP is disabled on interface */
1101 			m_freem(m);
1102 			return;
1103 		}
1104 		isr = NETISR_ARP;
1105 		break;
1106 #endif
1107 
1108 #ifdef INET6
1109 	case ETHERTYPE_IPV6:
1110 		isr = NETISR_IPV6;
1111 		break;
1112 #endif
1113 
1114 #ifdef MPLS
1115 	case ETHERTYPE_MPLS:
1116 	case ETHERTYPE_MPLS_MCAST:
1117 		/* Should have been set by ether_input(). */
1118 		KKASSERT(m->m_flags & M_MPLSLABELED);
1119 		isr = NETISR_MPLS;
1120 		break;
1121 #endif
1122 
1123 	default:
1124 		/*
1125 		 * The accurate msgport is not determined before
1126 		 * we reach here, so recharacterize packet.
1127 		 */
1128 		m->m_flags &= ~M_HASH;
1129 		if (ng_ether_input_orphan_p != NULL) {
1130 			/*
1131 			 * Put back the ethernet header so netgraph has a
1132 			 * consistent view of inbound packets.
1133 			 */
1134 			M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT);
1135 			if (m == NULL) {
1136 				/*
1137 				 * M_PREPEND frees the mbuf in case of failure.
1138 				 */
1139 				return;
1140 			}
1141 			/*
1142 			 * Hold BGL and recheck ng_ether_input_orphan_p
1143 			 */
1144 			get_mplock();
1145 			if (ng_ether_input_orphan_p != NULL) {
1146 				ng_ether_input_orphan_p(ifp, m);
1147 				rel_mplock();
1148 				return;
1149 			}
1150 			rel_mplock();
1151 		}
1152 		m_freem(m);
1153 		return;
1154 	}
1155 
1156 	if (m->m_flags & M_HASH) {
1157 		if (&curthread->td_msgport ==
1158 		    netisr_hashport(m->m_pkthdr.hash)) {
1159 			netisr_handle(isr, m);
1160 			return;
1161 		} else {
1162 			/*
1163 			 * XXX Something is wrong,
1164 			 * we probably should panic here!
1165 			 */
1166 			m->m_flags &= ~M_HASH;
1167 			atomic_add_long(&ether_input_wronghash, 1);
1168 		}
1169 	}
1170 #ifdef RSS_DEBUG
1171 	atomic_add_long(&ether_input_requeue, 1);
1172 #endif
1173 	netisr_queue(isr, m);
1174 }
1175 
1176 /*
1177  * First we perform any link layer operations, then continue to the
1178  * upper layers with ether_demux_oncpu().
1179  */
1180 static void
1181 ether_input_oncpu(struct ifnet *ifp, struct mbuf *m)
1182 {
1183 #ifdef CARP
1184 	void *carp;
1185 #endif
1186 
1187 	if ((ifp->if_flags & (IFF_UP | IFF_MONITOR)) != IFF_UP) {
1188 		/*
1189 		 * Receiving interface's flags are changed, when this
1190 		 * packet is waiting for processing; discard it.
1191 		 */
1192 		m_freem(m);
1193 		return;
1194 	}
1195 
1196 	/*
1197 	 * Tap the packet off here for a bridge.  bridge_input()
1198 	 * will return NULL if it has consumed the packet, otherwise
1199 	 * it gets processed as normal.  Note that bridge_input()
1200 	 * will always return the original packet if we need to
1201 	 * process it locally.
1202 	 */
1203 	if (ifp->if_bridge) {
1204 		KASSERT(bridge_input_p != NULL,
1205 			("%s: if_bridge not loaded!", __func__));
1206 
1207 		if(m->m_flags & M_ETHER_BRIDGED) {
1208 			m->m_flags &= ~M_ETHER_BRIDGED;
1209 		} else {
1210 			m = bridge_input_p(ifp, m);
1211 			if (m == NULL)
1212 				return;
1213 
1214 			KASSERT(ifp == m->m_pkthdr.rcvif,
1215 				("bridge_input_p changed rcvif"));
1216 		}
1217 	}
1218 
1219 #ifdef CARP
1220 	carp = ifp->if_carp;
1221 	if (carp) {
1222 		m = carp_input(carp, m);
1223 		if (m == NULL)
1224 			return;
1225 		KASSERT(ifp == m->m_pkthdr.rcvif,
1226 		    ("carp_input changed rcvif"));
1227 	}
1228 #endif
1229 
1230 	/* Handle ng_ether(4) processing, if any */
1231 	if (ng_ether_input_p != NULL) {
1232 		/*
1233 		 * Hold BGL and recheck ng_ether_input_p
1234 		 */
1235 		get_mplock();
1236 		if (ng_ether_input_p != NULL)
1237 			ng_ether_input_p(ifp, &m);
1238 		rel_mplock();
1239 
1240 		if (m == NULL)
1241 			return;
1242 	}
1243 
1244 	/* Continue with upper layer processing */
1245 	ether_demux_oncpu(ifp, m);
1246 }
1247 
1248 /*
1249  * Perform certain functions of ether_input():
1250  * - Test IFF_UP
1251  * - Update statistics
1252  * - Run bpf(4) tap if requested
1253  * Then pass the packet to ether_input_oncpu().
1254  *
1255  * This function should be used by pseudo interface (e.g. vlan(4)),
1256  * when it tries to claim that the packet is received by it.
1257  *
1258  * REINPUT_KEEPRCVIF
1259  * REINPUT_RUNBPF
1260  */
1261 void
1262 ether_reinput_oncpu(struct ifnet *ifp, struct mbuf *m, int reinput_flags)
1263 {
1264 	/* Discard packet if interface is not up */
1265 	if (!(ifp->if_flags & IFF_UP)) {
1266 		m_freem(m);
1267 		return;
1268 	}
1269 
1270 	/*
1271 	 * Change receiving interface.  The bridge will often pass a flag to
1272 	 * ask that this not be done so ARPs get applied to the correct
1273 	 * side.
1274 	 */
1275 	if ((reinput_flags & REINPUT_KEEPRCVIF) == 0 ||
1276 	    m->m_pkthdr.rcvif == NULL) {
1277 		m->m_pkthdr.rcvif = ifp;
1278 	}
1279 
1280 	/* Update statistics */
1281 	IFNET_STAT_INC(ifp, ipackets, 1);
1282 	IFNET_STAT_INC(ifp, ibytes, m->m_pkthdr.len);
1283 	if (m->m_flags & (M_MCAST | M_BCAST))
1284 		IFNET_STAT_INC(ifp, imcasts, 1);
1285 
1286 	if (reinput_flags & REINPUT_RUNBPF)
1287 		BPF_MTAP(ifp, m);
1288 
1289 	ether_input_oncpu(ifp, m);
1290 }
1291 
1292 static __inline boolean_t
1293 ether_vlancheck(struct mbuf **m0)
1294 {
1295 	struct mbuf *m = *m0;
1296 	struct ether_header *eh;
1297 	uint16_t ether_type;
1298 
1299 	eh = mtod(m, struct ether_header *);
1300 	ether_type = ntohs(eh->ether_type);
1301 
1302 	if (ether_type == ETHERTYPE_VLAN && (m->m_flags & M_VLANTAG) == 0) {
1303 		/*
1304 		 * Extract vlan tag if hardware does not do it for us
1305 		 */
1306 		vlan_ether_decap(&m);
1307 		if (m == NULL)
1308 			goto failed;
1309 
1310 		eh = mtod(m, struct ether_header *);
1311 		ether_type = ntohs(eh->ether_type);
1312 	}
1313 
1314 	if (ether_type == ETHERTYPE_VLAN && (m->m_flags & M_VLANTAG)) {
1315 		/*
1316 		 * To prevent possible dangerous recursion,
1317 		 * we don't do vlan-in-vlan
1318 		 */
1319 		IFNET_STAT_INC(m->m_pkthdr.rcvif, noproto, 1);
1320 		goto failed;
1321 	}
1322 	KKASSERT(ether_type != ETHERTYPE_VLAN);
1323 
1324 	m->m_flags |= M_ETHER_VLANCHECKED;
1325 	*m0 = m;
1326 	return TRUE;
1327 failed:
1328 	if (m != NULL)
1329 		m_freem(m);
1330 	*m0 = NULL;
1331 	return FALSE;
1332 }
1333 
1334 static void
1335 ether_input_handler(netmsg_t nmsg)
1336 {
1337 	struct netmsg_packet *nmp = &nmsg->packet;	/* actual size */
1338 	struct ether_header *eh;
1339 	struct ifnet *ifp;
1340 	struct mbuf *m;
1341 
1342 	m = nmp->nm_packet;
1343 	M_ASSERTPKTHDR(m);
1344 
1345 	if ((m->m_flags & M_ETHER_VLANCHECKED) == 0) {
1346 		if (!ether_vlancheck(&m)) {
1347 			KKASSERT(m == NULL);
1348 			return;
1349 		}
1350 	}
1351 	if ((m->m_flags & (M_HASH | M_CKHASH)) == (M_HASH | M_CKHASH) ||
1352 	    __predict_false(ether_input_ckhash)) {
1353 		int isr;
1354 
1355 		/*
1356 		 * Need to verify the hash supplied by the hardware
1357 		 * which could be wrong.
1358 		 */
1359 		m->m_flags &= ~(M_HASH | M_CKHASH);
1360 		isr = ether_characterize(&m);
1361 		if (m == NULL)
1362 			return;
1363 		KKASSERT(m->m_flags & M_HASH);
1364 
1365 		if (netisr_hashcpu(m->m_pkthdr.hash) != mycpuid) {
1366 			/*
1367 			 * Wrong hardware supplied hash; redispatch
1368 			 */
1369 			ether_dispatch(isr, m, -1);
1370 			if (__predict_false(ether_input_ckhash))
1371 				atomic_add_long(&ether_input_wronghwhash, 1);
1372 			return;
1373 		}
1374 	}
1375 	ifp = m->m_pkthdr.rcvif;
1376 
1377 	eh = mtod(m, struct ether_header *);
1378 	if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
1379 		if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
1380 			 ifp->if_addrlen) == 0)
1381 			m->m_flags |= M_BCAST;
1382 		else
1383 			m->m_flags |= M_MCAST;
1384 		IFNET_STAT_INC(ifp, imcasts, 1);
1385 	}
1386 
1387 	ether_input_oncpu(ifp, m);
1388 }
1389 
1390 /*
1391  * Send the packet to the target netisr msgport
1392  *
1393  * At this point the packet must be characterized (M_HASH set),
1394  * so we know which netisr to send it to.
1395  */
1396 static void
1397 ether_dispatch(int isr, struct mbuf *m, int cpuid)
1398 {
1399 	struct netmsg_packet *pmsg;
1400 	int target_cpuid;
1401 
1402 	KKASSERT(m->m_flags & M_HASH);
1403 	target_cpuid = netisr_hashcpu(m->m_pkthdr.hash);
1404 
1405 	pmsg = &m->m_hdr.mh_netmsg;
1406 	netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport,
1407 		    0, ether_input_handler);
1408 	pmsg->nm_packet = m;
1409 	pmsg->base.lmsg.u.ms_result = isr;
1410 
1411 	logether(disp_beg, NULL);
1412 	if (target_cpuid == cpuid) {
1413 		lwkt_sendmsg_oncpu(netisr_cpuport(target_cpuid),
1414 		    &pmsg->base.lmsg);
1415 	} else {
1416 		lwkt_sendmsg(netisr_cpuport(target_cpuid),
1417 		    &pmsg->base.lmsg);
1418 	}
1419 	logether(disp_end, NULL);
1420 }
1421 
1422 /*
1423  * Process a received Ethernet packet.
1424  *
1425  * The ethernet header is assumed to be in the mbuf so the caller
1426  * MUST MAKE SURE that there are at least sizeof(struct ether_header)
1427  * bytes in the first mbuf.
1428  *
1429  * If the caller knows that the current thread is stick to the current
1430  * cpu, e.g. the interrupt thread or the netisr thread, the current cpuid
1431  * (mycpuid) should be passed through 'cpuid' argument.  Else -1 should
1432  * be passed as 'cpuid' argument.
1433  */
1434 void
1435 ether_input(struct ifnet *ifp, struct mbuf *m, const struct pktinfo *pi,
1436     int cpuid)
1437 {
1438 	int isr;
1439 
1440 	M_ASSERTPKTHDR(m);
1441 
1442 	/* Discard packet if interface is not up */
1443 	if (!(ifp->if_flags & IFF_UP)) {
1444 		m_freem(m);
1445 		return;
1446 	}
1447 
1448 	if (m->m_len < sizeof(struct ether_header)) {
1449 		/* XXX error in the caller. */
1450 		m_freem(m);
1451 		return;
1452 	}
1453 
1454 	m->m_pkthdr.rcvif = ifp;
1455 
1456 	logether(pkt_beg, ifp);
1457 
1458 	ETHER_BPF_MTAP(ifp, m);
1459 
1460 	IFNET_STAT_INC(ifp, ibytes, m->m_pkthdr.len);
1461 
1462 	if (ifp->if_flags & IFF_MONITOR) {
1463 		struct ether_header *eh;
1464 
1465 		eh = mtod(m, struct ether_header *);
1466 		if (ETHER_IS_MULTICAST(eh->ether_dhost))
1467 			IFNET_STAT_INC(ifp, imcasts, 1);
1468 
1469 		/*
1470 		 * Interface marked for monitoring; discard packet.
1471 		 */
1472 		m_freem(m);
1473 
1474 		logether(pkt_end, ifp);
1475 		return;
1476 	}
1477 
1478 	/*
1479 	 * If the packet has been characterized (pi->pi_netisr / M_HASH)
1480 	 * we can dispatch it immediately with trivial checks.
1481 	 */
1482 	if (pi != NULL && (m->m_flags & M_HASH)) {
1483 #ifdef RSS_DEBUG
1484 		atomic_add_long(&ether_pktinfo_try, 1);
1485 #endif
1486 		netisr_hashcheck(pi->pi_netisr, m, pi);
1487 		if (m->m_flags & M_HASH) {
1488 			ether_dispatch(pi->pi_netisr, m, cpuid);
1489 #ifdef RSS_DEBUG
1490 			atomic_add_long(&ether_pktinfo_hit, 1);
1491 #endif
1492 			logether(pkt_end, ifp);
1493 			return;
1494 		}
1495 	}
1496 #ifdef RSS_DEBUG
1497 	else if (ifp->if_capenable & IFCAP_RSS) {
1498 		if (pi == NULL)
1499 			atomic_add_long(&ether_rss_nopi, 1);
1500 		else
1501 			atomic_add_long(&ether_rss_nohash, 1);
1502 	}
1503 #endif
1504 
1505 	/*
1506 	 * Packet hash will be recalculated by software, so clear
1507 	 * the M_HASH and M_CKHASH flag set by the driver; the hash
1508 	 * value calculated by the hardware may not be exactly what
1509 	 * we want.
1510 	 */
1511 	m->m_flags &= ~(M_HASH | M_CKHASH);
1512 
1513 	if (!ether_vlancheck(&m)) {
1514 		KKASSERT(m == NULL);
1515 		logether(pkt_end, ifp);
1516 		return;
1517 	}
1518 
1519 	isr = ether_characterize(&m);
1520 	if (m == NULL) {
1521 		logether(pkt_end, ifp);
1522 		return;
1523 	}
1524 
1525 	/*
1526 	 * Finally dispatch it
1527 	 */
1528 	ether_dispatch(isr, m, cpuid);
1529 
1530 	logether(pkt_end, ifp);
1531 }
1532 
1533 static int
1534 ether_characterize(struct mbuf **m0)
1535 {
1536 	struct mbuf *m = *m0;
1537 	struct ether_header *eh;
1538 	uint16_t ether_type;
1539 	int isr;
1540 
1541 	eh = mtod(m, struct ether_header *);
1542 	ether_type = ntohs(eh->ether_type);
1543 
1544 	/*
1545 	 * Map ether type to netisr id.
1546 	 */
1547 	switch (ether_type) {
1548 #ifdef INET
1549 	case ETHERTYPE_IP:
1550 		isr = NETISR_IP;
1551 		break;
1552 
1553 	case ETHERTYPE_ARP:
1554 		isr = NETISR_ARP;
1555 		break;
1556 #endif
1557 
1558 #ifdef INET6
1559 	case ETHERTYPE_IPV6:
1560 		isr = NETISR_IPV6;
1561 		break;
1562 #endif
1563 
1564 #ifdef MPLS
1565 	case ETHERTYPE_MPLS:
1566 	case ETHERTYPE_MPLS_MCAST:
1567 		m->m_flags |= M_MPLSLABELED;
1568 		isr = NETISR_MPLS;
1569 		break;
1570 #endif
1571 
1572 	default:
1573 		/*
1574 		 * NETISR_MAX is an invalid value; it is chosen to let
1575 		 * netisr_characterize() know that we have no clear
1576 		 * idea where this packet should go.
1577 		 */
1578 		isr = NETISR_MAX;
1579 		break;
1580 	}
1581 
1582 	/*
1583 	 * Ask the isr to characterize the packet since we couldn't.
1584 	 * This is an attempt to optimally get us onto the correct protocol
1585 	 * thread.
1586 	 */
1587 	netisr_characterize(isr, &m, sizeof(struct ether_header));
1588 
1589 	*m0 = m;
1590 	return isr;
1591 }
1592 
1593 static void
1594 ether_demux_handler(netmsg_t nmsg)
1595 {
1596 	struct netmsg_packet *nmp = &nmsg->packet;	/* actual size */
1597 	struct ifnet *ifp;
1598 	struct mbuf *m;
1599 
1600 	m = nmp->nm_packet;
1601 	M_ASSERTPKTHDR(m);
1602 	ifp = m->m_pkthdr.rcvif;
1603 
1604 	ether_demux_oncpu(ifp, m);
1605 }
1606 
1607 void
1608 ether_demux(struct mbuf *m)
1609 {
1610 	struct netmsg_packet *pmsg;
1611 	int isr;
1612 
1613 	isr = ether_characterize(&m);
1614 	if (m == NULL)
1615 		return;
1616 
1617 	KKASSERT(m->m_flags & M_HASH);
1618 	pmsg = &m->m_hdr.mh_netmsg;
1619 	netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport,
1620 	    0, ether_demux_handler);
1621 	pmsg->nm_packet = m;
1622 	pmsg->base.lmsg.u.ms_result = isr;
1623 
1624 	lwkt_sendmsg(netisr_hashport(m->m_pkthdr.hash), &pmsg->base.lmsg);
1625 }
1626 
1627 u_char *
1628 kether_aton(const char *macstr, u_char *addr)
1629 {
1630         unsigned int o0, o1, o2, o3, o4, o5;
1631         int n;
1632 
1633         if (macstr == NULL || addr == NULL)
1634                 return NULL;
1635 
1636         n = ksscanf(macstr, "%x:%x:%x:%x:%x:%x", &o0, &o1, &o2,
1637             &o3, &o4, &o5);
1638         if (n != 6)
1639                 return NULL;
1640 
1641         addr[0] = o0;
1642         addr[1] = o1;
1643         addr[2] = o2;
1644         addr[3] = o3;
1645         addr[4] = o4;
1646         addr[5] = o5;
1647 
1648         return addr;
1649 }
1650 
1651 char *
1652 kether_ntoa(const u_char *addr, char *buf)
1653 {
1654         int len = ETHER_ADDRSTRLEN + 1;
1655         int n;
1656 
1657         n = ksnprintf(buf, len, "%02x:%02x:%02x:%02x:%02x:%02x", addr[0],
1658             addr[1], addr[2], addr[3], addr[4], addr[5]);
1659 
1660         if (n < 17)
1661                 return NULL;
1662 
1663         return buf;
1664 }
1665 
1666 MODULE_VERSION(ether, 1);
1667