Home | History | Annotate | Line # | Download | only in netinet
      1 /*	$NetBSD: if_arp.c,v 1.321 2026/07/23 21:49:19 riastradh Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1998, 2000, 2008 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Public Access Networks Corporation ("Panix").  It was developed under
      9  * contract to Panix by Eric Haszlakiewicz and Thor Lancelot Simon.
     10  *
     11  * Redistribution and use in source and binary forms, with or without
     12  * modification, are permitted provided that the following conditions
     13  * are met:
     14  * 1. Redistributions of source code must retain the above copyright
     15  *    notice, this list of conditions and the following disclaimer.
     16  * 2. Redistributions in binary form must reproduce the above copyright
     17  *    notice, this list of conditions and the following disclaimer in the
     18  *    documentation and/or other materials provided with the distribution.
     19  *
     20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     30  * POSSIBILITY OF SUCH DAMAGE.
     31  */
     32 
     33 /*
     34  * Copyright (c) 1982, 1986, 1988, 1993
     35  *	The Regents of the University of California.  All rights reserved.
     36  *
     37  * Redistribution and use in source and binary forms, with or without
     38  * modification, are permitted provided that the following conditions
     39  * are met:
     40  * 1. Redistributions of source code must retain the above copyright
     41  *    notice, this list of conditions and the following disclaimer.
     42  * 2. Redistributions in binary form must reproduce the above copyright
     43  *    notice, this list of conditions and the following disclaimer in the
     44  *    documentation and/or other materials provided with the distribution.
     45  * 3. Neither the name of the University nor the names of its contributors
     46  *    may be used to endorse or promote products derived from this software
     47  *    without specific prior written permission.
     48  *
     49  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     59  * SUCH DAMAGE.
     60  *
     61  *	@(#)if_ether.c	8.2 (Berkeley) 9/26/94
     62  */
     63 
     64 /*
     65  * Ethernet address resolution protocol.
     66  * TODO:
     67  *	add "inuse/lock" bit (or ref. count) along with valid bit
     68  */
     69 
     70 #include <sys/cdefs.h>
     71 __KERNEL_RCSID(0, "$NetBSD: if_arp.c,v 1.321 2026/07/23 21:49:19 riastradh Exp $");
     72 
     73 #ifdef _KERNEL_OPT
     74 #include "opt_ddb.h"
     75 #include "opt_inet.h"
     76 #include "opt_net_mpsafe.h"
     77 #endif
     78 
     79 #ifdef INET
     80 
     81 #include "arp.h"
     82 #include "bridge.h"
     83 
     84 #include <sys/param.h>
     85 #include <sys/systm.h>
     86 #include <sys/callout.h>
     87 #include <sys/kmem.h>
     88 #include <sys/mbuf.h>
     89 #include <sys/socket.h>
     90 #include <sys/time.h>
     91 #include <sys/timetc.h>
     92 #include <sys/kernel.h>
     93 #include <sys/errno.h>
     94 #include <sys/ioctl.h>
     95 #include <sys/syslog.h>
     96 #include <sys/proc.h>
     97 #include <sys/protosw.h>
     98 #include <sys/domain.h>
     99 #include <sys/sysctl.h>
    100 #include <sys/socketvar.h>
    101 #include <sys/percpu.h>
    102 #include <sys/cprng.h>
    103 #include <sys/kmem.h>
    104 
    105 #include <net/ethertypes.h>
    106 #include <net/if.h>
    107 #include <net/if_dl.h>
    108 #include <net/if_types.h>
    109 #include <net/if_ether.h>
    110 #include <net/if_llatbl.h>
    111 #include <net/nd.h>
    112 #include <net/route.h>
    113 #include <net/net_stats.h>
    114 
    115 #include <netinet/in.h>
    116 #include <netinet/in_systm.h>
    117 #include <netinet/in_var.h>
    118 #include <netinet/ip.h>
    119 #include <netinet/if_inarp.h>
    120 
    121 #include "arcnet.h"
    122 #if NARCNET > 0
    123 #include <net/if_arc.h>
    124 #endif
    125 #include "carp.h"
    126 #if NCARP > 0
    127 #include <netinet/ip_carp.h>
    128 #endif
    129 
    130 /*
    131  * ARP trailer negotiation.  Trailer protocol is not IP specific,
    132  * but ARP request/response use IP addresses.
    133  */
    134 #define ETHERTYPE_IPTRAILERS ETHERTYPE_TRAIL
    135 
    136 /* timers */
    137 static int arp_reachable = REACHABLE_TIME;
    138 static int arp_retrans = RETRANS_TIMER;
    139 static int arp_perform_nud = 1;
    140 
    141 static bool arp_nud_enabled(struct ifnet *);
    142 static unsigned int arp_llinfo_reachable(struct ifnet *);
    143 static unsigned int arp_llinfo_retrans(struct ifnet *);
    144 static union l3addr *arp_llinfo_holdsrc(struct llentry *, union l3addr *);
    145 static void arp_llinfo_output(struct ifnet *, const union l3addr *,
    146     const union l3addr *, const uint8_t *, const union l3addr *);
    147 static void arp_llinfo_missed(struct ifnet *, const union l3addr *,
    148     int16_t, struct mbuf *);
    149 static void arp_free(struct llentry *, int);
    150 
    151 static struct nd_domain arp_nd_domain = {
    152 	.nd_family = AF_INET,
    153 	.nd_delay = 5,		/* delay first probe time 5 second */
    154 	.nd_mmaxtries = 3,	/* maximum broadcast query */
    155 	.nd_umaxtries = 3,	/* maximum unicast query */
    156 	.nd_retransmultiple = BACKOFF_MULTIPLE,
    157 	.nd_maxretrans = MAX_RETRANS_TIMER,
    158 	.nd_maxnudhint = 0,	/* max # of subsequent upper layer hints */
    159 	.nd_maxqueuelen = 1,	/* max # of packets in unresolved ND entries */
    160 	.nd_gctimer = 24*60*60, /* stale neighbor GC timer duration */
    161 	.nd_nud_enabled = arp_nud_enabled,
    162 	.nd_reachable = arp_llinfo_reachable,
    163 	.nd_retrans = arp_llinfo_retrans,
    164 	.nd_holdsrc = arp_llinfo_holdsrc,
    165 	.nd_output = arp_llinfo_output,
    166 	.nd_missed = arp_llinfo_missed,
    167 	.nd_free = arp_free,
    168 };
    169 
    170 int ip_dad_count = PROBE_NUM;
    171 #ifdef ARP_DEBUG
    172 int arp_debug = 1;
    173 #else
    174 int arp_debug = 0;
    175 #endif
    176 
    177 static void arp_init(void);
    178 static void arp_dad_init(void);
    179 
    180 static void arprequest(struct ifnet *,
    181     const struct in_addr *, const struct in_addr *,
    182     const uint8_t *, const uint8_t *);
    183 static void arpannounce1(struct ifaddr *);
    184 static struct sockaddr *arp_setgate(struct rtentry *, struct sockaddr *,
    185     const struct sockaddr *);
    186 static struct llentry *arpcreate(struct ifnet *,
    187     const struct in_addr *, const struct sockaddr *, int);
    188 static void in_arpinput(struct mbuf *);
    189 static void in_revarpinput(struct mbuf *);
    190 static void revarprequest(struct ifnet *);
    191 
    192 static void arp_drainstub(void);
    193 
    194 struct dadq;
    195 static void arp_dad_timer(struct dadq *);
    196 static void arp_dad_start(struct ifaddr *);
    197 static void arp_dad_stop(struct ifaddr *);
    198 static void arp_dad_duplicated(struct ifaddr *, const struct sockaddr_dl *);
    199 
    200 #define	ARP_MAXQLEN	50
    201 pktqueue_t *		arp_pktq		__read_mostly;
    202 
    203 static int useloopback = 1;	/* use loopback interface for local traffic */
    204 
    205 static percpu_t *arpstat_percpu;
    206 
    207 #define	ARP_STAT_GETREF()	_NET_STAT_GETREF(arpstat_percpu)
    208 #define	ARP_STAT_PUTREF()	_NET_STAT_PUTREF(arpstat_percpu)
    209 
    210 #define	ARP_STATINC(x)		_NET_STATINC(arpstat_percpu, x)
    211 #define	ARP_STATADD(x, v)	_NET_STATADD(arpstat_percpu, x, v)
    212 
    213 /* revarp state */
    214 static struct in_addr myip, srv_ip;
    215 static int myip_initialized = 0;
    216 static int revarp_in_progress = 0;
    217 static struct ifnet *myip_ifp = NULL;
    218 
    219 static int arp_drainwanted;
    220 
    221 static int log_movements = 0;
    222 static int log_permanent_modify = 1;
    223 static int log_wrong_iface = 1;
    224 
    225 DOMAIN_DEFINE(arpdomain);	/* forward declare and add to link set */
    226 
    227 static void
    228 arp_fasttimo(void)
    229 {
    230 	if (arp_drainwanted) {
    231 		arp_drain();
    232 		arp_drainwanted = 0;
    233 	}
    234 }
    235 
    236 static const struct protosw arpsw[] = {
    237 	{
    238 		.pr_type = 0,
    239 		.pr_domain = &arpdomain,
    240 		.pr_protocol = 0,
    241 		.pr_flags = 0,
    242 		.pr_input = 0,
    243 		.pr_ctlinput = 0,
    244 		.pr_ctloutput = 0,
    245 		.pr_usrreqs = 0,
    246 		.pr_init = arp_init,
    247 		.pr_fasttimo = arp_fasttimo,
    248 		.pr_slowtimo = 0,
    249 		.pr_drain = arp_drainstub,
    250 	}
    251 };
    252 
    253 struct domain arpdomain = {
    254 	.dom_family = PF_ARP,
    255 	.dom_name = "arp",
    256 	.dom_protosw = arpsw,
    257 	.dom_protoswNPROTOSW = &arpsw[__arraycount(arpsw)],
    258 #ifdef MBUFTRACE
    259 	.dom_mowner = MOWNER_INIT("internet", "arp"),
    260 #endif
    261 };
    262 
    263 static void sysctl_net_inet_arp_setup(struct sysctllog **);
    264 
    265 void
    266 arp_init(void)
    267 {
    268 
    269 	arp_pktq = pktq_create(ARP_MAXQLEN, arpintr, NULL);
    270 	KASSERT(arp_pktq != NULL);
    271 
    272 	sysctl_net_inet_arp_setup(NULL);
    273 	arpstat_percpu = percpu_alloc(sizeof(uint64_t) * ARP_NSTATS);
    274 
    275 #ifdef MBUFTRACE
    276 	MOWNER_ATTACH(&arpdomain.dom_mowner);
    277 #endif
    278 
    279 	nd_attach_domain(&arp_nd_domain);
    280 	arp_dad_init();
    281 }
    282 
    283 static void
    284 arp_drainstub(void)
    285 {
    286 	arp_drainwanted = 1;
    287 }
    288 
    289 /*
    290  * ARP protocol drain routine.  Called when memory is in short supply.
    291  * Called at splvm();  don't acquire softnet_lock as can be called from
    292  * hardware interrupt handlers.
    293  */
    294 void
    295 arp_drain(void)
    296 {
    297 
    298 	lltable_drain(AF_INET);
    299 }
    300 
    301 /*
    302  * We set the gateway for RTF_CLONING routes to a "prototype"
    303  * link-layer sockaddr whose interface type (if_type) and interface
    304  * index (if_index) fields are prepared.
    305  */
    306 static struct sockaddr *
    307 arp_setgate(struct rtentry *rt, struct sockaddr *gate,
    308     const struct sockaddr *netmask)
    309 {
    310 	const struct ifnet *ifp = rt->rt_ifp;
    311 	uint8_t namelen = strlen(ifp->if_xname);
    312 	uint8_t addrlen = ifp->if_addrlen;
    313 
    314 	/*
    315 	 * XXX: If this is a manually added route to interface
    316 	 * such as older version of routed or gated might provide,
    317 	 * restore cloning bit.
    318 	 */
    319 	if ((rt->rt_flags & RTF_HOST) == 0 && netmask != NULL &&
    320 	    satocsin(netmask)->sin_addr.s_addr != 0xffffffff)
    321 		rt->rt_flags |= RTF_CONNECTED;
    322 
    323 	if ((rt->rt_flags & (RTF_CONNECTED | RTF_LOCAL))) {
    324 		union {
    325 			struct sockaddr sa;
    326 			struct sockaddr_storage ss;
    327 			struct sockaddr_dl sdl;
    328 		} u;
    329 		/*
    330 		 * Case 1: This route should come from a route to iface.
    331 		 */
    332 		sockaddr_dl_init(&u.sdl, sizeof(u.ss),
    333 		    ifp->if_index, ifp->if_type, NULL, namelen, NULL, addrlen);
    334 		rt_setgate(rt, &u.sa);
    335 		gate = rt->rt_gateway;
    336 	}
    337 	return gate;
    338 }
    339 
    340 /*
    341  * Parallel to llc_rtrequest.
    342  */
    343 void
    344 arp_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info)
    345 {
    346 	struct sockaddr *gate = rt->rt_gateway;
    347 	struct in_ifaddr *ia;
    348 	struct ifaddr *ifa;
    349 	struct ifnet *ifp = rt->rt_ifp;
    350 	int bound;
    351 	int s;
    352 
    353 	if (req == RTM_LLINFO_UPD) {
    354 		if ((ifa = info->rti_ifa) != NULL)
    355 			arpannounce1(ifa);
    356 		return;
    357 	}
    358 
    359 	if ((rt->rt_flags & RTF_GATEWAY) != 0) {
    360 		if (req != RTM_ADD)
    361 			return;
    362 
    363 		/*
    364 		 * linklayers with particular link MTU limitation.
    365 		 */
    366 		switch(ifp->if_type) {
    367 #if NARCNET > 0
    368 		case IFT_ARCNET:
    369 		    {
    370 			int arcipifmtu;
    371 
    372 			if (ifp->if_flags & IFF_LINK0)
    373 				arcipifmtu = arc_ipmtu;
    374 			else
    375 				arcipifmtu = ARCMTU;
    376 			if (ifp->if_mtu > arcipifmtu)
    377 				rt->rt_rmx.rmx_mtu = arcipifmtu;
    378 			break;
    379 		    }
    380 #endif
    381 		}
    382 		return;
    383 	}
    384 
    385 	switch (req) {
    386 	case RTM_SETGATE:
    387 		gate = arp_setgate(rt, gate, info->rti_info[RTAX_NETMASK]);
    388 		break;
    389 	case RTM_ADD:
    390 		gate = arp_setgate(rt, gate, info->rti_info[RTAX_NETMASK]);
    391 		if (gate == NULL) {
    392 			log(LOG_ERR, "%s: arp_setgate failed\n", __func__);
    393 			break;
    394 		}
    395 		if ((rt->rt_flags & RTF_CONNECTED) ||
    396 		    (rt->rt_flags & RTF_LOCAL)) {
    397 			/*
    398 			 * linklayers with particular link MTU limitation.
    399 			 */
    400 			switch (ifp->if_type) {
    401 #if NARCNET > 0
    402 			case IFT_ARCNET:
    403 			    {
    404 				int arcipifmtu;
    405 				if (ifp->if_flags & IFF_LINK0)
    406 					arcipifmtu = arc_ipmtu;
    407 				else
    408 					arcipifmtu = ARCMTU;
    409 
    410 				if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0 &&
    411 				    (rt->rt_rmx.rmx_mtu > arcipifmtu ||
    412 				     (rt->rt_rmx.rmx_mtu == 0 &&
    413 				      ifp->if_mtu > arcipifmtu)))
    414 					rt->rt_rmx.rmx_mtu = arcipifmtu;
    415 				break;
    416 			    }
    417 #endif
    418 			}
    419 			if (rt->rt_flags & RTF_CONNECTED)
    420 				break;
    421 		}
    422 
    423 		bound = curlwp_bind();
    424 		/* Announce a new entry if requested. */
    425 		if (rt->rt_flags & RTF_ANNOUNCE) {
    426 			struct psref psref;
    427 			ia = in_get_ia_on_iface_psref(
    428 			    satocsin(rt_getkey(rt))->sin_addr, ifp, &psref);
    429 			if (ia != NULL) {
    430 				arpannounce(ifp, &ia->ia_ifa,
    431 				    CLLADDR(satocsdl(gate)));
    432 				ia4_release(ia, &psref);
    433 			}
    434 		}
    435 
    436 		if (gate->sa_family != AF_LINK ||
    437 		    gate->sa_len < sockaddr_dl_measure(0, ifp->if_addrlen)) {
    438 			log(LOG_DEBUG, "%s: bad gateway value\n", __func__);
    439 			goto out;
    440 		}
    441 
    442 		satosdl(gate)->sdl_type = ifp->if_type;
    443 		satosdl(gate)->sdl_index = ifp->if_index;
    444 
    445 		/*
    446 		 * If the route is for a broadcast address mark it as such.
    447 		 * This way we can avoid an expensive call to in_broadcast()
    448 		 * in ip_output() most of the time (because the route passed
    449 		 * to ip_output() is almost always a host route).
    450 		 */
    451 		if (rt->rt_flags & RTF_HOST &&
    452 		    !(rt->rt_flags & RTF_BROADCAST) &&
    453 		    in_broadcast(satocsin(rt_getkey(rt))->sin_addr, rt->rt_ifp))
    454 			rt->rt_flags |= RTF_BROADCAST;
    455 		/* There is little point in resolving the broadcast address */
    456 		if (rt->rt_flags & RTF_BROADCAST)
    457 			goto out;
    458 
    459 		/*
    460 		 * When called from rt_ifa_addlocal, we cannot depend on that
    461 		 * the address (rt_getkey(rt)) exits in the address list of the
    462 		 * interface. So check RTF_LOCAL instead.
    463 		 */
    464 		if (rt->rt_flags & RTF_LOCAL) {
    465 			if (useloopback) {
    466 				rt->rt_ifp = lo0ifp;
    467 				rt->rt_rmx.rmx_mtu = 0;
    468 			}
    469 			goto out;
    470 		}
    471 
    472 		s = pserialize_read_enter();
    473 		ia = in_get_ia_on_iface(satocsin(rt_getkey(rt))->sin_addr, ifp);
    474 		if (ia == NULL) {
    475 			pserialize_read_exit(s);
    476 			goto out;
    477 		}
    478 
    479 		if (useloopback) {
    480 			rt->rt_ifp = lo0ifp;
    481 			rt->rt_rmx.rmx_mtu = 0;
    482 		}
    483 		rt->rt_flags |= RTF_LOCAL;
    484 
    485 		if (ISSET(info->rti_flags, RTF_DONTCHANGEIFA)) {
    486 			pserialize_read_exit(s);
    487 			goto out;
    488 		}
    489 		/*
    490 		 * make sure to set rt->rt_ifa to the interface
    491 		 * address we are using, otherwise we will have trouble
    492 		 * with source address selection.
    493 		 */
    494 		ifa = &ia->ia_ifa;
    495 		if (ifa != rt->rt_ifa)
    496 			/* Assume it doesn't sleep */
    497 			rt_replace_ifa(rt, ifa);
    498 		pserialize_read_exit(s);
    499 	out:
    500 		curlwp_bindx(bound);
    501 		break;
    502 	}
    503 }
    504 
    505 /*
    506  * Broadcast an ARP request. Caller specifies:
    507  *	- arp header source ip address
    508  *	- arp header target ip address
    509  *	- arp header source ethernet address
    510  */
    511 static void
    512 arprequest(struct ifnet *ifp,
    513     const struct in_addr *sip, const struct in_addr *tip,
    514     const uint8_t *saddr, const uint8_t *taddr)
    515 {
    516 	struct mbuf *m;
    517 	struct arphdr *ah;
    518 	struct sockaddr sa;
    519 	net_stat_ref_t arps;
    520 
    521 	KASSERT(sip != NULL);
    522 	KASSERT(tip != NULL);
    523 	KASSERT(saddr != NULL);
    524 
    525 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
    526 		return;
    527 	MCLAIM(m, &arpdomain.dom_mowner);
    528 	switch (ifp->if_type) {
    529 	case IFT_IEEE1394:
    530 		m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
    531 		    ifp->if_addrlen;
    532 		break;
    533 	default:
    534 		m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
    535 		    2 * ifp->if_addrlen;
    536 		break;
    537 	}
    538 	m->m_pkthdr.len = m->m_len;
    539 	m_align(m, m->m_len);
    540 	ah = mtod(m, struct arphdr *);
    541 	memset(ah, 0, m->m_len);
    542 	switch (ifp->if_type) {
    543 	case IFT_IEEE1394:	/* RFC2734 */
    544 		/* fill it now for ar_tpa computation */
    545 		ah->ar_hrd = htons(ARPHRD_IEEE1394);
    546 		break;
    547 	default:
    548 		/* ifp->if_output will fill ar_hrd */
    549 		break;
    550 	}
    551 	ah->ar_pro = htons(ETHERTYPE_IP);
    552 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
    553 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
    554 	ah->ar_op = htons(ARPOP_REQUEST);
    555 	memcpy(ar_sha(ah), saddr, ah->ar_hln);
    556 	if (taddr == NULL)
    557 		m->m_flags |= M_BCAST;
    558 	else
    559 		memcpy(ar_tha(ah), taddr, ah->ar_hln);
    560 	memcpy(ar_spa(ah), sip, ah->ar_pln);
    561 	memcpy(ar_tpa(ah), tip, ah->ar_pln);
    562 	sa.sa_family = AF_ARP;
    563 	sa.sa_len = 2;
    564 	arps = ARP_STAT_GETREF();
    565 	_NET_STATINC_REF(arps, ARP_STAT_SNDTOTAL);
    566 	_NET_STATINC_REF(arps, ARP_STAT_SENDREQUEST);
    567 	ARP_STAT_PUTREF();
    568 	if_output_lock(ifp, ifp, m, &sa, NULL);
    569 }
    570 
    571 void
    572 arpannounce(struct ifnet *ifp, struct ifaddr *ifa, const uint8_t *enaddr)
    573 {
    574 	struct in_ifaddr *ia = ifatoia(ifa);
    575 	struct in_addr *ip = &IA_SIN(ifa)->sin_addr;
    576 
    577 	if (ia->ia4_flags & (IN_IFF_NOTREADY | IN_IFF_DETACHED)) {
    578 		ARPLOG(LOG_DEBUG, "%s not ready\n", ARPLOGADDR(ip));
    579 		return;
    580 	}
    581 	arprequest(ifp, ip, ip, enaddr, NULL);
    582 }
    583 
    584 static void
    585 arpannounce1(struct ifaddr *ifa)
    586 {
    587 
    588 	arpannounce(ifa->ifa_ifp, ifa, CLLADDR(ifa->ifa_ifp->if_sadl));
    589 }
    590 
    591 /*
    592  * Resolve an IP address into an ethernet address.  If success, desten is
    593  * filled in. If there is no entry in arptab, set one up and broadcast a
    594  * request for the IP address. Hold onto this mbuf and resend it once the
    595  * address is finally resolved.
    596  *
    597  * A return value of 0 indicates that desten has been filled in and the packet
    598  * should be sent normally; a return value of EWOULDBLOCK indicates that the
    599  * packet has been held pending resolution. Any other value indicates an
    600  * error.
    601  */
    602 int
    603 arpresolve(struct ifnet *ifp, const struct rtentry *rt, struct mbuf *m,
    604     const struct sockaddr *dst, void *desten, size_t destlen)
    605 {
    606 	struct llentry *la;
    607 	const char *create_lookup;
    608 	int error;
    609 
    610 #if NCARP > 0
    611 	if (rt != NULL && rt->rt_ifp->if_type == IFT_CARP)
    612 		ifp = rt->rt_ifp;
    613 #endif
    614 
    615 	KASSERT(m != NULL);
    616 
    617 	la = arplookup(ifp, NULL, dst, 0);
    618 	if (la == NULL)
    619 		goto notfound;
    620 
    621 	if (la->la_flags & LLE_VALID && la->ln_state == ND_LLINFO_REACHABLE) {
    622 		KASSERT(destlen >= ifp->if_addrlen);
    623 		memcpy(desten, &la->ll_addr, ifp->if_addrlen);
    624 		LLE_RUNLOCK(la);
    625 		return 0;
    626 	}
    627 
    628 notfound:
    629 	if (ifp->if_flags & IFF_NOARP) {
    630 		if (la != NULL)
    631 			LLE_RUNLOCK(la);
    632 		error = ENOTSUP;
    633 		goto bad;
    634 	}
    635 
    636 	if (la == NULL) {
    637 		struct rtentry *_rt;
    638 
    639 		create_lookup = "create";
    640 		_rt = rtalloc1(dst, 0);
    641 		IF_AFDATA_WLOCK(ifp);
    642 		la = lla_create(LLTABLE(ifp), LLE_EXCLUSIVE, dst, _rt);
    643 		IF_AFDATA_WUNLOCK(ifp);
    644 		if (_rt != NULL)
    645 			rt_unref(_rt);
    646 		if (la == NULL)
    647 			ARP_STATINC(ARP_STAT_ALLOCFAIL);
    648 		else
    649 			la->ln_state = ND_LLINFO_NOSTATE;
    650 	} else if (LLE_TRY_UPGRADE(la) == 0) {
    651 		create_lookup = "lookup";
    652 		LLE_RUNLOCK(la);
    653 		IF_AFDATA_RLOCK(ifp);
    654 		la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
    655 		IF_AFDATA_RUNLOCK(ifp);
    656 	}
    657 
    658 	error = EINVAL;
    659 	if (la == NULL) {
    660 		log(LOG_DEBUG,
    661 		    "%s: failed to %s llentry for %s on %s\n",
    662 		    __func__, create_lookup, inet_ntoa(satocsin(dst)->sin_addr),
    663 		    ifp->if_xname);
    664 		goto bad;
    665 	}
    666 
    667 	error = nd_resolve(la, rt, m, desten, destlen);
    668 	return error;
    669 
    670 bad:
    671 	m_freem(m);
    672 	return error;
    673 }
    674 
    675 /*
    676  * Common length and type checks are done here,
    677  * then the protocol-specific routine is called.
    678  */
    679 void
    680 arpintr(void *arg __unused)
    681 {
    682 	struct mbuf *m;
    683 	struct arphdr *ar;
    684 	int s;
    685 	int arplen;
    686 	struct ifnet *rcvif;
    687 	bool badhrd;
    688 
    689 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
    690 	while ((m = pktq_dequeue(arp_pktq)) != NULL) {
    691 		if ((m->m_flags & M_PKTHDR) == 0)
    692 			panic("arpintr");
    693 
    694 		MCLAIM(m, &arpdomain.dom_mowner);
    695 		ARP_STATINC(ARP_STAT_RCVTOTAL);
    696 
    697 		if (__predict_false(m->m_len < sizeof(*ar))) {
    698 			if ((m = m_pullup(m, sizeof(*ar))) == NULL)
    699 				goto badlen;
    700 		}
    701 		ar = mtod(m, struct arphdr *);
    702 		KASSERT(ACCESSIBLE_POINTER(ar, struct arphdr));
    703 
    704 		rcvif = m_get_rcvif(m, &s);
    705 		if (__predict_false(rcvif == NULL)) {
    706 			ARP_STATINC(ARP_STAT_RCVNOINT);
    707 			goto free;
    708 		}
    709 
    710 		/*
    711 		 * We don't want non-IEEE1394 ARP packets on IEEE1394
    712 		 * interfaces, and vice versa. Our life depends on that.
    713 		 */
    714 		if (ntohs(ar->ar_hrd) == ARPHRD_IEEE1394)
    715 			badhrd = rcvif->if_type != IFT_IEEE1394;
    716 		else
    717 			badhrd = rcvif->if_type == IFT_IEEE1394;
    718 
    719 		m_put_rcvif(rcvif, &s);
    720 
    721 		if (badhrd) {
    722 			ARP_STATINC(ARP_STAT_RCVBADPROTO);
    723 			goto free;
    724 		}
    725 
    726 		arplen = sizeof(*ar) + 2 * ar->ar_hln + 2 * ar->ar_pln;
    727 		if (__predict_false(m->m_len < arplen)) {
    728 			if ((m = m_pullup(m, arplen)) == NULL)
    729 				goto badlen;
    730 			ar = mtod(m, struct arphdr *);
    731 			KASSERT(ACCESSIBLE_POINTER(ar, struct arphdr));
    732 		}
    733 
    734 		switch (ntohs(ar->ar_pro)) {
    735 		case ETHERTYPE_IP:
    736 		case ETHERTYPE_IPTRAILERS:
    737 			switch (ntohs(ar->ar_op)) {
    738 			case ARPOP_REQUEST:
    739 			case ARPOP_REPLY:
    740 				break;
    741 			default:
    742 				ARP_STATINC(ARP_STAT_RCVBADPROTO);
    743 				goto free;
    744 			}
    745 			in_arpinput(m);
    746 			continue;
    747 		default:
    748 			ARP_STATINC(ARP_STAT_RCVBADPROTO);
    749 			goto free;
    750 		}
    751 
    752 badlen:
    753 		ARP_STATINC(ARP_STAT_RCVBADLEN);
    754 free:
    755 		m_freem(m);
    756 	}
    757 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    758 	return; /* XXX gcc */
    759 }
    760 
    761 /*
    762  * ARP for Internet protocols on 10 Mb/s Ethernet. Algorithm is that given in
    763  * RFC 826. In addition, a sanity check is performed on the sender protocol
    764  * address, to catch impersonators.
    765  *
    766  * We no longer handle negotiations for use of trailer protocol: formerly, ARP
    767  * replied for protocol type ETHERTYPE_TRAIL sent along with IP replies if we
    768  * wanted trailers sent to us, and also sent them in response to IP replies.
    769  * This allowed either end to announce the desire to receive trailer packets.
    770  *
    771  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either, but
    772  * formerly didn't normally send requests.
    773  */
    774 static void
    775 in_arpinput(struct mbuf *m)
    776 {
    777 	struct arphdr *ah;
    778 	struct ifnet *ifp, *rcvif = NULL;
    779 	struct llentry *la = NULL;
    780 	struct in_ifaddr *ia = NULL;
    781 #if NBRIDGE > 0
    782 	struct in_ifaddr *bridge_ia = NULL;
    783 #endif
    784 #if NCARP > 0
    785 	uint32_t count = 0, index = 0;
    786 #endif
    787 	struct sockaddr sa;
    788 	struct in_addr isaddr, itaddr, myaddr;
    789 	int op, rt_cmd, new_state = 0;
    790 	void *tha;
    791 	net_stat_ref_t arps;
    792 	struct psref psref, psref_ia;
    793 	int s;
    794 	char ipbuf[INET_ADDRSTRLEN];
    795 	bool find_source, do_dad;
    796 
    797 	if (__predict_false(m_makewritable(&m, 0, m->m_pkthdr.len, M_DONTWAIT)))
    798 		goto out;
    799 	ah = mtod(m, struct arphdr *);
    800 	op = ntohs(ah->ar_op);
    801 	KASSERT(op == ARPOP_REPLY || op == ARPOP_REQUEST);
    802 
    803 	if (ah->ar_pln != sizeof(struct in_addr))
    804 		goto out;
    805 
    806 	/* RFC5227 2.4 says any of the host's own interface addresses
    807 	 * are not conflicting ARP packets. */
    808 	ifp = if_get_bylla(ar_sha(ah), ah->ar_hln, &psref);
    809 	if (ifp) {
    810 		/* it's from me, ignore it. */
    811 		if_put(ifp, &psref);
    812 		ARP_STATINC(ARP_STAT_RCVLOCALSHA);
    813 		goto out;
    814 	}
    815 
    816 	rcvif = ifp = m_get_rcvif_psref(m, &psref);
    817 	if (__predict_false(rcvif == NULL))
    818 		goto out;
    819 	if (rcvif->if_flags & IFF_NOARP)
    820 		goto out;
    821 
    822 	memcpy(&isaddr, ar_spa(ah), sizeof(isaddr));
    823 	memcpy(&itaddr, ar_tpa(ah), sizeof(itaddr));
    824 
    825 	if (m->m_flags & (M_BCAST|M_MCAST))
    826 		ARP_STATINC(ARP_STAT_RCVMCAST);
    827 
    828 	/*
    829 	 * Search for a matching interface address
    830 	 * or any address on the interface to use
    831 	 * as a dummy address in the rest of this function.
    832 	 *
    833 	 * First try and find the source address for early
    834 	 * duplicate address detection.
    835 	 */
    836 	if (in_nullhost(isaddr)) {
    837 		if (in_nullhost(itaddr)) /* very bogus ARP */
    838 			goto out;
    839 		find_source = false;
    840 		myaddr = itaddr;
    841 	} else {
    842 		find_source = true;
    843 		myaddr = isaddr;
    844 	}
    845 	s = pserialize_read_enter();
    846 again:
    847 	IN_ADDRHASH_READER_FOREACH(ia, myaddr.s_addr) {
    848 		if (!in_hosteq(ia->ia_addr.sin_addr, myaddr))
    849 			continue;
    850 #if NCARP > 0
    851 		if (ia->ia_ifp->if_type == IFT_CARP &&
    852 		    ((ia->ia_ifp->if_flags & (IFF_UP|IFF_RUNNING)) ==
    853 		    (IFF_UP|IFF_RUNNING))) {
    854 			index++;
    855 			/* XXX: ar_hln? */
    856 			if (ia->ia_ifp == rcvif && (ah->ar_hln >= 6) &&
    857 			    carp_iamatch(ia, ar_sha(ah),
    858 			    &count, index)) {
    859 				break;
    860 			}
    861 		} else
    862 #endif
    863 		if (ia->ia_ifp == rcvif)
    864 			break;
    865 #if NBRIDGE > 0
    866 		/*
    867 		 * If the interface we received the packet on
    868 		 * is part of a bridge, check to see if we need
    869 		 * to "bridge" the packet to ourselves at this
    870 		 * layer.  Note we still prefer a perfect match,
    871 		 * but allow this weaker match if necessary.
    872 		 */
    873 		if (rcvif->if_bridge != NULL &&
    874 		    rcvif->if_bridge == ia->ia_ifp->if_bridge)
    875 			bridge_ia = ia;
    876 #endif
    877 	}
    878 
    879 #if NBRIDGE > 0
    880 	if (ia == NULL && bridge_ia != NULL) {
    881 		ia = bridge_ia;
    882 		m_put_rcvif_psref(rcvif, &psref);
    883 		rcvif = NULL;
    884 		/* FIXME */
    885 		ifp = bridge_ia->ia_ifp;
    886 	}
    887 #endif
    888 
    889 	/* If we failed to find the source address then find
    890 	 * the target address. */
    891 	if (ia == NULL && find_source && !in_nullhost(itaddr)) {
    892 		find_source = false;
    893 		myaddr = itaddr;
    894 		goto again;
    895 	}
    896 
    897 	if (ia != NULL)
    898 		ia4_acquire(ia, &psref_ia);
    899 	pserialize_read_exit(s);
    900 
    901 	if (ah->ar_hln != ifp->if_addrlen) {
    902 		ARP_STATINC(ARP_STAT_RCVBADLEN);
    903 		log(LOG_WARNING,
    904 		    "arp from %s: addr len: new %d, i/f %d (ignored)\n",
    905 		    IN_PRINT(ipbuf, &isaddr), ah->ar_hln, ifp->if_addrlen);
    906 		goto out;
    907 	}
    908 
    909 	/* Only do DaD if we have a matching address. */
    910 	do_dad = (ia != NULL);
    911 
    912 	if (ia == NULL) {
    913 		ia = in_get_ia_on_iface_psref(isaddr, rcvif, &psref_ia);
    914 		if (ia == NULL) {
    915 			ia = in_get_ia_from_ifp_psref(ifp, &psref_ia);
    916 			if (ia == NULL) {
    917 				ARP_STATINC(ARP_STAT_RCVNOINT);
    918 				goto out;
    919 			}
    920 		}
    921 	}
    922 
    923 	myaddr = ia->ia_addr.sin_addr;
    924 
    925 	/* XXX checks for bridge case? */
    926 	if (!memcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
    927 		ARP_STATINC(ARP_STAT_RCVBCASTSHA);
    928 		log(LOG_ERR,
    929 		    "%s: arp: link address is broadcast for IP address %s!\n",
    930 		    ifp->if_xname, IN_PRINT(ipbuf, &isaddr));
    931 		goto out;
    932 	}
    933 
    934 	/*
    935 	 * If the source IP address is zero, this is an RFC 5227 ARP probe
    936 	 */
    937 	if (in_nullhost(isaddr))
    938 		ARP_STATINC(ARP_STAT_RCVZEROSPA);
    939 	else if (in_hosteq(isaddr, myaddr)) {
    940 		ARP_STATINC(ARP_STAT_RCVLOCALSPA);
    941 		/* This is the original behavior prior to supporting IPv4 DAD */
    942 		if (!ip_dad_enabled()) {
    943 			char llabuf[LLA_ADDRSTRLEN];
    944 			log(LOG_ERR,
    945 			    "duplicate IP address %s sent from link address %s\n",
    946 			    IN_PRINT(ipbuf, &isaddr),
    947 			    lla_snprintf(llabuf, sizeof(llabuf), ar_sha(ah),
    948 			                 ah->ar_hln));
    949 			itaddr = myaddr;
    950 			goto reply;
    951 		}
    952 	}
    953 
    954 	if (in_nullhost(itaddr))
    955 		ARP_STATINC(ARP_STAT_RCVZEROTPA);
    956 
    957 	/*
    958 	 * DAD check, RFC 5227.
    959 	 * ARP sender hardware address must match the interface
    960 	 * address of the interface sending the packet.
    961 	 * Collision on sender address is always a duplicate.
    962 	 * Collision on target address is only a duplicate
    963 	 * IF the sender address is the null host (ie a DAD probe)
    964 	 * AND the message was broadcast
    965 	 * AND our address is either tentative or duplicated
    966 	 * If it was unicast then it's a valid Unicast Poll from RFC 1122.
    967 	 */
    968 	if (ip_dad_enabled() && do_dad &&
    969 	    (in_hosteq(isaddr, myaddr) ||
    970 	    (in_nullhost(isaddr) && in_hosteq(itaddr, myaddr) &&
    971 	     m->m_flags & M_BCAST &&
    972 	     ia->ia4_flags & (IN_IFF_TENTATIVE | IN_IFF_DUPLICATED))))
    973 	{
    974 		struct m_tag *mtag;
    975 
    976 		mtag = m_tag_find(m, PACKET_TAG_ETHERNET_SRC);
    977 		if (mtag == NULL || (ah->ar_hln == ETHER_ADDR_LEN &&
    978 		    memcmp(mtag + 1, ar_sha(ah), ah->ar_hln) == 0)) {
    979 			struct sockaddr_dl sdl, *sdlp;
    980 
    981 			sdlp = sockaddr_dl_init(&sdl, sizeof(sdl),
    982 			    ifp->if_index, ifp->if_type,
    983 			    NULL, 0, ar_sha(ah), ah->ar_hln);
    984 			arp_dad_duplicated((struct ifaddr *)ia, sdlp);
    985 			goto out;
    986 		}
    987 	}
    988 
    989 	/*
    990 	 * If the target IP address is zero, ignore the packet.
    991 	 * This prevents the code below from trying to answer
    992 	 * when we are using IP address zero (booting).
    993 	 */
    994 	if (in_nullhost(itaddr))
    995 		goto out;
    996 
    997 	if (in_nullhost(isaddr))
    998 		goto reply;
    999 
   1000 	if (in_hosteq(itaddr, myaddr))
   1001 		la = arpcreate(ifp, &isaddr, NULL, 1);
   1002 	else
   1003 		la = arplookup(ifp, &isaddr, NULL, 1);
   1004 	if (la == NULL)
   1005 		goto reply;
   1006 
   1007 	if ((la->la_flags & LLE_VALID) &&
   1008 	    memcmp(ar_sha(ah), &la->ll_addr, ifp->if_addrlen))
   1009 	{
   1010 		char llabuf[LLA_ADDRSTRLEN], *llastr;
   1011 
   1012 		llastr = lla_snprintf(llabuf, sizeof(llabuf),
   1013 		    ar_sha(ah), ah->ar_hln);
   1014 
   1015 		if (la->la_flags & LLE_STATIC) {
   1016 			ARP_STATINC(ARP_STAT_RCVOVERPERM);
   1017 			if (!log_permanent_modify)
   1018 				goto out;
   1019 			log(LOG_INFO,
   1020 			    "%s tried to overwrite permanent arp info"
   1021 			    " for %s\n", llastr, IN_PRINT(ipbuf, &isaddr));
   1022 			goto out;
   1023 		} else if (la->lle_tbl->llt_ifp != ifp) {
   1024 			/* XXX should not happen? */
   1025 			ARP_STATINC(ARP_STAT_RCVOVERINT);
   1026 			if (!log_wrong_iface)
   1027 				goto out;
   1028 			log(LOG_INFO,
   1029 			    "%s on %s tried to overwrite "
   1030 			    "arp info for %s on %s\n",
   1031 			    llastr,
   1032 			    ifp->if_xname, IN_PRINT(ipbuf, &isaddr),
   1033 			    la->lle_tbl->llt_ifp->if_xname);
   1034 				goto out;
   1035 		} else {
   1036 			ARP_STATINC(ARP_STAT_RCVOVER);
   1037 			if (log_movements)
   1038 				log(LOG_INFO, "arp info overwritten "
   1039 				    "for %s by %s\n",
   1040 				    IN_PRINT(ipbuf, &isaddr), llastr);
   1041 		}
   1042 		rt_cmd = RTM_CHANGE;
   1043 		new_state = ND_LLINFO_STALE;
   1044 	} else {
   1045 		if (op == ARPOP_REPLY && in_hosteq(itaddr, myaddr)) {
   1046 			/* This was a solicited ARP reply. */
   1047 			la->ln_byhint = 0;
   1048 			new_state = ND_LLINFO_REACHABLE;
   1049 		} else if (op == ARPOP_REQUEST &&
   1050 		           (la->ln_state == ND_LLINFO_NOSTATE ||
   1051 			    la->ln_state == ND_LLINFO_INCOMPLETE)) {
   1052 			/*
   1053 			 * If an ARP request comes but there is no entry
   1054 			 * and a new one has been created or an entry exists
   1055 			 * but incomplete, make it stale to allow to send
   1056 			 * packets to the requester without an ARP resolution.
   1057 			 */
   1058 			la->ln_byhint = 0;
   1059 			new_state = ND_LLINFO_STALE;
   1060 		}
   1061 		rt_cmd = la->la_flags & LLE_VALID ? 0 : RTM_ADD;
   1062 	}
   1063 
   1064 	KASSERT(ifp->if_sadl->sdl_alen == ifp->if_addrlen);
   1065 
   1066 	KASSERT(sizeof(la->ll_addr) >= ifp->if_addrlen);
   1067 	memcpy(&la->ll_addr, ar_sha(ah), ifp->if_addrlen);
   1068 	la->la_flags |= LLE_VALID;
   1069 	la->la_flags &= ~LLE_UNRESOLVED;
   1070 	la->ln_asked = 0;
   1071 	if (new_state != 0) {
   1072 		la->ln_state = new_state;
   1073 
   1074 		if (new_state != ND_LLINFO_REACHABLE ||
   1075 		    !(la->la_flags & LLE_STATIC))
   1076 		{
   1077 			int timer = ND_TIMER_GC;
   1078 
   1079 			if (new_state == ND_LLINFO_REACHABLE)
   1080 				timer = ND_TIMER_REACHABLE;
   1081 			nd_set_timer(la, timer);
   1082 		}
   1083 	}
   1084 
   1085 	if (rt_cmd != 0) {
   1086 		struct sockaddr_in sin;
   1087 
   1088 		sockaddr_in_init(&sin, &la->r_l3addr.addr4, 0);
   1089 		rt_clonedmsg(rt_cmd, NULL, sintosa(&sin), ar_sha(ah), ifp);
   1090 	}
   1091 
   1092 	if (la->la_hold != NULL) {
   1093 		int n = la->la_numheld;
   1094 		struct mbuf *m_hold, *m_hold_next;
   1095 		struct sockaddr_in sin;
   1096 
   1097 		sockaddr_in_init(&sin, &la->r_l3addr.addr4, 0);
   1098 
   1099 		m_hold = la->la_hold;
   1100 		la->la_hold = NULL;
   1101 		la->la_numheld = 0;
   1102 		/*
   1103 		 * We have to unlock here because if_output would call
   1104 		 * arpresolve
   1105 		 */
   1106 		LLE_WUNLOCK(la);
   1107 		ARP_STATADD(ARP_STAT_DFRSENT, n);
   1108 		ARP_STATADD(ARP_STAT_DFRTOTAL, n);
   1109 		for (; m_hold != NULL; m_hold = m_hold_next) {
   1110 			m_hold_next = m_hold->m_nextpkt;
   1111 			m_hold->m_nextpkt = NULL;
   1112 			if_output_lock(ifp, ifp, m_hold, sintosa(&sin), NULL);
   1113 		}
   1114 	} else
   1115 		LLE_WUNLOCK(la);
   1116 	la = NULL;
   1117 
   1118 reply:
   1119 	if (la != NULL) {
   1120 		LLE_WUNLOCK(la);
   1121 		la = NULL;
   1122 	}
   1123 	if (op != ARPOP_REQUEST) {
   1124 		KASSERT(op == ARPOP_REPLY);
   1125 		ARP_STATINC(ARP_STAT_RCVREPLY);
   1126 		goto out;
   1127 	}
   1128 	ARP_STATINC(ARP_STAT_RCVREQUEST);
   1129 	if (in_hosteq(itaddr, myaddr)) {
   1130 		/* If our address is unusable, don't reply */
   1131 		if (ia->ia4_flags & (IN_IFF_NOTREADY | IN_IFF_DETACHED))
   1132 			goto out;
   1133 		/* I am the target */
   1134 		tha = ar_tha(ah);
   1135 		if (tha)
   1136 			memcpy(tha, ar_sha(ah), ah->ar_hln);
   1137 		memcpy(ar_sha(ah), CLLADDR(ifp->if_sadl), ah->ar_hln);
   1138 	} else {
   1139 		/* Proxy ARP */
   1140 		struct llentry *lle = NULL;
   1141 		struct sockaddr_in sin;
   1142 
   1143 #if NCARP > 0
   1144 		if (ifp->if_type == IFT_CARP) {
   1145 			struct ifnet *_rcvif = m_get_rcvif(m, &s);
   1146 			int iftype = 0;
   1147 			if (__predict_true(_rcvif != NULL))
   1148 				iftype = _rcvif->if_type;
   1149 			m_put_rcvif(_rcvif, &s);
   1150 			if (iftype != IFT_CARP)
   1151 				goto out;
   1152 		}
   1153 #endif
   1154 
   1155 		tha = ar_tha(ah);
   1156 
   1157 		sockaddr_in_init(&sin, &itaddr, 0);
   1158 
   1159 		IF_AFDATA_RLOCK(ifp);
   1160 		lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
   1161 		IF_AFDATA_RUNLOCK(ifp);
   1162 
   1163 		if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
   1164 			if (tha)
   1165 				memcpy(tha, ar_sha(ah), ah->ar_hln);
   1166 			memcpy(ar_sha(ah), &lle->ll_addr, ah->ar_hln);
   1167 			LLE_RUNLOCK(lle);
   1168 		} else {
   1169 			if (lle != NULL)
   1170 				LLE_RUNLOCK(lle);
   1171 			goto out;
   1172 		}
   1173 	}
   1174 	ia4_release(ia, &psref_ia);
   1175 
   1176 	/*
   1177 	 * XXX XXX: Here we're recycling the mbuf. But the mbuf could have
   1178 	 * other mbufs in its chain, and just overwriting m->m_pkthdr.len
   1179 	 * would be wrong in this case (the length becomes smaller than the
   1180 	 * real chain size).
   1181 	 *
   1182 	 * This can theoretically cause bugs in the lower layers (drivers,
   1183 	 * and L2encap), in some corner cases.
   1184 	 */
   1185 	memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
   1186 	memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
   1187 	ah->ar_op = htons(ARPOP_REPLY);
   1188 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
   1189 	switch (ifp->if_type) {
   1190 	case IFT_IEEE1394:
   1191 		/* ieee1394 arp reply is broadcast */
   1192 		m->m_flags &= ~M_MCAST;
   1193 		m->m_flags |= M_BCAST;
   1194 		m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + ah->ar_hln;
   1195 		break;
   1196 	default:
   1197 		m->m_flags &= ~(M_BCAST|M_MCAST); /* never reply by broadcast */
   1198 		m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
   1199 		break;
   1200 	}
   1201 	m->m_pkthdr.len = m->m_len;
   1202 	m->m_pkthdr.csum_flags = 0; /* Clear any in-bound checksum flags. */
   1203 	sa.sa_family = AF_ARP;
   1204 	sa.sa_len = 2;
   1205 	arps = ARP_STAT_GETREF();
   1206 	_NET_STATINC_REF(arps, ARP_STAT_SNDTOTAL);
   1207 	_NET_STATINC_REF(arps, ARP_STAT_SNDREPLY);
   1208 	ARP_STAT_PUTREF();
   1209 	if_output_lock(ifp, ifp, m, &sa, NULL);
   1210 	if (rcvif != NULL)
   1211 		m_put_rcvif_psref(rcvif, &psref);
   1212 	return;
   1213 
   1214 out:
   1215 	if (la != NULL)
   1216 		LLE_WUNLOCK(la);
   1217 	if (ia != NULL)
   1218 		ia4_release(ia, &psref_ia);
   1219 	if (rcvif != NULL)
   1220 		m_put_rcvif_psref(rcvif, &psref);
   1221 	m_freem(m);
   1222 }
   1223 
   1224 /*
   1225  * Lookup or a new address in arptab.
   1226  */
   1227 struct llentry *
   1228 arplookup(struct ifnet *ifp, const struct in_addr *addr,
   1229     const struct sockaddr *sa, int wlock)
   1230 {
   1231 	struct sockaddr_in sin;
   1232 	struct llentry *la;
   1233 	int flags = wlock ? LLE_EXCLUSIVE : 0;
   1234 
   1235 	if (sa == NULL) {
   1236 		KASSERT(addr != NULL);
   1237 		sockaddr_in_init(&sin, addr, 0);
   1238 		sa = sintocsa(&sin);
   1239 	}
   1240 
   1241 	IF_AFDATA_RLOCK(ifp);
   1242 	la = lla_lookup(LLTABLE(ifp), flags, sa);
   1243 	IF_AFDATA_RUNLOCK(ifp);
   1244 
   1245 	return la;
   1246 }
   1247 
   1248 static struct llentry *
   1249 arpcreate(struct ifnet *ifp, const struct in_addr *addr,
   1250     const struct sockaddr *sa, int wlock)
   1251 {
   1252 	struct sockaddr_in sin;
   1253 	struct llentry *la;
   1254 	int flags = wlock ? LLE_EXCLUSIVE : 0;
   1255 
   1256 	if (sa == NULL) {
   1257 		KASSERT(addr != NULL);
   1258 		sockaddr_in_init(&sin, addr, 0);
   1259 		sa = sintocsa(&sin);
   1260 	}
   1261 
   1262 	la = arplookup(ifp, addr, sa, wlock);
   1263 
   1264 	if (la == NULL) {
   1265 		struct rtentry *rt;
   1266 
   1267 		rt = rtalloc1(sa, 0);
   1268 		IF_AFDATA_WLOCK(ifp);
   1269 		la = lla_create(LLTABLE(ifp), flags, sa, rt);
   1270 		IF_AFDATA_WUNLOCK(ifp);
   1271 		if (rt != NULL)
   1272 			rt_unref(rt);
   1273 
   1274 		if (la != NULL)
   1275 			la->ln_state = ND_LLINFO_NOSTATE;
   1276 	}
   1277 
   1278 	return la;
   1279 }
   1280 
   1281 int
   1282 arpioctl(u_long cmd, void *data)
   1283 {
   1284 
   1285 	return EOPNOTSUPP;
   1286 }
   1287 
   1288 void
   1289 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
   1290 {
   1291 	struct in_ifaddr *ia = (struct in_ifaddr *)ifa;
   1292 
   1293 	ifa->ifa_rtrequest = arp_rtrequest;
   1294 	ifa->ifa_flags |= RTF_CONNECTED;
   1295 
   1296 	/* ARP will handle DAD for this address. */
   1297 	if (in_nullhost(IA_SIN(ifa)->sin_addr)) {
   1298 		if (ia->ia_dad_stop != NULL)	/* safety */
   1299 			ia->ia_dad_stop(ifa);
   1300 		ia->ia_dad_start = NULL;
   1301 		ia->ia_dad_stop = NULL;
   1302 		ia->ia4_flags &= ~IN_IFF_TENTATIVE;
   1303 	} else {
   1304 		ia->ia_dad_start = arp_dad_start;
   1305 		ia->ia_dad_stop = arp_dad_stop;
   1306 		if (ia->ia4_flags & IN_IFF_TRYTENTATIVE && ip_dad_enabled())
   1307 			ia->ia4_flags |= IN_IFF_TENTATIVE;
   1308 		else
   1309 			arpannounce1(ifa);
   1310 	}
   1311 }
   1312 
   1313 static bool
   1314 arp_nud_enabled(__unused struct ifnet *ifp)
   1315 {
   1316 
   1317 	return arp_perform_nud != 0;
   1318 }
   1319 
   1320 static unsigned int
   1321 arp_llinfo_reachable(__unused struct ifnet *ifp)
   1322 {
   1323 
   1324 	return arp_reachable;
   1325 }
   1326 
   1327 static unsigned int
   1328 arp_llinfo_retrans(__unused struct ifnet *ifp)
   1329 {
   1330 
   1331 	return arp_retrans;
   1332 }
   1333 
   1334 /*
   1335  * Gets source address of the first packet in hold queue
   1336  * and stores it in @src.
   1337  * Returns pointer to @src (if hold queue is not empty) or NULL.
   1338  */
   1339 static union l3addr *
   1340 arp_llinfo_holdsrc(struct llentry *ln, union l3addr *src)
   1341 {
   1342 	struct ip *ip;
   1343 
   1344 	if (ln == NULL || ln->ln_hold == NULL)
   1345 		return NULL;
   1346 
   1347 	/*
   1348 	 * assuming every packet in ln_hold has the same IP header
   1349 	 */
   1350 	ip = mtod(ln->ln_hold, struct ip *);
   1351 	/* XXX pullup? */
   1352 	if (sizeof(*ip) < ln->ln_hold->m_len)
   1353 		src->addr4 = ip->ip_src;
   1354 	else
   1355 		src = NULL;
   1356 
   1357 	return src;
   1358 }
   1359 
   1360 static void
   1361 arp_llinfo_output(struct ifnet *ifp, __unused const union l3addr *daddr,
   1362     const union l3addr *taddr, const uint8_t *tlladdr,
   1363     const union l3addr *hsrc)
   1364 {
   1365 	struct in_addr tip = taddr->addr4, sip = zeroin_addr;
   1366 	const uint8_t *slladdr = CLLADDR(ifp->if_sadl);
   1367 
   1368 	if (hsrc != NULL) {
   1369 		struct in_ifaddr *ia;
   1370 		struct psref psref;
   1371 
   1372 		ia = in_get_ia_on_iface_psref(hsrc->addr4, ifp, &psref);
   1373 		if (ia != NULL) {
   1374 			sip = hsrc->addr4;
   1375 			ia4_release(ia, &psref);
   1376 		}
   1377 	}
   1378 
   1379 	if (sip.s_addr == INADDR_ANY) {
   1380 		struct sockaddr_in dst;
   1381 		struct rtentry *rt;
   1382 
   1383 		sockaddr_in_init(&dst, &tip, 0);
   1384 		rt = rtalloc1(sintosa(&dst), 0);
   1385 		if (rt != NULL) {
   1386 			if (rt->rt_ifp == ifp &&
   1387 			    rt->rt_ifa != NULL &&
   1388 			    rt->rt_ifa->ifa_addr->sa_family == AF_INET)
   1389 				sip = satosin(rt->rt_ifa->ifa_addr)->sin_addr;
   1390 			rt_unref(rt);
   1391 		}
   1392 		if (sip.s_addr == INADDR_ANY) {
   1393 			char ipbuf[INET_ADDRSTRLEN];
   1394 
   1395 			log(LOG_DEBUG, "%s: source can't be "
   1396 			    "determined: dst=%s\n", __func__,
   1397 			    IN_PRINT(ipbuf, &tip));
   1398 			return;
   1399 		}
   1400 	}
   1401 
   1402 	arprequest(ifp, &sip, &tip, slladdr, tlladdr);
   1403 }
   1404 
   1405 
   1406 static void
   1407 arp_llinfo_missed(struct ifnet *ifp, const union l3addr *taddr,
   1408     __unused int16_t type, struct mbuf *m)
   1409 {
   1410 	struct in_addr mdaddr = zeroin_addr;
   1411 	struct sockaddr_in dsin, tsin;
   1412 	struct sockaddr *sa;
   1413 
   1414 	if (m != NULL) {
   1415 		struct ip *ip = mtod(m, struct ip *);
   1416 
   1417 		if (sizeof(*ip) < m->m_len)
   1418 			mdaddr = ip->ip_src;
   1419 
   1420 		/* ip_input() will send ICMP_UNREACH_HOST, not us. */
   1421 		m_freem(m);
   1422 	}
   1423 
   1424 	if (mdaddr.s_addr != INADDR_ANY) {
   1425 		sockaddr_in_init(&dsin, &mdaddr, 0);
   1426 		sa = sintosa(&dsin);
   1427 	} else
   1428 		sa = NULL;
   1429 
   1430 	sockaddr_in_init(&tsin, &taddr->addr4, 0);
   1431 	rt_clonedmsg(RTM_MISS, sa, sintosa(&tsin), NULL, ifp);
   1432 }
   1433 
   1434 static void
   1435 arp_free(struct llentry *ln, int gc)
   1436 {
   1437 	struct ifnet *ifp;
   1438 
   1439 	KASSERT(ln != NULL);
   1440 	LLE_WLOCK_ASSERT(ln);
   1441 
   1442 	ifp = ln->lle_tbl->llt_ifp;
   1443 
   1444 	if (ln->la_flags & LLE_VALID || gc) {
   1445 		struct sockaddr_in sin;
   1446 		const char *lladdr;
   1447 
   1448 		sockaddr_in_init(&sin, &ln->r_l3addr.addr4, 0);
   1449 		lladdr = ln->la_flags & LLE_VALID ?
   1450 		    (const char *)&ln->ll_addr : NULL;
   1451 		rt_clonedmsg(RTM_DELETE, NULL, sintosa(&sin), lladdr, ifp);
   1452 	}
   1453 
   1454 	/*
   1455 	 * Save to unlock. We still hold an extra reference and will not
   1456 	 * free(9) in llentry_free() if someone else holds one as well.
   1457 	 */
   1458 	LLE_WUNLOCK(ln);
   1459 	IF_AFDATA_LOCK(ifp);
   1460 	LLE_WLOCK(ln);
   1461 
   1462 	lltable_free_entry(LLTABLE(ifp), ln);
   1463 
   1464 	IF_AFDATA_UNLOCK(ifp);
   1465 }
   1466 
   1467 /*
   1468  * Upper-layer reachability hint for Neighbor Unreachability Detection.
   1469  *
   1470  * XXX cost-effective methods?
   1471  */
   1472 void
   1473 arp_nud_hint(struct rtentry *rt)
   1474 {
   1475 	struct llentry *ln;
   1476 	struct ifnet *ifp;
   1477 
   1478 	if (rt == NULL)
   1479 		return;
   1480 
   1481 	ifp = rt->rt_ifp;
   1482 	ln = arplookup(ifp, NULL, rt_getkey(rt), 1);
   1483 	nd_nud_hint(ln);
   1484 }
   1485 
   1486 TAILQ_HEAD(dadq_head, dadq);
   1487 struct dadq {
   1488 	TAILQ_ENTRY(dadq) dad_list;
   1489 	struct ifaddr *dad_ifa;
   1490 	int dad_count;		/* max ARP to send */
   1491 	int dad_arp_tcount;	/* # of trials to send ARP */
   1492 	int dad_arp_ocount;	/* ARP sent so far */
   1493 	int dad_arp_announce;	/* max ARP announcements */
   1494 	int dad_arp_acount;	/* # of announcements */
   1495 	struct callout dad_timer_ch;
   1496 };
   1497 
   1498 static struct dadq_head dadq;
   1499 static int dad_maxtry = 15;     /* max # of *tries* to transmit DAD packet */
   1500 static kmutex_t arp_dad_lock;
   1501 
   1502 static void
   1503 arp_dad_init(void)
   1504 {
   1505 
   1506 	TAILQ_INIT(&dadq);
   1507 	mutex_init(&arp_dad_lock, MUTEX_DEFAULT, IPL_NONE);
   1508 }
   1509 
   1510 static struct dadq *
   1511 arp_dad_find(struct ifaddr *ifa)
   1512 {
   1513 	struct dadq *dp;
   1514 
   1515 	KASSERT(mutex_owned(&arp_dad_lock));
   1516 
   1517 	TAILQ_FOREACH(dp, &dadq, dad_list) {
   1518 		if (dp->dad_ifa == ifa)
   1519 			return dp;
   1520 	}
   1521 	return NULL;
   1522 }
   1523 
   1524 static void
   1525 arp_dad_starttimer(struct dadq *dp, int ticks)
   1526 {
   1527 
   1528 	callout_reset(&dp->dad_timer_ch, ticks,
   1529 	    (void (*)(void *))arp_dad_timer, dp);
   1530 }
   1531 
   1532 static void
   1533 arp_dad_stoptimer(struct dadq *dp)
   1534 {
   1535 
   1536 	KASSERT(mutex_owned(&arp_dad_lock));
   1537 
   1538 	TAILQ_REMOVE(&dadq, dp, dad_list);
   1539 	/* Tell the timer that dp is being destroyed. */
   1540 	dp->dad_ifa = NULL;
   1541 	callout_halt(&dp->dad_timer_ch, &arp_dad_lock);
   1542 }
   1543 
   1544 static void
   1545 arp_dad_destroytimer(struct dadq *dp)
   1546 {
   1547 
   1548 	callout_destroy(&dp->dad_timer_ch);
   1549 	KASSERT(dp->dad_ifa == NULL);
   1550 	kmem_intr_free(dp, sizeof(*dp));
   1551 }
   1552 
   1553 static void
   1554 arp_dad_output(struct dadq *dp, struct ifaddr *ifa)
   1555 {
   1556 	struct in_ifaddr *ia = (struct in_ifaddr *)ifa;
   1557 	struct ifnet *ifp = ifa->ifa_ifp;
   1558 	struct in_addr sip;
   1559 
   1560 	dp->dad_arp_tcount++;
   1561 	if ((ifp->if_flags & IFF_UP) == 0)
   1562 		return;
   1563 	if ((ifp->if_flags & IFF_RUNNING) == 0)
   1564 		return;
   1565 
   1566 	dp->dad_arp_tcount = 0;
   1567 	dp->dad_arp_ocount++;
   1568 
   1569 	memset(&sip, 0, sizeof(sip));
   1570 	arprequest(ifa->ifa_ifp, &sip, &ia->ia_addr.sin_addr,
   1571 	    CLLADDR(ifa->ifa_ifp->if_sadl), NULL);
   1572 }
   1573 
   1574 /*
   1575  * Start Duplicate Address Detection (DAD) for specified interface address.
   1576  */
   1577 static void
   1578 arp_dad_start(struct ifaddr *ifa)
   1579 {
   1580 	struct in_ifaddr *ia = (struct in_ifaddr *)ifa;
   1581 	struct dadq *dp;
   1582 	char ipbuf[INET_ADDRSTRLEN];
   1583 
   1584 	/*
   1585 	 * If we don't need DAD, don't do it.
   1586 	 * - DAD is disabled
   1587 	 */
   1588 	if (!(ia->ia4_flags & IN_IFF_TENTATIVE)) {
   1589 		log(LOG_DEBUG,
   1590 		    "%s: called with non-tentative address %s(%s)\n", __func__,
   1591 		    IN_PRINT(ipbuf, &ia->ia_addr.sin_addr),
   1592 		    ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
   1593 		return;
   1594 	}
   1595 	if (!ip_dad_enabled()) {
   1596 		ia->ia4_flags &= ~IN_IFF_TENTATIVE;
   1597 		rt_addrmsg(RTM_NEWADDR, ifa);
   1598 		arpannounce1(ifa);
   1599 		return;
   1600 	}
   1601 	KASSERT(ifa->ifa_ifp != NULL);
   1602 	if (!(ifa->ifa_ifp->if_flags & IFF_UP))
   1603 		return;
   1604 
   1605 	dp = kmem_intr_alloc(sizeof(*dp), KM_NOSLEEP);
   1606 
   1607 	mutex_enter(&arp_dad_lock);
   1608 	if (arp_dad_find(ifa) != NULL) {
   1609 		mutex_exit(&arp_dad_lock);
   1610 		/* DAD already in progress */
   1611 		if (dp != NULL)
   1612 			kmem_intr_free(dp, sizeof(*dp));
   1613 		return;
   1614 	}
   1615 
   1616 	if (dp == NULL) {
   1617 		mutex_exit(&arp_dad_lock);
   1618 		log(LOG_ERR, "%s: memory allocation failed for %s(%s)\n",
   1619 		    __func__, IN_PRINT(ipbuf, &ia->ia_addr.sin_addr),
   1620 		    ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
   1621 		return;
   1622 	}
   1623 
   1624 	/*
   1625 	 * Send ARP packet for DAD, ip_dad_count times.
   1626 	 * Note that we must delay the first transmission.
   1627 	 */
   1628 	callout_init(&dp->dad_timer_ch, CALLOUT_MPSAFE);
   1629 	dp->dad_ifa = ifa;
   1630 	ifaref(ifa);	/* just for safety */
   1631 	dp->dad_count = ip_dad_count;
   1632 	dp->dad_arp_announce = 0; /* Will be set when starting to announce */
   1633 	dp->dad_arp_acount = dp->dad_arp_ocount = dp->dad_arp_tcount = 0;
   1634 	TAILQ_INSERT_TAIL(&dadq, (struct dadq *)dp, dad_list);
   1635 
   1636 	ARPLOG(LOG_DEBUG, "%s: starting DAD for %s\n", if_name(ifa->ifa_ifp),
   1637 	    ARPLOGADDR(&ia->ia_addr.sin_addr));
   1638 
   1639 	arp_dad_starttimer(dp, cprng_fast32() % (PROBE_WAIT * hz));
   1640 
   1641 	mutex_exit(&arp_dad_lock);
   1642 }
   1643 
   1644 /*
   1645  * terminate DAD unconditionally.  used for address removals.
   1646  */
   1647 static void
   1648 arp_dad_stop(struct ifaddr *ifa)
   1649 {
   1650 	struct dadq *dp;
   1651 
   1652 	mutex_enter(&arp_dad_lock);
   1653 	dp = arp_dad_find(ifa);
   1654 	if (dp == NULL) {
   1655 		mutex_exit(&arp_dad_lock);
   1656 		/* DAD wasn't started yet */
   1657 		return;
   1658 	}
   1659 
   1660 	arp_dad_stoptimer(dp);
   1661 
   1662 	mutex_exit(&arp_dad_lock);
   1663 
   1664 	arp_dad_destroytimer(dp);
   1665 	ifafree(ifa);
   1666 }
   1667 
   1668 static void
   1669 arp_dad_timer(struct dadq *dp)
   1670 {
   1671 	struct ifaddr *ifa;
   1672 	struct in_ifaddr *ia;
   1673 	char ipbuf[INET_ADDRSTRLEN];
   1674 	bool need_free = false;
   1675 
   1676 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
   1677 	mutex_enter(&arp_dad_lock);
   1678 
   1679 	ifa = dp->dad_ifa;
   1680 	if (ifa == NULL) {
   1681 		/* dp is being destroyed by someone.  Do nothing. */
   1682 		goto done;
   1683 	}
   1684 
   1685 	ia = (struct in_ifaddr *)ifa;
   1686 	if (ia->ia4_flags & IN_IFF_DUPLICATED) {
   1687 		log(LOG_ERR, "%s: called with duplicate address %s(%s)\n",
   1688 		    __func__, IN_PRINT(ipbuf, &ia->ia_addr.sin_addr),
   1689 		    ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
   1690 		goto done;
   1691 	}
   1692 	if ((ia->ia4_flags & IN_IFF_TENTATIVE) == 0 && dp->dad_arp_acount == 0)
   1693 	{
   1694 		log(LOG_ERR, "%s: called with non-tentative address %s(%s)\n",
   1695 		    __func__, IN_PRINT(ipbuf, &ia->ia_addr.sin_addr),
   1696 		    ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
   1697 		goto done;
   1698 	}
   1699 
   1700 	/* timeouted with IFF_{RUNNING,UP} check */
   1701 	if (dp->dad_arp_tcount > dad_maxtry) {
   1702 		ARPLOG(LOG_INFO, "%s: could not run DAD, driver problem?\n",
   1703 		    if_name(ifa->ifa_ifp));
   1704 
   1705 		arp_dad_stoptimer(dp);
   1706 		need_free = true;
   1707 		goto done;
   1708 	}
   1709 
   1710 	/* Need more checks? */
   1711 	if (dp->dad_arp_ocount < dp->dad_count) {
   1712 		int adelay;
   1713 
   1714 		/*
   1715 		 * We have more ARP to go.  Send ARP packet for DAD.
   1716 		 */
   1717 		arp_dad_output(dp, ifa);
   1718 		if (dp->dad_arp_ocount < dp->dad_count)
   1719 			adelay = (PROBE_MIN * hz) +
   1720 			    (cprng_fast32() %
   1721 			    ((PROBE_MAX * hz) - (PROBE_MIN * hz)));
   1722 		else
   1723 			adelay = ANNOUNCE_WAIT * hz;
   1724 		arp_dad_starttimer(dp, adelay);
   1725 		goto done;
   1726 	} else if (dp->dad_arp_acount == 0) {
   1727 		/*
   1728 		 * We are done with DAD.
   1729 		 * No duplicate address found.
   1730 		 */
   1731 		ia->ia4_flags &= ~IN_IFF_TENTATIVE;
   1732 		rt_addrmsg(RTM_NEWADDR, ifa);
   1733 		ARPLOG(LOG_DEBUG,
   1734 		    "%s: DAD complete for %s - no duplicates found\n",
   1735 		    if_name(ifa->ifa_ifp), ARPLOGADDR(&ia->ia_addr.sin_addr));
   1736 		dp->dad_arp_announce = ANNOUNCE_NUM;
   1737 		goto announce;
   1738 	} else if (dp->dad_arp_acount < dp->dad_arp_announce) {
   1739 announce:
   1740 		/*
   1741 		 * Announce the address.
   1742 		 */
   1743 		arpannounce1(ifa);
   1744 		dp->dad_arp_acount++;
   1745 		if (dp->dad_arp_acount < dp->dad_arp_announce) {
   1746 			arp_dad_starttimer(dp, ANNOUNCE_INTERVAL * hz);
   1747 			goto done;
   1748 		}
   1749 		ARPLOG(LOG_DEBUG,
   1750 		    "%s: ARP announcement complete for %s\n",
   1751 		    if_name(ifa->ifa_ifp), ARPLOGADDR(&ia->ia_addr.sin_addr));
   1752 	}
   1753 
   1754 	arp_dad_stoptimer(dp);
   1755 	need_free = true;
   1756 done:
   1757 	mutex_exit(&arp_dad_lock);
   1758 
   1759 	if (need_free) {
   1760 		arp_dad_destroytimer(dp);
   1761 		KASSERT(ifa != NULL);
   1762 		ifafree(ifa);
   1763 	}
   1764 
   1765 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
   1766 }
   1767 
   1768 static void
   1769 arp_dad_duplicated(struct ifaddr *ifa, const struct sockaddr_dl *from)
   1770 {
   1771 	struct in_ifaddr *ia = ifatoia(ifa);
   1772 	struct ifnet *ifp = ifa->ifa_ifp;
   1773 	char ipbuf[INET_ADDRSTRLEN], llabuf[LLA_ADDRSTRLEN];
   1774 	const char *iastr, *llastr;
   1775 
   1776 	iastr = IN_PRINT(ipbuf, &ia->ia_addr.sin_addr);
   1777 	if (__predict_false(from == NULL))
   1778 		llastr = NULL;
   1779 	else
   1780 		llastr = lla_snprintf(llabuf, sizeof(llabuf),
   1781 		    CLLADDR(from), from->sdl_alen);
   1782 
   1783 	if (ia->ia4_flags & (IN_IFF_TENTATIVE|IN_IFF_DUPLICATED)) {
   1784 		log(LOG_ERR,
   1785 		    "%s: DAD duplicate address %s from %s\n",
   1786 		    if_name(ifp), iastr, llastr);
   1787 	} else if (ia->ia_dad_defended == 0 ||
   1788 		   ia->ia_dad_defended < time_uptime - DEFEND_INTERVAL) {
   1789 		ia->ia_dad_defended = time_uptime;
   1790 		arpannounce1(ifa);
   1791 		log(LOG_ERR,
   1792 		    "%s: DAD defended address %s from %s\n",
   1793 		    if_name(ifp), iastr, llastr);
   1794 		return;
   1795 	} else {
   1796 		/* If DAD is disabled, just report the duplicate. */
   1797 		if (!ip_dad_enabled()) {
   1798 			log(LOG_ERR,
   1799 			    "%s: DAD ignoring duplicate address %s from %s\n",
   1800 			    if_name(ifp), iastr, llastr);
   1801 			return;
   1802 		}
   1803 		log(LOG_ERR,
   1804 		    "%s: DAD defence failed for %s from %s\n",
   1805 		    if_name(ifp), iastr, llastr);
   1806 	}
   1807 
   1808 	arp_dad_stop(ifa);
   1809 
   1810 	ia->ia4_flags &= ~IN_IFF_TENTATIVE;
   1811 	if ((ia->ia4_flags & IN_IFF_DUPLICATED) == 0) {
   1812 		ia->ia4_flags |= IN_IFF_DUPLICATED;
   1813 		/* Inform the routing socket of the duplicate address */
   1814 		rt_addrmsg_src(RTM_NEWADDR, ifa, (const struct sockaddr *)from);
   1815 	}
   1816 }
   1817 
   1818 /*
   1819  * Called from 10 Mb/s Ethernet interrupt handlers
   1820  * when ether packet type ETHERTYPE_REVARP
   1821  * is received.  Common length and type checks are done here,
   1822  * then the protocol-specific routine is called.
   1823  */
   1824 void
   1825 revarpinput(struct mbuf *m)
   1826 {
   1827 	struct arphdr *ar;
   1828 	int arplen;
   1829 
   1830 	arplen = sizeof(struct arphdr);
   1831 	if (m->m_len < arplen && (m = m_pullup(m, arplen)) == NULL)
   1832 		return;
   1833 	ar = mtod(m, struct arphdr *);
   1834 
   1835 	if (ntohs(ar->ar_hrd) == ARPHRD_IEEE1394) {
   1836 		goto out;
   1837 	}
   1838 
   1839 	arplen = sizeof(struct arphdr) + 2 * (ar->ar_hln + ar->ar_pln);
   1840 	if (m->m_len < arplen && (m = m_pullup(m, arplen)) == NULL)
   1841 		return;
   1842 	ar = mtod(m, struct arphdr *);
   1843 
   1844 	switch (ntohs(ar->ar_pro)) {
   1845 	case ETHERTYPE_IP:
   1846 	case ETHERTYPE_IPTRAILERS:
   1847 		in_revarpinput(m);
   1848 		return;
   1849 
   1850 	default:
   1851 		break;
   1852 	}
   1853 
   1854 out:
   1855 	m_freem(m);
   1856 }
   1857 
   1858 /*
   1859  * RARP for Internet protocols on 10 Mb/s Ethernet.
   1860  * Algorithm is that given in RFC 903.
   1861  * We are only using for bootstrap purposes to get an ip address for one of
   1862  * our interfaces.  Thus we support no user-interface.
   1863  *
   1864  * Since the contents of the RARP reply are specific to the interface that
   1865  * sent the request, this code must ensure that they are properly associated.
   1866  *
   1867  * Note: also supports ARP via RARP packets, per the RFC.
   1868  */
   1869 void
   1870 in_revarpinput(struct mbuf *m)
   1871 {
   1872 	struct arphdr *ah;
   1873 	void *tha;
   1874 	int op;
   1875 	struct ifnet *rcvif;
   1876 	int s;
   1877 
   1878 	ah = mtod(m, struct arphdr *);
   1879 	op = ntohs(ah->ar_op);
   1880 
   1881 	rcvif = m_get_rcvif(m, &s);
   1882 	if (__predict_false(rcvif == NULL))
   1883 		goto out;
   1884 	if (rcvif->if_flags & IFF_NOARP)
   1885 		goto out;
   1886 
   1887 	switch (rcvif->if_type) {
   1888 	case IFT_IEEE1394:
   1889 		/* ARP without target hardware address is not supported */
   1890 		goto out;
   1891 	default:
   1892 		break;
   1893 	}
   1894 
   1895 	switch (op) {
   1896 	case ARPOP_REQUEST:
   1897 	case ARPOP_REPLY:	/* per RFC */
   1898 		m_put_rcvif(rcvif, &s);
   1899 		in_arpinput(m);
   1900 		return;
   1901 	case ARPOP_REVREPLY:
   1902 		break;
   1903 	case ARPOP_REVREQUEST:	/* handled by rarpd(8) */
   1904 	default:
   1905 		goto out;
   1906 	}
   1907 	if (!revarp_in_progress)
   1908 		goto out;
   1909 	if (rcvif != myip_ifp) /* !same interface */
   1910 		goto out;
   1911 	if (myip_initialized)
   1912 		goto wake;
   1913 	tha = ar_tha(ah);
   1914 	if (tha == NULL)
   1915 		goto out;
   1916 	if (ah->ar_pln != sizeof(struct in_addr))
   1917 		goto out;
   1918 	if (ah->ar_hln != rcvif->if_sadl->sdl_alen)
   1919 		goto out;
   1920 	if (memcmp(tha, CLLADDR(rcvif->if_sadl), rcvif->if_sadl->sdl_alen))
   1921 		goto out;
   1922 	memcpy(&srv_ip, ar_spa(ah), sizeof(srv_ip));
   1923 	memcpy(&myip, ar_tpa(ah), sizeof(myip));
   1924 	myip_initialized = 1;
   1925 wake:	/* Do wakeup every time in case it was missed. */
   1926 	wakeup((void *)&myip);
   1927 
   1928 out:
   1929 	m_put_rcvif(rcvif, &s);
   1930 	m_freem(m);
   1931 }
   1932 
   1933 /*
   1934  * Send a RARP request for the ip address of the specified interface.
   1935  * The request should be RFC 903-compliant.
   1936  */
   1937 static void
   1938 revarprequest(struct ifnet *ifp)
   1939 {
   1940 	struct sockaddr sa;
   1941 	struct mbuf *m;
   1942 	struct arphdr *ah;
   1943 	void *tha;
   1944 
   1945 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
   1946 		return;
   1947 	MCLAIM(m, &arpdomain.dom_mowner);
   1948 	m->m_len = sizeof(*ah) + 2*sizeof(struct in_addr) +
   1949 	    2*ifp->if_addrlen;
   1950 	m->m_pkthdr.len = m->m_len;
   1951 	m_align(m, m->m_len);
   1952 	ah = mtod(m, struct arphdr *);
   1953 	memset(ah, 0, m->m_len);
   1954 	ah->ar_pro = htons(ETHERTYPE_IP);
   1955 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
   1956 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
   1957 	ah->ar_op = htons(ARPOP_REVREQUEST);
   1958 
   1959 	memcpy(ar_sha(ah), CLLADDR(ifp->if_sadl), ah->ar_hln);
   1960 	tha = ar_tha(ah);
   1961 	if (tha == NULL) {
   1962 		m_free(m);
   1963 		return;
   1964 	}
   1965 	memcpy(tha, CLLADDR(ifp->if_sadl), ah->ar_hln);
   1966 
   1967 	sa.sa_family = AF_ARP;
   1968 	sa.sa_len = 2;
   1969 	m->m_flags |= M_BCAST;
   1970 
   1971 	if_output_lock(ifp, ifp, m, &sa, NULL);
   1972 }
   1973 
   1974 /*
   1975  * RARP for the ip address of the specified interface, but also
   1976  * save the ip address of the server that sent the answer.
   1977  * Timeout if no response is received.
   1978  */
   1979 int
   1980 revarpwhoarewe(struct ifnet *ifp, struct in_addr *serv_in,
   1981     struct in_addr *clnt_in)
   1982 {
   1983 	int result, count = 20;
   1984 
   1985 	myip_initialized = 0;
   1986 	myip_ifp = ifp;
   1987 
   1988 	revarp_in_progress = 1;
   1989 	while (count--) {
   1990 		revarprequest(ifp);
   1991 		result = tsleep((void *)&myip, PSOCK, "revarp", hz/2);
   1992 		if (result != EWOULDBLOCK)
   1993 			break;
   1994 	}
   1995 	revarp_in_progress = 0;
   1996 
   1997 	if (!myip_initialized)
   1998 		return ENETUNREACH;
   1999 
   2000 	memcpy(serv_in, &srv_ip, sizeof(*serv_in));
   2001 	memcpy(clnt_in, &myip, sizeof(*clnt_in));
   2002 	return 0;
   2003 }
   2004 
   2005 void
   2006 arp_stat_add(int type, uint64_t count)
   2007 {
   2008 	ARP_STATADD(type, count);
   2009 }
   2010 
   2011 static int
   2012 sysctl_net_inet_arp_stats(SYSCTLFN_ARGS)
   2013 {
   2014 
   2015 	return NETSTAT_SYSCTL(arpstat_percpu, ARP_NSTATS);
   2016 }
   2017 
   2018 static void
   2019 sysctl_net_inet_arp_setup(struct sysctllog **clog)
   2020 {
   2021 	const struct sysctlnode *node;
   2022 
   2023 	sysctl_createv(clog, 0, NULL, NULL,
   2024 			CTLFLAG_PERMANENT,
   2025 			CTLTYPE_NODE, "inet", NULL,
   2026 			NULL, 0, NULL, 0,
   2027 			CTL_NET, PF_INET, CTL_EOL);
   2028 	sysctl_createv(clog, 0, NULL, &node,
   2029 			CTLFLAG_PERMANENT,
   2030 			CTLTYPE_NODE, "arp",
   2031 			SYSCTL_DESCR("Address Resolution Protocol"),
   2032 			NULL, 0, NULL, 0,
   2033 			CTL_NET, PF_INET, CTL_CREATE, CTL_EOL);
   2034 
   2035 	sysctl_createv(clog, 0, NULL, NULL,
   2036 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2037 		       CTLTYPE_INT, "nd_delay",
   2038 		       SYSCTL_DESCR("First probe delay time"),
   2039 		       NULL, 0, &arp_nd_domain.nd_delay, 0,
   2040 		       CTL_NET, PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2041 	sysctl_createv(clog, 0, NULL, NULL,
   2042 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2043 		       CTLTYPE_INT, "nd_bmaxtries",
   2044 		       SYSCTL_DESCR("Number of broadcast discovery attempts"),
   2045 		       NULL, 0, &arp_nd_domain.nd_mmaxtries, 0,
   2046 		       CTL_NET, PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2047 	sysctl_createv(clog, 0, NULL, NULL,
   2048 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2049 		       CTLTYPE_INT, "nd_umaxtries",
   2050 		       SYSCTL_DESCR("Number of unicast discovery attempts"),
   2051 		       NULL, 0, &arp_nd_domain.nd_umaxtries, 0,
   2052 		       CTL_NET, PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2053 	sysctl_createv(clog, 0, NULL, NULL,
   2054 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2055 		       CTLTYPE_INT, "nd_reachable",
   2056 		       SYSCTL_DESCR("Reachable time"),
   2057 		       NULL, 0, &arp_reachable, 0,
   2058 		       CTL_NET, PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2059 	sysctl_createv(clog, 0, NULL, NULL,
   2060 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2061 		       CTLTYPE_INT, "nd_retrans",
   2062 		       SYSCTL_DESCR("Retransmission time"),
   2063 		       NULL, 0, &arp_retrans, 0,
   2064 		       CTL_NET, PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2065 	sysctl_createv(clog, 0, NULL, NULL,
   2066 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2067 		       CTLTYPE_INT, "nd_nud",
   2068 		       SYSCTL_DESCR("Perform neighbour unreachability detection"),
   2069 		       NULL, 0, &arp_perform_nud, 0,
   2070 		       CTL_NET, PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2071 	sysctl_createv(clog, 0, NULL, NULL,
   2072 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2073 		       CTLTYPE_INT, "nd_maxnudhint",
   2074 		       SYSCTL_DESCR("Maximum neighbor unreachable hint count"),
   2075 		       NULL, 0, &arp_nd_domain.nd_maxnudhint, 0,
   2076 		       CTL_NET, PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2077 	sysctl_createv(clog, 0, NULL, NULL,
   2078 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2079 		       CTLTYPE_INT, "maxqueuelen",
   2080 		       SYSCTL_DESCR("max packet queue len for a unresolved ARP"),
   2081 		       NULL, 1, &arp_nd_domain.nd_maxqueuelen, 0,
   2082 		       CTL_NET, PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2083 
   2084 	sysctl_createv(clog, 0, NULL, NULL,
   2085 			CTLFLAG_PERMANENT,
   2086 			CTLTYPE_STRUCT, "stats",
   2087 			SYSCTL_DESCR("ARP statistics"),
   2088 			sysctl_net_inet_arp_stats, 0, NULL, 0,
   2089 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2090 
   2091 	sysctl_createv(clog, 0, NULL, NULL,
   2092 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2093 			CTLTYPE_INT, "log_movements",
   2094 			SYSCTL_DESCR("log ARP replies from MACs different than"
   2095 			    " the one in the cache"),
   2096 			NULL, 0, &log_movements, 0,
   2097 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2098 
   2099 	sysctl_createv(clog, 0, NULL, NULL,
   2100 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2101 			CTLTYPE_INT, "log_permanent_modify",
   2102 			SYSCTL_DESCR("log ARP replies from MACs different than"
   2103 			    " the one in the permanent arp entry"),
   2104 			NULL, 0, &log_permanent_modify, 0,
   2105 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2106 
   2107 	sysctl_createv(clog, 0, NULL, NULL,
   2108 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2109 			CTLTYPE_INT, "log_wrong_iface",
   2110 			SYSCTL_DESCR("log ARP packets arriving on the wrong"
   2111 			    " interface"),
   2112 			NULL, 0, &log_wrong_iface, 0,
   2113 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2114 
   2115 	sysctl_createv(clog, 0, NULL, NULL,
   2116 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2117 			CTLTYPE_INT, "debug",
   2118 			SYSCTL_DESCR("Enable ARP DAD debug output"),
   2119 			NULL, 0, &arp_debug, 0,
   2120 			CTL_NET, PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2121 }
   2122 
   2123 #endif /* INET */
   2124