Home | History | Annotate | Line # | Graph | Download | only in netinet
      1 /*	$NetBSD: if_arp.c,v 1.231 2016/10/18 07:30:31 ozaki-r 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 1.231 2016/10/18 07:30:31 ozaki-r 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/malloc.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_token.h>
    109 #include <net/if_types.h>
    110 #include <net/if_ether.h>
    111 #include <net/if_llatbl.h>
    112 #include <net/net_osdep.h>
    113 #include <net/route.h>
    114 #include <net/net_stats.h>
    115 
    116 #include <netinet/in.h>
    117 #include <netinet/in_systm.h>
    118 #include <netinet/in_var.h>
    119 #include <netinet/ip.h>
    120 #include <netinet/if_inarp.h>
    121 
    122 #include "arcnet.h"
    123 #if NARCNET > 0
    124 #include <net/if_arc.h>
    125 #endif
    126 #include "fddi.h"
    127 #if NFDDI > 0
    128 #include <net/if_fddi.h>
    129 #endif
    130 #include "token.h"
    131 #include "carp.h"
    132 #if NCARP > 0
    133 #include <netinet/ip_carp.h>
    134 #endif
    135 
    136 #define SIN(s) ((struct sockaddr_in *)s)
    137 #define SRP(s) ((struct sockaddr_inarp *)s)
    138 
    139 /*
    140  * ARP trailer negotiation.  Trailer protocol is not IP specific,
    141  * but ARP request/response use IP addresses.
    142  */
    143 #define ETHERTYPE_IPTRAILERS ETHERTYPE_TRAIL
    144 
    145 /* timer values */
    146 static int	arpt_keep = (20*60);	/* once resolved, good for 20 more minutes */
    147 static int	arpt_down = 20;		/* once declared down, don't send for 20 secs */
    148 static int	arp_maxhold = 1;	/* number of packets to hold per ARP entry */
    149 #define	rt_expire rt_rmx.rmx_expire
    150 #define	rt_pksent rt_rmx.rmx_pksent
    151 
    152 int		ip_dad_count = PROBE_NUM;
    153 #ifdef ARP_DEBUG
    154 int		arp_debug = 1;
    155 #else
    156 int		arp_debug = 0;
    157 #endif
    158 
    159 static	void arp_init(void);
    160 
    161 static	void arprequest(struct ifnet *,
    162     const struct in_addr *, const struct in_addr *,
    163     const u_int8_t *);
    164 static	void arpannounce1(struct ifaddr *);
    165 static	struct sockaddr *arp_setgate(struct rtentry *, struct sockaddr *,
    166 	    const struct sockaddr *);
    167 static	void arptimer(void *);
    168 static	void arp_settimer(struct llentry *, int);
    169 static	struct llentry *arplookup(struct ifnet *, struct mbuf *,
    170 	    const struct in_addr *, const struct sockaddr *, int);
    171 static	struct llentry *arpcreate(struct ifnet *, struct mbuf *,
    172 	    const struct in_addr *, const struct sockaddr *, int);
    173 static	void in_arpinput(struct mbuf *);
    174 static	void in_revarpinput(struct mbuf *);
    175 static	void revarprequest(struct ifnet *);
    176 
    177 static	void arp_drainstub(void);
    178 
    179 static void arp_dad_timer(struct ifaddr *);
    180 static void arp_dad_start(struct ifaddr *);
    181 static void arp_dad_stop(struct ifaddr *);
    182 static void arp_dad_duplicated(struct ifaddr *, const char *);
    183 
    184 static void arp_init_llentry(struct ifnet *, struct llentry *);
    185 #if NTOKEN > 0
    186 static void arp_free_llentry_tokenring(struct llentry *);
    187 #endif
    188 
    189 struct	ifqueue arpintrq = {
    190 	.ifq_head = NULL,
    191 	.ifq_tail = NULL,
    192 	.ifq_len = 0,
    193 	.ifq_maxlen = 50,
    194 	.ifq_drops = 0,
    195 };
    196 static int	arp_maxtries = 5;
    197 static int	useloopback = 1;	/* use loopback interface for local traffic */
    198 
    199 static percpu_t *arpstat_percpu;
    200 
    201 #define	ARP_STAT_GETREF()	_NET_STAT_GETREF(arpstat_percpu)
    202 #define	ARP_STAT_PUTREF()	_NET_STAT_PUTREF(arpstat_percpu)
    203 
    204 #define	ARP_STATINC(x)		_NET_STATINC(arpstat_percpu, x)
    205 #define	ARP_STATADD(x, v)	_NET_STATADD(arpstat_percpu, x, v)
    206 
    207 /* revarp state */
    208 static struct	in_addr myip, srv_ip;
    209 static int	myip_initialized = 0;
    210 static int	revarp_in_progress = 0;
    211 static struct	ifnet *myip_ifp = NULL;
    212 
    213 static int arp_drainwanted;
    214 
    215 static int log_movements = 1;
    216 static int log_permanent_modify = 1;
    217 static int log_wrong_iface = 1;
    218 static int log_unknown_network = 1;
    219 
    220 /*
    221  * this should be elsewhere.
    222  */
    223 
    224 static char *
    225 lla_snprintf(u_int8_t *, int);
    226 
    227 static char *
    228 lla_snprintf(u_int8_t *adrp, int len)
    229 {
    230 #define NUMBUFS 3
    231 	static char buf[NUMBUFS][16*3];
    232 	static int bnum = 0;
    233 
    234 	int i;
    235 	char *p;
    236 
    237 	p = buf[bnum];
    238 
    239 	*p++ = hexdigits[(*adrp)>>4];
    240 	*p++ = hexdigits[(*adrp++)&0xf];
    241 
    242 	for (i=1; i<len && i<16; i++) {
    243 		*p++ = ':';
    244 		*p++ = hexdigits[(*adrp)>>4];
    245 		*p++ = hexdigits[(*adrp++)&0xf];
    246 	}
    247 
    248 	*p = 0;
    249 	p = buf[bnum];
    250 	bnum = (bnum + 1) % NUMBUFS;
    251 	return p;
    252 }
    253 
    254 DOMAIN_DEFINE(arpdomain);	/* forward declare and add to link set */
    255 
    256 static void
    257 arp_fasttimo(void)
    258 {
    259 	if (arp_drainwanted) {
    260 		arp_drain();
    261 		arp_drainwanted = 0;
    262 	}
    263 }
    264 
    265 const struct protosw arpsw[] = {
    266 	{ .pr_type = 0,
    267 	  .pr_domain = &arpdomain,
    268 	  .pr_protocol = 0,
    269 	  .pr_flags = 0,
    270 	  .pr_input = 0,
    271 	  .pr_ctlinput = 0,
    272 	  .pr_ctloutput = 0,
    273 	  .pr_usrreqs = 0,
    274 	  .pr_init = arp_init,
    275 	  .pr_fasttimo = arp_fasttimo,
    276 	  .pr_slowtimo = 0,
    277 	  .pr_drain = arp_drainstub,
    278 	}
    279 };
    280 
    281 struct domain arpdomain = {
    282 	.dom_family = PF_ARP,
    283 	.dom_name = "arp",
    284 	.dom_protosw = arpsw,
    285 	.dom_protoswNPROTOSW = &arpsw[__arraycount(arpsw)],
    286 };
    287 
    288 static void sysctl_net_inet_arp_setup(struct sysctllog **);
    289 
    290 void
    291 arp_init(void)
    292 {
    293 
    294 	sysctl_net_inet_arp_setup(NULL);
    295 	arpstat_percpu = percpu_alloc(sizeof(uint64_t) * ARP_NSTATS);
    296 	IFQ_LOCK_INIT(&arpintrq);
    297 }
    298 
    299 static void
    300 arp_drainstub(void)
    301 {
    302 	arp_drainwanted = 1;
    303 }
    304 
    305 /*
    306  * ARP protocol drain routine.  Called when memory is in short supply.
    307  * Called at splvm();  don't acquire softnet_lock as can be called from
    308  * hardware interrupt handlers.
    309  */
    310 void
    311 arp_drain(void)
    312 {
    313 
    314 	lltable_drain(AF_INET);
    315 }
    316 
    317 static void
    318 arptimer(void *arg)
    319 {
    320 	struct llentry *lle = arg;
    321 	struct ifnet *ifp;
    322 
    323 	if (lle == NULL)
    324 		return;
    325 
    326 	if (lle->la_flags & LLE_STATIC)
    327 		return;
    328 
    329 	LLE_WLOCK(lle);
    330 	if (callout_pending(&lle->la_timer)) {
    331 		/*
    332 		 * Here we are a bit odd here in the treatment of
    333 		 * active/pending. If the pending bit is set, it got
    334 		 * rescheduled before I ran. The active
    335 		 * bit we ignore, since if it was stopped
    336 		 * in ll_tablefree() and was currently running
    337 		 * it would have return 0 so the code would
    338 		 * not have deleted it since the callout could
    339 		 * not be stopped so we want to go through
    340 		 * with the delete here now. If the callout
    341 		 * was restarted, the pending bit will be back on and
    342 		 * we just want to bail since the callout_reset would
    343 		 * return 1 and our reference would have been removed
    344 		 * by arpresolve() below.
    345 		 */
    346 		LLE_WUNLOCK(lle);
    347 		return;
    348 	}
    349 	ifp = lle->lle_tbl->llt_ifp;
    350 
    351 	callout_stop(&lle->la_timer);
    352 
    353 	/* XXX: LOR avoidance. We still have ref on lle. */
    354 	LLE_WUNLOCK(lle);
    355 
    356 	IF_AFDATA_LOCK(ifp);
    357 	LLE_WLOCK(lle);
    358 
    359 	/* Guard against race with other llentry_free(). */
    360 	if (lle->la_flags & LLE_LINKED) {
    361 		size_t pkts_dropped;
    362 
    363 		LLE_REMREF(lle);
    364 		pkts_dropped = llentry_free(lle);
    365 		ARP_STATADD(ARP_STAT_DFRDROPPED, pkts_dropped);
    366 		ARP_STATADD(ARP_STAT_DFRTOTAL, pkts_dropped);
    367 	} else {
    368 		LLE_FREE_LOCKED(lle);
    369 	}
    370 
    371 	IF_AFDATA_UNLOCK(ifp);
    372 }
    373 
    374 static void
    375 arp_settimer(struct llentry *la, int sec)
    376 {
    377 
    378 	LLE_WLOCK_ASSERT(la);
    379 	LLE_ADDREF(la);
    380 	callout_reset(&la->la_timer, hz * sec, arptimer, la);
    381 }
    382 
    383 /*
    384  * We set the gateway for RTF_CLONING routes to a "prototype"
    385  * link-layer sockaddr whose interface type (if_type) and interface
    386  * index (if_index) fields are prepared.
    387  */
    388 static struct sockaddr *
    389 arp_setgate(struct rtentry *rt, struct sockaddr *gate,
    390     const struct sockaddr *netmask)
    391 {
    392 	const struct ifnet *ifp = rt->rt_ifp;
    393 	uint8_t namelen = strlen(ifp->if_xname);
    394 	uint8_t addrlen = ifp->if_addrlen;
    395 
    396 	/*
    397 	 * XXX: If this is a manually added route to interface
    398 	 * such as older version of routed or gated might provide,
    399 	 * restore cloning bit.
    400 	 */
    401 	if ((rt->rt_flags & RTF_HOST) == 0 && netmask != NULL &&
    402 	    satocsin(netmask)->sin_addr.s_addr != 0xffffffff)
    403 		rt->rt_flags |= RTF_CONNECTED;
    404 
    405 	if ((rt->rt_flags & (RTF_CONNECTED | RTF_LOCAL))) {
    406 		union {
    407 			struct sockaddr sa;
    408 			struct sockaddr_storage ss;
    409 			struct sockaddr_dl sdl;
    410 		} u;
    411 		/*
    412 		 * Case 1: This route should come from a route to iface.
    413 		 */
    414 		sockaddr_dl_init(&u.sdl, sizeof(u.ss),
    415 		    ifp->if_index, ifp->if_type, NULL, namelen, NULL, addrlen);
    416 		rt_setgate(rt, &u.sa);
    417 		gate = rt->rt_gateway;
    418 	}
    419 	return gate;
    420 }
    421 
    422 static void
    423 arp_init_llentry(struct ifnet *ifp, struct llentry *lle)
    424 {
    425 
    426 	switch (ifp->if_type) {
    427 #if NTOKEN > 0
    428 	case IFT_ISO88025:
    429 		lle->la_opaque = kmem_intr_alloc(sizeof(struct token_rif),
    430 		    KM_NOSLEEP);
    431 		lle->lle_ll_free = arp_free_llentry_tokenring;
    432 		break;
    433 #endif
    434 	}
    435 }
    436 
    437 #if NTOKEN > 0
    438 static void
    439 arp_free_llentry_tokenring(struct llentry *lle)
    440 {
    441 
    442 	kmem_intr_free(lle->la_opaque, sizeof(struct token_rif));
    443 }
    444 #endif
    445 
    446 /*
    447  * Parallel to llc_rtrequest.
    448  */
    449 void
    450 arp_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info)
    451 {
    452 	struct sockaddr *gate = rt->rt_gateway;
    453 	struct in_ifaddr *ia;
    454 	struct ifaddr *ifa;
    455 	struct ifnet *ifp = rt->rt_ifp;
    456 	int bound;
    457 	int s;
    458 
    459 	if (req == RTM_LLINFO_UPD) {
    460 		if ((ifa = info->rti_ifa) != NULL)
    461 			arpannounce1(ifa);
    462 		return;
    463 	}
    464 
    465 	if ((rt->rt_flags & RTF_GATEWAY) != 0) {
    466 		if (req != RTM_ADD)
    467 			return;
    468 
    469 		/*
    470 		 * linklayers with particular link MTU limitation.
    471 		 */
    472 		switch(ifp->if_type) {
    473 #if NFDDI > 0
    474 		case IFT_FDDI:
    475 			if (ifp->if_mtu > FDDIIPMTU)
    476 				rt->rt_rmx.rmx_mtu = FDDIIPMTU;
    477 			break;
    478 #endif
    479 #if NARCNET > 0
    480 		case IFT_ARCNET:
    481 		    {
    482 			int arcipifmtu;
    483 
    484 			if (ifp->if_flags & IFF_LINK0)
    485 				arcipifmtu = arc_ipmtu;
    486 			else
    487 				arcipifmtu = ARCMTU;
    488 			if (ifp->if_mtu > arcipifmtu)
    489 				rt->rt_rmx.rmx_mtu = arcipifmtu;
    490 			break;
    491 		    }
    492 #endif
    493 		}
    494 		return;
    495 	}
    496 
    497 	switch (req) {
    498 	case RTM_SETGATE:
    499 		gate = arp_setgate(rt, gate, info->rti_info[RTAX_NETMASK]);
    500 		break;
    501 	case RTM_ADD:
    502 		gate = arp_setgate(rt, gate, info->rti_info[RTAX_NETMASK]);
    503 		if (gate == NULL) {
    504 			log(LOG_ERR, "%s: arp_setgate failed\n", __func__);
    505 			break;
    506 		}
    507 		if ((rt->rt_flags & RTF_CONNECTED) ||
    508 		    (rt->rt_flags & RTF_LOCAL)) {
    509 			/*
    510 			 * Give this route an expiration time, even though
    511 			 * it's a "permanent" route, so that routes cloned
    512 			 * from it do not need their expiration time set.
    513 			 */
    514 			KASSERT(time_uptime != 0);
    515 			rt->rt_expire = time_uptime;
    516 			/*
    517 			 * linklayers with particular link MTU limitation.
    518 			 */
    519 			switch (ifp->if_type) {
    520 #if NFDDI > 0
    521 			case IFT_FDDI:
    522 				if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0 &&
    523 				    (rt->rt_rmx.rmx_mtu > FDDIIPMTU ||
    524 				     (rt->rt_rmx.rmx_mtu == 0 &&
    525 				      ifp->if_mtu > FDDIIPMTU)))
    526 					rt->rt_rmx.rmx_mtu = FDDIIPMTU;
    527 				break;
    528 #endif
    529 #if NARCNET > 0
    530 			case IFT_ARCNET:
    531 			    {
    532 				int arcipifmtu;
    533 				if (ifp->if_flags & IFF_LINK0)
    534 					arcipifmtu = arc_ipmtu;
    535 				else
    536 					arcipifmtu = ARCMTU;
    537 
    538 				if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0 &&
    539 				    (rt->rt_rmx.rmx_mtu > arcipifmtu ||
    540 				     (rt->rt_rmx.rmx_mtu == 0 &&
    541 				      ifp->if_mtu > arcipifmtu)))
    542 					rt->rt_rmx.rmx_mtu = arcipifmtu;
    543 				break;
    544 			    }
    545 #endif
    546 			}
    547 			if (rt->rt_flags & RTF_CONNECTED)
    548 				break;
    549 		}
    550 
    551 		bound = curlwp_bind();
    552 		/* Announce a new entry if requested. */
    553 		if (rt->rt_flags & RTF_ANNOUNCE) {
    554 			struct psref psref;
    555 			ia = in_get_ia_on_iface_psref(
    556 			    satocsin(rt_getkey(rt))->sin_addr, ifp, &psref);
    557 			if (ia != NULL) {
    558 				arpannounce(ifp, &ia->ia_ifa,
    559 				    CLLADDR(satocsdl(gate)));
    560 				ia4_release(ia, &psref);
    561 			}
    562 		}
    563 
    564 		if (gate->sa_family != AF_LINK ||
    565 		    gate->sa_len < sockaddr_dl_measure(0, ifp->if_addrlen)) {
    566 			log(LOG_DEBUG, "%s: bad gateway value\n", __func__);
    567 			goto out;
    568 		}
    569 
    570 		satosdl(gate)->sdl_type = ifp->if_type;
    571 		satosdl(gate)->sdl_index = ifp->if_index;
    572 
    573 		/* If the route is for a broadcast address mark it as such.
    574 		 * This way we can avoid an expensive call to in_broadcast()
    575 		 * in ip_output() most of the time (because the route passed
    576 		 * to ip_output() is almost always a host route). */
    577 		if (rt->rt_flags & RTF_HOST &&
    578 		    !(rt->rt_flags & RTF_BROADCAST) &&
    579 		    in_broadcast(satocsin(rt_getkey(rt))->sin_addr, rt->rt_ifp))
    580 			rt->rt_flags |= RTF_BROADCAST;
    581 		/* There is little point in resolving the broadcast address */
    582 		if (rt->rt_flags & RTF_BROADCAST)
    583 			goto out;
    584 
    585 		/*
    586 		 * When called from rt_ifa_addlocal, we cannot depend on that
    587 		 * the address (rt_getkey(rt)) exits in the address list of the
    588 		 * interface. So check RTF_LOCAL instead.
    589 		 */
    590 		if (rt->rt_flags & RTF_LOCAL) {
    591 			rt->rt_expire = 0;
    592 			if (useloopback) {
    593 				rt->rt_ifp = lo0ifp;
    594 				rt->rt_rmx.rmx_mtu = 0;
    595 			}
    596 			goto out;
    597 		}
    598 
    599 		s = pserialize_read_enter();
    600 		ia = in_get_ia_on_iface(satocsin(rt_getkey(rt))->sin_addr, ifp);
    601 		if (ia == NULL) {
    602 			pserialize_read_exit(s);
    603 			goto out;
    604 		}
    605 
    606 		rt->rt_expire = 0;
    607 		if (useloopback) {
    608 			rt->rt_ifp = lo0ifp;
    609 			rt->rt_rmx.rmx_mtu = 0;
    610 		}
    611 		rt->rt_flags |= RTF_LOCAL;
    612 		/*
    613 		 * make sure to set rt->rt_ifa to the interface
    614 		 * address we are using, otherwise we will have trouble
    615 		 * with source address selection.
    616 		 */
    617 		ifa = &ia->ia_ifa;
    618 		if (ifa != rt->rt_ifa)
    619 			/* Assume it doesn't sleep */
    620 			rt_replace_ifa(rt, ifa);
    621 		pserialize_read_exit(s);
    622 	out:
    623 		curlwp_bindx(bound);
    624 		break;
    625 	}
    626 }
    627 
    628 /*
    629  * Broadcast an ARP request. Caller specifies:
    630  *	- arp header source ip address
    631  *	- arp header target ip address
    632  *	- arp header source ethernet address
    633  */
    634 static void
    635 arprequest(struct ifnet *ifp,
    636     const struct in_addr *sip, const struct in_addr *tip,
    637     const u_int8_t *enaddr)
    638 {
    639 	struct mbuf *m;
    640 	struct arphdr *ah;
    641 	struct sockaddr sa;
    642 	uint64_t *arps;
    643 
    644 	KASSERT(sip != NULL);
    645 	KASSERT(tip != NULL);
    646 	KASSERT(enaddr != NULL);
    647 
    648 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
    649 		return;
    650 	MCLAIM(m, &arpdomain.dom_mowner);
    651 	switch (ifp->if_type) {
    652 	case IFT_IEEE1394:
    653 		m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
    654 		    ifp->if_addrlen;
    655 		break;
    656 	default:
    657 		m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
    658 		    2 * ifp->if_addrlen;
    659 		break;
    660 	}
    661 	m->m_pkthdr.len = m->m_len;
    662 	MH_ALIGN(m, m->m_len);
    663 	ah = mtod(m, struct arphdr *);
    664 	memset(ah, 0, m->m_len);
    665 	switch (ifp->if_type) {
    666 	case IFT_IEEE1394:	/* RFC2734 */
    667 		/* fill it now for ar_tpa computation */
    668 		ah->ar_hrd = htons(ARPHRD_IEEE1394);
    669 		break;
    670 	default:
    671 		/* ifp->if_output will fill ar_hrd */
    672 		break;
    673 	}
    674 	ah->ar_pro = htons(ETHERTYPE_IP);
    675 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
    676 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
    677 	ah->ar_op = htons(ARPOP_REQUEST);
    678 	memcpy(ar_sha(ah), enaddr, ah->ar_hln);
    679 	memcpy(ar_spa(ah), sip, ah->ar_pln);
    680 	memcpy(ar_tpa(ah), tip, ah->ar_pln);
    681 	sa.sa_family = AF_ARP;
    682 	sa.sa_len = 2;
    683 	m->m_flags |= M_BCAST;
    684 	arps = ARP_STAT_GETREF();
    685 	arps[ARP_STAT_SNDTOTAL]++;
    686 	arps[ARP_STAT_SENDREQUEST]++;
    687 	ARP_STAT_PUTREF();
    688 	if_output_lock(ifp, ifp, m, &sa, NULL);
    689 }
    690 
    691 void
    692 arpannounce(struct ifnet *ifp, struct ifaddr *ifa, const uint8_t *enaddr)
    693 {
    694 	struct in_ifaddr *ia = ifatoia(ifa);
    695 	struct in_addr *ip = &IA_SIN(ifa)->sin_addr;
    696 
    697 	if (ia->ia4_flags & (IN_IFF_NOTREADY | IN_IFF_DETACHED)) {
    698 		arplog(LOG_DEBUG, "%s not ready\n", in_fmtaddr(*ip));
    699 		return;
    700 	}
    701 	arprequest(ifp, ip, ip, enaddr);
    702 }
    703 
    704 static void
    705 arpannounce1(struct ifaddr *ifa)
    706 {
    707 
    708 	arpannounce(ifa->ifa_ifp, ifa, CLLADDR(ifa->ifa_ifp->if_sadl));
    709 }
    710 
    711 /*
    712  * Resolve an IP address into an ethernet address.  If success,
    713  * desten is filled in.  If there is no entry in arptab,
    714  * set one up and broadcast a request for the IP address.
    715  * Hold onto this mbuf and resend it once the address
    716  * is finally resolved.  A return value of 0 indicates
    717  * that desten has been filled in and the packet should be sent
    718  * normally; a return value of EWOULDBLOCK indicates that the packet has been
    719  * held pending resolution.
    720  * Any other value indicates an error.
    721  */
    722 int
    723 arpresolve(struct ifnet *ifp, const struct rtentry *rt, struct mbuf *m,
    724     const struct sockaddr *dst, void *desten, size_t destlen)
    725 {
    726 	struct llentry *la;
    727 	const char *create_lookup;
    728 	bool renew;
    729 	int error;
    730 
    731 	KASSERT(m != NULL);
    732 
    733 	la = arplookup(ifp, m, NULL, dst, 0);
    734 	if (la == NULL)
    735 		goto notfound;
    736 
    737 	if ((la->la_flags & LLE_VALID) &&
    738 	    ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
    739 		KASSERT(destlen >= ifp->if_addrlen);
    740 		memcpy(desten, &la->ll_addr, ifp->if_addrlen);
    741 		LLE_RUNLOCK(la);
    742 		return 0;
    743 	}
    744 
    745 notfound:
    746 #ifdef IFF_STATICARP /* FreeBSD */
    747 #define _IFF_NOARP (IFF_NOARP | IFF_STATICARP)
    748 #else
    749 #define _IFF_NOARP IFF_NOARP
    750 #endif
    751 	if (ifp->if_flags & _IFF_NOARP) {
    752 		if (la != NULL)
    753 			LLE_RUNLOCK(la);
    754 		error = ENOTSUP;
    755 		goto bad;
    756 	}
    757 #undef _IFF_NOARP
    758 	if (la == NULL) {
    759 		create_lookup = "create";
    760 		IF_AFDATA_WLOCK(ifp);
    761 		la = lla_create(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
    762 		IF_AFDATA_WUNLOCK(ifp);
    763 		if (la == NULL)
    764 			ARP_STATINC(ARP_STAT_ALLOCFAIL);
    765 		else
    766 			arp_init_llentry(ifp, la);
    767 	} else if (LLE_TRY_UPGRADE(la) == 0) {
    768 		create_lookup = "lookup";
    769 		LLE_RUNLOCK(la);
    770 		IF_AFDATA_RLOCK(ifp);
    771 		la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
    772 		IF_AFDATA_RUNLOCK(ifp);
    773 	}
    774 
    775 	error = EINVAL;
    776 	if (la == NULL) {
    777 		log(LOG_DEBUG,
    778 		    "%s: failed to %s llentry for %s on %s\n",
    779 		    __func__, create_lookup, inet_ntoa(satocsin(dst)->sin_addr),
    780 		    ifp->if_xname);
    781 		goto bad;
    782 	}
    783 
    784 	if ((la->la_flags & LLE_VALID) &&
    785 	    ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime))
    786 	{
    787 		KASSERT(destlen >= ifp->if_addrlen);
    788 		memcpy(desten, &la->ll_addr, ifp->if_addrlen);
    789 		renew = false;
    790 		/*
    791 		 * If entry has an expiry time and it is approaching,
    792 		 * see if we need to send an ARP request within this
    793 		 * arpt_down interval.
    794 		 */
    795 		if (!(la->la_flags & LLE_STATIC) &&
    796 		    time_uptime + la->la_preempt > la->la_expire)
    797 		{
    798 			renew = true;
    799 			la->la_preempt--;
    800 		}
    801 
    802 		LLE_WUNLOCK(la);
    803 
    804 		if (renew) {
    805 			const u_int8_t *enaddr =
    806 #if NCARP > 0
    807 			    (ifp->if_type == IFT_CARP) ?
    808 			    CLLADDR(ifp->if_sadl):
    809 #endif
    810 			    CLLADDR(ifp->if_sadl);
    811 			arprequest(ifp,
    812 			    &satocsin(rt->rt_ifa->ifa_addr)->sin_addr,
    813 			    &satocsin(dst)->sin_addr, enaddr);
    814 		}
    815 
    816 		return 0;
    817 	}
    818 
    819 	if (la->la_flags & LLE_STATIC) {   /* should not happen! */
    820 		LLE_RUNLOCK(la);
    821 		log(LOG_DEBUG, "%s: ouch, empty static llinfo for %s\n",
    822 		    __func__, inet_ntoa(satocsin(dst)->sin_addr));
    823 		error = EINVAL;
    824 		goto bad;
    825 	}
    826 
    827 	renew = (la->la_asked == 0 || la->la_expire != time_uptime);
    828 
    829 	/*
    830 	 * There is an arptab entry, but no ethernet address
    831 	 * response yet.  Add the mbuf to the list, dropping
    832 	 * the oldest packet if we have exceeded the system
    833 	 * setting.
    834 	 */
    835 	LLE_WLOCK_ASSERT(la);
    836 	if (la->la_numheld >= arp_maxhold) {
    837 		if (la->la_hold != NULL) {
    838 			struct mbuf *next = la->la_hold->m_nextpkt;
    839 			m_freem(la->la_hold);
    840 			la->la_hold = next;
    841 			la->la_numheld--;
    842 			ARP_STATINC(ARP_STAT_DFRDROPPED);
    843 			ARP_STATINC(ARP_STAT_DFRTOTAL);
    844 		}
    845 	}
    846 	if (la->la_hold != NULL) {
    847 		struct mbuf *curr = la->la_hold;
    848 		while (curr->m_nextpkt != NULL)
    849 			curr = curr->m_nextpkt;
    850 		curr->m_nextpkt = m;
    851 	} else
    852 		la->la_hold = m;
    853 	la->la_numheld++;
    854 	if (!renew)
    855 		LLE_DOWNGRADE(la);
    856 
    857 	/*
    858 	 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
    859 	 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
    860 	 * if we have already sent arp_maxtries ARP requests. Retransmit the
    861 	 * ARP request, but not faster than one request per second.
    862 	 */
    863 	if (la->la_asked < arp_maxtries)
    864 		error = EWOULDBLOCK;	/* First request. */
    865 	else
    866 		error = (rt != NULL && rt->rt_flags & RTF_GATEWAY) ?
    867 		    EHOSTUNREACH : EHOSTDOWN;
    868 
    869 	if (renew) {
    870 		const u_int8_t *enaddr =
    871 #if NCARP > 0
    872 		    (rt != NULL && rt->rt_ifp->if_type == IFT_CARP) ?
    873 		    CLLADDR(rt->rt_ifp->if_sadl):
    874 #endif
    875 		    CLLADDR(ifp->if_sadl);
    876 		la->la_expire = time_uptime;
    877 		arp_settimer(la, arpt_down);
    878 		la->la_asked++;
    879 		LLE_WUNLOCK(la);
    880 
    881 		if (rt != NULL) {
    882 			arprequest(ifp, &satocsin(rt->rt_ifa->ifa_addr)->sin_addr,
    883 			    &satocsin(dst)->sin_addr, enaddr);
    884 		} else {
    885 			struct sockaddr_in sin;
    886 			struct rtentry *_rt;
    887 
    888 			sockaddr_in_init(&sin, &la->r_l3addr.addr4, 0);
    889 
    890 			/* XXX */
    891 			_rt = rtalloc1((struct sockaddr *)&sin, 0);
    892 			if (_rt == NULL)
    893 				goto bad;
    894 			arprequest(ifp,
    895 			    &satocsin(_rt->rt_ifa->ifa_addr)->sin_addr,
    896 			    &satocsin(dst)->sin_addr, enaddr);
    897 			rtfree(_rt);
    898 		}
    899 		return error;
    900 	}
    901 
    902 	LLE_RUNLOCK(la);
    903 	return error;
    904 
    905 bad:
    906 	m_freem(m);
    907 	return error;
    908 }
    909 
    910 /*
    911  * Common length and type checks are done here,
    912  * then the protocol-specific routine is called.
    913  */
    914 void
    915 arpintr(void)
    916 {
    917 	struct mbuf *m;
    918 	struct arphdr *ar;
    919 	int s;
    920 	int arplen;
    921 
    922 #ifndef NET_MPSAFE
    923 	mutex_enter(softnet_lock);
    924 	KERNEL_LOCK(1, NULL);
    925 #endif
    926 	for (;;) {
    927 		struct ifnet *rcvif;
    928 
    929 		IFQ_LOCK(&arpintrq);
    930 		IF_DEQUEUE(&arpintrq, m);
    931 		IFQ_UNLOCK(&arpintrq);
    932 		if (m == NULL)
    933 			goto out;
    934 		if ((m->m_flags & M_PKTHDR) == 0)
    935 			panic("arpintr");
    936 
    937 		MCLAIM(m, &arpdomain.dom_mowner);
    938 		ARP_STATINC(ARP_STAT_RCVTOTAL);
    939 
    940 		/*
    941 		 * First, make sure we have at least struct arphdr.
    942 		 */
    943 		if (m->m_len < sizeof(struct arphdr) ||
    944 		    (ar = mtod(m, struct arphdr *)) == NULL)
    945 			goto badlen;
    946 
    947 		rcvif = m_get_rcvif(m, &s);
    948 		switch (rcvif->if_type) {
    949 		case IFT_IEEE1394:
    950 			arplen = sizeof(struct arphdr) +
    951 			    ar->ar_hln + 2 * ar->ar_pln;
    952 			break;
    953 		default:
    954 			arplen = sizeof(struct arphdr) +
    955 			    2 * ar->ar_hln + 2 * ar->ar_pln;
    956 			break;
    957 		}
    958 		m_put_rcvif(rcvif, &s);
    959 
    960 		if (/* XXX ntohs(ar->ar_hrd) == ARPHRD_ETHER && */
    961 		    m->m_len >= arplen)
    962 			switch (ntohs(ar->ar_pro)) {
    963 			case ETHERTYPE_IP:
    964 			case ETHERTYPE_IPTRAILERS:
    965 				in_arpinput(m);
    966 				continue;
    967 			default:
    968 				ARP_STATINC(ARP_STAT_RCVBADPROTO);
    969 			}
    970 		else {
    971 badlen:
    972 			ARP_STATINC(ARP_STAT_RCVBADLEN);
    973 		}
    974 		m_freem(m);
    975 	}
    976 out:
    977 #ifndef NET_MPSAFE
    978 	KERNEL_UNLOCK_ONE(NULL);
    979 	mutex_exit(softnet_lock);
    980 #else
    981 	return; /* XXX gcc */
    982 #endif
    983 }
    984 
    985 /*
    986  * ARP for Internet protocols on 10 Mb/s Ethernet.
    987  * Algorithm is that given in RFC 826.
    988  * In addition, a sanity check is performed on the sender
    989  * protocol address, to catch impersonators.
    990  * We no longer handle negotiations for use of trailer protocol:
    991  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
    992  * along with IP replies if we wanted trailers sent to us,
    993  * and also sent them in response to IP replies.
    994  * This allowed either end to announce the desire to receive
    995  * trailer packets.
    996  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
    997  * but formerly didn't normally send requests.
    998  */
    999 static void
   1000 in_arpinput(struct mbuf *m)
   1001 {
   1002 	struct arphdr *ah;
   1003 	struct ifnet *ifp, *rcvif = NULL;
   1004 	struct llentry *la = NULL;
   1005 	struct in_ifaddr *ia = NULL;
   1006 #if NBRIDGE > 0
   1007 	struct in_ifaddr *bridge_ia = NULL;
   1008 #endif
   1009 #if NCARP > 0
   1010 	u_int32_t count = 0, index = 0;
   1011 #endif
   1012 	struct sockaddr sa;
   1013 	struct in_addr isaddr, itaddr, myaddr;
   1014 	int op;
   1015 	void *tha;
   1016 	uint64_t *arps;
   1017 	struct psref psref, psref_ia;
   1018 	int s;
   1019 
   1020 	if (__predict_false(m_makewritable(&m, 0, m->m_pkthdr.len, M_DONTWAIT)))
   1021 		goto out;
   1022 	ah = mtod(m, struct arphdr *);
   1023 	op = ntohs(ah->ar_op);
   1024 
   1025 	rcvif = ifp = m_get_rcvif_psref(m, &psref);
   1026 	if (__predict_false(rcvif == NULL))
   1027 		goto drop;
   1028 	/*
   1029 	 * Fix up ah->ar_hrd if necessary, before using ar_tha() or
   1030 	 * ar_tpa().
   1031 	 */
   1032 	switch (ifp->if_type) {
   1033 	case IFT_IEEE1394:
   1034 		if (ntohs(ah->ar_hrd) == ARPHRD_IEEE1394)
   1035 			;
   1036 		else {
   1037 			/* XXX this is to make sure we compute ar_tha right */
   1038 			/* XXX check ar_hrd more strictly? */
   1039 			ah->ar_hrd = htons(ARPHRD_IEEE1394);
   1040 		}
   1041 		break;
   1042 	default:
   1043 		/* XXX check ar_hrd? */
   1044 		break;
   1045 	}
   1046 
   1047 	memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
   1048 	memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
   1049 
   1050 	if (m->m_flags & (M_BCAST|M_MCAST))
   1051 		ARP_STATINC(ARP_STAT_RCVMCAST);
   1052 
   1053 
   1054 	/*
   1055 	 * Search for a matching interface address
   1056 	 * or any address on the interface to use
   1057 	 * as a dummy address in the rest of this function
   1058 	 */
   1059 	s = pserialize_read_enter();
   1060 	IN_ADDRHASH_READER_FOREACH(ia, itaddr.s_addr) {
   1061 		if (!in_hosteq(ia->ia_addr.sin_addr, itaddr))
   1062 			continue;
   1063 #if NCARP > 0
   1064 		if (ia->ia_ifp->if_type == IFT_CARP &&
   1065 		    ((ia->ia_ifp->if_flags & (IFF_UP|IFF_RUNNING)) ==
   1066 		    (IFF_UP|IFF_RUNNING))) {
   1067 			index++;
   1068 			if (ia->ia_ifp == rcvif &&
   1069 			    carp_iamatch(ia, ar_sha(ah),
   1070 			    &count, index)) {
   1071 				break;
   1072 			}
   1073 		} else
   1074 #endif
   1075 		if (ia->ia_ifp == rcvif)
   1076 			break;
   1077 #if NBRIDGE > 0
   1078 		/*
   1079 		 * If the interface we received the packet on
   1080 		 * is part of a bridge, check to see if we need
   1081 		 * to "bridge" the packet to ourselves at this
   1082 		 * layer.  Note we still prefer a perfect match,
   1083 		 * but allow this weaker match if necessary.
   1084 		 */
   1085 		if (rcvif->if_bridge != NULL &&
   1086 		    rcvif->if_bridge == ia->ia_ifp->if_bridge)
   1087 			bridge_ia = ia;
   1088 #endif /* NBRIDGE > 0 */
   1089 	}
   1090 
   1091 #if NBRIDGE > 0
   1092 	if (ia == NULL && bridge_ia != NULL) {
   1093 		ia = bridge_ia;
   1094 		m_put_rcvif_psref(rcvif, &psref);
   1095 		rcvif = NULL;
   1096 		/* FIXME */
   1097 		ifp = bridge_ia->ia_ifp;
   1098 	}
   1099 #endif
   1100 	if (ia != NULL)
   1101 		ia4_acquire(ia, &psref_ia);
   1102 	pserialize_read_exit(s);
   1103 
   1104 	if (ia == NULL) {
   1105 		ia = in_get_ia_on_iface_psref(isaddr, rcvif, &psref_ia);
   1106 		if (ia == NULL) {
   1107 			ia = in_get_ia_from_ifp_psref(ifp, &psref_ia);
   1108 			if (ia == NULL) {
   1109 				ARP_STATINC(ARP_STAT_RCVNOINT);
   1110 				goto out;
   1111 			}
   1112 		}
   1113 	}
   1114 
   1115 	myaddr = ia->ia_addr.sin_addr;
   1116 
   1117 	/* XXX checks for bridge case? */
   1118 	if (!memcmp(ar_sha(ah), CLLADDR(ifp->if_sadl), ifp->if_addrlen)) {
   1119 		ARP_STATINC(ARP_STAT_RCVLOCALSHA);
   1120 		goto out;	/* it's from me, ignore it. */
   1121 	}
   1122 
   1123 	/* XXX checks for bridge case? */
   1124 	if (!memcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
   1125 		ARP_STATINC(ARP_STAT_RCVBCASTSHA);
   1126 		log(LOG_ERR,
   1127 		    "%s: arp: link address is broadcast for IP address %s!\n",
   1128 		    ifp->if_xname, in_fmtaddr(isaddr));
   1129 		goto out;
   1130 	}
   1131 
   1132 	/*
   1133 	 * If the source IP address is zero, this is an RFC 5227 ARP probe
   1134 	 */
   1135 	if (in_nullhost(isaddr))
   1136 		ARP_STATINC(ARP_STAT_RCVZEROSPA);
   1137 	else if (in_hosteq(isaddr, myaddr))
   1138 		ARP_STATINC(ARP_STAT_RCVLOCALSPA);
   1139 
   1140 	/*
   1141 	 * If the target IP address is zero, ignore the packet.
   1142 	 * This prevents the code below from tring to answer
   1143 	 * when we are using IP address zero (booting).
   1144 	 */
   1145 	if (in_nullhost(itaddr)) {
   1146 		ARP_STATINC(ARP_STAT_RCVZEROTPA);
   1147 		goto out;
   1148 	}
   1149 
   1150 	/* DAD check, RFC 5227 */
   1151 	if (in_hosteq(isaddr, myaddr) ||
   1152 	    (in_nullhost(isaddr) && in_hosteq(itaddr, myaddr)))
   1153 	{
   1154 		arp_dad_duplicated((struct ifaddr *)ia,
   1155 		    lla_snprintf(ar_sha(ah), ah->ar_hln));
   1156 		goto out;
   1157 	}
   1158 
   1159 	if (in_nullhost(isaddr))
   1160 		goto reply;
   1161 
   1162 	if (in_hosteq(itaddr, myaddr))
   1163 		la = arpcreate(ifp, m, &isaddr, NULL, 1);
   1164 	else
   1165 		la = arplookup(ifp, m, &isaddr, NULL, 1);
   1166 	if (la == NULL)
   1167 		goto reply;
   1168 
   1169 	if ((la->la_flags & LLE_VALID) &&
   1170 	    memcmp(ar_sha(ah), &la->ll_addr, ifp->if_addrlen)) {
   1171 		if (la->la_flags & LLE_STATIC) {
   1172 			ARP_STATINC(ARP_STAT_RCVOVERPERM);
   1173 			if (!log_permanent_modify)
   1174 				goto out;
   1175 			log(LOG_INFO,
   1176 			    "%s tried to overwrite permanent arp info"
   1177 			    " for %s\n",
   1178 			    lla_snprintf(ar_sha(ah), ah->ar_hln),
   1179 			    in_fmtaddr(isaddr));
   1180 			goto out;
   1181 		} else if (la->lle_tbl->llt_ifp != ifp) {
   1182 			/* XXX should not happen? */
   1183 			ARP_STATINC(ARP_STAT_RCVOVERINT);
   1184 			if (!log_wrong_iface)
   1185 				goto out;
   1186 			log(LOG_INFO,
   1187 			    "%s on %s tried to overwrite "
   1188 			    "arp info for %s on %s\n",
   1189 			    lla_snprintf(ar_sha(ah), ah->ar_hln),
   1190 			    ifp->if_xname, in_fmtaddr(isaddr),
   1191 			    la->lle_tbl->llt_ifp->if_xname);
   1192 				goto out;
   1193 		} else {
   1194 			ARP_STATINC(ARP_STAT_RCVOVER);
   1195 			if (log_movements)
   1196 				log(LOG_INFO, "arp info overwritten "
   1197 				    "for %s by %s\n",
   1198 				    in_fmtaddr(isaddr),
   1199 				    lla_snprintf(ar_sha(ah),
   1200 				    ah->ar_hln));
   1201 		}
   1202 	}
   1203 
   1204 	/* XXX llentry should have addrlen? */
   1205 #if 0
   1206 	/*
   1207 	 * sanity check for the address length.
   1208 	 * XXX this does not work for protocols with variable address
   1209 	 * length. -is
   1210 	 */
   1211 	if (sdl->sdl_alen && sdl->sdl_alen != ah->ar_hln) {
   1212 		ARP_STATINC(ARP_STAT_RCVLENCHG);
   1213 		log(LOG_WARNING,
   1214 		    "arp from %s: new addr len %d, was %d\n",
   1215 		    in_fmtaddr(isaddr), ah->ar_hln, sdl->sdl_alen);
   1216 	}
   1217 #endif
   1218 
   1219 	if (ifp->if_addrlen != ah->ar_hln) {
   1220 		ARP_STATINC(ARP_STAT_RCVBADLEN);
   1221 		log(LOG_WARNING,
   1222 		    "arp from %s: addr len: new %d, i/f %d (ignored)\n",
   1223 		    in_fmtaddr(isaddr), ah->ar_hln,
   1224 		    ifp->if_addrlen);
   1225 		goto reply;
   1226 	}
   1227 
   1228 #if NTOKEN > 0
   1229 	/*
   1230 	 * XXX uses m_data and assumes the complete answer including
   1231 	 * XXX token-ring headers is in the same buf
   1232 	 */
   1233 	if (ifp->if_type == IFT_ISO88025) {
   1234 		struct token_header *trh;
   1235 
   1236 		trh = (struct token_header *)M_TRHSTART(m);
   1237 		if (trh->token_shost[0] & TOKEN_RI_PRESENT) {
   1238 			struct token_rif *rif;
   1239 			size_t riflen;
   1240 
   1241 			rif = TOKEN_RIF(trh);
   1242 			riflen = (ntohs(rif->tr_rcf) &
   1243 			    TOKEN_RCF_LEN_MASK) >> 8;
   1244 
   1245 			if (riflen > 2 &&
   1246 			    riflen < sizeof(struct token_rif) &&
   1247 			    (riflen & 1) == 0) {
   1248 				rif->tr_rcf ^= htons(TOKEN_RCF_DIRECTION);
   1249 				rif->tr_rcf &= htons(~TOKEN_RCF_BROADCAST_MASK);
   1250 				memcpy(TOKEN_RIF_LLE(la), rif, riflen);
   1251 			}
   1252 		}
   1253 	}
   1254 #endif /* NTOKEN > 0 */
   1255 
   1256 	KASSERT(sizeof(la->ll_addr) >= ifp->if_addrlen);
   1257 	(void)memcpy(&la->ll_addr, ar_sha(ah), ifp->if_addrlen);
   1258 	la->la_flags |= LLE_VALID;
   1259 	if ((la->la_flags & LLE_STATIC) == 0) {
   1260 		la->la_expire = time_uptime + arpt_keep;
   1261 		arp_settimer(la, arpt_keep);
   1262 	}
   1263 	la->la_asked = 0;
   1264 	/* rt->rt_flags &= ~RTF_REJECT; */
   1265 
   1266 	if (la->la_hold != NULL) {
   1267 		int n = la->la_numheld;
   1268 		struct mbuf *m_hold, *m_hold_next;
   1269 		struct sockaddr_in sin;
   1270 
   1271 		sockaddr_in_init(&sin, &la->r_l3addr.addr4, 0);
   1272 
   1273 		m_hold = la->la_hold;
   1274 		la->la_hold = NULL;
   1275 		la->la_numheld = 0;
   1276 		/*
   1277 		 * We have to unlock here because if_output would call
   1278 		 * arpresolve
   1279 		 */
   1280 		LLE_WUNLOCK(la);
   1281 		ARP_STATADD(ARP_STAT_DFRSENT, n);
   1282 		ARP_STATADD(ARP_STAT_DFRTOTAL, n);
   1283 		for (; m_hold != NULL; m_hold = m_hold_next) {
   1284 			m_hold_next = m_hold->m_nextpkt;
   1285 			m_hold->m_nextpkt = NULL;
   1286 			if_output_lock(ifp, ifp, m_hold, sintosa(&sin), NULL);
   1287 		}
   1288 	} else
   1289 		LLE_WUNLOCK(la);
   1290 	la = NULL;
   1291 
   1292 reply:
   1293 	if (la != NULL) {
   1294 		LLE_WUNLOCK(la);
   1295 		la = NULL;
   1296 	}
   1297 	if (op != ARPOP_REQUEST) {
   1298 		if (op == ARPOP_REPLY)
   1299 			ARP_STATINC(ARP_STAT_RCVREPLY);
   1300 		goto out;
   1301 	}
   1302 	ARP_STATINC(ARP_STAT_RCVREQUEST);
   1303 	if (in_hosteq(itaddr, myaddr)) {
   1304 		/* If our address is unuseable, don't reply */
   1305 		if (ia->ia4_flags & (IN_IFF_NOTREADY | IN_IFF_DETACHED))
   1306 			goto out;
   1307 		/* I am the target */
   1308 		tha = ar_tha(ah);
   1309 		if (tha)
   1310 			memcpy(tha, ar_sha(ah), ah->ar_hln);
   1311 		memcpy(ar_sha(ah), CLLADDR(ifp->if_sadl), ah->ar_hln);
   1312 	} else {
   1313 		/* Proxy ARP */
   1314 		struct llentry *lle = NULL;
   1315 		struct sockaddr_in sin;
   1316 #if NCARP > 0
   1317 		struct ifnet *_rcvif = m_get_rcvif(m, &s);
   1318 		if (ifp->if_type == IFT_CARP && _rcvif->if_type != IFT_CARP)
   1319 			goto out;
   1320 		m_put_rcvif(_rcvif, &s);
   1321 #endif
   1322 
   1323 		tha = ar_tha(ah);
   1324 
   1325 		sockaddr_in_init(&sin, &itaddr, 0);
   1326 
   1327 		IF_AFDATA_RLOCK(ifp);
   1328 		lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
   1329 		IF_AFDATA_RUNLOCK(ifp);
   1330 
   1331 		if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
   1332 			(void)memcpy(tha, ar_sha(ah), ah->ar_hln);
   1333 			(void)memcpy(ar_sha(ah), &lle->ll_addr, ah->ar_hln);
   1334 			LLE_RUNLOCK(lle);
   1335 		} else {
   1336 			if (lle != NULL)
   1337 				LLE_RUNLOCK(lle);
   1338 			goto drop;
   1339 		}
   1340 	}
   1341 	ia4_release(ia, &psref_ia);
   1342 
   1343 	memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
   1344 	memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
   1345 	ah->ar_op = htons(ARPOP_REPLY);
   1346 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
   1347 	switch (ifp->if_type) {
   1348 	case IFT_IEEE1394:
   1349 		/*
   1350 		 * ieee1394 arp reply is broadcast
   1351 		 */
   1352 		m->m_flags &= ~M_MCAST;
   1353 		m->m_flags |= M_BCAST;
   1354 		m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + ah->ar_hln;
   1355 		break;
   1356 	default:
   1357 		m->m_flags &= ~(M_BCAST|M_MCAST); /* never reply by broadcast */
   1358 		m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
   1359 		break;
   1360 	}
   1361 	m->m_pkthdr.len = m->m_len;
   1362 	sa.sa_family = AF_ARP;
   1363 	sa.sa_len = 2;
   1364 	arps = ARP_STAT_GETREF();
   1365 	arps[ARP_STAT_SNDTOTAL]++;
   1366 	arps[ARP_STAT_SNDREPLY]++;
   1367 	ARP_STAT_PUTREF();
   1368 	if_output_lock(ifp, ifp, m, &sa, NULL);
   1369 	if (rcvif != NULL)
   1370 		m_put_rcvif_psref(rcvif, &psref);
   1371 	return;
   1372 
   1373 out:
   1374 	if (la != NULL)
   1375 		LLE_WUNLOCK(la);
   1376 drop:
   1377 	if (ia != NULL)
   1378 		ia4_release(ia, &psref_ia);
   1379 	if (rcvif != NULL)
   1380 		m_put_rcvif_psref(rcvif, &psref);
   1381 	m_freem(m);
   1382 }
   1383 
   1384 /*
   1385  * Lookup or a new address in arptab.
   1386  */
   1387 static struct llentry *
   1388 arplookup(struct ifnet *ifp, struct mbuf *m, const struct in_addr *addr,
   1389     const struct sockaddr *sa, int wlock)
   1390 {
   1391 	struct sockaddr_in sin;
   1392 	struct llentry *la;
   1393 	int flags = wlock ? LLE_EXCLUSIVE : 0;
   1394 
   1395 
   1396 	if (sa == NULL) {
   1397 		KASSERT(addr != NULL);
   1398 		sockaddr_in_init(&sin, addr, 0);
   1399 		sa = sintocsa(&sin);
   1400 	}
   1401 
   1402 	IF_AFDATA_RLOCK(ifp);
   1403 	la = lla_lookup(LLTABLE(ifp), flags, sa);
   1404 	IF_AFDATA_RUNLOCK(ifp);
   1405 
   1406 	return la;
   1407 }
   1408 
   1409 static struct llentry *
   1410 arpcreate(struct ifnet *ifp, struct mbuf *m, const struct in_addr *addr,
   1411     const struct sockaddr *sa, int wlock)
   1412 {
   1413 	struct sockaddr_in sin;
   1414 	struct llentry *la;
   1415 	int flags = wlock ? LLE_EXCLUSIVE : 0;
   1416 
   1417 	if (sa == NULL) {
   1418 		KASSERT(addr != NULL);
   1419 		sockaddr_in_init(&sin, addr, 0);
   1420 		sa = sintocsa(&sin);
   1421 	}
   1422 
   1423 	la = arplookup(ifp, m, addr, sa, wlock);
   1424 
   1425 	if (la == NULL) {
   1426 		IF_AFDATA_WLOCK(ifp);
   1427 		la = lla_create(LLTABLE(ifp), flags, sa);
   1428 		IF_AFDATA_WUNLOCK(ifp);
   1429 
   1430 		if (la != NULL)
   1431 			arp_init_llentry(ifp, la);
   1432 	}
   1433 
   1434 	return la;
   1435 }
   1436 
   1437 int
   1438 arpioctl(u_long cmd, void *data)
   1439 {
   1440 
   1441 	return EOPNOTSUPP;
   1442 }
   1443 
   1444 void
   1445 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
   1446 {
   1447 	struct in_addr *ip;
   1448 	struct in_ifaddr *ia = (struct in_ifaddr *)ifa;
   1449 
   1450 	/*
   1451 	 * Warn the user if another station has this IP address,
   1452 	 * but only if the interface IP address is not zero.
   1453 	 */
   1454 	ip = &IA_SIN(ifa)->sin_addr;
   1455 	if (!in_nullhost(*ip) &&
   1456 	    (ia->ia4_flags & (IN_IFF_NOTREADY | IN_IFF_DETACHED)) == 0) {
   1457 		struct llentry *lle;
   1458 
   1459 		/*
   1460 		 * interface address is considered static entry
   1461 		 * because the output of the arp utility shows
   1462 		 * that L2 entry as permanent
   1463 		 */
   1464 		IF_AFDATA_WLOCK(ifp);
   1465 		lle = lla_create(LLTABLE(ifp), (LLE_IFADDR | LLE_STATIC),
   1466 				 (struct sockaddr *)IA_SIN(ifa));
   1467 		IF_AFDATA_WUNLOCK(ifp);
   1468 		if (lle == NULL)
   1469 			log(LOG_INFO, "%s: cannot create arp entry for"
   1470 			    " interface address\n", __func__);
   1471 		else {
   1472 			arp_init_llentry(ifp, lle);
   1473 			LLE_RUNLOCK(lle);
   1474 		}
   1475 	}
   1476 
   1477 	ifa->ifa_rtrequest = arp_rtrequest;
   1478 	ifa->ifa_flags |= RTF_CONNECTED;
   1479 
   1480 	/* ARP will handle DAD for this address. */
   1481 	if (in_nullhost(*ip)) {
   1482 		if (ia->ia_dad_stop != NULL)	/* safety */
   1483 			ia->ia_dad_stop(ifa);
   1484 		ia->ia_dad_start = NULL;
   1485 		ia->ia_dad_stop = NULL;
   1486 		ia->ia4_flags &= ~IN_IFF_TENTATIVE;
   1487 	} else {
   1488 		ia->ia_dad_start = arp_dad_start;
   1489 		ia->ia_dad_stop = arp_dad_stop;
   1490 		if (ia->ia4_flags & IN_IFF_TRYTENTATIVE)
   1491 			ia->ia4_flags |= IN_IFF_TENTATIVE;
   1492 		else
   1493 			arpannounce1(ifa);
   1494 	}
   1495 }
   1496 
   1497 TAILQ_HEAD(dadq_head, dadq);
   1498 struct dadq {
   1499 	TAILQ_ENTRY(dadq) dad_list;
   1500 	struct ifaddr *dad_ifa;
   1501 	int dad_count;		/* max ARP to send */
   1502 	int dad_arp_tcount;	/* # of trials to send ARP */
   1503 	int dad_arp_ocount;	/* ARP sent so far */
   1504 	int dad_arp_announce;	/* max ARP announcements */
   1505 	int dad_arp_acount;	/* # of announcements */
   1506 	struct callout dad_timer_ch;
   1507 };
   1508 MALLOC_JUSTDEFINE(M_IPARP, "ARP DAD", "ARP DAD Structure");
   1509 
   1510 static struct dadq_head dadq;
   1511 static int dad_init = 0;
   1512 static int dad_maxtry = 15;     /* max # of *tries* to transmit DAD packet */
   1513 static kmutex_t arp_dad_lock;
   1514 
   1515 static struct dadq *
   1516 arp_dad_find(struct ifaddr *ifa)
   1517 {
   1518 	struct dadq *dp;
   1519 
   1520 	KASSERT(mutex_owned(&arp_dad_lock));
   1521 
   1522 	TAILQ_FOREACH(dp, &dadq, dad_list) {
   1523 		if (dp->dad_ifa == ifa)
   1524 			return dp;
   1525 	}
   1526 	return NULL;
   1527 }
   1528 
   1529 static void
   1530 arp_dad_starttimer(struct dadq *dp, int ticks)
   1531 {
   1532 
   1533 	callout_reset(&dp->dad_timer_ch, ticks,
   1534 	    (void (*)(void *))arp_dad_timer, (void *)dp->dad_ifa);
   1535 }
   1536 
   1537 static void
   1538 arp_dad_stoptimer(struct dadq *dp)
   1539 {
   1540 
   1541 #ifdef NET_MPSAFE
   1542 	callout_halt(&dp->dad_timer_ch, NULL);
   1543 #else
   1544 	callout_halt(&dp->dad_timer_ch, softnet_lock);
   1545 #endif
   1546 }
   1547 
   1548 static void
   1549 arp_dad_output(struct dadq *dp, struct ifaddr *ifa)
   1550 {
   1551 	struct in_ifaddr *ia = (struct in_ifaddr *)ifa;
   1552 	struct ifnet *ifp = ifa->ifa_ifp;
   1553 	struct in_addr sip;
   1554 
   1555 	dp->dad_arp_tcount++;
   1556 	if ((ifp->if_flags & IFF_UP) == 0)
   1557 		return;
   1558 	if ((ifp->if_flags & IFF_RUNNING) == 0)
   1559 		return;
   1560 
   1561 	dp->dad_arp_tcount = 0;
   1562 	dp->dad_arp_ocount++;
   1563 
   1564 	memset(&sip, 0, sizeof(sip));
   1565 	arprequest(ifa->ifa_ifp, &sip, &ia->ia_addr.sin_addr,
   1566 	    CLLADDR(ifa->ifa_ifp->if_sadl));
   1567 }
   1568 
   1569 /*
   1570  * Start Duplicate Address Detection (DAD) for specified interface address.
   1571  */
   1572 static void
   1573 arp_dad_start(struct ifaddr *ifa)
   1574 {
   1575 	struct in_ifaddr *ia = (struct in_ifaddr *)ifa;
   1576 	struct dadq *dp;
   1577 
   1578 	if (!dad_init) {
   1579 		TAILQ_INIT(&dadq);
   1580 		mutex_init(&arp_dad_lock, MUTEX_DEFAULT, IPL_NONE);
   1581 		dad_init++;
   1582 	}
   1583 
   1584 	/*
   1585 	 * If we don't need DAD, don't do it.
   1586 	 * - DAD is disabled (ip_dad_count == 0)
   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_fmtaddr(ia->ia_addr.sin_addr),
   1592 		    ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
   1593 		return;
   1594 	}
   1595 	if (!ip_dad_count) {
   1596 		ia->ia4_flags &= ~IN_IFF_TENTATIVE;
   1597 		rt_newaddrmsg(RTM_NEWADDR, ifa, 0, NULL);
   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 	mutex_enter(&arp_dad_lock);
   1606 	if (arp_dad_find(ifa) != NULL) {
   1607 		mutex_exit(&arp_dad_lock);
   1608 		/* DAD already in progress */
   1609 		return;
   1610 	}
   1611 
   1612 	dp = malloc(sizeof(*dp), M_IPARP, M_NOWAIT);
   1613 	if (dp == NULL) {
   1614 		mutex_exit(&arp_dad_lock);
   1615 		log(LOG_ERR, "%s: memory allocation failed for %s(%s)\n",
   1616 		    __func__, in_fmtaddr(ia->ia_addr.sin_addr),
   1617 		    ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
   1618 		return;
   1619 	}
   1620 	memset(dp, 0, sizeof(*dp));
   1621 	callout_init(&dp->dad_timer_ch, CALLOUT_MPSAFE);
   1622 
   1623 	/*
   1624 	 * Send ARP packet for DAD, ip_dad_count times.
   1625 	 * Note that we must delay the first transmission.
   1626 	 */
   1627 	dp->dad_ifa = ifa;
   1628 	ifaref(ifa);	/* just for safety */
   1629 	dp->dad_count = ip_dad_count;
   1630 	dp->dad_arp_announce = 0; /* Will be set when starting to announce */
   1631 	dp->dad_arp_acount = dp->dad_arp_ocount = dp->dad_arp_tcount = 0;
   1632 	TAILQ_INSERT_TAIL(&dadq, (struct dadq *)dp, dad_list);
   1633 
   1634 	arplog(LOG_DEBUG, "%s: starting DAD for %s\n", if_name(ifa->ifa_ifp),
   1635 	    in_fmtaddr(ia->ia_addr.sin_addr));
   1636 
   1637 	arp_dad_starttimer(dp, cprng_fast32() % (PROBE_WAIT * hz));
   1638 
   1639 	mutex_exit(&arp_dad_lock);
   1640 }
   1641 
   1642 /*
   1643  * terminate DAD unconditionally.  used for address removals.
   1644  */
   1645 static void
   1646 arp_dad_stop(struct ifaddr *ifa)
   1647 {
   1648 	struct dadq *dp;
   1649 
   1650 	if (!dad_init)
   1651 		return;
   1652 
   1653 	mutex_enter(&arp_dad_lock);
   1654 	dp = arp_dad_find(ifa);
   1655 	if (dp == NULL) {
   1656 		mutex_exit(&arp_dad_lock);
   1657 		/* DAD wasn't started yet */
   1658 		return;
   1659 	}
   1660 
   1661 	/* Prevent the timer from running anymore. */
   1662 	TAILQ_REMOVE(&dadq, dp, dad_list);
   1663 	mutex_exit(&arp_dad_lock);
   1664 
   1665 	arp_dad_stoptimer(dp);
   1666 
   1667 	free(dp, M_IPARP);
   1668 	dp = NULL;
   1669 	ifafree(ifa);
   1670 }
   1671 
   1672 static void
   1673 arp_dad_timer(struct ifaddr *ifa)
   1674 {
   1675 	struct in_ifaddr *ia = (struct in_ifaddr *)ifa;
   1676 	struct dadq *dp;
   1677 
   1678 	mutex_enter(softnet_lock);
   1679 	KERNEL_LOCK(1, NULL);
   1680 	mutex_enter(&arp_dad_lock);
   1681 
   1682 	/* Sanity check */
   1683 	if (ia == NULL) {
   1684 		log(LOG_ERR, "%s: called with null parameter\n", __func__);
   1685 		goto done;
   1686 	}
   1687 	dp = arp_dad_find(ifa);
   1688 	if (dp == NULL) {
   1689 		/* DAD seems to be stopping, so do nothing. */
   1690 		goto done;
   1691 	}
   1692 	if (ia->ia4_flags & IN_IFF_DUPLICATED) {
   1693 		log(LOG_ERR, "%s: called with duplicate address %s(%s)\n",
   1694 		    __func__, in_fmtaddr(ia->ia_addr.sin_addr),
   1695 		    ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
   1696 		goto done;
   1697 	}
   1698 	if ((ia->ia4_flags & IN_IFF_TENTATIVE) == 0 && dp->dad_arp_acount == 0)
   1699 	{
   1700 		log(LOG_ERR, "%s: called with non-tentative address %s(%s)\n",
   1701 		    __func__, in_fmtaddr(ia->ia_addr.sin_addr),
   1702 		    ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
   1703 		goto done;
   1704 	}
   1705 
   1706 	/* timeouted with IFF_{RUNNING,UP} check */
   1707 	if (dp->dad_arp_tcount > dad_maxtry) {
   1708 		arplog(LOG_INFO, "%s: could not run DAD, driver problem?\n",
   1709 		    if_name(ifa->ifa_ifp));
   1710 
   1711 		TAILQ_REMOVE(&dadq, dp, dad_list);
   1712 		free(dp, M_IPARP);
   1713 		dp = NULL;
   1714 		ifafree(ifa);
   1715 		goto done;
   1716 	}
   1717 
   1718 	/* Need more checks? */
   1719 	if (dp->dad_arp_ocount < dp->dad_count) {
   1720 		int adelay;
   1721 
   1722 		/*
   1723 		 * We have more ARP to go.  Send ARP packet for DAD.
   1724 		 */
   1725 		arp_dad_output(dp, ifa);
   1726 		if (dp->dad_arp_ocount < dp->dad_count)
   1727 			adelay = (PROBE_MIN * hz) +
   1728 			    (cprng_fast32() %
   1729 			    ((PROBE_MAX * hz) - (PROBE_MIN * hz)));
   1730 		else
   1731 			adelay = ANNOUNCE_WAIT * hz;
   1732 		arp_dad_starttimer(dp, adelay);
   1733 		goto done;
   1734 	} else if (dp->dad_arp_acount == 0) {
   1735 		/*
   1736 		 * We are done with DAD.
   1737 		 * No duplicate address found.
   1738 		 */
   1739 		ia->ia4_flags &= ~IN_IFF_TENTATIVE;
   1740 		rt_newaddrmsg(RTM_NEWADDR, ifa, 0, NULL);
   1741 		arplog(LOG_DEBUG,
   1742 		    "%s: DAD complete for %s - no duplicates found\n",
   1743 		    if_name(ifa->ifa_ifp),
   1744 		    in_fmtaddr(ia->ia_addr.sin_addr));
   1745 		dp->dad_arp_announce = ANNOUNCE_NUM;
   1746 		goto announce;
   1747 	} else if (dp->dad_arp_acount < dp->dad_arp_announce) {
   1748 announce:
   1749 		/*
   1750 		 * Announce the address.
   1751 		 */
   1752 		arpannounce1(ifa);
   1753 		dp->dad_arp_acount++;
   1754 		if (dp->dad_arp_acount < dp->dad_arp_announce) {
   1755 			arp_dad_starttimer(dp, ANNOUNCE_INTERVAL * hz);
   1756 			goto done;
   1757 		}
   1758 		arplog(LOG_DEBUG,
   1759 		    "%s: ARP announcement complete for %s\n",
   1760 		    if_name(ifa->ifa_ifp),
   1761 		    in_fmtaddr(ia->ia_addr.sin_addr));
   1762 	}
   1763 
   1764 	TAILQ_REMOVE(&dadq, dp, dad_list);
   1765 	free(dp, M_IPARP);
   1766 	dp = NULL;
   1767 	ifafree(ifa);
   1768 
   1769 done:
   1770 	mutex_exit(&arp_dad_lock);
   1771 	KERNEL_UNLOCK_ONE(NULL);
   1772 	mutex_exit(softnet_lock);
   1773 }
   1774 
   1775 static void
   1776 arp_dad_duplicated(struct ifaddr *ifa, const char *sha)
   1777 {
   1778 	struct in_ifaddr *ia = (struct in_ifaddr *)ifa;
   1779 	struct ifnet *ifp = ifa->ifa_ifp;
   1780 	const char *iastr = in_fmtaddr(ia->ia_addr.sin_addr);
   1781 
   1782 	if (ia->ia4_flags & (IN_IFF_TENTATIVE|IN_IFF_DUPLICATED)) {
   1783 		log(LOG_ERR,
   1784 		    "%s: DAD duplicate address %s from %s\n",
   1785 		    if_name(ifp), iastr, sha);
   1786 	} else if (ia->ia_dad_defended == 0 ||
   1787 		   ia->ia_dad_defended < time_uptime - DEFEND_INTERVAL) {
   1788 		ia->ia_dad_defended = time_uptime;
   1789 		arpannounce1(ifa);
   1790 		log(LOG_ERR,
   1791 		    "%s: DAD defended address %s from %s\n",
   1792 		    if_name(ifp), iastr, sha);
   1793 		return;
   1794 	} else {
   1795 		/* If DAD is disabled, just report the duplicate. */
   1796 		if (ip_dad_count == 0) {
   1797 			log(LOG_ERR,
   1798 			    "%s: DAD ignoring duplicate address %s from %s\n",
   1799 			    if_name(ifp), iastr, sha);
   1800 			return;
   1801 		}
   1802 		log(LOG_ERR,
   1803 		    "%s: DAD defence failed for %s from %s\n",
   1804 		    if_name(ifp), iastr, sha);
   1805 	}
   1806 
   1807 	arp_dad_stop(ifa);
   1808 
   1809 	ia->ia4_flags &= ~IN_IFF_TENTATIVE;
   1810 	if ((ia->ia4_flags & IN_IFF_DUPLICATED) == 0) {
   1811 		ia->ia4_flags |= IN_IFF_DUPLICATED;
   1812 		/* Inform the routing socket of the duplicate address */
   1813 		rt_newaddrmsg(RTM_NEWADDR, ifa, 0, NULL);
   1814 	}
   1815 }
   1816 
   1817 /*
   1818  * Called from 10 Mb/s Ethernet interrupt handlers
   1819  * when ether packet type ETHERTYPE_REVARP
   1820  * is received.  Common length and type checks are done here,
   1821  * then the protocol-specific routine is called.
   1822  */
   1823 void
   1824 revarpinput(struct mbuf *m)
   1825 {
   1826 	struct arphdr *ar;
   1827 
   1828 	if (m->m_len < sizeof(struct arphdr))
   1829 		goto out;
   1830 	ar = mtod(m, struct arphdr *);
   1831 #if 0 /* XXX I don't think we need this... and it will prevent other LL */
   1832 	if (ntohs(ar->ar_hrd) != ARPHRD_ETHER)
   1833 		goto out;
   1834 #endif
   1835 	if (m->m_len < sizeof(struct arphdr) + 2 * (ar->ar_hln + ar->ar_pln))
   1836 		goto out;
   1837 	switch (ntohs(ar->ar_pro)) {
   1838 	case ETHERTYPE_IP:
   1839 	case ETHERTYPE_IPTRAILERS:
   1840 		in_revarpinput(m);
   1841 		return;
   1842 
   1843 	default:
   1844 		break;
   1845 	}
   1846 out:
   1847 	m_freem(m);
   1848 }
   1849 
   1850 /*
   1851  * RARP for Internet protocols on 10 Mb/s Ethernet.
   1852  * Algorithm is that given in RFC 903.
   1853  * We are only using for bootstrap purposes to get an ip address for one of
   1854  * our interfaces.  Thus we support no user-interface.
   1855  *
   1856  * Since the contents of the RARP reply are specific to the interface that
   1857  * sent the request, this code must ensure that they are properly associated.
   1858  *
   1859  * Note: also supports ARP via RARP packets, per the RFC.
   1860  */
   1861 void
   1862 in_revarpinput(struct mbuf *m)
   1863 {
   1864 	struct arphdr *ah;
   1865 	void *tha;
   1866 	int op;
   1867 	struct ifnet *rcvif;
   1868 	int s;
   1869 
   1870 	ah = mtod(m, struct arphdr *);
   1871 	op = ntohs(ah->ar_op);
   1872 
   1873 	rcvif = m_get_rcvif(m, &s);
   1874 	switch (rcvif->if_type) {
   1875 	case IFT_IEEE1394:
   1876 		/* ARP without target hardware address is not supported */
   1877 		goto out;
   1878 	default:
   1879 		break;
   1880 	}
   1881 
   1882 	switch (op) {
   1883 	case ARPOP_REQUEST:
   1884 	case ARPOP_REPLY:	/* per RFC */
   1885 		m_put_rcvif(rcvif, &s);
   1886 		in_arpinput(m);
   1887 		return;
   1888 	case ARPOP_REVREPLY:
   1889 		break;
   1890 	case ARPOP_REVREQUEST:	/* handled by rarpd(8) */
   1891 	default:
   1892 		goto out;
   1893 	}
   1894 	if (!revarp_in_progress)
   1895 		goto out;
   1896 	if (rcvif != myip_ifp) /* !same interface */
   1897 		goto out;
   1898 	if (myip_initialized)
   1899 		goto wake;
   1900 	tha = ar_tha(ah);
   1901 	if (tha == NULL)
   1902 		goto out;
   1903 	if (memcmp(tha, CLLADDR(rcvif->if_sadl), rcvif->if_sadl->sdl_alen))
   1904 		goto out;
   1905 	memcpy(&srv_ip, ar_spa(ah), sizeof(srv_ip));
   1906 	memcpy(&myip, ar_tpa(ah), sizeof(myip));
   1907 	myip_initialized = 1;
   1908 wake:	/* Do wakeup every time in case it was missed. */
   1909 	wakeup((void *)&myip);
   1910 
   1911 out:
   1912 	m_put_rcvif(rcvif, &s);
   1913 	m_freem(m);
   1914 }
   1915 
   1916 /*
   1917  * Send a RARP request for the ip address of the specified interface.
   1918  * The request should be RFC 903-compliant.
   1919  */
   1920 static void
   1921 revarprequest(struct ifnet *ifp)
   1922 {
   1923 	struct sockaddr sa;
   1924 	struct mbuf *m;
   1925 	struct arphdr *ah;
   1926 	void *tha;
   1927 
   1928 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
   1929 		return;
   1930 	MCLAIM(m, &arpdomain.dom_mowner);
   1931 	m->m_len = sizeof(*ah) + 2*sizeof(struct in_addr) +
   1932 	    2*ifp->if_addrlen;
   1933 	m->m_pkthdr.len = m->m_len;
   1934 	MH_ALIGN(m, m->m_len);
   1935 	ah = mtod(m, struct arphdr *);
   1936 	memset(ah, 0, m->m_len);
   1937 	ah->ar_pro = htons(ETHERTYPE_IP);
   1938 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
   1939 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
   1940 	ah->ar_op = htons(ARPOP_REVREQUEST);
   1941 
   1942 	memcpy(ar_sha(ah), CLLADDR(ifp->if_sadl), ah->ar_hln);
   1943 	tha = ar_tha(ah);
   1944 	if (tha == NULL) {
   1945 		m_free(m);
   1946 		return;
   1947 	}
   1948 	memcpy(tha, CLLADDR(ifp->if_sadl), ah->ar_hln);
   1949 
   1950 	sa.sa_family = AF_ARP;
   1951 	sa.sa_len = 2;
   1952 	m->m_flags |= M_BCAST;
   1953 
   1954 	if_output_lock(ifp, ifp, m, &sa, NULL);
   1955 }
   1956 
   1957 /*
   1958  * RARP for the ip address of the specified interface, but also
   1959  * save the ip address of the server that sent the answer.
   1960  * Timeout if no response is received.
   1961  */
   1962 int
   1963 revarpwhoarewe(struct ifnet *ifp, struct in_addr *serv_in,
   1964     struct in_addr *clnt_in)
   1965 {
   1966 	int result, count = 20;
   1967 
   1968 	myip_initialized = 0;
   1969 	myip_ifp = ifp;
   1970 
   1971 	revarp_in_progress = 1;
   1972 	while (count--) {
   1973 		revarprequest(ifp);
   1974 		result = tsleep((void *)&myip, PSOCK, "revarp", hz/2);
   1975 		if (result != EWOULDBLOCK)
   1976 			break;
   1977 	}
   1978 	revarp_in_progress = 0;
   1979 
   1980 	if (!myip_initialized)
   1981 		return ENETUNREACH;
   1982 
   1983 	memcpy(serv_in, &srv_ip, sizeof(*serv_in));
   1984 	memcpy(clnt_in, &myip, sizeof(*clnt_in));
   1985 	return 0;
   1986 }
   1987 
   1988 void
   1989 arp_stat_add(int type, uint64_t count)
   1990 {
   1991 	ARP_STATADD(type, count);
   1992 }
   1993 
   1994 static int
   1995 sysctl_net_inet_arp_stats(SYSCTLFN_ARGS)
   1996 {
   1997 
   1998 	return NETSTAT_SYSCTL(arpstat_percpu, ARP_NSTATS);
   1999 }
   2000 
   2001 static void
   2002 sysctl_net_inet_arp_setup(struct sysctllog **clog)
   2003 {
   2004 	const struct sysctlnode *node;
   2005 
   2006 	sysctl_createv(clog, 0, NULL, NULL,
   2007 			CTLFLAG_PERMANENT,
   2008 			CTLTYPE_NODE, "inet", NULL,
   2009 			NULL, 0, NULL, 0,
   2010 			CTL_NET, PF_INET, CTL_EOL);
   2011 	sysctl_createv(clog, 0, NULL, &node,
   2012 			CTLFLAG_PERMANENT,
   2013 			CTLTYPE_NODE, "arp",
   2014 			SYSCTL_DESCR("Address Resolution Protocol"),
   2015 			NULL, 0, NULL, 0,
   2016 			CTL_NET, PF_INET, CTL_CREATE, CTL_EOL);
   2017 
   2018 	sysctl_createv(clog, 0, NULL, NULL,
   2019 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2020 			CTLTYPE_INT, "keep",
   2021 			SYSCTL_DESCR("Valid ARP entry lifetime in seconds"),
   2022 			NULL, 0, &arpt_keep, 0,
   2023 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2024 
   2025 	sysctl_createv(clog, 0, NULL, NULL,
   2026 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2027 			CTLTYPE_INT, "down",
   2028 			SYSCTL_DESCR("Failed ARP entry lifetime in seconds"),
   2029 			NULL, 0, &arpt_down, 0,
   2030 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2031 
   2032 	sysctl_createv(clog, 0, NULL, NULL,
   2033 			CTLFLAG_PERMANENT,
   2034 			CTLTYPE_STRUCT, "stats",
   2035 			SYSCTL_DESCR("ARP statistics"),
   2036 			sysctl_net_inet_arp_stats, 0, NULL, 0,
   2037 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2038 
   2039 	sysctl_createv(clog, 0, NULL, NULL,
   2040 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2041 			CTLTYPE_INT, "log_movements",
   2042 			SYSCTL_DESCR("log ARP replies from MACs different than"
   2043 			    " the one in the cache"),
   2044 			NULL, 0, &log_movements, 0,
   2045 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2046 
   2047 	sysctl_createv(clog, 0, NULL, NULL,
   2048 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2049 			CTLTYPE_INT, "log_permanent_modify",
   2050 			SYSCTL_DESCR("log ARP replies from MACs different than"
   2051 			    " the one in the permanent arp entry"),
   2052 			NULL, 0, &log_permanent_modify, 0,
   2053 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2054 
   2055 	sysctl_createv(clog, 0, NULL, NULL,
   2056 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2057 			CTLTYPE_INT, "log_wrong_iface",
   2058 			SYSCTL_DESCR("log ARP packets arriving on the wrong"
   2059 			    " interface"),
   2060 			NULL, 0, &log_wrong_iface, 0,
   2061 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2062 
   2063 	sysctl_createv(clog, 0, NULL, NULL,
   2064 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2065 			CTLTYPE_INT, "log_unknown_network",
   2066 			SYSCTL_DESCR("log ARP packets from non-local network"),
   2067 			NULL, 0, &log_unknown_network, 0,
   2068 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2069 
   2070 	sysctl_createv(clog, 0, NULL, NULL,
   2071 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2072 		       CTLTYPE_INT, "debug",
   2073 		       SYSCTL_DESCR("Enable ARP DAD debug output"),
   2074 		       NULL, 0, &arp_debug, 0,
   2075 		       CTL_NET, PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   2076 }
   2077 
   2078 #endif /* INET */
   2079