xref: /dragonfly/sys/netinet/if_ether.c (revision 7485684f)
1 /*
2  * Copyright (c) 2004, 2005 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Jeffrey M. Hsu.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of The DragonFly Project nor the names of its
16  *    contributors may be used to endorse or promote products derived
17  *    from this software without specific, prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
22  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
23  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
25  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
26  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
27  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
28  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
29  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * 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	8.1 (Berkeley) 6/10/93
62  * $FreeBSD: src/sys/netinet/if_ether.c,v 1.64.2.23 2003/04/11 07:23:15 fjoe Exp $
63  */
64 
65 /*
66  * Ethernet address resolution protocol.
67  * TODO:
68  *	add "inuse/lock" bit (or ref. count) along with valid bit
69  */
70 
71 #include "opt_inet.h"
72 #include "opt_carp.h"
73 
74 #include <sys/param.h>
75 #include <sys/kernel.h>
76 #include <sys/queue.h>
77 #include <sys/sysctl.h>
78 #include <sys/systm.h>
79 #include <sys/mbuf.h>
80 #include <sys/malloc.h>
81 #include <sys/socket.h>
82 #include <sys/syslog.h>
83 #include <sys/lock.h>
84 
85 #include <net/if.h>
86 #include <net/if_dl.h>
87 #include <net/if_types.h>
88 #include <net/route.h>
89 #include <net/netisr.h>
90 #include <net/if_llc.h>
91 
92 #include <netinet/in.h>
93 #include <netinet/in_var.h>
94 #include <netinet/if_ether.h>
95 
96 #include <sys/thread2.h>
97 #include <sys/msgport2.h>
98 #include <net/netmsg2.h>
99 #include <net/netisr2.h>
100 
101 #ifdef CARP
102 #include <netinet/ip_carp.h>
103 #endif
104 
105 #define SIN(s) ((struct sockaddr_in *)s)
106 #define SDL(s) ((struct sockaddr_dl *)s)
107 
108 MALLOC_DEFINE(M_ARP, "arp", "ARP");
109 
110 SYSCTL_DECL(_net_link_ether);
111 SYSCTL_NODE(_net_link_ether, PF_INET, inet, CTLFLAG_RW, 0, "");
112 
113 /* timer values */
114 static int arpt_prune = (5*60*1); /* walk list every 5 minutes */
115 static int arpt_keep = (20*60); /* once resolved, good for 20 more minutes */
116 static int arpt_down = 20;	/* once declared down, don't send for 20 sec */
117 
118 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, prune_intvl, CTLFLAG_RW,
119 	   &arpt_prune, 0, "");
120 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_RW,
121 	   &arpt_keep, 0, "");
122 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, host_down_time, CTLFLAG_RW,
123 	   &arpt_down, 0, "");
124 
125 #define	rt_expire	rt_rmx.rmx_expire
126 
127 struct llinfo_arp {
128 	LIST_ENTRY(llinfo_arp) la_le;
129 	struct	rtentry *la_rt;
130 	struct	mbuf *la_hold;	/* last packet until resolved/timeout */
131 	u_short	la_preempt;	/* countdown for pre-expiry arps */
132 	u_short	la_asked;	/* #times we QUERIED following expiration */
133 };
134 
135 static int	arp_maxtries = 5;
136 static int	useloopback = 1; /* use loopback interface for local traffic */
137 static int	arp_proxyall = 0;
138 static int	arp_refresh = 60; /* refresh arp cache ~60 (not impl yet) */
139 static int	arp_restricted_match = 0;
140 static int	arp_ignore_probes = 1;
141 
142 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_RW,
143 	   &arp_maxtries, 0, "ARP resolution attempts before returning error");
144 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, useloopback, CTLFLAG_RW,
145 	   &useloopback, 0, "Use the loopback interface for local traffic");
146 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_RW,
147 	   &arp_proxyall, 0, "Enable proxy ARP for all suitable requests");
148 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, restricted_match, CTLFLAG_RW,
149 	   &arp_restricted_match, 0, "Only match against the sender");
150 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, refresh, CTLFLAG_RW,
151 	   &arp_refresh, 0, "Preemptively refresh the ARP");
152 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, ignore_probes, CTLFLAG_RW,
153 	   &arp_ignore_probes, 0, "Ignore ARP probes");
154 
155 static void	arp_rtrequest(int, struct rtentry *);
156 static void	arprequest(struct ifnet *, const struct in_addr *,
157 			   const struct in_addr *, const u_char *);
158 static void	arprequest_async(struct ifnet *, const struct in_addr *,
159 				 const struct in_addr *, const u_char *);
160 static void	arpintr(netmsg_t msg);
161 static void	arptfree(struct llinfo_arp *);
162 static void	arptimer(void *);
163 static struct llinfo_arp *
164 		arplookup(in_addr_t, boolean_t, boolean_t);
165 #ifdef INET
166 static void	in_arpinput(struct mbuf *);
167 static void	in_arpreply(struct mbuf *m, in_addr_t, in_addr_t);
168 static void	arp_update_msghandler(netmsg_t);
169 static void	arp_reply_msghandler(netmsg_t);
170 #endif
171 
172 struct arp_pcpu_data {
173 	LIST_HEAD(, llinfo_arp) llinfo_list;
174 	struct callout		timer_ch;
175 	struct netmsg_base	timer_nmsg;
176 };
177 
178 static struct arp_pcpu_data	*arp_data[MAXCPU];
179 
180 /*
181  * Timeout routine.  Age arp_tab entries periodically.
182  */
183 static void
arptimer_dispatch(netmsg_t nmsg)184 arptimer_dispatch(netmsg_t nmsg)
185 {
186 	struct arp_pcpu_data *ad = nmsg->lmsg.u.ms_resultp;
187 	struct llinfo_arp *la, *nla;
188 
189 	ASSERT_NETISR_NCPUS(mycpuid);
190 
191 	/* Reply ASAP */
192 	crit_enter();
193 	netisr_replymsg(&nmsg->base, 0);
194 	crit_exit();
195 
196 	LIST_FOREACH_MUTABLE(la, &ad->llinfo_list, la_le, nla) {
197 		if (la->la_rt->rt_expire && la->la_rt->rt_expire <= time_uptime)
198 			arptfree(la);
199 	}
200 	callout_reset(&ad->timer_ch, arpt_prune * hz, arptimer, &ad->timer_nmsg);
201 }
202 
203 static void
arptimer(void * xnm)204 arptimer(void *xnm)
205 {
206 	struct netmsg_base *nm = xnm;
207 
208 	KKASSERT(mycpuid < netisr_ncpus);
209 
210 	crit_enter();
211 	if (nm->lmsg.ms_flags & MSGF_DONE)
212 		netisr_sendmsg_oncpu(nm);
213 	crit_exit();
214 }
215 
216 /*
217  * Parallel to llc_rtrequest.
218  *
219  * Called after a route is successfully added to the tree to fix-up the
220  * route and initiate arp operations if required.
221  */
222 static void
arp_rtrequest(int req,struct rtentry * rt)223 arp_rtrequest(int req, struct rtentry *rt)
224 {
225 	struct sockaddr *gate = rt->rt_gateway;
226 	struct llinfo_arp *la = rt->rt_llinfo;
227 
228 	struct sockaddr_dl null_sdl = { sizeof null_sdl, AF_LINK };
229 
230 	if (rt->rt_flags & RTF_GATEWAY)
231 		return;
232 
233 	switch (req) {
234 	case RTM_ADD:
235 		/*
236 		 * XXX: If this is a manually added route to interface
237 		 * such as older version of routed or gated might provide,
238 		 * restore cloning bit.
239 		 */
240 		if (!(rt->rt_flags & RTF_HOST) &&
241 		    SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
242 			rt->rt_flags |= RTF_CLONING;
243 		if (rt->rt_flags & RTF_CLONING) {
244 			/*
245 			 * Case 1: This route should come from a route to iface.
246 			 */
247 			rt_setgate(rt, rt_key(rt),
248 				   (struct sockaddr *)&null_sdl);
249 			gate = rt->rt_gateway;
250 			SDL(gate)->sdl_type = rt->rt_ifp->if_type;
251 			SDL(gate)->sdl_index = rt->rt_ifp->if_index;
252 			rt->rt_expire = time_uptime;
253 			break;
254 		}
255 		/*
256 		 * Announce a new entry if requested, and only announce it
257 		 * once on cpu0.
258 		 */
259 		if ((rt->rt_flags & RTF_ANNOUNCE) && mycpuid == 0) {
260 			arprequest(rt->rt_ifp,
261 			    &SIN(rt_key(rt))->sin_addr,
262 			    &SIN(rt_key(rt))->sin_addr,
263 			    LLADDR(SDL(gate)));
264 		}
265 		/*FALLTHROUGH*/
266 	case RTM_RESOLVE:
267 		if (gate->sa_family != AF_LINK ||
268 		    gate->sa_len < sizeof(struct sockaddr_dl)) {
269 			log(LOG_DEBUG, "arp_rtrequest: bad gateway value\n");
270 			break;
271 		}
272 		SDL(gate)->sdl_type = rt->rt_ifp->if_type;
273 		SDL(gate)->sdl_index = rt->rt_ifp->if_index;
274 		if (la != NULL)
275 			break; /* This happens on a route change */
276 		/*
277 		 * Case 2:  This route may come from cloning, or a manual route
278 		 * add with a LL address.
279 		 */
280 		R_Malloc(la, struct llinfo_arp *, sizeof *la);
281 		rt->rt_llinfo = la;
282 		if (la == NULL) {
283 			log(LOG_DEBUG, "arp_rtrequest: malloc failed\n");
284 			break;
285 		}
286 		bzero(la, sizeof *la);
287 		la->la_rt = rt;
288 		rt->rt_flags |= RTF_LLINFO;
289 		LIST_INSERT_HEAD(&arp_data[mycpuid]->llinfo_list, la, la_le);
290 
291 #ifdef INET
292 		/*
293 		 * This keeps the multicast addresses from showing up
294 		 * in `arp -a' listings as unresolved.  It's not actually
295 		 * functional.  Then the same for broadcast.
296 		 */
297 		if (IN_MULTICAST(ntohl(SIN(rt_key(rt))->sin_addr.s_addr))) {
298 			ETHER_MAP_IP_MULTICAST(&SIN(rt_key(rt))->sin_addr,
299 					       LLADDR(SDL(gate)));
300 			SDL(gate)->sdl_alen = 6;
301 			rt->rt_expire = 0;
302 		}
303 		if (in_broadcast(SIN(rt_key(rt))->sin_addr, rt->rt_ifp)) {
304 			memcpy(LLADDR(SDL(gate)), rt->rt_ifp->if_broadcastaddr,
305 			       rt->rt_ifp->if_addrlen);
306 			SDL(gate)->sdl_alen = rt->rt_ifp->if_addrlen;
307 			rt->rt_expire = 0;
308 		}
309 #endif
310 
311 		/*
312 		 * This fixes up the routing interface for local addresses.
313 		 * The route is adjusted to point at lo0 and the expiration
314 		 * timer is disabled.
315 		 *
316 		 * NOTE: This prevents locally targetted traffic from going
317 		 *	 out the hardware interface, which is inefficient
318 		 *	 and might not work if the hardware cannot listen
319 		 *	 to its own transmitted packets.   Setting
320 		 *	 net.link.ether.inet.useloopback to 0 will force
321 		 *	 packets for local addresses out the hardware (and
322 		 *	 it is expected to receive its own packet).
323 		 *
324 		 * XXX We should just be able to test RTF_LOCAL here instead
325 		 *     of having to compare IPs.
326 		 */
327 		if (SIN(rt_key(rt))->sin_addr.s_addr ==
328 		    (IA_SIN(rt->rt_ifa))->sin_addr.s_addr) {
329 			rt->rt_expire = 0;
330 			SDL(gate)->sdl_alen = rt->rt_ifp->if_addrlen;
331 			bcopy(IF_LLADDR(rt->rt_ifp), LLADDR(SDL(gate)),
332 			      rt->rt_ifp->if_addrlen);
333 			if (useloopback)
334 				rt->rt_ifp = loif;
335 		}
336 		break;
337 
338 	case RTM_DELETE:
339 		if (la == NULL)
340 			break;
341 		LIST_REMOVE(la, la_le);
342 		rt->rt_llinfo = NULL;
343 		rt->rt_flags &= ~RTF_LLINFO;
344 		if (la->la_hold != NULL)
345 			m_freem(la->la_hold);
346 		R_Free(la);
347 		break;
348 	}
349 }
350 
351 static struct mbuf *
arpreq_alloc(struct ifnet * ifp,const struct in_addr * sip,const struct in_addr * tip,const u_char * enaddr)352 arpreq_alloc(struct ifnet *ifp, const struct in_addr *sip,
353 	     const struct in_addr *tip, const u_char *enaddr)
354 {
355 	struct mbuf *m;
356 	struct arphdr *ah;
357 	u_short ar_hrd;
358 
359 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
360 		return NULL;
361 	m->m_pkthdr.rcvif = NULL;
362 
363 	switch (ifp->if_type) {
364 	case IFT_ETHER:
365 		/*
366 		 * This may not be correct for types not explicitly
367 		 * listed, but this is our best guess
368 		 */
369 	default:
370 		ar_hrd = htons(ARPHRD_ETHER);
371 
372 		m->m_len = arphdr_len2(ifp->if_addrlen, sizeof(struct in_addr));
373 		m->m_pkthdr.len = m->m_len;
374 		MH_ALIGN(m, m->m_len);
375 
376 		ah = mtod(m, struct arphdr *);
377 		break;
378 	}
379 
380 	ah->ar_hrd = ar_hrd;
381 	ah->ar_pro = htons(ETHERTYPE_IP);
382 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
383 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
384 	ah->ar_op = htons(ARPOP_REQUEST);
385 	memcpy(ar_sha(ah), enaddr, ah->ar_hln);
386 	memset(ar_tha(ah), 0, ah->ar_hln);
387 	memcpy(ar_spa(ah), sip, ah->ar_pln);
388 	memcpy(ar_tpa(ah), tip, ah->ar_pln);
389 
390 	return m;
391 }
392 
393 static void
arpreq_send(struct ifnet * ifp,struct mbuf * m)394 arpreq_send(struct ifnet *ifp, struct mbuf *m)
395 {
396 	struct sockaddr sa;
397 	struct ether_header *eh;
398 
399 	ASSERT_NETISR_NCPUS(mycpuid);
400 
401 	switch (ifp->if_type) {
402 	case IFT_ETHER:
403 		/*
404 		 * This may not be correct for types not explicitly
405 		 * listed, but this is our best guess
406 		 */
407 	default:
408 		eh = (struct ether_header *)sa.sa_data;
409 		/* if_output() will not swap */
410 		eh->ether_type = htons(ETHERTYPE_ARP);
411 		memcpy(eh->ether_dhost, ifp->if_broadcastaddr, ifp->if_addrlen);
412 		break;
413 	}
414 
415 	sa.sa_family = AF_UNSPEC;
416 	sa.sa_len = sizeof(sa);
417 	ifp->if_output(ifp, m, &sa, NULL);
418 }
419 
420 static void
arpreq_send_handler(netmsg_t msg)421 arpreq_send_handler(netmsg_t msg)
422 {
423 	struct mbuf *m = msg->packet.nm_packet;
424 	struct ifnet *ifp = msg->lmsg.u.ms_resultp;
425 
426 	arpreq_send(ifp, m);
427 	/* nmsg was embedded in the mbuf, do not reply! */
428 }
429 
430 /*
431  * Broadcast an ARP request. Caller specifies:
432  *	- arp header source ip address
433  *	- arp header target ip address
434  *	- arp header source ethernet address
435  *
436  * NOTE: Caller MUST NOT hold ifp's serializer
437  */
438 static void
arprequest(struct ifnet * ifp,const struct in_addr * sip,const struct in_addr * tip,const u_char * enaddr)439 arprequest(struct ifnet *ifp, const struct in_addr *sip,
440 	   const struct in_addr *tip, const u_char *enaddr)
441 {
442 	struct mbuf *m;
443 
444 	ASSERT_NETISR_NCPUS(mycpuid);
445 
446 	if (enaddr == NULL) {
447 		if (ifp->if_bridge) {
448 			enaddr = IF_LLADDR(ether_bridge_interface(ifp));
449 		} else {
450 			enaddr = IF_LLADDR(ifp);
451 		}
452 	}
453 
454 	m = arpreq_alloc(ifp, sip, tip, enaddr);
455 	if (m == NULL)
456 		return;
457 	arpreq_send(ifp, m);
458 }
459 
460 /*
461  * Same as arprequest(), except:
462  * - Caller is allowed to hold ifp's serializer
463  * - Network output is done in protocol thead
464  */
465 static void
arprequest_async(struct ifnet * ifp,const struct in_addr * sip,const struct in_addr * tip,const u_char * enaddr)466 arprequest_async(struct ifnet *ifp, const struct in_addr *sip,
467 		 const struct in_addr *tip, const u_char *enaddr)
468 {
469 	struct mbuf *m;
470 	struct netmsg_packet *pmsg;
471 	int cpu;
472 
473 	if (enaddr == NULL) {
474 		if (ifp->if_bridge) {
475 			enaddr = IF_LLADDR(ether_bridge_interface(ifp));
476 		} else {
477 			enaddr = IF_LLADDR(ifp);
478 		}
479 	}
480 	m = arpreq_alloc(ifp, sip, tip, enaddr);
481 	if (m == NULL)
482 		return;
483 
484 	pmsg = &m->m_hdr.mh_netmsg;
485 	netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport,
486 		    0, arpreq_send_handler);
487 	pmsg->nm_packet = m;
488 	pmsg->base.lmsg.u.ms_resultp = ifp;
489 
490 	if (mycpuid < netisr_ncpus)
491 		cpu = mycpuid;
492 	else
493 		cpu = 0;
494 	lwkt_sendmsg(netisr_cpuport(cpu), &pmsg->base.lmsg);
495 }
496 
497 /*
498  * Resolve an IP address into an ethernet address.  If success,
499  * desten is filled in.  If there is no entry in arptab,
500  * set one up and broadcast a request for the IP address.
501  * Hold onto this mbuf and resend it once the address
502  * is finally resolved.  A return value of 1 indicates
503  * that desten has been filled in and the packet should be sent
504  * normally; a 0 return indicates that the packet has been
505  * taken over here, either now or for later transmission.
506  */
507 int
arpresolve(struct ifnet * ifp,struct rtentry * rt0,struct mbuf * m,struct sockaddr * dst,u_char * desten)508 arpresolve(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
509 	   struct sockaddr *dst, u_char *desten)
510 {
511 	struct rtentry *rt = NULL;
512 	struct llinfo_arp *la = NULL;
513 	struct sockaddr_dl *sdl;
514 	int error;
515 
516 	if (m->m_flags & M_BCAST) {	/* broadcast */
517 		memcpy(desten, ifp->if_broadcastaddr, ifp->if_addrlen);
518 		return 0;
519 	}
520 	if (m->m_flags & M_MCAST) {/* multicast */
521 		ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
522 		return 0;
523 	}
524 	if (rt0 != NULL) {
525 		error = rt_llroute(dst, rt0, &rt);
526 		if (error != 0) {
527 			m_freem(m);
528 			return error;
529 		}
530 		la = rt->rt_llinfo;
531 	}
532 	if (la == NULL) {
533 		la = arplookup(SIN(dst)->sin_addr.s_addr, TRUE, FALSE);
534 		if (la != NULL)
535 			rt = la->la_rt;
536 	}
537 	if (la == NULL || rt == NULL) {
538 		char addr[INET_ADDRSTRLEN];
539 
540 		log(LOG_DEBUG, "arpresolve: can't allocate llinfo for %s%s%s\n",
541 		    kinet_ntoa(SIN(dst)->sin_addr, addr), la ? "la" : " ",
542 		    rt ? "rt" : "");
543 		m_freem(m);
544 		return ENOBUFS;
545 	}
546 
547 	/*
548 	 * Check the address family and length is valid, the address
549 	 * is resolved; otherwise, try to resolve.
550 	 */
551 	sdl = SDL(rt->rt_gateway);
552 	if ((rt->rt_expire == 0 || rt->rt_expire > time_uptime) &&
553 	    sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0) {
554 		/*
555 		 * If entry has an expiry time and it is approaching,
556 		 * see if we need to send an ARP request within this
557 		 * arpt_down interval.
558 		 */
559 		if ((rt->rt_expire != 0) &&
560 		    (time_uptime + la->la_preempt > rt->rt_expire)) {
561 			arprequest(ifp,
562 				   &SIN(rt->rt_ifa->ifa_addr)->sin_addr,
563 				   &SIN(dst)->sin_addr,
564 				   NULL);
565 			la->la_preempt--;
566 		}
567 
568 		bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
569 		return 0;
570 	}
571 
572 	/*
573 	 * If ARP is disabled or static on this interface, stop.
574 	 * XXX
575 	 * Probably should not allocate empty llinfo struct if we are
576 	 * not going to be sending out an arp request.
577 	 */
578 	if (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) {
579 		m_freem(m);
580 		return ifp->if_flags & IFF_NOARP ? ENOTSUP : EINVAL;
581 	}
582 
583 	/*
584 	 * There is an arptab entry, but no ethernet address
585 	 * response yet.  Replace the held mbuf with this
586 	 * latest one.
587 	 */
588 	if (la->la_hold != NULL)
589 		m_freem(la->la_hold);
590 	la->la_hold = m;
591 
592 	/*
593 	 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
594 	 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
595 	 * if we have already sent arp_maxtries ARP requests. Retransmit the
596 	 * ARP request, but not faster than one request per second.
597 	 */
598 	if (la->la_asked < arp_maxtries)
599 		error = EWOULDBLOCK;
600 	else
601 		error = (rt != NULL && rt->rt_flags & RTF_GATEWAY) ?
602 		    EHOSTUNREACH : EHOSTDOWN;
603 
604 	if (rt->rt_expire || ((rt->rt_flags & RTF_STATIC) && !sdl->sdl_alen)) {
605 		rt->rt_flags &= ~RTF_REJECT;
606 		if (la->la_asked == 0 || rt->rt_expire != time_uptime) {
607 			rt->rt_expire = time_uptime;
608 			arprequest(ifp,
609 				   &SIN(rt->rt_ifa->ifa_addr)->sin_addr,
610 				   &SIN(dst)->sin_addr,
611 				   NULL);
612 			if (la->la_asked++ >= arp_maxtries) {
613 				rt->rt_expire += arpt_down;
614 				la->la_preempt = arp_maxtries;
615 				rt_rtmsg(RTM_MISS, rt, rt->rt_ifp, 0);
616 			}
617 		}
618 	}
619 	return error;
620 }
621 
622 /*
623  * Common length and type checks are done here,
624  * then the protocol-specific routine is called.
625  */
626 static void
arpintr(netmsg_t msg)627 arpintr(netmsg_t msg)
628 {
629 	struct mbuf *m = msg->packet.nm_packet;
630 	struct arphdr *ar;
631 	u_short ar_hrd;
632 	char hexstr[6];
633 
634 	if (m->m_len < sizeof(struct arphdr) &&
635 	    (m = m_pullup(m, sizeof(struct arphdr))) == NULL) {
636 		log(LOG_ERR, "arp: runt packet -- m_pullup failed\n");
637 		return;
638 	}
639 	ar = mtod(m, struct arphdr *);
640 
641 	ar_hrd = ntohs(ar->ar_hrd);
642 	if (ar_hrd != ARPHRD_ETHER && ar_hrd != ARPHRD_IEEE802) {
643 		hexncpy((unsigned char *)&ar->ar_hrd, 2, hexstr, 5, NULL);
644 		log(LOG_ERR, "arp: unknown hardware address format (0x%s)\n",
645 		    hexstr);
646 		m_freem(m);
647 		return;
648 	}
649 
650 	if (m->m_pkthdr.len < arphdr_len(ar)) {
651 		if ((m = m_pullup(m, arphdr_len(ar))) == NULL) {
652 			log(LOG_ERR, "arp: runt packet\n");
653 			return;
654 		}
655 		ar = mtod(m, struct arphdr *);
656 	}
657 
658 	switch (ntohs(ar->ar_pro)) {
659 #ifdef INET
660 	case ETHERTYPE_IP:
661 		in_arpinput(m);
662 		return;
663 #endif
664 	}
665 	m_freem(m);
666 	/* msg was embedded in the mbuf, do not reply! */
667 }
668 
669 #ifdef INET
670 /*
671  * ARP for Internet protocols on 10 Mb/s Ethernet.
672  * Algorithm is that given in RFC 826.
673  * In addition, a sanity check is performed on the sender
674  * protocol address, to catch impersonators.
675  * We no longer handle negotiations for use of trailer protocol:
676  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
677  * along with IP replies if we wanted trailers sent to us,
678  * and also sent them in response to IP replies.
679  * This allowed either end to announce the desire to receive
680  * trailer packets.
681  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
682  * but formerly didn't normally send requests.
683  */
684 
685 static int	log_arp_wrong_iface = 1;
686 static int	log_arp_movements = 1;
687 static int	log_arp_permanent_modify = 1;
688 static int	log_arp_creation_failure = 1;
689 
690 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
691 	   &log_arp_wrong_iface, 0,
692 	   "Log arp packets arriving on the wrong interface");
693 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
694 	   &log_arp_movements, 0,
695 	   "Log arp replies from MACs different than the one in the cache");
696 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
697 	   &log_arp_permanent_modify, 0,
698 	   "Log arp replies from MACs different than the one "
699 	   "in the permanent arp entry");
700 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_creation_failure, CTLFLAG_RW,
701 	   &log_arp_creation_failure, 0, "Log arp creation failure");
702 
703 /*
704  * Returns non-zero if the routine updated anything.
705  */
706 static int
arp_update_oncpu(struct mbuf * m,in_addr_t saddr,boolean_t create,boolean_t dologging)707 arp_update_oncpu(struct mbuf *m, in_addr_t saddr, boolean_t create,
708 		 boolean_t dologging)
709 {
710 	struct arphdr *ah = mtod(m, struct arphdr *);
711 	struct ifnet *ifp = m->m_pkthdr.rcvif;
712 	struct llinfo_arp *la;
713 	struct sockaddr_dl *sdl;
714 	struct rtentry *rt;
715 	char hexstr[2][64];
716 	char sbuf[INET_ADDRSTRLEN];
717 	int changed = create;
718 
719 	KASSERT(curthread->td_type == TD_TYPE_NETISR,
720 	    ("arp update not in netisr"));
721 
722 	la = arplookup(saddr, create, FALSE);
723 	if (la && (rt = la->la_rt) && (sdl = SDL(rt->rt_gateway))) {
724 		struct in_addr isaddr = { saddr };
725 		int rt_cmd = sdl->sdl_alen == 0 ? RTM_ADD : RTM_CHANGE;
726 		bool do_rtmsg = false;
727 
728 		/*
729 		 * Normally arps coming in on the wrong interface are ignored,
730 		 * but if we are bridging and the two interfaces belong to
731 		 * the same bridge, or one is a member of the bridge which
732 		 * is the other, then it isn't an error.
733 		 */
734 		if (rt->rt_ifp != ifp) {
735 			/*
736 			 * (1) ifp and rt_ifp both members of same bridge
737 			 * (2) rt_ifp member of bridge ifp
738 			 * (3) ifp member of bridge rt_ifp
739 			 *
740 			 * Always replace rt_ifp with the bridge ifc.
741 			 */
742 			struct ifnet *nifp;
743 
744 			if (ifp->if_bridge &&
745 			    rt->rt_ifp->if_bridge == ifp->if_bridge) {
746 				nifp = ether_bridge_interface(ifp);
747 			} else if (rt->rt_ifp->if_bridge &&
748 				   ether_bridge_interface(rt->rt_ifp) == ifp) {
749 				nifp = ifp;
750 			} else if (ifp->if_bridge &&
751 				   ether_bridge_interface(ifp) == rt->rt_ifp) {
752 				nifp = rt->rt_ifp;
753 			} else {
754 				nifp = NULL;
755 			}
756 
757 			if ((log_arp_wrong_iface == 1 && nifp == NULL) ||
758 			    log_arp_wrong_iface == 2) {
759 				hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
760 				    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
761 				log(LOG_ERR,
762 				    "arp: %s is on %s "
763 				    "but got reply from %s on %s\n",
764 				    kinet_ntoa(isaddr, sbuf),
765 				    rt->rt_ifp->if_xname, hexstr[0],
766 				    ifp->if_xname);
767 			}
768 			if (nifp == NULL)
769 				return 0;
770 
771 			/*
772 			 * nifp is our man!  Replace rt_ifp and adjust
773 			 * the sdl.
774 			 */
775 			ifp = rt->rt_ifp = nifp;
776 			if (sdl->sdl_type != ifp->if_type) {
777 				sdl->sdl_type = ifp->if_type;
778 				changed = 1;
779 				do_rtmsg = true;
780 			}
781 			if (sdl->sdl_index != ifp->if_index) {
782 				sdl->sdl_index = ifp->if_index;
783 				changed = 1;
784 				do_rtmsg = true;
785 			}
786 		}
787 		if (sdl->sdl_alen &&
788 		    bcmp(ar_sha(ah), LLADDR(sdl), sdl->sdl_alen)) {
789 			changed = 1;
790 			if (rt->rt_expire != 0) {
791 				if (dologging && log_arp_movements) {
792 					hexncpy((u_char *)LLADDR(sdl), ifp->if_addrlen,
793 					    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
794 					hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
795 					    hexstr[1], HEX_NCPYLEN(ifp->if_addrlen), ":");
796 					log(LOG_INFO,
797 					    "arp: %s moved from %s to %s on %s\n",
798 					    kinet_ntoa(isaddr, sbuf), hexstr[0], hexstr[1],
799 					    ifp->if_xname);
800 				}
801 			} else {
802 				if (dologging && log_arp_permanent_modify) {
803 					hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
804 					    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
805 					log(LOG_ERR,
806 					"arp: %s attempts to modify "
807 					"permanent entry for %s on %s\n",
808 					hexstr[0], kinet_ntoa(isaddr, sbuf), ifp->if_xname);
809 				}
810 				return changed;
811 			}
812 			do_rtmsg = true;
813 		}
814 		/*
815 		 * sanity check for the address length.
816 		 * XXX this does not work for protocols with variable address
817 		 * length. -is
818 		 */
819 		if (dologging && sdl->sdl_alen && sdl->sdl_alen != ah->ar_hln) {
820 			hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
821 			    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
822 			log(LOG_WARNING,
823 			    "arp from %s: new addr len %d, was %d",
824 			    hexstr[0], ah->ar_hln, sdl->sdl_alen);
825 		}
826 		if (ifp->if_addrlen != ah->ar_hln) {
827 			if (dologging) {
828 				hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
829 				    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
830 				log(LOG_WARNING,
831 				"arp from %s: addr len: new %d, i/f %d "
832 				"(ignored)", hexstr[0],
833 				ah->ar_hln, ifp->if_addrlen);
834 			}
835 			return changed;
836 		}
837 		if (sdl->sdl_alen == 0)
838 			do_rtmsg = true;
839 		memcpy(LLADDR(sdl), ar_sha(ah), sdl->sdl_alen = ah->ar_hln);
840 		if (rt->rt_expire != 0) {
841 			if (rt->rt_expire != time_uptime + arpt_keep &&
842 			    rt->rt_expire != time_uptime + arpt_keep - 1) {
843 				rt->rt_expire = time_uptime + arpt_keep;
844 				changed = 1;
845 			}
846 		}
847 		if (rt->rt_flags & RTF_REJECT) {
848 			rt->rt_flags &= ~RTF_REJECT;
849 			changed = 1;
850 		}
851 		if (la->la_asked != 0) {
852 			la->la_asked = 0;
853 			changed = 1;
854 		}
855 		if (la->la_preempt != arp_maxtries) {
856 			la->la_preempt = arp_maxtries;
857 			changed = 1;
858 		}
859 
860 		/*
861 		 * This particular cpu might have been holding an mbuf
862 		 * pending ARP resolution.  If so, transmit the mbuf now.
863 		 */
864 		if (la->la_hold != NULL) {
865 			struct mbuf *m = la->la_hold;
866 
867 			la->la_hold = NULL;
868 			m_adj(m, sizeof(struct ether_header));
869 			ifp->if_output(ifp, m, rt_key(rt), rt);
870 			changed = 1;
871 		}
872 
873 		if (do_rtmsg && mycpuid == 0)
874 			rt_rtmsg(rt_cmd, rt, rt->rt_ifp, 0);
875 	}
876 	return changed;
877 }
878 
879 /*
880  * Called from arpintr() - this routine is run from a single cpu.
881  */
882 static void
in_arpinput(struct mbuf * m)883 in_arpinput(struct mbuf *m)
884 {
885 	struct arphdr *ah;
886 	struct ifnet *ifp = m->m_pkthdr.rcvif;
887 	struct ifaddr_container *ifac;
888 	struct in_ifaddr_container *iac;
889 	struct in_ifaddr *ia = NULL;
890 	struct in_addr isaddr, itaddr, myaddr;
891 	uint8_t *enaddr = NULL;
892 	int req_len;
893 	int changed;
894 	char hexstr[64], sbuf[INET_ADDRSTRLEN];
895 
896 	req_len = arphdr_len2(ifp->if_addrlen, sizeof(struct in_addr));
897 	if (m->m_len < req_len && (m = m_pullup(m, req_len)) == NULL) {
898 		log(LOG_ERR, "in_arp: runt packet -- m_pullup failed\n");
899 		return;
900 	}
901 
902 	ah = mtod(m, struct arphdr *);
903 	memcpy(&isaddr, ar_spa(ah), sizeof isaddr);
904 	memcpy(&itaddr, ar_tpa(ah), sizeof itaddr);
905 
906 	/*
907 	 * Check both target and sender IP addresses:
908 	 *
909 	 * If we receive the packet on the interface owning the address,
910 	 * then accept the address.
911 	 *
912 	 * For a bridge, we accept the address if the receive interface and
913 	 * the interface owning the address are on the same bridge, and
914 	 * use the bridge MAC as the is-at response.  The bridge will be
915 	 * responsible for handling the packet.
916 	 *
917 	 * (0) Check target IP against CARP IPs
918 	 */
919 #ifdef CARP
920 	LIST_FOREACH(iac, INADDR_HASH(itaddr.s_addr), ia_hash) {
921 		int is_match = 0, is_parent = 0;
922 
923 		ia = iac->ia;
924 
925 		/* Skip all ia's which don't match */
926 		if (itaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
927 			continue;
928 
929 		if (ia->ia_ifp->if_type != IFT_CARP)
930 			continue;
931 
932 		if (carp_parent(ia->ia_ifp) == ifp)
933 			is_parent = 1;
934 		if (is_parent || ia->ia_ifp == ifp)
935 			is_match = carp_iamatch(ia);
936 
937 		if (is_match) {
938 			if (is_parent) {
939 				/*
940 				 * The parent interface will also receive
941 				 * the ethernet broadcast packets, e.g. ARP
942 				 * REQUEST, so if we could find a CARP
943 				 * interface of the parent that could match
944 				 * the target IP address, we then drop the
945 				 * packets, which is delieverd to us through
946 				 * the parent interface.
947 				 */
948 				m_freem(m);
949 				return;
950 			}
951 			goto match;
952 		}
953 	}
954 #endif	/* CARP */
955 
956 	/*
957 	 * (1) Check target IP against our local IPs
958 	 */
959 	LIST_FOREACH(iac, INADDR_HASH(itaddr.s_addr), ia_hash) {
960 		ia = iac->ia;
961 
962 		/* Skip all ia's which don't match */
963 		if (itaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
964 			continue;
965 
966 #ifdef CARP
967 		/* CARP interfaces are checked in (0) */
968 		if (ia->ia_ifp->if_type == IFT_CARP)
969 			continue;
970 #endif
971 
972 		if (ifp->if_bridge && ia->ia_ifp &&
973 		    ifp->if_bridge == ia->ia_ifp->if_bridge) {
974 			ifp = ether_bridge_interface(ifp);
975 			goto match;
976 		}
977 		if (ia->ia_ifp && ia->ia_ifp->if_bridge &&
978 		    ether_bridge_interface(ia->ia_ifp) == ifp) {
979 			goto match;
980 		}
981 		if (ifp->if_bridge && ether_bridge_interface(ifp) ==
982 		    ia->ia_ifp) {
983 			goto match;
984 		}
985 		if (ia->ia_ifp == ifp) {
986 			goto match;
987 		}
988 	}
989 
990 	/*
991 	 * (2) Check sender IP against our local IPs
992 	 */
993 	LIST_FOREACH(iac, INADDR_HASH(isaddr.s_addr), ia_hash) {
994 		ia = iac->ia;
995 
996 		/* Skip all ia's which don't match */
997 		if (isaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
998 			continue;
999 
1000 		if (ifp->if_bridge && ia->ia_ifp &&
1001 		    ifp->if_bridge == ia->ia_ifp->if_bridge) {
1002 			ifp = ether_bridge_interface(ifp);
1003 			goto match;
1004 		}
1005 		if (ia->ia_ifp && ia->ia_ifp->if_bridge &&
1006 		    ether_bridge_interface(ia->ia_ifp) == ifp) {
1007 			goto match;
1008 		}
1009 		if (ifp->if_bridge && ether_bridge_interface(ifp) ==
1010 		    ia->ia_ifp) {
1011 			goto match;
1012 		}
1013 
1014 		if (ia->ia_ifp == ifp)
1015 			goto match;
1016 	}
1017 
1018 	/*
1019 	 * No match, use the first inet address on the receive interface
1020 	 * as a dummy address for the rest of the function.
1021 	 */
1022 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1023 		struct ifaddr *ifa = ifac->ifa;
1024 
1025 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
1026 			ia = ifatoia(ifa);
1027 			goto match;
1028 		}
1029 	}
1030 
1031 	/*
1032 	 * If we got here, we didn't find any suitable interface,
1033 	 * so drop the packet.
1034 	 */
1035 	m_freem(m);
1036 	return;
1037 
1038 match:
1039 	if (!enaddr)
1040 		enaddr = (uint8_t *)IF_LLADDR(ifp);
1041 	myaddr = ia->ia_addr.sin_addr;
1042 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen)) {
1043 		m_freem(m);	/* it's from me, ignore it. */
1044 		return;
1045 	}
1046 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
1047 		log(LOG_ERR,
1048 		    "arp: link address is broadcast for IP address %s!\n",
1049 		    kinet_ntoa(isaddr, sbuf));
1050 		m_freem(m);
1051 		return;
1052 	}
1053 	if (isaddr.s_addr == myaddr.s_addr && myaddr.s_addr != 0) {
1054 		hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
1055 		    hexstr, HEX_NCPYLEN(ifp->if_addrlen), ":");
1056 		log(LOG_ERR,
1057 		   "arp: %s is using my IP address %s!\n",
1058 		    hexstr, kinet_ntoa(isaddr, sbuf));
1059 		itaddr = myaddr;
1060 		goto reply;
1061 	}
1062 	if (ifp->if_flags & IFF_STATICARP)
1063 		goto reply;
1064 
1065 	/*
1066 	 * When arp_restricted_match is true and the ARP response is not
1067 	 * specifically targetted to me, ignore it.  Otherwise the entry
1068 	 * timeout may be updated for an old MAC.
1069 	 */
1070 	if (arp_restricted_match && itaddr.s_addr != myaddr.s_addr) {
1071 		m_freem(m);
1072 		return;
1073 	}
1074 
1075 	/*
1076 	 * Update all CPU's routing tables with this ARP packet.
1077 	 *
1078 	 * However, we only need to generate rtmsg on CPU0.
1079 	 */
1080 	ASSERT_NETISR0;
1081 	changed = arp_update_oncpu(m, isaddr.s_addr,
1082 				   itaddr.s_addr == myaddr.s_addr,
1083 				   TRUE);
1084 
1085 	if (netisr_ncpus > 1 && changed) {
1086 		struct netmsg_inarp *msg = &m->m_hdr.mh_arpmsg;
1087 
1088 		netmsg_init(&msg->base, NULL, &netisr_apanic_rport,
1089 			    0, arp_update_msghandler);
1090 		msg->m = m;
1091 		msg->saddr = isaddr.s_addr;
1092 		msg->taddr = itaddr.s_addr;
1093 		msg->myaddr = myaddr.s_addr;
1094 		lwkt_sendmsg(netisr_cpuport(1), &msg->base.lmsg);
1095 	} else {
1096 		goto reply;
1097 	}
1098 
1099 	/*
1100 	 * Just return here; after all CPUs's routing tables are
1101 	 * properly updated by this ARP packet, an ARP reply will
1102 	 * be generated if appropriate.
1103 	 */
1104 	return;
1105 reply:
1106 	in_arpreply(m, itaddr.s_addr, myaddr.s_addr);
1107 }
1108 
1109 static void
arp_reply_msghandler(netmsg_t msg)1110 arp_reply_msghandler(netmsg_t msg)
1111 {
1112 	struct netmsg_inarp *rmsg = (struct netmsg_inarp *)msg;
1113 
1114 	in_arpreply(rmsg->m, rmsg->taddr, rmsg->myaddr);
1115 	/* Don't reply this netmsg; netmsg_inarp is embedded in mbuf */
1116 }
1117 
1118 static void
arp_update_msghandler(netmsg_t msg)1119 arp_update_msghandler(netmsg_t msg)
1120 {
1121 	struct netmsg_inarp *rmsg = (struct netmsg_inarp *)msg;
1122 	int nextcpu;
1123 
1124 	ASSERT_NETISR_NCPUS(mycpuid);
1125 
1126 	/*
1127 	 * This message handler will be called on all of the APs;
1128 	 * no need to generate rtmsg on them.
1129 	 */
1130 	KASSERT(mycpuid > 0, ("arp update msg on cpu%d", mycpuid));
1131 	arp_update_oncpu(rmsg->m, rmsg->saddr,
1132 			 rmsg->taddr == rmsg->myaddr,
1133 			 FALSE);
1134 
1135 	nextcpu = mycpuid + 1;
1136 	if (nextcpu < netisr_ncpus) {
1137 		lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg);
1138 	} else {
1139 		struct mbuf *m = rmsg->m;
1140 		in_addr_t saddr = rmsg->saddr;
1141 		in_addr_t taddr = rmsg->taddr;
1142 		in_addr_t myaddr = rmsg->myaddr;
1143 
1144 		/*
1145 		 * Dispatch this mbuf to netisr0 to perform ARP reply,
1146 		 * if appropriate.
1147 		 * NOTE: netmsg_inarp is embedded in this mbuf.
1148 		 */
1149 		netmsg_init(&rmsg->base, NULL, &netisr_apanic_rport,
1150 		    0, arp_reply_msghandler);
1151 		rmsg->m = m;
1152 		rmsg->saddr = saddr;
1153 		rmsg->taddr = taddr;
1154 		rmsg->myaddr = myaddr;
1155 		lwkt_sendmsg(netisr_cpuport(0), &rmsg->base.lmsg);
1156 	}
1157 }
1158 
1159 /*
1160  * Reply to an arp request
1161  */
1162 static void
in_arpreply(struct mbuf * m,in_addr_t taddr,in_addr_t myaddr)1163 in_arpreply(struct mbuf *m, in_addr_t taddr, in_addr_t myaddr)
1164 {
1165 	struct ifnet *ifp = m->m_pkthdr.rcvif;
1166 	const uint8_t *enaddr;
1167 	struct arphdr *ah;
1168 	struct sockaddr sa;
1169 	struct ether_header *eh;
1170 
1171 	ASSERT_NETISR0;
1172 
1173 	ah = mtod(m, struct arphdr *);
1174 	if (ntohs(ah->ar_op) != ARPOP_REQUEST) {
1175 		m_freem(m);
1176 		return;
1177 	}
1178 
1179 	enaddr = (const uint8_t *)IF_LLADDR(ifp);
1180 	if (taddr == myaddr) {
1181 		/* I am the target */
1182 		memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1183 		memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1184 	} else {
1185 		struct llinfo_arp *la;
1186 		struct rtentry *rt;
1187 
1188 		la = arplookup(taddr, FALSE, SIN_PROXY);
1189 		if (la == NULL) {
1190 			struct sockaddr_in sin;
1191 #ifdef DEBUG_PROXY
1192 			char tbuf[INET_ADDRSTRLEN];
1193 #endif
1194 
1195 			if (!arp_proxyall) {
1196 				m_freem(m);
1197 				return;
1198 			}
1199 
1200 			bzero(&sin, sizeof sin);
1201 			sin.sin_family = AF_INET;
1202 			sin.sin_len = sizeof sin;
1203 			sin.sin_addr.s_addr = taddr;
1204 
1205 			rt = rtpurelookup((struct sockaddr *)&sin);
1206 			if (rt == NULL) {
1207 				m_freem(m);
1208 				return;
1209 			}
1210 			--rt->rt_refcnt;
1211 
1212 			/*
1213 			 * Don't send proxies for nodes on the same interface
1214 			 * as this one came out of, or we'll get into a fight
1215 			 * over who claims what Ether address.
1216 			 *
1217 			 * If the rt entry is associated with a bridge, we
1218 			 * count it as the 'same' interface if ifp is
1219 			 * associated with the bridge.
1220 			 */
1221 			if (rt->rt_ifp == ifp || rt->rt_ifp == ifp->if_bridge) {
1222 				m_freem(m);
1223 				return;
1224 			}
1225 			memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1226 			memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1227 #ifdef DEBUG_PROXY
1228 			kprintf("arp: proxying for %s\n",
1229 			    kinet_ntoa(itaddr, tbuf));
1230 #endif
1231 		} else {
1232 			struct sockaddr_dl *sdl;
1233 
1234 			rt = la->la_rt;
1235 			memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1236 			sdl = SDL(rt->rt_gateway);
1237 			memcpy(ar_sha(ah), LLADDR(sdl), ah->ar_hln);
1238 		}
1239 	}
1240 
1241 	memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1242 	memcpy(ar_spa(ah), &taddr, ah->ar_pln);
1243 	ah->ar_op = htons(ARPOP_REPLY);
1244 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1245 	switch (ifp->if_type) {
1246 	case IFT_ETHER:
1247 		/*
1248 		 * May not be correct for types not explictly
1249 		 * listed, but it is our best guess.
1250 		 */
1251 	default:
1252 		eh = (struct ether_header *)sa.sa_data;
1253 		memcpy(eh->ether_dhost, ar_tha(ah), sizeof eh->ether_dhost);
1254 		eh->ether_type = htons(ETHERTYPE_ARP);
1255 		break;
1256 	}
1257 	sa.sa_family = AF_UNSPEC;
1258 	sa.sa_len = sizeof sa;
1259 	ifp->if_output(ifp, m, &sa, NULL);
1260 }
1261 
1262 #endif	/* INET */
1263 
1264 /*
1265  * Free an arp entry.  If the arp entry is actively referenced or represents
1266  * a static entry we only clear it back to an unresolved state, otherwise
1267  * we destroy the entry entirely.
1268  *
1269  * Note that static entries are created when route add ... -interface is used
1270  * to create an interface route to a (direct) destination.
1271  */
1272 static void
arptfree(struct llinfo_arp * la)1273 arptfree(struct llinfo_arp *la)
1274 {
1275 	struct rtentry *rt = la->la_rt;
1276 	struct sockaddr_dl *sdl;
1277 
1278 	if (rt == NULL)
1279 		panic("arptfree");
1280 	sdl = SDL(rt->rt_gateway);
1281 	if (sdl != NULL &&
1282 	    ((rt->rt_refcnt > 0 && sdl->sdl_family == AF_LINK) ||
1283 	     (rt->rt_flags & RTF_STATIC))) {
1284 		sdl->sdl_alen = 0;
1285 		la->la_preempt = la->la_asked = 0;
1286 		rt->rt_flags &= ~RTF_REJECT;
1287 		return;
1288 	}
1289 
1290 	/*
1291 	 * ARP expiry happens under one big timer.
1292 	 * To avoid overflowing the route socket, don't report this.
1293 	 * Now that RTM_MISS is reported when an address is unresolvable
1294 	 * the benefit of reporting this deletion is questionable.
1295 	 */
1296 	rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1297 }
1298 
1299 /*
1300  * Lookup or enter a new address in arptab.
1301  */
1302 static struct llinfo_arp *
arplookup(in_addr_t addr,boolean_t create,boolean_t proxy)1303 arplookup(in_addr_t addr, boolean_t create,
1304 	  boolean_t proxy)
1305 {
1306 	struct rtentry *rt;
1307 	struct sockaddr_inarp sin = { sizeof sin, AF_INET };
1308 	const char *why = NULL;
1309 
1310 	/* Check ARP probes, e.g. from Cisco switches. */
1311 	if (addr == INADDR_ANY && arp_ignore_probes)
1312 		return (NULL);
1313 
1314 	sin.sin_addr.s_addr = addr;
1315 	sin.sin_other = proxy ? SIN_PROXY : 0;
1316 	if (create) {
1317 		rt = rtlookup((struct sockaddr *)&sin);
1318 	} else {
1319 		rt = rtpurelookup((struct sockaddr *)&sin);
1320 	}
1321 	if (rt == NULL)
1322 		return (NULL);
1323 	rt->rt_refcnt--;
1324 
1325 	if (rt->rt_flags & RTF_GATEWAY)
1326 		why = "host is not on local network";
1327 	else if (!(rt->rt_flags & RTF_LLINFO))
1328 		why = "could not allocate llinfo";
1329 	else if (rt->rt_gateway->sa_family != AF_LINK)
1330 		why = "gateway route is not ours";
1331 
1332 	if (why) {
1333 		if (create && log_arp_creation_failure) {
1334 			char abuf[INET_ADDRSTRLEN];
1335 
1336 			log(LOG_DEBUG, "arplookup %s failed: %s\n",
1337 			    kinet_ntoa(sin.sin_addr, abuf), why);
1338 		}
1339 		if (rt->rt_refcnt <= 0 && (rt->rt_flags & RTF_WASCLONED)) {
1340 			/* No references to this route.  Purge it. */
1341 			rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1342 				  rt_mask(rt), rt->rt_flags, NULL);
1343 		}
1344 		return (NULL);
1345 	}
1346 	return (rt->rt_llinfo);
1347 }
1348 
1349 void
arp_ifinit(struct ifnet * ifp,struct ifaddr * ifa)1350 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1351 {
1352 	ifa->ifa_rtrequest = arp_rtrequest;
1353 	ifa->ifa_flags |= RTF_CLONING;
1354 }
1355 
1356 void
arp_gratuitous(struct ifnet * ifp,struct ifaddr * ifa)1357 arp_gratuitous(struct ifnet *ifp, struct ifaddr *ifa)
1358 {
1359 	if (IA_SIN(ifa)->sin_addr.s_addr != INADDR_ANY) {
1360 		if (IN_NETISR_NCPUS(mycpuid)) {
1361 			arprequest(ifp, &IA_SIN(ifa)->sin_addr,
1362 			    &IA_SIN(ifa)->sin_addr, NULL);
1363 		} else {
1364 			arprequest_async(ifp, &IA_SIN(ifa)->sin_addr,
1365 			    &IA_SIN(ifa)->sin_addr, NULL);
1366 		}
1367 	}
1368 }
1369 
1370 static void
arp_ifaddr(void * arg __unused,struct ifnet * ifp,enum ifaddr_event event,struct ifaddr * ifa)1371 arp_ifaddr(void *arg __unused, struct ifnet *ifp,
1372     enum ifaddr_event event, struct ifaddr *ifa)
1373 {
1374 	if (ifa->ifa_rtrequest != arp_rtrequest) /* XXX need a generic way */
1375 		return;
1376 	if (ifa->ifa_addr->sa_family != AF_INET)
1377 		return;
1378 	if (event == IFADDR_EVENT_DELETE)
1379 		return;
1380 
1381 	/*
1382 	 * - CARP interfaces will take care of gratuitous ARP themselves.
1383 	 * - If we are the CARP interface's parent, don't send gratuitous
1384 	 *   ARP to avoid unnecessary confusion.
1385 	 */
1386 #ifdef CARP
1387 	if (ifp->if_type != IFT_CARP && ifp->if_carp == NULL)
1388 #endif
1389 	{
1390 		arp_gratuitous(ifp, ifa);
1391 	}
1392 }
1393 
1394 static void
arp_init_dispatch(netmsg_t nm)1395 arp_init_dispatch(netmsg_t nm)
1396 {
1397 	struct arp_pcpu_data *ad;
1398 
1399 	ASSERT_NETISR_NCPUS(mycpuid);
1400 
1401 	ad = kmalloc(sizeof(*ad), M_ARP, M_WAITOK | M_ZERO);
1402 
1403 	LIST_INIT(&ad->llinfo_list);
1404 	callout_init_mp(&ad->timer_ch);
1405 	netmsg_init(&ad->timer_nmsg, NULL, &netisr_adone_rport,
1406 	    MSGF_PRIORITY, arptimer_dispatch);
1407 	ad->timer_nmsg.lmsg.u.ms_resultp = ad;
1408 
1409 	arp_data[mycpuid] = ad;
1410 
1411 	callout_reset(&ad->timer_ch, hz, arptimer, &ad->timer_nmsg);
1412 
1413 	netisr_forwardmsg(&nm->base, mycpuid + 1);
1414 }
1415 
1416 static void
arp_init(void)1417 arp_init(void)
1418 {
1419 	struct netmsg_base nm;
1420 
1421 	netmsg_init(&nm, NULL, &curthread->td_msgport, 0, arp_init_dispatch);
1422 	netisr_domsg_global(&nm);
1423 
1424 	netisr_register(NETISR_ARP, arpintr, NULL);
1425 
1426 	EVENTHANDLER_REGISTER(ifaddr_event, arp_ifaddr, NULL,
1427 	    EVENTHANDLER_PRI_LAST);
1428 }
1429 SYSINIT(arp, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, arp_init, 0);
1430