2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1993
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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 University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)if_ether.c 8.1 (Berkeley) 6/10/93
35 * Ethernet address resolution protocol.
37 * add "inuse/lock" bit (or ref. count) along with valid bit
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
45 #include <sys/param.h>
46 #include <sys/eventhandler.h>
47 #include <sys/kernel.h>
49 #include <sys/queue.h>
50 #include <sys/sysctl.h>
51 #include <sys/systm.h>
53 #include <sys/malloc.h>
55 #include <sys/rmlock.h>
56 #include <sys/socket.h>
57 #include <sys/syslog.h>
60 #include <net/if_var.h>
61 #include <net/if_dl.h>
62 #include <net/if_types.h>
63 #include <net/netisr.h>
64 #include <net/ethernet.h>
65 #include <net/route.h>
66 #include <net/route/nhop.h>
69 #include <netinet/in.h>
70 #include <netinet/in_fib.h>
71 #include <netinet/in_var.h>
72 #include <net/if_llatbl.h>
73 #include <netinet/if_ether.h>
75 #include <netinet/ip_carp.h>
78 #include <security/mac/mac_framework.h>
80 #define SIN(s) ((const struct sockaddr_in *)(s))
82 static struct timeval arp_lastlog;
83 static int arp_curpps;
84 static int arp_maxpps = 1;
86 /* Simple ARP state machine */
87 enum arp_llinfo_state {
88 ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */
89 ARP_LLINFO_REACHABLE, /* LLE is valid */
90 ARP_LLINFO_VERIFY, /* LLE is valid, need refresh */
91 ARP_LLINFO_DELETED, /* LLE is deleted */
94 SYSCTL_DECL(_net_link_ether);
95 static SYSCTL_NODE(_net_link_ether, PF_INET, inet,
96 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
98 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp,
99 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
103 VNET_DEFINE_STATIC(int, arpt_keep) = (20*60); /* once resolved, good for 20
105 VNET_DEFINE_STATIC(int, arp_maxtries) = 5;
106 VNET_DEFINE_STATIC(int, arp_proxyall) = 0;
107 VNET_DEFINE_STATIC(int, arpt_down) = 20; /* keep incomplete entries for
109 VNET_DEFINE_STATIC(int, arpt_rexmit) = 1; /* retransmit arp entries, sec*/
110 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat); /* ARP statistics, see if_arp.h */
111 VNET_PCPUSTAT_SYSINIT(arpstat);
114 VNET_PCPUSTAT_SYSUNINIT(arpstat);
117 VNET_DEFINE_STATIC(int, arp_maxhold) = 1;
119 #define V_arpt_keep VNET(arpt_keep)
120 #define V_arpt_down VNET(arpt_down)
121 #define V_arpt_rexmit VNET(arpt_rexmit)
122 #define V_arp_maxtries VNET(arp_maxtries)
123 #define V_arp_proxyall VNET(arp_proxyall)
124 #define V_arp_maxhold VNET(arp_maxhold)
126 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
127 &VNET_NAME(arpt_keep), 0,
128 "ARP entry lifetime in seconds");
129 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
130 &VNET_NAME(arp_maxtries), 0,
131 "ARP resolution attempts before returning error");
132 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
133 &VNET_NAME(arp_proxyall), 0,
134 "Enable proxy ARP for all suitable requests");
135 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
136 &VNET_NAME(arpt_down), 0,
137 "Incomplete ARP entry lifetime in seconds");
138 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
139 arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
140 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
141 &VNET_NAME(arp_maxhold), 0,
142 "Number of packets to hold per ARP entry");
143 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
144 CTLFLAG_RW, &arp_maxpps, 0,
145 "Maximum number of remotely triggered ARP messages that can be "
146 "logged per second");
149 * Due to the exponential backoff algorithm used for the interval between GARP
150 * retransmissions, the maximum number of retransmissions is limited for
151 * sanity. This limit corresponds to a maximum interval between retransmissions
152 * of 2^16 seconds ~= 18 hours.
154 * Making this limit more dynamic is more complicated than worthwhile,
155 * especially since sending out GARPs spaced days apart would be of little
156 * use. A maximum dynamic limit would look something like:
158 * const int max = fls(INT_MAX / hz) - 1;
160 #define MAX_GARP_RETRANSMITS 16
161 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS);
162 static int garp_rexmit_count = 0; /* GARP retransmission setting. */
164 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count,
165 CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE,
166 &garp_rexmit_count, 0, sysctl_garp_rexmit, "I",
167 "Number of times to retransmit GARP packets;"
168 " 0 to disable, maximum of 16");
170 VNET_DEFINE_STATIC(int, arp_log_level) = LOG_INFO; /* Min. log(9) level. */
171 #define V_arp_log_level VNET(arp_log_level)
172 SYSCTL_INT(_net_link_ether_arp, OID_AUTO, log_level, CTLFLAG_VNET | CTLFLAG_RW,
173 &VNET_NAME(arp_log_level), 0,
174 "Minimum log(9) level for recording rate limited arp log messages. "
175 "The higher will be log more (emerg=0, info=6 (default), debug=7).");
176 #define ARP_LOG(pri, ...) do { \
177 if ((pri) <= V_arp_log_level && \
178 ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps)) \
179 log((pri), "arp: " __VA_ARGS__); \
182 static void arpintr(struct mbuf *);
183 static void arptimer(void *);
185 static void in_arpinput(struct mbuf *);
188 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
189 struct ifnet *ifp, int bridged, struct llentry *la);
190 static void arp_mark_lle_reachable(struct llentry *la);
191 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
193 static eventhandler_tag iflladdr_tag;
195 static const struct netisr_handler arp_nh = {
197 .nh_handler = arpintr,
198 .nh_proto = NETISR_ARP,
199 .nh_policy = NETISR_POLICY_SOURCE,
203 * Timeout routine. Age arp_tab entries periodically.
208 struct llentry *lle = (struct llentry *)arg;
212 if (lle->la_flags & LLE_STATIC) {
216 if (callout_pending(&lle->lle_timer)) {
218 * Here we are a bit odd here in the treatment of
219 * active/pending. If the pending bit is set, it got
220 * rescheduled before I ran. The active
221 * bit we ignore, since if it was stopped
222 * in ll_tablefree() and was currently running
223 * it would have return 0 so the code would
224 * not have deleted it since the callout could
225 * not be stopped so we want to go through
226 * with the delete here now. If the callout
227 * was restarted, the pending bit will be back on and
228 * we just want to bail since the callout_reset would
229 * return 1 and our reference would have been removed
230 * by arpresolve() below.
235 ifp = lle->lle_tbl->llt_ifp;
236 CURVNET_SET(ifp->if_vnet);
238 switch (lle->ln_state) {
239 case ARP_LLINFO_REACHABLE:
242 * Expiration time is approaching.
243 * Let's try to refresh entry if it is still
246 * Set r_skip_req to get feedback from
247 * fast path. Change state and re-schedule
253 lle->ln_state = ARP_LLINFO_VERIFY;
254 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
258 case ARP_LLINFO_VERIFY:
260 r_skip_req = lle->r_skip_req;
263 if (r_skip_req == 0 && lle->la_preempt > 0) {
264 /* Entry was used, issue refresh request */
265 struct epoch_tracker et;
268 dst = lle->r_l3addr.addr4;
270 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
273 arprequest(ifp, NULL, &dst, NULL);
278 /* Nothing happened. Reschedule if not too late */
279 if (lle->la_expire > time_uptime) {
280 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
286 case ARP_LLINFO_INCOMPLETE:
287 case ARP_LLINFO_DELETED:
291 if ((lle->la_flags & LLE_DELETED) == 0) {
294 if (lle->la_flags & LLE_VALID)
295 evt = LLENTRY_EXPIRED;
297 evt = LLENTRY_TIMEDOUT;
298 EVENTHANDLER_INVOKE(lle_event, lle, evt);
301 callout_stop(&lle->lle_timer);
303 /* XXX: LOR avoidance. We still have ref on lle. */
308 /* Guard against race with other llentry_free(). */
309 if (lle->la_flags & LLE_LINKED) {
311 lltable_unlink_entry(lle->lle_tbl, lle);
313 IF_AFDATA_UNLOCK(ifp);
315 size_t pkts_dropped = llentry_free(lle);
317 ARPSTAT_ADD(dropped, pkts_dropped);
318 ARPSTAT_INC(timeouts);
324 * Stores link-layer header for @ifp in format suitable for if_output()
325 * into buffer @buf. Resulting header length is stored in @bufsize.
327 * Returns 0 on success.
330 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
333 struct if_encap_req ereq;
336 bzero(buf, *bufsize);
337 bzero(&ereq, sizeof(ereq));
339 ereq.bufsize = *bufsize;
340 ereq.rtype = IFENCAP_LL;
341 ereq.family = AF_ARP;
342 ereq.lladdr = ar_tha(ah);
343 ereq.hdata = (u_char *)ah;
345 ereq.flags = IFENCAP_FLAG_BROADCAST;
346 error = ifp->if_requestencap(ifp, &ereq);
348 *bufsize = ereq.bufsize;
354 * Broadcast an ARP request. Caller specifies:
355 * - arp header source ip address
356 * - arp header target ip address
357 * - arp header source ethernet address
360 arprequest_internal(struct ifnet *ifp, const struct in_addr *sip,
361 const struct in_addr *tip, u_char *enaddr)
366 u_char *carpaddr = NULL;
367 uint8_t linkhdr[LLE_MAX_LINKHDR];
376 * The caller did not supply a source address, try to find
377 * a compatible one among those assigned to this interface.
381 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
382 if (ifa->ifa_addr->sa_family != AF_INET)
386 if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
388 sip = &IA_SIN(ifa)->sin_addr;
391 sip = &IA_SIN(ifa)->sin_addr;
394 if (0 == ((sip->s_addr ^ tip->s_addr) &
395 IA_MASKSIN(ifa)->sin_addr.s_addr))
396 break; /* found it. */
399 printf("%s: cannot find matching address\n", __func__);
400 return (EADDRNOTAVAIL);
404 enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
406 if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
408 m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
410 m->m_pkthdr.len = m->m_len;
411 M_ALIGN(m, m->m_len);
412 ah = mtod(m, struct arphdr *);
413 bzero((caddr_t)ah, m->m_len);
415 mac_netinet_arp_send(ifp, m);
417 ah->ar_pro = htons(ETHERTYPE_IP);
418 ah->ar_hln = ifp->if_addrlen; /* hardware address length */
419 ah->ar_pln = sizeof(struct in_addr); /* protocol address length */
420 ah->ar_op = htons(ARPOP_REQUEST);
421 bcopy(enaddr, ar_sha(ah), ah->ar_hln);
422 bcopy(sip, ar_spa(ah), ah->ar_pln);
423 bcopy(tip, ar_tpa(ah), ah->ar_pln);
424 sa.sa_family = AF_ARP;
427 /* Calculate link header for sending frame */
428 bzero(&ro, sizeof(ro));
429 linkhdrsize = sizeof(linkhdr);
430 error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
431 if (error != 0 && error != EAFNOSUPPORT) {
432 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
433 if_name(ifp), error);
437 ro.ro_prepend = linkhdr;
438 ro.ro_plen = linkhdrsize;
441 m->m_flags |= M_BCAST;
442 m_clrprotoflags(m); /* Avoid confusing lower layers. */
443 error = (*ifp->if_output)(ifp, m, &sa, &ro);
444 ARPSTAT_INC(txrequests);
446 ARPSTAT_INC(txerrors);
447 ARP_LOG(LOG_DEBUG, "Failed to send ARP packet on %s: %d\n",
448 if_name(ifp), error);
454 arprequest(struct ifnet *ifp, const struct in_addr *sip,
455 const struct in_addr *tip, u_char *enaddr)
458 (void) arprequest_internal(ifp, sip, tip, enaddr);
462 * Resolve an IP address into an ethernet address - heavy version.
463 * Used internally by arpresolve().
464 * We have already checked that we can't use an existing lle without
465 * modification so we have to acquire an LLE_EXCLUSIVE lle lock.
467 * On success, desten and pflags are filled in and the function returns 0;
468 * If the packet must be held pending resolution, we return EWOULDBLOCK
469 * On other errors, we return the corresponding error code.
470 * Note that m_freem() handles NULL.
473 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
474 const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
475 struct llentry **plle)
477 struct llentry *la = NULL, *la_tmp;
478 struct mbuf *curr = NULL;
479 struct mbuf *next = NULL;
491 if ((flags & LLE_CREATE) == 0)
492 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
493 if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
494 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
496 char addrbuf[INET_ADDRSTRLEN];
499 "arpresolve: can't allocate llinfo for %s on %s\n",
500 inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
506 IF_AFDATA_WLOCK(ifp);
508 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
509 /* Prefer ANY existing lle over newly-created one */
511 lltable_link_entry(LLTABLE(ifp), la);
512 IF_AFDATA_WUNLOCK(ifp);
513 if (la_tmp != NULL) {
514 lltable_free_entry(LLTABLE(ifp), la);
523 if ((la->la_flags & LLE_VALID) &&
524 ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
525 if (flags & LLE_ADDRONLY) {
526 lladdr = la->ll_addr;
527 ll_len = ifp->if_addrlen;
529 lladdr = la->r_linkdata;
530 ll_len = la->r_hdrlen;
532 bcopy(lladdr, desten, ll_len);
534 /* Notify LLE code that the entry was used by datapath */
535 llentry_mark_used(la);
537 *pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
546 renew = (la->la_asked == 0 || la->la_expire != time_uptime);
548 * There is an arptab entry, but no ethernet address
549 * response yet. Add the mbuf to the list, dropping
550 * the oldest packet if we have exceeded the system
554 if (la->la_numheld >= V_arp_maxhold) {
555 if (la->la_hold != NULL) {
556 next = la->la_hold->m_nextpkt;
557 m_freem(la->la_hold);
560 ARPSTAT_INC(dropped);
563 if (la->la_hold != NULL) {
565 while (curr->m_nextpkt != NULL)
566 curr = curr->m_nextpkt;
573 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
574 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
575 * if we have already sent arp_maxtries ARP requests. Retransmit the
576 * ARP request, but not faster than one request per second.
578 if (la->la_asked < V_arp_maxtries)
579 error = EWOULDBLOCK; /* First request. */
581 error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
587 la->la_expire = time_uptime;
588 canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
594 e = arprequest_internal(ifp, NULL, &SIN(dst)->sin_addr, NULL);
596 * Only overwrite 'error' in case of error; in case of success
597 * the proper return value was already set above.
609 * Lookups link header based on an IP address.
611 * ifp is the interface we use
612 * is_gw != 0 if @dst represents gateway to some destination
613 * m is the mbuf. May be NULL if we don't have a packet.
614 * dst is the next hop,
615 * desten is the storage to put LL header.
616 * flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
618 * On success, full/partial link header and flags are filled in and
619 * the function returns 0.
620 * If the packet must be held pending resolution, we return EWOULDBLOCK
621 * On other errors, we return the corresponding error code.
622 * Note that m_freem() handles NULL.
625 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
626 const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
627 struct llentry **plle)
629 struct llentry *la = NULL;
639 if (m->m_flags & M_BCAST) {
642 ifp->if_broadcastaddr, ifp->if_addrlen);
645 if (m->m_flags & M_MCAST) {
647 ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
652 la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
653 if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
654 /* Entry found, let's copy lle info */
655 bcopy(la->r_linkdata, desten, la->r_hdrlen);
657 *pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
658 /* Notify the LLE handling code that the entry was used. */
659 llentry_mark_used(la);
670 return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
671 desten, pflags, plle));
675 * Common length and type checks are done here,
676 * then the protocol-specific routine is called.
679 arpintr(struct mbuf *m)
686 ifp = m->m_pkthdr.rcvif;
688 if (m->m_len < sizeof(struct arphdr) &&
689 ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
690 ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
694 ar = mtod(m, struct arphdr *);
696 /* Check if length is sufficient */
697 if (m->m_len < arphdr_len(ar)) {
698 m = m_pullup(m, arphdr_len(ar));
700 ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
704 ar = mtod(m, struct arphdr *);
709 switch (ntohs(ar->ar_hrd)) {
711 hlen = ETHER_ADDR_LEN; /* RFC 826 */
714 case ARPHRD_INFINIBAND:
715 hlen = 20; /* RFC 4391, INFINIBAND_ALEN */
716 layer = "infiniband";
718 case ARPHRD_IEEE1394:
719 hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
723 * Restrict too long hardware addresses.
724 * Currently we are capable of handling 20-byte
725 * addresses ( sizeof(lle->ll_addr) )
727 if (ar->ar_hln >= 20)
732 "packet with unknown hardware format 0x%02d received on "
733 "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
738 if (hlen != 0 && hlen != ar->ar_hln) {
740 "packet with invalid %s address length %d received on %s\n",
741 layer, ar->ar_hln, if_name(ifp));
746 ARPSTAT_INC(received);
747 switch (ntohs(ar->ar_pro)) {
759 * ARP for Internet protocols on 10 Mb/s Ethernet.
760 * Algorithm is that given in RFC 826.
761 * In addition, a sanity check is performed on the sender
762 * protocol address, to catch impersonators.
763 * We no longer handle negotiations for use of trailer protocol:
764 * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
765 * along with IP replies if we wanted trailers sent to us,
766 * and also sent them in response to IP replies.
767 * This allowed either end to announce the desire to receive
769 * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
770 * but formerly didn't normally send requests.
772 static int log_arp_wrong_iface = 1;
773 static int log_arp_movements = 1;
774 static int log_arp_permanent_modify = 1;
775 static int allow_multicast = 0;
777 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
778 &log_arp_wrong_iface, 0,
779 "log arp packets arriving on the wrong interface");
780 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
781 &log_arp_movements, 0,
782 "log arp replies from MACs different than the one in the cache");
783 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
784 &log_arp_permanent_modify, 0,
785 "log arp replies from MACs different than the one in the permanent arp entry");
786 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
787 &allow_multicast, 0, "accept multicast addresses");
790 in_arpinput(struct mbuf *m)
792 struct rm_priotracker in_ifa_tracker;
794 struct ifnet *ifp = m->m_pkthdr.rcvif;
795 struct llentry *la = NULL, *la_tmp;
797 struct in_ifaddr *ia;
799 struct in_addr isaddr, itaddr, myaddr;
800 u_int8_t *enaddr = NULL;
802 int bridged = 0, is_bridge = 0;
804 struct sockaddr_in sin;
805 struct sockaddr *dst;
806 struct nhop_object *nh;
807 uint8_t linkhdr[LLE_MAX_LINKHDR];
812 char addrbuf[INET_ADDRSTRLEN];
816 sin.sin_len = sizeof(struct sockaddr_in);
817 sin.sin_family = AF_INET;
818 sin.sin_addr.s_addr = 0;
822 if (ifp->if_type == IFT_BRIDGE)
826 * We already have checked that mbuf contains enough contiguous data
827 * to hold entire arp message according to the arp header.
829 ah = mtod(m, struct arphdr *);
832 * ARP is only for IPv4 so we can reject packets with
833 * a protocol length not equal to an IPv4 address.
835 if (ah->ar_pln != sizeof(struct in_addr)) {
836 ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
837 sizeof(struct in_addr));
841 if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
842 ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
843 ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
847 op = ntohs(ah->ar_op);
848 (void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
849 (void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
851 if (op == ARPOP_REPLY)
852 ARPSTAT_INC(rxreplies);
855 * For a bridge, we want to check the address irrespective
856 * of the receive interface. (This will change slightly
857 * when we have clusters of interfaces).
859 IN_IFADDR_RLOCK(&in_ifa_tracker);
860 LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
861 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
862 ia->ia_ifp == ifp) &&
863 itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
864 (ia->ia_ifa.ifa_carp == NULL ||
865 (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
866 ifa_ref(&ia->ia_ifa);
867 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
871 LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
872 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
873 ia->ia_ifp == ifp) &&
874 isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
875 ifa_ref(&ia->ia_ifa);
876 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
880 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia) \
881 (ia->ia_ifp->if_bridge == ifp->if_softc && \
882 !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) && \
883 addr == ia->ia_addr.sin_addr.s_addr)
885 * Check the case when bridge shares its MAC address with
886 * some of its children, so packets are claimed by bridge
887 * itself (bridge_input() does it first), but they are really
888 * meant to be destined to the bridge member.
891 LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
892 if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
893 ifa_ref(&ia->ia_ifa);
895 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
900 #undef BDG_MEMBER_MATCHES_ARP
901 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
904 * No match, use the first inet address on the receive interface
905 * as a dummy address for the rest of the function.
907 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
908 if (ifa->ifa_addr->sa_family == AF_INET &&
909 (ifa->ifa_carp == NULL ||
910 (*carp_iamatch_p)(ifa, &enaddr))) {
917 * If bridging, fall back to using any inet address.
919 IN_IFADDR_RLOCK(&in_ifa_tracker);
920 if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL) {
921 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
924 ifa_ref(&ia->ia_ifa);
925 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
928 enaddr = (u_int8_t *)IF_LLADDR(ifp);
929 carped = (ia->ia_ifa.ifa_carp != NULL);
930 myaddr = ia->ia_addr.sin_addr;
931 ifa_free(&ia->ia_ifa);
932 if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
933 goto drop; /* it's from me, ignore it. */
934 if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
935 ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
936 "%s!\n", inet_ntoa_r(isaddr, addrbuf));
940 if (ifp->if_addrlen != ah->ar_hln) {
941 ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
942 "i/f %d (ignored)\n", ifp->if_addrlen,
943 (u_char *) ar_sha(ah), ":", ah->ar_hln,
949 * Warn if another host is using the same IP address, but only if the
950 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
951 * case we suppress the warning to avoid false positive complaints of
952 * potential misconfiguration.
954 if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
955 myaddr.s_addr != 0) {
956 ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
957 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
958 inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
963 if (ifp->if_flags & IFF_STATICARP)
966 bzero(&sin, sizeof(sin));
967 sin.sin_len = sizeof(struct sockaddr_in);
968 sin.sin_family = AF_INET;
969 sin.sin_addr = isaddr;
970 dst = (struct sockaddr *)&sin;
971 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
973 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
974 else if (itaddr.s_addr == myaddr.s_addr) {
976 * Request/reply to our address, but no lle exists yet.
977 * Calculate full link prepend to use in lle.
979 linkhdrsize = sizeof(linkhdr);
980 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
981 &linkhdrsize, &lladdr_off) != 0)
984 /* Allocate new entry */
985 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
988 * lle creation may fail if source address belongs
989 * to non-directly connected subnet. However, we
990 * will try to answer the request instead of dropping
995 lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
998 IF_AFDATA_WLOCK(ifp);
1000 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1003 * Check if lle still does not exists.
1004 * If it does, that means that we either
1005 * 1) have configured it explicitly, via
1006 * 1a) 'arp -s' static entry or
1007 * 1b) interface address static record
1009 * 2) it was the result of sending first packet to-host
1011 * 3) it was another arp reply packet we handled in
1014 * In all cases except 3) we definitely need to prefer
1015 * existing lle. For the sake of simplicity, prefer any
1016 * existing lle over newly-create one.
1019 lltable_link_entry(LLTABLE(ifp), la);
1020 IF_AFDATA_WUNLOCK(ifp);
1022 if (la_tmp == NULL) {
1023 arp_mark_lle_reachable(la);
1026 /* Free newly-create entry and handle packet */
1027 lltable_free_entry(LLTABLE(ifp), la);
1030 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1031 /* arp_check_update_lle() returns @la unlocked */
1036 if (op != ARPOP_REQUEST)
1038 ARPSTAT_INC(rxrequests);
1040 if (itaddr.s_addr == myaddr.s_addr) {
1041 /* Shortcut.. the receiving interface is the target. */
1042 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1043 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1046 * Destination address is not ours. Check if
1047 * proxyarp entry exists or proxyarp is turned on globally.
1049 struct llentry *lle;
1051 sin.sin_addr = itaddr;
1052 lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1054 if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1055 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1056 (void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1062 if (!V_arp_proxyall)
1066 nh = fib4_lookup(ifp->if_fib, itaddr, 0, 0, 0);
1071 * Don't send proxies for nodes on the same interface
1072 * as this one came out of, or we'll get into a fight
1073 * over who claims what Ether address.
1075 if (nh->nh_ifp == ifp)
1078 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1079 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1082 * Also check that the node which sent the ARP packet
1083 * is on the interface we expect it to be on. This
1084 * avoids ARP chaos if an interface is connected to the
1088 nh = fib4_lookup(ifp->if_fib, isaddr, 0, 0, 0);
1091 if (nh->nh_ifp != ifp) {
1092 ARP_LOG(LOG_INFO, "proxy: ignoring request"
1093 " from %s via %s\n",
1094 inet_ntoa_r(isaddr, addrbuf),
1100 printf("arp: proxying for %s\n",
1101 inet_ntoa_r(itaddr, addrbuf));
1106 if (itaddr.s_addr == myaddr.s_addr &&
1107 IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1108 /* RFC 3927 link-local IPv4; always reply by broadcast. */
1109 #ifdef DEBUG_LINKLOCAL
1110 printf("arp: sending reply for link-local addr %s\n",
1111 inet_ntoa_r(itaddr, addrbuf));
1113 m->m_flags |= M_BCAST;
1114 m->m_flags &= ~M_MCAST;
1116 /* default behaviour; never reply by broadcast. */
1117 m->m_flags &= ~(M_BCAST|M_MCAST);
1119 (void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1120 (void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1121 ah->ar_op = htons(ARPOP_REPLY);
1122 ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1123 m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1124 m->m_pkthdr.len = m->m_len;
1125 m->m_pkthdr.rcvif = NULL;
1126 sa.sa_family = AF_ARP;
1129 /* Calculate link header for sending frame */
1130 bzero(&ro, sizeof(ro));
1131 linkhdrsize = sizeof(linkhdr);
1132 error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1135 * arp_fillheader() may fail due to lack of support inside encap request
1136 * routing. This is not necessary an error, AF_ARP can/should be handled
1139 if (error != 0 && error != EAFNOSUPPORT) {
1140 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1141 if_name(ifp), error);
1145 ro.ro_prepend = linkhdr;
1146 ro.ro_plen = linkhdrsize;
1149 m_clrprotoflags(m); /* Avoid confusing lower layers. */
1150 (*ifp->if_output)(ifp, m, &sa, &ro);
1151 ARPSTAT_INC(txreplies);
1160 * Checks received arp data against existing @la.
1161 * Updates lle state/performs notification if necessary.
1164 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1165 int bridged, struct llentry *la)
1168 struct mbuf *m_hold, *m_hold_next;
1169 uint8_t linkhdr[LLE_MAX_LINKHDR];
1172 char addrbuf[INET_ADDRSTRLEN];
1174 LLE_WLOCK_ASSERT(la);
1176 /* the following is not an error when doing bridging */
1177 if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1178 if (log_arp_wrong_iface)
1179 ARP_LOG(LOG_WARNING, "%s is on %s "
1180 "but got reply from %*D on %s\n",
1181 inet_ntoa_r(isaddr, addrbuf),
1182 la->lle_tbl->llt_ifp->if_xname,
1183 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1188 if ((la->la_flags & LLE_VALID) &&
1189 bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1190 if (la->la_flags & LLE_STATIC) {
1192 if (log_arp_permanent_modify)
1194 "%*D attempts to modify "
1195 "permanent entry for %s on %s\n",
1197 (u_char *)ar_sha(ah), ":",
1198 inet_ntoa_r(isaddr, addrbuf),
1202 if (log_arp_movements) {
1203 ARP_LOG(LOG_INFO, "%s moved from %*D "
1205 inet_ntoa_r(isaddr, addrbuf),
1207 (u_char *)la->ll_addr, ":",
1208 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1213 /* Calculate full link prepend to use in lle */
1214 linkhdrsize = sizeof(linkhdr);
1215 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1216 &linkhdrsize, &lladdr_off) != 0)
1219 /* Check if something has changed */
1220 if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1221 (la->la_flags & LLE_VALID) == 0) {
1222 /* Try to perform LLE update */
1223 if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1227 /* Clear fast path feedback request if set */
1231 arp_mark_lle_reachable(la);
1234 * The packets are all freed within the call to the output
1237 * NB: The lock MUST be released before the call to the
1240 if (la->la_hold != NULL) {
1241 m_hold = la->la_hold;
1244 lltable_fill_sa_entry(la, &sa);
1246 for (; m_hold != NULL; m_hold = m_hold_next) {
1247 m_hold_next = m_hold->m_nextpkt;
1248 m_hold->m_nextpkt = NULL;
1249 /* Avoid confusing lower layers. */
1250 m_clrprotoflags(m_hold);
1251 (*ifp->if_output)(ifp, m_hold, &sa, NULL);
1258 arp_mark_lle_reachable(struct llentry *la)
1260 int canceled, wtime;
1262 LLE_WLOCK_ASSERT(la);
1264 la->ln_state = ARP_LLINFO_REACHABLE;
1265 EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1267 if (!(la->la_flags & LLE_STATIC)) {
1269 la->la_expire = time_uptime + V_arpt_keep;
1270 wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1272 wtime = V_arpt_keep;
1273 canceled = callout_reset(&la->lle_timer,
1274 hz * wtime, arptimer, la);
1279 la->la_preempt = V_arp_maxtries;
1283 * Add permanent link-layer record for given interface address.
1285 static __noinline void
1286 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1288 struct llentry *lle, *lle_tmp;
1291 * Interface address LLE record is considered static
1292 * because kernel code relies on LLE_STATIC flag to check
1293 * if these entries can be rewriten by arp updates.
1295 lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1297 log(LOG_INFO, "arp_ifinit: cannot create arp "
1298 "entry for interface address\n");
1302 IF_AFDATA_WLOCK(ifp);
1304 /* Unlink any entry if exists */
1305 lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1306 if (lle_tmp != NULL)
1307 lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1309 lltable_link_entry(LLTABLE(ifp), lle);
1310 IF_AFDATA_WUNLOCK(ifp);
1312 if (lle_tmp != NULL)
1313 EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1315 EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1317 if (lle_tmp != NULL)
1318 lltable_free_entry(LLTABLE(ifp), lle_tmp);
1322 * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1326 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1329 int rexmit_count = *(int *)arg1;
1331 error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1333 /* Enforce limits on any new value that may have been set. */
1334 if (!error && req->newptr) {
1335 /* A new value was set. */
1336 if (rexmit_count < 0) {
1338 } else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1339 rexmit_count = MAX_GARP_RETRANSMITS;
1341 *(int *)arg1 = rexmit_count;
1348 * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1349 * retransmit it again. A pending callout owns a reference to the ifa.
1352 garp_rexmit(void *arg)
1354 struct in_ifaddr *ia = arg;
1356 if (callout_pending(&ia->ia_garp_timer) ||
1357 !callout_active(&ia->ia_garp_timer)) {
1358 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1359 ifa_free(&ia->ia_ifa);
1363 CURVNET_SET(ia->ia_ifa.ifa_ifp->if_vnet);
1366 * Drop lock while the ARP request is generated.
1368 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1370 arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1371 &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1374 * Increment the count of retransmissions. If the count has reached the
1375 * maximum value, stop sending the GARP packets. Otherwise, schedule
1376 * the callout to retransmit another GARP packet.
1378 ++ia->ia_garp_count;
1379 if (ia->ia_garp_count >= garp_rexmit_count) {
1380 ifa_free(&ia->ia_ifa);
1383 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1384 rescheduled = callout_reset(&ia->ia_garp_timer,
1385 (1 << ia->ia_garp_count) * hz,
1387 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1389 ifa_free(&ia->ia_ifa);
1397 * Start the GARP retransmit timer.
1399 * A single GARP is always transmitted when an IPv4 address is added
1400 * to an interface and that is usually sufficient. However, in some
1401 * circumstances, such as when a shared address is passed between
1402 * cluster nodes, this single GARP may occasionally be dropped or
1403 * lost. This can lead to neighbors on the network link working with a
1404 * stale ARP cache and sending packets destined for that address to
1405 * the node that previously owned the address, which may not respond.
1407 * To avoid this situation, GARP retransmits can be enabled by setting
1408 * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1409 * than zero. The setting represents the maximum number of
1410 * retransmissions. The interval between retransmissions is calculated
1411 * using an exponential backoff algorithm, doubling each time, so the
1412 * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1415 garp_timer_start(struct ifaddr *ifa)
1417 struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1419 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1420 ia->ia_garp_count = 0;
1421 if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1422 garp_rexmit, ia) == 0) {
1425 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1429 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1431 struct epoch_tracker et;
1432 const struct sockaddr_in *dst_in;
1433 const struct sockaddr *dst;
1435 if (ifa->ifa_carp != NULL)
1438 dst = ifa->ifa_addr;
1439 dst_in = (const struct sockaddr_in *)dst;
1441 if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1443 NET_EPOCH_ENTER(et);
1444 arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1446 if (garp_rexmit_count > 0) {
1447 garp_timer_start(ifa);
1450 arp_add_ifa_lle(ifp, dst);
1454 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1457 if (ntohl(addr.s_addr) != INADDR_ANY)
1458 arprequest(ifp, &addr, &addr, enaddr);
1462 * Sends gratuitous ARPs for each ifaddr to notify other
1463 * nodes about the address change.
1465 static __noinline void
1466 arp_handle_ifllchange(struct ifnet *ifp)
1470 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1471 if (ifa->ifa_addr->sa_family == AF_INET)
1472 arp_ifinit(ifp, ifa);
1477 * A handler for interface link layer address change event.
1480 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1483 lltable_update_ifaddr(LLTABLE(ifp));
1485 if ((ifp->if_flags & IFF_UP) != 0)
1486 arp_handle_ifllchange(ifp);
1493 if (IS_DEFAULT_VNET(curvnet)) {
1494 netisr_register(&arp_nh);
1495 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1496 arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1500 netisr_register_vnet(&arp_nh);
1503 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1508 * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1509 * lookups after destroying the hash. Ideally this would go on SI_ORDER_3.5.
1512 vnet_arp_destroy(__unused void *arg)
1515 netisr_unregister_vnet(&arp_nh);
1517 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1518 vnet_arp_destroy, NULL);