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
33 * Ethernet address resolution protocol.
35 * add "inuse/lock" bit (or ref. count) along with valid bit
38 #include <sys/cdefs.h>
41 #include <sys/param.h>
42 #include <sys/eventhandler.h>
43 #include <sys/kernel.h>
45 #include <sys/queue.h>
46 #include <sys/sysctl.h>
47 #include <sys/systm.h>
49 #include <sys/malloc.h>
51 #include <sys/socket.h>
52 #include <sys/syslog.h>
55 #include <net/if_var.h>
56 #include <net/if_dl.h>
57 #include <net/if_private.h>
58 #include <net/if_types.h>
59 #include <net/netisr.h>
60 #include <net/ethernet.h>
61 #include <net/route.h>
62 #include <net/route/nhop.h>
65 #include <netinet/in.h>
66 #include <netinet/in_fib.h>
67 #include <netinet/in_var.h>
68 #include <net/if_llatbl.h>
69 #include <netinet/if_ether.h>
71 #include <netinet/ip_carp.h>
74 #include <security/mac/mac_framework.h>
76 #define SIN(s) ((const struct sockaddr_in *)(s))
78 static struct timeval arp_lastlog;
79 static int arp_curpps;
80 static int arp_maxpps = 1;
82 /* Simple ARP state machine */
83 enum arp_llinfo_state {
84 ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */
85 ARP_LLINFO_REACHABLE, /* LLE is valid */
86 ARP_LLINFO_VERIFY, /* LLE is valid, need refresh */
87 ARP_LLINFO_DELETED, /* LLE is deleted */
90 SYSCTL_DECL(_net_link_ether);
91 static SYSCTL_NODE(_net_link_ether, PF_INET, inet,
92 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
94 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp,
95 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
99 VNET_DEFINE_STATIC(int, arpt_keep) = (20*60); /* once resolved, good for 20
101 VNET_DEFINE_STATIC(int, arp_maxtries) = 5;
102 VNET_DEFINE_STATIC(int, arp_proxyall) = 0;
103 VNET_DEFINE_STATIC(int, arpt_down) = 20; /* keep incomplete entries for
105 VNET_DEFINE_STATIC(int, arpt_rexmit) = 1; /* retransmit arp entries, sec*/
106 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat); /* ARP statistics, see if_arp.h */
107 VNET_PCPUSTAT_SYSINIT(arpstat);
110 VNET_PCPUSTAT_SYSUNINIT(arpstat);
113 VNET_DEFINE_STATIC(int, arp_maxhold) = 16;
115 #define V_arpt_keep VNET(arpt_keep)
116 #define V_arpt_down VNET(arpt_down)
117 #define V_arpt_rexmit VNET(arpt_rexmit)
118 #define V_arp_maxtries VNET(arp_maxtries)
119 #define V_arp_proxyall VNET(arp_proxyall)
120 #define V_arp_maxhold VNET(arp_maxhold)
122 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
123 &VNET_NAME(arpt_keep), 0,
124 "ARP entry lifetime in seconds");
125 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
126 &VNET_NAME(arp_maxtries), 0,
127 "ARP resolution attempts before returning error");
128 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
129 &VNET_NAME(arp_proxyall), 0,
130 "Enable proxy ARP for all suitable requests");
131 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
132 &VNET_NAME(arpt_down), 0,
133 "Incomplete ARP entry lifetime in seconds");
134 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
135 arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
136 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
137 &VNET_NAME(arp_maxhold), 0,
138 "Number of packets to hold per ARP entry");
139 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
140 CTLFLAG_RW, &arp_maxpps, 0,
141 "Maximum number of remotely triggered ARP messages that can be "
142 "logged per second");
145 * Due to the exponential backoff algorithm used for the interval between GARP
146 * retransmissions, the maximum number of retransmissions is limited for
147 * sanity. This limit corresponds to a maximum interval between retransmissions
148 * of 2^16 seconds ~= 18 hours.
150 * Making this limit more dynamic is more complicated than worthwhile,
151 * especially since sending out GARPs spaced days apart would be of little
152 * use. A maximum dynamic limit would look something like:
154 * const int max = fls(INT_MAX / hz) - 1;
156 #define MAX_GARP_RETRANSMITS 16
157 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS);
158 static int garp_rexmit_count = 0; /* GARP retransmission setting. */
160 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count,
161 CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE,
162 &garp_rexmit_count, 0, sysctl_garp_rexmit, "I",
163 "Number of times to retransmit GARP packets;"
164 " 0 to disable, maximum of 16");
166 VNET_DEFINE_STATIC(int, arp_log_level) = LOG_INFO; /* Min. log(9) level. */
167 #define V_arp_log_level VNET(arp_log_level)
168 SYSCTL_INT(_net_link_ether_arp, OID_AUTO, log_level, CTLFLAG_VNET | CTLFLAG_RW,
169 &VNET_NAME(arp_log_level), 0,
170 "Minimum log(9) level for recording rate limited arp log messages. "
171 "The higher will be log more (emerg=0, info=6 (default), debug=7).");
172 #define ARP_LOG(pri, ...) do { \
173 if ((pri) <= V_arp_log_level && \
174 ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps)) \
175 log((pri), "arp: " __VA_ARGS__); \
178 static void arpintr(struct mbuf *);
179 static void arptimer(void *);
181 static void in_arpinput(struct mbuf *);
184 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
185 struct ifnet *ifp, int bridged, struct llentry *la);
186 static void arp_mark_lle_reachable(struct llentry *la, struct ifnet *ifp);
187 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
189 static eventhandler_tag iflladdr_tag;
191 static const struct netisr_handler arp_nh = {
193 .nh_handler = arpintr,
194 .nh_proto = NETISR_ARP,
195 .nh_policy = NETISR_POLICY_SOURCE,
199 * Timeout routine. Age arp_tab entries periodically.
204 struct llentry *lle = (struct llentry *)arg;
207 if (lle->la_flags & LLE_STATIC) {
211 if (callout_pending(&lle->lle_timer)) {
213 * Here we are a bit odd here in the treatment of
214 * active/pending. If the pending bit is set, it got
215 * rescheduled before I ran. The active
216 * bit we ignore, since if it was stopped
217 * in ll_tablefree() and was currently running
218 * it would have return 0 so the code would
219 * not have deleted it since the callout could
220 * not be stopped so we want to go through
221 * with the delete here now. If the callout
222 * was restarted, the pending bit will be back on and
223 * we just want to bail since the callout_reset would
224 * return 1 and our reference would have been removed
225 * by arpresolve() below.
230 ifp = lle->lle_tbl->llt_ifp;
231 CURVNET_SET(ifp->if_vnet);
233 switch (lle->ln_state) {
234 case ARP_LLINFO_REACHABLE:
237 * Expiration time is approaching.
238 * Request usage feedback from the datapath.
239 * Change state and re-schedule ourselves.
241 llentry_request_feedback(lle);
242 lle->ln_state = ARP_LLINFO_VERIFY;
243 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
247 case ARP_LLINFO_VERIFY:
248 if (llentry_get_hittime(lle) > 0 && lle->la_preempt > 0) {
249 /* Entry was used, issue refresh request */
250 struct epoch_tracker et;
253 dst = lle->r_l3addr.addr4;
255 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
258 arprequest(ifp, NULL, &dst, NULL);
263 /* Nothing happened. Reschedule if not too late */
264 if (lle->la_expire > time_uptime) {
265 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
271 case ARP_LLINFO_INCOMPLETE:
272 case ARP_LLINFO_DELETED:
276 if ((lle->la_flags & LLE_DELETED) == 0) {
279 if (lle->la_flags & LLE_VALID)
280 evt = LLENTRY_EXPIRED;
282 evt = LLENTRY_TIMEDOUT;
283 EVENTHANDLER_INVOKE(lle_event, lle, evt);
286 callout_stop(&lle->lle_timer);
288 /* XXX: LOR avoidance. We still have ref on lle. */
293 /* Guard against race with other llentry_free(). */
294 if (lle->la_flags & LLE_LINKED) {
296 lltable_unlink_entry(lle->lle_tbl, lle);
298 IF_AFDATA_UNLOCK(ifp);
300 size_t pkts_dropped = llentry_free(lle);
302 ARPSTAT_ADD(dropped, pkts_dropped);
303 ARPSTAT_INC(timeouts);
309 * Stores link-layer header for @ifp in format suitable for if_output()
310 * into buffer @buf. Resulting header length is stored in @bufsize.
312 * Returns 0 on success.
315 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
318 struct if_encap_req ereq;
321 bzero(buf, *bufsize);
322 bzero(&ereq, sizeof(ereq));
324 ereq.bufsize = *bufsize;
325 ereq.rtype = IFENCAP_LL;
326 ereq.family = AF_ARP;
327 ereq.lladdr = ar_tha(ah);
328 ereq.hdata = (u_char *)ah;
330 ereq.flags = IFENCAP_FLAG_BROADCAST;
331 error = ifp->if_requestencap(ifp, &ereq);
333 *bufsize = ereq.bufsize;
339 * Broadcast an ARP request. Caller specifies:
340 * - arp header source ip address
341 * - arp header target ip address
342 * - arp header source ethernet address
345 arprequest_internal(struct ifnet *ifp, const struct in_addr *sip,
346 const struct in_addr *tip, u_char *enaddr)
351 u_char *carpaddr = NULL;
352 uint8_t linkhdr[LLE_MAX_LINKHDR];
361 * The caller did not supply a source address, try to find
362 * a compatible one among those assigned to this interface.
366 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
367 if (ifa->ifa_addr->sa_family != AF_INET)
371 if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
373 sip = &IA_SIN(ifa)->sin_addr;
376 sip = &IA_SIN(ifa)->sin_addr;
379 if (0 == ((sip->s_addr ^ tip->s_addr) &
380 IA_MASKSIN(ifa)->sin_addr.s_addr))
381 break; /* found it. */
384 printf("%s: cannot find matching address\n", __func__);
385 return (EADDRNOTAVAIL);
389 enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
391 if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
393 m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
395 m->m_pkthdr.len = m->m_len;
396 M_ALIGN(m, m->m_len);
397 ah = mtod(m, struct arphdr *);
398 bzero((caddr_t)ah, m->m_len);
400 mac_netinet_arp_send(ifp, m);
402 ah->ar_pro = htons(ETHERTYPE_IP);
403 ah->ar_hln = ifp->if_addrlen; /* hardware address length */
404 ah->ar_pln = sizeof(struct in_addr); /* protocol address length */
405 ah->ar_op = htons(ARPOP_REQUEST);
406 bcopy(enaddr, ar_sha(ah), ah->ar_hln);
407 bcopy(sip, ar_spa(ah), ah->ar_pln);
408 bcopy(tip, ar_tpa(ah), ah->ar_pln);
409 sa.sa_family = AF_ARP;
412 /* Calculate link header for sending frame */
413 bzero(&ro, sizeof(ro));
414 linkhdrsize = sizeof(linkhdr);
415 error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
416 if (error != 0 && error != EAFNOSUPPORT) {
418 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
419 if_name(ifp), error);
423 ro.ro_prepend = linkhdr;
424 ro.ro_plen = linkhdrsize;
427 m->m_flags |= M_BCAST;
428 m_clrprotoflags(m); /* Avoid confusing lower layers. */
429 error = (*ifp->if_output)(ifp, m, &sa, &ro);
430 ARPSTAT_INC(txrequests);
432 ARPSTAT_INC(txerrors);
433 ARP_LOG(LOG_DEBUG, "Failed to send ARP packet on %s: %d\n",
434 if_name(ifp), error);
440 arprequest(struct ifnet *ifp, const struct in_addr *sip,
441 const struct in_addr *tip, u_char *enaddr)
444 (void) arprequest_internal(ifp, sip, tip, enaddr);
448 * Resolve an IP address into an ethernet address - heavy version.
449 * Used internally by arpresolve().
450 * We have already checked that we can't use an existing lle without
451 * modification so we have to acquire an LLE_EXCLUSIVE lle lock.
453 * On success, desten and pflags are filled in and the function returns 0;
454 * If the packet must be held pending resolution, we return EWOULDBLOCK
455 * On other errors, we return the corresponding error code.
456 * Note that m_freem() handles NULL.
459 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
460 const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
461 struct llentry **plle)
463 struct llentry *la = NULL, *la_tmp;
475 if ((flags & LLE_CREATE) == 0)
476 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
477 if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
478 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
480 char addrbuf[INET_ADDRSTRLEN];
483 "arpresolve: can't allocate llinfo for %s on %s\n",
484 inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
490 IF_AFDATA_WLOCK(ifp);
492 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
493 /* Prefer ANY existing lle over newly-created one */
495 lltable_link_entry(LLTABLE(ifp), la);
496 IF_AFDATA_WUNLOCK(ifp);
497 if (la_tmp != NULL) {
498 lltable_free_entry(LLTABLE(ifp), la);
507 if ((la->la_flags & LLE_VALID) &&
508 ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
509 if (flags & LLE_ADDRONLY) {
510 lladdr = la->ll_addr;
511 ll_len = ifp->if_addrlen;
513 lladdr = la->r_linkdata;
514 ll_len = la->r_hdrlen;
516 bcopy(lladdr, desten, ll_len);
518 /* Notify LLE code that the entry was used by datapath */
519 llentry_provide_feedback(la);
521 *pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
530 renew = (la->la_asked == 0 || la->la_expire != time_uptime);
533 * There is an arptab entry, but no ethernet address
534 * response yet. Add the mbuf to the list, dropping
535 * the oldest packet if we have exceeded the system
539 size_t dropped = lltable_append_entry_queue(la, m, V_arp_maxhold);
540 ARPSTAT_ADD(dropped, dropped);
544 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
545 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
546 * if we have already sent arp_maxtries ARP requests. Retransmit the
547 * ARP request, but not faster than one request per second.
549 if (la->la_asked < V_arp_maxtries)
550 error = EWOULDBLOCK; /* First request. */
552 error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
558 la->la_expire = time_uptime;
559 canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
565 e = arprequest_internal(ifp, NULL, &SIN(dst)->sin_addr, NULL);
567 * Only overwrite 'error' in case of error; in case of success
568 * the proper return value was already set above.
580 * Lookups link header based on an IP address.
582 * ifp is the interface we use
583 * is_gw != 0 if @dst represents gateway to some destination
584 * m is the mbuf. May be NULL if we don't have a packet.
585 * dst is the next hop,
586 * desten is the storage to put LL header.
587 * flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
589 * On success, full/partial link header and flags are filled in and
590 * the function returns 0.
591 * If the packet must be held pending resolution, we return EWOULDBLOCK
592 * On other errors, we return the corresponding error code.
593 * Note that m_freem() handles NULL.
596 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
597 const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
598 struct llentry **plle)
600 struct llentry *la = NULL;
610 if (m->m_flags & M_BCAST) {
613 ifp->if_broadcastaddr, ifp->if_addrlen);
616 if (m->m_flags & M_MCAST) {
618 ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
623 la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
624 if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
625 /* Entry found, let's copy lle info */
626 bcopy(la->r_linkdata, desten, la->r_hdrlen);
628 *pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
629 /* Notify the LLE handling code that the entry was used. */
630 llentry_provide_feedback(la);
641 return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
642 desten, pflags, plle));
646 * Common length and type checks are done here,
647 * then the protocol-specific routine is called.
650 arpintr(struct mbuf *m)
657 ifp = m->m_pkthdr.rcvif;
659 if (m->m_len < sizeof(struct arphdr) &&
660 ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
661 ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
665 ar = mtod(m, struct arphdr *);
667 /* Check if length is sufficient */
668 if (m->m_len < arphdr_len(ar)) {
669 m = m_pullup(m, arphdr_len(ar));
671 ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
675 ar = mtod(m, struct arphdr *);
680 switch (ntohs(ar->ar_hrd)) {
682 hlen = ETHER_ADDR_LEN; /* RFC 826 */
685 case ARPHRD_INFINIBAND:
686 hlen = 20; /* RFC 4391, INFINIBAND_ALEN */
687 layer = "infiniband";
689 case ARPHRD_IEEE1394:
690 hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
694 * Restrict too long hardware addresses.
695 * Currently we are capable of handling 20-byte
696 * addresses ( sizeof(lle->ll_addr) )
698 if (ar->ar_hln >= 20)
703 "packet with unknown hardware format 0x%02d received on "
704 "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
709 if (hlen != 0 && hlen != ar->ar_hln) {
711 "packet with invalid %s address length %d received on %s\n",
712 layer, ar->ar_hln, if_name(ifp));
717 ARPSTAT_INC(received);
718 switch (ntohs(ar->ar_pro)) {
730 * ARP for Internet protocols on 10 Mb/s Ethernet.
731 * Algorithm is that given in RFC 826.
732 * In addition, a sanity check is performed on the sender
733 * protocol address, to catch impersonators.
734 * We no longer handle negotiations for use of trailer protocol:
735 * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
736 * along with IP replies if we wanted trailers sent to us,
737 * and also sent them in response to IP replies.
738 * This allowed either end to announce the desire to receive
740 * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
741 * but formerly didn't normally send requests.
743 static int log_arp_wrong_iface = 1;
744 static int log_arp_movements = 1;
745 static int log_arp_permanent_modify = 1;
746 static int allow_multicast = 0;
748 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
749 &log_arp_wrong_iface, 0,
750 "log arp packets arriving on the wrong interface");
751 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
752 &log_arp_movements, 0,
753 "log arp replies from MACs different than the one in the cache");
754 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
755 &log_arp_permanent_modify, 0,
756 "log arp replies from MACs different than the one in the permanent arp entry");
757 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
758 &allow_multicast, 0, "accept multicast addresses");
761 in_arpinput(struct mbuf *m)
764 struct ifnet *ifp = m->m_pkthdr.rcvif;
765 struct llentry *la = NULL, *la_tmp;
767 struct in_ifaddr *ia;
769 struct in_addr isaddr, itaddr, myaddr;
770 u_int8_t *enaddr = NULL;
772 int bridged = 0, is_bridge = 0;
774 struct sockaddr_in sin;
775 struct sockaddr *dst;
776 struct nhop_object *nh;
777 uint8_t linkhdr[LLE_MAX_LINKHDR];
782 char addrbuf[INET_ADDRSTRLEN];
786 sin.sin_len = sizeof(struct sockaddr_in);
787 sin.sin_family = AF_INET;
788 sin.sin_addr.s_addr = 0;
792 if (ifp->if_type == IFT_BRIDGE)
796 * We already have checked that mbuf contains enough contiguous data
797 * to hold entire arp message according to the arp header.
799 ah = mtod(m, struct arphdr *);
802 * ARP is only for IPv4 so we can reject packets with
803 * a protocol length not equal to an IPv4 address.
805 if (ah->ar_pln != sizeof(struct in_addr)) {
806 ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
807 sizeof(struct in_addr));
811 if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
812 ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
813 ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
817 op = ntohs(ah->ar_op);
818 (void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
819 (void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
821 if (op == ARPOP_REPLY)
822 ARPSTAT_INC(rxreplies);
825 * For a bridge, we want to check the address irrespective
826 * of the receive interface. (This will change slightly
827 * when we have clusters of interfaces).
829 CK_LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
830 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
831 ia->ia_ifp == ifp) &&
832 itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
833 (ia->ia_ifa.ifa_carp == NULL ||
834 (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
835 ifa_ref(&ia->ia_ifa);
839 CK_LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
840 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
841 ia->ia_ifp == ifp) &&
842 isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
843 ifa_ref(&ia->ia_ifa);
847 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia) \
848 (ia->ia_ifp->if_bridge == ifp->if_softc && \
849 !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) && \
850 addr == ia->ia_addr.sin_addr.s_addr)
852 * Check the case when bridge shares its MAC address with
853 * some of its children, so packets are claimed by bridge
854 * itself (bridge_input() does it first), but they are really
855 * meant to be destined to the bridge member.
858 CK_LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
859 if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
860 ifa_ref(&ia->ia_ifa);
866 #undef BDG_MEMBER_MATCHES_ARP
869 * No match, use the first inet address on the receive interface
870 * as a dummy address for the rest of the function.
872 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
873 if (ifa->ifa_addr->sa_family == AF_INET &&
874 (ifa->ifa_carp == NULL ||
875 (*carp_iamatch_p)(ifa, &enaddr))) {
882 * If bridging, fall back to using any inet address.
884 if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL)
886 ifa_ref(&ia->ia_ifa);
889 enaddr = (u_int8_t *)IF_LLADDR(ifp);
890 carped = (ia->ia_ifa.ifa_carp != NULL);
891 myaddr = ia->ia_addr.sin_addr;
892 ifa_free(&ia->ia_ifa);
893 if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
894 goto drop; /* it's from me, ignore it. */
895 if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
896 ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
897 "%s!\n", inet_ntoa_r(isaddr, addrbuf));
901 if (ifp->if_addrlen != ah->ar_hln) {
902 ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
903 "i/f %d (ignored)\n", ifp->if_addrlen,
904 (u_char *) ar_sha(ah), ":", ah->ar_hln,
910 * Warn if another host is using the same IP address, but only if the
911 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
912 * case we suppress the warning to avoid false positive complaints of
913 * potential misconfiguration.
915 if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
916 myaddr.s_addr != 0) {
917 ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
918 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
919 inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
924 if (ifp->if_flags & IFF_STATICARP)
927 bzero(&sin, sizeof(sin));
928 sin.sin_len = sizeof(struct sockaddr_in);
929 sin.sin_family = AF_INET;
930 sin.sin_addr = isaddr;
931 dst = (struct sockaddr *)&sin;
932 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
934 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
935 else if (itaddr.s_addr == myaddr.s_addr) {
937 * Request/reply to our address, but no lle exists yet.
938 * Calculate full link prepend to use in lle.
940 linkhdrsize = sizeof(linkhdr);
941 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
942 &linkhdrsize, &lladdr_off) != 0)
945 /* Allocate new entry */
946 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
949 * lle creation may fail if source address belongs
950 * to non-directly connected subnet. However, we
951 * will try to answer the request instead of dropping
956 lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
959 IF_AFDATA_WLOCK(ifp);
961 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
964 * Check if lle still does not exists.
965 * If it does, that means that we either
966 * 1) have configured it explicitly, via
967 * 1a) 'arp -s' static entry or
968 * 1b) interface address static record
970 * 2) it was the result of sending first packet to-host
972 * 3) it was another arp reply packet we handled in
975 * In all cases except 3) we definitely need to prefer
976 * existing lle. For the sake of simplicity, prefer any
977 * existing lle over newly-create one.
980 lltable_link_entry(LLTABLE(ifp), la);
981 IF_AFDATA_WUNLOCK(ifp);
983 if (la_tmp == NULL) {
984 arp_mark_lle_reachable(la, ifp);
987 /* Free newly-create entry and handle packet */
988 lltable_free_entry(LLTABLE(ifp), la);
991 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
992 /* arp_check_update_lle() returns @la unlocked */
997 if (op != ARPOP_REQUEST)
999 ARPSTAT_INC(rxrequests);
1001 if (itaddr.s_addr == myaddr.s_addr) {
1002 /* Shortcut.. the receiving interface is the target. */
1003 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1004 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1007 * Destination address is not ours. Check if
1008 * proxyarp entry exists or proxyarp is turned on globally.
1010 struct llentry *lle;
1012 sin.sin_addr = itaddr;
1013 lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1015 if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1016 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1017 (void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1023 if (!V_arp_proxyall)
1027 nh = fib4_lookup(ifp->if_fib, itaddr, 0, 0, 0);
1032 * Don't send proxies for nodes on the same interface
1033 * as this one came out of, or we'll get into a fight
1034 * over who claims what Ether address.
1036 if (nh->nh_ifp == ifp)
1039 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1040 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1043 * Also check that the node which sent the ARP packet
1044 * is on the interface we expect it to be on. This
1045 * avoids ARP chaos if an interface is connected to the
1049 nh = fib4_lookup(ifp->if_fib, isaddr, 0, 0, 0);
1052 if (nh->nh_ifp != ifp) {
1053 ARP_LOG(LOG_INFO, "proxy: ignoring request"
1054 " from %s via %s\n",
1055 inet_ntoa_r(isaddr, addrbuf),
1061 printf("arp: proxying for %s\n",
1062 inet_ntoa_r(itaddr, addrbuf));
1067 if (itaddr.s_addr == myaddr.s_addr &&
1068 IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1069 /* RFC 3927 link-local IPv4; always reply by broadcast. */
1070 #ifdef DEBUG_LINKLOCAL
1071 printf("arp: sending reply for link-local addr %s\n",
1072 inet_ntoa_r(itaddr, addrbuf));
1074 m->m_flags |= M_BCAST;
1075 m->m_flags &= ~M_MCAST;
1077 /* default behaviour; never reply by broadcast. */
1078 m->m_flags &= ~(M_BCAST|M_MCAST);
1080 (void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1081 (void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1082 ah->ar_op = htons(ARPOP_REPLY);
1083 ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1084 m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1085 m->m_pkthdr.len = m->m_len;
1086 m->m_pkthdr.rcvif = NULL;
1087 sa.sa_family = AF_ARP;
1090 /* Calculate link header for sending frame */
1091 bzero(&ro, sizeof(ro));
1092 linkhdrsize = sizeof(linkhdr);
1093 error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1096 * arp_fillheader() may fail due to lack of support inside encap request
1097 * routing. This is not necessary an error, AF_ARP can/should be handled
1100 if (error != 0 && error != EAFNOSUPPORT) {
1101 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1102 if_name(ifp), error);
1106 ro.ro_prepend = linkhdr;
1107 ro.ro_plen = linkhdrsize;
1110 m_clrprotoflags(m); /* Avoid confusing lower layers. */
1111 (*ifp->if_output)(ifp, m, &sa, &ro);
1112 ARPSTAT_INC(txreplies);
1120 static struct mbuf *
1121 arp_grab_holdchain(struct llentry *la)
1125 LLE_WLOCK_ASSERT(la);
1127 chain = la->la_hold;
1135 arp_flush_holdchain(struct ifnet *ifp, struct llentry *la, struct mbuf *chain)
1137 struct mbuf *m_hold, *m_hold_next;
1138 struct sockaddr_in sin;
1143 .ro_prepend = la->r_linkdata,
1144 .ro_plen = la->r_hdrlen,
1147 lltable_fill_sa_entry(la, (struct sockaddr *)&sin);
1149 for (m_hold = chain; m_hold != NULL; m_hold = m_hold_next) {
1150 m_hold_next = m_hold->m_nextpkt;
1151 m_hold->m_nextpkt = NULL;
1152 /* Avoid confusing lower layers. */
1153 m_clrprotoflags(m_hold);
1154 (*ifp->if_output)(ifp, m_hold, (struct sockaddr *)&sin, &ro);
1159 * Checks received arp data against existing @la.
1160 * Updates lle state/performs notification if necessary.
1163 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1164 int bridged, struct llentry *la)
1166 uint8_t linkhdr[LLE_MAX_LINKHDR];
1169 char addrbuf[INET_ADDRSTRLEN];
1171 LLE_WLOCK_ASSERT(la);
1173 /* the following is not an error when doing bridging */
1174 if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1175 if (log_arp_wrong_iface)
1176 ARP_LOG(LOG_WARNING, "%s is on %s "
1177 "but got reply from %*D on %s\n",
1178 inet_ntoa_r(isaddr, addrbuf),
1179 la->lle_tbl->llt_ifp->if_xname,
1180 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1185 if ((la->la_flags & LLE_VALID) &&
1186 bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1187 if (la->la_flags & LLE_STATIC) {
1189 if (log_arp_permanent_modify)
1191 "%*D attempts to modify "
1192 "permanent entry for %s on %s\n",
1194 (u_char *)ar_sha(ah), ":",
1195 inet_ntoa_r(isaddr, addrbuf),
1199 if (log_arp_movements) {
1200 ARP_LOG(LOG_INFO, "%s moved from %*D "
1202 inet_ntoa_r(isaddr, addrbuf),
1204 (u_char *)la->ll_addr, ":",
1205 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1210 /* Calculate full link prepend to use in lle */
1211 linkhdrsize = sizeof(linkhdr);
1212 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1213 &linkhdrsize, &lladdr_off) != 0) {
1218 /* Check if something has changed */
1219 if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1220 (la->la_flags & LLE_VALID) == 0) {
1221 /* Try to perform LLE update */
1222 if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1228 /* Clear fast path feedback request if set */
1229 llentry_mark_used(la);
1232 arp_mark_lle_reachable(la, ifp);
1235 * The packets are all freed within the call to the output
1238 * NB: The lock MUST be released before the call to the
1241 if (la->la_hold != NULL) {
1244 chain = arp_grab_holdchain(la);
1246 arp_flush_holdchain(ifp, la, chain);
1252 arp_mark_lle_reachable(struct llentry *la, struct ifnet *ifp)
1254 int canceled, wtime;
1256 LLE_WLOCK_ASSERT(la);
1258 la->ln_state = ARP_LLINFO_REACHABLE;
1259 EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1261 if ((ifp->if_flags & IFF_STICKYARP) != 0)
1262 la->la_flags |= LLE_STATIC;
1264 if (!(la->la_flags & LLE_STATIC)) {
1266 la->la_expire = time_uptime + V_arpt_keep;
1267 wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1269 wtime = V_arpt_keep;
1270 canceled = callout_reset(&la->lle_timer,
1271 hz * wtime, arptimer, la);
1276 la->la_preempt = V_arp_maxtries;
1280 * Add permanent link-layer record for given interface address.
1282 static __noinline void
1283 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1285 struct llentry *lle, *lle_tmp;
1288 * Interface address LLE record is considered static
1289 * because kernel code relies on LLE_STATIC flag to check
1290 * if these entries can be rewriten by arp updates.
1292 lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1294 log(LOG_INFO, "arp_ifinit: cannot create arp "
1295 "entry for interface address\n");
1299 IF_AFDATA_WLOCK(ifp);
1301 /* Unlink any entry if exists */
1302 lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1303 if (lle_tmp != NULL)
1304 lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1306 lltable_link_entry(LLTABLE(ifp), lle);
1307 IF_AFDATA_WUNLOCK(ifp);
1309 if (lle_tmp != NULL)
1310 EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1312 EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1314 if (lle_tmp != NULL)
1315 lltable_free_entry(LLTABLE(ifp), lle_tmp);
1319 * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1323 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1326 int rexmit_count = *(int *)arg1;
1328 error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1330 /* Enforce limits on any new value that may have been set. */
1331 if (!error && req->newptr) {
1332 /* A new value was set. */
1333 if (rexmit_count < 0) {
1335 } else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1336 rexmit_count = MAX_GARP_RETRANSMITS;
1338 *(int *)arg1 = rexmit_count;
1345 * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1346 * retransmit it again. A pending callout owns a reference to the ifa.
1349 garp_rexmit(void *arg)
1351 struct in_ifaddr *ia = arg;
1353 if (callout_pending(&ia->ia_garp_timer) ||
1354 !callout_active(&ia->ia_garp_timer)) {
1355 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1356 ifa_free(&ia->ia_ifa);
1360 CURVNET_SET(ia->ia_ifa.ifa_ifp->if_vnet);
1363 * Drop lock while the ARP request is generated.
1365 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1367 arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1368 &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1371 * Increment the count of retransmissions. If the count has reached the
1372 * maximum value, stop sending the GARP packets. Otherwise, schedule
1373 * the callout to retransmit another GARP packet.
1375 ++ia->ia_garp_count;
1376 if (ia->ia_garp_count >= garp_rexmit_count) {
1377 ifa_free(&ia->ia_ifa);
1380 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1381 rescheduled = callout_reset(&ia->ia_garp_timer,
1382 (1 << ia->ia_garp_count) * hz,
1384 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1386 ifa_free(&ia->ia_ifa);
1394 * Start the GARP retransmit timer.
1396 * A single GARP is always transmitted when an IPv4 address is added
1397 * to an interface and that is usually sufficient. However, in some
1398 * circumstances, such as when a shared address is passed between
1399 * cluster nodes, this single GARP may occasionally be dropped or
1400 * lost. This can lead to neighbors on the network link working with a
1401 * stale ARP cache and sending packets destined for that address to
1402 * the node that previously owned the address, which may not respond.
1404 * To avoid this situation, GARP retransmits can be enabled by setting
1405 * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1406 * than zero. The setting represents the maximum number of
1407 * retransmissions. The interval between retransmissions is calculated
1408 * using an exponential backoff algorithm, doubling each time, so the
1409 * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1412 garp_timer_start(struct ifaddr *ifa)
1414 struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1416 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1417 ia->ia_garp_count = 0;
1418 if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1419 garp_rexmit, ia) == 0) {
1422 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1426 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1428 struct epoch_tracker et;
1429 const struct sockaddr_in *dst_in;
1430 const struct sockaddr *dst;
1432 if (ifa->ifa_carp != NULL)
1435 dst = ifa->ifa_addr;
1436 dst_in = (const struct sockaddr_in *)dst;
1438 if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1440 NET_EPOCH_ENTER(et);
1441 arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1443 if (garp_rexmit_count > 0) {
1444 garp_timer_start(ifa);
1447 arp_add_ifa_lle(ifp, dst);
1451 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1454 if (ntohl(addr.s_addr) != INADDR_ANY)
1455 arprequest(ifp, &addr, &addr, enaddr);
1459 * Sends gratuitous ARPs for each ifaddr to notify other
1460 * nodes about the address change.
1462 static __noinline void
1463 arp_handle_ifllchange(struct ifnet *ifp)
1467 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1468 if (ifa->ifa_addr->sa_family == AF_INET)
1469 arp_ifinit(ifp, ifa);
1474 * A handler for interface link layer address change event.
1477 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1479 /* if_bridge can update its lladdr during if_vmove(), after we've done
1480 * if_detach_internal()/dom_ifdetach(). */
1481 if (ifp->if_afdata[AF_INET] == NULL)
1484 lltable_update_ifaddr(LLTABLE(ifp));
1486 if ((ifp->if_flags & IFF_UP) != 0)
1487 arp_handle_ifllchange(ifp);
1494 if (IS_DEFAULT_VNET(curvnet)) {
1495 netisr_register(&arp_nh);
1496 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1497 arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1501 netisr_register_vnet(&arp_nh);
1504 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1509 * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1510 * lookups after destroying the hash. Ideally this would go on SI_ORDER_3.5.
1513 vnet_arp_destroy(__unused void *arg)
1516 netisr_unregister_vnet(&arp_nh);
1518 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1519 vnet_arp_destroy, NULL);