2 * Copyright (c) 1982, 1986, 1988, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)if_ether.c 8.1 (Berkeley) 6/10/93
33 * Ethernet address resolution protocol.
35 * add "inuse/lock" bit (or ref. count) along with valid bit
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
43 #include <sys/param.h>
44 #include <sys/kernel.h>
46 #include <sys/queue.h>
47 #include <sys/sysctl.h>
48 #include <sys/systm.h>
50 #include <sys/malloc.h>
52 #include <sys/rmlock.h>
53 #include <sys/socket.h>
54 #include <sys/syslog.h>
57 #include <net/if_var.h>
58 #include <net/if_dl.h>
59 #include <net/if_types.h>
60 #include <net/netisr.h>
61 #include <net/ethernet.h>
62 #include <net/route.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, CTLFLAG_RW, 0, "");
92 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp, CTLFLAG_RW, 0, "");
95 static VNET_DEFINE(int, arpt_keep) = (20*60); /* once resolved, good for 20
97 static VNET_DEFINE(int, arp_maxtries) = 5;
98 static VNET_DEFINE(int, arp_proxyall) = 0;
99 static VNET_DEFINE(int, arpt_down) = 20; /* keep incomplete entries for
101 static VNET_DEFINE(int, arpt_rexmit) = 1; /* retransmit arp entries, sec*/
102 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat); /* ARP statistics, see if_arp.h */
103 VNET_PCPUSTAT_SYSINIT(arpstat);
106 VNET_PCPUSTAT_SYSUNINIT(arpstat);
109 static VNET_DEFINE(int, arp_maxhold) = 1;
111 #define V_arpt_keep VNET(arpt_keep)
112 #define V_arpt_down VNET(arpt_down)
113 #define V_arpt_rexmit VNET(arpt_rexmit)
114 #define V_arp_maxtries VNET(arp_maxtries)
115 #define V_arp_proxyall VNET(arp_proxyall)
116 #define V_arp_maxhold VNET(arp_maxhold)
118 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
119 &VNET_NAME(arpt_keep), 0,
120 "ARP entry lifetime in seconds");
121 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
122 &VNET_NAME(arp_maxtries), 0,
123 "ARP resolution attempts before returning error");
124 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
125 &VNET_NAME(arp_proxyall), 0,
126 "Enable proxy ARP for all suitable requests");
127 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
128 &VNET_NAME(arpt_down), 0,
129 "Incomplete ARP entry lifetime in seconds");
130 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
131 arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
132 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
133 &VNET_NAME(arp_maxhold), 0,
134 "Number of packets to hold per ARP entry");
135 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
136 CTLFLAG_RW, &arp_maxpps, 0,
137 "Maximum number of remotely triggered ARP messages that can be "
138 "logged per second");
141 * Due to the exponential backoff algorithm used for the interval between GARP
142 * retransmissions, the maximum number of retransmissions is limited for
143 * sanity. This limit corresponds to a maximum interval between retransmissions
144 * of 2^16 seconds ~= 18 hours.
146 * Making this limit more dynamic is more complicated than worthwhile,
147 * especially since sending out GARPs spaced days apart would be of little
148 * use. A maximum dynamic limit would look something like:
150 * const int max = fls(INT_MAX / hz) - 1;
152 #define MAX_GARP_RETRANSMITS 16
153 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS);
154 static int garp_rexmit_count = 0; /* GARP retransmission setting. */
156 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count,
157 CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE,
158 &garp_rexmit_count, 0, sysctl_garp_rexmit, "I",
159 "Number of times to retransmit GARP packets;"
160 " 0 to disable, maximum of 16");
162 #define ARP_LOG(pri, ...) do { \
163 if (ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps)) \
164 log((pri), "arp: " __VA_ARGS__); \
168 static void arpintr(struct mbuf *);
169 static void arptimer(void *);
171 static void in_arpinput(struct mbuf *);
174 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
175 struct ifnet *ifp, int bridged, struct llentry *la);
176 static void arp_mark_lle_reachable(struct llentry *la);
177 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
179 static eventhandler_tag iflladdr_tag;
181 static const struct netisr_handler arp_nh = {
183 .nh_handler = arpintr,
184 .nh_proto = NETISR_ARP,
185 .nh_policy = NETISR_POLICY_SOURCE,
189 * Timeout routine. Age arp_tab entries periodically.
194 struct llentry *lle = (struct llentry *)arg;
198 if (lle->la_flags & LLE_STATIC) {
202 if (callout_pending(&lle->lle_timer)) {
204 * Here we are a bit odd here in the treatment of
205 * active/pending. If the pending bit is set, it got
206 * rescheduled before I ran. The active
207 * bit we ignore, since if it was stopped
208 * in ll_tablefree() and was currently running
209 * it would have return 0 so the code would
210 * not have deleted it since the callout could
211 * not be stopped so we want to go through
212 * with the delete here now. If the callout
213 * was restarted, the pending bit will be back on and
214 * we just want to bail since the callout_reset would
215 * return 1 and our reference would have been removed
216 * by arpresolve() below.
221 ifp = lle->lle_tbl->llt_ifp;
222 CURVNET_SET(ifp->if_vnet);
224 switch (lle->ln_state) {
225 case ARP_LLINFO_REACHABLE:
228 * Expiration time is approaching.
229 * Let's try to refresh entry if it is still
232 * Set r_skip_req to get feedback from
233 * fast path. Change state and re-schedule
239 lle->ln_state = ARP_LLINFO_VERIFY;
240 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
244 case ARP_LLINFO_VERIFY:
246 r_skip_req = lle->r_skip_req;
249 if (r_skip_req == 0 && lle->la_preempt > 0) {
250 /* Entry was used, issue refresh request */
252 dst = lle->r_l3addr.addr4;
254 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
256 arprequest(ifp, NULL, &dst, NULL);
260 /* Nothing happened. Reschedule if not too late */
261 if (lle->la_expire > time_uptime) {
262 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
268 case ARP_LLINFO_INCOMPLETE:
269 case ARP_LLINFO_DELETED:
273 if ((lle->la_flags & LLE_DELETED) == 0) {
276 if (lle->la_flags & LLE_VALID)
277 evt = LLENTRY_EXPIRED;
279 evt = LLENTRY_TIMEDOUT;
280 EVENTHANDLER_INVOKE(lle_event, lle, evt);
283 callout_stop(&lle->lle_timer);
285 /* XXX: LOR avoidance. We still have ref on lle. */
290 /* Guard against race with other llentry_free(). */
291 if (lle->la_flags & LLE_LINKED) {
293 lltable_unlink_entry(lle->lle_tbl, lle);
295 IF_AFDATA_UNLOCK(ifp);
297 size_t pkts_dropped = llentry_free(lle);
299 ARPSTAT_ADD(dropped, pkts_dropped);
300 ARPSTAT_INC(timeouts);
306 * Stores link-layer header for @ifp in format suitable for if_output()
307 * into buffer @buf. Resulting header length is stored in @bufsize.
309 * Returns 0 on success.
312 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
315 struct if_encap_req ereq;
318 bzero(buf, *bufsize);
319 bzero(&ereq, sizeof(ereq));
321 ereq.bufsize = *bufsize;
322 ereq.rtype = IFENCAP_LL;
323 ereq.family = AF_ARP;
324 ereq.lladdr = ar_tha(ah);
325 ereq.hdata = (u_char *)ah;
327 ereq.flags = IFENCAP_FLAG_BROADCAST;
328 error = ifp->if_requestencap(ifp, &ereq);
330 *bufsize = ereq.bufsize;
337 * Broadcast an ARP request. Caller specifies:
338 * - arp header source ip address
339 * - arp header target ip address
340 * - arp header source ethernet address
343 arprequest(struct ifnet *ifp, const struct in_addr *sip,
344 const struct in_addr *tip, u_char *enaddr)
349 u_char *carpaddr = NULL;
350 uint8_t linkhdr[LLE_MAX_LINKHDR];
357 * The caller did not supply a source address, try to find
358 * a compatible one among those assigned to this interface.
363 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
364 if (ifa->ifa_addr->sa_family != AF_INET)
368 if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
370 sip = &IA_SIN(ifa)->sin_addr;
373 sip = &IA_SIN(ifa)->sin_addr;
376 if (0 == ((sip->s_addr ^ tip->s_addr) &
377 IA_MASKSIN(ifa)->sin_addr.s_addr))
378 break; /* found it. */
380 IF_ADDR_RUNLOCK(ifp);
382 printf("%s: cannot find matching address\n", __func__);
387 enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
389 if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
391 m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
393 m->m_pkthdr.len = m->m_len;
394 M_ALIGN(m, m->m_len);
395 ah = mtod(m, struct arphdr *);
396 bzero((caddr_t)ah, m->m_len);
398 mac_netinet_arp_send(ifp, m);
400 ah->ar_pro = htons(ETHERTYPE_IP);
401 ah->ar_hln = ifp->if_addrlen; /* hardware address length */
402 ah->ar_pln = sizeof(struct in_addr); /* protocol address length */
403 ah->ar_op = htons(ARPOP_REQUEST);
404 bcopy(enaddr, ar_sha(ah), ah->ar_hln);
405 bcopy(sip, ar_spa(ah), ah->ar_pln);
406 bcopy(tip, ar_tpa(ah), ah->ar_pln);
407 sa.sa_family = AF_ARP;
410 /* Calculate link header for sending frame */
411 bzero(&ro, sizeof(ro));
412 linkhdrsize = sizeof(linkhdr);
413 error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
414 if (error != 0 && error != EAFNOSUPPORT) {
415 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
416 if_name(ifp), error);
420 ro.ro_prepend = linkhdr;
421 ro.ro_plen = linkhdrsize;
424 m->m_flags |= M_BCAST;
425 m_clrprotoflags(m); /* Avoid confusing lower layers. */
426 (*ifp->if_output)(ifp, m, &sa, &ro);
427 ARPSTAT_INC(txrequests);
432 * Resolve an IP address into an ethernet address - heavy version.
433 * Used internally by arpresolve().
434 * We have already checked than we can't use existing lle without
435 * modification so we have to acquire LLE_EXCLUSIVE lle lock.
437 * On success, desten and flags are filled in and the function returns 0;
438 * If the packet must be held pending resolution, we return EWOULDBLOCK
439 * On other errors, we return the corresponding error code.
440 * Note that m_freem() handles NULL.
443 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
444 const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
445 struct llentry **plle)
447 struct llentry *la = NULL, *la_tmp;
448 struct mbuf *curr = NULL;
449 struct mbuf *next = NULL;
459 if ((flags & LLE_CREATE) == 0) {
460 IF_AFDATA_RLOCK(ifp);
461 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
462 IF_AFDATA_RUNLOCK(ifp);
464 if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
465 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
467 char addrbuf[INET_ADDRSTRLEN];
470 "arpresolve: can't allocate llinfo for %s on %s\n",
471 inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
477 IF_AFDATA_WLOCK(ifp);
479 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
480 /* Prefer ANY existing lle over newly-created one */
482 lltable_link_entry(LLTABLE(ifp), la);
483 IF_AFDATA_WUNLOCK(ifp);
484 if (la_tmp != NULL) {
485 lltable_free_entry(LLTABLE(ifp), la);
494 if ((la->la_flags & LLE_VALID) &&
495 ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
496 if (flags & LLE_ADDRONLY) {
497 lladdr = la->ll_addr;
498 ll_len = ifp->if_addrlen;
500 lladdr = la->r_linkdata;
501 ll_len = la->r_hdrlen;
503 bcopy(lladdr, desten, ll_len);
505 /* Check if we have feedback request from arptimer() */
506 if (la->r_skip_req != 0) {
508 la->r_skip_req = 0; /* Notify that entry was used */
512 *pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
521 renew = (la->la_asked == 0 || la->la_expire != time_uptime);
523 * There is an arptab entry, but no ethernet address
524 * response yet. Add the mbuf to the list, dropping
525 * the oldest packet if we have exceeded the system
529 if (la->la_numheld >= V_arp_maxhold) {
530 if (la->la_hold != NULL) {
531 next = la->la_hold->m_nextpkt;
532 m_freem(la->la_hold);
535 ARPSTAT_INC(dropped);
538 if (la->la_hold != NULL) {
540 while (curr->m_nextpkt != NULL)
541 curr = curr->m_nextpkt;
548 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
549 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
550 * if we have already sent arp_maxtries ARP requests. Retransmit the
551 * ARP request, but not faster than one request per second.
553 if (la->la_asked < V_arp_maxtries)
554 error = EWOULDBLOCK; /* First request. */
556 error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
562 la->la_expire = time_uptime;
563 canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
569 arprequest(ifp, NULL, &SIN(dst)->sin_addr, NULL);
578 * Resolve an IP address into an ethernet address.
581 arpresolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
582 char *desten, uint32_t *pflags, struct llentry **plle)
586 flags |= LLE_ADDRONLY;
587 error = arpresolve_full(ifp, 0, flags, NULL, dst, desten, pflags, plle);
593 * Lookups link header based on an IP address.
595 * ifp is the interface we use
596 * is_gw != 0 if @dst represents gateway to some destination
597 * m is the mbuf. May be NULL if we don't have a packet.
598 * dst is the next hop,
599 * desten is the storage to put LL header.
600 * flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
602 * On success, full/partial link header and flags are filled in and
603 * the function returns 0.
604 * If the packet must be held pending resolution, we return EWOULDBLOCK
605 * On other errors, we return the corresponding error code.
606 * Note that m_freem() handles NULL.
609 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
610 const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
611 struct llentry **plle)
613 struct llentry *la = NULL;
621 if (m->m_flags & M_BCAST) {
624 ifp->if_broadcastaddr, ifp->if_addrlen);
627 if (m->m_flags & M_MCAST) {
629 ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
634 IF_AFDATA_RLOCK(ifp);
635 la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
636 if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
637 /* Entry found, let's copy lle info */
638 bcopy(la->r_linkdata, desten, la->r_hdrlen);
640 *pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
641 /* Check if we have feedback request from arptimer() */
642 if (la->r_skip_req != 0) {
644 la->r_skip_req = 0; /* Notify that entry was used */
652 IF_AFDATA_RUNLOCK(ifp);
657 IF_AFDATA_RUNLOCK(ifp);
659 return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
660 desten, pflags, plle));
664 * Common length and type checks are done here,
665 * then the protocol-specific routine is called.
668 arpintr(struct mbuf *m)
675 ifp = m->m_pkthdr.rcvif;
677 if (m->m_len < sizeof(struct arphdr) &&
678 ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
679 ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
683 ar = mtod(m, struct arphdr *);
685 /* Check if length is sufficient */
686 if (m->m_len < arphdr_len(ar)) {
687 m = m_pullup(m, arphdr_len(ar));
689 ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
693 ar = mtod(m, struct arphdr *);
698 switch (ntohs(ar->ar_hrd)) {
700 hlen = ETHER_ADDR_LEN; /* RFC 826 */
704 hlen = 6; /* RFC 1390, FDDI_ADDR_LEN */
708 hlen = 1; /* RFC 1201, ARC_ADDR_LEN */
711 case ARPHRD_INFINIBAND:
712 hlen = 20; /* RFC 4391, INFINIBAND_ALEN */
713 layer = "infiniband";
715 case ARPHRD_IEEE1394:
716 hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
720 * Restrict too long hardware addresses.
721 * Currently we are capable of handling 20-byte
722 * addresses ( sizeof(lle->ll_addr) )
724 if (ar->ar_hln >= 20)
729 "packet with unknown hardware format 0x%02d received on "
730 "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
735 if (hlen != 0 && hlen != ar->ar_hln) {
737 "packet with invalid %s address length %d received on %s\n",
738 layer, ar->ar_hln, if_name(ifp));
743 ARPSTAT_INC(received);
744 switch (ntohs(ar->ar_pro)) {
756 * ARP for Internet protocols on 10 Mb/s Ethernet.
757 * Algorithm is that given in RFC 826.
758 * In addition, a sanity check is performed on the sender
759 * protocol address, to catch impersonators.
760 * We no longer handle negotiations for use of trailer protocol:
761 * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
762 * along with IP replies if we wanted trailers sent to us,
763 * and also sent them in response to IP replies.
764 * This allowed either end to announce the desire to receive
766 * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
767 * but formerly didn't normally send requests.
769 static int log_arp_wrong_iface = 1;
770 static int log_arp_movements = 1;
771 static int log_arp_permanent_modify = 1;
772 static int allow_multicast = 0;
774 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
775 &log_arp_wrong_iface, 0,
776 "log arp packets arriving on the wrong interface");
777 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
778 &log_arp_movements, 0,
779 "log arp replies from MACs different than the one in the cache");
780 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
781 &log_arp_permanent_modify, 0,
782 "log arp replies from MACs different than the one in the permanent arp entry");
783 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
784 &allow_multicast, 0, "accept multicast addresses");
787 in_arpinput(struct mbuf *m)
789 struct rm_priotracker in_ifa_tracker;
791 struct ifnet *ifp = m->m_pkthdr.rcvif;
792 struct llentry *la = NULL, *la_tmp;
794 struct in_ifaddr *ia;
796 struct in_addr isaddr, itaddr, myaddr;
797 u_int8_t *enaddr = NULL;
799 int bridged = 0, is_bridge = 0;
801 struct sockaddr_in sin;
802 struct sockaddr *dst;
803 struct nhop4_basic nh4;
804 uint8_t linkhdr[LLE_MAX_LINKHDR];
809 char addrbuf[INET_ADDRSTRLEN];
811 sin.sin_len = sizeof(struct sockaddr_in);
812 sin.sin_family = AF_INET;
813 sin.sin_addr.s_addr = 0;
817 if (ifp->if_type == IFT_BRIDGE)
821 * We already have checked that mbuf contains enough contiguous data
822 * to hold entire arp message according to the arp header.
824 ah = mtod(m, struct arphdr *);
827 * ARP is only for IPv4 so we can reject packets with
828 * a protocol length not equal to an IPv4 address.
830 if (ah->ar_pln != sizeof(struct in_addr)) {
831 ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
832 sizeof(struct in_addr));
836 if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
837 ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
838 ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
842 op = ntohs(ah->ar_op);
843 (void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
844 (void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
846 if (op == ARPOP_REPLY)
847 ARPSTAT_INC(rxreplies);
850 * For a bridge, we want to check the address irrespective
851 * of the receive interface. (This will change slightly
852 * when we have clusters of interfaces).
854 IN_IFADDR_RLOCK(&in_ifa_tracker);
855 LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
856 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
857 ia->ia_ifp == ifp) &&
858 itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
859 (ia->ia_ifa.ifa_carp == NULL ||
860 (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
861 ifa_ref(&ia->ia_ifa);
862 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
866 LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
867 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
868 ia->ia_ifp == ifp) &&
869 isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
870 ifa_ref(&ia->ia_ifa);
871 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
875 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia) \
876 (ia->ia_ifp->if_bridge == ifp->if_softc && \
877 !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) && \
878 addr == ia->ia_addr.sin_addr.s_addr)
880 * Check the case when bridge shares its MAC address with
881 * some of its children, so packets are claimed by bridge
882 * itself (bridge_input() does it first), but they are really
883 * meant to be destined to the bridge member.
886 LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
887 if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
888 ifa_ref(&ia->ia_ifa);
890 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
895 #undef BDG_MEMBER_MATCHES_ARP
896 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
899 * No match, use the first inet address on the receive interface
900 * as a dummy address for the rest of the function.
903 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
904 if (ifa->ifa_addr->sa_family == AF_INET &&
905 (ifa->ifa_carp == NULL ||
906 (*carp_iamatch_p)(ifa, &enaddr))) {
909 IF_ADDR_RUNLOCK(ifp);
912 IF_ADDR_RUNLOCK(ifp);
915 * If bridging, fall back to using any inet address.
917 IN_IFADDR_RLOCK(&in_ifa_tracker);
918 if (!bridged || (ia = TAILQ_FIRST(&V_in_ifaddrhead)) == NULL) {
919 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
922 ifa_ref(&ia->ia_ifa);
923 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
926 enaddr = (u_int8_t *)IF_LLADDR(ifp);
927 carped = (ia->ia_ifa.ifa_carp != NULL);
928 myaddr = ia->ia_addr.sin_addr;
929 ifa_free(&ia->ia_ifa);
930 if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
931 goto drop; /* it's from me, ignore it. */
932 if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
933 ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
934 "%s!\n", inet_ntoa_r(isaddr, addrbuf));
938 if (ifp->if_addrlen != ah->ar_hln) {
939 ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
940 "i/f %d (ignored)\n", ifp->if_addrlen,
941 (u_char *) ar_sha(ah), ":", ah->ar_hln,
947 * Warn if another host is using the same IP address, but only if the
948 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
949 * case we suppress the warning to avoid false positive complaints of
950 * potential misconfiguration.
952 if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
953 myaddr.s_addr != 0) {
954 ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
955 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
956 inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
961 if (ifp->if_flags & IFF_STATICARP)
964 bzero(&sin, sizeof(sin));
965 sin.sin_len = sizeof(struct sockaddr_in);
966 sin.sin_family = AF_INET;
967 sin.sin_addr = isaddr;
968 dst = (struct sockaddr *)&sin;
969 IF_AFDATA_RLOCK(ifp);
970 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
971 IF_AFDATA_RUNLOCK(ifp);
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);
989 * lle creation may fail if source address belongs
990 * to non-directly connected subnet. However, we
991 * will try to answer the request instead of dropping
996 lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
999 IF_AFDATA_WLOCK(ifp);
1001 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1004 * Check if lle still does not exists.
1005 * If it does, that means that we either
1006 * 1) have configured it explicitly, via
1007 * 1a) 'arp -s' static entry or
1008 * 1b) interface address static record
1010 * 2) it was the result of sending first packet to-host
1012 * 3) it was another arp reply packet we handled in
1015 * In all cases except 3) we definitely need to prefer
1016 * existing lle. For the sake of simplicity, prefer any
1017 * existing lle over newly-create one.
1020 lltable_link_entry(LLTABLE(ifp), la);
1021 IF_AFDATA_WUNLOCK(ifp);
1023 if (la_tmp == NULL) {
1024 arp_mark_lle_reachable(la);
1027 /* Free newly-create entry and handle packet */
1028 lltable_free_entry(LLTABLE(ifp), la);
1031 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1032 /* arp_check_update_lle() returns @la unlocked */
1037 if (op != ARPOP_REQUEST)
1039 ARPSTAT_INC(rxrequests);
1041 if (itaddr.s_addr == myaddr.s_addr) {
1042 /* Shortcut.. the receiving interface is the target. */
1043 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1044 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1046 struct llentry *lle = NULL;
1048 sin.sin_addr = itaddr;
1049 IF_AFDATA_RLOCK(ifp);
1050 lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1051 IF_AFDATA_RUNLOCK(ifp);
1053 if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1054 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1055 (void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1062 if (!V_arp_proxyall)
1065 /* XXX MRT use table 0 for arp reply */
1066 if (fib4_lookup_nh_basic(0, itaddr, 0, 0, &nh4) != 0)
1070 * Don't send proxies for nodes on the same interface
1071 * as this one came out of, or we'll get into a fight
1072 * over who claims what Ether address.
1074 if (nh4.nh_ifp == ifp)
1077 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1078 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1081 * Also check that the node which sent the ARP packet
1082 * is on the interface we expect it to be on. This
1083 * avoids ARP chaos if an interface is connected to the
1087 /* XXX MRT use table 0 for arp checks */
1088 if (fib4_lookup_nh_basic(0, isaddr, 0, 0, &nh4) != 0)
1090 if (nh4.nh_ifp != ifp) {
1091 ARP_LOG(LOG_INFO, "proxy: ignoring request"
1092 " from %s via %s\n",
1093 inet_ntoa_r(isaddr, addrbuf),
1099 printf("arp: proxying for %s\n",
1100 inet_ntoa_r(itaddr, addrbuf));
1105 if (itaddr.s_addr == myaddr.s_addr &&
1106 IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1107 /* RFC 3927 link-local IPv4; always reply by broadcast. */
1108 #ifdef DEBUG_LINKLOCAL
1109 printf("arp: sending reply for link-local addr %s\n",
1110 inet_ntoa_r(itaddr, addrbuf));
1112 m->m_flags |= M_BCAST;
1113 m->m_flags &= ~M_MCAST;
1115 /* default behaviour; never reply by broadcast. */
1116 m->m_flags &= ~(M_BCAST|M_MCAST);
1118 (void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1119 (void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1120 ah->ar_op = htons(ARPOP_REPLY);
1121 ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1122 m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1123 m->m_pkthdr.len = m->m_len;
1124 m->m_pkthdr.rcvif = NULL;
1125 sa.sa_family = AF_ARP;
1128 /* Calculate link header for sending frame */
1129 bzero(&ro, sizeof(ro));
1130 linkhdrsize = sizeof(linkhdr);
1131 error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1134 * arp_fillheader() may fail due to lack of support inside encap request
1135 * routing. This is not necessary an error, AF_ARP can/should be handled
1138 if (error != 0 && error != EAFNOSUPPORT) {
1139 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1140 if_name(ifp), error);
1144 ro.ro_prepend = linkhdr;
1145 ro.ro_plen = linkhdrsize;
1148 m_clrprotoflags(m); /* Avoid confusing lower layers. */
1149 (*ifp->if_output)(ifp, m, &sa, &ro);
1150 ARPSTAT_INC(txreplies);
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)
1167 struct mbuf *m_hold, *m_hold_next;
1168 uint8_t linkhdr[LLE_MAX_LINKHDR];
1171 char addrbuf[INET_ADDRSTRLEN];
1173 LLE_WLOCK_ASSERT(la);
1175 /* the following is not an error when doing bridging */
1176 if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1177 if (log_arp_wrong_iface)
1178 ARP_LOG(LOG_WARNING, "%s is on %s "
1179 "but got reply from %*D on %s\n",
1180 inet_ntoa_r(isaddr, addrbuf),
1181 la->lle_tbl->llt_ifp->if_xname,
1182 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1187 if ((la->la_flags & LLE_VALID) &&
1188 bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1189 if (la->la_flags & LLE_STATIC) {
1191 if (log_arp_permanent_modify)
1193 "%*D attempts to modify "
1194 "permanent entry for %s on %s\n",
1196 (u_char *)ar_sha(ah), ":",
1197 inet_ntoa_r(isaddr, addrbuf),
1201 if (log_arp_movements) {
1202 ARP_LOG(LOG_INFO, "%s moved from %*D "
1204 inet_ntoa_r(isaddr, addrbuf),
1206 (u_char *)la->ll_addr, ":",
1207 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1212 /* Calculate full link prepend to use in lle */
1213 linkhdrsize = sizeof(linkhdr);
1214 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1215 &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,
1226 /* Clear fast path feedback request if set */
1230 arp_mark_lle_reachable(la);
1233 * The packets are all freed within the call to the output
1236 * NB: The lock MUST be released before the call to the
1239 if (la->la_hold != NULL) {
1240 m_hold = la->la_hold;
1243 lltable_fill_sa_entry(la, &sa);
1245 for (; m_hold != NULL; m_hold = m_hold_next) {
1246 m_hold_next = m_hold->m_nextpkt;
1247 m_hold->m_nextpkt = NULL;
1248 /* Avoid confusing lower layers. */
1249 m_clrprotoflags(m_hold);
1250 (*ifp->if_output)(ifp, m_hold, &sa, NULL);
1257 arp_mark_lle_reachable(struct llentry *la)
1259 int canceled, wtime;
1261 LLE_WLOCK_ASSERT(la);
1263 la->ln_state = ARP_LLINFO_REACHABLE;
1264 EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1266 if (!(la->la_flags & LLE_STATIC)) {
1268 la->la_expire = time_uptime + V_arpt_keep;
1269 wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1271 wtime = V_arpt_keep;
1272 canceled = callout_reset(&la->lle_timer,
1273 hz * wtime, arptimer, la);
1278 la->la_preempt = V_arp_maxtries;
1282 * Add permanent link-layer record for given interface address.
1284 static __noinline void
1285 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1287 struct llentry *lle, *lle_tmp;
1290 * Interface address LLE record is considered static
1291 * because kernel code relies on LLE_STATIC flag to check
1292 * if these entries can be rewriten by arp updates.
1294 lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1296 log(LOG_INFO, "arp_ifinit: cannot create arp "
1297 "entry for interface address\n");
1301 IF_AFDATA_WLOCK(ifp);
1303 /* Unlink any entry if exists */
1304 lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1305 if (lle_tmp != NULL)
1306 lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1308 lltable_link_entry(LLTABLE(ifp), lle);
1309 IF_AFDATA_WUNLOCK(ifp);
1311 if (lle_tmp != NULL)
1312 EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1314 EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1316 if (lle_tmp != NULL)
1317 lltable_free_entry(LLTABLE(ifp), lle_tmp);
1321 * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1325 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1328 int rexmit_count = *(int *)arg1;
1330 error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1332 /* Enforce limits on any new value that may have been set. */
1333 if (!error && req->newptr) {
1334 /* A new value was set. */
1335 if (rexmit_count < 0) {
1337 } else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1338 rexmit_count = MAX_GARP_RETRANSMITS;
1340 *(int *)arg1 = rexmit_count;
1347 * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1348 * retransmit it again. A pending callout owns a reference to the ifa.
1351 garp_rexmit(void *arg)
1353 struct in_ifaddr *ia = arg;
1355 if (callout_pending(&ia->ia_garp_timer) ||
1356 !callout_active(&ia->ia_garp_timer)) {
1357 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1358 ifa_free(&ia->ia_ifa);
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);
1392 * Start the GARP retransmit timer.
1394 * A single GARP is always transmitted when an IPv4 address is added
1395 * to an interface and that is usually sufficient. However, in some
1396 * circumstances, such as when a shared address is passed between
1397 * cluster nodes, this single GARP may occasionally be dropped or
1398 * lost. This can lead to neighbors on the network link working with a
1399 * stale ARP cache and sending packets destined for that address to
1400 * the node that previously owned the address, which may not respond.
1402 * To avoid this situation, GARP retransmits can be enabled by setting
1403 * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1404 * than zero. The setting represents the maximum number of
1405 * retransmissions. The interval between retransmissions is calculated
1406 * using an exponential backoff algorithm, doubling each time, so the
1407 * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1410 garp_timer_start(struct ifaddr *ifa)
1412 struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1414 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1415 ia->ia_garp_count = 0;
1416 if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1417 garp_rexmit, ia) == 0) {
1420 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1424 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1426 const struct sockaddr_in *dst_in;
1427 const struct sockaddr *dst;
1429 if (ifa->ifa_carp != NULL)
1432 dst = ifa->ifa_addr;
1433 dst_in = (const struct sockaddr_in *)dst;
1435 if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1437 arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1438 if (garp_rexmit_count > 0) {
1439 garp_timer_start(ifa);
1442 arp_add_ifa_lle(ifp, dst);
1446 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1449 if (ntohl(addr.s_addr) != INADDR_ANY)
1450 arprequest(ifp, &addr, &addr, enaddr);
1454 * Sends gratuitous ARPs for each ifaddr to notify other
1455 * nodes about the address change.
1457 static __noinline void
1458 arp_handle_ifllchange(struct ifnet *ifp)
1462 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1463 if (ifa->ifa_addr->sa_family == AF_INET)
1464 arp_ifinit(ifp, ifa);
1469 * A handler for interface link layer address change event.
1472 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1475 lltable_update_ifaddr(LLTABLE(ifp));
1477 if ((ifp->if_flags & IFF_UP) != 0)
1478 arp_handle_ifllchange(ifp);
1485 if (IS_DEFAULT_VNET(curvnet)) {
1486 netisr_register(&arp_nh);
1487 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1488 arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1492 netisr_register_vnet(&arp_nh);
1495 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1500 * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1501 * lookups after destroying the hash. Ideally this would go on SI_ORDER_3.5.
1504 vnet_arp_destroy(__unused void *arg)
1507 netisr_unregister_vnet(&arp_nh);
1509 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1510 vnet_arp_destroy, NULL);