1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1988, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the 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.
18 *
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
29 * SUCH DAMAGE.
30 */
31
32 /*
33 * Ethernet address resolution protocol.
34 * TODO:
35 * add "inuse/lock" bit (or ref. count) along with valid bit
36 */
37
38 #include <sys/cdefs.h>
39 #include "opt_inet.h"
40
41 #include <sys/param.h>
42 #include <sys/eventhandler.h>
43 #include <sys/kernel.h>
44 #include <sys/lock.h>
45 #include <sys/queue.h>
46 #include <sys/sysctl.h>
47 #include <sys/systm.h>
48 #include <sys/mbuf.h>
49 #include <sys/malloc.h>
50 #include <sys/proc.h>
51 #include <sys/socket.h>
52 #include <sys/syslog.h>
53
54 #include <net/if.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>
63 #include <net/vnet.h>
64
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>
70 #ifdef INET
71 #include <netinet/ip_carp.h>
72 #endif
73
74 #include <security/mac/mac_framework.h>
75
76 #define SIN(s) ((const struct sockaddr_in *)(s))
77
78 static struct timeval arp_lastlog;
79 static int arp_curpps;
80 static int arp_maxpps = 1;
81
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 */
88 };
89
90 SYSCTL_DECL(_net_link_ether);
91 static SYSCTL_NODE(_net_link_ether, PF_INET, inet,
92 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
93 "");
94 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp,
95 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
96 "");
97
98 /* timer values */
99 VNET_DEFINE_STATIC(int, arpt_keep) = (20*60); /* once resolved, good for 20
100 * minutes */
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
104 * 20 seconds */
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);
108
109 #ifdef VIMAGE
110 VNET_PCPUSTAT_SYSUNINIT(arpstat);
111 #endif /* VIMAGE */
112
113 VNET_DEFINE_STATIC(int, arp_maxhold) = 16;
114
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)
121
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");
143
144 /*
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.
149 *
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:
153 *
154 * const int max = fls(INT_MAX / hz) - 1;
155 */
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. */
159
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");
165
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__); \
176 } while (0)
177
178 static void arpintr(struct mbuf *);
179 static void arptimer(void *);
180 #ifdef INET
181 static void in_arpinput(struct mbuf *);
182 #endif
183
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);
188
189 static eventhandler_tag iflladdr_tag;
190
191 static const struct netisr_handler arp_nh = {
192 .nh_name = "arp",
193 .nh_handler = arpintr,
194 .nh_proto = NETISR_ARP,
195 .nh_policy = NETISR_POLICY_SOURCE,
196 };
197
198 /*
199 * Timeout routine. Age arp_tab entries periodically.
200 */
201 static void
arptimer(void * arg)202 arptimer(void *arg)
203 {
204 struct llentry *lle = (struct llentry *)arg;
205 struct ifnet *ifp;
206
207 if (lle->la_flags & LLE_STATIC) {
208 return;
209 }
210 LLE_WLOCK(lle);
211 if (callout_pending(&lle->lle_timer)) {
212 /*
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.
226 */
227 LLE_WUNLOCK(lle);
228 return;
229 }
230 ifp = lle->lle_tbl->llt_ifp;
231 CURVNET_SET(ifp->if_vnet);
232
233 switch (lle->ln_state) {
234 case ARP_LLINFO_REACHABLE:
235
236 /*
237 * Expiration time is approaching.
238 * Request usage feedback from the datapath.
239 * Change state and re-schedule ourselves.
240 */
241 llentry_request_feedback(lle);
242 lle->ln_state = ARP_LLINFO_VERIFY;
243 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
244 LLE_WUNLOCK(lle);
245 CURVNET_RESTORE();
246 return;
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;
251 struct in_addr dst;
252
253 dst = lle->r_l3addr.addr4;
254 lle->la_preempt--;
255 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
256 LLE_WUNLOCK(lle);
257 NET_EPOCH_ENTER(et);
258 arprequest(ifp, NULL, &dst, NULL);
259 NET_EPOCH_EXIT(et);
260 CURVNET_RESTORE();
261 return;
262 }
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);
266 LLE_WUNLOCK(lle);
267 CURVNET_RESTORE();
268 return;
269 }
270 break;
271 case ARP_LLINFO_INCOMPLETE:
272 case ARP_LLINFO_DELETED:
273 break;
274 }
275
276 if ((lle->la_flags & LLE_DELETED) == 0) {
277 int evt;
278
279 if (lle->la_flags & LLE_VALID)
280 evt = LLENTRY_EXPIRED;
281 else
282 evt = LLENTRY_TIMEDOUT;
283 EVENTHANDLER_INVOKE(lle_event, lle, evt);
284 }
285
286 callout_stop(&lle->lle_timer);
287
288 /* XXX: LOR avoidance. We still have ref on lle. */
289 LLE_WUNLOCK(lle);
290 IF_AFDATA_LOCK(ifp);
291 LLE_WLOCK(lle);
292
293 /* Guard against race with other llentry_free(). */
294 if (lle->la_flags & LLE_LINKED) {
295 LLE_REMREF(lle);
296 lltable_unlink_entry(lle->lle_tbl, lle);
297 }
298 IF_AFDATA_UNLOCK(ifp);
299
300 size_t pkts_dropped = llentry_free(lle);
301
302 ARPSTAT_ADD(dropped, pkts_dropped);
303 ARPSTAT_INC(timeouts);
304
305 CURVNET_RESTORE();
306 }
307
308 /*
309 * Stores link-layer header for @ifp in format suitable for if_output()
310 * into buffer @buf. Resulting header length is stored in @bufsize.
311 *
312 * Returns 0 on success.
313 */
314 static int
arp_fillheader(struct ifnet * ifp,struct arphdr * ah,int bcast,u_char * buf,size_t * bufsize)315 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
316 size_t *bufsize)
317 {
318 struct if_encap_req ereq;
319 int error;
320
321 bzero(buf, *bufsize);
322 bzero(&ereq, sizeof(ereq));
323 ereq.buf = buf;
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;
329 if (bcast)
330 ereq.flags = IFENCAP_FLAG_BROADCAST;
331 error = ifp->if_requestencap(ifp, &ereq);
332 if (error == 0)
333 *bufsize = ereq.bufsize;
334
335 return (error);
336 }
337
338 /*
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
343 */
344 static int
arprequest_internal(struct ifnet * ifp,const struct in_addr * sip,const struct in_addr * tip,u_char * enaddr)345 arprequest_internal(struct ifnet *ifp, const struct in_addr *sip,
346 const struct in_addr *tip, u_char *enaddr)
347 {
348 struct mbuf *m;
349 struct arphdr *ah;
350 struct sockaddr sa;
351 u_char *carpaddr = NULL;
352 uint8_t linkhdr[LLE_MAX_LINKHDR];
353 size_t linkhdrsize;
354 struct route ro;
355 int error;
356
357 NET_EPOCH_ASSERT();
358
359 if (sip == NULL) {
360 /*
361 * The caller did not supply a source address, try to find
362 * a compatible one among those assigned to this interface.
363 */
364 struct ifaddr *ifa;
365
366 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
367 if (ifa->ifa_addr->sa_family != AF_INET)
368 continue;
369
370 if (ifa->ifa_carp) {
371 if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
372 continue;
373 sip = &IA_SIN(ifa)->sin_addr;
374 } else {
375 carpaddr = NULL;
376 sip = &IA_SIN(ifa)->sin_addr;
377 }
378
379 if (0 == ((sip->s_addr ^ tip->s_addr) &
380 IA_MASKSIN(ifa)->sin_addr.s_addr))
381 break; /* found it. */
382 }
383 if (sip == NULL) {
384 printf("%s: cannot find matching address\n", __func__);
385 return (EADDRNOTAVAIL);
386 }
387 }
388 if (enaddr == NULL)
389 enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
390
391 if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
392 return (ENOMEM);
393 m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
394 2 * ifp->if_addrlen;
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);
399 #ifdef MAC
400 mac_netinet_arp_send(ifp, m);
401 #endif
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;
410 sa.sa_len = 2;
411
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) {
417 m_freem(m);
418 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
419 if_name(ifp), error);
420 return (error);
421 }
422
423 ro.ro_prepend = linkhdr;
424 ro.ro_plen = linkhdrsize;
425 ro.ro_flags = 0;
426
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);
431 if (error) {
432 ARPSTAT_INC(txerrors);
433 ARP_LOG(LOG_DEBUG, "Failed to send ARP packet on %s: %d\n",
434 if_name(ifp), error);
435 }
436 return (error);
437 }
438
439 void
arprequest(struct ifnet * ifp,const struct in_addr * sip,const struct in_addr * tip,u_char * enaddr)440 arprequest(struct ifnet *ifp, const struct in_addr *sip,
441 const struct in_addr *tip, u_char *enaddr)
442 {
443
444 (void) arprequest_internal(ifp, sip, tip, enaddr);
445 }
446
447 /*
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.
452 *
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.
457 */
458 static int
arpresolve_full(struct ifnet * ifp,int is_gw,int flags,struct mbuf * m,const struct sockaddr * dst,u_char * desten,uint32_t * pflags,struct llentry ** plle)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)
462 {
463 struct llentry *la = NULL, *la_tmp;
464 int error, renew;
465 char *lladdr;
466 int ll_len;
467
468 NET_EPOCH_ASSERT();
469
470 if (pflags != NULL)
471 *pflags = 0;
472 if (plle != NULL)
473 *plle = NULL;
474
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);
479 if (la == NULL) {
480 char addrbuf[INET_ADDRSTRLEN];
481
482 log(LOG_DEBUG,
483 "arpresolve: can't allocate llinfo for %s on %s\n",
484 inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
485 if_name(ifp));
486 m_freem(m);
487 return (EINVAL);
488 }
489
490 IF_AFDATA_WLOCK(ifp);
491 LLE_WLOCK(la);
492 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
493 /* Prefer ANY existing lle over newly-created one */
494 if (la_tmp == NULL)
495 lltable_link_entry(LLTABLE(ifp), la);
496 IF_AFDATA_WUNLOCK(ifp);
497 if (la_tmp != NULL) {
498 lltable_free_entry(LLTABLE(ifp), la);
499 la = la_tmp;
500 }
501 }
502 if (la == NULL) {
503 m_freem(m);
504 return (EINVAL);
505 }
506
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;
512 } else {
513 lladdr = la->r_linkdata;
514 ll_len = la->r_hdrlen;
515 }
516 bcopy(lladdr, desten, ll_len);
517
518 /* Notify LLE code that the entry was used by datapath */
519 llentry_provide_feedback(la);
520 if (pflags != NULL)
521 *pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
522 if (plle) {
523 LLE_ADDREF(la);
524 *plle = la;
525 }
526 LLE_WUNLOCK(la);
527 return (0);
528 }
529
530 renew = (la->la_asked == 0 || la->la_expire != time_uptime);
531
532 /*
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
536 * setting.
537 */
538 if (m != NULL) {
539 size_t dropped = lltable_append_entry_queue(la, m, V_arp_maxhold);
540 ARPSTAT_ADD(dropped, dropped);
541 }
542
543 /*
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.
548 */
549 if (la->la_asked < V_arp_maxtries)
550 error = EWOULDBLOCK; /* First request. */
551 else
552 error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
553
554 if (renew) {
555 int canceled, e;
556
557 LLE_ADDREF(la);
558 la->la_expire = time_uptime;
559 canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
560 arptimer, la);
561 if (canceled)
562 LLE_REMREF(la);
563 la->la_asked++;
564 LLE_WUNLOCK(la);
565 e = arprequest_internal(ifp, NULL, &SIN(dst)->sin_addr, NULL);
566 /*
567 * Only overwrite 'error' in case of error; in case of success
568 * the proper return value was already set above.
569 */
570 if (e != 0)
571 return (e);
572 return (error);
573 }
574
575 LLE_WUNLOCK(la);
576 return (error);
577 }
578
579 /*
580 * Lookups link header based on an IP address.
581 * On input:
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
588 *
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.
594 */
595 int
arpresolve(struct ifnet * ifp,int is_gw,struct mbuf * m,const struct sockaddr * dst,u_char * desten,uint32_t * pflags,struct llentry ** plle)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)
599 {
600 struct llentry *la = NULL;
601
602 NET_EPOCH_ASSERT();
603
604 if (pflags != NULL)
605 *pflags = 0;
606 if (plle != NULL)
607 *plle = NULL;
608
609 if (m != NULL) {
610 if (m->m_flags & M_BCAST) {
611 /* broadcast */
612 (void)memcpy(desten,
613 ifp->if_broadcastaddr, ifp->if_addrlen);
614 return (0);
615 }
616 if (m->m_flags & M_MCAST) {
617 /* multicast */
618 ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
619 return (0);
620 }
621 }
622
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);
627 if (pflags != NULL)
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);
631 if (plle) {
632 LLE_ADDREF(la);
633 *plle = la;
634 LLE_WUNLOCK(la);
635 }
636 return (0);
637 }
638 if (plle && la)
639 LLE_WUNLOCK(la);
640
641 return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
642 desten, pflags, plle));
643 }
644
645 /*
646 * Common length and type checks are done here,
647 * then the protocol-specific routine is called.
648 */
649 static void
arpintr(struct mbuf * m)650 arpintr(struct mbuf *m)
651 {
652 struct arphdr *ar;
653 struct ifnet *ifp;
654 char *layer;
655 int hlen;
656
657 ifp = m->m_pkthdr.rcvif;
658
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",
662 if_name(ifp));
663 return;
664 }
665 ar = mtod(m, struct arphdr *);
666
667 /* Check if length is sufficient */
668 if (m->m_len < arphdr_len(ar)) {
669 m = m_pullup(m, arphdr_len(ar));
670 if (m == NULL) {
671 ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
672 if_name(ifp));
673 return;
674 }
675 ar = mtod(m, struct arphdr *);
676 }
677
678 hlen = 0;
679 layer = "";
680 switch (ntohs(ar->ar_hrd)) {
681 case ARPHRD_ETHER:
682 hlen = ETHER_ADDR_LEN; /* RFC 826 */
683 layer = "ethernet";
684 break;
685 case ARPHRD_IEEE802:
686 hlen = ETHER_ADDR_LEN;
687 layer = "ieee802";
688 break;
689 case ARPHRD_INFINIBAND:
690 hlen = 20; /* RFC 4391, INFINIBAND_ALEN */
691 layer = "infiniband";
692 break;
693 case ARPHRD_IEEE1394:
694 hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
695 layer = "firewire";
696
697 /*
698 * Restrict too long hardware addresses.
699 * Currently we are capable of handling 20-byte
700 * addresses ( sizeof(lle->ll_addr) )
701 */
702 if (ar->ar_hln >= 20)
703 hlen = 16;
704 break;
705 default:
706 ARP_LOG(LOG_NOTICE,
707 "packet with unknown hardware format 0x%02d received on "
708 "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
709 m_freem(m);
710 return;
711 }
712
713 if (hlen != 0 && hlen != ar->ar_hln) {
714 ARP_LOG(LOG_NOTICE,
715 "packet with invalid %s address length %d received on %s\n",
716 layer, ar->ar_hln, if_name(ifp));
717 m_freem(m);
718 return;
719 }
720
721 ARPSTAT_INC(received);
722 switch (ntohs(ar->ar_pro)) {
723 #ifdef INET
724 case ETHERTYPE_IP:
725 in_arpinput(m);
726 return;
727 #endif
728 }
729 m_freem(m);
730 }
731
732 #ifdef INET
733 /*
734 * ARP for Internet protocols on 10 Mb/s Ethernet.
735 * Algorithm is that given in RFC 826.
736 * In addition, a sanity check is performed on the sender
737 * protocol address, to catch impersonators.
738 * We no longer handle negotiations for use of trailer protocol:
739 * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
740 * along with IP replies if we wanted trailers sent to us,
741 * and also sent them in response to IP replies.
742 * This allowed either end to announce the desire to receive
743 * trailer packets.
744 * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
745 * but formerly didn't normally send requests.
746 */
747 static int log_arp_wrong_iface = 1;
748 static int log_arp_movements = 1;
749 static int log_arp_permanent_modify = 1;
750 static int allow_multicast = 0;
751
752 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
753 &log_arp_wrong_iface, 0,
754 "log arp packets arriving on the wrong interface");
755 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
756 &log_arp_movements, 0,
757 "log arp replies from MACs different than the one in the cache");
758 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
759 &log_arp_permanent_modify, 0,
760 "log arp replies from MACs different than the one in the permanent arp entry");
761 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
762 &allow_multicast, 0, "accept multicast addresses");
763
764 static void
in_arpinput(struct mbuf * m)765 in_arpinput(struct mbuf *m)
766 {
767 struct arphdr *ah;
768 struct ifnet *ifp = m->m_pkthdr.rcvif;
769 struct llentry *la = NULL, *la_tmp;
770 struct ifaddr *ifa;
771 struct in_ifaddr *ia;
772 struct sockaddr sa;
773 struct in_addr isaddr, itaddr, myaddr;
774 u_int8_t *enaddr = NULL;
775 int op;
776 int bridged = 0, is_bridge = 0;
777 int carped;
778 struct sockaddr_in sin;
779 struct sockaddr *dst;
780 struct nhop_object *nh;
781 uint8_t linkhdr[LLE_MAX_LINKHDR];
782 struct route ro;
783 size_t linkhdrsize;
784 int lladdr_off;
785 int error;
786 char addrbuf[INET_ADDRSTRLEN];
787
788 NET_EPOCH_ASSERT();
789
790 sin.sin_len = sizeof(struct sockaddr_in);
791 sin.sin_family = AF_INET;
792 sin.sin_addr.s_addr = 0;
793
794 if (ifp->if_bridge)
795 bridged = 1;
796 if (ifp->if_type == IFT_BRIDGE)
797 is_bridge = 1;
798
799 /*
800 * We already have checked that mbuf contains enough contiguous data
801 * to hold entire arp message according to the arp header.
802 */
803 ah = mtod(m, struct arphdr *);
804
805 /*
806 * ARP is only for IPv4 so we can reject packets with
807 * a protocol length not equal to an IPv4 address.
808 */
809 if (ah->ar_pln != sizeof(struct in_addr)) {
810 ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
811 sizeof(struct in_addr));
812 goto drop;
813 }
814
815 if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
816 ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
817 ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
818 goto drop;
819 }
820
821 op = ntohs(ah->ar_op);
822 (void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
823 (void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
824
825 if (op == ARPOP_REPLY)
826 ARPSTAT_INC(rxreplies);
827
828 /*
829 * For a bridge, we want to check the address irrespective
830 * of the receive interface. (This will change slightly
831 * when we have clusters of interfaces).
832 */
833 CK_LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
834 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
835 ia->ia_ifp == ifp) &&
836 itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
837 (ia->ia_ifa.ifa_carp == NULL ||
838 (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
839 ifa_ref(&ia->ia_ifa);
840 goto match;
841 }
842 }
843 CK_LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
844 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
845 ia->ia_ifp == ifp) &&
846 isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
847 ifa_ref(&ia->ia_ifa);
848 goto match;
849 }
850
851 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia) \
852 (ia->ia_ifp->if_bridge == ifp->if_softc && \
853 !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) && \
854 addr == ia->ia_addr.sin_addr.s_addr)
855 /*
856 * Check the case when bridge shares its MAC address with
857 * some of its children, so packets are claimed by bridge
858 * itself (bridge_input() does it first), but they are really
859 * meant to be destined to the bridge member.
860 */
861 if (is_bridge) {
862 CK_LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
863 if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
864 ifa_ref(&ia->ia_ifa);
865 ifp = ia->ia_ifp;
866 goto match;
867 }
868 }
869 }
870 #undef BDG_MEMBER_MATCHES_ARP
871
872 /*
873 * No match, use the first inet address on the receive interface
874 * as a dummy address for the rest of the function.
875 */
876 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
877 if (ifa->ifa_addr->sa_family == AF_INET &&
878 (ifa->ifa_carp == NULL ||
879 (*carp_iamatch_p)(ifa, &enaddr))) {
880 ia = ifatoia(ifa);
881 ifa_ref(ifa);
882 goto match;
883 }
884
885 /*
886 * If bridging, fall back to using any inet address.
887 */
888 if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL)
889 goto drop;
890 ifa_ref(&ia->ia_ifa);
891 match:
892 if (!enaddr)
893 enaddr = (u_int8_t *)IF_LLADDR(ifp);
894 carped = (ia->ia_ifa.ifa_carp != NULL);
895 myaddr = ia->ia_addr.sin_addr;
896 ifa_free(&ia->ia_ifa);
897 if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
898 goto drop; /* it's from me, ignore it. */
899 if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
900 ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
901 "%s!\n", inet_ntoa_r(isaddr, addrbuf));
902 goto drop;
903 }
904
905 if (ifp->if_addrlen != ah->ar_hln) {
906 ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
907 "i/f %d (ignored)\n", ifp->if_addrlen,
908 (u_char *) ar_sha(ah), ":", ah->ar_hln,
909 ifp->if_addrlen);
910 goto drop;
911 }
912
913 /*
914 * Warn if another host is using the same IP address, but only if the
915 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
916 * case we suppress the warning to avoid false positive complaints of
917 * potential misconfiguration.
918 */
919 if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
920 myaddr.s_addr != 0) {
921 ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
922 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
923 inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
924 itaddr = myaddr;
925 ARPSTAT_INC(dupips);
926 goto reply;
927 }
928 if (ifp->if_flags & IFF_STATICARP)
929 goto reply;
930
931 bzero(&sin, sizeof(sin));
932 sin.sin_len = sizeof(struct sockaddr_in);
933 sin.sin_family = AF_INET;
934 sin.sin_addr = isaddr;
935 dst = (struct sockaddr *)&sin;
936 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
937 if (la != NULL)
938 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
939 else if (itaddr.s_addr == myaddr.s_addr) {
940 /*
941 * Request/reply to our address, but no lle exists yet.
942 * Calculate full link prepend to use in lle.
943 */
944 linkhdrsize = sizeof(linkhdr);
945 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
946 &linkhdrsize, &lladdr_off) != 0)
947 goto reply;
948
949 /* Allocate new entry */
950 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
951 if (la == NULL) {
952 /*
953 * lle creation may fail if source address belongs
954 * to non-directly connected subnet. However, we
955 * will try to answer the request instead of dropping
956 * frame.
957 */
958 goto reply;
959 }
960 lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
961 lladdr_off);
962
963 IF_AFDATA_WLOCK(ifp);
964 LLE_WLOCK(la);
965 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
966
967 /*
968 * Check if lle still does not exists.
969 * If it does, that means that we either
970 * 1) have configured it explicitly, via
971 * 1a) 'arp -s' static entry or
972 * 1b) interface address static record
973 * or
974 * 2) it was the result of sending first packet to-host
975 * or
976 * 3) it was another arp reply packet we handled in
977 * different thread.
978 *
979 * In all cases except 3) we definitely need to prefer
980 * existing lle. For the sake of simplicity, prefer any
981 * existing lle over newly-create one.
982 */
983 if (la_tmp == NULL)
984 lltable_link_entry(LLTABLE(ifp), la);
985 IF_AFDATA_WUNLOCK(ifp);
986
987 if (la_tmp == NULL) {
988 arp_mark_lle_reachable(la, ifp);
989 LLE_WUNLOCK(la);
990 } else {
991 /* Free newly-create entry and handle packet */
992 lltable_free_entry(LLTABLE(ifp), la);
993 la = la_tmp;
994 la_tmp = NULL;
995 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
996 /* arp_check_update_lle() returns @la unlocked */
997 }
998 la = NULL;
999 }
1000 reply:
1001 if (op != ARPOP_REQUEST)
1002 goto drop;
1003 ARPSTAT_INC(rxrequests);
1004
1005 if (itaddr.s_addr == myaddr.s_addr) {
1006 /* Shortcut.. the receiving interface is the target. */
1007 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1008 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1009 } else {
1010 /*
1011 * Destination address is not ours. Check if
1012 * proxyarp entry exists or proxyarp is turned on globally.
1013 */
1014 struct llentry *lle;
1015
1016 sin.sin_addr = itaddr;
1017 lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1018
1019 if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1020 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1021 (void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1022 LLE_RUNLOCK(lle);
1023 } else {
1024 if (lle != NULL)
1025 LLE_RUNLOCK(lle);
1026
1027 if (!V_arp_proxyall)
1028 goto drop;
1029
1030 NET_EPOCH_ASSERT();
1031 nh = fib4_lookup(ifp->if_fib, itaddr, 0, 0, 0);
1032 if (nh == NULL)
1033 goto drop;
1034
1035 /*
1036 * Don't send proxies for nodes on the same interface
1037 * as this one came out of, or we'll get into a fight
1038 * over who claims what Ether address.
1039 */
1040 if (nh->nh_ifp == ifp)
1041 goto drop;
1042
1043 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1044 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1045
1046 /*
1047 * Also check that the node which sent the ARP packet
1048 * is on the interface we expect it to be on. This
1049 * avoids ARP chaos if an interface is connected to the
1050 * wrong network.
1051 */
1052
1053 nh = fib4_lookup(ifp->if_fib, isaddr, 0, 0, 0);
1054 if (nh == NULL)
1055 goto drop;
1056 if (nh->nh_ifp != ifp) {
1057 ARP_LOG(LOG_INFO, "proxy: ignoring request"
1058 " from %s via %s\n",
1059 inet_ntoa_r(isaddr, addrbuf),
1060 ifp->if_xname);
1061 goto drop;
1062 }
1063
1064 #ifdef DEBUG_PROXY
1065 printf("arp: proxying for %s\n",
1066 inet_ntoa_r(itaddr, addrbuf));
1067 #endif
1068 }
1069 }
1070
1071 if (itaddr.s_addr == myaddr.s_addr &&
1072 IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1073 /* RFC 3927 link-local IPv4; always reply by broadcast. */
1074 #ifdef DEBUG_LINKLOCAL
1075 printf("arp: sending reply for link-local addr %s\n",
1076 inet_ntoa_r(itaddr, addrbuf));
1077 #endif
1078 m->m_flags |= M_BCAST;
1079 m->m_flags &= ~M_MCAST;
1080 } else {
1081 /* default behaviour; never reply by broadcast. */
1082 m->m_flags &= ~(M_BCAST|M_MCAST);
1083 }
1084 (void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1085 (void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1086 ah->ar_op = htons(ARPOP_REPLY);
1087 ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1088 m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1089 m->m_pkthdr.len = m->m_len;
1090 m->m_pkthdr.rcvif = NULL;
1091 sa.sa_family = AF_ARP;
1092 sa.sa_len = 2;
1093
1094 /* Calculate link header for sending frame */
1095 bzero(&ro, sizeof(ro));
1096 linkhdrsize = sizeof(linkhdr);
1097 error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1098
1099 /*
1100 * arp_fillheader() may fail due to lack of support inside encap request
1101 * routing. This is not necessary an error, AF_ARP can/should be handled
1102 * by if_output().
1103 */
1104 if (error != 0 && error != EAFNOSUPPORT) {
1105 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1106 if_name(ifp), error);
1107 goto drop;
1108 }
1109
1110 ro.ro_prepend = linkhdr;
1111 ro.ro_plen = linkhdrsize;
1112 ro.ro_flags = 0;
1113
1114 m_clrprotoflags(m); /* Avoid confusing lower layers. */
1115 (*ifp->if_output)(ifp, m, &sa, &ro);
1116 ARPSTAT_INC(txreplies);
1117 return;
1118
1119 drop:
1120 m_freem(m);
1121 }
1122 #endif
1123
1124 static struct mbuf *
arp_grab_holdchain(struct llentry * la)1125 arp_grab_holdchain(struct llentry *la)
1126 {
1127 struct mbuf *chain;
1128
1129 LLE_WLOCK_ASSERT(la);
1130
1131 chain = la->la_hold;
1132 la->la_hold = NULL;
1133 la->la_numheld = 0;
1134
1135 return (chain);
1136 }
1137
1138 static void
arp_flush_holdchain(struct ifnet * ifp,struct llentry * la,struct mbuf * chain)1139 arp_flush_holdchain(struct ifnet *ifp, struct llentry *la, struct mbuf *chain)
1140 {
1141 struct mbuf *m_hold, *m_hold_next;
1142 struct sockaddr_in sin;
1143
1144 NET_EPOCH_ASSERT();
1145
1146 struct route ro = {
1147 .ro_prepend = la->r_linkdata,
1148 .ro_plen = la->r_hdrlen,
1149 };
1150
1151 lltable_fill_sa_entry(la, (struct sockaddr *)&sin);
1152
1153 for (m_hold = chain; m_hold != NULL; m_hold = m_hold_next) {
1154 m_hold_next = m_hold->m_nextpkt;
1155 m_hold->m_nextpkt = NULL;
1156 /* Avoid confusing lower layers. */
1157 m_clrprotoflags(m_hold);
1158 (*ifp->if_output)(ifp, m_hold, (struct sockaddr *)&sin, &ro);
1159 }
1160 }
1161
1162 /*
1163 * Checks received arp data against existing @la.
1164 * Updates lle state/performs notification if necessary.
1165 */
1166 static void
arp_check_update_lle(struct arphdr * ah,struct in_addr isaddr,struct ifnet * ifp,int bridged,struct llentry * la)1167 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1168 int bridged, struct llentry *la)
1169 {
1170 uint8_t linkhdr[LLE_MAX_LINKHDR];
1171 size_t linkhdrsize;
1172 int lladdr_off;
1173 char addrbuf[INET_ADDRSTRLEN];
1174
1175 LLE_WLOCK_ASSERT(la);
1176
1177 /* the following is not an error when doing bridging */
1178 if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1179 if (log_arp_wrong_iface)
1180 ARP_LOG(LOG_WARNING, "%s is on %s "
1181 "but got reply from %*D on %s\n",
1182 inet_ntoa_r(isaddr, addrbuf),
1183 la->lle_tbl->llt_ifp->if_xname,
1184 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1185 ifp->if_xname);
1186 LLE_WUNLOCK(la);
1187 return;
1188 }
1189 if ((la->la_flags & LLE_VALID) &&
1190 bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1191 if (la->la_flags & LLE_STATIC) {
1192 LLE_WUNLOCK(la);
1193 if (log_arp_permanent_modify)
1194 ARP_LOG(LOG_ERR,
1195 "%*D attempts to modify "
1196 "permanent entry for %s on %s\n",
1197 ifp->if_addrlen,
1198 (u_char *)ar_sha(ah), ":",
1199 inet_ntoa_r(isaddr, addrbuf),
1200 ifp->if_xname);
1201 return;
1202 }
1203 if (log_arp_movements) {
1204 ARP_LOG(LOG_INFO, "%s moved from %*D "
1205 "to %*D on %s\n",
1206 inet_ntoa_r(isaddr, addrbuf),
1207 ifp->if_addrlen,
1208 (u_char *)la->ll_addr, ":",
1209 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1210 ifp->if_xname);
1211 }
1212 }
1213
1214 /* Calculate full link prepend to use in lle */
1215 linkhdrsize = sizeof(linkhdr);
1216 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1217 &linkhdrsize, &lladdr_off) != 0) {
1218 LLE_WUNLOCK(la);
1219 return;
1220 }
1221
1222 /* Check if something has changed */
1223 if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1224 (la->la_flags & LLE_VALID) == 0) {
1225 /* Try to perform LLE update */
1226 if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1227 lladdr_off) == 0) {
1228 LLE_WUNLOCK(la);
1229 return;
1230 }
1231
1232 /* Clear fast path feedback request if set */
1233 llentry_mark_used(la);
1234 }
1235
1236 arp_mark_lle_reachable(la, ifp);
1237
1238 /*
1239 * The packets are all freed within the call to the output
1240 * routine.
1241 *
1242 * NB: The lock MUST be released before the call to the
1243 * output routine.
1244 */
1245 if (la->la_hold != NULL) {
1246 struct mbuf *chain;
1247
1248 chain = arp_grab_holdchain(la);
1249 LLE_WUNLOCK(la);
1250 arp_flush_holdchain(ifp, la, chain);
1251 } else
1252 LLE_WUNLOCK(la);
1253 }
1254
1255 static void
arp_mark_lle_reachable(struct llentry * la,struct ifnet * ifp)1256 arp_mark_lle_reachable(struct llentry *la, struct ifnet *ifp)
1257 {
1258 int canceled, wtime;
1259
1260 LLE_WLOCK_ASSERT(la);
1261
1262 la->ln_state = ARP_LLINFO_REACHABLE;
1263 EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1264
1265 if ((ifp->if_flags & IFF_STICKYARP) != 0)
1266 la->la_flags |= LLE_STATIC;
1267
1268 if (!(la->la_flags & LLE_STATIC)) {
1269 LLE_ADDREF(la);
1270 la->la_expire = time_uptime + V_arpt_keep;
1271 wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1272 if (wtime < 0)
1273 wtime = V_arpt_keep;
1274 canceled = callout_reset(&la->lle_timer,
1275 hz * wtime, arptimer, la);
1276 if (canceled)
1277 LLE_REMREF(la);
1278 }
1279 la->la_asked = 0;
1280 la->la_preempt = V_arp_maxtries;
1281 }
1282
1283 /*
1284 * Add permanent link-layer record for given interface address.
1285 */
1286 static __noinline void
arp_add_ifa_lle(struct ifnet * ifp,const struct sockaddr * dst)1287 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1288 {
1289 struct llentry *lle, *lle_tmp;
1290
1291 /*
1292 * Interface address LLE record is considered static
1293 * because kernel code relies on LLE_STATIC flag to check
1294 * if these entries can be rewriten by arp updates.
1295 */
1296 lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1297 if (lle == NULL) {
1298 log(LOG_INFO, "arp_ifinit: cannot create arp "
1299 "entry for interface address\n");
1300 return;
1301 }
1302
1303 IF_AFDATA_WLOCK(ifp);
1304 LLE_WLOCK(lle);
1305 /* Unlink any entry if exists */
1306 lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1307 if (lle_tmp != NULL)
1308 lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1309
1310 lltable_link_entry(LLTABLE(ifp), lle);
1311 IF_AFDATA_WUNLOCK(ifp);
1312
1313 if (lle_tmp != NULL)
1314 EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1315
1316 EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1317 LLE_WUNLOCK(lle);
1318 if (lle_tmp != NULL)
1319 lltable_free_entry(LLTABLE(ifp), lle_tmp);
1320 }
1321
1322 /*
1323 * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1324 * of valid values.
1325 */
1326 static int
sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)1327 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1328 {
1329 int error;
1330 int rexmit_count = *(int *)arg1;
1331
1332 error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1333
1334 /* Enforce limits on any new value that may have been set. */
1335 if (!error && req->newptr) {
1336 /* A new value was set. */
1337 if (rexmit_count < 0) {
1338 rexmit_count = 0;
1339 } else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1340 rexmit_count = MAX_GARP_RETRANSMITS;
1341 }
1342 *(int *)arg1 = rexmit_count;
1343 }
1344
1345 return (error);
1346 }
1347
1348 /*
1349 * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1350 * retransmit it again. A pending callout owns a reference to the ifa.
1351 */
1352 static void
garp_rexmit(void * arg)1353 garp_rexmit(void *arg)
1354 {
1355 struct in_ifaddr *ia = arg;
1356
1357 if (callout_pending(&ia->ia_garp_timer) ||
1358 !callout_active(&ia->ia_garp_timer)) {
1359 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1360 ifa_free(&ia->ia_ifa);
1361 return;
1362 }
1363
1364 CURVNET_SET(ia->ia_ifa.ifa_ifp->if_vnet);
1365
1366 /*
1367 * Drop lock while the ARP request is generated.
1368 */
1369 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1370
1371 arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1372 &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1373
1374 /*
1375 * Increment the count of retransmissions. If the count has reached the
1376 * maximum value, stop sending the GARP packets. Otherwise, schedule
1377 * the callout to retransmit another GARP packet.
1378 */
1379 ++ia->ia_garp_count;
1380 if (ia->ia_garp_count >= garp_rexmit_count) {
1381 ifa_free(&ia->ia_ifa);
1382 } else {
1383 int rescheduled;
1384 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1385 rescheduled = callout_reset(&ia->ia_garp_timer,
1386 (1 << ia->ia_garp_count) * hz,
1387 garp_rexmit, ia);
1388 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1389 if (rescheduled) {
1390 ifa_free(&ia->ia_ifa);
1391 }
1392 }
1393
1394 CURVNET_RESTORE();
1395 }
1396
1397 /*
1398 * Start the GARP retransmit timer.
1399 *
1400 * A single GARP is always transmitted when an IPv4 address is added
1401 * to an interface and that is usually sufficient. However, in some
1402 * circumstances, such as when a shared address is passed between
1403 * cluster nodes, this single GARP may occasionally be dropped or
1404 * lost. This can lead to neighbors on the network link working with a
1405 * stale ARP cache and sending packets destined for that address to
1406 * the node that previously owned the address, which may not respond.
1407 *
1408 * To avoid this situation, GARP retransmits can be enabled by setting
1409 * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1410 * than zero. The setting represents the maximum number of
1411 * retransmissions. The interval between retransmissions is calculated
1412 * using an exponential backoff algorithm, doubling each time, so the
1413 * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1414 */
1415 static void
garp_timer_start(struct ifaddr * ifa)1416 garp_timer_start(struct ifaddr *ifa)
1417 {
1418 struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1419
1420 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1421 ia->ia_garp_count = 0;
1422 if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1423 garp_rexmit, ia) == 0) {
1424 ifa_ref(ifa);
1425 }
1426 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1427 }
1428
1429 void
arp_ifinit(struct ifnet * ifp,struct ifaddr * ifa)1430 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1431 {
1432 struct epoch_tracker et;
1433 const struct sockaddr_in *dst_in;
1434 const struct sockaddr *dst;
1435
1436 if (ifa->ifa_carp != NULL)
1437 return;
1438
1439 dst = ifa->ifa_addr;
1440 dst_in = (const struct sockaddr_in *)dst;
1441
1442 if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1443 return;
1444 NET_EPOCH_ENTER(et);
1445 arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1446 NET_EPOCH_EXIT(et);
1447 if (garp_rexmit_count > 0) {
1448 garp_timer_start(ifa);
1449 }
1450
1451 arp_add_ifa_lle(ifp, dst);
1452 }
1453
1454 void
arp_announce_ifaddr(struct ifnet * ifp,struct in_addr addr,u_char * enaddr)1455 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1456 {
1457
1458 if (ntohl(addr.s_addr) != INADDR_ANY)
1459 arprequest(ifp, &addr, &addr, enaddr);
1460 }
1461
1462 /*
1463 * Sends gratuitous ARPs for each ifaddr to notify other
1464 * nodes about the address change.
1465 */
1466 static __noinline void
arp_handle_ifllchange(struct ifnet * ifp)1467 arp_handle_ifllchange(struct ifnet *ifp)
1468 {
1469 struct ifaddr *ifa;
1470
1471 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1472 if (ifa->ifa_addr->sa_family == AF_INET)
1473 arp_ifinit(ifp, ifa);
1474 }
1475 }
1476
1477 /*
1478 * A handler for interface link layer address change event.
1479 */
1480 static void
arp_iflladdr(void * arg __unused,struct ifnet * ifp)1481 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1482 {
1483 /* if_bridge can update its lladdr during if_vmove(), after we've done
1484 * if_detach_internal()/dom_ifdetach(). */
1485 if (ifp->if_afdata[AF_INET] == NULL)
1486 return;
1487
1488 lltable_update_ifaddr(LLTABLE(ifp));
1489
1490 if ((ifp->if_flags & IFF_UP) != 0)
1491 arp_handle_ifllchange(ifp);
1492 }
1493
1494 static void
vnet_arp_init(void)1495 vnet_arp_init(void)
1496 {
1497
1498 if (IS_DEFAULT_VNET(curvnet)) {
1499 netisr_register(&arp_nh);
1500 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1501 arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1502 }
1503 #ifdef VIMAGE
1504 else
1505 netisr_register_vnet(&arp_nh);
1506 #endif
1507 }
1508 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1509 vnet_arp_init, 0);
1510
1511 #ifdef VIMAGE
1512 /*
1513 * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1514 * lookups after destroying the hash. Ideally this would go on SI_ORDER_3.5.
1515 */
1516 static void
vnet_arp_destroy(__unused void * arg)1517 vnet_arp_destroy(__unused void *arg)
1518 {
1519
1520 netisr_unregister_vnet(&arp_nh);
1521 }
1522 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1523 vnet_arp_destroy, NULL);
1524 #endif
1525