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 * @(#)if_ether.c 8.1 (Berkeley) 6/10/93
32 */
33
34 /*
35 * Ethernet address resolution protocol.
36 * TODO:
37 * add "inuse/lock" bit (or ref. count) along with valid bit
38 */
39
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42
43 #include "opt_inet.h"
44
45 #include <sys/param.h>
46 #include <sys/eventhandler.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/queue.h>
50 #include <sys/sysctl.h>
51 #include <sys/systm.h>
52 #include <sys/mbuf.h>
53 #include <sys/malloc.h>
54 #include <sys/proc.h>
55 #include <sys/rmlock.h>
56 #include <sys/socket.h>
57 #include <sys/syslog.h>
58
59 #include <net/if.h>
60 #include <net/if_var.h>
61 #include <net/if_dl.h>
62 #include <net/if_types.h>
63 #include <net/netisr.h>
64 #include <net/ethernet.h>
65 #include <net/route.h>
66 #include <net/route/nhop.h>
67 #include <net/vnet.h>
68
69 #include <netinet/in.h>
70 #include <netinet/in_fib.h>
71 #include <netinet/in_var.h>
72 #include <net/if_llatbl.h>
73 #include <netinet/if_ether.h>
74 #ifdef INET
75 #include <netinet/ip_carp.h>
76 #endif
77
78 #include <security/mac/mac_framework.h>
79
80 #define SIN(s) ((const struct sockaddr_in *)(s))
81
82 static struct timeval arp_lastlog;
83 static int arp_curpps;
84 static int arp_maxpps = 1;
85
86 /* Simple ARP state machine */
87 enum arp_llinfo_state {
88 ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */
89 ARP_LLINFO_REACHABLE, /* LLE is valid */
90 ARP_LLINFO_VERIFY, /* LLE is valid, need refresh */
91 ARP_LLINFO_DELETED, /* LLE is deleted */
92 };
93
94 SYSCTL_DECL(_net_link_ether);
95 static SYSCTL_NODE(_net_link_ether, PF_INET, inet,
96 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
97 "");
98 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp,
99 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
100 "");
101
102 /* timer values */
103 VNET_DEFINE_STATIC(int, arpt_keep) = (20*60); /* once resolved, good for 20
104 * minutes */
105 VNET_DEFINE_STATIC(int, arp_maxtries) = 5;
106 VNET_DEFINE_STATIC(int, arp_proxyall) = 0;
107 VNET_DEFINE_STATIC(int, arpt_down) = 20; /* keep incomplete entries for
108 * 20 seconds */
109 VNET_DEFINE_STATIC(int, arpt_rexmit) = 1; /* retransmit arp entries, sec*/
110 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat); /* ARP statistics, see if_arp.h */
111 VNET_PCPUSTAT_SYSINIT(arpstat);
112
113 #ifdef VIMAGE
114 VNET_PCPUSTAT_SYSUNINIT(arpstat);
115 #endif /* VIMAGE */
116
117 VNET_DEFINE_STATIC(int, arp_maxhold) = 16;
118
119 #define V_arpt_keep VNET(arpt_keep)
120 #define V_arpt_down VNET(arpt_down)
121 #define V_arpt_rexmit VNET(arpt_rexmit)
122 #define V_arp_maxtries VNET(arp_maxtries)
123 #define V_arp_proxyall VNET(arp_proxyall)
124 #define V_arp_maxhold VNET(arp_maxhold)
125
126 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
127 &VNET_NAME(arpt_keep), 0,
128 "ARP entry lifetime in seconds");
129 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
130 &VNET_NAME(arp_maxtries), 0,
131 "ARP resolution attempts before returning error");
132 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
133 &VNET_NAME(arp_proxyall), 0,
134 "Enable proxy ARP for all suitable requests");
135 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
136 &VNET_NAME(arpt_down), 0,
137 "Incomplete ARP entry lifetime in seconds");
138 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
139 arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
140 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
141 &VNET_NAME(arp_maxhold), 0,
142 "Number of packets to hold per ARP entry");
143 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
144 CTLFLAG_RW, &arp_maxpps, 0,
145 "Maximum number of remotely triggered ARP messages that can be "
146 "logged per second");
147
148 /*
149 * Due to the exponential backoff algorithm used for the interval between GARP
150 * retransmissions, the maximum number of retransmissions is limited for
151 * sanity. This limit corresponds to a maximum interval between retransmissions
152 * of 2^16 seconds ~= 18 hours.
153 *
154 * Making this limit more dynamic is more complicated than worthwhile,
155 * especially since sending out GARPs spaced days apart would be of little
156 * use. A maximum dynamic limit would look something like:
157 *
158 * const int max = fls(INT_MAX / hz) - 1;
159 */
160 #define MAX_GARP_RETRANSMITS 16
161 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS);
162 static int garp_rexmit_count = 0; /* GARP retransmission setting. */
163
164 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count,
165 CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE,
166 &garp_rexmit_count, 0, sysctl_garp_rexmit, "I",
167 "Number of times to retransmit GARP packets;"
168 " 0 to disable, maximum of 16");
169
170 VNET_DEFINE_STATIC(int, arp_log_level) = LOG_INFO; /* Min. log(9) level. */
171 #define V_arp_log_level VNET(arp_log_level)
172 SYSCTL_INT(_net_link_ether_arp, OID_AUTO, log_level, CTLFLAG_VNET | CTLFLAG_RW,
173 &VNET_NAME(arp_log_level), 0,
174 "Minimum log(9) level for recording rate limited arp log messages. "
175 "The higher will be log more (emerg=0, info=6 (default), debug=7).");
176 #define ARP_LOG(pri, ...) do { \
177 if ((pri) <= V_arp_log_level && \
178 ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps)) \
179 log((pri), "arp: " __VA_ARGS__); \
180 } while (0)
181
182 static void arpintr(struct mbuf *);
183 static void arptimer(void *);
184 #ifdef INET
185 static void in_arpinput(struct mbuf *);
186 #endif
187
188 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
189 struct ifnet *ifp, int bridged, struct llentry *la);
190 static void arp_mark_lle_reachable(struct llentry *la);
191 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
192
193 static eventhandler_tag iflladdr_tag;
194
195 static const struct netisr_handler arp_nh = {
196 .nh_name = "arp",
197 .nh_handler = arpintr,
198 .nh_proto = NETISR_ARP,
199 .nh_policy = NETISR_POLICY_SOURCE,
200 };
201
202 /*
203 * Timeout routine. Age arp_tab entries periodically.
204 */
205 static void
arptimer(void * arg)206 arptimer(void *arg)
207 {
208 struct llentry *lle = (struct llentry *)arg;
209 struct ifnet *ifp;
210 int r_skip_req;
211
212 if (lle->la_flags & LLE_STATIC) {
213 return;
214 }
215 LLE_WLOCK(lle);
216 if (callout_pending(&lle->lle_timer)) {
217 /*
218 * Here we are a bit odd here in the treatment of
219 * active/pending. If the pending bit is set, it got
220 * rescheduled before I ran. The active
221 * bit we ignore, since if it was stopped
222 * in ll_tablefree() and was currently running
223 * it would have return 0 so the code would
224 * not have deleted it since the callout could
225 * not be stopped so we want to go through
226 * with the delete here now. If the callout
227 * was restarted, the pending bit will be back on and
228 * we just want to bail since the callout_reset would
229 * return 1 and our reference would have been removed
230 * by arpresolve() below.
231 */
232 LLE_WUNLOCK(lle);
233 return;
234 }
235 ifp = lle->lle_tbl->llt_ifp;
236 CURVNET_SET(ifp->if_vnet);
237
238 switch (lle->ln_state) {
239 case ARP_LLINFO_REACHABLE:
240
241 /*
242 * Expiration time is approaching.
243 * Let's try to refresh entry if it is still
244 * in use.
245 *
246 * Set r_skip_req to get feedback from
247 * fast path. Change state and re-schedule
248 * ourselves.
249 */
250 LLE_REQ_LOCK(lle);
251 lle->r_skip_req = 1;
252 LLE_REQ_UNLOCK(lle);
253 lle->ln_state = ARP_LLINFO_VERIFY;
254 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
255 LLE_WUNLOCK(lle);
256 CURVNET_RESTORE();
257 return;
258 case ARP_LLINFO_VERIFY:
259 LLE_REQ_LOCK(lle);
260 r_skip_req = lle->r_skip_req;
261 LLE_REQ_UNLOCK(lle);
262
263 if (r_skip_req == 0 && lle->la_preempt > 0) {
264 /* Entry was used, issue refresh request */
265 struct epoch_tracker et;
266 struct in_addr dst;
267
268 dst = lle->r_l3addr.addr4;
269 lle->la_preempt--;
270 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
271 LLE_WUNLOCK(lle);
272 NET_EPOCH_ENTER(et);
273 arprequest(ifp, NULL, &dst, NULL);
274 NET_EPOCH_EXIT(et);
275 CURVNET_RESTORE();
276 return;
277 }
278 /* Nothing happened. Reschedule if not too late */
279 if (lle->la_expire > time_uptime) {
280 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
281 LLE_WUNLOCK(lle);
282 CURVNET_RESTORE();
283 return;
284 }
285 break;
286 case ARP_LLINFO_INCOMPLETE:
287 case ARP_LLINFO_DELETED:
288 break;
289 }
290
291 if ((lle->la_flags & LLE_DELETED) == 0) {
292 int evt;
293
294 if (lle->la_flags & LLE_VALID)
295 evt = LLENTRY_EXPIRED;
296 else
297 evt = LLENTRY_TIMEDOUT;
298 EVENTHANDLER_INVOKE(lle_event, lle, evt);
299 }
300
301 callout_stop(&lle->lle_timer);
302
303 /* XXX: LOR avoidance. We still have ref on lle. */
304 LLE_WUNLOCK(lle);
305 IF_AFDATA_LOCK(ifp);
306 LLE_WLOCK(lle);
307
308 /* Guard against race with other llentry_free(). */
309 if (lle->la_flags & LLE_LINKED) {
310 LLE_REMREF(lle);
311 lltable_unlink_entry(lle->lle_tbl, lle);
312 }
313 IF_AFDATA_UNLOCK(ifp);
314
315 size_t pkts_dropped = llentry_free(lle);
316
317 ARPSTAT_ADD(dropped, pkts_dropped);
318 ARPSTAT_INC(timeouts);
319
320 CURVNET_RESTORE();
321 }
322
323 /*
324 * Stores link-layer header for @ifp in format suitable for if_output()
325 * into buffer @buf. Resulting header length is stored in @bufsize.
326 *
327 * Returns 0 on success.
328 */
329 static int
arp_fillheader(struct ifnet * ifp,struct arphdr * ah,int bcast,u_char * buf,size_t * bufsize)330 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
331 size_t *bufsize)
332 {
333 struct if_encap_req ereq;
334 int error;
335
336 bzero(buf, *bufsize);
337 bzero(&ereq, sizeof(ereq));
338 ereq.buf = buf;
339 ereq.bufsize = *bufsize;
340 ereq.rtype = IFENCAP_LL;
341 ereq.family = AF_ARP;
342 ereq.lladdr = ar_tha(ah);
343 ereq.hdata = (u_char *)ah;
344 if (bcast)
345 ereq.flags = IFENCAP_FLAG_BROADCAST;
346 error = ifp->if_requestencap(ifp, &ereq);
347 if (error == 0)
348 *bufsize = ereq.bufsize;
349
350 return (error);
351 }
352
353 /*
354 * Broadcast an ARP request. Caller specifies:
355 * - arp header source ip address
356 * - arp header target ip address
357 * - arp header source ethernet address
358 */
359 static int
arprequest_internal(struct ifnet * ifp,const struct in_addr * sip,const struct in_addr * tip,u_char * enaddr)360 arprequest_internal(struct ifnet *ifp, const struct in_addr *sip,
361 const struct in_addr *tip, u_char *enaddr)
362 {
363 struct mbuf *m;
364 struct arphdr *ah;
365 struct sockaddr sa;
366 u_char *carpaddr = NULL;
367 uint8_t linkhdr[LLE_MAX_LINKHDR];
368 size_t linkhdrsize;
369 struct route ro;
370 int error;
371
372 NET_EPOCH_ASSERT();
373
374 if (sip == NULL) {
375 /*
376 * The caller did not supply a source address, try to find
377 * a compatible one among those assigned to this interface.
378 */
379 struct ifaddr *ifa;
380
381 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
382 if (ifa->ifa_addr->sa_family != AF_INET)
383 continue;
384
385 if (ifa->ifa_carp) {
386 if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
387 continue;
388 sip = &IA_SIN(ifa)->sin_addr;
389 } else {
390 carpaddr = NULL;
391 sip = &IA_SIN(ifa)->sin_addr;
392 }
393
394 if (0 == ((sip->s_addr ^ tip->s_addr) &
395 IA_MASKSIN(ifa)->sin_addr.s_addr))
396 break; /* found it. */
397 }
398 if (sip == NULL) {
399 printf("%s: cannot find matching address\n", __func__);
400 return (EADDRNOTAVAIL);
401 }
402 }
403 if (enaddr == NULL)
404 enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
405
406 if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
407 return (ENOMEM);
408 m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
409 2 * ifp->if_addrlen;
410 m->m_pkthdr.len = m->m_len;
411 M_ALIGN(m, m->m_len);
412 ah = mtod(m, struct arphdr *);
413 bzero((caddr_t)ah, m->m_len);
414 #ifdef MAC
415 mac_netinet_arp_send(ifp, m);
416 #endif
417 ah->ar_pro = htons(ETHERTYPE_IP);
418 ah->ar_hln = ifp->if_addrlen; /* hardware address length */
419 ah->ar_pln = sizeof(struct in_addr); /* protocol address length */
420 ah->ar_op = htons(ARPOP_REQUEST);
421 bcopy(enaddr, ar_sha(ah), ah->ar_hln);
422 bcopy(sip, ar_spa(ah), ah->ar_pln);
423 bcopy(tip, ar_tpa(ah), ah->ar_pln);
424 sa.sa_family = AF_ARP;
425 sa.sa_len = 2;
426
427 /* Calculate link header for sending frame */
428 bzero(&ro, sizeof(ro));
429 linkhdrsize = sizeof(linkhdr);
430 error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
431 if (error != 0 && error != EAFNOSUPPORT) {
432 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
433 if_name(ifp), error);
434 return (error);
435 }
436
437 ro.ro_prepend = linkhdr;
438 ro.ro_plen = linkhdrsize;
439 ro.ro_flags = 0;
440
441 m->m_flags |= M_BCAST;
442 m_clrprotoflags(m); /* Avoid confusing lower layers. */
443 error = (*ifp->if_output)(ifp, m, &sa, &ro);
444 ARPSTAT_INC(txrequests);
445 if (error) {
446 ARPSTAT_INC(txerrors);
447 ARP_LOG(LOG_DEBUG, "Failed to send ARP packet on %s: %d\n",
448 if_name(ifp), error);
449 }
450 return (error);
451 }
452
453 void
arprequest(struct ifnet * ifp,const struct in_addr * sip,const struct in_addr * tip,u_char * enaddr)454 arprequest(struct ifnet *ifp, const struct in_addr *sip,
455 const struct in_addr *tip, u_char *enaddr)
456 {
457
458 (void) arprequest_internal(ifp, sip, tip, enaddr);
459 }
460
461 /*
462 * Resolve an IP address into an ethernet address - heavy version.
463 * Used internally by arpresolve().
464 * We have already checked that we can't use an existing lle without
465 * modification so we have to acquire an LLE_EXCLUSIVE lle lock.
466 *
467 * On success, desten and pflags are filled in and the function returns 0;
468 * If the packet must be held pending resolution, we return EWOULDBLOCK
469 * On other errors, we return the corresponding error code.
470 * Note that m_freem() handles NULL.
471 */
472 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)473 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
474 const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
475 struct llentry **plle)
476 {
477 struct llentry *la = NULL, *la_tmp;
478 struct mbuf *curr = NULL;
479 struct mbuf *next = NULL;
480 int error, renew;
481 char *lladdr;
482 int ll_len;
483
484 NET_EPOCH_ASSERT();
485
486 if (pflags != NULL)
487 *pflags = 0;
488 if (plle != NULL)
489 *plle = NULL;
490
491 if ((flags & LLE_CREATE) == 0)
492 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
493 if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
494 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
495 if (la == NULL) {
496 char addrbuf[INET_ADDRSTRLEN];
497
498 log(LOG_DEBUG,
499 "arpresolve: can't allocate llinfo for %s on %s\n",
500 inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
501 if_name(ifp));
502 m_freem(m);
503 return (EINVAL);
504 }
505
506 IF_AFDATA_WLOCK(ifp);
507 LLE_WLOCK(la);
508 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
509 /* Prefer ANY existing lle over newly-created one */
510 if (la_tmp == NULL)
511 lltable_link_entry(LLTABLE(ifp), la);
512 IF_AFDATA_WUNLOCK(ifp);
513 if (la_tmp != NULL) {
514 lltable_free_entry(LLTABLE(ifp), la);
515 la = la_tmp;
516 }
517 }
518 if (la == NULL) {
519 m_freem(m);
520 return (EINVAL);
521 }
522
523 if ((la->la_flags & LLE_VALID) &&
524 ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
525 if (flags & LLE_ADDRONLY) {
526 lladdr = la->ll_addr;
527 ll_len = ifp->if_addrlen;
528 } else {
529 lladdr = la->r_linkdata;
530 ll_len = la->r_hdrlen;
531 }
532 bcopy(lladdr, desten, ll_len);
533
534 /* Notify LLE code that the entry was used by datapath */
535 llentry_mark_used(la);
536 if (pflags != NULL)
537 *pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
538 if (plle) {
539 LLE_ADDREF(la);
540 *plle = la;
541 }
542 LLE_WUNLOCK(la);
543 return (0);
544 }
545
546 renew = (la->la_asked == 0 || la->la_expire != time_uptime);
547 /*
548 * There is an arptab entry, but no ethernet address
549 * response yet. Add the mbuf to the list, dropping
550 * the oldest packet if we have exceeded the system
551 * setting.
552 */
553 if (m != NULL) {
554 if (la->la_numheld >= V_arp_maxhold) {
555 if (la->la_hold != NULL) {
556 next = la->la_hold->m_nextpkt;
557 m_freem(la->la_hold);
558 la->la_hold = next;
559 la->la_numheld--;
560 ARPSTAT_INC(dropped);
561 }
562 }
563 if (la->la_hold != NULL) {
564 curr = la->la_hold;
565 while (curr->m_nextpkt != NULL)
566 curr = curr->m_nextpkt;
567 curr->m_nextpkt = m;
568 } else
569 la->la_hold = m;
570 la->la_numheld++;
571 }
572 /*
573 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
574 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
575 * if we have already sent arp_maxtries ARP requests. Retransmit the
576 * ARP request, but not faster than one request per second.
577 */
578 if (la->la_asked < V_arp_maxtries)
579 error = EWOULDBLOCK; /* First request. */
580 else
581 error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
582
583 if (renew) {
584 int canceled, e;
585
586 LLE_ADDREF(la);
587 la->la_expire = time_uptime;
588 canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
589 arptimer, la);
590 if (canceled)
591 LLE_REMREF(la);
592 la->la_asked++;
593 LLE_WUNLOCK(la);
594 e = arprequest_internal(ifp, NULL, &SIN(dst)->sin_addr, NULL);
595 /*
596 * Only overwrite 'error' in case of error; in case of success
597 * the proper return value was already set above.
598 */
599 if (e != 0)
600 return (e);
601 return (error);
602 }
603
604 LLE_WUNLOCK(la);
605 return (error);
606 }
607
608 /*
609 * Lookups link header based on an IP address.
610 * On input:
611 * ifp is the interface we use
612 * is_gw != 0 if @dst represents gateway to some destination
613 * m is the mbuf. May be NULL if we don't have a packet.
614 * dst is the next hop,
615 * desten is the storage to put LL header.
616 * flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
617 *
618 * On success, full/partial link header and flags are filled in and
619 * the function returns 0.
620 * If the packet must be held pending resolution, we return EWOULDBLOCK
621 * On other errors, we return the corresponding error code.
622 * Note that m_freem() handles NULL.
623 */
624 int
arpresolve(struct ifnet * ifp,int is_gw,struct mbuf * m,const struct sockaddr * dst,u_char * desten,uint32_t * pflags,struct llentry ** plle)625 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
626 const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
627 struct llentry **plle)
628 {
629 struct llentry *la = NULL;
630
631 NET_EPOCH_ASSERT();
632
633 if (pflags != NULL)
634 *pflags = 0;
635 if (plle != NULL)
636 *plle = NULL;
637
638 if (m != NULL) {
639 if (m->m_flags & M_BCAST) {
640 /* broadcast */
641 (void)memcpy(desten,
642 ifp->if_broadcastaddr, ifp->if_addrlen);
643 return (0);
644 }
645 if (m->m_flags & M_MCAST) {
646 /* multicast */
647 ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
648 return (0);
649 }
650 }
651
652 la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
653 if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
654 /* Entry found, let's copy lle info */
655 bcopy(la->r_linkdata, desten, la->r_hdrlen);
656 if (pflags != NULL)
657 *pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
658 /* Notify the LLE handling code that the entry was used. */
659 llentry_mark_used(la);
660 if (plle) {
661 LLE_ADDREF(la);
662 *plle = la;
663 LLE_WUNLOCK(la);
664 }
665 return (0);
666 }
667 if (plle && la)
668 LLE_WUNLOCK(la);
669
670 return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
671 desten, pflags, plle));
672 }
673
674 /*
675 * Common length and type checks are done here,
676 * then the protocol-specific routine is called.
677 */
678 static void
arpintr(struct mbuf * m)679 arpintr(struct mbuf *m)
680 {
681 struct arphdr *ar;
682 struct ifnet *ifp;
683 char *layer;
684 int hlen;
685
686 ifp = m->m_pkthdr.rcvif;
687
688 if (m->m_len < sizeof(struct arphdr) &&
689 ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
690 ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
691 if_name(ifp));
692 return;
693 }
694 ar = mtod(m, struct arphdr *);
695
696 /* Check if length is sufficient */
697 if (m->m_len < arphdr_len(ar)) {
698 m = m_pullup(m, arphdr_len(ar));
699 if (m == NULL) {
700 ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
701 if_name(ifp));
702 return;
703 }
704 ar = mtod(m, struct arphdr *);
705 }
706
707 hlen = 0;
708 layer = "";
709 switch (ntohs(ar->ar_hrd)) {
710 case ARPHRD_ETHER:
711 hlen = ETHER_ADDR_LEN; /* RFC 826 */
712 layer = "ethernet";
713 break;
714 case ARPHRD_INFINIBAND:
715 hlen = 20; /* RFC 4391, INFINIBAND_ALEN */
716 layer = "infiniband";
717 break;
718 case ARPHRD_IEEE1394:
719 hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
720 layer = "firewire";
721
722 /*
723 * Restrict too long hardware addresses.
724 * Currently we are capable of handling 20-byte
725 * addresses ( sizeof(lle->ll_addr) )
726 */
727 if (ar->ar_hln >= 20)
728 hlen = 16;
729 break;
730 default:
731 ARP_LOG(LOG_NOTICE,
732 "packet with unknown hardware format 0x%02d received on "
733 "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
734 m_freem(m);
735 return;
736 }
737
738 if (hlen != 0 && hlen != ar->ar_hln) {
739 ARP_LOG(LOG_NOTICE,
740 "packet with invalid %s address length %d received on %s\n",
741 layer, ar->ar_hln, if_name(ifp));
742 m_freem(m);
743 return;
744 }
745
746 ARPSTAT_INC(received);
747 switch (ntohs(ar->ar_pro)) {
748 #ifdef INET
749 case ETHERTYPE_IP:
750 in_arpinput(m);
751 return;
752 #endif
753 }
754 m_freem(m);
755 }
756
757 #ifdef INET
758 /*
759 * ARP for Internet protocols on 10 Mb/s Ethernet.
760 * Algorithm is that given in RFC 826.
761 * In addition, a sanity check is performed on the sender
762 * protocol address, to catch impersonators.
763 * We no longer handle negotiations for use of trailer protocol:
764 * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
765 * along with IP replies if we wanted trailers sent to us,
766 * and also sent them in response to IP replies.
767 * This allowed either end to announce the desire to receive
768 * trailer packets.
769 * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
770 * but formerly didn't normally send requests.
771 */
772 static int log_arp_wrong_iface = 1;
773 static int log_arp_movements = 1;
774 static int log_arp_permanent_modify = 1;
775 static int allow_multicast = 0;
776
777 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
778 &log_arp_wrong_iface, 0,
779 "log arp packets arriving on the wrong interface");
780 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
781 &log_arp_movements, 0,
782 "log arp replies from MACs different than the one in the cache");
783 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
784 &log_arp_permanent_modify, 0,
785 "log arp replies from MACs different than the one in the permanent arp entry");
786 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
787 &allow_multicast, 0, "accept multicast addresses");
788
789 #pragma GCC diagnostic ignored "-Wformat"
790 #pragma GCC diagnostic ignored "-Wformat-extra-args"
791 static void
in_arpinput(struct mbuf * m)792 in_arpinput(struct mbuf *m)
793 {
794 struct rm_priotracker in_ifa_tracker;
795 struct arphdr *ah;
796 struct ifnet *ifp = m->m_pkthdr.rcvif;
797 struct llentry *la = NULL, *la_tmp;
798 struct ifaddr *ifa;
799 struct in_ifaddr *ia;
800 struct sockaddr sa;
801 struct in_addr isaddr, itaddr, myaddr;
802 u_int8_t *enaddr = NULL;
803 int op;
804 int bridged = 0, is_bridge = 0;
805 int carped;
806 struct sockaddr_in sin;
807 struct sockaddr *dst;
808 struct nhop_object *nh;
809 uint8_t linkhdr[LLE_MAX_LINKHDR];
810 struct route ro;
811 size_t linkhdrsize;
812 int lladdr_off;
813 int error;
814 char addrbuf[INET_ADDRSTRLEN];
815
816 NET_EPOCH_ASSERT();
817
818 sin.sin_len = sizeof(struct sockaddr_in);
819 sin.sin_family = AF_INET;
820 sin.sin_addr.s_addr = 0;
821
822 if (ifp->if_bridge)
823 bridged = 1;
824 if (ifp->if_type == IFT_BRIDGE)
825 is_bridge = 1;
826
827 /*
828 * We already have checked that mbuf contains enough contiguous data
829 * to hold entire arp message according to the arp header.
830 */
831 ah = mtod(m, struct arphdr *);
832
833 /*
834 * ARP is only for IPv4 so we can reject packets with
835 * a protocol length not equal to an IPv4 address.
836 */
837 if (ah->ar_pln != sizeof(struct in_addr)) {
838 ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
839 sizeof(struct in_addr));
840 goto drop;
841 }
842
843 if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
844 ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
845 ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
846 goto drop;
847 }
848
849 op = ntohs(ah->ar_op);
850 (void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
851 (void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
852
853 if (op == ARPOP_REPLY)
854 ARPSTAT_INC(rxreplies);
855
856 /*
857 * For a bridge, we want to check the address irrespective
858 * of the receive interface. (This will change slightly
859 * when we have clusters of interfaces).
860 */
861 IN_IFADDR_RLOCK(&in_ifa_tracker);
862 LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
863 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
864 ia->ia_ifp == ifp) &&
865 itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
866 (ia->ia_ifa.ifa_carp == NULL ||
867 (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
868 ifa_ref(&ia->ia_ifa);
869 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
870 goto match;
871 }
872 }
873 LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
874 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
875 ia->ia_ifp == ifp) &&
876 isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
877 ifa_ref(&ia->ia_ifa);
878 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
879 goto match;
880 }
881
882 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia) \
883 (ia->ia_ifp->if_bridge == ifp->if_softc && \
884 !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) && \
885 addr == ia->ia_addr.sin_addr.s_addr)
886 /*
887 * Check the case when bridge shares its MAC address with
888 * some of its children, so packets are claimed by bridge
889 * itself (bridge_input() does it first), but they are really
890 * meant to be destined to the bridge member.
891 */
892 if (is_bridge) {
893 LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
894 if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
895 ifa_ref(&ia->ia_ifa);
896 ifp = ia->ia_ifp;
897 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
898 goto match;
899 }
900 }
901 }
902 #undef BDG_MEMBER_MATCHES_ARP
903 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
904
905 /*
906 * No match, use the first inet address on the receive interface
907 * as a dummy address for the rest of the function.
908 */
909 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
910 if (ifa->ifa_addr->sa_family == AF_INET &&
911 (ifa->ifa_carp == NULL ||
912 (*carp_iamatch_p)(ifa, &enaddr))) {
913 ia = ifatoia(ifa);
914 ifa_ref(ifa);
915 goto match;
916 }
917
918 /*
919 * If bridging, fall back to using any inet address.
920 */
921 IN_IFADDR_RLOCK(&in_ifa_tracker);
922 if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL) {
923 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
924 goto drop;
925 }
926 ifa_ref(&ia->ia_ifa);
927 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
928 match:
929 if (!enaddr)
930 enaddr = (u_int8_t *)IF_LLADDR(ifp);
931 carped = (ia->ia_ifa.ifa_carp != NULL);
932 myaddr = ia->ia_addr.sin_addr;
933 ifa_free(&ia->ia_ifa);
934 if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
935 goto drop; /* it's from me, ignore it. */
936 if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
937 ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
938 "%s!\n", inet_ntoa_r(isaddr, addrbuf));
939 goto drop;
940 }
941
942 if (ifp->if_addrlen != ah->ar_hln) {
943 ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
944 "i/f %d (ignored)\n", ifp->if_addrlen,
945 (u_char *) ar_sha(ah), ":", ah->ar_hln,
946 ifp->if_addrlen);
947 goto drop;
948 }
949
950 /*
951 * Warn if another host is using the same IP address, but only if the
952 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
953 * case we suppress the warning to avoid false positive complaints of
954 * potential misconfiguration.
955 */
956 if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
957 myaddr.s_addr != 0) {
958 ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
959 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
960 inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
961 itaddr = myaddr;
962 ARPSTAT_INC(dupips);
963 goto reply;
964 }
965 if (ifp->if_flags & IFF_STATICARP)
966 goto reply;
967
968 bzero(&sin, sizeof(sin));
969 sin.sin_len = sizeof(struct sockaddr_in);
970 sin.sin_family = AF_INET;
971 sin.sin_addr = isaddr;
972 dst = (struct sockaddr *)&sin;
973 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
974 if (la != NULL)
975 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
976 else if (itaddr.s_addr == myaddr.s_addr) {
977 /*
978 * Request/reply to our address, but no lle exists yet.
979 * Calculate full link prepend to use in lle.
980 */
981 linkhdrsize = sizeof(linkhdr);
982 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
983 &linkhdrsize, &lladdr_off) != 0)
984 goto reply;
985
986 /* Allocate new entry */
987 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
988 if (la == NULL) {
989 /*
990 * lle creation may fail if source address belongs
991 * to non-directly connected subnet. However, we
992 * will try to answer the request instead of dropping
993 * frame.
994 */
995 goto reply;
996 }
997 lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
998 lladdr_off);
999
1000 IF_AFDATA_WLOCK(ifp);
1001 LLE_WLOCK(la);
1002 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1003
1004 /*
1005 * Check if lle still does not exists.
1006 * If it does, that means that we either
1007 * 1) have configured it explicitly, via
1008 * 1a) 'arp -s' static entry or
1009 * 1b) interface address static record
1010 * or
1011 * 2) it was the result of sending first packet to-host
1012 * or
1013 * 3) it was another arp reply packet we handled in
1014 * different thread.
1015 *
1016 * In all cases except 3) we definitely need to prefer
1017 * existing lle. For the sake of simplicity, prefer any
1018 * existing lle over newly-create one.
1019 */
1020 if (la_tmp == NULL)
1021 lltable_link_entry(LLTABLE(ifp), la);
1022 IF_AFDATA_WUNLOCK(ifp);
1023
1024 if (la_tmp == NULL) {
1025 arp_mark_lle_reachable(la);
1026 LLE_WUNLOCK(la);
1027 } else {
1028 /* Free newly-create entry and handle packet */
1029 lltable_free_entry(LLTABLE(ifp), la);
1030 la = la_tmp;
1031 la_tmp = NULL;
1032 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1033 /* arp_check_update_lle() returns @la unlocked */
1034 }
1035 la = NULL;
1036 }
1037 reply:
1038 if (op != ARPOP_REQUEST)
1039 goto drop;
1040 ARPSTAT_INC(rxrequests);
1041
1042 if (itaddr.s_addr == myaddr.s_addr) {
1043 /* Shortcut.. the receiving interface is the target. */
1044 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1045 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1046 } else {
1047 /*
1048 * Destination address is not ours. Check if
1049 * proxyarp entry exists or proxyarp is turned on globally.
1050 */
1051 struct llentry *lle;
1052
1053 sin.sin_addr = itaddr;
1054 lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1055
1056 if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1057 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1058 (void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1059 LLE_RUNLOCK(lle);
1060 } else {
1061 if (lle != NULL)
1062 LLE_RUNLOCK(lle);
1063
1064 if (!V_arp_proxyall)
1065 goto drop;
1066
1067 NET_EPOCH_ASSERT();
1068 nh = fib4_lookup(ifp->if_fib, itaddr, 0, 0, 0);
1069 if (nh == NULL)
1070 goto drop;
1071
1072 /*
1073 * Don't send proxies for nodes on the same interface
1074 * as this one came out of, or we'll get into a fight
1075 * over who claims what Ether address.
1076 */
1077 if (nh->nh_ifp == ifp)
1078 goto drop;
1079
1080 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1081 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1082
1083 /*
1084 * Also check that the node which sent the ARP packet
1085 * is on the interface we expect it to be on. This
1086 * avoids ARP chaos if an interface is connected to the
1087 * wrong network.
1088 */
1089
1090 nh = fib4_lookup(ifp->if_fib, isaddr, 0, 0, 0);
1091 if (nh == NULL)
1092 goto drop;
1093 if (nh->nh_ifp != ifp) {
1094 ARP_LOG(LOG_INFO, "proxy: ignoring request"
1095 " from %s via %s\n",
1096 inet_ntoa_r(isaddr, addrbuf),
1097 ifp->if_xname);
1098 goto drop;
1099 }
1100
1101 #ifdef DEBUG_PROXY
1102 printf("arp: proxying for %s\n",
1103 inet_ntoa_r(itaddr, addrbuf));
1104 #endif
1105 }
1106 }
1107
1108 if (itaddr.s_addr == myaddr.s_addr &&
1109 IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1110 /* RFC 3927 link-local IPv4; always reply by broadcast. */
1111 #ifdef DEBUG_LINKLOCAL
1112 printf("arp: sending reply for link-local addr %s\n",
1113 inet_ntoa_r(itaddr, addrbuf));
1114 #endif
1115 m->m_flags |= M_BCAST;
1116 m->m_flags &= ~M_MCAST;
1117 } else {
1118 /* default behaviour; never reply by broadcast. */
1119 m->m_flags &= ~(M_BCAST|M_MCAST);
1120 }
1121 (void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1122 (void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1123 ah->ar_op = htons(ARPOP_REPLY);
1124 ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1125 m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1126 m->m_pkthdr.len = m->m_len;
1127 m->m_pkthdr.rcvif = NULL;
1128 sa.sa_family = AF_ARP;
1129 sa.sa_len = 2;
1130
1131 /* Calculate link header for sending frame */
1132 bzero(&ro, sizeof(ro));
1133 linkhdrsize = sizeof(linkhdr);
1134 error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1135
1136 /*
1137 * arp_fillheader() may fail due to lack of support inside encap request
1138 * routing. This is not necessary an error, AF_ARP can/should be handled
1139 * by if_output().
1140 */
1141 if (error != 0 && error != EAFNOSUPPORT) {
1142 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1143 if_name(ifp), error);
1144 return;
1145 }
1146
1147 ro.ro_prepend = linkhdr;
1148 ro.ro_plen = linkhdrsize;
1149 ro.ro_flags = 0;
1150
1151 m_clrprotoflags(m); /* Avoid confusing lower layers. */
1152 (*ifp->if_output)(ifp, m, &sa, &ro);
1153 ARPSTAT_INC(txreplies);
1154 return;
1155
1156 drop:
1157 m_freem(m);
1158 }
1159 #endif
1160
1161 /*
1162 * Checks received arp data against existing @la.
1163 * Updates lle state/performs notification if necessary.
1164 */
1165 static void
arp_check_update_lle(struct arphdr * ah,struct in_addr isaddr,struct ifnet * ifp,int bridged,struct llentry * la)1166 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1167 int bridged, struct llentry *la)
1168 {
1169 struct sockaddr sa;
1170 struct mbuf *m_hold, *m_hold_next;
1171 uint8_t linkhdr[LLE_MAX_LINKHDR];
1172 size_t linkhdrsize;
1173 int lladdr_off;
1174 char addrbuf[INET_ADDRSTRLEN];
1175
1176 LLE_WLOCK_ASSERT(la);
1177
1178 /* the following is not an error when doing bridging */
1179 if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1180 if (log_arp_wrong_iface)
1181 ARP_LOG(LOG_WARNING, "%s is on %s "
1182 "but got reply from %*D on %s\n",
1183 inet_ntoa_r(isaddr, addrbuf),
1184 la->lle_tbl->llt_ifp->if_xname,
1185 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1186 ifp->if_xname);
1187 LLE_WUNLOCK(la);
1188 return;
1189 }
1190 if ((la->la_flags & LLE_VALID) &&
1191 bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1192 if (la->la_flags & LLE_STATIC) {
1193 LLE_WUNLOCK(la);
1194 if (log_arp_permanent_modify)
1195 ARP_LOG(LOG_ERR,
1196 "%*D attempts to modify "
1197 "permanent entry for %s on %s\n",
1198 ifp->if_addrlen,
1199 (u_char *)ar_sha(ah), ":",
1200 inet_ntoa_r(isaddr, addrbuf),
1201 ifp->if_xname);
1202 return;
1203 }
1204 if (log_arp_movements) {
1205 ARP_LOG(LOG_INFO, "%s moved from %*D "
1206 "to %*D on %s\n",
1207 inet_ntoa_r(isaddr, addrbuf),
1208 ifp->if_addrlen,
1209 (u_char *)la->ll_addr, ":",
1210 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1211 ifp->if_xname);
1212 }
1213 }
1214
1215 /* Calculate full link prepend to use in lle */
1216 linkhdrsize = sizeof(linkhdr);
1217 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1218 &linkhdrsize, &lladdr_off) != 0)
1219 return;
1220
1221 /* Check if something has changed */
1222 if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1223 (la->la_flags & LLE_VALID) == 0) {
1224 /* Try to perform LLE update */
1225 if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1226 lladdr_off) == 0)
1227 return;
1228
1229 /* Clear fast path feedback request if set */
1230 la->r_skip_req = 0;
1231 }
1232
1233 arp_mark_lle_reachable(la);
1234
1235 /*
1236 * The packets are all freed within the call to the output
1237 * routine.
1238 *
1239 * NB: The lock MUST be released before the call to the
1240 * output routine.
1241 */
1242 if (la->la_hold != NULL) {
1243 m_hold = la->la_hold;
1244 la->la_hold = NULL;
1245 la->la_numheld = 0;
1246 lltable_fill_sa_entry(la, &sa);
1247 LLE_WUNLOCK(la);
1248 for (; m_hold != NULL; m_hold = m_hold_next) {
1249 m_hold_next = m_hold->m_nextpkt;
1250 m_hold->m_nextpkt = NULL;
1251 /* Avoid confusing lower layers. */
1252 m_clrprotoflags(m_hold);
1253 (*ifp->if_output)(ifp, m_hold, &sa, NULL);
1254 }
1255 } else
1256 LLE_WUNLOCK(la);
1257 }
1258 #pragma GCC diagnostic error "-Wformat"
1259 #pragma GCC diagnostic error "-Wformat-extra-args"
1260
1261 static void
arp_mark_lle_reachable(struct llentry * la)1262 arp_mark_lle_reachable(struct llentry *la)
1263 {
1264 int canceled, wtime;
1265
1266 LLE_WLOCK_ASSERT(la);
1267
1268 la->ln_state = ARP_LLINFO_REACHABLE;
1269 EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1270
1271 if (!(la->la_flags & LLE_STATIC)) {
1272 LLE_ADDREF(la);
1273 la->la_expire = time_uptime + V_arpt_keep;
1274 wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1275 if (wtime < 0)
1276 wtime = V_arpt_keep;
1277 canceled = callout_reset(&la->lle_timer,
1278 hz * wtime, arptimer, la);
1279 if (canceled)
1280 LLE_REMREF(la);
1281 }
1282 la->la_asked = 0;
1283 la->la_preempt = V_arp_maxtries;
1284 }
1285
1286 /*
1287 * Add permanent link-layer record for given interface address.
1288 */
1289 static __noinline void
arp_add_ifa_lle(struct ifnet * ifp,const struct sockaddr * dst)1290 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1291 {
1292 struct llentry *lle, *lle_tmp;
1293
1294 /*
1295 * Interface address LLE record is considered static
1296 * because kernel code relies on LLE_STATIC flag to check
1297 * if these entries can be rewriten by arp updates.
1298 */
1299 lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1300 if (lle == NULL) {
1301 log(LOG_INFO, "arp_ifinit: cannot create arp "
1302 "entry for interface address\n");
1303 return;
1304 }
1305
1306 IF_AFDATA_WLOCK(ifp);
1307 LLE_WLOCK(lle);
1308 /* Unlink any entry if exists */
1309 lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1310 if (lle_tmp != NULL)
1311 lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1312
1313 lltable_link_entry(LLTABLE(ifp), lle);
1314 IF_AFDATA_WUNLOCK(ifp);
1315
1316 if (lle_tmp != NULL)
1317 EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1318
1319 EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1320 LLE_WUNLOCK(lle);
1321 if (lle_tmp != NULL)
1322 lltable_free_entry(LLTABLE(ifp), lle_tmp);
1323 }
1324
1325 /*
1326 * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1327 * of valid values.
1328 */
1329 static int
sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)1330 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1331 {
1332 int error;
1333 int rexmit_count = *(int *)arg1;
1334
1335 error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1336
1337 /* Enforce limits on any new value that may have been set. */
1338 if (!error && req->newptr) {
1339 /* A new value was set. */
1340 if (rexmit_count < 0) {
1341 rexmit_count = 0;
1342 } else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1343 rexmit_count = MAX_GARP_RETRANSMITS;
1344 }
1345 *(int *)arg1 = rexmit_count;
1346 }
1347
1348 return (error);
1349 }
1350
1351 /*
1352 * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1353 * retransmit it again. A pending callout owns a reference to the ifa.
1354 */
1355 static void
garp_rexmit(void * arg)1356 garp_rexmit(void *arg)
1357 {
1358 struct in_ifaddr *ia = arg;
1359
1360 if (callout_pending(&ia->ia_garp_timer) ||
1361 !callout_active(&ia->ia_garp_timer)) {
1362 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1363 ifa_free(&ia->ia_ifa);
1364 return;
1365 }
1366
1367 CURVNET_SET(ia->ia_ifa.ifa_ifp->if_vnet);
1368
1369 /*
1370 * Drop lock while the ARP request is generated.
1371 */
1372 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1373
1374 arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1375 &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1376
1377 /*
1378 * Increment the count of retransmissions. If the count has reached the
1379 * maximum value, stop sending the GARP packets. Otherwise, schedule
1380 * the callout to retransmit another GARP packet.
1381 */
1382 ++ia->ia_garp_count;
1383 if (ia->ia_garp_count >= garp_rexmit_count) {
1384 ifa_free(&ia->ia_ifa);
1385 } else {
1386 int rescheduled;
1387 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1388 rescheduled = callout_reset(&ia->ia_garp_timer,
1389 (1 << ia->ia_garp_count) * hz,
1390 garp_rexmit, ia);
1391 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1392 if (rescheduled) {
1393 ifa_free(&ia->ia_ifa);
1394 }
1395 }
1396
1397 CURVNET_RESTORE();
1398 }
1399
1400 /*
1401 * Start the GARP retransmit timer.
1402 *
1403 * A single GARP is always transmitted when an IPv4 address is added
1404 * to an interface and that is usually sufficient. However, in some
1405 * circumstances, such as when a shared address is passed between
1406 * cluster nodes, this single GARP may occasionally be dropped or
1407 * lost. This can lead to neighbors on the network link working with a
1408 * stale ARP cache and sending packets destined for that address to
1409 * the node that previously owned the address, which may not respond.
1410 *
1411 * To avoid this situation, GARP retransmits can be enabled by setting
1412 * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1413 * than zero. The setting represents the maximum number of
1414 * retransmissions. The interval between retransmissions is calculated
1415 * using an exponential backoff algorithm, doubling each time, so the
1416 * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1417 */
1418 static void
garp_timer_start(struct ifaddr * ifa)1419 garp_timer_start(struct ifaddr *ifa)
1420 {
1421 struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1422
1423 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1424 ia->ia_garp_count = 0;
1425 if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1426 garp_rexmit, ia) == 0) {
1427 ifa_ref(ifa);
1428 }
1429 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1430 }
1431
1432 void
arp_ifinit(struct ifnet * ifp,struct ifaddr * ifa)1433 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1434 {
1435 struct epoch_tracker et;
1436 const struct sockaddr_in *dst_in;
1437 const struct sockaddr *dst;
1438
1439 if (ifa->ifa_carp != NULL)
1440 return;
1441
1442 dst = ifa->ifa_addr;
1443 dst_in = (const struct sockaddr_in *)dst;
1444
1445 if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1446 return;
1447 NET_EPOCH_ENTER(et);
1448 arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1449 NET_EPOCH_EXIT(et);
1450 if (garp_rexmit_count > 0) {
1451 garp_timer_start(ifa);
1452 }
1453
1454 arp_add_ifa_lle(ifp, dst);
1455 }
1456
1457 void
arp_announce_ifaddr(struct ifnet * ifp,struct in_addr addr,u_char * enaddr)1458 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1459 {
1460
1461 if (ntohl(addr.s_addr) != INADDR_ANY)
1462 arprequest(ifp, &addr, &addr, enaddr);
1463 }
1464
1465 /*
1466 * Sends gratuitous ARPs for each ifaddr to notify other
1467 * nodes about the address change.
1468 */
1469 static __noinline void
arp_handle_ifllchange(struct ifnet * ifp)1470 arp_handle_ifllchange(struct ifnet *ifp)
1471 {
1472 struct ifaddr *ifa;
1473
1474 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1475 if (ifa->ifa_addr->sa_family == AF_INET)
1476 arp_ifinit(ifp, ifa);
1477 }
1478 }
1479
1480 /*
1481 * A handler for interface link layer address change event.
1482 */
1483 static void
arp_iflladdr(void * arg __unused,struct ifnet * ifp)1484 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1485 {
1486
1487 lltable_update_ifaddr(LLTABLE(ifp));
1488
1489 if ((ifp->if_flags & IFF_UP) != 0)
1490 arp_handle_ifllchange(ifp);
1491 }
1492
1493 static void
vnet_arp_init(void)1494 vnet_arp_init(void)
1495 {
1496
1497 if (IS_DEFAULT_VNET(curvnet)) {
1498 netisr_register(&arp_nh);
1499 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1500 arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1501 }
1502 #ifdef VIMAGE
1503 else
1504 netisr_register_vnet(&arp_nh);
1505 #endif
1506 }
1507 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1508 vnet_arp_init, 0);
1509
1510 #ifdef VIMAGE
1511 /*
1512 * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1513 * lookups after destroying the hash. Ideally this would go on SI_ORDER_3.5.
1514 */
1515 static void
vnet_arp_destroy(__unused void * arg)1516 vnet_arp_destroy(__unused void *arg)
1517 {
1518
1519 netisr_unregister_vnet(&arp_nh);
1520 }
1521 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1522 vnet_arp_destroy, NULL);
1523 #endif
1524