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/if_bridgevar.h>
60 #include <net/netisr.h>
61 #include <net/ethernet.h>
62 #include <net/route.h>
63 #include <net/route/nhop.h>
64 #include <net/vnet.h>
65
66 #include <netinet/in.h>
67 #include <netinet/in_fib.h>
68 #include <netinet/in_var.h>
69 #include <net/if_llatbl.h>
70 #include <netinet/if_ether.h>
71 #ifdef INET
72 #include <netinet/ip_carp.h>
73 #endif
74
75 #include <security/mac/mac_framework.h>
76
77 #define SIN(s) ((const struct sockaddr_in *)(s))
78
79 static struct timeval arp_lastlog;
80 static int arp_curpps;
81 static int arp_maxpps = 1;
82
83 /* Simple ARP state machine */
84 enum arp_llinfo_state {
85 ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */
86 ARP_LLINFO_REACHABLE, /* LLE is valid */
87 ARP_LLINFO_VERIFY, /* LLE is valid, need refresh */
88 ARP_LLINFO_DELETED, /* LLE is deleted */
89 };
90
91 SYSCTL_DECL(_net_link_ether);
92 static SYSCTL_NODE(_net_link_ether, PF_INET, inet,
93 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
94 "");
95 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp,
96 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
97 "");
98
99 /* timer values */
100 VNET_DEFINE_STATIC(int, arpt_keep) = (20*60); /* once resolved, good for 20
101 * minutes */
102 VNET_DEFINE_STATIC(int, arp_maxtries) = 5;
103 VNET_DEFINE_STATIC(int, arp_proxyall) = 0;
104 VNET_DEFINE_STATIC(int, arpt_down) = 20; /* keep incomplete entries for
105 * 20 seconds */
106 VNET_DEFINE_STATIC(int, arpt_rexmit) = 1; /* retransmit arp entries, sec*/
107 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat); /* ARP statistics, see if_arp.h */
108 VNET_PCPUSTAT_SYSINIT(arpstat);
109
110 #ifdef VIMAGE
111 VNET_PCPUSTAT_SYSUNINIT(arpstat);
112 #endif /* VIMAGE */
113
114 VNET_DEFINE_STATIC(int, arp_maxhold) = 16;
115
116 #define V_arpt_keep VNET(arpt_keep)
117 #define V_arpt_down VNET(arpt_down)
118 #define V_arpt_rexmit VNET(arpt_rexmit)
119 #define V_arp_maxtries VNET(arp_maxtries)
120 #define V_arp_proxyall VNET(arp_proxyall)
121 #define V_arp_maxhold VNET(arp_maxhold)
122
123 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
124 &VNET_NAME(arpt_keep), 0,
125 "ARP entry lifetime in seconds");
126 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
127 &VNET_NAME(arp_maxtries), 0,
128 "ARP resolution attempts before returning error");
129 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
130 &VNET_NAME(arp_proxyall), 0,
131 "Enable proxy ARP for all suitable requests");
132 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
133 &VNET_NAME(arpt_down), 0,
134 "Incomplete ARP entry lifetime in seconds");
135 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
136 arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
137 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
138 &VNET_NAME(arp_maxhold), 0,
139 "Number of packets to hold per ARP entry");
140 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
141 CTLFLAG_RW, &arp_maxpps, 0,
142 "Maximum number of remotely triggered ARP messages that can be "
143 "logged per second");
144
145 /*
146 * Due to the exponential backoff algorithm used for the interval between GARP
147 * retransmissions, the maximum number of retransmissions is limited for
148 * sanity. This limit corresponds to a maximum interval between retransmissions
149 * of 2^16 seconds ~= 18 hours.
150 *
151 * Making this limit more dynamic is more complicated than worthwhile,
152 * especially since sending out GARPs spaced days apart would be of little
153 * use. A maximum dynamic limit would look something like:
154 *
155 * const int max = fls(INT_MAX / hz) - 1;
156 */
157 #define MAX_GARP_RETRANSMITS 16
158 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS);
159 VNET_DEFINE_STATIC(int, garp_rexmit_count) = 0; /* GARP retransmission setting. */
160 #define V_garp_rexmit_count VNET(garp_rexmit_count)
161
162 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count,
163 CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE|CTLFLAG_VNET,
164 &VNET_NAME(garp_rexmit_count), 0, sysctl_garp_rexmit, "I",
165 "Number of times to retransmit GARP packets;"
166 " 0 to disable, maximum of 16");
167
168 VNET_DEFINE_STATIC(int, arp_log_level) = LOG_INFO; /* Min. log(9) level. */
169 #define V_arp_log_level VNET(arp_log_level)
170 SYSCTL_INT(_net_link_ether_arp, OID_AUTO, log_level, CTLFLAG_VNET | CTLFLAG_RW,
171 &VNET_NAME(arp_log_level), 0,
172 "Minimum log(9) level for recording rate limited arp log messages. "
173 "The higher will be log more (emerg=0, info=6 (default), debug=7).");
174 #define ARP_LOG(pri, ...) do { \
175 if ((pri) <= V_arp_log_level && \
176 ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps)) \
177 log((pri), "arp: " __VA_ARGS__); \
178 } while (0)
179
180 static void arpintr(struct mbuf *);
181 static void arptimer(void *);
182 #ifdef INET
183 static void in_arpinput(struct mbuf *);
184 #endif
185
186 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
187 struct ifnet *ifp, int bridged, struct llentry *la);
188 static void arp_mark_lle_reachable(struct llentry *la, struct ifnet *ifp);
189 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
190
191 static eventhandler_tag iflladdr_tag;
192
193 static const struct netisr_handler arp_nh = {
194 .nh_name = "arp",
195 .nh_handler = arpintr,
196 .nh_proto = NETISR_ARP,
197 .nh_policy = NETISR_POLICY_SOURCE,
198 };
199
200 /*
201 * Timeout routine. Age arp_tab entries periodically.
202 */
203 static void
arptimer(void * arg)204 arptimer(void *arg)
205 {
206 struct llentry *lle = (struct llentry *)arg;
207 struct ifnet *ifp;
208
209 if (lle->la_flags & LLE_STATIC) {
210 return;
211 }
212 LLE_WLOCK(lle);
213 if (callout_pending(&lle->lle_timer)) {
214 /*
215 * Here we are a bit odd here in the treatment of
216 * active/pending. If the pending bit is set, it got
217 * rescheduled before I ran. The active
218 * bit we ignore, since if it was stopped
219 * in ll_tablefree() and was currently running
220 * it would have return 0 so the code would
221 * not have deleted it since the callout could
222 * not be stopped so we want to go through
223 * with the delete here now. If the callout
224 * was restarted, the pending bit will be back on and
225 * we just want to bail since the callout_reset would
226 * return 1 and our reference would have been removed
227 * by arpresolve() below.
228 */
229 LLE_WUNLOCK(lle);
230 return;
231 }
232 ifp = lle->lle_tbl->llt_ifp;
233 CURVNET_SET(ifp->if_vnet);
234
235 switch (lle->ln_state) {
236 case ARP_LLINFO_REACHABLE:
237
238 /*
239 * Expiration time is approaching.
240 * Request usage feedback from the datapath.
241 * Change state and re-schedule ourselves.
242 */
243 llentry_request_feedback(lle);
244 lle->ln_state = ARP_LLINFO_VERIFY;
245 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
246 LLE_WUNLOCK(lle);
247 CURVNET_RESTORE();
248 return;
249 case ARP_LLINFO_VERIFY:
250 if (llentry_get_hittime(lle) > 0 && lle->la_preempt > 0) {
251 /* Entry was used, issue refresh request */
252 struct epoch_tracker et;
253 struct in_addr dst;
254
255 dst = lle->r_l3addr.addr4;
256 lle->la_preempt--;
257 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
258 LLE_WUNLOCK(lle);
259 NET_EPOCH_ENTER(et);
260 arprequest(ifp, NULL, &dst, NULL);
261 NET_EPOCH_EXIT(et);
262 CURVNET_RESTORE();
263 return;
264 }
265 /* Nothing happened. Reschedule if not too late */
266 if (lle->la_expire > time_uptime) {
267 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
268 LLE_WUNLOCK(lle);
269 CURVNET_RESTORE();
270 return;
271 }
272 break;
273 case ARP_LLINFO_INCOMPLETE:
274 case ARP_LLINFO_DELETED:
275 break;
276 }
277
278 if ((lle->la_flags & LLE_DELETED) == 0) {
279 int evt;
280
281 if (lle->la_flags & LLE_VALID)
282 evt = LLENTRY_EXPIRED;
283 else
284 evt = LLENTRY_TIMEDOUT;
285 EVENTHANDLER_INVOKE(lle_event, lle, evt);
286 }
287
288 callout_stop(&lle->lle_timer);
289
290 /* XXX: LOR avoidance. We still have ref on lle. */
291 LLE_WUNLOCK(lle);
292 IF_AFDATA_LOCK(ifp);
293 LLE_WLOCK(lle);
294
295 /* Guard against race with other llentry_free(). */
296 if (lle->la_flags & LLE_LINKED) {
297 LLE_REMREF(lle);
298 lltable_unlink_entry(lle->lle_tbl, lle);
299 }
300 IF_AFDATA_UNLOCK(ifp);
301
302 size_t pkts_dropped = llentry_free(lle);
303
304 ARPSTAT_ADD(dropped, pkts_dropped);
305 ARPSTAT_INC(timeouts);
306
307 CURVNET_RESTORE();
308 }
309
310 /*
311 * Stores link-layer header for @ifp in format suitable for if_output()
312 * into buffer @buf. Resulting header length is stored in @bufsize.
313 *
314 * Returns 0 on success.
315 */
316 static int
arp_fillheader(struct ifnet * ifp,struct arphdr * ah,int bcast,u_char * buf,size_t * bufsize)317 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
318 size_t *bufsize)
319 {
320 struct if_encap_req ereq;
321 int error;
322
323 bzero(buf, *bufsize);
324 bzero(&ereq, sizeof(ereq));
325 ereq.buf = buf;
326 ereq.bufsize = *bufsize;
327 ereq.rtype = IFENCAP_LL;
328 ereq.family = AF_ARP;
329 ereq.lladdr = ar_tha(ah);
330 ereq.hdata = (u_char *)ah;
331 if (bcast)
332 ereq.flags = IFENCAP_FLAG_BROADCAST;
333 error = ifp->if_requestencap(ifp, &ereq);
334 if (error == 0)
335 *bufsize = ereq.bufsize;
336
337 return (error);
338 }
339
340 /*
341 * Broadcast an ARP request. Caller specifies:
342 * - arp header source ip address
343 * - arp header target ip address
344 * - arp header source ethernet address
345 */
346 static int
arprequest_internal(struct ifnet * ifp,const struct in_addr * sip,const struct in_addr * tip,u_char * enaddr)347 arprequest_internal(struct ifnet *ifp, const struct in_addr *sip,
348 const struct in_addr *tip, u_char *enaddr)
349 {
350 struct mbuf *m;
351 struct arphdr *ah;
352 struct sockaddr sa;
353 u_char *carpaddr = NULL;
354 uint8_t linkhdr[LLE_MAX_LINKHDR];
355 size_t linkhdrsize;
356 struct route ro;
357 int error;
358
359 NET_EPOCH_ASSERT();
360
361 if (sip == NULL) {
362 /*
363 * The caller did not supply a source address, try to find
364 * a compatible one among those assigned to this interface.
365 */
366 struct ifaddr *ifa;
367
368 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
369 if (ifa->ifa_addr->sa_family != AF_INET)
370 continue;
371
372 if (ifa->ifa_carp) {
373 if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
374 continue;
375 sip = &IA_SIN(ifa)->sin_addr;
376 } else {
377 carpaddr = NULL;
378 sip = &IA_SIN(ifa)->sin_addr;
379 }
380
381 if (0 == ((sip->s_addr ^ tip->s_addr) &
382 IA_MASKSIN(ifa)->sin_addr.s_addr))
383 break; /* found it. */
384 }
385 if (sip == NULL) {
386 printf("%s: cannot find matching address\n", __func__);
387 return (EADDRNOTAVAIL);
388 }
389 }
390 if (enaddr == NULL)
391 enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
392
393 if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
394 return (ENOMEM);
395 m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
396 2 * ifp->if_addrlen;
397 m->m_pkthdr.len = m->m_len;
398 M_ALIGN(m, m->m_len);
399 ah = mtod(m, struct arphdr *);
400 bzero((caddr_t)ah, m->m_len);
401 #ifdef MAC
402 mac_netinet_arp_send(ifp, m);
403 #endif
404 ah->ar_pro = htons(ETHERTYPE_IP);
405 ah->ar_hln = ifp->if_addrlen; /* hardware address length */
406 ah->ar_pln = sizeof(struct in_addr); /* protocol address length */
407 ah->ar_op = htons(ARPOP_REQUEST);
408 bcopy(enaddr, ar_sha(ah), ah->ar_hln);
409 bcopy(sip, ar_spa(ah), ah->ar_pln);
410 bcopy(tip, ar_tpa(ah), ah->ar_pln);
411 sa.sa_family = AF_ARP;
412 sa.sa_len = 2;
413
414 /* Calculate link header for sending frame */
415 bzero(&ro, sizeof(ro));
416 linkhdrsize = sizeof(linkhdr);
417 error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
418 if (error != 0 && error != EAFNOSUPPORT) {
419 m_freem(m);
420 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
421 if_name(ifp), error);
422 return (error);
423 }
424
425 ro.ro_prepend = linkhdr;
426 ro.ro_plen = linkhdrsize;
427 ro.ro_flags = 0;
428
429 m->m_flags |= M_BCAST;
430 m_clrprotoflags(m); /* Avoid confusing lower layers. */
431 error = (*ifp->if_output)(ifp, m, &sa, &ro);
432 ARPSTAT_INC(txrequests);
433 if (error) {
434 ARPSTAT_INC(txerrors);
435 ARP_LOG(LOG_DEBUG, "Failed to send ARP packet on %s: %d\n",
436 if_name(ifp), error);
437 }
438 return (error);
439 }
440
441 void
arprequest(struct ifnet * ifp,const struct in_addr * sip,const struct in_addr * tip,u_char * enaddr)442 arprequest(struct ifnet *ifp, const struct in_addr *sip,
443 const struct in_addr *tip, u_char *enaddr)
444 {
445
446 (void) arprequest_internal(ifp, sip, tip, enaddr);
447 }
448
449 /*
450 * Resolve an IP address into an ethernet address - heavy version.
451 * Used internally by arpresolve().
452 * We have already checked that we can't use an existing lle without
453 * modification so we have to acquire an LLE_EXCLUSIVE lle lock.
454 *
455 * On success, desten and pflags are filled in and the function returns 0;
456 * If the packet must be held pending resolution, we return EWOULDBLOCK
457 * On other errors, we return the corresponding error code.
458 * Note that m_freem() handles NULL.
459 */
460 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)461 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
462 const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
463 struct llentry **plle)
464 {
465 struct llentry *la = NULL, *la_tmp;
466 int error, renew;
467 char *lladdr;
468 int ll_len;
469
470 NET_EPOCH_ASSERT();
471
472 if (pflags != NULL)
473 *pflags = 0;
474 if (plle != NULL)
475 *plle = NULL;
476
477 if ((flags & LLE_CREATE) == 0)
478 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
479 if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
480 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
481 if (la == NULL) {
482 char addrbuf[INET_ADDRSTRLEN];
483
484 log(LOG_DEBUG,
485 "arpresolve: can't allocate llinfo for %s on %s\n",
486 inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
487 if_name(ifp));
488 m_freem(m);
489 return (EINVAL);
490 }
491
492 IF_AFDATA_WLOCK(ifp);
493 LLE_WLOCK(la);
494 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
495 /* Prefer ANY existing lle over newly-created one */
496 if (la_tmp == NULL)
497 lltable_link_entry(LLTABLE(ifp), la);
498 IF_AFDATA_WUNLOCK(ifp);
499 if (la_tmp != NULL) {
500 lltable_free_entry(LLTABLE(ifp), la);
501 la = la_tmp;
502 }
503 }
504 if (la == NULL) {
505 m_freem(m);
506 return (EINVAL);
507 }
508
509 if ((la->la_flags & LLE_VALID) &&
510 ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
511 if (flags & LLE_ADDRONLY) {
512 lladdr = la->ll_addr;
513 ll_len = ifp->if_addrlen;
514 } else {
515 lladdr = la->r_linkdata;
516 ll_len = la->r_hdrlen;
517 }
518 bcopy(lladdr, desten, ll_len);
519
520 /* Notify LLE code that the entry was used by datapath */
521 llentry_provide_feedback(la);
522 if (pflags != NULL)
523 *pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
524 if (plle) {
525 LLE_ADDREF(la);
526 *plle = la;
527 }
528 LLE_WUNLOCK(la);
529 return (0);
530 }
531
532 renew = (la->la_asked == 0 || la->la_expire != time_uptime);
533
534 /*
535 * There is an arptab entry, but no ethernet address
536 * response yet. Add the mbuf to the list, dropping
537 * the oldest packet if we have exceeded the system
538 * setting.
539 */
540 if (m != NULL) {
541 size_t dropped = lltable_append_entry_queue(la, m, V_arp_maxhold);
542 ARPSTAT_ADD(dropped, dropped);
543 }
544
545 /*
546 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
547 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
548 * if we have already sent arp_maxtries ARP requests. Retransmit the
549 * ARP request, but not faster than one request per second.
550 */
551 if (la->la_asked < V_arp_maxtries)
552 error = EWOULDBLOCK; /* First request. */
553 else
554 error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
555
556 if (renew) {
557 int canceled, e;
558
559 LLE_ADDREF(la);
560 la->la_expire = time_uptime;
561 canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
562 arptimer, la);
563 if (canceled)
564 LLE_REMREF(la);
565 la->la_asked++;
566 LLE_WUNLOCK(la);
567 e = arprequest_internal(ifp, NULL, &SIN(dst)->sin_addr, NULL);
568 /*
569 * Only overwrite 'error' in case of error; in case of success
570 * the proper return value was already set above.
571 */
572 if (e != 0)
573 return (e);
574 return (error);
575 }
576
577 LLE_WUNLOCK(la);
578 return (error);
579 }
580
581 /*
582 * Lookups link header based on an IP address.
583 * On input:
584 * ifp is the interface we use
585 * is_gw != 0 if @dst represents gateway to some destination
586 * m is the mbuf. May be NULL if we don't have a packet.
587 * dst is the next hop,
588 * desten is the storage to put LL header.
589 * flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
590 *
591 * On success, full/partial link header and flags are filled in and
592 * the function returns 0.
593 * If the packet must be held pending resolution, we return EWOULDBLOCK
594 * On other errors, we return the corresponding error code.
595 * Note that m_freem() handles NULL.
596 */
597 int
arpresolve(struct ifnet * ifp,int is_gw,struct mbuf * m,const struct sockaddr * dst,u_char * desten,uint32_t * pflags,struct llentry ** plle)598 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
599 const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
600 struct llentry **plle)
601 {
602 struct llentry *la = NULL;
603
604 NET_EPOCH_ASSERT();
605
606 if (pflags != NULL)
607 *pflags = 0;
608 if (plle != NULL)
609 *plle = NULL;
610
611 if (m != NULL) {
612 if (m->m_flags & M_BCAST) {
613 /* broadcast */
614 (void)memcpy(desten,
615 ifp->if_broadcastaddr, ifp->if_addrlen);
616 return (0);
617 }
618 if (m->m_flags & M_MCAST) {
619 /* multicast */
620 ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
621 return (0);
622 }
623 }
624
625 la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
626 if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
627 /* Entry found, let's copy lle info */
628 bcopy(la->r_linkdata, desten, la->r_hdrlen);
629 if (pflags != NULL)
630 *pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
631 /* Notify the LLE handling code that the entry was used. */
632 llentry_provide_feedback(la);
633 if (plle) {
634 LLE_ADDREF(la);
635 *plle = la;
636 LLE_WUNLOCK(la);
637 }
638 return (0);
639 }
640 if (plle && la)
641 LLE_WUNLOCK(la);
642
643 return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
644 desten, pflags, plle));
645 }
646
647 /*
648 * Common length and type checks are done here,
649 * then the protocol-specific routine is called.
650 */
651 static void
arpintr(struct mbuf * m)652 arpintr(struct mbuf *m)
653 {
654 struct arphdr *ar;
655 struct ifnet *ifp;
656 char *layer;
657 int hlen;
658
659 ifp = m->m_pkthdr.rcvif;
660
661 if (m->m_len < sizeof(struct arphdr) &&
662 ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
663 ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
664 if_name(ifp));
665 return;
666 }
667 ar = mtod(m, struct arphdr *);
668
669 /* Check if length is sufficient */
670 if (m->m_len < arphdr_len(ar)) {
671 m = m_pullup(m, arphdr_len(ar));
672 if (m == NULL) {
673 ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
674 if_name(ifp));
675 return;
676 }
677 ar = mtod(m, struct arphdr *);
678 }
679
680 hlen = 0;
681 layer = "";
682 switch (ntohs(ar->ar_hrd)) {
683 case ARPHRD_ETHER:
684 hlen = ETHER_ADDR_LEN; /* RFC 826 */
685 layer = "ethernet";
686 break;
687 case ARPHRD_IEEE802:
688 hlen = ETHER_ADDR_LEN;
689 layer = "ieee802";
690 break;
691 case ARPHRD_INFINIBAND:
692 hlen = 20; /* RFC 4391, INFINIBAND_ALEN */
693 layer = "infiniband";
694 break;
695 case ARPHRD_IEEE1394:
696 hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
697 layer = "firewire";
698
699 /*
700 * Restrict too long hardware addresses.
701 * Currently we are capable of handling 20-byte
702 * addresses ( sizeof(lle->ll_addr) )
703 */
704 if (ar->ar_hln >= 20)
705 hlen = 16;
706 break;
707 default:
708 ARP_LOG(LOG_NOTICE,
709 "packet with unknown hardware format 0x%02d received on "
710 "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
711 m_freem(m);
712 return;
713 }
714
715 if (hlen != 0 && hlen != ar->ar_hln) {
716 ARP_LOG(LOG_NOTICE,
717 "packet with invalid %s address length %d received on %s\n",
718 layer, ar->ar_hln, if_name(ifp));
719 m_freem(m);
720 return;
721 }
722
723 ARPSTAT_INC(received);
724 switch (ntohs(ar->ar_pro)) {
725 #ifdef INET
726 case ETHERTYPE_IP:
727 in_arpinput(m);
728 return;
729 #endif
730 }
731 m_freem(m);
732 }
733
734 #ifdef INET
735 /*
736 * ARP for Internet protocols on 10 Mb/s Ethernet.
737 * Algorithm is that given in RFC 826.
738 * In addition, a sanity check is performed on the sender
739 * protocol address, to catch impersonators.
740 * We no longer handle negotiations for use of trailer protocol:
741 * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
742 * along with IP replies if we wanted trailers sent to us,
743 * and also sent them in response to IP replies.
744 * This allowed either end to announce the desire to receive
745 * trailer packets.
746 * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
747 * but formerly didn't normally send requests.
748 */
749 static int log_arp_wrong_iface = 1;
750 static int log_arp_movements = 1;
751 static int log_arp_permanent_modify = 1;
752 static int allow_multicast = 0;
753
754 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
755 &log_arp_wrong_iface, 0,
756 "log arp packets arriving on the wrong interface");
757 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
758 &log_arp_movements, 0,
759 "log arp replies from MACs different than the one in the cache");
760 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
761 &log_arp_permanent_modify, 0,
762 "log arp replies from MACs different than the one in the permanent arp entry");
763 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
764 &allow_multicast, 0, "accept multicast addresses");
765
766 static void
in_arpinput(struct mbuf * m)767 in_arpinput(struct mbuf *m)
768 {
769 struct arphdr *ah;
770 struct ifnet *ifp = m->m_pkthdr.rcvif;
771 struct llentry *la = NULL, *la_tmp;
772 struct ifaddr *ifa;
773 struct in_ifaddr *ia;
774 struct sockaddr sa;
775 struct in_addr isaddr, itaddr, myaddr;
776 u_int8_t *enaddr = NULL;
777 int op;
778 int bridged = 0, is_bridge = 0;
779 int carped;
780 struct sockaddr_in sin;
781 struct sockaddr *dst;
782 struct nhop_object *nh;
783 uint8_t linkhdr[LLE_MAX_LINKHDR];
784 struct route ro;
785 size_t linkhdrsize;
786 int lladdr_off;
787 int error;
788 char addrbuf[INET_ADDRSTRLEN];
789
790 NET_EPOCH_ASSERT();
791
792 sin.sin_len = sizeof(struct sockaddr_in);
793 sin.sin_family = AF_INET;
794 sin.sin_addr.s_addr = 0;
795
796 if (ifp->if_bridge)
797 bridged = 1;
798 if (ifp->if_type == IFT_BRIDGE)
799 is_bridge = 1;
800
801 /*
802 * We already have checked that mbuf contains enough contiguous data
803 * to hold entire arp message according to the arp header.
804 */
805 ah = mtod(m, struct arphdr *);
806
807 /*
808 * ARP is only for IPv4 so we can reject packets with
809 * a protocol length not equal to an IPv4 address.
810 */
811 if (ah->ar_pln != sizeof(struct in_addr)) {
812 ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
813 sizeof(struct in_addr));
814 goto drop;
815 }
816
817 if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
818 ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
819 ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
820 goto drop;
821 }
822
823 op = ntohs(ah->ar_op);
824 (void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
825 (void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
826
827 if (op == ARPOP_REPLY)
828 ARPSTAT_INC(rxreplies);
829
830 /*
831 * For a bridge, we want to check the address irrespective
832 * of the receive interface. (This will change slightly
833 * when we have clusters of interfaces).
834 */
835 CK_LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
836 if (((bridged && bridge_same_p(ia->ia_ifp->if_bridge, ifp->if_bridge)) ||
837 ia->ia_ifp == ifp) &&
838 itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
839 (ia->ia_ifa.ifa_carp == NULL ||
840 (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
841 ifa_ref(&ia->ia_ifa);
842 goto match;
843 }
844 }
845 CK_LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
846 if (((bridged && bridge_same_p(ia->ia_ifp->if_bridge, ifp->if_bridge)) ||
847 ia->ia_ifp == ifp) &&
848 isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
849 ifa_ref(&ia->ia_ifa);
850 goto match;
851 }
852
853 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia) \
854 (bridge_get_softc_p(ia->ia_ifp) == ifp->if_softc && \
855 !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) && \
856 addr == ia->ia_addr.sin_addr.s_addr)
857 /*
858 * Check the case when bridge shares its MAC address with
859 * some of its children, so packets are claimed by bridge
860 * itself (bridge_input() does it first), but they are really
861 * meant to be destined to the bridge member.
862 */
863 if (is_bridge) {
864 CK_LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
865 if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
866 ifa_ref(&ia->ia_ifa);
867 ifp = ia->ia_ifp;
868 goto match;
869 }
870 }
871 }
872 #undef BDG_MEMBER_MATCHES_ARP
873
874 /*
875 * No match, use the first inet address on the receive interface
876 * as a dummy address for the rest of the function.
877 */
878 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
879 if (ifa->ifa_addr->sa_family == AF_INET &&
880 (ifa->ifa_carp == NULL ||
881 (*carp_iamatch_p)(ifa, &enaddr))) {
882 ia = ifatoia(ifa);
883 ifa_ref(ifa);
884 goto match;
885 }
886
887 /*
888 * If bridging, fall back to using any inet address.
889 */
890 if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL)
891 goto drop;
892 ifa_ref(&ia->ia_ifa);
893 match:
894 if (!enaddr)
895 enaddr = (u_int8_t *)IF_LLADDR(ifp);
896 carped = (ia->ia_ifa.ifa_carp != NULL);
897 myaddr = ia->ia_addr.sin_addr;
898 ifa_free(&ia->ia_ifa);
899 if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
900 goto drop; /* it's from me, ignore it. */
901 if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
902 ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
903 "%s!\n", inet_ntoa_r(isaddr, addrbuf));
904 goto drop;
905 }
906
907 if (ifp->if_addrlen != ah->ar_hln) {
908 ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
909 "i/f %d (ignored)\n", ifp->if_addrlen,
910 (u_char *) ar_sha(ah), ":", ah->ar_hln,
911 ifp->if_addrlen);
912 goto drop;
913 }
914
915 /*
916 * Warn if another host is using the same IP address, but only if the
917 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
918 * case we suppress the warning to avoid false positive complaints of
919 * potential misconfiguration.
920 */
921 if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
922 myaddr.s_addr != 0) {
923 ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
924 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
925 inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
926 itaddr = myaddr;
927 ARPSTAT_INC(dupips);
928 goto reply;
929 }
930 if (ifp->if_flags & IFF_STATICARP)
931 goto reply;
932
933 bzero(&sin, sizeof(sin));
934 sin.sin_len = sizeof(struct sockaddr_in);
935 sin.sin_family = AF_INET;
936 sin.sin_addr = isaddr;
937 dst = (struct sockaddr *)&sin;
938 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
939 if (la != NULL)
940 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
941 else if (itaddr.s_addr == myaddr.s_addr) {
942 /*
943 * Request/reply to our address, but no lle exists yet.
944 * Calculate full link prepend to use in lle.
945 */
946 linkhdrsize = sizeof(linkhdr);
947 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
948 &linkhdrsize, &lladdr_off) != 0)
949 goto reply;
950
951 /* Allocate new entry */
952 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
953 if (la == NULL) {
954 /*
955 * lle creation may fail if source address belongs
956 * to non-directly connected subnet. However, we
957 * will try to answer the request instead of dropping
958 * frame.
959 */
960 goto reply;
961 }
962 lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
963 lladdr_off);
964
965 IF_AFDATA_WLOCK(ifp);
966 LLE_WLOCK(la);
967 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
968
969 /*
970 * Check if lle still does not exists.
971 * If it does, that means that we either
972 * 1) have configured it explicitly, via
973 * 1a) 'arp -s' static entry or
974 * 1b) interface address static record
975 * or
976 * 2) it was the result of sending first packet to-host
977 * or
978 * 3) it was another arp reply packet we handled in
979 * different thread.
980 *
981 * In all cases except 3) we definitely need to prefer
982 * existing lle. For the sake of simplicity, prefer any
983 * existing lle over newly-create one.
984 */
985 if (la_tmp == NULL)
986 lltable_link_entry(LLTABLE(ifp), la);
987 IF_AFDATA_WUNLOCK(ifp);
988
989 if (la_tmp == NULL) {
990 arp_mark_lle_reachable(la, ifp);
991 LLE_WUNLOCK(la);
992 } else {
993 /* Free newly-create entry and handle packet */
994 lltable_free_entry(LLTABLE(ifp), la);
995 la = la_tmp;
996 la_tmp = NULL;
997 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
998 /* arp_check_update_lle() returns @la unlocked */
999 }
1000 la = NULL;
1001 }
1002 reply:
1003 if (op != ARPOP_REQUEST)
1004 goto drop;
1005 ARPSTAT_INC(rxrequests);
1006
1007 if (itaddr.s_addr == myaddr.s_addr) {
1008 /* Shortcut.. the receiving interface is the target. */
1009 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1010 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1011 } else {
1012 /*
1013 * Destination address is not ours. Check if
1014 * proxyarp entry exists or proxyarp is turned on globally.
1015 */
1016 struct llentry *lle;
1017
1018 sin.sin_addr = itaddr;
1019 lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1020
1021 if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1022 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1023 (void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1024 LLE_RUNLOCK(lle);
1025 } else {
1026 if (lle != NULL)
1027 LLE_RUNLOCK(lle);
1028
1029 if (!V_arp_proxyall)
1030 goto drop;
1031
1032 NET_EPOCH_ASSERT();
1033 nh = fib4_lookup(ifp->if_fib, itaddr, 0, 0, 0);
1034 if (nh == NULL)
1035 goto drop;
1036
1037 /*
1038 * Don't send proxies for nodes on the same interface
1039 * as this one came out of, or we'll get into a fight
1040 * over who claims what Ether address.
1041 */
1042 if (nh->nh_ifp == ifp)
1043 goto drop;
1044
1045 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1046 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1047
1048 /*
1049 * Also check that the node which sent the ARP packet
1050 * is on the interface we expect it to be on. This
1051 * avoids ARP chaos if an interface is connected to the
1052 * wrong network.
1053 */
1054
1055 nh = fib4_lookup(ifp->if_fib, isaddr, 0, 0, 0);
1056 if (nh == NULL)
1057 goto drop;
1058 if (nh->nh_ifp != ifp) {
1059 ARP_LOG(LOG_INFO, "proxy: ignoring request"
1060 " from %s via %s\n",
1061 inet_ntoa_r(isaddr, addrbuf),
1062 ifp->if_xname);
1063 goto drop;
1064 }
1065
1066 #ifdef DEBUG_PROXY
1067 printf("arp: proxying for %s\n",
1068 inet_ntoa_r(itaddr, addrbuf));
1069 #endif
1070 }
1071 }
1072
1073 if (itaddr.s_addr == myaddr.s_addr &&
1074 IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1075 /* RFC 3927 link-local IPv4; always reply by broadcast. */
1076 #ifdef DEBUG_LINKLOCAL
1077 printf("arp: sending reply for link-local addr %s\n",
1078 inet_ntoa_r(itaddr, addrbuf));
1079 #endif
1080 m->m_flags |= M_BCAST;
1081 m->m_flags &= ~M_MCAST;
1082 } else {
1083 /* default behaviour; never reply by broadcast. */
1084 m->m_flags &= ~(M_BCAST|M_MCAST);
1085 }
1086 (void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1087 (void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1088 ah->ar_op = htons(ARPOP_REPLY);
1089 ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1090 m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1091 m->m_pkthdr.len = m->m_len;
1092 m->m_pkthdr.rcvif = NULL;
1093 sa.sa_family = AF_ARP;
1094 sa.sa_len = 2;
1095
1096 /* Calculate link header for sending frame */
1097 bzero(&ro, sizeof(ro));
1098 linkhdrsize = sizeof(linkhdr);
1099 error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1100
1101 /*
1102 * arp_fillheader() may fail due to lack of support inside encap request
1103 * routing. This is not necessary an error, AF_ARP can/should be handled
1104 * by if_output().
1105 */
1106 if (error != 0 && error != EAFNOSUPPORT) {
1107 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1108 if_name(ifp), error);
1109 goto drop;
1110 }
1111
1112 ro.ro_prepend = linkhdr;
1113 ro.ro_plen = linkhdrsize;
1114 ro.ro_flags = 0;
1115
1116 m_clrprotoflags(m); /* Avoid confusing lower layers. */
1117 (*ifp->if_output)(ifp, m, &sa, &ro);
1118 ARPSTAT_INC(txreplies);
1119 return;
1120
1121 drop:
1122 m_freem(m);
1123 }
1124 #endif
1125
1126 static struct mbuf *
arp_grab_holdchain(struct llentry * la)1127 arp_grab_holdchain(struct llentry *la)
1128 {
1129 struct mbuf *chain;
1130
1131 LLE_WLOCK_ASSERT(la);
1132
1133 chain = la->la_hold;
1134 la->la_hold = NULL;
1135 la->la_numheld = 0;
1136
1137 return (chain);
1138 }
1139
1140 static void
arp_flush_holdchain(struct ifnet * ifp,struct llentry * la,struct mbuf * chain)1141 arp_flush_holdchain(struct ifnet *ifp, struct llentry *la, struct mbuf *chain)
1142 {
1143 struct mbuf *m_hold, *m_hold_next;
1144 struct sockaddr_in sin;
1145
1146 NET_EPOCH_ASSERT();
1147
1148 struct route ro = {
1149 .ro_prepend = la->r_linkdata,
1150 .ro_plen = la->r_hdrlen,
1151 };
1152
1153 lltable_fill_sa_entry(la, (struct sockaddr *)&sin);
1154
1155 for (m_hold = chain; m_hold != NULL; m_hold = m_hold_next) {
1156 m_hold_next = m_hold->m_nextpkt;
1157 m_hold->m_nextpkt = NULL;
1158 /* Avoid confusing lower layers. */
1159 m_clrprotoflags(m_hold);
1160 (*ifp->if_output)(ifp, m_hold, (struct sockaddr *)&sin, &ro);
1161 }
1162 }
1163
1164 /*
1165 * Checks received arp data against existing @la.
1166 * Updates lle state/performs notification if necessary.
1167 */
1168 static void
arp_check_update_lle(struct arphdr * ah,struct in_addr isaddr,struct ifnet * ifp,int bridged,struct llentry * la)1169 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1170 int bridged, struct llentry *la)
1171 {
1172 uint8_t linkhdr[LLE_MAX_LINKHDR];
1173 size_t linkhdrsize;
1174 int lladdr_off;
1175 char addrbuf[INET_ADDRSTRLEN];
1176
1177 LLE_WLOCK_ASSERT(la);
1178
1179 /* the following is not an error when doing bridging */
1180 if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1181 if (log_arp_wrong_iface)
1182 ARP_LOG(LOG_WARNING, "%s is on %s "
1183 "but got reply from %*D on %s\n",
1184 inet_ntoa_r(isaddr, addrbuf),
1185 la->lle_tbl->llt_ifp->if_xname,
1186 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1187 ifp->if_xname);
1188 LLE_WUNLOCK(la);
1189 return;
1190 }
1191 if ((la->la_flags & LLE_VALID) &&
1192 bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1193 if (la->la_flags & LLE_STATIC) {
1194 LLE_WUNLOCK(la);
1195 if (log_arp_permanent_modify)
1196 ARP_LOG(LOG_ERR,
1197 "%*D attempts to modify "
1198 "permanent entry for %s on %s\n",
1199 ifp->if_addrlen,
1200 (u_char *)ar_sha(ah), ":",
1201 inet_ntoa_r(isaddr, addrbuf),
1202 ifp->if_xname);
1203 return;
1204 }
1205 if (log_arp_movements) {
1206 ARP_LOG(LOG_INFO, "%s moved from %*D "
1207 "to %*D on %s\n",
1208 inet_ntoa_r(isaddr, addrbuf),
1209 ifp->if_addrlen,
1210 (u_char *)la->ll_addr, ":",
1211 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1212 ifp->if_xname);
1213 }
1214 }
1215
1216 /* Calculate full link prepend to use in lle */
1217 linkhdrsize = sizeof(linkhdr);
1218 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1219 &linkhdrsize, &lladdr_off) != 0) {
1220 LLE_WUNLOCK(la);
1221 return;
1222 }
1223
1224 /* Check if something has changed */
1225 if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1226 (la->la_flags & LLE_VALID) == 0) {
1227 /* Try to perform LLE update */
1228 if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1229 lladdr_off) == 0) {
1230 LLE_WUNLOCK(la);
1231 return;
1232 }
1233
1234 /* Clear fast path feedback request if set */
1235 llentry_mark_used(la);
1236 }
1237
1238 arp_mark_lle_reachable(la, ifp);
1239
1240 /*
1241 * The packets are all freed within the call to the output
1242 * routine.
1243 *
1244 * NB: The lock MUST be released before the call to the
1245 * output routine.
1246 */
1247 if (la->la_hold != NULL) {
1248 struct mbuf *chain;
1249
1250 chain = arp_grab_holdchain(la);
1251 LLE_WUNLOCK(la);
1252 arp_flush_holdchain(ifp, la, chain);
1253 } else
1254 LLE_WUNLOCK(la);
1255 }
1256
1257 static void
arp_mark_lle_reachable(struct llentry * la,struct ifnet * ifp)1258 arp_mark_lle_reachable(struct llentry *la, struct ifnet *ifp)
1259 {
1260 int canceled, wtime;
1261
1262 LLE_WLOCK_ASSERT(la);
1263
1264 la->ln_state = ARP_LLINFO_REACHABLE;
1265 EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1266
1267 if ((ifp->if_flags & IFF_STICKYARP) != 0)
1268 la->la_flags |= LLE_STATIC;
1269
1270 if (!(la->la_flags & LLE_STATIC)) {
1271 LLE_ADDREF(la);
1272 la->la_expire = time_uptime + V_arpt_keep;
1273 wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1274 if (wtime < 0)
1275 wtime = V_arpt_keep;
1276 canceled = callout_reset(&la->lle_timer,
1277 hz * wtime, arptimer, la);
1278 if (canceled)
1279 LLE_REMREF(la);
1280 }
1281 la->la_asked = 0;
1282 la->la_preempt = V_arp_maxtries;
1283 }
1284
1285 /*
1286 * Add permanent link-layer record for given interface address.
1287 */
1288 static __noinline void
arp_add_ifa_lle(struct ifnet * ifp,const struct sockaddr * dst)1289 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1290 {
1291 struct llentry *lle, *lle_tmp;
1292
1293 /*
1294 * Interface address LLE record is considered static
1295 * because kernel code relies on LLE_STATIC flag to check
1296 * if these entries can be rewriten by arp updates.
1297 */
1298 lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1299 if (lle == NULL) {
1300 log(LOG_INFO, "arp_ifinit: cannot create arp "
1301 "entry for interface address\n");
1302 return;
1303 }
1304
1305 IF_AFDATA_WLOCK(ifp);
1306 LLE_WLOCK(lle);
1307 /* Unlink any entry if exists */
1308 lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1309 if (lle_tmp != NULL)
1310 lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1311
1312 lltable_link_entry(LLTABLE(ifp), lle);
1313 IF_AFDATA_WUNLOCK(ifp);
1314
1315 if (lle_tmp != NULL)
1316 EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1317
1318 EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1319 LLE_WUNLOCK(lle);
1320 if (lle_tmp != NULL)
1321 lltable_free_entry(LLTABLE(ifp), lle_tmp);
1322 }
1323
1324 /*
1325 * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1326 * of valid values.
1327 */
1328 static int
sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)1329 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1330 {
1331 int error;
1332 int rexmit_count = *(int *)arg1;
1333
1334 error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1335
1336 /* Enforce limits on any new value that may have been set. */
1337 if (!error && req->newptr) {
1338 /* A new value was set. */
1339 if (rexmit_count < 0) {
1340 rexmit_count = 0;
1341 } else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1342 rexmit_count = MAX_GARP_RETRANSMITS;
1343 }
1344 *(int *)arg1 = rexmit_count;
1345 }
1346
1347 return (error);
1348 }
1349
1350 /*
1351 * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1352 * retransmit it again. A pending callout owns a reference to the ifa.
1353 */
1354 static void
garp_rexmit(void * arg)1355 garp_rexmit(void *arg)
1356 {
1357 struct epoch_tracker et;
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 NET_EPOCH_ENTER(et);
1368 CURVNET_SET(ia->ia_ifa.ifa_ifp->if_vnet);
1369
1370 /*
1371 * Drop lock while the ARP request is generated.
1372 */
1373 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1374
1375 arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1376 &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1377
1378 /*
1379 * Increment the count of retransmissions. If the count has reached the
1380 * maximum value, stop sending the GARP packets. Otherwise, schedule
1381 * the callout to retransmit another GARP packet.
1382 */
1383 ++ia->ia_garp_count;
1384 if (ia->ia_garp_count >= V_garp_rexmit_count) {
1385 ifa_free(&ia->ia_ifa);
1386 } else {
1387 int rescheduled;
1388 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1389 rescheduled = callout_reset(&ia->ia_garp_timer,
1390 (1 << ia->ia_garp_count) * hz,
1391 garp_rexmit, ia);
1392 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1393 if (rescheduled) {
1394 ifa_free(&ia->ia_ifa);
1395 }
1396 }
1397
1398 CURVNET_RESTORE();
1399 NET_EPOCH_EXIT(et);
1400 }
1401
1402 /*
1403 * Start the GARP retransmit timer.
1404 *
1405 * A single GARP is always transmitted when an IPv4 address is added
1406 * to an interface and that is usually sufficient. However, in some
1407 * circumstances, such as when a shared address is passed between
1408 * cluster nodes, this single GARP may occasionally be dropped or
1409 * lost. This can lead to neighbors on the network link working with a
1410 * stale ARP cache and sending packets destined for that address to
1411 * the node that previously owned the address, which may not respond.
1412 *
1413 * To avoid this situation, GARP retransmits can be enabled by setting
1414 * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1415 * than zero. The setting represents the maximum number of
1416 * retransmissions. The interval between retransmissions is calculated
1417 * using an exponential backoff algorithm, doubling each time, so the
1418 * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1419 */
1420 static void
garp_timer_start(struct ifaddr * ifa)1421 garp_timer_start(struct ifaddr *ifa)
1422 {
1423 struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1424
1425 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1426 ia->ia_garp_count = 0;
1427 if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1428 garp_rexmit, ia) == 0) {
1429 ifa_ref(ifa);
1430 }
1431 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1432 }
1433
1434 void
arp_ifinit(struct ifnet * ifp,struct ifaddr * ifa)1435 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1436 {
1437 struct epoch_tracker et;
1438 const struct sockaddr_in *dst_in;
1439 const struct sockaddr *dst;
1440
1441 if (ifa->ifa_carp != NULL)
1442 return;
1443
1444 dst = ifa->ifa_addr;
1445 dst_in = (const struct sockaddr_in *)dst;
1446
1447 if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1448 return;
1449 NET_EPOCH_ENTER(et);
1450 arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1451 NET_EPOCH_EXIT(et);
1452 if (V_garp_rexmit_count > 0) {
1453 garp_timer_start(ifa);
1454 }
1455
1456 arp_add_ifa_lle(ifp, dst);
1457 }
1458
1459 void
arp_announce_ifaddr(struct ifnet * ifp,struct in_addr addr,u_char * enaddr)1460 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1461 {
1462
1463 if (ntohl(addr.s_addr) != INADDR_ANY)
1464 arprequest(ifp, &addr, &addr, enaddr);
1465 }
1466
1467 /*
1468 * Sends gratuitous ARPs for each ifaddr to notify other
1469 * nodes about the address change.
1470 */
1471 static __noinline void
arp_handle_ifllchange(struct ifnet * ifp)1472 arp_handle_ifllchange(struct ifnet *ifp)
1473 {
1474 struct ifaddr *ifa;
1475
1476 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1477 if (ifa->ifa_addr->sa_family == AF_INET)
1478 arp_ifinit(ifp, ifa);
1479 }
1480 }
1481
1482 /*
1483 * A handler for interface link layer address change event.
1484 */
1485 static void
arp_iflladdr(void * arg __unused,struct ifnet * ifp)1486 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1487 {
1488 /* if_bridge can update its lladdr during if_vmove(), after we've done
1489 * if_detach_internal()/dom_ifdetach(). */
1490 if (ifp->if_afdata[AF_INET] == NULL)
1491 return;
1492
1493 lltable_update_ifaddr(LLTABLE(ifp));
1494
1495 if ((ifp->if_flags & IFF_UP) != 0)
1496 arp_handle_ifllchange(ifp);
1497 }
1498
1499 static void
vnet_arp_init(void)1500 vnet_arp_init(void)
1501 {
1502
1503 if (IS_DEFAULT_VNET(curvnet)) {
1504 netisr_register(&arp_nh);
1505 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1506 arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1507 }
1508 #ifdef VIMAGE
1509 else
1510 netisr_register_vnet(&arp_nh);
1511 #endif
1512 }
1513 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1514 vnet_arp_init, NULL);
1515
1516 #ifdef VIMAGE
1517 /*
1518 * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1519 * lookups after destroying the hash. Ideally this would go on SI_ORDER_3.5.
1520 */
1521 static void
vnet_arp_destroy(__unused void * arg)1522 vnet_arp_destroy(__unused void *arg)
1523 {
1524
1525 netisr_unregister_vnet(&arp_nh);
1526 }
1527 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1528 vnet_arp_destroy, NULL);
1529 #endif
1530