1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1988, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 */
31
32 /*
33 * Ethernet address resolution protocol.
34 * TODO:
35 * add "inuse/lock" bit (or ref. count) along with valid bit
36 */
37
38 #include <sys/cdefs.h>
39 #include "opt_inet.h"
40
41 #include <sys/param.h>
42 #include <sys/eventhandler.h>
43 #include <sys/kernel.h>
44 #include <sys/lock.h>
45 #include <sys/queue.h>
46 #include <sys/sysctl.h>
47 #include <sys/systm.h>
48 #include <sys/mbuf.h>
49 #include <sys/malloc.h>
50 #include <sys/proc.h>
51 #include <sys/socket.h>
52 #include <sys/syslog.h>
53
54 #include <net/if.h>
55 #include <net/if_var.h>
56 #include <net/if_dl.h>
57 #include <net/if_private.h>
58 #include <net/if_types.h>
59 #include <net/netisr.h>
60 #include <net/ethernet.h>
61 #include <net/route.h>
62 #include <net/route/nhop.h>
63 #include <net/vnet.h>
64
65 #include <netinet/in.h>
66 #include <netinet/in_fib.h>
67 #include <netinet/in_var.h>
68 #include <net/if_llatbl.h>
69 #include <netinet/if_ether.h>
70 #ifdef INET
71 #include <netinet/ip_carp.h>
72 #endif
73
74 #include <security/mac/mac_framework.h>
75
76 #define SIN(s) ((const struct sockaddr_in *)(s))
77
78 static struct timeval arp_lastlog;
79 static int arp_curpps;
80 static int arp_maxpps = 1;
81
82 /* Simple ARP state machine */
83 enum arp_llinfo_state {
84 ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */
85 ARP_LLINFO_REACHABLE, /* LLE is valid */
86 ARP_LLINFO_VERIFY, /* LLE is valid, need refresh */
87 ARP_LLINFO_DELETED, /* LLE is deleted */
88 };
89
90 SYSCTL_DECL(_net_link_ether);
91 static SYSCTL_NODE(_net_link_ether, PF_INET, inet,
92 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
93 "");
94 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp,
95 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
96 "");
97
98 /* timer values */
99 VNET_DEFINE_STATIC(int, arpt_keep) = (20*60); /* once resolved, good for 20
100 * minutes */
101 VNET_DEFINE_STATIC(int, arp_maxtries) = 5;
102 VNET_DEFINE_STATIC(int, arp_proxyall) = 0;
103 VNET_DEFINE_STATIC(int, arpt_down) = 20; /* keep incomplete entries for
104 * 20 seconds */
105 VNET_DEFINE_STATIC(int, arpt_rexmit) = 1; /* retransmit arp entries, sec*/
106 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat); /* ARP statistics, see if_arp.h */
107 VNET_PCPUSTAT_SYSINIT(arpstat);
108
109 #ifdef VIMAGE
110 VNET_PCPUSTAT_SYSUNINIT(arpstat);
111 #endif /* VIMAGE */
112
113 VNET_DEFINE_STATIC(int, arp_maxhold) = 16;
114
115 #define V_arpt_keep VNET(arpt_keep)
116 #define V_arpt_down VNET(arpt_down)
117 #define V_arpt_rexmit VNET(arpt_rexmit)
118 #define V_arp_maxtries VNET(arp_maxtries)
119 #define V_arp_proxyall VNET(arp_proxyall)
120 #define V_arp_maxhold VNET(arp_maxhold)
121
122 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
123 &VNET_NAME(arpt_keep), 0,
124 "ARP entry lifetime in seconds");
125 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
126 &VNET_NAME(arp_maxtries), 0,
127 "ARP resolution attempts before returning error");
128 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
129 &VNET_NAME(arp_proxyall), 0,
130 "Enable proxy ARP for all suitable requests");
131 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
132 &VNET_NAME(arpt_down), 0,
133 "Incomplete ARP entry lifetime in seconds");
134 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
135 arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
136 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
137 &VNET_NAME(arp_maxhold), 0,
138 "Number of packets to hold per ARP entry");
139 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
140 CTLFLAG_RW, &arp_maxpps, 0,
141 "Maximum number of remotely triggered ARP messages that can be "
142 "logged per second");
143
144 /*
145 * Due to the exponential backoff algorithm used for the interval between GARP
146 * retransmissions, the maximum number of retransmissions is limited for
147 * sanity. This limit corresponds to a maximum interval between retransmissions
148 * of 2^16 seconds ~= 18 hours.
149 *
150 * Making this limit more dynamic is more complicated than worthwhile,
151 * especially since sending out GARPs spaced days apart would be of little
152 * use. A maximum dynamic limit would look something like:
153 *
154 * const int max = fls(INT_MAX / hz) - 1;
155 */
156 #define MAX_GARP_RETRANSMITS 16
157 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS);
158 VNET_DEFINE_STATIC(int, garp_rexmit_count) = 0; /* GARP retransmission setting. */
159 #define V_garp_rexmit_count VNET(garp_rexmit_count)
160
161 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count,
162 CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE|CTLFLAG_VNET,
163 &VNET_NAME(garp_rexmit_count), 0, sysctl_garp_rexmit, "I",
164 "Number of times to retransmit GARP packets;"
165 " 0 to disable, maximum of 16");
166
167 VNET_DEFINE_STATIC(int, arp_log_level) = LOG_INFO; /* Min. log(9) level. */
168 #define V_arp_log_level VNET(arp_log_level)
169 SYSCTL_INT(_net_link_ether_arp, OID_AUTO, log_level, CTLFLAG_VNET | CTLFLAG_RW,
170 &VNET_NAME(arp_log_level), 0,
171 "Minimum log(9) level for recording rate limited arp log messages. "
172 "The higher will be log more (emerg=0, info=6 (default), debug=7).");
173 #define ARP_LOG(pri, ...) do { \
174 if ((pri) <= V_arp_log_level && \
175 ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps)) \
176 log((pri), "arp: " __VA_ARGS__); \
177 } while (0)
178
179 static void arpintr(struct mbuf *);
180 static void arptimer(void *);
181 #ifdef INET
182 static void in_arpinput(struct mbuf *);
183 #endif
184
185 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
186 struct ifnet *ifp, int bridged, struct llentry *la);
187 static void arp_mark_lle_reachable(struct llentry *la, struct ifnet *ifp);
188 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
189
190 static eventhandler_tag iflladdr_tag;
191
192 static const struct netisr_handler arp_nh = {
193 .nh_name = "arp",
194 .nh_handler = arpintr,
195 .nh_proto = NETISR_ARP,
196 .nh_policy = NETISR_POLICY_SOURCE,
197 };
198
199 /*
200 * Timeout routine. Age arp_tab entries periodically.
201 */
202 static void
arptimer(void * arg)203 arptimer(void *arg)
204 {
205 struct llentry *lle = (struct llentry *)arg;
206 struct ifnet *ifp;
207
208 if (lle->la_flags & LLE_STATIC) {
209 return;
210 }
211 LLE_WLOCK(lle);
212 if (callout_pending(&lle->lle_timer)) {
213 /*
214 * Here we are a bit odd here in the treatment of
215 * active/pending. If the pending bit is set, it got
216 * rescheduled before I ran. The active
217 * bit we ignore, since if it was stopped
218 * in ll_tablefree() and was currently running
219 * it would have return 0 so the code would
220 * not have deleted it since the callout could
221 * not be stopped so we want to go through
222 * with the delete here now. If the callout
223 * was restarted, the pending bit will be back on and
224 * we just want to bail since the callout_reset would
225 * return 1 and our reference would have been removed
226 * by arpresolve() below.
227 */
228 LLE_WUNLOCK(lle);
229 return;
230 }
231 ifp = lle->lle_tbl->llt_ifp;
232 CURVNET_SET(ifp->if_vnet);
233
234 switch (lle->ln_state) {
235 case ARP_LLINFO_REACHABLE:
236
237 /*
238 * Expiration time is approaching.
239 * Request usage feedback from the datapath.
240 * Change state and re-schedule ourselves.
241 */
242 llentry_request_feedback(lle);
243 lle->ln_state = ARP_LLINFO_VERIFY;
244 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
245 LLE_WUNLOCK(lle);
246 CURVNET_RESTORE();
247 return;
248 case ARP_LLINFO_VERIFY:
249 if (llentry_get_hittime(lle) > 0 && lle->la_preempt > 0) {
250 /* Entry was used, issue refresh request */
251 struct epoch_tracker et;
252 struct in_addr dst;
253
254 dst = lle->r_l3addr.addr4;
255 lle->la_preempt--;
256 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
257 LLE_WUNLOCK(lle);
258 NET_EPOCH_ENTER(et);
259 arprequest(ifp, NULL, &dst, NULL);
260 NET_EPOCH_EXIT(et);
261 CURVNET_RESTORE();
262 return;
263 }
264 /* Nothing happened. Reschedule if not too late */
265 if (lle->la_expire > time_uptime) {
266 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
267 LLE_WUNLOCK(lle);
268 CURVNET_RESTORE();
269 return;
270 }
271 break;
272 case ARP_LLINFO_INCOMPLETE:
273 case ARP_LLINFO_DELETED:
274 break;
275 }
276
277 if ((lle->la_flags & LLE_DELETED) == 0) {
278 int evt;
279
280 if (lle->la_flags & LLE_VALID)
281 evt = LLENTRY_EXPIRED;
282 else
283 evt = LLENTRY_TIMEDOUT;
284 EVENTHANDLER_INVOKE(lle_event, lle, evt);
285 }
286
287 callout_stop(&lle->lle_timer);
288
289 /* XXX: LOR avoidance. We still have ref on lle. */
290 LLE_WUNLOCK(lle);
291 IF_AFDATA_LOCK(ifp);
292 LLE_WLOCK(lle);
293
294 /* Guard against race with other llentry_free(). */
295 if (lle->la_flags & LLE_LINKED) {
296 LLE_REMREF(lle);
297 lltable_unlink_entry(lle->lle_tbl, lle);
298 }
299 IF_AFDATA_UNLOCK(ifp);
300
301 size_t pkts_dropped = llentry_free(lle);
302
303 ARPSTAT_ADD(dropped, pkts_dropped);
304 ARPSTAT_INC(timeouts);
305
306 CURVNET_RESTORE();
307 }
308
309 /*
310 * Stores link-layer header for @ifp in format suitable for if_output()
311 * into buffer @buf. Resulting header length is stored in @bufsize.
312 *
313 * Returns 0 on success.
314 */
315 static int
arp_fillheader(struct ifnet * ifp,struct arphdr * ah,int bcast,u_char * buf,size_t * bufsize)316 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
317 size_t *bufsize)
318 {
319 struct if_encap_req ereq;
320 int error;
321
322 bzero(buf, *bufsize);
323 bzero(&ereq, sizeof(ereq));
324 ereq.buf = buf;
325 ereq.bufsize = *bufsize;
326 ereq.rtype = IFENCAP_LL;
327 ereq.family = AF_ARP;
328 ereq.lladdr = ar_tha(ah);
329 ereq.hdata = (u_char *)ah;
330 if (bcast)
331 ereq.flags = IFENCAP_FLAG_BROADCAST;
332 error = ifp->if_requestencap(ifp, &ereq);
333 if (error == 0)
334 *bufsize = ereq.bufsize;
335
336 return (error);
337 }
338
339 /*
340 * Broadcast an ARP request. Caller specifies:
341 * - arp header source ip address
342 * - arp header target ip address
343 * - arp header source ethernet address
344 */
345 static int
arprequest_internal(struct ifnet * ifp,const struct in_addr * sip,const struct in_addr * tip,u_char * enaddr)346 arprequest_internal(struct ifnet *ifp, const struct in_addr *sip,
347 const struct in_addr *tip, u_char *enaddr)
348 {
349 struct mbuf *m;
350 struct arphdr *ah;
351 struct sockaddr sa;
352 u_char *carpaddr = NULL;
353 uint8_t linkhdr[LLE_MAX_LINKHDR];
354 size_t linkhdrsize;
355 struct route ro;
356 int error;
357
358 NET_EPOCH_ASSERT();
359
360 if (sip == NULL) {
361 /*
362 * The caller did not supply a source address, try to find
363 * a compatible one among those assigned to this interface.
364 */
365 struct ifaddr *ifa;
366
367 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
368 if (ifa->ifa_addr->sa_family != AF_INET)
369 continue;
370
371 if (ifa->ifa_carp) {
372 if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
373 continue;
374 sip = &IA_SIN(ifa)->sin_addr;
375 } else {
376 carpaddr = NULL;
377 sip = &IA_SIN(ifa)->sin_addr;
378 }
379
380 if (0 == ((sip->s_addr ^ tip->s_addr) &
381 IA_MASKSIN(ifa)->sin_addr.s_addr))
382 break; /* found it. */
383 }
384 if (sip == NULL) {
385 printf("%s: cannot find matching address\n", __func__);
386 return (EADDRNOTAVAIL);
387 }
388 }
389 if (enaddr == NULL)
390 enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
391
392 if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
393 return (ENOMEM);
394 m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
395 2 * ifp->if_addrlen;
396 m->m_pkthdr.len = m->m_len;
397 M_ALIGN(m, m->m_len);
398 ah = mtod(m, struct arphdr *);
399 bzero((caddr_t)ah, m->m_len);
400 #ifdef MAC
401 mac_netinet_arp_send(ifp, m);
402 #endif
403 ah->ar_pro = htons(ETHERTYPE_IP);
404 ah->ar_hln = ifp->if_addrlen; /* hardware address length */
405 ah->ar_pln = sizeof(struct in_addr); /* protocol address length */
406 ah->ar_op = htons(ARPOP_REQUEST);
407 bcopy(enaddr, ar_sha(ah), ah->ar_hln);
408 bcopy(sip, ar_spa(ah), ah->ar_pln);
409 bcopy(tip, ar_tpa(ah), ah->ar_pln);
410 sa.sa_family = AF_ARP;
411 sa.sa_len = 2;
412
413 /* Calculate link header for sending frame */
414 bzero(&ro, sizeof(ro));
415 linkhdrsize = sizeof(linkhdr);
416 error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
417 if (error != 0 && error != EAFNOSUPPORT) {
418 m_freem(m);
419 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
420 if_name(ifp), error);
421 return (error);
422 }
423
424 ro.ro_prepend = linkhdr;
425 ro.ro_plen = linkhdrsize;
426 ro.ro_flags = 0;
427
428 m->m_flags |= M_BCAST;
429 m_clrprotoflags(m); /* Avoid confusing lower layers. */
430 error = (*ifp->if_output)(ifp, m, &sa, &ro);
431 ARPSTAT_INC(txrequests);
432 if (error) {
433 ARPSTAT_INC(txerrors);
434 ARP_LOG(LOG_DEBUG, "Failed to send ARP packet on %s: %d\n",
435 if_name(ifp), error);
436 }
437 return (error);
438 }
439
440 void
arprequest(struct ifnet * ifp,const struct in_addr * sip,const struct in_addr * tip,u_char * enaddr)441 arprequest(struct ifnet *ifp, const struct in_addr *sip,
442 const struct in_addr *tip, u_char *enaddr)
443 {
444
445 (void) arprequest_internal(ifp, sip, tip, enaddr);
446 }
447
448 /*
449 * Resolve an IP address into an ethernet address - heavy version.
450 * Used internally by arpresolve().
451 * We have already checked that we can't use an existing lle without
452 * modification so we have to acquire an LLE_EXCLUSIVE lle lock.
453 *
454 * On success, desten and pflags are filled in and the function returns 0;
455 * If the packet must be held pending resolution, we return EWOULDBLOCK
456 * On other errors, we return the corresponding error code.
457 * Note that m_freem() handles NULL.
458 */
459 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)460 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
461 const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
462 struct llentry **plle)
463 {
464 struct llentry *la = NULL, *la_tmp;
465 int error, renew;
466 char *lladdr;
467 int ll_len;
468
469 NET_EPOCH_ASSERT();
470
471 if (pflags != NULL)
472 *pflags = 0;
473 if (plle != NULL)
474 *plle = NULL;
475
476 if ((flags & LLE_CREATE) == 0)
477 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
478 if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
479 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
480 if (la == NULL) {
481 char addrbuf[INET_ADDRSTRLEN];
482
483 log(LOG_DEBUG,
484 "arpresolve: can't allocate llinfo for %s on %s\n",
485 inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
486 if_name(ifp));
487 m_freem(m);
488 return (EINVAL);
489 }
490
491 IF_AFDATA_WLOCK(ifp);
492 LLE_WLOCK(la);
493 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
494 /* Prefer ANY existing lle over newly-created one */
495 if (la_tmp == NULL)
496 lltable_link_entry(LLTABLE(ifp), la);
497 IF_AFDATA_WUNLOCK(ifp);
498 if (la_tmp != NULL) {
499 lltable_free_entry(LLTABLE(ifp), la);
500 la = la_tmp;
501 }
502 }
503 if (la == NULL) {
504 m_freem(m);
505 return (EINVAL);
506 }
507
508 if ((la->la_flags & LLE_VALID) &&
509 ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
510 if (flags & LLE_ADDRONLY) {
511 lladdr = la->ll_addr;
512 ll_len = ifp->if_addrlen;
513 } else {
514 lladdr = la->r_linkdata;
515 ll_len = la->r_hdrlen;
516 }
517 bcopy(lladdr, desten, ll_len);
518
519 /* Notify LLE code that the entry was used by datapath */
520 llentry_provide_feedback(la);
521 if (pflags != NULL)
522 *pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
523 if (plle) {
524 LLE_ADDREF(la);
525 *plle = la;
526 }
527 LLE_WUNLOCK(la);
528 return (0);
529 }
530
531 renew = (la->la_asked == 0 || la->la_expire != time_uptime);
532
533 /*
534 * There is an arptab entry, but no ethernet address
535 * response yet. Add the mbuf to the list, dropping
536 * the oldest packet if we have exceeded the system
537 * setting.
538 */
539 if (m != NULL) {
540 size_t dropped = lltable_append_entry_queue(la, m, V_arp_maxhold);
541 ARPSTAT_ADD(dropped, dropped);
542 }
543
544 /*
545 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
546 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
547 * if we have already sent arp_maxtries ARP requests. Retransmit the
548 * ARP request, but not faster than one request per second.
549 */
550 if (la->la_asked < V_arp_maxtries)
551 error = EWOULDBLOCK; /* First request. */
552 else
553 error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
554
555 if (renew) {
556 int canceled, e;
557
558 LLE_ADDREF(la);
559 la->la_expire = time_uptime;
560 canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
561 arptimer, la);
562 if (canceled)
563 LLE_REMREF(la);
564 la->la_asked++;
565 LLE_WUNLOCK(la);
566 e = arprequest_internal(ifp, NULL, &SIN(dst)->sin_addr, NULL);
567 /*
568 * Only overwrite 'error' in case of error; in case of success
569 * the proper return value was already set above.
570 */
571 if (e != 0)
572 return (e);
573 return (error);
574 }
575
576 LLE_WUNLOCK(la);
577 return (error);
578 }
579
580 /*
581 * Lookups link header based on an IP address.
582 * On input:
583 * ifp is the interface we use
584 * is_gw != 0 if @dst represents gateway to some destination
585 * m is the mbuf. May be NULL if we don't have a packet.
586 * dst is the next hop,
587 * desten is the storage to put LL header.
588 * flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
589 *
590 * On success, full/partial link header and flags are filled in and
591 * the function returns 0.
592 * If the packet must be held pending resolution, we return EWOULDBLOCK
593 * On other errors, we return the corresponding error code.
594 * Note that m_freem() handles NULL.
595 */
596 int
arpresolve(struct ifnet * ifp,int is_gw,struct mbuf * m,const struct sockaddr * dst,u_char * desten,uint32_t * pflags,struct llentry ** plle)597 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
598 const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
599 struct llentry **plle)
600 {
601 struct llentry *la = NULL;
602
603 NET_EPOCH_ASSERT();
604
605 if (pflags != NULL)
606 *pflags = 0;
607 if (plle != NULL)
608 *plle = NULL;
609
610 if (m != NULL) {
611 if (m->m_flags & M_BCAST) {
612 /* broadcast */
613 (void)memcpy(desten,
614 ifp->if_broadcastaddr, ifp->if_addrlen);
615 return (0);
616 }
617 if (m->m_flags & M_MCAST) {
618 /* multicast */
619 ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
620 return (0);
621 }
622 }
623
624 la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
625 if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
626 /* Entry found, let's copy lle info */
627 bcopy(la->r_linkdata, desten, la->r_hdrlen);
628 if (pflags != NULL)
629 *pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
630 /* Notify the LLE handling code that the entry was used. */
631 llentry_provide_feedback(la);
632 if (plle) {
633 LLE_ADDREF(la);
634 *plle = la;
635 LLE_WUNLOCK(la);
636 }
637 return (0);
638 }
639 if (plle && la)
640 LLE_WUNLOCK(la);
641
642 return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
643 desten, pflags, plle));
644 }
645
646 /*
647 * Common length and type checks are done here,
648 * then the protocol-specific routine is called.
649 */
650 static void
arpintr(struct mbuf * m)651 arpintr(struct mbuf *m)
652 {
653 struct arphdr *ar;
654 struct ifnet *ifp;
655 char *layer;
656 int hlen;
657
658 ifp = m->m_pkthdr.rcvif;
659
660 if (m->m_len < sizeof(struct arphdr) &&
661 ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
662 ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
663 if_name(ifp));
664 return;
665 }
666 ar = mtod(m, struct arphdr *);
667
668 /* Check if length is sufficient */
669 if (m->m_len < arphdr_len(ar)) {
670 m = m_pullup(m, arphdr_len(ar));
671 if (m == NULL) {
672 ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
673 if_name(ifp));
674 return;
675 }
676 ar = mtod(m, struct arphdr *);
677 }
678
679 hlen = 0;
680 layer = "";
681 switch (ntohs(ar->ar_hrd)) {
682 case ARPHRD_ETHER:
683 hlen = ETHER_ADDR_LEN; /* RFC 826 */
684 layer = "ethernet";
685 break;
686 case ARPHRD_IEEE802:
687 hlen = ETHER_ADDR_LEN;
688 layer = "ieee802";
689 break;
690 case ARPHRD_INFINIBAND:
691 hlen = 20; /* RFC 4391, INFINIBAND_ALEN */
692 layer = "infiniband";
693 break;
694 case ARPHRD_IEEE1394:
695 hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
696 layer = "firewire";
697
698 /*
699 * Restrict too long hardware addresses.
700 * Currently we are capable of handling 20-byte
701 * addresses ( sizeof(lle->ll_addr) )
702 */
703 if (ar->ar_hln >= 20)
704 hlen = 16;
705 break;
706 default:
707 ARP_LOG(LOG_NOTICE,
708 "packet with unknown hardware format 0x%02d received on "
709 "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
710 m_freem(m);
711 return;
712 }
713
714 if (hlen != 0 && hlen != ar->ar_hln) {
715 ARP_LOG(LOG_NOTICE,
716 "packet with invalid %s address length %d received on %s\n",
717 layer, ar->ar_hln, if_name(ifp));
718 m_freem(m);
719 return;
720 }
721
722 ARPSTAT_INC(received);
723 switch (ntohs(ar->ar_pro)) {
724 #ifdef INET
725 case ETHERTYPE_IP:
726 in_arpinput(m);
727 return;
728 #endif
729 }
730 m_freem(m);
731 }
732
733 #ifdef INET
734 /*
735 * ARP for Internet protocols on 10 Mb/s Ethernet.
736 * Algorithm is that given in RFC 826.
737 * In addition, a sanity check is performed on the sender
738 * protocol address, to catch impersonators.
739 * We no longer handle negotiations for use of trailer protocol:
740 * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
741 * along with IP replies if we wanted trailers sent to us,
742 * and also sent them in response to IP replies.
743 * This allowed either end to announce the desire to receive
744 * trailer packets.
745 * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
746 * but formerly didn't normally send requests.
747 */
748 static int log_arp_wrong_iface = 1;
749 static int log_arp_movements = 1;
750 static int log_arp_permanent_modify = 1;
751 static int allow_multicast = 0;
752
753 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
754 &log_arp_wrong_iface, 0,
755 "log arp packets arriving on the wrong interface");
756 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
757 &log_arp_movements, 0,
758 "log arp replies from MACs different than the one in the cache");
759 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
760 &log_arp_permanent_modify, 0,
761 "log arp replies from MACs different than the one in the permanent arp entry");
762 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
763 &allow_multicast, 0, "accept multicast addresses");
764
765 static void
in_arpinput(struct mbuf * m)766 in_arpinput(struct mbuf *m)
767 {
768 struct arphdr *ah;
769 struct ifnet *ifp = m->m_pkthdr.rcvif;
770 struct llentry *la = NULL, *la_tmp;
771 struct ifaddr *ifa;
772 struct in_ifaddr *ia;
773 struct sockaddr sa;
774 struct in_addr isaddr, itaddr, myaddr;
775 u_int8_t *enaddr = NULL;
776 int op;
777 int bridged = 0, is_bridge = 0;
778 int carped;
779 struct sockaddr_in sin;
780 struct sockaddr *dst;
781 struct nhop_object *nh;
782 uint8_t linkhdr[LLE_MAX_LINKHDR];
783 struct route ro;
784 size_t linkhdrsize;
785 int lladdr_off;
786 int error;
787 char addrbuf[INET_ADDRSTRLEN];
788
789 NET_EPOCH_ASSERT();
790
791 sin.sin_len = sizeof(struct sockaddr_in);
792 sin.sin_family = AF_INET;
793 sin.sin_addr.s_addr = 0;
794
795 if (ifp->if_bridge)
796 bridged = 1;
797 if (ifp->if_type == IFT_BRIDGE)
798 is_bridge = 1;
799
800 /*
801 * We already have checked that mbuf contains enough contiguous data
802 * to hold entire arp message according to the arp header.
803 */
804 ah = mtod(m, struct arphdr *);
805
806 /*
807 * ARP is only for IPv4 so we can reject packets with
808 * a protocol length not equal to an IPv4 address.
809 */
810 if (ah->ar_pln != sizeof(struct in_addr)) {
811 ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
812 sizeof(struct in_addr));
813 goto drop;
814 }
815
816 if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
817 ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
818 ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
819 goto drop;
820 }
821
822 op = ntohs(ah->ar_op);
823 (void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
824 (void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
825
826 if (op == ARPOP_REPLY)
827 ARPSTAT_INC(rxreplies);
828
829 /*
830 * For a bridge, we want to check the address irrespective
831 * of the receive interface. (This will change slightly
832 * when we have clusters of interfaces).
833 */
834 CK_LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
835 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
836 ia->ia_ifp == ifp) &&
837 itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
838 (ia->ia_ifa.ifa_carp == NULL ||
839 (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
840 ifa_ref(&ia->ia_ifa);
841 goto match;
842 }
843 }
844 CK_LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
845 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
846 ia->ia_ifp == ifp) &&
847 isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
848 ifa_ref(&ia->ia_ifa);
849 goto match;
850 }
851
852 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia) \
853 (ia->ia_ifp->if_bridge == ifp->if_softc && \
854 !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) && \
855 addr == ia->ia_addr.sin_addr.s_addr)
856 /*
857 * Check the case when bridge shares its MAC address with
858 * some of its children, so packets are claimed by bridge
859 * itself (bridge_input() does it first), but they are really
860 * meant to be destined to the bridge member.
861 */
862 if (is_bridge) {
863 CK_LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
864 if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
865 ifa_ref(&ia->ia_ifa);
866 ifp = ia->ia_ifp;
867 goto match;
868 }
869 }
870 }
871 #undef BDG_MEMBER_MATCHES_ARP
872
873 /*
874 * No match, use the first inet address on the receive interface
875 * as a dummy address for the rest of the function.
876 */
877 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
878 if (ifa->ifa_addr->sa_family == AF_INET &&
879 (ifa->ifa_carp == NULL ||
880 (*carp_iamatch_p)(ifa, &enaddr))) {
881 ia = ifatoia(ifa);
882 ifa_ref(ifa);
883 goto match;
884 }
885
886 /*
887 * If bridging, fall back to using any inet address.
888 */
889 if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL)
890 goto drop;
891 ifa_ref(&ia->ia_ifa);
892 match:
893 if (!enaddr)
894 enaddr = (u_int8_t *)IF_LLADDR(ifp);
895 carped = (ia->ia_ifa.ifa_carp != NULL);
896 myaddr = ia->ia_addr.sin_addr;
897 ifa_free(&ia->ia_ifa);
898 if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
899 goto drop; /* it's from me, ignore it. */
900 if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
901 ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
902 "%s!\n", inet_ntoa_r(isaddr, addrbuf));
903 goto drop;
904 }
905
906 if (ifp->if_addrlen != ah->ar_hln) {
907 ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
908 "i/f %d (ignored)\n", ifp->if_addrlen,
909 (u_char *) ar_sha(ah), ":", ah->ar_hln,
910 ifp->if_addrlen);
911 goto drop;
912 }
913
914 /*
915 * Warn if another host is using the same IP address, but only if the
916 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
917 * case we suppress the warning to avoid false positive complaints of
918 * potential misconfiguration.
919 */
920 if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
921 myaddr.s_addr != 0) {
922 ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
923 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
924 inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
925 itaddr = myaddr;
926 ARPSTAT_INC(dupips);
927 goto reply;
928 }
929 if (ifp->if_flags & IFF_STATICARP)
930 goto reply;
931
932 bzero(&sin, sizeof(sin));
933 sin.sin_len = sizeof(struct sockaddr_in);
934 sin.sin_family = AF_INET;
935 sin.sin_addr = isaddr;
936 dst = (struct sockaddr *)&sin;
937 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
938 if (la != NULL)
939 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
940 else if (itaddr.s_addr == myaddr.s_addr) {
941 /*
942 * Request/reply to our address, but no lle exists yet.
943 * Calculate full link prepend to use in lle.
944 */
945 linkhdrsize = sizeof(linkhdr);
946 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
947 &linkhdrsize, &lladdr_off) != 0)
948 goto reply;
949
950 /* Allocate new entry */
951 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
952 if (la == NULL) {
953 /*
954 * lle creation may fail if source address belongs
955 * to non-directly connected subnet. However, we
956 * will try to answer the request instead of dropping
957 * frame.
958 */
959 goto reply;
960 }
961 lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
962 lladdr_off);
963
964 IF_AFDATA_WLOCK(ifp);
965 LLE_WLOCK(la);
966 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
967
968 /*
969 * Check if lle still does not exists.
970 * If it does, that means that we either
971 * 1) have configured it explicitly, via
972 * 1a) 'arp -s' static entry or
973 * 1b) interface address static record
974 * or
975 * 2) it was the result of sending first packet to-host
976 * or
977 * 3) it was another arp reply packet we handled in
978 * different thread.
979 *
980 * In all cases except 3) we definitely need to prefer
981 * existing lle. For the sake of simplicity, prefer any
982 * existing lle over newly-create one.
983 */
984 if (la_tmp == NULL)
985 lltable_link_entry(LLTABLE(ifp), la);
986 IF_AFDATA_WUNLOCK(ifp);
987
988 if (la_tmp == NULL) {
989 arp_mark_lle_reachable(la, ifp);
990 LLE_WUNLOCK(la);
991 } else {
992 /* Free newly-create entry and handle packet */
993 lltable_free_entry(LLTABLE(ifp), la);
994 la = la_tmp;
995 la_tmp = NULL;
996 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
997 /* arp_check_update_lle() returns @la unlocked */
998 }
999 la = NULL;
1000 }
1001 reply:
1002 if (op != ARPOP_REQUEST)
1003 goto drop;
1004 ARPSTAT_INC(rxrequests);
1005
1006 if (itaddr.s_addr == myaddr.s_addr) {
1007 /* Shortcut.. the receiving interface is the target. */
1008 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1009 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1010 } else {
1011 /*
1012 * Destination address is not ours. Check if
1013 * proxyarp entry exists or proxyarp is turned on globally.
1014 */
1015 struct llentry *lle;
1016
1017 sin.sin_addr = itaddr;
1018 lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1019
1020 if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1021 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1022 (void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1023 LLE_RUNLOCK(lle);
1024 } else {
1025 if (lle != NULL)
1026 LLE_RUNLOCK(lle);
1027
1028 if (!V_arp_proxyall)
1029 goto drop;
1030
1031 NET_EPOCH_ASSERT();
1032 nh = fib4_lookup(ifp->if_fib, itaddr, 0, 0, 0);
1033 if (nh == NULL)
1034 goto drop;
1035
1036 /*
1037 * Don't send proxies for nodes on the same interface
1038 * as this one came out of, or we'll get into a fight
1039 * over who claims what Ether address.
1040 */
1041 if (nh->nh_ifp == ifp)
1042 goto drop;
1043
1044 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1045 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1046
1047 /*
1048 * Also check that the node which sent the ARP packet
1049 * is on the interface we expect it to be on. This
1050 * avoids ARP chaos if an interface is connected to the
1051 * wrong network.
1052 */
1053
1054 nh = fib4_lookup(ifp->if_fib, isaddr, 0, 0, 0);
1055 if (nh == NULL)
1056 goto drop;
1057 if (nh->nh_ifp != ifp) {
1058 ARP_LOG(LOG_INFO, "proxy: ignoring request"
1059 " from %s via %s\n",
1060 inet_ntoa_r(isaddr, addrbuf),
1061 ifp->if_xname);
1062 goto drop;
1063 }
1064
1065 #ifdef DEBUG_PROXY
1066 printf("arp: proxying for %s\n",
1067 inet_ntoa_r(itaddr, addrbuf));
1068 #endif
1069 }
1070 }
1071
1072 if (itaddr.s_addr == myaddr.s_addr &&
1073 IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1074 /* RFC 3927 link-local IPv4; always reply by broadcast. */
1075 #ifdef DEBUG_LINKLOCAL
1076 printf("arp: sending reply for link-local addr %s\n",
1077 inet_ntoa_r(itaddr, addrbuf));
1078 #endif
1079 m->m_flags |= M_BCAST;
1080 m->m_flags &= ~M_MCAST;
1081 } else {
1082 /* default behaviour; never reply by broadcast. */
1083 m->m_flags &= ~(M_BCAST|M_MCAST);
1084 }
1085 (void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1086 (void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1087 ah->ar_op = htons(ARPOP_REPLY);
1088 ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1089 m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1090 m->m_pkthdr.len = m->m_len;
1091 m->m_pkthdr.rcvif = NULL;
1092 sa.sa_family = AF_ARP;
1093 sa.sa_len = 2;
1094
1095 /* Calculate link header for sending frame */
1096 bzero(&ro, sizeof(ro));
1097 linkhdrsize = sizeof(linkhdr);
1098 error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1099
1100 /*
1101 * arp_fillheader() may fail due to lack of support inside encap request
1102 * routing. This is not necessary an error, AF_ARP can/should be handled
1103 * by if_output().
1104 */
1105 if (error != 0 && error != EAFNOSUPPORT) {
1106 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1107 if_name(ifp), error);
1108 goto drop;
1109 }
1110
1111 ro.ro_prepend = linkhdr;
1112 ro.ro_plen = linkhdrsize;
1113 ro.ro_flags = 0;
1114
1115 m_clrprotoflags(m); /* Avoid confusing lower layers. */
1116 (*ifp->if_output)(ifp, m, &sa, &ro);
1117 ARPSTAT_INC(txreplies);
1118 return;
1119
1120 drop:
1121 m_freem(m);
1122 }
1123 #endif
1124
1125 static struct mbuf *
arp_grab_holdchain(struct llentry * la)1126 arp_grab_holdchain(struct llentry *la)
1127 {
1128 struct mbuf *chain;
1129
1130 LLE_WLOCK_ASSERT(la);
1131
1132 chain = la->la_hold;
1133 la->la_hold = NULL;
1134 la->la_numheld = 0;
1135
1136 return (chain);
1137 }
1138
1139 static void
arp_flush_holdchain(struct ifnet * ifp,struct llentry * la,struct mbuf * chain)1140 arp_flush_holdchain(struct ifnet *ifp, struct llentry *la, struct mbuf *chain)
1141 {
1142 struct mbuf *m_hold, *m_hold_next;
1143 struct sockaddr_in sin;
1144
1145 NET_EPOCH_ASSERT();
1146
1147 struct route ro = {
1148 .ro_prepend = la->r_linkdata,
1149 .ro_plen = la->r_hdrlen,
1150 };
1151
1152 lltable_fill_sa_entry(la, (struct sockaddr *)&sin);
1153
1154 for (m_hold = chain; m_hold != NULL; m_hold = m_hold_next) {
1155 m_hold_next = m_hold->m_nextpkt;
1156 m_hold->m_nextpkt = NULL;
1157 /* Avoid confusing lower layers. */
1158 m_clrprotoflags(m_hold);
1159 (*ifp->if_output)(ifp, m_hold, (struct sockaddr *)&sin, &ro);
1160 }
1161 }
1162
1163 /*
1164 * Checks received arp data against existing @la.
1165 * Updates lle state/performs notification if necessary.
1166 */
1167 static void
arp_check_update_lle(struct arphdr * ah,struct in_addr isaddr,struct ifnet * ifp,int bridged,struct llentry * la)1168 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1169 int bridged, struct llentry *la)
1170 {
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 LLE_WUNLOCK(la);
1220 return;
1221 }
1222
1223 /* Check if something has changed */
1224 if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1225 (la->la_flags & LLE_VALID) == 0) {
1226 /* Try to perform LLE update */
1227 if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1228 lladdr_off) == 0) {
1229 LLE_WUNLOCK(la);
1230 return;
1231 }
1232
1233 /* Clear fast path feedback request if set */
1234 llentry_mark_used(la);
1235 }
1236
1237 arp_mark_lle_reachable(la, ifp);
1238
1239 /*
1240 * The packets are all freed within the call to the output
1241 * routine.
1242 *
1243 * NB: The lock MUST be released before the call to the
1244 * output routine.
1245 */
1246 if (la->la_hold != NULL) {
1247 struct mbuf *chain;
1248
1249 chain = arp_grab_holdchain(la);
1250 LLE_WUNLOCK(la);
1251 arp_flush_holdchain(ifp, la, chain);
1252 } else
1253 LLE_WUNLOCK(la);
1254 }
1255
1256 static void
arp_mark_lle_reachable(struct llentry * la,struct ifnet * ifp)1257 arp_mark_lle_reachable(struct llentry *la, struct ifnet *ifp)
1258 {
1259 int canceled, wtime;
1260
1261 LLE_WLOCK_ASSERT(la);
1262
1263 la->ln_state = ARP_LLINFO_REACHABLE;
1264 EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1265
1266 if ((ifp->if_flags & IFF_STICKYARP) != 0)
1267 la->la_flags |= LLE_STATIC;
1268
1269 if (!(la->la_flags & LLE_STATIC)) {
1270 LLE_ADDREF(la);
1271 la->la_expire = time_uptime + V_arpt_keep;
1272 wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1273 if (wtime < 0)
1274 wtime = V_arpt_keep;
1275 canceled = callout_reset(&la->lle_timer,
1276 hz * wtime, arptimer, la);
1277 if (canceled)
1278 LLE_REMREF(la);
1279 }
1280 la->la_asked = 0;
1281 la->la_preempt = V_arp_maxtries;
1282 }
1283
1284 /*
1285 * Add permanent link-layer record for given interface address.
1286 */
1287 static __noinline void
arp_add_ifa_lle(struct ifnet * ifp,const struct sockaddr * dst)1288 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1289 {
1290 struct llentry *lle, *lle_tmp;
1291
1292 /*
1293 * Interface address LLE record is considered static
1294 * because kernel code relies on LLE_STATIC flag to check
1295 * if these entries can be rewriten by arp updates.
1296 */
1297 lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1298 if (lle == NULL) {
1299 log(LOG_INFO, "arp_ifinit: cannot create arp "
1300 "entry for interface address\n");
1301 return;
1302 }
1303
1304 IF_AFDATA_WLOCK(ifp);
1305 LLE_WLOCK(lle);
1306 /* Unlink any entry if exists */
1307 lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1308 if (lle_tmp != NULL)
1309 lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1310
1311 lltable_link_entry(LLTABLE(ifp), lle);
1312 IF_AFDATA_WUNLOCK(ifp);
1313
1314 if (lle_tmp != NULL)
1315 EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1316
1317 EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1318 LLE_WUNLOCK(lle);
1319 if (lle_tmp != NULL)
1320 lltable_free_entry(LLTABLE(ifp), lle_tmp);
1321 }
1322
1323 /*
1324 * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1325 * of valid values.
1326 */
1327 static int
sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)1328 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1329 {
1330 int error;
1331 int rexmit_count = *(int *)arg1;
1332
1333 error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1334
1335 /* Enforce limits on any new value that may have been set. */
1336 if (!error && req->newptr) {
1337 /* A new value was set. */
1338 if (rexmit_count < 0) {
1339 rexmit_count = 0;
1340 } else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1341 rexmit_count = MAX_GARP_RETRANSMITS;
1342 }
1343 *(int *)arg1 = rexmit_count;
1344 }
1345
1346 return (error);
1347 }
1348
1349 /*
1350 * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1351 * retransmit it again. A pending callout owns a reference to the ifa.
1352 */
1353 static void
garp_rexmit(void * arg)1354 garp_rexmit(void *arg)
1355 {
1356 struct epoch_tracker et;
1357 struct in_ifaddr *ia = arg;
1358
1359 if (callout_pending(&ia->ia_garp_timer) ||
1360 !callout_active(&ia->ia_garp_timer)) {
1361 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1362 ifa_free(&ia->ia_ifa);
1363 return;
1364 }
1365
1366 NET_EPOCH_ENTER(et);
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 >= V_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 NET_EPOCH_EXIT(et);
1399 }
1400
1401 /*
1402 * Start the GARP retransmit timer.
1403 *
1404 * A single GARP is always transmitted when an IPv4 address is added
1405 * to an interface and that is usually sufficient. However, in some
1406 * circumstances, such as when a shared address is passed between
1407 * cluster nodes, this single GARP may occasionally be dropped or
1408 * lost. This can lead to neighbors on the network link working with a
1409 * stale ARP cache and sending packets destined for that address to
1410 * the node that previously owned the address, which may not respond.
1411 *
1412 * To avoid this situation, GARP retransmits can be enabled by setting
1413 * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1414 * than zero. The setting represents the maximum number of
1415 * retransmissions. The interval between retransmissions is calculated
1416 * using an exponential backoff algorithm, doubling each time, so the
1417 * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1418 */
1419 static void
garp_timer_start(struct ifaddr * ifa)1420 garp_timer_start(struct ifaddr *ifa)
1421 {
1422 struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1423
1424 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1425 ia->ia_garp_count = 0;
1426 if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1427 garp_rexmit, ia) == 0) {
1428 ifa_ref(ifa);
1429 }
1430 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1431 }
1432
1433 void
arp_ifinit(struct ifnet * ifp,struct ifaddr * ifa)1434 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1435 {
1436 struct epoch_tracker et;
1437 const struct sockaddr_in *dst_in;
1438 const struct sockaddr *dst;
1439
1440 if (ifa->ifa_carp != NULL)
1441 return;
1442
1443 dst = ifa->ifa_addr;
1444 dst_in = (const struct sockaddr_in *)dst;
1445
1446 if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1447 return;
1448 NET_EPOCH_ENTER(et);
1449 arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1450 NET_EPOCH_EXIT(et);
1451 if (V_garp_rexmit_count > 0) {
1452 garp_timer_start(ifa);
1453 }
1454
1455 arp_add_ifa_lle(ifp, dst);
1456 }
1457
1458 void
arp_announce_ifaddr(struct ifnet * ifp,struct in_addr addr,u_char * enaddr)1459 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1460 {
1461
1462 if (ntohl(addr.s_addr) != INADDR_ANY)
1463 arprequest(ifp, &addr, &addr, enaddr);
1464 }
1465
1466 /*
1467 * Sends gratuitous ARPs for each ifaddr to notify other
1468 * nodes about the address change.
1469 */
1470 static __noinline void
arp_handle_ifllchange(struct ifnet * ifp)1471 arp_handle_ifllchange(struct ifnet *ifp)
1472 {
1473 struct ifaddr *ifa;
1474
1475 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1476 if (ifa->ifa_addr->sa_family == AF_INET)
1477 arp_ifinit(ifp, ifa);
1478 }
1479 }
1480
1481 /*
1482 * A handler for interface link layer address change event.
1483 */
1484 static void
arp_iflladdr(void * arg __unused,struct ifnet * ifp)1485 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1486 {
1487 /* if_bridge can update its lladdr during if_vmove(), after we've done
1488 * if_detach_internal()/dom_ifdetach(). */
1489 if (ifp->if_afdata[AF_INET] == NULL)
1490 return;
1491
1492 lltable_update_ifaddr(LLTABLE(ifp));
1493
1494 if ((ifp->if_flags & IFF_UP) != 0)
1495 arp_handle_ifllchange(ifp);
1496 }
1497
1498 static void
vnet_arp_init(void)1499 vnet_arp_init(void)
1500 {
1501
1502 if (IS_DEFAULT_VNET(curvnet)) {
1503 netisr_register(&arp_nh);
1504 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1505 arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1506 }
1507 #ifdef VIMAGE
1508 else
1509 netisr_register_vnet(&arp_nh);
1510 #endif
1511 }
1512 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1513 vnet_arp_init, NULL);
1514
1515 #ifdef VIMAGE
1516 /*
1517 * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1518 * lookups after destroying the hash. Ideally this would go on SI_ORDER_3.5.
1519 */
1520 static void
vnet_arp_destroy(__unused void * arg)1521 vnet_arp_destroy(__unused void *arg)
1522 {
1523
1524 netisr_unregister_vnet(&arp_nh);
1525 }
1526 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1527 vnet_arp_destroy, NULL);
1528 #endif
1529