1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1988, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * @(#)if_ether.c 8.1 (Berkeley) 6/10/93
32 */
33
34 /*
35 * Ethernet address resolution protocol.
36 * TODO:
37 * add "inuse/lock" bit (or ref. count) along with valid bit
38 */
39
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42
43 #include "opt_inet.h"
44
45 #include <sys/param.h>
46 #include <sys/eventhandler.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/queue.h>
50 #include <sys/sysctl.h>
51 #include <sys/systm.h>
52 #include <sys/mbuf.h>
53 #include <sys/malloc.h>
54 #include <sys/proc.h>
55 #include <sys/rmlock.h>
56 #include <sys/socket.h>
57 #include <sys/syslog.h>
58
59 #include <net/if.h>
60 #include <net/if_var.h>
61 #include <net/if_dl.h>
62 #include <net/if_types.h>
63 #include <net/netisr.h>
64 #include <net/ethernet.h>
65 #include <net/route.h>
66 #include <net/route/nhop.h>
67 #include <net/vnet.h>
68
69 #include <netinet/in.h>
70 #include <netinet/in_fib.h>
71 #include <netinet/in_var.h>
72 #include <net/if_llatbl.h>
73 #include <netinet/if_ether.h>
74 #ifdef INET
75 #include <netinet/ip_carp.h>
76 #endif
77
78 #include <security/mac/mac_framework.h>
79
80 #define SIN(s) ((const struct sockaddr_in *)(s))
81
82 static struct timeval arp_lastlog;
83 static int arp_curpps;
84 static int arp_maxpps = 1;
85
86 /* Simple ARP state machine */
87 enum arp_llinfo_state {
88 ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */
89 ARP_LLINFO_REACHABLE, /* LLE is valid */
90 ARP_LLINFO_VERIFY, /* LLE is valid, need refresh */
91 ARP_LLINFO_DELETED, /* LLE is deleted */
92 };
93
94 SYSCTL_DECL(_net_link_ether);
95 static SYSCTL_NODE(_net_link_ether, PF_INET, inet,
96 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
97 "");
98 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp,
99 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
100 "");
101
102 /* timer values */
103 VNET_DEFINE_STATIC(int, arpt_keep) = (20*60); /* once resolved, good for 20
104 * minutes */
105 VNET_DEFINE_STATIC(int, arp_maxtries) = 5;
106 VNET_DEFINE_STATIC(int, arp_proxyall) = 0;
107 VNET_DEFINE_STATIC(int, arpt_down) = 20; /* keep incomplete entries for
108 * 20 seconds */
109 VNET_DEFINE_STATIC(int, arpt_rexmit) = 1; /* retransmit arp entries, sec*/
110 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat); /* ARP statistics, see if_arp.h */
111 VNET_PCPUSTAT_SYSINIT(arpstat);
112
113 #ifdef VIMAGE
114 VNET_PCPUSTAT_SYSUNINIT(arpstat);
115 #endif /* VIMAGE */
116
117 VNET_DEFINE_STATIC(int, arp_maxhold) = 16;
118
119 #define V_arpt_keep VNET(arpt_keep)
120 #define V_arpt_down VNET(arpt_down)
121 #define V_arpt_rexmit VNET(arpt_rexmit)
122 #define V_arp_maxtries VNET(arp_maxtries)
123 #define V_arp_proxyall VNET(arp_proxyall)
124 #define V_arp_maxhold VNET(arp_maxhold)
125
126 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
127 &VNET_NAME(arpt_keep), 0,
128 "ARP entry lifetime in seconds");
129 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
130 &VNET_NAME(arp_maxtries), 0,
131 "ARP resolution attempts before returning error");
132 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
133 &VNET_NAME(arp_proxyall), 0,
134 "Enable proxy ARP for all suitable requests");
135 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
136 &VNET_NAME(arpt_down), 0,
137 "Incomplete ARP entry lifetime in seconds");
138 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
139 arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
140 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
141 &VNET_NAME(arp_maxhold), 0,
142 "Number of packets to hold per ARP entry");
143 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
144 CTLFLAG_RW, &arp_maxpps, 0,
145 "Maximum number of remotely triggered ARP messages that can be "
146 "logged per second");
147
148 /*
149 * Due to the exponential backoff algorithm used for the interval between GARP
150 * retransmissions, the maximum number of retransmissions is limited for
151 * sanity. This limit corresponds to a maximum interval between retransmissions
152 * of 2^16 seconds ~= 18 hours.
153 *
154 * Making this limit more dynamic is more complicated than worthwhile,
155 * especially since sending out GARPs spaced days apart would be of little
156 * use. A maximum dynamic limit would look something like:
157 *
158 * const int max = fls(INT_MAX / hz) - 1;
159 */
160 #define MAX_GARP_RETRANSMITS 16
161 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS);
162 static int garp_rexmit_count = 0; /* GARP retransmission setting. */
163
164 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count,
165 CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE,
166 &garp_rexmit_count, 0, sysctl_garp_rexmit, "I",
167 "Number of times to retransmit GARP packets;"
168 " 0 to disable, maximum of 16");
169
170 VNET_DEFINE_STATIC(int, arp_log_level) = LOG_INFO; /* Min. log(9) level. */
171 #define V_arp_log_level VNET(arp_log_level)
172 SYSCTL_INT(_net_link_ether_arp, OID_AUTO, log_level, CTLFLAG_VNET | CTLFLAG_RW,
173 &VNET_NAME(arp_log_level), 0,
174 "Minimum log(9) level for recording rate limited arp log messages. "
175 "The higher will be log more (emerg=0, info=6 (default), debug=7).");
176 #define ARP_LOG(pri, ...) do { \
177 if ((pri) <= V_arp_log_level && \
178 ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps)) \
179 log((pri), "arp: " __VA_ARGS__); \
180 } while (0)
181
182 static void arpintr(struct mbuf *);
183 static void arptimer(void *);
184 #ifdef INET
185 static void in_arpinput(struct mbuf *);
186 #endif
187
188 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
189 struct ifnet *ifp, int bridged, struct llentry *la);
190 static void arp_mark_lle_reachable(struct llentry *la);
191 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
192
193 static eventhandler_tag iflladdr_tag;
194
195 static const struct netisr_handler arp_nh = {
196 .nh_name = "arp",
197 .nh_handler = arpintr,
198 .nh_proto = NETISR_ARP,
199 .nh_policy = NETISR_POLICY_SOURCE,
200 };
201
202 /*
203 * Timeout routine. Age arp_tab entries periodically.
204 */
205 static void
arptimer(void * arg)206 arptimer(void *arg)
207 {
208 struct llentry *lle = (struct llentry *)arg;
209 struct ifnet *ifp;
210
211 if (lle->la_flags & LLE_STATIC) {
212 return;
213 }
214 LLE_WLOCK(lle);
215 if (callout_pending(&lle->lle_timer)) {
216 /*
217 * Here we are a bit odd here in the treatment of
218 * active/pending. If the pending bit is set, it got
219 * rescheduled before I ran. The active
220 * bit we ignore, since if it was stopped
221 * in ll_tablefree() and was currently running
222 * it would have return 0 so the code would
223 * not have deleted it since the callout could
224 * not be stopped so we want to go through
225 * with the delete here now. If the callout
226 * was restarted, the pending bit will be back on and
227 * we just want to bail since the callout_reset would
228 * return 1 and our reference would have been removed
229 * by arpresolve() below.
230 */
231 LLE_WUNLOCK(lle);
232 return;
233 }
234 ifp = lle->lle_tbl->llt_ifp;
235 CURVNET_SET(ifp->if_vnet);
236
237 switch (lle->ln_state) {
238 case ARP_LLINFO_REACHABLE:
239
240 /*
241 * Expiration time is approaching.
242 * Request usage feedback from the datapath.
243 * Change state and re-schedule ourselves.
244 */
245 llentry_request_feedback(lle);
246 lle->ln_state = ARP_LLINFO_VERIFY;
247 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
248 LLE_WUNLOCK(lle);
249 CURVNET_RESTORE();
250 return;
251 case ARP_LLINFO_VERIFY:
252 if (llentry_get_hittime(lle) > 0 && lle->la_preempt > 0) {
253 /* Entry was used, issue refresh request */
254 struct epoch_tracker et;
255 struct in_addr dst;
256
257 dst = lle->r_l3addr.addr4;
258 lle->la_preempt--;
259 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
260 LLE_WUNLOCK(lle);
261 NET_EPOCH_ENTER(et);
262 arprequest(ifp, NULL, &dst, NULL);
263 NET_EPOCH_EXIT(et);
264 CURVNET_RESTORE();
265 return;
266 }
267 /* Nothing happened. Reschedule if not too late */
268 if (lle->la_expire > time_uptime) {
269 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
270 LLE_WUNLOCK(lle);
271 CURVNET_RESTORE();
272 return;
273 }
274 break;
275 case ARP_LLINFO_INCOMPLETE:
276 case ARP_LLINFO_DELETED:
277 break;
278 }
279
280 if ((lle->la_flags & LLE_DELETED) == 0) {
281 int evt;
282
283 if (lle->la_flags & LLE_VALID)
284 evt = LLENTRY_EXPIRED;
285 else
286 evt = LLENTRY_TIMEDOUT;
287 EVENTHANDLER_INVOKE(lle_event, lle, evt);
288 }
289
290 callout_stop(&lle->lle_timer);
291
292 /* XXX: LOR avoidance. We still have ref on lle. */
293 LLE_WUNLOCK(lle);
294 IF_AFDATA_LOCK(ifp);
295 LLE_WLOCK(lle);
296
297 /* Guard against race with other llentry_free(). */
298 if (lle->la_flags & LLE_LINKED) {
299 LLE_REMREF(lle);
300 lltable_unlink_entry(lle->lle_tbl, lle);
301 }
302 IF_AFDATA_UNLOCK(ifp);
303
304 size_t pkts_dropped = llentry_free(lle);
305
306 ARPSTAT_ADD(dropped, pkts_dropped);
307 ARPSTAT_INC(timeouts);
308
309 CURVNET_RESTORE();
310 }
311
312 /*
313 * Stores link-layer header for @ifp in format suitable for if_output()
314 * into buffer @buf. Resulting header length is stored in @bufsize.
315 *
316 * Returns 0 on success.
317 */
318 static int
arp_fillheader(struct ifnet * ifp,struct arphdr * ah,int bcast,u_char * buf,size_t * bufsize)319 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
320 size_t *bufsize)
321 {
322 struct if_encap_req ereq;
323 int error;
324
325 bzero(buf, *bufsize);
326 bzero(&ereq, sizeof(ereq));
327 ereq.buf = buf;
328 ereq.bufsize = *bufsize;
329 ereq.rtype = IFENCAP_LL;
330 ereq.family = AF_ARP;
331 ereq.lladdr = ar_tha(ah);
332 ereq.hdata = (u_char *)ah;
333 if (bcast)
334 ereq.flags = IFENCAP_FLAG_BROADCAST;
335 error = ifp->if_requestencap(ifp, &ereq);
336 if (error == 0)
337 *bufsize = ereq.bufsize;
338
339 return (error);
340 }
341
342 /*
343 * Broadcast an ARP request. Caller specifies:
344 * - arp header source ip address
345 * - arp header target ip address
346 * - arp header source ethernet address
347 */
348 static int
arprequest_internal(struct ifnet * ifp,const struct in_addr * sip,const struct in_addr * tip,u_char * enaddr)349 arprequest_internal(struct ifnet *ifp, const struct in_addr *sip,
350 const struct in_addr *tip, u_char *enaddr)
351 {
352 struct mbuf *m;
353 struct arphdr *ah;
354 struct sockaddr sa;
355 u_char *carpaddr = NULL;
356 uint8_t linkhdr[LLE_MAX_LINKHDR];
357 size_t linkhdrsize;
358 struct route ro;
359 int error;
360
361 NET_EPOCH_ASSERT();
362
363 if (sip == NULL) {
364 /*
365 * The caller did not supply a source address, try to find
366 * a compatible one among those assigned to this interface.
367 */
368 struct ifaddr *ifa;
369
370 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
371 if (ifa->ifa_addr->sa_family != AF_INET)
372 continue;
373
374 if (ifa->ifa_carp) {
375 if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
376 continue;
377 sip = &IA_SIN(ifa)->sin_addr;
378 } else {
379 carpaddr = NULL;
380 sip = &IA_SIN(ifa)->sin_addr;
381 }
382
383 if (0 == ((sip->s_addr ^ tip->s_addr) &
384 IA_MASKSIN(ifa)->sin_addr.s_addr))
385 break; /* found it. */
386 }
387 if (sip == NULL) {
388 printf("%s: cannot find matching address\n", __func__);
389 return (EADDRNOTAVAIL);
390 }
391 }
392 if (enaddr == NULL)
393 enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
394
395 if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
396 return (ENOMEM);
397 m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
398 2 * ifp->if_addrlen;
399 m->m_pkthdr.len = m->m_len;
400 M_ALIGN(m, m->m_len);
401 ah = mtod(m, struct arphdr *);
402 bzero((caddr_t)ah, m->m_len);
403 #ifdef MAC
404 mac_netinet_arp_send(ifp, m);
405 #endif
406 ah->ar_pro = htons(ETHERTYPE_IP);
407 ah->ar_hln = ifp->if_addrlen; /* hardware address length */
408 ah->ar_pln = sizeof(struct in_addr); /* protocol address length */
409 ah->ar_op = htons(ARPOP_REQUEST);
410 bcopy(enaddr, ar_sha(ah), ah->ar_hln);
411 bcopy(sip, ar_spa(ah), ah->ar_pln);
412 bcopy(tip, ar_tpa(ah), ah->ar_pln);
413 sa.sa_family = AF_ARP;
414 sa.sa_len = 2;
415
416 /* Calculate link header for sending frame */
417 bzero(&ro, sizeof(ro));
418 linkhdrsize = sizeof(linkhdr);
419 error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
420 if (error != 0 && error != EAFNOSUPPORT) {
421 m_freem(m);
422 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
423 if_name(ifp), error);
424 return (error);
425 }
426
427 ro.ro_prepend = linkhdr;
428 ro.ro_plen = linkhdrsize;
429 ro.ro_flags = 0;
430
431 m->m_flags |= M_BCAST;
432 m_clrprotoflags(m); /* Avoid confusing lower layers. */
433 error = (*ifp->if_output)(ifp, m, &sa, &ro);
434 ARPSTAT_INC(txrequests);
435 if (error) {
436 ARPSTAT_INC(txerrors);
437 ARP_LOG(LOG_DEBUG, "Failed to send ARP packet on %s: %d\n",
438 if_name(ifp), error);
439 }
440 return (error);
441 }
442
443 void
arprequest(struct ifnet * ifp,const struct in_addr * sip,const struct in_addr * tip,u_char * enaddr)444 arprequest(struct ifnet *ifp, const struct in_addr *sip,
445 const struct in_addr *tip, u_char *enaddr)
446 {
447
448 (void) arprequest_internal(ifp, sip, tip, enaddr);
449 }
450
451 /*
452 * Resolve an IP address into an ethernet address - heavy version.
453 * Used internally by arpresolve().
454 * We have already checked that we can't use an existing lle without
455 * modification so we have to acquire an LLE_EXCLUSIVE lle lock.
456 *
457 * On success, desten and pflags are filled in and the function returns 0;
458 * If the packet must be held pending resolution, we return EWOULDBLOCK
459 * On other errors, we return the corresponding error code.
460 * Note that m_freem() handles NULL.
461 */
462 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)463 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
464 const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
465 struct llentry **plle)
466 {
467 struct llentry *la = NULL, *la_tmp;
468 struct mbuf *curr = NULL;
469 struct mbuf *next = NULL;
470 int error, renew;
471 char *lladdr;
472 int ll_len;
473
474 NET_EPOCH_ASSERT();
475
476 if (pflags != NULL)
477 *pflags = 0;
478 if (plle != NULL)
479 *plle = NULL;
480
481 if ((flags & LLE_CREATE) == 0)
482 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
483 if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
484 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
485 if (la == NULL) {
486 char addrbuf[INET_ADDRSTRLEN];
487
488 log(LOG_DEBUG,
489 "arpresolve: can't allocate llinfo for %s on %s\n",
490 inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
491 if_name(ifp));
492 m_freem(m);
493 return (EINVAL);
494 }
495
496 IF_AFDATA_WLOCK(ifp);
497 LLE_WLOCK(la);
498 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
499 /* Prefer ANY existing lle over newly-created one */
500 if (la_tmp == NULL)
501 lltable_link_entry(LLTABLE(ifp), la);
502 IF_AFDATA_WUNLOCK(ifp);
503 if (la_tmp != NULL) {
504 lltable_free_entry(LLTABLE(ifp), la);
505 la = la_tmp;
506 }
507 }
508 if (la == NULL) {
509 m_freem(m);
510 return (EINVAL);
511 }
512
513 if ((la->la_flags & LLE_VALID) &&
514 ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
515 if (flags & LLE_ADDRONLY) {
516 lladdr = la->ll_addr;
517 ll_len = ifp->if_addrlen;
518 } else {
519 lladdr = la->r_linkdata;
520 ll_len = la->r_hdrlen;
521 }
522 bcopy(lladdr, desten, ll_len);
523
524 /* Notify LLE code that the entry was used by datapath */
525 llentry_provide_feedback(la);
526 if (pflags != NULL)
527 *pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
528 if (plle) {
529 LLE_ADDREF(la);
530 *plle = la;
531 }
532 LLE_WUNLOCK(la);
533 return (0);
534 }
535
536 renew = (la->la_asked == 0 || la->la_expire != time_uptime);
537 /*
538 * There is an arptab entry, but no ethernet address
539 * response yet. Add the mbuf to the list, dropping
540 * the oldest packet if we have exceeded the system
541 * setting.
542 */
543 if (m != NULL) {
544 if (la->la_numheld >= V_arp_maxhold) {
545 if (la->la_hold != NULL) {
546 next = la->la_hold->m_nextpkt;
547 m_freem(la->la_hold);
548 la->la_hold = next;
549 la->la_numheld--;
550 ARPSTAT_INC(dropped);
551 }
552 }
553 if (la->la_hold != NULL) {
554 curr = la->la_hold;
555 while (curr->m_nextpkt != NULL)
556 curr = curr->m_nextpkt;
557 curr->m_nextpkt = m;
558 } else
559 la->la_hold = m;
560 la->la_numheld++;
561 }
562 /*
563 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
564 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
565 * if we have already sent arp_maxtries ARP requests. Retransmit the
566 * ARP request, but not faster than one request per second.
567 */
568 if (la->la_asked < V_arp_maxtries)
569 error = EWOULDBLOCK; /* First request. */
570 else
571 error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
572
573 if (renew) {
574 int canceled, e;
575
576 LLE_ADDREF(la);
577 la->la_expire = time_uptime;
578 canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
579 arptimer, la);
580 if (canceled)
581 LLE_REMREF(la);
582 la->la_asked++;
583 LLE_WUNLOCK(la);
584 e = arprequest_internal(ifp, NULL, &SIN(dst)->sin_addr, NULL);
585 /*
586 * Only overwrite 'error' in case of error; in case of success
587 * the proper return value was already set above.
588 */
589 if (e != 0)
590 return (e);
591 return (error);
592 }
593
594 LLE_WUNLOCK(la);
595 return (error);
596 }
597
598 /*
599 * Lookups link header based on an IP address.
600 * On input:
601 * ifp is the interface we use
602 * is_gw != 0 if @dst represents gateway to some destination
603 * m is the mbuf. May be NULL if we don't have a packet.
604 * dst is the next hop,
605 * desten is the storage to put LL header.
606 * flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
607 *
608 * On success, full/partial link header and flags are filled in and
609 * the function returns 0.
610 * If the packet must be held pending resolution, we return EWOULDBLOCK
611 * On other errors, we return the corresponding error code.
612 * Note that m_freem() handles NULL.
613 */
614 int
arpresolve(struct ifnet * ifp,int is_gw,struct mbuf * m,const struct sockaddr * dst,u_char * desten,uint32_t * pflags,struct llentry ** plle)615 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
616 const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
617 struct llentry **plle)
618 {
619 struct llentry *la = NULL;
620
621 NET_EPOCH_ASSERT();
622
623 if (pflags != NULL)
624 *pflags = 0;
625 if (plle != NULL)
626 *plle = NULL;
627
628 if (m != NULL) {
629 if (m->m_flags & M_BCAST) {
630 /* broadcast */
631 (void)memcpy(desten,
632 ifp->if_broadcastaddr, ifp->if_addrlen);
633 return (0);
634 }
635 if (m->m_flags & M_MCAST) {
636 /* multicast */
637 ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
638 return (0);
639 }
640 }
641
642 la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
643 if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
644 /* Entry found, let's copy lle info */
645 bcopy(la->r_linkdata, desten, la->r_hdrlen);
646 if (pflags != NULL)
647 *pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
648 /* Notify the LLE handling code that the entry was used. */
649 llentry_provide_feedback(la);
650 if (plle) {
651 LLE_ADDREF(la);
652 *plle = la;
653 LLE_WUNLOCK(la);
654 }
655 return (0);
656 }
657 if (plle && la)
658 LLE_WUNLOCK(la);
659
660 return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
661 desten, pflags, plle));
662 }
663
664 /*
665 * Common length and type checks are done here,
666 * then the protocol-specific routine is called.
667 */
668 static void
arpintr(struct mbuf * m)669 arpintr(struct mbuf *m)
670 {
671 struct arphdr *ar;
672 struct ifnet *ifp;
673 char *layer;
674 int hlen;
675
676 ifp = m->m_pkthdr.rcvif;
677
678 if (m->m_len < sizeof(struct arphdr) &&
679 ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
680 ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
681 if_name(ifp));
682 return;
683 }
684 ar = mtod(m, struct arphdr *);
685
686 /* Check if length is sufficient */
687 if (m->m_len < arphdr_len(ar)) {
688 m = m_pullup(m, arphdr_len(ar));
689 if (m == NULL) {
690 ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
691 if_name(ifp));
692 return;
693 }
694 ar = mtod(m, struct arphdr *);
695 }
696
697 hlen = 0;
698 layer = "";
699 switch (ntohs(ar->ar_hrd)) {
700 case ARPHRD_ETHER:
701 hlen = ETHER_ADDR_LEN; /* RFC 826 */
702 layer = "ethernet";
703 break;
704 case ARPHRD_INFINIBAND:
705 hlen = 20; /* RFC 4391, INFINIBAND_ALEN */
706 layer = "infiniband";
707 break;
708 case ARPHRD_IEEE1394:
709 hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
710 layer = "firewire";
711
712 /*
713 * Restrict too long hardware addresses.
714 * Currently we are capable of handling 20-byte
715 * addresses ( sizeof(lle->ll_addr) )
716 */
717 if (ar->ar_hln >= 20)
718 hlen = 16;
719 break;
720 default:
721 ARP_LOG(LOG_NOTICE,
722 "packet with unknown hardware format 0x%02d received on "
723 "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
724 m_freem(m);
725 return;
726 }
727
728 if (hlen != 0 && hlen != ar->ar_hln) {
729 ARP_LOG(LOG_NOTICE,
730 "packet with invalid %s address length %d received on %s\n",
731 layer, ar->ar_hln, if_name(ifp));
732 m_freem(m);
733 return;
734 }
735
736 ARPSTAT_INC(received);
737 switch (ntohs(ar->ar_pro)) {
738 #ifdef INET
739 case ETHERTYPE_IP:
740 in_arpinput(m);
741 return;
742 #endif
743 }
744 m_freem(m);
745 }
746
747 #ifdef INET
748 /*
749 * ARP for Internet protocols on 10 Mb/s Ethernet.
750 * Algorithm is that given in RFC 826.
751 * In addition, a sanity check is performed on the sender
752 * protocol address, to catch impersonators.
753 * We no longer handle negotiations for use of trailer protocol:
754 * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
755 * along with IP replies if we wanted trailers sent to us,
756 * and also sent them in response to IP replies.
757 * This allowed either end to announce the desire to receive
758 * trailer packets.
759 * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
760 * but formerly didn't normally send requests.
761 */
762 static int log_arp_wrong_iface = 1;
763 static int log_arp_movements = 1;
764 static int log_arp_permanent_modify = 1;
765 static int allow_multicast = 0;
766
767 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
768 &log_arp_wrong_iface, 0,
769 "log arp packets arriving on the wrong interface");
770 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
771 &log_arp_movements, 0,
772 "log arp replies from MACs different than the one in the cache");
773 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
774 &log_arp_permanent_modify, 0,
775 "log arp replies from MACs different than the one in the permanent arp entry");
776 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
777 &allow_multicast, 0, "accept multicast addresses");
778
779 static void
in_arpinput(struct mbuf * m)780 in_arpinput(struct mbuf *m)
781 {
782 struct rm_priotracker in_ifa_tracker;
783 struct arphdr *ah;
784 struct ifnet *ifp = m->m_pkthdr.rcvif;
785 struct llentry *la = NULL, *la_tmp;
786 struct ifaddr *ifa;
787 struct in_ifaddr *ia;
788 struct sockaddr sa;
789 struct in_addr isaddr, itaddr, myaddr;
790 u_int8_t *enaddr = NULL;
791 int op;
792 int bridged = 0, is_bridge = 0;
793 int carped;
794 struct sockaddr_in sin;
795 struct sockaddr *dst;
796 struct nhop_object *nh;
797 uint8_t linkhdr[LLE_MAX_LINKHDR];
798 struct route ro;
799 size_t linkhdrsize;
800 int lladdr_off;
801 int error;
802 char addrbuf[INET_ADDRSTRLEN];
803
804 NET_EPOCH_ASSERT();
805
806 sin.sin_len = sizeof(struct sockaddr_in);
807 sin.sin_family = AF_INET;
808 sin.sin_addr.s_addr = 0;
809
810 if (ifp->if_bridge)
811 bridged = 1;
812 if (ifp->if_type == IFT_BRIDGE)
813 is_bridge = 1;
814
815 /*
816 * We already have checked that mbuf contains enough contiguous data
817 * to hold entire arp message according to the arp header.
818 */
819 ah = mtod(m, struct arphdr *);
820
821 /*
822 * ARP is only for IPv4 so we can reject packets with
823 * a protocol length not equal to an IPv4 address.
824 */
825 if (ah->ar_pln != sizeof(struct in_addr)) {
826 ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
827 sizeof(struct in_addr));
828 goto drop;
829 }
830
831 if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
832 ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
833 ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
834 goto drop;
835 }
836
837 op = ntohs(ah->ar_op);
838 (void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
839 (void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
840
841 if (op == ARPOP_REPLY)
842 ARPSTAT_INC(rxreplies);
843
844 /*
845 * For a bridge, we want to check the address irrespective
846 * of the receive interface. (This will change slightly
847 * when we have clusters of interfaces).
848 */
849 IN_IFADDR_RLOCK(&in_ifa_tracker);
850 LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
851 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
852 ia->ia_ifp == ifp) &&
853 itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
854 (ia->ia_ifa.ifa_carp == NULL ||
855 (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
856 ifa_ref(&ia->ia_ifa);
857 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
858 goto match;
859 }
860 }
861 LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
862 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
863 ia->ia_ifp == ifp) &&
864 isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
865 ifa_ref(&ia->ia_ifa);
866 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
867 goto match;
868 }
869
870 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia) \
871 (ia->ia_ifp->if_bridge == ifp->if_softc && \
872 !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) && \
873 addr == ia->ia_addr.sin_addr.s_addr)
874 /*
875 * Check the case when bridge shares its MAC address with
876 * some of its children, so packets are claimed by bridge
877 * itself (bridge_input() does it first), but they are really
878 * meant to be destined to the bridge member.
879 */
880 if (is_bridge) {
881 LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
882 if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
883 ifa_ref(&ia->ia_ifa);
884 ifp = ia->ia_ifp;
885 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
886 goto match;
887 }
888 }
889 }
890 #undef BDG_MEMBER_MATCHES_ARP
891 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
892
893 /*
894 * No match, use the first inet address on the receive interface
895 * as a dummy address for the rest of the function.
896 */
897 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
898 if (ifa->ifa_addr->sa_family == AF_INET &&
899 (ifa->ifa_carp == NULL ||
900 (*carp_iamatch_p)(ifa, &enaddr))) {
901 ia = ifatoia(ifa);
902 ifa_ref(ifa);
903 goto match;
904 }
905
906 /*
907 * If bridging, fall back to using any inet address.
908 */
909 IN_IFADDR_RLOCK(&in_ifa_tracker);
910 if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL) {
911 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
912 goto drop;
913 }
914 ifa_ref(&ia->ia_ifa);
915 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
916 match:
917 if (!enaddr)
918 enaddr = (u_int8_t *)IF_LLADDR(ifp);
919 carped = (ia->ia_ifa.ifa_carp != NULL);
920 myaddr = ia->ia_addr.sin_addr;
921 ifa_free(&ia->ia_ifa);
922 if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
923 goto drop; /* it's from me, ignore it. */
924 if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
925 ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
926 "%s!\n", inet_ntoa_r(isaddr, addrbuf));
927 goto drop;
928 }
929
930 if (ifp->if_addrlen != ah->ar_hln) {
931 ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
932 "i/f %d (ignored)\n", ifp->if_addrlen,
933 (u_char *) ar_sha(ah), ":", ah->ar_hln,
934 ifp->if_addrlen);
935 goto drop;
936 }
937
938 /*
939 * Warn if another host is using the same IP address, but only if the
940 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
941 * case we suppress the warning to avoid false positive complaints of
942 * potential misconfiguration.
943 */
944 if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
945 myaddr.s_addr != 0) {
946 ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
947 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
948 inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
949 itaddr = myaddr;
950 ARPSTAT_INC(dupips);
951 goto reply;
952 }
953 if (ifp->if_flags & IFF_STATICARP)
954 goto reply;
955
956 bzero(&sin, sizeof(sin));
957 sin.sin_len = sizeof(struct sockaddr_in);
958 sin.sin_family = AF_INET;
959 sin.sin_addr = isaddr;
960 dst = (struct sockaddr *)&sin;
961 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
962 if (la != NULL)
963 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
964 else if (itaddr.s_addr == myaddr.s_addr) {
965 /*
966 * Request/reply to our address, but no lle exists yet.
967 * Calculate full link prepend to use in lle.
968 */
969 linkhdrsize = sizeof(linkhdr);
970 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
971 &linkhdrsize, &lladdr_off) != 0)
972 goto reply;
973
974 /* Allocate new entry */
975 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
976 if (la == NULL) {
977 /*
978 * lle creation may fail if source address belongs
979 * to non-directly connected subnet. However, we
980 * will try to answer the request instead of dropping
981 * frame.
982 */
983 goto reply;
984 }
985 lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
986 lladdr_off);
987
988 IF_AFDATA_WLOCK(ifp);
989 LLE_WLOCK(la);
990 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
991
992 /*
993 * Check if lle still does not exists.
994 * If it does, that means that we either
995 * 1) have configured it explicitly, via
996 * 1a) 'arp -s' static entry or
997 * 1b) interface address static record
998 * or
999 * 2) it was the result of sending first packet to-host
1000 * or
1001 * 3) it was another arp reply packet we handled in
1002 * different thread.
1003 *
1004 * In all cases except 3) we definitely need to prefer
1005 * existing lle. For the sake of simplicity, prefer any
1006 * existing lle over newly-create one.
1007 */
1008 if (la_tmp == NULL)
1009 lltable_link_entry(LLTABLE(ifp), la);
1010 IF_AFDATA_WUNLOCK(ifp);
1011
1012 if (la_tmp == NULL) {
1013 arp_mark_lle_reachable(la);
1014 LLE_WUNLOCK(la);
1015 } else {
1016 /* Free newly-create entry and handle packet */
1017 lltable_free_entry(LLTABLE(ifp), la);
1018 la = la_tmp;
1019 la_tmp = NULL;
1020 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1021 /* arp_check_update_lle() returns @la unlocked */
1022 }
1023 la = NULL;
1024 }
1025 reply:
1026 if (op != ARPOP_REQUEST)
1027 goto drop;
1028 ARPSTAT_INC(rxrequests);
1029
1030 if (itaddr.s_addr == myaddr.s_addr) {
1031 /* Shortcut.. the receiving interface is the target. */
1032 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1033 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1034 } else {
1035 /*
1036 * Destination address is not ours. Check if
1037 * proxyarp entry exists or proxyarp is turned on globally.
1038 */
1039 struct llentry *lle;
1040
1041 sin.sin_addr = itaddr;
1042 lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1043
1044 if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1045 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1046 (void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1047 LLE_RUNLOCK(lle);
1048 } else {
1049 if (lle != NULL)
1050 LLE_RUNLOCK(lle);
1051
1052 if (!V_arp_proxyall)
1053 goto drop;
1054
1055 NET_EPOCH_ASSERT();
1056 nh = fib4_lookup(ifp->if_fib, itaddr, 0, 0, 0);
1057 if (nh == NULL)
1058 goto drop;
1059
1060 /*
1061 * Don't send proxies for nodes on the same interface
1062 * as this one came out of, or we'll get into a fight
1063 * over who claims what Ether address.
1064 */
1065 if (nh->nh_ifp == ifp)
1066 goto drop;
1067
1068 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1069 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1070
1071 /*
1072 * Also check that the node which sent the ARP packet
1073 * is on the interface we expect it to be on. This
1074 * avoids ARP chaos if an interface is connected to the
1075 * wrong network.
1076 */
1077
1078 nh = fib4_lookup(ifp->if_fib, isaddr, 0, 0, 0);
1079 if (nh == NULL)
1080 goto drop;
1081 if (nh->nh_ifp != ifp) {
1082 ARP_LOG(LOG_INFO, "proxy: ignoring request"
1083 " from %s via %s\n",
1084 inet_ntoa_r(isaddr, addrbuf),
1085 ifp->if_xname);
1086 goto drop;
1087 }
1088
1089 #ifdef DEBUG_PROXY
1090 printf("arp: proxying for %s\n",
1091 inet_ntoa_r(itaddr, addrbuf));
1092 #endif
1093 }
1094 }
1095
1096 if (itaddr.s_addr == myaddr.s_addr &&
1097 IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1098 /* RFC 3927 link-local IPv4; always reply by broadcast. */
1099 #ifdef DEBUG_LINKLOCAL
1100 printf("arp: sending reply for link-local addr %s\n",
1101 inet_ntoa_r(itaddr, addrbuf));
1102 #endif
1103 m->m_flags |= M_BCAST;
1104 m->m_flags &= ~M_MCAST;
1105 } else {
1106 /* default behaviour; never reply by broadcast. */
1107 m->m_flags &= ~(M_BCAST|M_MCAST);
1108 }
1109 (void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1110 (void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1111 ah->ar_op = htons(ARPOP_REPLY);
1112 ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1113 m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1114 m->m_pkthdr.len = m->m_len;
1115 m->m_pkthdr.rcvif = NULL;
1116 sa.sa_family = AF_ARP;
1117 sa.sa_len = 2;
1118
1119 /* Calculate link header for sending frame */
1120 bzero(&ro, sizeof(ro));
1121 linkhdrsize = sizeof(linkhdr);
1122 error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1123
1124 /*
1125 * arp_fillheader() may fail due to lack of support inside encap request
1126 * routing. This is not necessary an error, AF_ARP can/should be handled
1127 * by if_output().
1128 */
1129 if (error != 0 && error != EAFNOSUPPORT) {
1130 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1131 if_name(ifp), error);
1132 goto drop;
1133 }
1134
1135 ro.ro_prepend = linkhdr;
1136 ro.ro_plen = linkhdrsize;
1137 ro.ro_flags = 0;
1138
1139 m_clrprotoflags(m); /* Avoid confusing lower layers. */
1140 (*ifp->if_output)(ifp, m, &sa, &ro);
1141 ARPSTAT_INC(txreplies);
1142 return;
1143
1144 drop:
1145 m_freem(m);
1146 }
1147 #endif
1148
1149 static struct mbuf *
arp_grab_holdchain(struct llentry * la)1150 arp_grab_holdchain(struct llentry *la)
1151 {
1152 struct mbuf *chain;
1153
1154 LLE_WLOCK_ASSERT(la);
1155
1156 chain = la->la_hold;
1157 la->la_hold = NULL;
1158 la->la_numheld = 0;
1159
1160 return (chain);
1161 }
1162
1163 static void
arp_flush_holdchain(struct ifnet * ifp,struct llentry * la,struct mbuf * chain)1164 arp_flush_holdchain(struct ifnet *ifp, struct llentry *la, struct mbuf *chain)
1165 {
1166 struct mbuf *m_hold, *m_hold_next;
1167 struct sockaddr_in sin;
1168
1169 NET_EPOCH_ASSERT();
1170
1171 struct route ro = {
1172 .ro_prepend = la->r_linkdata,
1173 .ro_plen = la->r_hdrlen,
1174 };
1175
1176 lltable_fill_sa_entry(la, (struct sockaddr *)&sin);
1177
1178 for (m_hold = chain; m_hold != NULL; m_hold = m_hold_next) {
1179 m_hold_next = m_hold->m_nextpkt;
1180 m_hold->m_nextpkt = NULL;
1181 /* Avoid confusing lower layers. */
1182 m_clrprotoflags(m_hold);
1183 (*ifp->if_output)(ifp, m_hold, (struct sockaddr *)&sin, &ro);
1184 }
1185 }
1186
1187 /*
1188 * Checks received arp data against existing @la.
1189 * Updates lle state/performs notification if necessary.
1190 */
1191 static void
arp_check_update_lle(struct arphdr * ah,struct in_addr isaddr,struct ifnet * ifp,int bridged,struct llentry * la)1192 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1193 int bridged, struct llentry *la)
1194 {
1195 uint8_t linkhdr[LLE_MAX_LINKHDR];
1196 size_t linkhdrsize;
1197 int lladdr_off;
1198 char addrbuf[INET_ADDRSTRLEN];
1199
1200 LLE_WLOCK_ASSERT(la);
1201
1202 /* the following is not an error when doing bridging */
1203 if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1204 if (log_arp_wrong_iface)
1205 ARP_LOG(LOG_WARNING, "%s is on %s "
1206 "but got reply from %*D on %s\n",
1207 inet_ntoa_r(isaddr, addrbuf),
1208 la->lle_tbl->llt_ifp->if_xname,
1209 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1210 ifp->if_xname);
1211 LLE_WUNLOCK(la);
1212 return;
1213 }
1214 if ((la->la_flags & LLE_VALID) &&
1215 bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1216 if (la->la_flags & LLE_STATIC) {
1217 LLE_WUNLOCK(la);
1218 if (log_arp_permanent_modify)
1219 ARP_LOG(LOG_ERR,
1220 "%*D attempts to modify "
1221 "permanent entry for %s on %s\n",
1222 ifp->if_addrlen,
1223 (u_char *)ar_sha(ah), ":",
1224 inet_ntoa_r(isaddr, addrbuf),
1225 ifp->if_xname);
1226 return;
1227 }
1228 if (log_arp_movements) {
1229 ARP_LOG(LOG_INFO, "%s moved from %*D "
1230 "to %*D on %s\n",
1231 inet_ntoa_r(isaddr, addrbuf),
1232 ifp->if_addrlen,
1233 (u_char *)la->ll_addr, ":",
1234 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1235 ifp->if_xname);
1236 }
1237 }
1238
1239 /* Calculate full link prepend to use in lle */
1240 linkhdrsize = sizeof(linkhdr);
1241 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1242 &linkhdrsize, &lladdr_off) != 0) {
1243 LLE_WUNLOCK(la);
1244 return;
1245 }
1246
1247 /* Check if something has changed */
1248 if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1249 (la->la_flags & LLE_VALID) == 0) {
1250 /* Try to perform LLE update */
1251 if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1252 lladdr_off) == 0) {
1253 LLE_WUNLOCK(la);
1254 return;
1255 }
1256
1257 /* Clear fast path feedback request if set */
1258 llentry_mark_used(la);
1259 }
1260
1261 arp_mark_lle_reachable(la);
1262
1263 /*
1264 * The packets are all freed within the call to the output
1265 * routine.
1266 *
1267 * NB: The lock MUST be released before the call to the
1268 * output routine.
1269 */
1270 if (la->la_hold != NULL) {
1271 struct mbuf *chain;
1272
1273 chain = arp_grab_holdchain(la);
1274 LLE_WUNLOCK(la);
1275 arp_flush_holdchain(ifp, la, chain);
1276 } else
1277 LLE_WUNLOCK(la);
1278 }
1279
1280 static void
arp_mark_lle_reachable(struct llentry * la)1281 arp_mark_lle_reachable(struct llentry *la)
1282 {
1283 int canceled, wtime;
1284
1285 LLE_WLOCK_ASSERT(la);
1286
1287 la->ln_state = ARP_LLINFO_REACHABLE;
1288 EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1289
1290 if (!(la->la_flags & LLE_STATIC)) {
1291 LLE_ADDREF(la);
1292 la->la_expire = time_uptime + V_arpt_keep;
1293 wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1294 if (wtime < 0)
1295 wtime = V_arpt_keep;
1296 canceled = callout_reset(&la->lle_timer,
1297 hz * wtime, arptimer, la);
1298 if (canceled)
1299 LLE_REMREF(la);
1300 }
1301 la->la_asked = 0;
1302 la->la_preempt = V_arp_maxtries;
1303 }
1304
1305 /*
1306 * Add permanent link-layer record for given interface address.
1307 */
1308 static __noinline void
arp_add_ifa_lle(struct ifnet * ifp,const struct sockaddr * dst)1309 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1310 {
1311 struct llentry *lle, *lle_tmp;
1312
1313 /*
1314 * Interface address LLE record is considered static
1315 * because kernel code relies on LLE_STATIC flag to check
1316 * if these entries can be rewriten by arp updates.
1317 */
1318 lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1319 if (lle == NULL) {
1320 log(LOG_INFO, "arp_ifinit: cannot create arp "
1321 "entry for interface address\n");
1322 return;
1323 }
1324
1325 IF_AFDATA_WLOCK(ifp);
1326 LLE_WLOCK(lle);
1327 /* Unlink any entry if exists */
1328 lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1329 if (lle_tmp != NULL)
1330 lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1331
1332 lltable_link_entry(LLTABLE(ifp), lle);
1333 IF_AFDATA_WUNLOCK(ifp);
1334
1335 if (lle_tmp != NULL)
1336 EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1337
1338 EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1339 LLE_WUNLOCK(lle);
1340 if (lle_tmp != NULL)
1341 lltable_free_entry(LLTABLE(ifp), lle_tmp);
1342 }
1343
1344 /*
1345 * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1346 * of valid values.
1347 */
1348 static int
sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)1349 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1350 {
1351 int error;
1352 int rexmit_count = *(int *)arg1;
1353
1354 error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1355
1356 /* Enforce limits on any new value that may have been set. */
1357 if (!error && req->newptr) {
1358 /* A new value was set. */
1359 if (rexmit_count < 0) {
1360 rexmit_count = 0;
1361 } else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1362 rexmit_count = MAX_GARP_RETRANSMITS;
1363 }
1364 *(int *)arg1 = rexmit_count;
1365 }
1366
1367 return (error);
1368 }
1369
1370 /*
1371 * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1372 * retransmit it again. A pending callout owns a reference to the ifa.
1373 */
1374 static void
garp_rexmit(void * arg)1375 garp_rexmit(void *arg)
1376 {
1377 struct in_ifaddr *ia = arg;
1378
1379 if (callout_pending(&ia->ia_garp_timer) ||
1380 !callout_active(&ia->ia_garp_timer)) {
1381 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1382 ifa_free(&ia->ia_ifa);
1383 return;
1384 }
1385
1386 CURVNET_SET(ia->ia_ifa.ifa_ifp->if_vnet);
1387
1388 /*
1389 * Drop lock while the ARP request is generated.
1390 */
1391 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1392
1393 arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1394 &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1395
1396 /*
1397 * Increment the count of retransmissions. If the count has reached the
1398 * maximum value, stop sending the GARP packets. Otherwise, schedule
1399 * the callout to retransmit another GARP packet.
1400 */
1401 ++ia->ia_garp_count;
1402 if (ia->ia_garp_count >= garp_rexmit_count) {
1403 ifa_free(&ia->ia_ifa);
1404 } else {
1405 int rescheduled;
1406 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1407 rescheduled = callout_reset(&ia->ia_garp_timer,
1408 (1 << ia->ia_garp_count) * hz,
1409 garp_rexmit, ia);
1410 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1411 if (rescheduled) {
1412 ifa_free(&ia->ia_ifa);
1413 }
1414 }
1415
1416 CURVNET_RESTORE();
1417 }
1418
1419 /*
1420 * Start the GARP retransmit timer.
1421 *
1422 * A single GARP is always transmitted when an IPv4 address is added
1423 * to an interface and that is usually sufficient. However, in some
1424 * circumstances, such as when a shared address is passed between
1425 * cluster nodes, this single GARP may occasionally be dropped or
1426 * lost. This can lead to neighbors on the network link working with a
1427 * stale ARP cache and sending packets destined for that address to
1428 * the node that previously owned the address, which may not respond.
1429 *
1430 * To avoid this situation, GARP retransmits can be enabled by setting
1431 * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1432 * than zero. The setting represents the maximum number of
1433 * retransmissions. The interval between retransmissions is calculated
1434 * using an exponential backoff algorithm, doubling each time, so the
1435 * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1436 */
1437 static void
garp_timer_start(struct ifaddr * ifa)1438 garp_timer_start(struct ifaddr *ifa)
1439 {
1440 struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1441
1442 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1443 ia->ia_garp_count = 0;
1444 if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1445 garp_rexmit, ia) == 0) {
1446 ifa_ref(ifa);
1447 }
1448 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1449 }
1450
1451 void
arp_ifinit(struct ifnet * ifp,struct ifaddr * ifa)1452 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1453 {
1454 struct epoch_tracker et;
1455 const struct sockaddr_in *dst_in;
1456 const struct sockaddr *dst;
1457
1458 if (ifa->ifa_carp != NULL)
1459 return;
1460
1461 dst = ifa->ifa_addr;
1462 dst_in = (const struct sockaddr_in *)dst;
1463
1464 if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1465 return;
1466 NET_EPOCH_ENTER(et);
1467 arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1468 NET_EPOCH_EXIT(et);
1469 if (garp_rexmit_count > 0) {
1470 garp_timer_start(ifa);
1471 }
1472
1473 arp_add_ifa_lle(ifp, dst);
1474 }
1475
1476 void
arp_announce_ifaddr(struct ifnet * ifp,struct in_addr addr,u_char * enaddr)1477 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1478 {
1479
1480 if (ntohl(addr.s_addr) != INADDR_ANY)
1481 arprequest(ifp, &addr, &addr, enaddr);
1482 }
1483
1484 /*
1485 * Sends gratuitous ARPs for each ifaddr to notify other
1486 * nodes about the address change.
1487 */
1488 static __noinline void
arp_handle_ifllchange(struct ifnet * ifp)1489 arp_handle_ifllchange(struct ifnet *ifp)
1490 {
1491 struct ifaddr *ifa;
1492
1493 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1494 if (ifa->ifa_addr->sa_family == AF_INET)
1495 arp_ifinit(ifp, ifa);
1496 }
1497 }
1498
1499 /*
1500 * A handler for interface link layer address change event.
1501 */
1502 static void
arp_iflladdr(void * arg __unused,struct ifnet * ifp)1503 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1504 {
1505 /* if_bridge can update its lladdr during if_vmove(), after we've done
1506 * if_detach_internal()/dom_ifdetach(). */
1507 if (ifp->if_afdata[AF_INET] == NULL)
1508 return;
1509
1510 lltable_update_ifaddr(LLTABLE(ifp));
1511
1512 if ((ifp->if_flags & IFF_UP) != 0)
1513 arp_handle_ifllchange(ifp);
1514 }
1515
1516 static void
vnet_arp_init(void)1517 vnet_arp_init(void)
1518 {
1519
1520 if (IS_DEFAULT_VNET(curvnet)) {
1521 netisr_register(&arp_nh);
1522 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1523 arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1524 }
1525 #ifdef VIMAGE
1526 else
1527 netisr_register_vnet(&arp_nh);
1528 #endif
1529 }
1530 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1531 vnet_arp_init, 0);
1532
1533 #ifdef VIMAGE
1534 /*
1535 * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1536 * lookups after destroying the hash. Ideally this would go on SI_ORDER_3.5.
1537 */
1538 static void
vnet_arp_destroy(__unused void * arg)1539 vnet_arp_destroy(__unused void *arg)
1540 {
1541
1542 netisr_unregister_vnet(&arp_nh);
1543 }
1544 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1545 vnet_arp_destroy, NULL);
1546 #endif
1547