1 /*-
2 * Copyright (c) 1982, 1986, 1988, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)if_ether.c 8.1 (Berkeley) 6/10/93
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 __FBSDID("$FreeBSD$");
40
41 #include "opt_inet.h"
42
43 #include <sys/param.h>
44 #include <sys/kernel.h>
45 #include <sys/lock.h>
46 #include <sys/queue.h>
47 #include <sys/sysctl.h>
48 #include <sys/systm.h>
49 #include <sys/mbuf.h>
50 #include <sys/malloc.h>
51 #include <sys/proc.h>
52 #include <sys/rmlock.h>
53 #include <sys/socket.h>
54 #include <sys/syslog.h>
55
56 #include <net/if.h>
57 #include <net/if_var.h>
58 #include <net/if_dl.h>
59 #include <net/if_types.h>
60 #include <net/netisr.h>
61 #include <net/ethernet.h>
62 #include <net/route.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, CTLFLAG_RW, 0, "");
92 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp, CTLFLAG_RW, 0, "");
93
94 /* timer values */
95 static VNET_DEFINE(int, arpt_keep) = (20*60); /* once resolved, good for 20
96 * minutes */
97 static VNET_DEFINE(int, arp_maxtries) = 5;
98 static VNET_DEFINE(int, arp_proxyall) = 0;
99 static VNET_DEFINE(int, arpt_down) = 20; /* keep incomplete entries for
100 * 20 seconds */
101 static VNET_DEFINE(int, arpt_rexmit) = 1; /* retransmit arp entries, sec*/
102 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat); /* ARP statistics, see if_arp.h */
103 VNET_PCPUSTAT_SYSINIT(arpstat);
104
105 #ifdef VIMAGE
106 VNET_PCPUSTAT_SYSUNINIT(arpstat);
107 #endif /* VIMAGE */
108
109 static VNET_DEFINE(int, arp_maxhold) = 1;
110
111 #define V_arpt_keep VNET(arpt_keep)
112 #define V_arpt_down VNET(arpt_down)
113 #define V_arpt_rexmit VNET(arpt_rexmit)
114 #define V_arp_maxtries VNET(arp_maxtries)
115 #define V_arp_proxyall VNET(arp_proxyall)
116 #define V_arp_maxhold VNET(arp_maxhold)
117
118 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
119 &VNET_NAME(arpt_keep), 0,
120 "ARP entry lifetime in seconds");
121 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
122 &VNET_NAME(arp_maxtries), 0,
123 "ARP resolution attempts before returning error");
124 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
125 &VNET_NAME(arp_proxyall), 0,
126 "Enable proxy ARP for all suitable requests");
127 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
128 &VNET_NAME(arpt_down), 0,
129 "Incomplete ARP entry lifetime in seconds");
130 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
131 arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
132 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
133 &VNET_NAME(arp_maxhold), 0,
134 "Number of packets to hold per ARP entry");
135 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
136 CTLFLAG_RW, &arp_maxpps, 0,
137 "Maximum number of remotely triggered ARP messages that can be "
138 "logged per second");
139
140 /*
141 * Due to the exponential backoff algorithm used for the interval between GARP
142 * retransmissions, the maximum number of retransmissions is limited for
143 * sanity. This limit corresponds to a maximum interval between retransmissions
144 * of 2^16 seconds ~= 18 hours.
145 *
146 * Making this limit more dynamic is more complicated than worthwhile,
147 * especially since sending out GARPs spaced days apart would be of little
148 * use. A maximum dynamic limit would look something like:
149 *
150 * const int max = fls(INT_MAX / hz) - 1;
151 */
152 #define MAX_GARP_RETRANSMITS 16
153 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS);
154 static int garp_rexmit_count = 0; /* GARP retransmission setting. */
155
156 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count,
157 CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE,
158 &garp_rexmit_count, 0, sysctl_garp_rexmit, "I",
159 "Number of times to retransmit GARP packets;"
160 " 0 to disable, maximum of 16");
161
162 #define ARP_LOG(pri, ...) do { \
163 if (ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps)) \
164 log((pri), "arp: " __VA_ARGS__); \
165 } while (0)
166
167
168 static void arpintr(struct mbuf *);
169 static void arptimer(void *);
170 #ifdef INET
171 static void in_arpinput(struct mbuf *);
172 #endif
173
174 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
175 struct ifnet *ifp, int bridged, struct llentry *la);
176 static void arp_mark_lle_reachable(struct llentry *la);
177 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
178
179 static eventhandler_tag iflladdr_tag;
180
181 static const struct netisr_handler arp_nh = {
182 .nh_name = "arp",
183 .nh_handler = arpintr,
184 .nh_proto = NETISR_ARP,
185 .nh_policy = NETISR_POLICY_SOURCE,
186 };
187
188 /*
189 * Timeout routine. Age arp_tab entries periodically.
190 */
191 static void
192 arptimer(void *arg)
193 {
194 struct llentry *lle = (struct llentry *)arg;
195 struct ifnet *ifp;
196 int r_skip_req;
197
198 if (lle->la_flags & LLE_STATIC) {
199 return;
200 }
201 LLE_WLOCK(lle);
202 if (callout_pending(&lle->lle_timer)) {
203 /*
204 * Here we are a bit odd here in the treatment of
205 * active/pending. If the pending bit is set, it got
206 * rescheduled before I ran. The active
207 * bit we ignore, since if it was stopped
208 * in ll_tablefree() and was currently running
209 * it would have return 0 so the code would
210 * not have deleted it since the callout could
211 * not be stopped so we want to go through
212 * with the delete here now. If the callout
213 * was restarted, the pending bit will be back on and
214 * we just want to bail since the callout_reset would
215 * return 1 and our reference would have been removed
216 * by arpresolve() below.
217 */
218 LLE_WUNLOCK(lle);
219 return;
220 }
221 ifp = lle->lle_tbl->llt_ifp;
222 CURVNET_SET(ifp->if_vnet);
223
224 switch (lle->ln_state) {
225 case ARP_LLINFO_REACHABLE:
226
227 /*
228 * Expiration time is approaching.
229 * Let's try to refresh entry if it is still
230 * in use.
231 *
232 * Set r_skip_req to get feedback from
233 * fast path. Change state and re-schedule
234 * ourselves.
235 */
236 LLE_REQ_LOCK(lle);
237 lle->r_skip_req = 1;
238 LLE_REQ_UNLOCK(lle);
239 lle->ln_state = ARP_LLINFO_VERIFY;
240 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
241 LLE_WUNLOCK(lle);
242 CURVNET_RESTORE();
243 return;
244 case ARP_LLINFO_VERIFY:
245 LLE_REQ_LOCK(lle);
246 r_skip_req = lle->r_skip_req;
247 LLE_REQ_UNLOCK(lle);
248
249 if (r_skip_req == 0 && lle->la_preempt > 0) {
250 /* Entry was used, issue refresh request */
251 struct in_addr dst;
252 dst = lle->r_l3addr.addr4;
253 lle->la_preempt--;
254 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
255 LLE_WUNLOCK(lle);
256 arprequest(ifp, NULL, &dst, NULL);
257 CURVNET_RESTORE();
258 return;
259 }
260 /* Nothing happened. Reschedule if not too late */
261 if (lle->la_expire > time_uptime) {
262 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
263 LLE_WUNLOCK(lle);
264 CURVNET_RESTORE();
265 return;
266 }
267 break;
268 case ARP_LLINFO_INCOMPLETE:
269 case ARP_LLINFO_DELETED:
270 break;
271 }
272
273 if ((lle->la_flags & LLE_DELETED) == 0) {
274 int evt;
275
276 if (lle->la_flags & LLE_VALID)
277 evt = LLENTRY_EXPIRED;
278 else
279 evt = LLENTRY_TIMEDOUT;
280 EVENTHANDLER_INVOKE(lle_event, lle, evt);
281 }
282
283 callout_stop(&lle->lle_timer);
284
285 /* XXX: LOR avoidance. We still have ref on lle. */
286 LLE_WUNLOCK(lle);
287 IF_AFDATA_LOCK(ifp);
288 LLE_WLOCK(lle);
289
290 /* Guard against race with other llentry_free(). */
291 if (lle->la_flags & LLE_LINKED) {
292 LLE_REMREF(lle);
293 lltable_unlink_entry(lle->lle_tbl, lle);
294 }
295 IF_AFDATA_UNLOCK(ifp);
296
297 size_t pkts_dropped = llentry_free(lle);
298
299 ARPSTAT_ADD(dropped, pkts_dropped);
300 ARPSTAT_INC(timeouts);
301
302 CURVNET_RESTORE();
303 }
304
305 /*
306 * Stores link-layer header for @ifp in format suitable for if_output()
307 * into buffer @buf. Resulting header length is stored in @bufsize.
308 *
309 * Returns 0 on success.
310 */
311 static int
312 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
313 size_t *bufsize)
314 {
315 struct if_encap_req ereq;
316 int error;
317
318 bzero(buf, *bufsize);
319 bzero(&ereq, sizeof(ereq));
320 ereq.buf = buf;
321 ereq.bufsize = *bufsize;
322 ereq.rtype = IFENCAP_LL;
323 ereq.family = AF_ARP;
324 ereq.lladdr = ar_tha(ah);
325 ereq.hdata = (u_char *)ah;
326 if (bcast)
327 ereq.flags = IFENCAP_FLAG_BROADCAST;
328 error = ifp->if_requestencap(ifp, &ereq);
329 if (error == 0)
330 *bufsize = ereq.bufsize;
331
332 return (error);
333 }
334
335
336 /*
337 * Broadcast an ARP request. Caller specifies:
338 * - arp header source ip address
339 * - arp header target ip address
340 * - arp header source ethernet address
341 */
342 void
343 arprequest(struct ifnet *ifp, const struct in_addr *sip,
344 const struct in_addr *tip, u_char *enaddr)
345 {
346 struct mbuf *m;
347 struct arphdr *ah;
348 struct sockaddr sa;
349 u_char *carpaddr = NULL;
350 uint8_t linkhdr[LLE_MAX_LINKHDR];
351 size_t linkhdrsize;
352 struct route ro;
353 int error;
354
355 if (sip == NULL) {
356 /*
357 * The caller did not supply a source address, try to find
358 * a compatible one among those assigned to this interface.
359 */
360 struct ifaddr *ifa;
361
362 IF_ADDR_RLOCK(ifp);
363 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
364 if (ifa->ifa_addr->sa_family != AF_INET)
365 continue;
366
367 if (ifa->ifa_carp) {
368 if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
369 continue;
370 sip = &IA_SIN(ifa)->sin_addr;
371 } else {
372 carpaddr = NULL;
373 sip = &IA_SIN(ifa)->sin_addr;
374 }
375
376 if (0 == ((sip->s_addr ^ tip->s_addr) &
377 IA_MASKSIN(ifa)->sin_addr.s_addr))
378 break; /* found it. */
379 }
380 IF_ADDR_RUNLOCK(ifp);
381 if (sip == NULL) {
382 printf("%s: cannot find matching address\n", __func__);
383 return;
384 }
385 }
386 if (enaddr == NULL)
387 enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
388
389 if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
390 return;
391 m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
392 2 * ifp->if_addrlen;
393 m->m_pkthdr.len = m->m_len;
394 M_ALIGN(m, m->m_len);
395 ah = mtod(m, struct arphdr *);
396 bzero((caddr_t)ah, m->m_len);
397 #ifdef MAC
398 mac_netinet_arp_send(ifp, m);
399 #endif
400 ah->ar_pro = htons(ETHERTYPE_IP);
401 ah->ar_hln = ifp->if_addrlen; /* hardware address length */
402 ah->ar_pln = sizeof(struct in_addr); /* protocol address length */
403 ah->ar_op = htons(ARPOP_REQUEST);
404 bcopy(enaddr, ar_sha(ah), ah->ar_hln);
405 bcopy(sip, ar_spa(ah), ah->ar_pln);
406 bcopy(tip, ar_tpa(ah), ah->ar_pln);
407 sa.sa_family = AF_ARP;
408 sa.sa_len = 2;
409
410 /* Calculate link header for sending frame */
411 bzero(&ro, sizeof(ro));
412 linkhdrsize = sizeof(linkhdr);
413 error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
414 if (error != 0 && error != EAFNOSUPPORT) {
415 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
416 if_name(ifp), error);
417 return;
418 }
419
420 ro.ro_prepend = linkhdr;
421 ro.ro_plen = linkhdrsize;
422 ro.ro_flags = 0;
423
424 m->m_flags |= M_BCAST;
425 m_clrprotoflags(m); /* Avoid confusing lower layers. */
426 (*ifp->if_output)(ifp, m, &sa, &ro);
427 ARPSTAT_INC(txrequests);
428 }
429
430
431 /*
432 * Resolve an IP address into an ethernet address - heavy version.
433 * Used internally by arpresolve().
434 * We have already checked that we can't use an existing lle without
435 * modification so we have to acquire an LLE_EXCLUSIVE lle lock.
436 *
437 * On success, desten and pflags are filled in and the function returns 0;
438 * If the packet must be held pending resolution, we return EWOULDBLOCK
439 * On other errors, we return the corresponding error code.
440 * Note that m_freem() handles NULL.
441 */
442 static int
443 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
444 const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
445 struct llentry **plle)
446 {
447 struct llentry *la = NULL, *la_tmp;
448 struct mbuf *curr = NULL;
449 struct mbuf *next = NULL;
450 int error, renew;
451 char *lladdr;
452 int ll_len;
453
454 if (pflags != NULL)
455 *pflags = 0;
456 if (plle != NULL)
457 *plle = NULL;
458
459 if ((flags & LLE_CREATE) == 0) {
460 IF_AFDATA_RLOCK(ifp);
461 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
462 IF_AFDATA_RUNLOCK(ifp);
463 }
464 if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
465 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
466 if (la == NULL) {
467 char addrbuf[INET_ADDRSTRLEN];
468
469 log(LOG_DEBUG,
470 "arpresolve: can't allocate llinfo for %s on %s\n",
471 inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
472 if_name(ifp));
473 m_freem(m);
474 return (EINVAL);
475 }
476
477 IF_AFDATA_WLOCK(ifp);
478 LLE_WLOCK(la);
479 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
480 /* Prefer ANY existing lle over newly-created one */
481 if (la_tmp == NULL)
482 lltable_link_entry(LLTABLE(ifp), la);
483 IF_AFDATA_WUNLOCK(ifp);
484 if (la_tmp != NULL) {
485 lltable_free_entry(LLTABLE(ifp), la);
486 la = la_tmp;
487 }
488 }
489 if (la == NULL) {
490 m_freem(m);
491 return (EINVAL);
492 }
493
494 if ((la->la_flags & LLE_VALID) &&
495 ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
496 if (flags & LLE_ADDRONLY) {
497 lladdr = la->ll_addr;
498 ll_len = ifp->if_addrlen;
499 } else {
500 lladdr = la->r_linkdata;
501 ll_len = la->r_hdrlen;
502 }
503 bcopy(lladdr, desten, ll_len);
504
505 /* Notify LLE code that the entry was used by datapath */
506 llentry_mark_used(la);
507 if (pflags != NULL)
508 *pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
509 if (plle) {
510 LLE_ADDREF(la);
511 *plle = la;
512 }
513 LLE_WUNLOCK(la);
514 return (0);
515 }
516
517 renew = (la->la_asked == 0 || la->la_expire != time_uptime);
518 /*
519 * There is an arptab entry, but no ethernet address
520 * response yet. Add the mbuf to the list, dropping
521 * the oldest packet if we have exceeded the system
522 * setting.
523 */
524 if (m != NULL) {
525 if (la->la_numheld >= V_arp_maxhold) {
526 if (la->la_hold != NULL) {
527 next = la->la_hold->m_nextpkt;
528 m_freem(la->la_hold);
529 la->la_hold = next;
530 la->la_numheld--;
531 ARPSTAT_INC(dropped);
532 }
533 }
534 if (la->la_hold != NULL) {
535 curr = la->la_hold;
536 while (curr->m_nextpkt != NULL)
537 curr = curr->m_nextpkt;
538 curr->m_nextpkt = m;
539 } else
540 la->la_hold = m;
541 la->la_numheld++;
542 }
543 /*
544 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
545 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
546 * if we have already sent arp_maxtries ARP requests. Retransmit the
547 * ARP request, but not faster than one request per second.
548 */
549 if (la->la_asked < V_arp_maxtries)
550 error = EWOULDBLOCK; /* First request. */
551 else
552 error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
553
554 if (renew) {
555 int canceled;
556
557 LLE_ADDREF(la);
558 la->la_expire = time_uptime;
559 canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
560 arptimer, la);
561 if (canceled)
562 LLE_REMREF(la);
563 la->la_asked++;
564 LLE_WUNLOCK(la);
565 arprequest(ifp, NULL, &SIN(dst)->sin_addr, NULL);
566 return (error);
567 }
568
569 LLE_WUNLOCK(la);
570 return (error);
571 }
572
573 /*
574 * Resolve an IP address into an ethernet address.
575 */
576 int
577 arpresolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
578 char *desten, uint32_t *pflags, struct llentry **plle)
579 {
580 int error;
581
582 flags |= LLE_ADDRONLY;
583 error = arpresolve_full(ifp, 0, flags, NULL, dst, desten, pflags, plle);
584 return (error);
585 }
586
587
588 /*
589 * Lookups link header based on an IP address.
590 * On input:
591 * ifp is the interface we use
592 * is_gw != 0 if @dst represents gateway to some destination
593 * m is the mbuf. May be NULL if we don't have a packet.
594 * dst is the next hop,
595 * desten is the storage to put LL header.
596 * flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
597 *
598 * On success, full/partial link header and flags are filled in and
599 * the function returns 0.
600 * If the packet must be held pending resolution, we return EWOULDBLOCK
601 * On other errors, we return the corresponding error code.
602 * Note that m_freem() handles NULL.
603 */
604 int
605 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
606 const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
607 struct llentry **plle)
608 {
609 struct llentry *la = NULL;
610
611 if (pflags != NULL)
612 *pflags = 0;
613 if (plle != NULL)
614 *plle = NULL;
615
616 if (m != NULL) {
617 if (m->m_flags & M_BCAST) {
618 /* broadcast */
619 (void)memcpy(desten,
620 ifp->if_broadcastaddr, ifp->if_addrlen);
621 return (0);
622 }
623 if (m->m_flags & M_MCAST) {
624 /* multicast */
625 ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
626 return (0);
627 }
628 }
629
630 IF_AFDATA_RLOCK(ifp);
631 la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
632 if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
633 /* Entry found, let's copy lle info */
634 bcopy(la->r_linkdata, desten, la->r_hdrlen);
635 if (pflags != NULL)
636 *pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
637 /* Notify the LLE handling code that the entry was used. */
638 llentry_mark_used(la);
639 if (plle) {
640 LLE_ADDREF(la);
641 *plle = la;
642 LLE_WUNLOCK(la);
643 }
644 IF_AFDATA_RUNLOCK(ifp);
645 return (0);
646 }
647 if (plle && la)
648 LLE_WUNLOCK(la);
649 IF_AFDATA_RUNLOCK(ifp);
650
651 return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
652 desten, pflags, plle));
653 }
654
655 /*
656 * Common length and type checks are done here,
657 * then the protocol-specific routine is called.
658 */
659 static void
660 arpintr(struct mbuf *m)
661 {
662 struct arphdr *ar;
663 struct ifnet *ifp;
664 char *layer;
665 int hlen;
666
667 ifp = m->m_pkthdr.rcvif;
668
669 if (m->m_len < sizeof(struct arphdr) &&
670 ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
671 ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
672 if_name(ifp));
673 return;
674 }
675 ar = mtod(m, struct arphdr *);
676
677 /* Check if length is sufficient */
678 if (m->m_len < arphdr_len(ar)) {
679 m = m_pullup(m, arphdr_len(ar));
680 if (m == NULL) {
681 ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
682 if_name(ifp));
683 return;
684 }
685 ar = mtod(m, struct arphdr *);
686 }
687
688 hlen = 0;
689 layer = "";
690 switch (ntohs(ar->ar_hrd)) {
691 case ARPHRD_ETHER:
692 hlen = ETHER_ADDR_LEN; /* RFC 826 */
693 layer = "ethernet";
694 break;
695 case ARPHRD_IEEE802:
696 hlen = 6; /* RFC 1390, FDDI_ADDR_LEN */
697 layer = "fddi";
698 break;
699 case ARPHRD_ARCNET:
700 hlen = 1; /* RFC 1201, ARC_ADDR_LEN */
701 layer = "arcnet";
702 break;
703 case ARPHRD_INFINIBAND:
704 hlen = 20; /* RFC 4391, INFINIBAND_ALEN */
705 layer = "infiniband";
706 break;
707 case ARPHRD_IEEE1394:
708 hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
709 layer = "firewire";
710
711 /*
712 * Restrict too long hardware addresses.
713 * Currently we are capable of handling 20-byte
714 * addresses ( sizeof(lle->ll_addr) )
715 */
716 if (ar->ar_hln >= 20)
717 hlen = 16;
718 break;
719 default:
720 ARP_LOG(LOG_NOTICE,
721 "packet with unknown hardware format 0x%02d received on "
722 "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
723 m_freem(m);
724 return;
725 }
726
727 if (hlen != 0 && hlen != ar->ar_hln) {
728 ARP_LOG(LOG_NOTICE,
729 "packet with invalid %s address length %d received on %s\n",
730 layer, ar->ar_hln, if_name(ifp));
731 m_freem(m);
732 return;
733 }
734
735 ARPSTAT_INC(received);
736 switch (ntohs(ar->ar_pro)) {
737 #ifdef INET
738 case ETHERTYPE_IP:
739 in_arpinput(m);
740 return;
741 #endif
742 }
743 m_freem(m);
744 }
745
746 #ifdef INET
747 /*
748 * ARP for Internet protocols on 10 Mb/s Ethernet.
749 * Algorithm is that given in RFC 826.
750 * In addition, a sanity check is performed on the sender
751 * protocol address, to catch impersonators.
752 * We no longer handle negotiations for use of trailer protocol:
753 * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
754 * along with IP replies if we wanted trailers sent to us,
755 * and also sent them in response to IP replies.
756 * This allowed either end to announce the desire to receive
757 * trailer packets.
758 * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
759 * but formerly didn't normally send requests.
760 */
761 static int log_arp_wrong_iface = 1;
762 static int log_arp_movements = 1;
763 static int log_arp_permanent_modify = 1;
764 static int allow_multicast = 0;
765
766 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
767 &log_arp_wrong_iface, 0,
768 "log arp packets arriving on the wrong interface");
769 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
770 &log_arp_movements, 0,
771 "log arp replies from MACs different than the one in the cache");
772 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
773 &log_arp_permanent_modify, 0,
774 "log arp replies from MACs different than the one in the permanent arp entry");
775 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
776 &allow_multicast, 0, "accept multicast addresses");
777
778 static void
779 in_arpinput(struct mbuf *m)
780 {
781 struct rm_priotracker in_ifa_tracker;
782 struct arphdr *ah;
783 struct ifnet *ifp = m->m_pkthdr.rcvif;
784 struct llentry *la = NULL, *la_tmp;
785 struct ifaddr *ifa;
786 struct in_ifaddr *ia;
787 struct sockaddr sa;
788 struct in_addr isaddr, itaddr, myaddr;
789 u_int8_t *enaddr = NULL;
790 int op;
791 int bridged = 0, is_bridge = 0;
792 int carped;
793 struct sockaddr_in sin;
794 struct sockaddr *dst;
795 struct nhop4_basic nh4;
796 uint8_t linkhdr[LLE_MAX_LINKHDR];
797 struct route ro;
798 size_t linkhdrsize;
799 int lladdr_off;
800 int error;
801 char addrbuf[INET_ADDRSTRLEN];
802
803 sin.sin_len = sizeof(struct sockaddr_in);
804 sin.sin_family = AF_INET;
805 sin.sin_addr.s_addr = 0;
806
807 if (ifp->if_bridge)
808 bridged = 1;
809 if (ifp->if_type == IFT_BRIDGE)
810 is_bridge = 1;
811
812 /*
813 * We already have checked that mbuf contains enough contiguous data
814 * to hold entire arp message according to the arp header.
815 */
816 ah = mtod(m, struct arphdr *);
817
818 /*
819 * ARP is only for IPv4 so we can reject packets with
820 * a protocol length not equal to an IPv4 address.
821 */
822 if (ah->ar_pln != sizeof(struct in_addr)) {
823 ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
824 sizeof(struct in_addr));
825 goto drop;
826 }
827
828 if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
829 ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
830 ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
831 goto drop;
832 }
833
834 op = ntohs(ah->ar_op);
835 (void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
836 (void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
837
838 if (op == ARPOP_REPLY)
839 ARPSTAT_INC(rxreplies);
840
841 /*
842 * For a bridge, we want to check the address irrespective
843 * of the receive interface. (This will change slightly
844 * when we have clusters of interfaces).
845 */
846 IN_IFADDR_RLOCK(&in_ifa_tracker);
847 LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
848 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
849 ia->ia_ifp == ifp) &&
850 itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
851 (ia->ia_ifa.ifa_carp == NULL ||
852 (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
853 ifa_ref(&ia->ia_ifa);
854 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
855 goto match;
856 }
857 }
858 LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
859 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
860 ia->ia_ifp == ifp) &&
861 isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
862 ifa_ref(&ia->ia_ifa);
863 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
864 goto match;
865 }
866
867 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia) \
868 (ia->ia_ifp->if_bridge == ifp->if_softc && \
869 !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) && \
870 addr == ia->ia_addr.sin_addr.s_addr)
871 /*
872 * Check the case when bridge shares its MAC address with
873 * some of its children, so packets are claimed by bridge
874 * itself (bridge_input() does it first), but they are really
875 * meant to be destined to the bridge member.
876 */
877 if (is_bridge) {
878 LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
879 if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
880 ifa_ref(&ia->ia_ifa);
881 ifp = ia->ia_ifp;
882 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
883 goto match;
884 }
885 }
886 }
887 #undef BDG_MEMBER_MATCHES_ARP
888 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
889
890 /*
891 * No match, use the first inet address on the receive interface
892 * as a dummy address for the rest of the function.
893 */
894 IF_ADDR_RLOCK(ifp);
895 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
896 if (ifa->ifa_addr->sa_family == AF_INET &&
897 (ifa->ifa_carp == NULL ||
898 (*carp_iamatch_p)(ifa, &enaddr))) {
899 ia = ifatoia(ifa);
900 ifa_ref(ifa);
901 IF_ADDR_RUNLOCK(ifp);
902 goto match;
903 }
904 IF_ADDR_RUNLOCK(ifp);
905
906 /*
907 * If bridging, fall back to using any inet address.
908 */
909 IN_IFADDR_RLOCK(&in_ifa_tracker);
910 if (!bridged || (ia = TAILQ_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 IF_AFDATA_RLOCK(ifp);
962 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
963 IF_AFDATA_RUNLOCK(ifp);
964 if (la != NULL)
965 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
966 else if (itaddr.s_addr == myaddr.s_addr) {
967 /*
968 * Request/reply to our address, but no lle exists yet.
969 * Calculate full link prepend to use in lle.
970 */
971 linkhdrsize = sizeof(linkhdr);
972 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
973 &linkhdrsize, &lladdr_off) != 0)
974 goto reply;
975
976 /* Allocate new entry */
977 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
978 if (la == NULL) {
979
980 /*
981 * lle creation may fail if source address belongs
982 * to non-directly connected subnet. However, we
983 * will try to answer the request instead of dropping
984 * frame.
985 */
986 goto reply;
987 }
988 lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
989 lladdr_off);
990
991 IF_AFDATA_WLOCK(ifp);
992 LLE_WLOCK(la);
993 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
994
995 /*
996 * Check if lle still does not exists.
997 * If it does, that means that we either
998 * 1) have configured it explicitly, via
999 * 1a) 'arp -s' static entry or
1000 * 1b) interface address static record
1001 * or
1002 * 2) it was the result of sending first packet to-host
1003 * or
1004 * 3) it was another arp reply packet we handled in
1005 * different thread.
1006 *
1007 * In all cases except 3) we definitely need to prefer
1008 * existing lle. For the sake of simplicity, prefer any
1009 * existing lle over newly-create one.
1010 */
1011 if (la_tmp == NULL)
1012 lltable_link_entry(LLTABLE(ifp), la);
1013 IF_AFDATA_WUNLOCK(ifp);
1014
1015 if (la_tmp == NULL) {
1016 arp_mark_lle_reachable(la);
1017 LLE_WUNLOCK(la);
1018 } else {
1019 /* Free newly-create entry and handle packet */
1020 lltable_free_entry(LLTABLE(ifp), la);
1021 la = la_tmp;
1022 la_tmp = NULL;
1023 arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1024 /* arp_check_update_lle() returns @la unlocked */
1025 }
1026 la = NULL;
1027 }
1028 reply:
1029 if (op != ARPOP_REQUEST)
1030 goto drop;
1031 ARPSTAT_INC(rxrequests);
1032
1033 if (itaddr.s_addr == myaddr.s_addr) {
1034 /* Shortcut.. the receiving interface is the target. */
1035 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1036 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1037 } else {
1038 struct llentry *lle = NULL;
1039
1040 sin.sin_addr = itaddr;
1041 IF_AFDATA_RLOCK(ifp);
1042 lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1043 IF_AFDATA_RUNLOCK(ifp);
1044
1045 if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1046 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1047 (void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1048 LLE_RUNLOCK(lle);
1049 } else {
1050
1051 if (lle != NULL)
1052 LLE_RUNLOCK(lle);
1053
1054 if (!V_arp_proxyall)
1055 goto drop;
1056
1057 /* XXX MRT use table 0 for arp reply */
1058 if (fib4_lookup_nh_basic(0, itaddr, 0, 0, &nh4) != 0)
1059 goto drop;
1060
1061 /*
1062 * Don't send proxies for nodes on the same interface
1063 * as this one came out of, or we'll get into a fight
1064 * over who claims what Ether address.
1065 */
1066 if (nh4.nh_ifp == ifp)
1067 goto drop;
1068
1069 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1070 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1071
1072 /*
1073 * Also check that the node which sent the ARP packet
1074 * is on the interface we expect it to be on. This
1075 * avoids ARP chaos if an interface is connected to the
1076 * wrong network.
1077 */
1078
1079 /* XXX MRT use table 0 for arp checks */
1080 if (fib4_lookup_nh_basic(0, isaddr, 0, 0, &nh4) != 0)
1081 goto drop;
1082 if (nh4.nh_ifp != ifp) {
1083 ARP_LOG(LOG_INFO, "proxy: ignoring request"
1084 " from %s via %s\n",
1085 inet_ntoa_r(isaddr, addrbuf),
1086 ifp->if_xname);
1087 goto drop;
1088 }
1089
1090 #ifdef DEBUG_PROXY
1091 printf("arp: proxying for %s\n",
1092 inet_ntoa_r(itaddr, addrbuf));
1093 #endif
1094 }
1095 }
1096
1097 if (itaddr.s_addr == myaddr.s_addr &&
1098 IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1099 /* RFC 3927 link-local IPv4; always reply by broadcast. */
1100 #ifdef DEBUG_LINKLOCAL
1101 printf("arp: sending reply for link-local addr %s\n",
1102 inet_ntoa_r(itaddr, addrbuf));
1103 #endif
1104 m->m_flags |= M_BCAST;
1105 m->m_flags &= ~M_MCAST;
1106 } else {
1107 /* default behaviour; never reply by broadcast. */
1108 m->m_flags &= ~(M_BCAST|M_MCAST);
1109 }
1110 (void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1111 (void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1112 ah->ar_op = htons(ARPOP_REPLY);
1113 ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1114 m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1115 m->m_pkthdr.len = m->m_len;
1116 m->m_pkthdr.rcvif = NULL;
1117 sa.sa_family = AF_ARP;
1118 sa.sa_len = 2;
1119
1120 /* Calculate link header for sending frame */
1121 bzero(&ro, sizeof(ro));
1122 linkhdrsize = sizeof(linkhdr);
1123 error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1124
1125 /*
1126 * arp_fillheader() may fail due to lack of support inside encap request
1127 * routing. This is not necessary an error, AF_ARP can/should be handled
1128 * by if_output().
1129 */
1130 if (error != 0 && error != EAFNOSUPPORT) {
1131 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1132 if_name(ifp), error);
1133 return;
1134 }
1135
1136 ro.ro_prepend = linkhdr;
1137 ro.ro_plen = linkhdrsize;
1138 ro.ro_flags = 0;
1139
1140 m_clrprotoflags(m); /* Avoid confusing lower layers. */
1141 (*ifp->if_output)(ifp, m, &sa, &ro);
1142 ARPSTAT_INC(txreplies);
1143 return;
1144
1145 drop:
1146 m_freem(m);
1147 }
1148 #endif
1149
1150 /*
1151 * Checks received arp data against existing @la.
1152 * Updates lle state/performs notification if necessary.
1153 */
1154 static void
1155 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1156 int bridged, struct llentry *la)
1157 {
1158 struct sockaddr sa;
1159 struct mbuf *m_hold, *m_hold_next;
1160 uint8_t linkhdr[LLE_MAX_LINKHDR];
1161 size_t linkhdrsize;
1162 int lladdr_off;
1163 char addrbuf[INET_ADDRSTRLEN];
1164
1165 LLE_WLOCK_ASSERT(la);
1166
1167 /* the following is not an error when doing bridging */
1168 if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1169 if (log_arp_wrong_iface)
1170 ARP_LOG(LOG_WARNING, "%s is on %s "
1171 "but got reply from %*D on %s\n",
1172 inet_ntoa_r(isaddr, addrbuf),
1173 la->lle_tbl->llt_ifp->if_xname,
1174 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1175 ifp->if_xname);
1176 LLE_WUNLOCK(la);
1177 return;
1178 }
1179 if ((la->la_flags & LLE_VALID) &&
1180 bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1181 if (la->la_flags & LLE_STATIC) {
1182 LLE_WUNLOCK(la);
1183 if (log_arp_permanent_modify)
1184 ARP_LOG(LOG_ERR,
1185 "%*D attempts to modify "
1186 "permanent entry for %s on %s\n",
1187 ifp->if_addrlen,
1188 (u_char *)ar_sha(ah), ":",
1189 inet_ntoa_r(isaddr, addrbuf),
1190 ifp->if_xname);
1191 return;
1192 }
1193 if (log_arp_movements) {
1194 ARP_LOG(LOG_INFO, "%s moved from %*D "
1195 "to %*D on %s\n",
1196 inet_ntoa_r(isaddr, addrbuf),
1197 ifp->if_addrlen,
1198 (u_char *)la->ll_addr, ":",
1199 ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1200 ifp->if_xname);
1201 }
1202 }
1203
1204 /* Calculate full link prepend to use in lle */
1205 linkhdrsize = sizeof(linkhdr);
1206 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1207 &linkhdrsize, &lladdr_off) != 0)
1208 return;
1209
1210 /* Check if something has changed */
1211 if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1212 (la->la_flags & LLE_VALID) == 0) {
1213 /* Try to perform LLE update */
1214 if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1215 lladdr_off) == 0)
1216 return;
1217
1218 /* Clear fast path feedback request if set */
1219 la->r_skip_req = 0;
1220 }
1221
1222 arp_mark_lle_reachable(la);
1223
1224 /*
1225 * The packets are all freed within the call to the output
1226 * routine.
1227 *
1228 * NB: The lock MUST be released before the call to the
1229 * output routine.
1230 */
1231 if (la->la_hold != NULL) {
1232 m_hold = la->la_hold;
1233 la->la_hold = NULL;
1234 la->la_numheld = 0;
1235 lltable_fill_sa_entry(la, &sa);
1236 LLE_WUNLOCK(la);
1237 for (; m_hold != NULL; m_hold = m_hold_next) {
1238 m_hold_next = m_hold->m_nextpkt;
1239 m_hold->m_nextpkt = NULL;
1240 /* Avoid confusing lower layers. */
1241 m_clrprotoflags(m_hold);
1242 (*ifp->if_output)(ifp, m_hold, &sa, NULL);
1243 }
1244 } else
1245 LLE_WUNLOCK(la);
1246 }
1247
1248 static void
1249 arp_mark_lle_reachable(struct llentry *la)
1250 {
1251 int canceled, wtime;
1252
1253 LLE_WLOCK_ASSERT(la);
1254
1255 la->ln_state = ARP_LLINFO_REACHABLE;
1256 EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1257
1258 if (!(la->la_flags & LLE_STATIC)) {
1259 LLE_ADDREF(la);
1260 la->la_expire = time_uptime + V_arpt_keep;
1261 wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1262 if (wtime < 0)
1263 wtime = V_arpt_keep;
1264 canceled = callout_reset(&la->lle_timer,
1265 hz * wtime, arptimer, la);
1266 if (canceled)
1267 LLE_REMREF(la);
1268 }
1269 la->la_asked = 0;
1270 la->la_preempt = V_arp_maxtries;
1271 }
1272
1273 /*
1274 * Add pernament link-layer record for given interface address.
1275 */
1276 static __noinline void
1277 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1278 {
1279 struct llentry *lle, *lle_tmp;
1280
1281 /*
1282 * Interface address LLE record is considered static
1283 * because kernel code relies on LLE_STATIC flag to check
1284 * if these entries can be rewriten by arp updates.
1285 */
1286 lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1287 if (lle == NULL) {
1288 log(LOG_INFO, "arp_ifinit: cannot create arp "
1289 "entry for interface address\n");
1290 return;
1291 }
1292
1293 IF_AFDATA_WLOCK(ifp);
1294 LLE_WLOCK(lle);
1295 /* Unlink any entry if exists */
1296 lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1297 if (lle_tmp != NULL)
1298 lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1299
1300 lltable_link_entry(LLTABLE(ifp), lle);
1301 IF_AFDATA_WUNLOCK(ifp);
1302
1303 if (lle_tmp != NULL)
1304 EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1305
1306 EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1307 LLE_WUNLOCK(lle);
1308 if (lle_tmp != NULL)
1309 lltable_free_entry(LLTABLE(ifp), lle_tmp);
1310 }
1311
1312 /*
1313 * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1314 * of valid values.
1315 */
1316 static int
1317 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1318 {
1319 int error;
1320 int rexmit_count = *(int *)arg1;
1321
1322 error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1323
1324 /* Enforce limits on any new value that may have been set. */
1325 if (!error && req->newptr) {
1326 /* A new value was set. */
1327 if (rexmit_count < 0) {
1328 rexmit_count = 0;
1329 } else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1330 rexmit_count = MAX_GARP_RETRANSMITS;
1331 }
1332 *(int *)arg1 = rexmit_count;
1333 }
1334
1335 return (error);
1336 }
1337
1338 /*
1339 * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1340 * retransmit it again. A pending callout owns a reference to the ifa.
1341 */
1342 static void
1343 garp_rexmit(void *arg)
1344 {
1345 struct in_ifaddr *ia = arg;
1346
1347 if (callout_pending(&ia->ia_garp_timer) ||
1348 !callout_active(&ia->ia_garp_timer)) {
1349 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1350 ifa_free(&ia->ia_ifa);
1351 return;
1352 }
1353
1354 CURVNET_SET(ia->ia_ifa.ifa_ifp->if_vnet);
1355
1356 /*
1357 * Drop lock while the ARP request is generated.
1358 */
1359 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1360
1361 arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1362 &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1363
1364 /*
1365 * Increment the count of retransmissions. If the count has reached the
1366 * maximum value, stop sending the GARP packets. Otherwise, schedule
1367 * the callout to retransmit another GARP packet.
1368 */
1369 ++ia->ia_garp_count;
1370 if (ia->ia_garp_count >= garp_rexmit_count) {
1371 ifa_free(&ia->ia_ifa);
1372 } else {
1373 int rescheduled;
1374 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1375 rescheduled = callout_reset(&ia->ia_garp_timer,
1376 (1 << ia->ia_garp_count) * hz,
1377 garp_rexmit, ia);
1378 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1379 if (rescheduled) {
1380 ifa_free(&ia->ia_ifa);
1381 }
1382 }
1383
1384 CURVNET_RESTORE();
1385 }
1386
1387 /*
1388 * Start the GARP retransmit timer.
1389 *
1390 * A single GARP is always transmitted when an IPv4 address is added
1391 * to an interface and that is usually sufficient. However, in some
1392 * circumstances, such as when a shared address is passed between
1393 * cluster nodes, this single GARP may occasionally be dropped or
1394 * lost. This can lead to neighbors on the network link working with a
1395 * stale ARP cache and sending packets destined for that address to
1396 * the node that previously owned the address, which may not respond.
1397 *
1398 * To avoid this situation, GARP retransmits can be enabled by setting
1399 * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1400 * than zero. The setting represents the maximum number of
1401 * retransmissions. The interval between retransmissions is calculated
1402 * using an exponential backoff algorithm, doubling each time, so the
1403 * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1404 */
1405 static void
1406 garp_timer_start(struct ifaddr *ifa)
1407 {
1408 struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1409
1410 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1411 ia->ia_garp_count = 0;
1412 if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1413 garp_rexmit, ia) == 0) {
1414 ifa_ref(ifa);
1415 }
1416 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1417 }
1418
1419 void
1420 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1421 {
1422 const struct sockaddr_in *dst_in;
1423 const struct sockaddr *dst;
1424
1425 if (ifa->ifa_carp != NULL)
1426 return;
1427
1428 dst = ifa->ifa_addr;
1429 dst_in = (const struct sockaddr_in *)dst;
1430
1431 if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1432 return;
1433 arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1434 if (garp_rexmit_count > 0) {
1435 garp_timer_start(ifa);
1436 }
1437
1438 arp_add_ifa_lle(ifp, dst);
1439 }
1440
1441 void
1442 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1443 {
1444
1445 if (ntohl(addr.s_addr) != INADDR_ANY)
1446 arprequest(ifp, &addr, &addr, enaddr);
1447 }
1448
1449 /*
1450 * Sends gratuitous ARPs for each ifaddr to notify other
1451 * nodes about the address change.
1452 */
1453 static __noinline void
1454 arp_handle_ifllchange(struct ifnet *ifp)
1455 {
1456 struct ifaddr *ifa;
1457
1458 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1459 if (ifa->ifa_addr->sa_family == AF_INET)
1460 arp_ifinit(ifp, ifa);
1461 }
1462 }
1463
1464 /*
1465 * A handler for interface link layer address change event.
1466 */
1467 static void
1468 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1469 {
1470
1471 lltable_update_ifaddr(LLTABLE(ifp));
1472
1473 if ((ifp->if_flags & IFF_UP) != 0)
1474 arp_handle_ifllchange(ifp);
1475 }
1476
1477 static void
1478 vnet_arp_init(void)
1479 {
1480
1481 if (IS_DEFAULT_VNET(curvnet)) {
1482 netisr_register(&arp_nh);
1483 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1484 arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1485 }
1486 #ifdef VIMAGE
1487 else
1488 netisr_register_vnet(&arp_nh);
1489 #endif
1490 }
1491 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1492 vnet_arp_init, 0);
1493
1494 #ifdef VIMAGE
1495 /*
1496 * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1497 * lookups after destroying the hash. Ideally this would go on SI_ORDER_3.5.
1498 */
1499 static void
1500 vnet_arp_destroy(__unused void *arg)
1501 {
1502
1503 netisr_unregister_vnet(&arp_nh);
1504 }
1505 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1506 vnet_arp_destroy, NULL);
1507 #endif
Cache object: 8dec912520056e130d8278a277eb8189
|