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/kernel.h> 47 #include <sys/lock.h> 48 #include <sys/queue.h> 49 #include <sys/sysctl.h> 50 #include <sys/systm.h> 51 #include <sys/mbuf.h> 52 #include <sys/malloc.h> 53 #include <sys/proc.h> 54 #include <sys/rmlock.h> 55 #include <sys/socket.h> 56 #include <sys/syslog.h> 57 58 #include <net/if.h> 59 #include <net/if_var.h> 60 #include <net/if_dl.h> 61 #include <net/if_types.h> 62 #include <net/netisr.h> 63 #include <net/ethernet.h> 64 #include <net/route.h> 65 #include <net/vnet.h> 66 67 #include <netinet/in.h> 68 #include <netinet/in_fib.h> 69 #include <netinet/in_var.h> 70 #include <net/if_llatbl.h> 71 #include <netinet/if_ether.h> 72 #ifdef INET 73 #include <netinet/ip_carp.h> 74 #endif 75 76 #include <security/mac/mac_framework.h> 77 78 #define SIN(s) ((const struct sockaddr_in *)(s)) 79 80 static struct timeval arp_lastlog; 81 static int arp_curpps; 82 static int arp_maxpps = 1; 83 84 /* Simple ARP state machine */ 85 enum arp_llinfo_state { 86 ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */ 87 ARP_LLINFO_REACHABLE, /* LLE is valid */ 88 ARP_LLINFO_VERIFY, /* LLE is valid, need refresh */ 89 ARP_LLINFO_DELETED, /* LLE is deleted */ 90 }; 91 92 SYSCTL_DECL(_net_link_ether); 93 static SYSCTL_NODE(_net_link_ether, PF_INET, inet, CTLFLAG_RW, 0, ""); 94 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp, CTLFLAG_RW, 0, ""); 95 96 /* timer values */ 97 VNET_DEFINE_STATIC(int, arpt_keep) = (20*60); /* once resolved, good for 20 98 * minutes */ 99 VNET_DEFINE_STATIC(int, arp_maxtries) = 5; 100 VNET_DEFINE_STATIC(int, arp_proxyall) = 0; 101 VNET_DEFINE_STATIC(int, arpt_down) = 20; /* keep incomplete entries for 102 * 20 seconds */ 103 VNET_DEFINE_STATIC(int, arpt_rexmit) = 1; /* retransmit arp entries, sec*/ 104 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat); /* ARP statistics, see if_arp.h */ 105 VNET_PCPUSTAT_SYSINIT(arpstat); 106 107 #ifdef VIMAGE 108 VNET_PCPUSTAT_SYSUNINIT(arpstat); 109 #endif /* VIMAGE */ 110 111 VNET_DEFINE_STATIC(int, arp_maxhold) = 1; 112 113 #define V_arpt_keep VNET(arpt_keep) 114 #define V_arpt_down VNET(arpt_down) 115 #define V_arpt_rexmit VNET(arpt_rexmit) 116 #define V_arp_maxtries VNET(arp_maxtries) 117 #define V_arp_proxyall VNET(arp_proxyall) 118 #define V_arp_maxhold VNET(arp_maxhold) 119 120 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW, 121 &VNET_NAME(arpt_keep), 0, 122 "ARP entry lifetime in seconds"); 123 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW, 124 &VNET_NAME(arp_maxtries), 0, 125 "ARP resolution attempts before returning error"); 126 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW, 127 &VNET_NAME(arp_proxyall), 0, 128 "Enable proxy ARP for all suitable requests"); 129 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW, 130 &VNET_NAME(arpt_down), 0, 131 "Incomplete ARP entry lifetime in seconds"); 132 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat, 133 arpstat, "ARP statistics (struct arpstat, net/if_arp.h)"); 134 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW, 135 &VNET_NAME(arp_maxhold), 0, 136 "Number of packets to hold per ARP entry"); 137 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second, 138 CTLFLAG_RW, &arp_maxpps, 0, 139 "Maximum number of remotely triggered ARP messages that can be " 140 "logged per second"); 141 142 /* 143 * Due to the exponential backoff algorithm used for the interval between GARP 144 * retransmissions, the maximum number of retransmissions is limited for 145 * sanity. This limit corresponds to a maximum interval between retransmissions 146 * of 2^16 seconds ~= 18 hours. 147 * 148 * Making this limit more dynamic is more complicated than worthwhile, 149 * especially since sending out GARPs spaced days apart would be of little 150 * use. A maximum dynamic limit would look something like: 151 * 152 * const int max = fls(INT_MAX / hz) - 1; 153 */ 154 #define MAX_GARP_RETRANSMITS 16 155 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS); 156 static int garp_rexmit_count = 0; /* GARP retransmission setting. */ 157 158 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count, 159 CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE, 160 &garp_rexmit_count, 0, sysctl_garp_rexmit, "I", 161 "Number of times to retransmit GARP packets;" 162 " 0 to disable, maximum of 16"); 163 164 VNET_DEFINE_STATIC(int, arp_log_level) = LOG_INFO; /* Min. log(9) level. */ 165 #define V_arp_log_level VNET(arp_log_level) 166 SYSCTL_INT(_net_link_ether_arp, OID_AUTO, log_level, CTLFLAG_VNET | CTLFLAG_RW, 167 &VNET_NAME(arp_log_level), 0, 168 "Minimum log(9) level for recording rate limited arp log messages. " 169 "The higher will be log more (emerg=0, info=6 (default), debug=7)."); 170 #define ARP_LOG(pri, ...) do { \ 171 if ((pri) <= V_arp_log_level && \ 172 ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps)) \ 173 log((pri), "arp: " __VA_ARGS__); \ 174 } while (0) 175 176 177 static void arpintr(struct mbuf *); 178 static void arptimer(void *); 179 #ifdef INET 180 static void in_arpinput(struct mbuf *); 181 #endif 182 183 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, 184 struct ifnet *ifp, int bridged, struct llentry *la); 185 static void arp_mark_lle_reachable(struct llentry *la); 186 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp); 187 188 static eventhandler_tag iflladdr_tag; 189 190 static const struct netisr_handler arp_nh = { 191 .nh_name = "arp", 192 .nh_handler = arpintr, 193 .nh_proto = NETISR_ARP, 194 .nh_policy = NETISR_POLICY_SOURCE, 195 }; 196 197 /* 198 * Timeout routine. Age arp_tab entries periodically. 199 */ 200 static void 201 arptimer(void *arg) 202 { 203 struct llentry *lle = (struct llentry *)arg; 204 struct ifnet *ifp; 205 int r_skip_req; 206 207 if (lle->la_flags & LLE_STATIC) { 208 return; 209 } 210 LLE_WLOCK(lle); 211 if (callout_pending(&lle->lle_timer)) { 212 /* 213 * Here we are a bit odd here in the treatment of 214 * active/pending. If the pending bit is set, it got 215 * rescheduled before I ran. The active 216 * bit we ignore, since if it was stopped 217 * in ll_tablefree() and was currently running 218 * it would have return 0 so the code would 219 * not have deleted it since the callout could 220 * not be stopped so we want to go through 221 * with the delete here now. If the callout 222 * was restarted, the pending bit will be back on and 223 * we just want to bail since the callout_reset would 224 * return 1 and our reference would have been removed 225 * by arpresolve() below. 226 */ 227 LLE_WUNLOCK(lle); 228 return; 229 } 230 ifp = lle->lle_tbl->llt_ifp; 231 CURVNET_SET(ifp->if_vnet); 232 233 switch (lle->ln_state) { 234 case ARP_LLINFO_REACHABLE: 235 236 /* 237 * Expiration time is approaching. 238 * Let's try to refresh entry if it is still 239 * in use. 240 * 241 * Set r_skip_req to get feedback from 242 * fast path. Change state and re-schedule 243 * ourselves. 244 */ 245 LLE_REQ_LOCK(lle); 246 lle->r_skip_req = 1; 247 LLE_REQ_UNLOCK(lle); 248 lle->ln_state = ARP_LLINFO_VERIFY; 249 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit); 250 LLE_WUNLOCK(lle); 251 CURVNET_RESTORE(); 252 return; 253 case ARP_LLINFO_VERIFY: 254 LLE_REQ_LOCK(lle); 255 r_skip_req = lle->r_skip_req; 256 LLE_REQ_UNLOCK(lle); 257 258 if (r_skip_req == 0 && lle->la_preempt > 0) { 259 /* Entry was used, issue refresh request */ 260 struct in_addr dst; 261 dst = lle->r_l3addr.addr4; 262 lle->la_preempt--; 263 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit); 264 LLE_WUNLOCK(lle); 265 arprequest(ifp, NULL, &dst, NULL); 266 CURVNET_RESTORE(); 267 return; 268 } 269 /* Nothing happened. Reschedule if not too late */ 270 if (lle->la_expire > time_uptime) { 271 callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit); 272 LLE_WUNLOCK(lle); 273 CURVNET_RESTORE(); 274 return; 275 } 276 break; 277 case ARP_LLINFO_INCOMPLETE: 278 case ARP_LLINFO_DELETED: 279 break; 280 } 281 282 if ((lle->la_flags & LLE_DELETED) == 0) { 283 int evt; 284 285 if (lle->la_flags & LLE_VALID) 286 evt = LLENTRY_EXPIRED; 287 else 288 evt = LLENTRY_TIMEDOUT; 289 EVENTHANDLER_INVOKE(lle_event, lle, evt); 290 } 291 292 callout_stop(&lle->lle_timer); 293 294 /* XXX: LOR avoidance. We still have ref on lle. */ 295 LLE_WUNLOCK(lle); 296 IF_AFDATA_LOCK(ifp); 297 LLE_WLOCK(lle); 298 299 /* Guard against race with other llentry_free(). */ 300 if (lle->la_flags & LLE_LINKED) { 301 LLE_REMREF(lle); 302 lltable_unlink_entry(lle->lle_tbl, lle); 303 } 304 IF_AFDATA_UNLOCK(ifp); 305 306 size_t pkts_dropped = llentry_free(lle); 307 308 ARPSTAT_ADD(dropped, pkts_dropped); 309 ARPSTAT_INC(timeouts); 310 311 CURVNET_RESTORE(); 312 } 313 314 /* 315 * Stores link-layer header for @ifp in format suitable for if_output() 316 * into buffer @buf. Resulting header length is stored in @bufsize. 317 * 318 * Returns 0 on success. 319 */ 320 static int 321 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf, 322 size_t *bufsize) 323 { 324 struct if_encap_req ereq; 325 int error; 326 327 bzero(buf, *bufsize); 328 bzero(&ereq, sizeof(ereq)); 329 ereq.buf = buf; 330 ereq.bufsize = *bufsize; 331 ereq.rtype = IFENCAP_LL; 332 ereq.family = AF_ARP; 333 ereq.lladdr = ar_tha(ah); 334 ereq.hdata = (u_char *)ah; 335 if (bcast) 336 ereq.flags = IFENCAP_FLAG_BROADCAST; 337 error = ifp->if_requestencap(ifp, &ereq); 338 if (error == 0) 339 *bufsize = ereq.bufsize; 340 341 return (error); 342 } 343 344 345 /* 346 * Broadcast an ARP request. Caller specifies: 347 * - arp header source ip address 348 * - arp header target ip address 349 * - arp header source ethernet address 350 */ 351 static int 352 arprequest_internal(struct ifnet *ifp, const struct in_addr *sip, 353 const struct in_addr *tip, u_char *enaddr) 354 { 355 struct mbuf *m; 356 struct arphdr *ah; 357 struct sockaddr sa; 358 u_char *carpaddr = NULL; 359 uint8_t linkhdr[LLE_MAX_LINKHDR]; 360 size_t linkhdrsize; 361 struct route ro; 362 int error; 363 364 if (sip == NULL) { 365 /* 366 * The caller did not supply a source address, try to find 367 * a compatible one among those assigned to this interface. 368 */ 369 struct epoch_tracker et; 370 struct ifaddr *ifa; 371 372 NET_EPOCH_ENTER(et); 373 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 374 if (ifa->ifa_addr->sa_family != AF_INET) 375 continue; 376 377 if (ifa->ifa_carp) { 378 if ((*carp_iamatch_p)(ifa, &carpaddr) == 0) 379 continue; 380 sip = &IA_SIN(ifa)->sin_addr; 381 } else { 382 carpaddr = NULL; 383 sip = &IA_SIN(ifa)->sin_addr; 384 } 385 386 if (0 == ((sip->s_addr ^ tip->s_addr) & 387 IA_MASKSIN(ifa)->sin_addr.s_addr)) 388 break; /* found it. */ 389 } 390 NET_EPOCH_EXIT(et); 391 if (sip == NULL) { 392 printf("%s: cannot find matching address\n", __func__); 393 return (EADDRNOTAVAIL); 394 } 395 } 396 if (enaddr == NULL) 397 enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp); 398 399 if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL) 400 return (ENOMEM); 401 m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) + 402 2 * ifp->if_addrlen; 403 m->m_pkthdr.len = m->m_len; 404 M_ALIGN(m, m->m_len); 405 ah = mtod(m, struct arphdr *); 406 bzero((caddr_t)ah, m->m_len); 407 #ifdef MAC 408 mac_netinet_arp_send(ifp, m); 409 #endif 410 ah->ar_pro = htons(ETHERTYPE_IP); 411 ah->ar_hln = ifp->if_addrlen; /* hardware address length */ 412 ah->ar_pln = sizeof(struct in_addr); /* protocol address length */ 413 ah->ar_op = htons(ARPOP_REQUEST); 414 bcopy(enaddr, ar_sha(ah), ah->ar_hln); 415 bcopy(sip, ar_spa(ah), ah->ar_pln); 416 bcopy(tip, ar_tpa(ah), ah->ar_pln); 417 sa.sa_family = AF_ARP; 418 sa.sa_len = 2; 419 420 /* Calculate link header for sending frame */ 421 bzero(&ro, sizeof(ro)); 422 linkhdrsize = sizeof(linkhdr); 423 error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize); 424 if (error != 0 && error != EAFNOSUPPORT) { 425 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n", 426 if_name(ifp), error); 427 return (error); 428 } 429 430 ro.ro_prepend = linkhdr; 431 ro.ro_plen = linkhdrsize; 432 ro.ro_flags = 0; 433 434 m->m_flags |= M_BCAST; 435 m_clrprotoflags(m); /* Avoid confusing lower layers. */ 436 error = (*ifp->if_output)(ifp, m, &sa, &ro); 437 ARPSTAT_INC(txrequests); 438 if (error) { 439 ARPSTAT_INC(txerrors); 440 ARP_LOG(LOG_DEBUG, "Failed to send ARP packet on %s: %d\n", 441 if_name(ifp), error); 442 } 443 return (error); 444 } 445 446 void 447 arprequest(struct ifnet *ifp, const struct in_addr *sip, 448 const struct in_addr *tip, u_char *enaddr) 449 { 450 451 (void) arprequest_internal(ifp, sip, tip, enaddr); 452 } 453 454 /* 455 * Resolve an IP address into an ethernet address - heavy version. 456 * Used internally by arpresolve(). 457 * We have already checked that we can't use an existing lle without 458 * modification so we have to acquire an LLE_EXCLUSIVE lle lock. 459 * 460 * On success, desten and pflags are filled in and the function returns 0; 461 * If the packet must be held pending resolution, we return EWOULDBLOCK 462 * On other errors, we return the corresponding error code. 463 * Note that m_freem() handles NULL. 464 */ 465 static int 466 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m, 467 const struct sockaddr *dst, u_char *desten, uint32_t *pflags, 468 struct llentry **plle) 469 { 470 struct llentry *la = NULL, *la_tmp; 471 struct mbuf *curr = NULL; 472 struct mbuf *next = NULL; 473 int error, renew; 474 char *lladdr; 475 int ll_len; 476 477 if (pflags != NULL) 478 *pflags = 0; 479 if (plle != NULL) 480 *plle = NULL; 481 482 if ((flags & LLE_CREATE) == 0) { 483 struct epoch_tracker et; 484 485 NET_EPOCH_ENTER(et); 486 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst); 487 NET_EPOCH_EXIT(et); 488 } 489 if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) { 490 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst); 491 if (la == NULL) { 492 char addrbuf[INET_ADDRSTRLEN]; 493 494 log(LOG_DEBUG, 495 "arpresolve: can't allocate llinfo for %s on %s\n", 496 inet_ntoa_r(SIN(dst)->sin_addr, addrbuf), 497 if_name(ifp)); 498 m_freem(m); 499 return (EINVAL); 500 } 501 502 IF_AFDATA_WLOCK(ifp); 503 LLE_WLOCK(la); 504 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst); 505 /* Prefer ANY existing lle over newly-created one */ 506 if (la_tmp == NULL) 507 lltable_link_entry(LLTABLE(ifp), la); 508 IF_AFDATA_WUNLOCK(ifp); 509 if (la_tmp != NULL) { 510 lltable_free_entry(LLTABLE(ifp), la); 511 la = la_tmp; 512 } 513 } 514 if (la == NULL) { 515 m_freem(m); 516 return (EINVAL); 517 } 518 519 if ((la->la_flags & LLE_VALID) && 520 ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) { 521 if (flags & LLE_ADDRONLY) { 522 lladdr = la->ll_addr; 523 ll_len = ifp->if_addrlen; 524 } else { 525 lladdr = la->r_linkdata; 526 ll_len = la->r_hdrlen; 527 } 528 bcopy(lladdr, desten, ll_len); 529 530 /* Notify LLE code that the entry was used by datapath */ 531 llentry_mark_used(la); 532 if (pflags != NULL) 533 *pflags = la->la_flags & (LLE_VALID|LLE_IFADDR); 534 if (plle) { 535 LLE_ADDREF(la); 536 *plle = la; 537 } 538 LLE_WUNLOCK(la); 539 return (0); 540 } 541 542 renew = (la->la_asked == 0 || la->la_expire != time_uptime); 543 /* 544 * There is an arptab entry, but no ethernet address 545 * response yet. Add the mbuf to the list, dropping 546 * the oldest packet if we have exceeded the system 547 * setting. 548 */ 549 if (m != NULL) { 550 if (la->la_numheld >= V_arp_maxhold) { 551 if (la->la_hold != NULL) { 552 next = la->la_hold->m_nextpkt; 553 m_freem(la->la_hold); 554 la->la_hold = next; 555 la->la_numheld--; 556 ARPSTAT_INC(dropped); 557 } 558 } 559 if (la->la_hold != NULL) { 560 curr = la->la_hold; 561 while (curr->m_nextpkt != NULL) 562 curr = curr->m_nextpkt; 563 curr->m_nextpkt = m; 564 } else 565 la->la_hold = m; 566 la->la_numheld++; 567 } 568 /* 569 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It 570 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH 571 * if we have already sent arp_maxtries ARP requests. Retransmit the 572 * ARP request, but not faster than one request per second. 573 */ 574 if (la->la_asked < V_arp_maxtries) 575 error = EWOULDBLOCK; /* First request. */ 576 else 577 error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN; 578 579 if (renew) { 580 int canceled, e; 581 582 LLE_ADDREF(la); 583 la->la_expire = time_uptime; 584 canceled = callout_reset(&la->lle_timer, hz * V_arpt_down, 585 arptimer, la); 586 if (canceled) 587 LLE_REMREF(la); 588 la->la_asked++; 589 LLE_WUNLOCK(la); 590 e = arprequest_internal(ifp, NULL, &SIN(dst)->sin_addr, NULL); 591 /* 592 * Only overwrite 'error' in case of error; in case of success 593 * the proper return value was already set above. 594 */ 595 if (e != 0) 596 return (e); 597 return (error); 598 } 599 600 LLE_WUNLOCK(la); 601 return (error); 602 } 603 604 /* 605 * Lookups link header based on an IP address. 606 * On input: 607 * ifp is the interface we use 608 * is_gw != 0 if @dst represents gateway to some destination 609 * m is the mbuf. May be NULL if we don't have a packet. 610 * dst is the next hop, 611 * desten is the storage to put LL header. 612 * flags returns subset of lle flags: LLE_VALID | LLE_IFADDR 613 * 614 * On success, full/partial link header and flags are filled in and 615 * the function returns 0. 616 * If the packet must be held pending resolution, we return EWOULDBLOCK 617 * On other errors, we return the corresponding error code. 618 * Note that m_freem() handles NULL. 619 */ 620 int 621 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m, 622 const struct sockaddr *dst, u_char *desten, uint32_t *pflags, 623 struct llentry **plle) 624 { 625 struct epoch_tracker et; 626 struct llentry *la = NULL; 627 628 if (pflags != NULL) 629 *pflags = 0; 630 if (plle != NULL) 631 *plle = NULL; 632 633 if (m != NULL) { 634 if (m->m_flags & M_BCAST) { 635 /* broadcast */ 636 (void)memcpy(desten, 637 ifp->if_broadcastaddr, ifp->if_addrlen); 638 return (0); 639 } 640 if (m->m_flags & M_MCAST) { 641 /* multicast */ 642 ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten); 643 return (0); 644 } 645 } 646 647 NET_EPOCH_ENTER(et); 648 la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst); 649 if (la != NULL && (la->r_flags & RLLE_VALID) != 0) { 650 /* Entry found, let's copy lle info */ 651 bcopy(la->r_linkdata, desten, la->r_hdrlen); 652 if (pflags != NULL) 653 *pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR); 654 /* Notify the LLE handling code that the entry was used. */ 655 llentry_mark_used(la); 656 if (plle) { 657 LLE_ADDREF(la); 658 *plle = la; 659 LLE_WUNLOCK(la); 660 } 661 NET_EPOCH_EXIT(et); 662 return (0); 663 } 664 if (plle && la) 665 LLE_WUNLOCK(la); 666 NET_EPOCH_EXIT(et); 667 668 return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst, 669 desten, pflags, plle)); 670 } 671 672 /* 673 * Common length and type checks are done here, 674 * then the protocol-specific routine is called. 675 */ 676 static void 677 arpintr(struct mbuf *m) 678 { 679 struct arphdr *ar; 680 struct ifnet *ifp; 681 char *layer; 682 int hlen; 683 684 ifp = m->m_pkthdr.rcvif; 685 686 if (m->m_len < sizeof(struct arphdr) && 687 ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) { 688 ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n", 689 if_name(ifp)); 690 return; 691 } 692 ar = mtod(m, struct arphdr *); 693 694 /* Check if length is sufficient */ 695 if (m->m_len < arphdr_len(ar)) { 696 m = m_pullup(m, arphdr_len(ar)); 697 if (m == NULL) { 698 ARP_LOG(LOG_NOTICE, "short packet received on %s\n", 699 if_name(ifp)); 700 return; 701 } 702 ar = mtod(m, struct arphdr *); 703 } 704 705 hlen = 0; 706 layer = ""; 707 switch (ntohs(ar->ar_hrd)) { 708 case ARPHRD_ETHER: 709 hlen = ETHER_ADDR_LEN; /* RFC 826 */ 710 layer = "ethernet"; 711 break; 712 case ARPHRD_INFINIBAND: 713 hlen = 20; /* RFC 4391, INFINIBAND_ALEN */ 714 layer = "infiniband"; 715 break; 716 case ARPHRD_IEEE1394: 717 hlen = 0; /* SHALL be 16 */ /* RFC 2734 */ 718 layer = "firewire"; 719 720 /* 721 * Restrict too long hardware addresses. 722 * Currently we are capable of handling 20-byte 723 * addresses ( sizeof(lle->ll_addr) ) 724 */ 725 if (ar->ar_hln >= 20) 726 hlen = 16; 727 break; 728 default: 729 ARP_LOG(LOG_NOTICE, 730 "packet with unknown hardware format 0x%02d received on " 731 "%s\n", ntohs(ar->ar_hrd), if_name(ifp)); 732 m_freem(m); 733 return; 734 } 735 736 if (hlen != 0 && hlen != ar->ar_hln) { 737 ARP_LOG(LOG_NOTICE, 738 "packet with invalid %s address length %d received on %s\n", 739 layer, ar->ar_hln, if_name(ifp)); 740 m_freem(m); 741 return; 742 } 743 744 ARPSTAT_INC(received); 745 switch (ntohs(ar->ar_pro)) { 746 #ifdef INET 747 case ETHERTYPE_IP: 748 in_arpinput(m); 749 return; 750 #endif 751 } 752 m_freem(m); 753 } 754 755 #ifdef INET 756 /* 757 * ARP for Internet protocols on 10 Mb/s Ethernet. 758 * Algorithm is that given in RFC 826. 759 * In addition, a sanity check is performed on the sender 760 * protocol address, to catch impersonators. 761 * We no longer handle negotiations for use of trailer protocol: 762 * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent 763 * along with IP replies if we wanted trailers sent to us, 764 * and also sent them in response to IP replies. 765 * This allowed either end to announce the desire to receive 766 * trailer packets. 767 * We no longer reply to requests for ETHERTYPE_TRAIL protocol either, 768 * but formerly didn't normally send requests. 769 */ 770 static int log_arp_wrong_iface = 1; 771 static int log_arp_movements = 1; 772 static int log_arp_permanent_modify = 1; 773 static int allow_multicast = 0; 774 775 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW, 776 &log_arp_wrong_iface, 0, 777 "log arp packets arriving on the wrong interface"); 778 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW, 779 &log_arp_movements, 0, 780 "log arp replies from MACs different than the one in the cache"); 781 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW, 782 &log_arp_permanent_modify, 0, 783 "log arp replies from MACs different than the one in the permanent arp entry"); 784 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW, 785 &allow_multicast, 0, "accept multicast addresses"); 786 787 static void 788 in_arpinput(struct mbuf *m) 789 { 790 struct rm_priotracker in_ifa_tracker; 791 struct arphdr *ah; 792 struct ifnet *ifp = m->m_pkthdr.rcvif; 793 struct llentry *la = NULL, *la_tmp; 794 struct ifaddr *ifa; 795 struct in_ifaddr *ia; 796 struct sockaddr sa; 797 struct in_addr isaddr, itaddr, myaddr; 798 u_int8_t *enaddr = NULL; 799 int op; 800 int bridged = 0, is_bridge = 0; 801 int carped; 802 struct sockaddr_in sin; 803 struct sockaddr *dst; 804 struct nhop4_basic nh4; 805 uint8_t linkhdr[LLE_MAX_LINKHDR]; 806 struct route ro; 807 size_t linkhdrsize; 808 int lladdr_off; 809 int error; 810 char addrbuf[INET_ADDRSTRLEN]; 811 struct epoch_tracker et; 812 813 sin.sin_len = sizeof(struct sockaddr_in); 814 sin.sin_family = AF_INET; 815 sin.sin_addr.s_addr = 0; 816 817 if (ifp->if_bridge) 818 bridged = 1; 819 if (ifp->if_type == IFT_BRIDGE) 820 is_bridge = 1; 821 822 /* 823 * We already have checked that mbuf contains enough contiguous data 824 * to hold entire arp message according to the arp header. 825 */ 826 ah = mtod(m, struct arphdr *); 827 828 /* 829 * ARP is only for IPv4 so we can reject packets with 830 * a protocol length not equal to an IPv4 address. 831 */ 832 if (ah->ar_pln != sizeof(struct in_addr)) { 833 ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n", 834 sizeof(struct in_addr)); 835 goto drop; 836 } 837 838 if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) { 839 ARP_LOG(LOG_NOTICE, "%*D is multicast\n", 840 ifp->if_addrlen, (u_char *)ar_sha(ah), ":"); 841 goto drop; 842 } 843 844 op = ntohs(ah->ar_op); 845 (void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr)); 846 (void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr)); 847 848 if (op == ARPOP_REPLY) 849 ARPSTAT_INC(rxreplies); 850 851 /* 852 * For a bridge, we want to check the address irrespective 853 * of the receive interface. (This will change slightly 854 * when we have clusters of interfaces). 855 */ 856 IN_IFADDR_RLOCK(&in_ifa_tracker); 857 LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) { 858 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) || 859 ia->ia_ifp == ifp) && 860 itaddr.s_addr == ia->ia_addr.sin_addr.s_addr && 861 (ia->ia_ifa.ifa_carp == NULL || 862 (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) { 863 ifa_ref(&ia->ia_ifa); 864 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 865 goto match; 866 } 867 } 868 LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash) 869 if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) || 870 ia->ia_ifp == ifp) && 871 isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) { 872 ifa_ref(&ia->ia_ifa); 873 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 874 goto match; 875 } 876 877 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia) \ 878 (ia->ia_ifp->if_bridge == ifp->if_softc && \ 879 !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) && \ 880 addr == ia->ia_addr.sin_addr.s_addr) 881 /* 882 * Check the case when bridge shares its MAC address with 883 * some of its children, so packets are claimed by bridge 884 * itself (bridge_input() does it first), but they are really 885 * meant to be destined to the bridge member. 886 */ 887 if (is_bridge) { 888 LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) { 889 if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) { 890 ifa_ref(&ia->ia_ifa); 891 ifp = ia->ia_ifp; 892 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 893 goto match; 894 } 895 } 896 } 897 #undef BDG_MEMBER_MATCHES_ARP 898 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 899 900 /* 901 * No match, use the first inet address on the receive interface 902 * as a dummy address for the rest of the function. 903 */ 904 NET_EPOCH_ENTER(et); 905 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) 906 if (ifa->ifa_addr->sa_family == AF_INET && 907 (ifa->ifa_carp == NULL || 908 (*carp_iamatch_p)(ifa, &enaddr))) { 909 ia = ifatoia(ifa); 910 ifa_ref(ifa); 911 NET_EPOCH_EXIT(et); 912 goto match; 913 } 914 NET_EPOCH_EXIT(et); 915 916 /* 917 * If bridging, fall back to using any inet address. 918 */ 919 IN_IFADDR_RLOCK(&in_ifa_tracker); 920 if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL) { 921 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 922 goto drop; 923 } 924 ifa_ref(&ia->ia_ifa); 925 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 926 match: 927 if (!enaddr) 928 enaddr = (u_int8_t *)IF_LLADDR(ifp); 929 carped = (ia->ia_ifa.ifa_carp != NULL); 930 myaddr = ia->ia_addr.sin_addr; 931 ifa_free(&ia->ia_ifa); 932 if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen)) 933 goto drop; /* it's from me, ignore it. */ 934 if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) { 935 ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address " 936 "%s!\n", inet_ntoa_r(isaddr, addrbuf)); 937 goto drop; 938 } 939 940 if (ifp->if_addrlen != ah->ar_hln) { 941 ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, " 942 "i/f %d (ignored)\n", ifp->if_addrlen, 943 (u_char *) ar_sha(ah), ":", ah->ar_hln, 944 ifp->if_addrlen); 945 goto drop; 946 } 947 948 /* 949 * Warn if another host is using the same IP address, but only if the 950 * IP address isn't 0.0.0.0, which is used for DHCP only, in which 951 * case we suppress the warning to avoid false positive complaints of 952 * potential misconfiguration. 953 */ 954 if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr && 955 myaddr.s_addr != 0) { 956 ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n", 957 ifp->if_addrlen, (u_char *)ar_sha(ah), ":", 958 inet_ntoa_r(isaddr, addrbuf), ifp->if_xname); 959 itaddr = myaddr; 960 ARPSTAT_INC(dupips); 961 goto reply; 962 } 963 if (ifp->if_flags & IFF_STATICARP) 964 goto reply; 965 966 bzero(&sin, sizeof(sin)); 967 sin.sin_len = sizeof(struct sockaddr_in); 968 sin.sin_family = AF_INET; 969 sin.sin_addr = isaddr; 970 dst = (struct sockaddr *)&sin; 971 NET_EPOCH_ENTER(et); 972 la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst); 973 NET_EPOCH_EXIT(et); 974 if (la != NULL) 975 arp_check_update_lle(ah, isaddr, ifp, bridged, la); 976 else if (itaddr.s_addr == myaddr.s_addr) { 977 /* 978 * Request/reply to our address, but no lle exists yet. 979 * Calculate full link prepend to use in lle. 980 */ 981 linkhdrsize = sizeof(linkhdr); 982 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr, 983 &linkhdrsize, &lladdr_off) != 0) 984 goto reply; 985 986 /* Allocate new entry */ 987 la = lltable_alloc_entry(LLTABLE(ifp), 0, dst); 988 if (la == NULL) { 989 990 /* 991 * lle creation may fail if source address belongs 992 * to non-directly connected subnet. However, we 993 * will try to answer the request instead of dropping 994 * frame. 995 */ 996 goto reply; 997 } 998 lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize, 999 lladdr_off); 1000 1001 IF_AFDATA_WLOCK(ifp); 1002 LLE_WLOCK(la); 1003 la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst); 1004 1005 /* 1006 * Check if lle still does not exists. 1007 * If it does, that means that we either 1008 * 1) have configured it explicitly, via 1009 * 1a) 'arp -s' static entry or 1010 * 1b) interface address static record 1011 * or 1012 * 2) it was the result of sending first packet to-host 1013 * or 1014 * 3) it was another arp reply packet we handled in 1015 * different thread. 1016 * 1017 * In all cases except 3) we definitely need to prefer 1018 * existing lle. For the sake of simplicity, prefer any 1019 * existing lle over newly-create one. 1020 */ 1021 if (la_tmp == NULL) 1022 lltable_link_entry(LLTABLE(ifp), la); 1023 IF_AFDATA_WUNLOCK(ifp); 1024 1025 if (la_tmp == NULL) { 1026 arp_mark_lle_reachable(la); 1027 LLE_WUNLOCK(la); 1028 } else { 1029 /* Free newly-create entry and handle packet */ 1030 lltable_free_entry(LLTABLE(ifp), la); 1031 la = la_tmp; 1032 la_tmp = NULL; 1033 arp_check_update_lle(ah, isaddr, ifp, bridged, la); 1034 /* arp_check_update_lle() returns @la unlocked */ 1035 } 1036 la = NULL; 1037 } 1038 reply: 1039 if (op != ARPOP_REQUEST) 1040 goto drop; 1041 ARPSTAT_INC(rxrequests); 1042 1043 if (itaddr.s_addr == myaddr.s_addr) { 1044 /* Shortcut.. the receiving interface is the target. */ 1045 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln); 1046 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln); 1047 } else { 1048 struct llentry *lle = NULL; 1049 1050 sin.sin_addr = itaddr; 1051 NET_EPOCH_ENTER(et); 1052 lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin); 1053 NET_EPOCH_EXIT(et); 1054 1055 if ((lle != NULL) && (lle->la_flags & LLE_PUB)) { 1056 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln); 1057 (void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln); 1058 LLE_RUNLOCK(lle); 1059 } else { 1060 1061 if (lle != NULL) 1062 LLE_RUNLOCK(lle); 1063 1064 if (!V_arp_proxyall) 1065 goto drop; 1066 1067 /* XXX MRT use table 0 for arp reply */ 1068 if (fib4_lookup_nh_basic(0, itaddr, 0, 0, &nh4) != 0) 1069 goto drop; 1070 1071 /* 1072 * Don't send proxies for nodes on the same interface 1073 * as this one came out of, or we'll get into a fight 1074 * over who claims what Ether address. 1075 */ 1076 if (nh4.nh_ifp == ifp) 1077 goto drop; 1078 1079 (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln); 1080 (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln); 1081 1082 /* 1083 * Also check that the node which sent the ARP packet 1084 * is on the interface we expect it to be on. This 1085 * avoids ARP chaos if an interface is connected to the 1086 * wrong network. 1087 */ 1088 1089 /* XXX MRT use table 0 for arp checks */ 1090 if (fib4_lookup_nh_basic(0, isaddr, 0, 0, &nh4) != 0) 1091 goto drop; 1092 if (nh4.nh_ifp != ifp) { 1093 ARP_LOG(LOG_INFO, "proxy: ignoring request" 1094 " from %s via %s\n", 1095 inet_ntoa_r(isaddr, addrbuf), 1096 ifp->if_xname); 1097 goto drop; 1098 } 1099 1100 #ifdef DEBUG_PROXY 1101 printf("arp: proxying for %s\n", 1102 inet_ntoa_r(itaddr, addrbuf)); 1103 #endif 1104 } 1105 } 1106 1107 if (itaddr.s_addr == myaddr.s_addr && 1108 IN_LINKLOCAL(ntohl(itaddr.s_addr))) { 1109 /* RFC 3927 link-local IPv4; always reply by broadcast. */ 1110 #ifdef DEBUG_LINKLOCAL 1111 printf("arp: sending reply for link-local addr %s\n", 1112 inet_ntoa_r(itaddr, addrbuf)); 1113 #endif 1114 m->m_flags |= M_BCAST; 1115 m->m_flags &= ~M_MCAST; 1116 } else { 1117 /* default behaviour; never reply by broadcast. */ 1118 m->m_flags &= ~(M_BCAST|M_MCAST); 1119 } 1120 (void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln); 1121 (void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln); 1122 ah->ar_op = htons(ARPOP_REPLY); 1123 ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */ 1124 m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln); 1125 m->m_pkthdr.len = m->m_len; 1126 m->m_pkthdr.rcvif = NULL; 1127 sa.sa_family = AF_ARP; 1128 sa.sa_len = 2; 1129 1130 /* Calculate link header for sending frame */ 1131 bzero(&ro, sizeof(ro)); 1132 linkhdrsize = sizeof(linkhdr); 1133 error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize); 1134 1135 /* 1136 * arp_fillheader() may fail due to lack of support inside encap request 1137 * routing. This is not necessary an error, AF_ARP can/should be handled 1138 * by if_output(). 1139 */ 1140 if (error != 0 && error != EAFNOSUPPORT) { 1141 ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n", 1142 if_name(ifp), error); 1143 return; 1144 } 1145 1146 ro.ro_prepend = linkhdr; 1147 ro.ro_plen = linkhdrsize; 1148 ro.ro_flags = 0; 1149 1150 m_clrprotoflags(m); /* Avoid confusing lower layers. */ 1151 (*ifp->if_output)(ifp, m, &sa, &ro); 1152 ARPSTAT_INC(txreplies); 1153 return; 1154 1155 drop: 1156 m_freem(m); 1157 } 1158 #endif 1159 1160 /* 1161 * Checks received arp data against existing @la. 1162 * Updates lle state/performs notification if necessary. 1163 */ 1164 static void 1165 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp, 1166 int bridged, struct llentry *la) 1167 { 1168 struct sockaddr sa; 1169 struct mbuf *m_hold, *m_hold_next; 1170 uint8_t linkhdr[LLE_MAX_LINKHDR]; 1171 size_t linkhdrsize; 1172 int lladdr_off; 1173 char addrbuf[INET_ADDRSTRLEN]; 1174 1175 LLE_WLOCK_ASSERT(la); 1176 1177 /* the following is not an error when doing bridging */ 1178 if (!bridged && la->lle_tbl->llt_ifp != ifp) { 1179 if (log_arp_wrong_iface) 1180 ARP_LOG(LOG_WARNING, "%s is on %s " 1181 "but got reply from %*D on %s\n", 1182 inet_ntoa_r(isaddr, addrbuf), 1183 la->lle_tbl->llt_ifp->if_xname, 1184 ifp->if_addrlen, (u_char *)ar_sha(ah), ":", 1185 ifp->if_xname); 1186 LLE_WUNLOCK(la); 1187 return; 1188 } 1189 if ((la->la_flags & LLE_VALID) && 1190 bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) { 1191 if (la->la_flags & LLE_STATIC) { 1192 LLE_WUNLOCK(la); 1193 if (log_arp_permanent_modify) 1194 ARP_LOG(LOG_ERR, 1195 "%*D attempts to modify " 1196 "permanent entry for %s on %s\n", 1197 ifp->if_addrlen, 1198 (u_char *)ar_sha(ah), ":", 1199 inet_ntoa_r(isaddr, addrbuf), 1200 ifp->if_xname); 1201 return; 1202 } 1203 if (log_arp_movements) { 1204 ARP_LOG(LOG_INFO, "%s moved from %*D " 1205 "to %*D on %s\n", 1206 inet_ntoa_r(isaddr, addrbuf), 1207 ifp->if_addrlen, 1208 (u_char *)la->ll_addr, ":", 1209 ifp->if_addrlen, (u_char *)ar_sha(ah), ":", 1210 ifp->if_xname); 1211 } 1212 } 1213 1214 /* Calculate full link prepend to use in lle */ 1215 linkhdrsize = sizeof(linkhdr); 1216 if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr, 1217 &linkhdrsize, &lladdr_off) != 0) 1218 return; 1219 1220 /* Check if something has changed */ 1221 if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 || 1222 (la->la_flags & LLE_VALID) == 0) { 1223 /* Try to perform LLE update */ 1224 if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize, 1225 lladdr_off) == 0) 1226 return; 1227 1228 /* Clear fast path feedback request if set */ 1229 la->r_skip_req = 0; 1230 } 1231 1232 arp_mark_lle_reachable(la); 1233 1234 /* 1235 * The packets are all freed within the call to the output 1236 * routine. 1237 * 1238 * NB: The lock MUST be released before the call to the 1239 * output routine. 1240 */ 1241 if (la->la_hold != NULL) { 1242 m_hold = la->la_hold; 1243 la->la_hold = NULL; 1244 la->la_numheld = 0; 1245 lltable_fill_sa_entry(la, &sa); 1246 LLE_WUNLOCK(la); 1247 for (; m_hold != NULL; m_hold = m_hold_next) { 1248 m_hold_next = m_hold->m_nextpkt; 1249 m_hold->m_nextpkt = NULL; 1250 /* Avoid confusing lower layers. */ 1251 m_clrprotoflags(m_hold); 1252 (*ifp->if_output)(ifp, m_hold, &sa, NULL); 1253 } 1254 } else 1255 LLE_WUNLOCK(la); 1256 } 1257 1258 static void 1259 arp_mark_lle_reachable(struct llentry *la) 1260 { 1261 int canceled, wtime; 1262 1263 LLE_WLOCK_ASSERT(la); 1264 1265 la->ln_state = ARP_LLINFO_REACHABLE; 1266 EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED); 1267 1268 if (!(la->la_flags & LLE_STATIC)) { 1269 LLE_ADDREF(la); 1270 la->la_expire = time_uptime + V_arpt_keep; 1271 wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit; 1272 if (wtime < 0) 1273 wtime = V_arpt_keep; 1274 canceled = callout_reset(&la->lle_timer, 1275 hz * wtime, arptimer, la); 1276 if (canceled) 1277 LLE_REMREF(la); 1278 } 1279 la->la_asked = 0; 1280 la->la_preempt = V_arp_maxtries; 1281 } 1282 1283 /* 1284 * Add permanent link-layer record for given interface address. 1285 */ 1286 static __noinline void 1287 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst) 1288 { 1289 struct llentry *lle, *lle_tmp; 1290 1291 /* 1292 * Interface address LLE record is considered static 1293 * because kernel code relies on LLE_STATIC flag to check 1294 * if these entries can be rewriten by arp updates. 1295 */ 1296 lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst); 1297 if (lle == NULL) { 1298 log(LOG_INFO, "arp_ifinit: cannot create arp " 1299 "entry for interface address\n"); 1300 return; 1301 } 1302 1303 IF_AFDATA_WLOCK(ifp); 1304 LLE_WLOCK(lle); 1305 /* Unlink any entry if exists */ 1306 lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst); 1307 if (lle_tmp != NULL) 1308 lltable_unlink_entry(LLTABLE(ifp), lle_tmp); 1309 1310 lltable_link_entry(LLTABLE(ifp), lle); 1311 IF_AFDATA_WUNLOCK(ifp); 1312 1313 if (lle_tmp != NULL) 1314 EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED); 1315 1316 EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED); 1317 LLE_WUNLOCK(lle); 1318 if (lle_tmp != NULL) 1319 lltable_free_entry(LLTABLE(ifp), lle_tmp); 1320 } 1321 1322 /* 1323 * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range 1324 * of valid values. 1325 */ 1326 static int 1327 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS) 1328 { 1329 int error; 1330 int rexmit_count = *(int *)arg1; 1331 1332 error = sysctl_handle_int(oidp, &rexmit_count, 0, req); 1333 1334 /* Enforce limits on any new value that may have been set. */ 1335 if (!error && req->newptr) { 1336 /* A new value was set. */ 1337 if (rexmit_count < 0) { 1338 rexmit_count = 0; 1339 } else if (rexmit_count > MAX_GARP_RETRANSMITS) { 1340 rexmit_count = MAX_GARP_RETRANSMITS; 1341 } 1342 *(int *)arg1 = rexmit_count; 1343 } 1344 1345 return (error); 1346 } 1347 1348 /* 1349 * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to 1350 * retransmit it again. A pending callout owns a reference to the ifa. 1351 */ 1352 static void 1353 garp_rexmit(void *arg) 1354 { 1355 struct in_ifaddr *ia = arg; 1356 1357 if (callout_pending(&ia->ia_garp_timer) || 1358 !callout_active(&ia->ia_garp_timer)) { 1359 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp); 1360 ifa_free(&ia->ia_ifa); 1361 return; 1362 } 1363 1364 CURVNET_SET(ia->ia_ifa.ifa_ifp->if_vnet); 1365 1366 /* 1367 * Drop lock while the ARP request is generated. 1368 */ 1369 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp); 1370 1371 arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr, 1372 &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp)); 1373 1374 /* 1375 * Increment the count of retransmissions. If the count has reached the 1376 * maximum value, stop sending the GARP packets. Otherwise, schedule 1377 * the callout to retransmit another GARP packet. 1378 */ 1379 ++ia->ia_garp_count; 1380 if (ia->ia_garp_count >= garp_rexmit_count) { 1381 ifa_free(&ia->ia_ifa); 1382 } else { 1383 int rescheduled; 1384 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp); 1385 rescheduled = callout_reset(&ia->ia_garp_timer, 1386 (1 << ia->ia_garp_count) * hz, 1387 garp_rexmit, ia); 1388 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp); 1389 if (rescheduled) { 1390 ifa_free(&ia->ia_ifa); 1391 } 1392 } 1393 1394 CURVNET_RESTORE(); 1395 } 1396 1397 /* 1398 * Start the GARP retransmit timer. 1399 * 1400 * A single GARP is always transmitted when an IPv4 address is added 1401 * to an interface and that is usually sufficient. However, in some 1402 * circumstances, such as when a shared address is passed between 1403 * cluster nodes, this single GARP may occasionally be dropped or 1404 * lost. This can lead to neighbors on the network link working with a 1405 * stale ARP cache and sending packets destined for that address to 1406 * the node that previously owned the address, which may not respond. 1407 * 1408 * To avoid this situation, GARP retransmits can be enabled by setting 1409 * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater 1410 * than zero. The setting represents the maximum number of 1411 * retransmissions. The interval between retransmissions is calculated 1412 * using an exponential backoff algorithm, doubling each time, so the 1413 * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds). 1414 */ 1415 static void 1416 garp_timer_start(struct ifaddr *ifa) 1417 { 1418 struct in_ifaddr *ia = (struct in_ifaddr *) ifa; 1419 1420 IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp); 1421 ia->ia_garp_count = 0; 1422 if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz, 1423 garp_rexmit, ia) == 0) { 1424 ifa_ref(ifa); 1425 } 1426 IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp); 1427 } 1428 1429 void 1430 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa) 1431 { 1432 const struct sockaddr_in *dst_in; 1433 const struct sockaddr *dst; 1434 1435 if (ifa->ifa_carp != NULL) 1436 return; 1437 1438 dst = ifa->ifa_addr; 1439 dst_in = (const struct sockaddr_in *)dst; 1440 1441 if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY) 1442 return; 1443 arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp)); 1444 if (garp_rexmit_count > 0) { 1445 garp_timer_start(ifa); 1446 } 1447 1448 arp_add_ifa_lle(ifp, dst); 1449 } 1450 1451 void 1452 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr) 1453 { 1454 1455 if (ntohl(addr.s_addr) != INADDR_ANY) 1456 arprequest(ifp, &addr, &addr, enaddr); 1457 } 1458 1459 /* 1460 * Sends gratuitous ARPs for each ifaddr to notify other 1461 * nodes about the address change. 1462 */ 1463 static __noinline void 1464 arp_handle_ifllchange(struct ifnet *ifp) 1465 { 1466 struct ifaddr *ifa; 1467 1468 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1469 if (ifa->ifa_addr->sa_family == AF_INET) 1470 arp_ifinit(ifp, ifa); 1471 } 1472 } 1473 1474 /* 1475 * A handler for interface link layer address change event. 1476 */ 1477 static void 1478 arp_iflladdr(void *arg __unused, struct ifnet *ifp) 1479 { 1480 1481 lltable_update_ifaddr(LLTABLE(ifp)); 1482 1483 if ((ifp->if_flags & IFF_UP) != 0) 1484 arp_handle_ifllchange(ifp); 1485 } 1486 1487 static void 1488 vnet_arp_init(void) 1489 { 1490 1491 if (IS_DEFAULT_VNET(curvnet)) { 1492 netisr_register(&arp_nh); 1493 iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event, 1494 arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY); 1495 } 1496 #ifdef VIMAGE 1497 else 1498 netisr_register_vnet(&arp_nh); 1499 #endif 1500 } 1501 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, 1502 vnet_arp_init, 0); 1503 1504 #ifdef VIMAGE 1505 /* 1506 * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH 1507 * lookups after destroying the hash. Ideally this would go on SI_ORDER_3.5. 1508 */ 1509 static void 1510 vnet_arp_destroy(__unused void *arg) 1511 { 1512 1513 netisr_unregister_vnet(&arp_nh); 1514 } 1515 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, 1516 vnet_arp_destroy, NULL); 1517 #endif 1518