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