1 /* 2 * Copyright (c) 1982, 1989, 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 * 3. 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_ethersubr.c 8.1 (Berkeley) 6/10/93 30 * $FreeBSD: src/sys/net/if_ethersubr.c,v 1.70.2.33 2003/04/28 15:45:53 archie Exp $ 31 */ 32 33 #include "opt_inet.h" 34 #include "opt_inet6.h" 35 #include "opt_mpls.h" 36 #include "opt_netgraph.h" 37 #include "opt_carp.h" 38 #include "opt_rss.h" 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/globaldata.h> 43 #include <sys/kernel.h> 44 #include <sys/ktr.h> 45 #include <sys/lock.h> 46 #include <sys/malloc.h> 47 #include <sys/mbuf.h> 48 #include <sys/msgport.h> 49 #include <sys/socket.h> 50 #include <sys/sockio.h> 51 #include <sys/sysctl.h> 52 #include <sys/thread.h> 53 54 #include <sys/thread2.h> 55 #include <sys/mplock2.h> 56 57 #include <net/if.h> 58 #include <net/netisr.h> 59 #include <net/route.h> 60 #include <net/if_llc.h> 61 #include <net/if_dl.h> 62 #include <net/if_types.h> 63 #include <net/ifq_var.h> 64 #include <net/bpf.h> 65 #include <net/ethernet.h> 66 #include <net/vlan/if_vlan_ether.h> 67 #include <net/vlan/if_vlan_var.h> 68 #include <net/netmsg2.h> 69 #include <net/netisr2.h> 70 71 #if defined(INET) || defined(INET6) 72 #include <netinet/in.h> 73 #include <netinet/ip_var.h> 74 #include <netinet/tcp_var.h> 75 #include <netinet/if_ether.h> 76 #include <netinet/ip_flow.h> 77 #include <net/ipfw/ip_fw.h> 78 #include <net/ipfw3/ip_fw.h> 79 #include <net/dummynet/ip_dummynet.h> 80 #endif 81 #ifdef INET6 82 #include <netinet6/nd6.h> 83 #endif 84 85 #ifdef CARP 86 #include <netinet/ip_carp.h> 87 #endif 88 89 #ifdef MPLS 90 #include <netproto/mpls/mpls.h> 91 #endif 92 93 /* netgraph node hooks for ng_ether(4) */ 94 void (*ng_ether_input_p)(struct ifnet *ifp, struct mbuf **mp); 95 void (*ng_ether_input_orphan_p)(struct ifnet *ifp, struct mbuf *m); 96 int (*ng_ether_output_p)(struct ifnet *ifp, struct mbuf **mp); 97 void (*ng_ether_attach_p)(struct ifnet *ifp); 98 void (*ng_ether_detach_p)(struct ifnet *ifp); 99 100 void (*vlan_input_p)(struct mbuf *); 101 102 static int ether_output(struct ifnet *, struct mbuf *, struct sockaddr *, 103 struct rtentry *); 104 static void ether_restore_header(struct mbuf **, const struct ether_header *, 105 const struct ether_header *); 106 static int ether_characterize(struct mbuf **); 107 static void ether_dispatch(int, struct mbuf *, int); 108 109 /* 110 * if_bridge support 111 */ 112 struct mbuf *(*bridge_input_p)(struct ifnet *, struct mbuf *); 113 int (*bridge_output_p)(struct ifnet *, struct mbuf *); 114 void (*bridge_dn_p)(struct mbuf *, struct ifnet *); 115 struct ifnet *(*bridge_interface_p)(void *if_bridge); 116 117 static int ether_resolvemulti(struct ifnet *, struct sockaddr **, 118 struct sockaddr *); 119 120 /* 121 * if_lagg(4) support 122 */ 123 void (*lagg_input_p)(struct ifnet *, struct mbuf *); 124 int (*lagg_output_p)(struct ifnet *, struct mbuf *); 125 126 const uint8_t etherbroadcastaddr[ETHER_ADDR_LEN] = { 127 0xff, 0xff, 0xff, 0xff, 0xff, 0xff 128 }; 129 130 #define gotoerr(e) do { error = (e); goto bad; } while (0) 131 #define IFP2AC(ifp) ((struct arpcom *)(ifp)) 132 133 static boolean_t ether_ipfw_chk(struct mbuf **m0, struct ifnet *dst, 134 struct ip_fw **rule, 135 const struct ether_header *eh); 136 137 static int ether_ipfw; 138 static u_long ether_restore_hdr; 139 static u_long ether_prepend_hdr; 140 static u_long ether_input_wronghash; 141 static int ether_debug; 142 143 #ifdef RSS_DEBUG 144 static u_long ether_pktinfo_try; 145 static u_long ether_pktinfo_hit; 146 static u_long ether_rss_nopi; 147 static u_long ether_rss_nohash; 148 static u_long ether_input_requeue; 149 #endif 150 static u_long ether_input_wronghwhash; 151 static int ether_input_ckhash; 152 153 #define ETHER_TSOLEN_DEFAULT (4 * ETHERMTU) 154 155 #define ETHER_NMBCLUSTERS_DEFMIN 32 156 #define ETHER_NMBCLUSTERS_DEFAULT 256 157 158 static int ether_tsolen_default = ETHER_TSOLEN_DEFAULT; 159 TUNABLE_INT("net.link.ether.tsolen", ðer_tsolen_default); 160 161 static int ether_nmbclusters_default = ETHER_NMBCLUSTERS_DEFAULT; 162 TUNABLE_INT("net.link.ether.nmbclusters", ðer_nmbclusters_default); 163 164 SYSCTL_DECL(_net_link); 165 SYSCTL_NODE(_net_link, IFT_ETHER, ether, CTLFLAG_RW, 0, "Ethernet"); 166 SYSCTL_INT(_net_link_ether, OID_AUTO, debug, CTLFLAG_RW, 167 ðer_debug, 0, "Ether debug"); 168 SYSCTL_INT(_net_link_ether, OID_AUTO, ipfw, CTLFLAG_RW, 169 ðer_ipfw, 0, "Pass ether pkts through firewall"); 170 SYSCTL_ULONG(_net_link_ether, OID_AUTO, restore_hdr, CTLFLAG_RW, 171 ðer_restore_hdr, 0, "# of ether header restoration"); 172 SYSCTL_ULONG(_net_link_ether, OID_AUTO, prepend_hdr, CTLFLAG_RW, 173 ðer_prepend_hdr, 0, 174 "# of ether header restoration which prepends mbuf"); 175 SYSCTL_ULONG(_net_link_ether, OID_AUTO, input_wronghash, CTLFLAG_RW, 176 ðer_input_wronghash, 0, "# of input packets with wrong hash"); 177 SYSCTL_INT(_net_link_ether, OID_AUTO, tsolen, CTLFLAG_RW, 178 ðer_tsolen_default, 0, "Default max TSO length"); 179 180 #ifdef RSS_DEBUG 181 SYSCTL_ULONG(_net_link_ether, OID_AUTO, rss_nopi, CTLFLAG_RW, 182 ðer_rss_nopi, 0, "# of packets do not have pktinfo"); 183 SYSCTL_ULONG(_net_link_ether, OID_AUTO, rss_nohash, CTLFLAG_RW, 184 ðer_rss_nohash, 0, "# of packets do not have hash"); 185 SYSCTL_ULONG(_net_link_ether, OID_AUTO, pktinfo_try, CTLFLAG_RW, 186 ðer_pktinfo_try, 0, 187 "# of tries to find packets' msgport using pktinfo"); 188 SYSCTL_ULONG(_net_link_ether, OID_AUTO, pktinfo_hit, CTLFLAG_RW, 189 ðer_pktinfo_hit, 0, 190 "# of packets whose msgport are found using pktinfo"); 191 SYSCTL_ULONG(_net_link_ether, OID_AUTO, input_requeue, CTLFLAG_RW, 192 ðer_input_requeue, 0, "# of input packets gets requeued"); 193 #endif 194 SYSCTL_ULONG(_net_link_ether, OID_AUTO, input_wronghwhash, CTLFLAG_RW, 195 ðer_input_wronghwhash, 0, "# of input packets with wrong hw hash"); 196 SYSCTL_INT(_net_link_ether, OID_AUTO, always_ckhash, CTLFLAG_RW, 197 ðer_input_ckhash, 0, "always check hash"); 198 199 #define ETHER_KTR_STR "ifp=%p" 200 #define ETHER_KTR_ARGS struct ifnet *ifp 201 #ifndef KTR_ETHERNET 202 #define KTR_ETHERNET KTR_ALL 203 #endif 204 KTR_INFO_MASTER(ether); 205 KTR_INFO(KTR_ETHERNET, ether, pkt_beg, 0, ETHER_KTR_STR, ETHER_KTR_ARGS); 206 KTR_INFO(KTR_ETHERNET, ether, pkt_end, 1, ETHER_KTR_STR, ETHER_KTR_ARGS); 207 KTR_INFO(KTR_ETHERNET, ether, disp_beg, 2, ETHER_KTR_STR, ETHER_KTR_ARGS); 208 KTR_INFO(KTR_ETHERNET, ether, disp_end, 3, ETHER_KTR_STR, ETHER_KTR_ARGS); 209 #define logether(name, arg) KTR_LOG(ether_ ## name, arg) 210 211 /* 212 * Ethernet output routine. 213 * Encapsulate a packet of type family for the local net. 214 * Use trailer local net encapsulation if enough data in first 215 * packet leaves a multiple of 512 bytes of data in remainder. 216 * Assumes that ifp is actually pointer to arpcom structure. 217 */ 218 static int 219 ether_output(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst, 220 struct rtentry *rt) 221 { 222 struct ether_header *eh, *deh; 223 u_char *edst; 224 int loop_copy = 0; 225 int hlen = ETHER_HDR_LEN; /* link layer header length */ 226 struct arpcom *ac = IFP2AC(ifp); 227 int error; 228 229 ASSERT_NETISR_NCPUS(mycpuid); 230 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp); 231 232 if (ifp->if_flags & IFF_MONITOR) 233 gotoerr(ENETDOWN); 234 if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) != (IFF_UP | IFF_RUNNING)) 235 gotoerr(ENETDOWN); 236 237 M_PREPEND(m, sizeof(struct ether_header), M_NOWAIT); 238 if (m == NULL) 239 return (ENOBUFS); 240 m->m_pkthdr.csum_lhlen = sizeof(struct ether_header); 241 eh = mtod(m, struct ether_header *); 242 edst = eh->ether_dhost; 243 244 /* 245 * Fill in the destination ethernet address and frame type. 246 */ 247 switch (dst->sa_family) { 248 #ifdef INET 249 case AF_INET: 250 if (!arpresolve(ifp, rt, m, dst, edst)) 251 return (0); /* if not yet resolved */ 252 #ifdef MPLS 253 if (m->m_flags & M_MPLSLABELED) 254 eh->ether_type = htons(ETHERTYPE_MPLS); 255 else 256 #endif 257 eh->ether_type = htons(ETHERTYPE_IP); 258 break; 259 #endif 260 #ifdef INET6 261 case AF_INET6: 262 if (!nd6_storelladdr(&ac->ac_if, rt, m, dst, edst)) 263 return (0); /* Something bad happenned. */ 264 eh->ether_type = htons(ETHERTYPE_IPV6); 265 break; 266 #endif 267 case pseudo_AF_HDRCMPLT: 268 case AF_UNSPEC: 269 loop_copy = -1; /* if this is for us, don't do it */ 270 deh = (struct ether_header *)dst->sa_data; 271 memcpy(edst, deh->ether_dhost, ETHER_ADDR_LEN); 272 eh->ether_type = deh->ether_type; 273 break; 274 275 default: 276 if_printf(ifp, "can't handle af%d\n", dst->sa_family); 277 gotoerr(EAFNOSUPPORT); 278 } 279 280 if (dst->sa_family == pseudo_AF_HDRCMPLT) /* unlikely */ 281 memcpy(eh->ether_shost, 282 ((struct ether_header *)dst->sa_data)->ether_shost, 283 ETHER_ADDR_LEN); 284 else 285 memcpy(eh->ether_shost, ac->ac_enaddr, ETHER_ADDR_LEN); 286 287 /* 288 * Bridges require special output handling. 289 */ 290 if (ifp->if_bridge) { 291 KASSERT(bridge_output_p != NULL, 292 ("%s: if_bridge not loaded!", __func__)); 293 return bridge_output_p(ifp, m); 294 } 295 #if 0 /* XXX */ 296 if (ifp->if_lagg) { 297 KASSERT(lagg_output_p != NULL, 298 ("%s: if_lagg not loaded!", __func__)); 299 return lagg_output_p(ifp, m); 300 } 301 #endif 302 303 /* 304 * If a simplex interface, and the packet is being sent to our 305 * Ethernet address or a broadcast address, loopback a copy. 306 * XXX To make a simplex device behave exactly like a duplex 307 * device, we should copy in the case of sending to our own 308 * ethernet address (thus letting the original actually appear 309 * on the wire). However, we don't do that here for security 310 * reasons and compatibility with the original behavior. 311 */ 312 if ((ifp->if_flags & IFF_SIMPLEX) && (loop_copy != -1)) { 313 int csum_flags = 0; 314 315 if (m->m_pkthdr.csum_flags & CSUM_IP) 316 csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID); 317 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) 318 csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 319 if ((m->m_flags & M_BCAST) || (loop_copy > 0)) { 320 struct mbuf *n; 321 322 if ((n = m_copypacket(m, M_NOWAIT)) != NULL) { 323 n->m_pkthdr.csum_flags |= csum_flags; 324 if (csum_flags & CSUM_DATA_VALID) 325 n->m_pkthdr.csum_data = 0xffff; 326 if_simloop(ifp, n, dst->sa_family, hlen); 327 } else 328 IFNET_STAT_INC(ifp, iqdrops, 1); 329 } else if (bcmp(eh->ether_dhost, eh->ether_shost, 330 ETHER_ADDR_LEN) == 0) { 331 m->m_pkthdr.csum_flags |= csum_flags; 332 if (csum_flags & CSUM_DATA_VALID) 333 m->m_pkthdr.csum_data = 0xffff; 334 if_simloop(ifp, m, dst->sa_family, hlen); 335 return (0); /* XXX */ 336 } 337 } 338 339 #ifdef CARP 340 if (ifp->if_type == IFT_CARP) { 341 ifp = carp_parent(ifp); 342 if (ifp == NULL) 343 gotoerr(ENETUNREACH); 344 345 ac = IFP2AC(ifp); 346 347 /* 348 * Check precondition again 349 */ 350 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp); 351 352 if (ifp->if_flags & IFF_MONITOR) 353 gotoerr(ENETDOWN); 354 if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) != 355 (IFF_UP | IFF_RUNNING)) 356 gotoerr(ENETDOWN); 357 } 358 #endif 359 360 /* Handle ng_ether(4) processing, if any */ 361 if (ng_ether_output_p != NULL) { 362 /* 363 * Hold BGL and recheck ng_ether_output_p 364 */ 365 get_mplock(); 366 if (ng_ether_output_p != NULL) { 367 if ((error = ng_ether_output_p(ifp, &m)) != 0) { 368 rel_mplock(); 369 goto bad; 370 } 371 if (m == NULL) { 372 rel_mplock(); 373 return (0); 374 } 375 } 376 rel_mplock(); 377 } 378 379 /* Continue with link-layer output */ 380 return ether_output_frame(ifp, m); 381 382 bad: 383 m_freem(m); 384 return (error); 385 } 386 387 /* 388 * Returns the bridge interface an ifp is associated 389 * with. 390 * 391 * Only call if ifp->if_bridge != NULL. 392 */ 393 struct ifnet * 394 ether_bridge_interface(struct ifnet *ifp) 395 { 396 if (bridge_interface_p) 397 return(bridge_interface_p(ifp->if_bridge)); 398 return (ifp); 399 } 400 401 /* 402 * Ethernet link layer output routine to send a raw frame to the device. 403 * 404 * This assumes that the 14 byte Ethernet header is present and contiguous 405 * in the first mbuf. 406 */ 407 int 408 ether_output_frame(struct ifnet *ifp, struct mbuf *m) 409 { 410 struct ip_fw *rule = NULL; 411 int error = 0; 412 struct altq_pktattr pktattr; 413 414 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp); 415 416 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) { 417 struct m_tag *mtag; 418 419 /* Extract info from dummynet tag */ 420 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL); 421 KKASSERT(mtag != NULL); 422 rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv; 423 KKASSERT(rule != NULL); 424 425 m_tag_delete(m, mtag); 426 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED; 427 } 428 429 if (ifq_is_enabled(&ifp->if_snd)) 430 altq_etherclassify(&ifp->if_snd, m, &pktattr); 431 crit_enter(); 432 if ((IPFW_LOADED || IPFW3_LOADED) && ether_ipfw != 0) { 433 struct ether_header save_eh, *eh; 434 435 eh = mtod(m, struct ether_header *); 436 save_eh = *eh; 437 m_adj(m, ETHER_HDR_LEN); 438 if (!ether_ipfw_chk(&m, ifp, &rule, eh)) { 439 crit_exit(); 440 if (m != NULL) { 441 m_freem(m); 442 return ENOBUFS; /* pkt dropped */ 443 } else 444 return 0; /* consumed e.g. in a pipe */ 445 } 446 447 /* packet was ok, restore the ethernet header */ 448 ether_restore_header(&m, eh, &save_eh); 449 if (m == NULL) { 450 crit_exit(); 451 return ENOBUFS; 452 } 453 } 454 crit_exit(); 455 456 /* 457 * Queue message on interface, update output statistics if 458 * successful, and start output if interface not yet active. 459 */ 460 error = ifq_dispatch(ifp, m, &pktattr); 461 return (error); 462 } 463 464 /* 465 * ipfw processing for ethernet packets (in and out). 466 * The second parameter is NULL from ether_demux(), and ifp from 467 * ether_output_frame(). 468 */ 469 static boolean_t 470 ether_ipfw_chk(struct mbuf **m0, struct ifnet *dst, struct ip_fw **rule, 471 const struct ether_header *eh) 472 { 473 struct ether_header save_eh = *eh; /* might be a ptr in *m0 */ 474 struct ip_fw_args args; 475 struct m_tag *mtag; 476 struct mbuf *m; 477 int i; 478 479 if (*rule != NULL && fw_one_pass) 480 return TRUE; /* dummynet packet, already partially processed */ 481 482 /* 483 * I need some amount of data to be contiguous. 484 */ 485 i = min((*m0)->m_pkthdr.len, max_protohdr); 486 if ((*m0)->m_len < i) { 487 *m0 = m_pullup(*m0, i); 488 if (*m0 == NULL) 489 return FALSE; 490 } 491 492 /* 493 * Clean up tags 494 */ 495 if ((mtag = m_tag_find(*m0, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL) 496 m_tag_delete(*m0, mtag); 497 if ((*m0)->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) { 498 mtag = m_tag_find(*m0, PACKET_TAG_IPFORWARD, NULL); 499 KKASSERT(mtag != NULL); 500 m_tag_delete(*m0, mtag); 501 (*m0)->m_pkthdr.fw_flags &= ~IPFORWARD_MBUF_TAGGED; 502 } 503 504 args.m = *m0; /* the packet we are looking at */ 505 args.oif = dst; /* destination, if any */ 506 args.rule = *rule; /* matching rule to restart */ 507 args.eh = &save_eh; /* MAC header for bridged/MAC packets */ 508 i = ip_fw_chk_ptr(&args); 509 *m0 = args.m; 510 *rule = args.rule; 511 512 if (*m0 == NULL) 513 return FALSE; 514 515 switch (i) { 516 case IP_FW_PASS: 517 return TRUE; 518 519 case IP_FW_DIVERT: 520 case IP_FW_TEE: 521 case IP_FW_DENY: 522 /* 523 * XXX at some point add support for divert/forward actions. 524 * If none of the above matches, we have to drop the pkt. 525 */ 526 return FALSE; 527 528 case IP_FW_DUMMYNET: 529 /* 530 * Pass the pkt to dummynet, which consumes it. 531 */ 532 m = *m0; /* pass the original to dummynet */ 533 *m0 = NULL; /* and nothing back to the caller */ 534 535 ether_restore_header(&m, eh, &save_eh); 536 if (m == NULL) 537 return FALSE; 538 539 ip_fw_dn_io_ptr(m, args.cookie, 540 dst ? DN_TO_ETH_OUT: DN_TO_ETH_DEMUX, &args); 541 ip_dn_queue(m); 542 return FALSE; 543 544 default: 545 panic("unknown ipfw return value: %d", i); 546 } 547 } 548 549 /* 550 * Perform common duties while attaching to interface list 551 */ 552 void 553 ether_ifattach(struct ifnet *ifp, const uint8_t *lla, 554 lwkt_serialize_t serializer) 555 { 556 ether_ifattach_bpf(ifp, lla, DLT_EN10MB, sizeof(struct ether_header), 557 serializer); 558 } 559 560 void 561 ether_ifattach_bpf(struct ifnet *ifp, const uint8_t *lla, 562 u_int dlt, u_int hdrlen, lwkt_serialize_t serializer) 563 { 564 struct sockaddr_dl *sdl; 565 char ethstr[ETHER_ADDRSTRLEN + 1]; 566 struct ifaltq *ifq; 567 int i; 568 569 /* 570 * If driver does not configure # of mbuf clusters/jclusters 571 * that could sit on the device queues for quite some time, 572 * we then assume: 573 * - The device queues only consume mbuf clusters. 574 * - No more than ether_nmbclusters_default (by default 256) 575 * mbuf clusters will sit on the device queues for quite 576 * some time. 577 */ 578 if (ifp->if_nmbclusters <= 0 && ifp->if_nmbjclusters <= 0) { 579 if (ether_nmbclusters_default < ETHER_NMBCLUSTERS_DEFMIN) { 580 kprintf("ether nmbclusters %d -> %d\n", 581 ether_nmbclusters_default, 582 ETHER_NMBCLUSTERS_DEFAULT); 583 ether_nmbclusters_default = ETHER_NMBCLUSTERS_DEFAULT; 584 } 585 ifp->if_nmbclusters = ether_nmbclusters_default; 586 } 587 588 ifp->if_type = IFT_ETHER; 589 ifp->if_addrlen = ETHER_ADDR_LEN; 590 ifp->if_hdrlen = ETHER_HDR_LEN; 591 if_attach(ifp, serializer); 592 ifq = &ifp->if_snd; 593 for (i = 0; i < ifq->altq_subq_cnt; ++i) { 594 struct ifaltq_subque *ifsq = ifq_get_subq(ifq, i); 595 596 ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * 597 (ETHER_MAX_LEN - ETHER_CRC_LEN); 598 } 599 ifp->if_mtu = ETHERMTU; 600 if (ifp->if_tsolen <= 0) { 601 if ((ether_tsolen_default / ETHERMTU) < 2) { 602 kprintf("ether TSO maxlen %d -> %d\n", 603 ether_tsolen_default, ETHER_TSOLEN_DEFAULT); 604 ether_tsolen_default = ETHER_TSOLEN_DEFAULT; 605 } 606 ifp->if_tsolen = ether_tsolen_default; 607 } 608 if (ifp->if_baudrate == 0) 609 ifp->if_baudrate = 10000000; 610 ifp->if_output = ether_output; 611 ifp->if_input = ether_input; 612 ifp->if_resolvemulti = ether_resolvemulti; 613 ifp->if_broadcastaddr = etherbroadcastaddr; 614 sdl = IF_LLSOCKADDR(ifp); 615 sdl->sdl_type = IFT_ETHER; 616 sdl->sdl_alen = ifp->if_addrlen; 617 bcopy(lla, LLADDR(sdl), ifp->if_addrlen); 618 /* 619 * XXX Keep the current drivers happy. 620 * XXX Remove once all drivers have been cleaned up 621 */ 622 if (lla != IFP2AC(ifp)->ac_enaddr) 623 bcopy(lla, IFP2AC(ifp)->ac_enaddr, ifp->if_addrlen); 624 bpfattach(ifp, dlt, hdrlen); 625 if (ng_ether_attach_p != NULL) 626 (*ng_ether_attach_p)(ifp); 627 628 if_printf(ifp, "MAC address: %s\n", kether_ntoa(lla, ethstr)); 629 } 630 631 /* 632 * Perform common duties while detaching an Ethernet interface 633 */ 634 void 635 ether_ifdetach(struct ifnet *ifp) 636 { 637 if_down(ifp); 638 639 if (ng_ether_detach_p != NULL) 640 (*ng_ether_detach_p)(ifp); 641 bpfdetach(ifp); 642 if_detach(ifp); 643 } 644 645 int 646 ether_ioctl(struct ifnet *ifp, u_long command, caddr_t data) 647 { 648 struct ifaddr *ifa = (struct ifaddr *) data; 649 struct ifreq *ifr = (struct ifreq *) data; 650 int error = 0; 651 652 #define IF_INIT(ifp) \ 653 do { \ 654 if (((ifp)->if_flags & IFF_UP) == 0) { \ 655 (ifp)->if_flags |= IFF_UP; \ 656 (ifp)->if_init((ifp)->if_softc); \ 657 } \ 658 } while (0) 659 660 ASSERT_IFNET_SERIALIZED_ALL(ifp); 661 662 switch (command) { 663 case SIOCSIFADDR: 664 switch (ifa->ifa_addr->sa_family) { 665 #ifdef INET 666 case AF_INET: 667 IF_INIT(ifp); /* before arpwhohas */ 668 arp_ifinit(ifp, ifa); 669 break; 670 #endif 671 default: 672 IF_INIT(ifp); 673 break; 674 } 675 break; 676 677 case SIOCGIFADDR: 678 bcopy(IFP2AC(ifp)->ac_enaddr, 679 ((struct sockaddr *)ifr->ifr_data)->sa_data, 680 ETHER_ADDR_LEN); 681 break; 682 683 case SIOCSIFMTU: 684 /* 685 * Set the interface MTU. 686 */ 687 if (ifr->ifr_mtu > ETHERMTU) { 688 error = EINVAL; 689 } else { 690 ifp->if_mtu = ifr->ifr_mtu; 691 } 692 break; 693 default: 694 error = EINVAL; 695 break; 696 } 697 return (error); 698 699 #undef IF_INIT 700 } 701 702 static int 703 ether_resolvemulti( 704 struct ifnet *ifp, 705 struct sockaddr **llsa, 706 struct sockaddr *sa) 707 { 708 struct sockaddr_dl *sdl; 709 #ifdef INET 710 struct sockaddr_in *sin; 711 #endif 712 #ifdef INET6 713 struct sockaddr_in6 *sin6; 714 #endif 715 u_char *e_addr; 716 717 switch(sa->sa_family) { 718 case AF_LINK: 719 /* 720 * No mapping needed. Just check that it's a valid MC address. 721 */ 722 sdl = (struct sockaddr_dl *)sa; 723 e_addr = LLADDR(sdl); 724 if ((e_addr[0] & 1) != 1) 725 return EADDRNOTAVAIL; 726 *llsa = NULL; 727 return 0; 728 729 #ifdef INET 730 case AF_INET: 731 sin = (struct sockaddr_in *)sa; 732 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) 733 return EADDRNOTAVAIL; 734 sdl = kmalloc(sizeof *sdl, M_IFMADDR, M_WAITOK | M_ZERO); 735 sdl->sdl_len = sizeof *sdl; 736 sdl->sdl_family = AF_LINK; 737 sdl->sdl_index = ifp->if_index; 738 sdl->sdl_type = IFT_ETHER; 739 sdl->sdl_alen = ETHER_ADDR_LEN; 740 e_addr = LLADDR(sdl); 741 ETHER_MAP_IP_MULTICAST(&sin->sin_addr, e_addr); 742 *llsa = (struct sockaddr *)sdl; 743 return 0; 744 #endif 745 #ifdef INET6 746 case AF_INET6: 747 sin6 = (struct sockaddr_in6 *)sa; 748 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 749 /* 750 * An IP6 address of 0 means listen to all 751 * of the Ethernet multicast address used for IP6. 752 * (This is used for multicast routers.) 753 */ 754 ifp->if_flags |= IFF_ALLMULTI; 755 *llsa = NULL; 756 return 0; 757 } 758 if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 759 return EADDRNOTAVAIL; 760 sdl = kmalloc(sizeof *sdl, M_IFMADDR, M_WAITOK | M_ZERO); 761 sdl->sdl_len = sizeof *sdl; 762 sdl->sdl_family = AF_LINK; 763 sdl->sdl_index = ifp->if_index; 764 sdl->sdl_type = IFT_ETHER; 765 sdl->sdl_alen = ETHER_ADDR_LEN; 766 e_addr = LLADDR(sdl); 767 ETHER_MAP_IPV6_MULTICAST(&sin6->sin6_addr, e_addr); 768 *llsa = (struct sockaddr *)sdl; 769 return 0; 770 #endif 771 772 default: 773 /* 774 * Well, the text isn't quite right, but it's the name 775 * that counts... 776 */ 777 return EAFNOSUPPORT; 778 } 779 } 780 781 #if 0 782 /* 783 * This is for reference. We have a table-driven version 784 * of the little-endian crc32 generator, which is faster 785 * than the double-loop. 786 */ 787 uint32_t 788 ether_crc32_le(const uint8_t *buf, size_t len) 789 { 790 uint32_t c, crc, carry; 791 size_t i, j; 792 793 crc = 0xffffffffU; /* initial value */ 794 795 for (i = 0; i < len; i++) { 796 c = buf[i]; 797 for (j = 0; j < 8; j++) { 798 carry = ((crc & 0x01) ? 1 : 0) ^ (c & 0x01); 799 crc >>= 1; 800 c >>= 1; 801 if (carry) 802 crc = (crc ^ ETHER_CRC_POLY_LE); 803 } 804 } 805 806 return (crc); 807 } 808 #else 809 uint32_t 810 ether_crc32_le(const uint8_t *buf, size_t len) 811 { 812 static const uint32_t crctab[] = { 813 0x00000000, 0x1db71064, 0x3b6e20c8, 0x26d930ac, 814 0x76dc4190, 0x6b6b51f4, 0x4db26158, 0x5005713c, 815 0xedb88320, 0xf00f9344, 0xd6d6a3e8, 0xcb61b38c, 816 0x9b64c2b0, 0x86d3d2d4, 0xa00ae278, 0xbdbdf21c 817 }; 818 uint32_t crc; 819 size_t i; 820 821 crc = 0xffffffffU; /* initial value */ 822 823 for (i = 0; i < len; i++) { 824 crc ^= buf[i]; 825 crc = (crc >> 4) ^ crctab[crc & 0xf]; 826 crc = (crc >> 4) ^ crctab[crc & 0xf]; 827 } 828 829 return (crc); 830 } 831 #endif 832 833 uint32_t 834 ether_crc32_be(const uint8_t *buf, size_t len) 835 { 836 uint32_t c, crc, carry; 837 size_t i, j; 838 839 crc = 0xffffffffU; /* initial value */ 840 841 for (i = 0; i < len; i++) { 842 c = buf[i]; 843 for (j = 0; j < 8; j++) { 844 carry = ((crc & 0x80000000U) ? 1 : 0) ^ (c & 0x01); 845 crc <<= 1; 846 c >>= 1; 847 if (carry) 848 crc = (crc ^ ETHER_CRC_POLY_BE) | carry; 849 } 850 } 851 852 return (crc); 853 } 854 855 /* 856 * find the size of ethernet header, and call classifier 857 */ 858 void 859 altq_etherclassify(struct ifaltq *ifq, struct mbuf *m, 860 struct altq_pktattr *pktattr) 861 { 862 struct ether_header *eh; 863 uint16_t ether_type; 864 int hlen, af, hdrsize; 865 866 hlen = sizeof(struct ether_header); 867 eh = mtod(m, struct ether_header *); 868 869 ether_type = ntohs(eh->ether_type); 870 if (ether_type < ETHERMTU) { 871 /* ick! LLC/SNAP */ 872 struct llc *llc = (struct llc *)(eh + 1); 873 hlen += 8; 874 875 if (m->m_len < hlen || 876 llc->llc_dsap != LLC_SNAP_LSAP || 877 llc->llc_ssap != LLC_SNAP_LSAP || 878 llc->llc_control != LLC_UI) 879 goto bad; /* not snap! */ 880 881 ether_type = ntohs(llc->llc_un.type_snap.ether_type); 882 } 883 884 if (ether_type == ETHERTYPE_IP) { 885 af = AF_INET; 886 hdrsize = 20; /* sizeof(struct ip) */ 887 #ifdef INET6 888 } else if (ether_type == ETHERTYPE_IPV6) { 889 af = AF_INET6; 890 hdrsize = 40; /* sizeof(struct ip6_hdr) */ 891 #endif 892 } else 893 goto bad; 894 895 while (m->m_len <= hlen) { 896 hlen -= m->m_len; 897 m = m->m_next; 898 } 899 if (m->m_len < hlen + hdrsize) { 900 /* 901 * ip header is not in a single mbuf. this should not 902 * happen in the current code. 903 * (todo: use m_pulldown in the future) 904 */ 905 goto bad; 906 } 907 m->m_data += hlen; 908 m->m_len -= hlen; 909 ifq_classify(ifq, m, af, pktattr); 910 m->m_data -= hlen; 911 m->m_len += hlen; 912 913 return; 914 915 bad: 916 pktattr->pattr_class = NULL; 917 pktattr->pattr_hdr = NULL; 918 pktattr->pattr_af = AF_UNSPEC; 919 } 920 921 static void 922 ether_restore_header(struct mbuf **m0, const struct ether_header *eh, 923 const struct ether_header *save_eh) 924 { 925 struct mbuf *m = *m0; 926 927 ether_restore_hdr++; 928 929 /* 930 * Prepend the header, optimize for the common case of 931 * eh pointing into the mbuf. 932 */ 933 if ((const void *)(eh + 1) == (void *)m->m_data) { 934 m->m_data -= ETHER_HDR_LEN; 935 m->m_len += ETHER_HDR_LEN; 936 m->m_pkthdr.len += ETHER_HDR_LEN; 937 } else { 938 ether_prepend_hdr++; 939 940 M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT); 941 if (m != NULL) { 942 bcopy(save_eh, mtod(m, struct ether_header *), 943 ETHER_HDR_LEN); 944 } 945 } 946 *m0 = m; 947 } 948 949 /* 950 * Upper layer processing for a received Ethernet packet. 951 */ 952 void 953 ether_demux_oncpu(struct ifnet *ifp, struct mbuf *m) 954 { 955 struct ether_header *eh; 956 int isr, discard = 0; 957 u_short ether_type; 958 struct ip_fw *rule = NULL; 959 960 M_ASSERTPKTHDR(m); 961 KASSERT(m->m_len >= ETHER_HDR_LEN, 962 ("ether header is not contiguous!")); 963 964 eh = mtod(m, struct ether_header *); 965 966 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) { 967 struct m_tag *mtag; 968 969 /* Extract info from dummynet tag */ 970 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL); 971 KKASSERT(mtag != NULL); 972 rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv; 973 KKASSERT(rule != NULL); 974 975 m_tag_delete(m, mtag); 976 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED; 977 978 /* packet is passing the second time */ 979 goto post_stats; 980 } 981 982 /* 983 * We got a packet which was unicast to a different Ethernet 984 * address. If the driver is working properly, then this 985 * situation can only happen when the interface is in 986 * promiscuous mode. We defer the packet discarding until the 987 * vlan processing is done, so that vlan/bridge or vlan/netgraph 988 * could work. 989 */ 990 if (((ifp->if_flags & (IFF_PROMISC | IFF_PPROMISC)) == IFF_PROMISC) && 991 !ETHER_IS_MULTICAST(eh->ether_dhost) && 992 bcmp(eh->ether_dhost, IFP2AC(ifp)->ac_enaddr, ETHER_ADDR_LEN)) { 993 if (ether_debug & 1) { 994 kprintf("%02x:%02x:%02x:%02x:%02x:%02x " 995 "%02x:%02x:%02x:%02x:%02x:%02x " 996 "%04x vs %02x:%02x:%02x:%02x:%02x:%02x\n", 997 eh->ether_dhost[0], 998 eh->ether_dhost[1], 999 eh->ether_dhost[2], 1000 eh->ether_dhost[3], 1001 eh->ether_dhost[4], 1002 eh->ether_dhost[5], 1003 eh->ether_shost[0], 1004 eh->ether_shost[1], 1005 eh->ether_shost[2], 1006 eh->ether_shost[3], 1007 eh->ether_shost[4], 1008 eh->ether_shost[5], 1009 eh->ether_type, 1010 ((u_char *)IFP2AC(ifp)->ac_enaddr)[0], 1011 ((u_char *)IFP2AC(ifp)->ac_enaddr)[1], 1012 ((u_char *)IFP2AC(ifp)->ac_enaddr)[2], 1013 ((u_char *)IFP2AC(ifp)->ac_enaddr)[3], 1014 ((u_char *)IFP2AC(ifp)->ac_enaddr)[4], 1015 ((u_char *)IFP2AC(ifp)->ac_enaddr)[5] 1016 ); 1017 } 1018 if ((ether_debug & 2) == 0) 1019 discard = 1; 1020 } 1021 1022 post_stats: 1023 if ((IPFW_LOADED || IPFW3_LOADED) && ether_ipfw != 0 && !discard) { 1024 struct ether_header save_eh = *eh; 1025 1026 /* XXX old crufty stuff, needs to be removed */ 1027 m_adj(m, sizeof(struct ether_header)); 1028 1029 if (!ether_ipfw_chk(&m, NULL, &rule, eh)) { 1030 m_freem(m); 1031 return; 1032 } 1033 1034 ether_restore_header(&m, eh, &save_eh); 1035 if (m == NULL) 1036 return; 1037 eh = mtod(m, struct ether_header *); 1038 } 1039 1040 ether_type = ntohs(eh->ether_type); 1041 KKASSERT(ether_type != ETHERTYPE_VLAN); 1042 1043 /* Handle input from a lagg(4) port */ 1044 if (ifp->if_type == IFT_IEEE8023ADLAG) { 1045 KASSERT(lagg_input_p != NULL, 1046 ("%s: if_lagg not loaded!", __func__)); 1047 (*lagg_input_p)(ifp, m); 1048 return; 1049 } 1050 1051 if (m->m_flags & M_VLANTAG) { 1052 void (*vlan_input_func)(struct mbuf *); 1053 1054 vlan_input_func = vlan_input_p; 1055 /* Make sure 'vlan_input_func' is really used. */ 1056 cpu_ccfence(); 1057 if (vlan_input_func != NULL) { 1058 vlan_input_func(m); 1059 } else { 1060 IFNET_STAT_INC(m->m_pkthdr.rcvif, noproto, 1); 1061 m_freem(m); 1062 } 1063 return; 1064 } 1065 1066 /* 1067 * If we have been asked to discard this packet 1068 * (e.g. not for us), drop it before entering 1069 * the upper layer. 1070 */ 1071 if (discard) { 1072 m_freem(m); 1073 return; 1074 } 1075 1076 /* 1077 * Clear protocol specific flags, 1078 * before entering the upper layer. 1079 */ 1080 m->m_flags &= ~M_ETHER_FLAGS; 1081 1082 /* Strip ethernet header. */ 1083 m_adj(m, sizeof(struct ether_header)); 1084 1085 switch (ether_type) { 1086 #ifdef INET 1087 case ETHERTYPE_IP: 1088 if ((m->m_flags & M_LENCHECKED) == 0) { 1089 if (!ip_lengthcheck(&m, 0)) 1090 return; 1091 } 1092 if (ipflow_fastforward(m)) 1093 return; 1094 isr = NETISR_IP; 1095 break; 1096 1097 case ETHERTYPE_ARP: 1098 if (ifp->if_flags & IFF_NOARP) { 1099 /* Discard packet if ARP is disabled on interface */ 1100 m_freem(m); 1101 return; 1102 } 1103 isr = NETISR_ARP; 1104 break; 1105 #endif 1106 1107 #ifdef INET6 1108 case ETHERTYPE_IPV6: 1109 isr = NETISR_IPV6; 1110 break; 1111 #endif 1112 1113 #ifdef MPLS 1114 case ETHERTYPE_MPLS: 1115 case ETHERTYPE_MPLS_MCAST: 1116 /* Should have been set by ether_input(). */ 1117 KKASSERT(m->m_flags & M_MPLSLABELED); 1118 isr = NETISR_MPLS; 1119 break; 1120 #endif 1121 1122 default: 1123 /* 1124 * The accurate msgport is not determined before 1125 * we reach here, so recharacterize packet. 1126 */ 1127 m->m_flags &= ~M_HASH; 1128 if (ng_ether_input_orphan_p != NULL) { 1129 /* 1130 * Put back the ethernet header so netgraph has a 1131 * consistent view of inbound packets. 1132 */ 1133 M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT); 1134 if (m == NULL) { 1135 /* 1136 * M_PREPEND frees the mbuf in case of failure. 1137 */ 1138 return; 1139 } 1140 /* 1141 * Hold BGL and recheck ng_ether_input_orphan_p 1142 */ 1143 get_mplock(); 1144 if (ng_ether_input_orphan_p != NULL) { 1145 ng_ether_input_orphan_p(ifp, m); 1146 rel_mplock(); 1147 return; 1148 } 1149 rel_mplock(); 1150 } 1151 m_freem(m); 1152 return; 1153 } 1154 1155 if (m->m_flags & M_HASH) { 1156 if (&curthread->td_msgport == 1157 netisr_hashport(m->m_pkthdr.hash)) { 1158 netisr_handle(isr, m); 1159 return; 1160 } else { 1161 /* 1162 * XXX Something is wrong, 1163 * we probably should panic here! 1164 */ 1165 m->m_flags &= ~M_HASH; 1166 atomic_add_long(ðer_input_wronghash, 1); 1167 } 1168 } 1169 #ifdef RSS_DEBUG 1170 atomic_add_long(ðer_input_requeue, 1); 1171 #endif 1172 netisr_queue(isr, m); 1173 } 1174 1175 /* 1176 * First we perform any link layer operations, then continue to the 1177 * upper layers with ether_demux_oncpu(). 1178 */ 1179 static void 1180 ether_input_oncpu(struct ifnet *ifp, struct mbuf *m) 1181 { 1182 #ifdef CARP 1183 void *carp; 1184 #endif 1185 1186 if ((ifp->if_flags & (IFF_UP | IFF_MONITOR)) != IFF_UP) { 1187 /* 1188 * Receiving interface's flags are changed, when this 1189 * packet is waiting for processing; discard it. 1190 */ 1191 m_freem(m); 1192 return; 1193 } 1194 1195 /* 1196 * Tap the packet off here for a bridge. bridge_input() 1197 * will return NULL if it has consumed the packet, otherwise 1198 * it gets processed as normal. Note that bridge_input() 1199 * will always return the original packet if we need to 1200 * process it locally. 1201 */ 1202 if (ifp->if_bridge) { 1203 KASSERT(bridge_input_p != NULL, 1204 ("%s: if_bridge not loaded!", __func__)); 1205 1206 if(m->m_flags & M_ETHER_BRIDGED) { 1207 m->m_flags &= ~M_ETHER_BRIDGED; 1208 } else { 1209 m = bridge_input_p(ifp, m); 1210 if (m == NULL) 1211 return; 1212 1213 KASSERT(ifp == m->m_pkthdr.rcvif, 1214 ("bridge_input_p changed rcvif")); 1215 } 1216 } 1217 1218 #ifdef CARP 1219 carp = ifp->if_carp; 1220 if (carp) { 1221 m = carp_input(carp, m); 1222 if (m == NULL) 1223 return; 1224 KASSERT(ifp == m->m_pkthdr.rcvif, 1225 ("carp_input changed rcvif")); 1226 } 1227 #endif 1228 1229 /* Handle ng_ether(4) processing, if any */ 1230 if (ng_ether_input_p != NULL) { 1231 /* 1232 * Hold BGL and recheck ng_ether_input_p 1233 */ 1234 get_mplock(); 1235 if (ng_ether_input_p != NULL) 1236 ng_ether_input_p(ifp, &m); 1237 rel_mplock(); 1238 1239 if (m == NULL) 1240 return; 1241 } 1242 1243 /* Continue with upper layer processing */ 1244 ether_demux_oncpu(ifp, m); 1245 } 1246 1247 /* 1248 * Perform certain functions of ether_input(): 1249 * - Test IFF_UP 1250 * - Update statistics 1251 * - Run bpf(4) tap if requested 1252 * Then pass the packet to ether_input_oncpu(). 1253 * 1254 * This function should be used by pseudo interface (e.g. vlan(4)), 1255 * when it tries to claim that the packet is received by it. 1256 * 1257 * REINPUT_KEEPRCVIF 1258 * REINPUT_RUNBPF 1259 */ 1260 void 1261 ether_reinput_oncpu(struct ifnet *ifp, struct mbuf *m, int reinput_flags) 1262 { 1263 /* Discard packet if interface is not up */ 1264 if (!(ifp->if_flags & IFF_UP)) { 1265 m_freem(m); 1266 return; 1267 } 1268 1269 /* 1270 * Change receiving interface. The bridge will often pass a flag to 1271 * ask that this not be done so ARPs get applied to the correct 1272 * side. 1273 */ 1274 if ((reinput_flags & REINPUT_KEEPRCVIF) == 0 || 1275 m->m_pkthdr.rcvif == NULL) { 1276 m->m_pkthdr.rcvif = ifp; 1277 } 1278 1279 /* Update statistics */ 1280 IFNET_STAT_INC(ifp, ipackets, 1); 1281 IFNET_STAT_INC(ifp, ibytes, m->m_pkthdr.len); 1282 if (m->m_flags & (M_MCAST | M_BCAST)) 1283 IFNET_STAT_INC(ifp, imcasts, 1); 1284 1285 if (reinput_flags & REINPUT_RUNBPF) 1286 BPF_MTAP(ifp, m); 1287 1288 ether_input_oncpu(ifp, m); 1289 } 1290 1291 static __inline boolean_t 1292 ether_vlancheck(struct mbuf **m0) 1293 { 1294 struct mbuf *m = *m0; 1295 struct ether_header *eh; 1296 uint16_t ether_type; 1297 1298 eh = mtod(m, struct ether_header *); 1299 ether_type = ntohs(eh->ether_type); 1300 1301 if (ether_type == ETHERTYPE_VLAN && (m->m_flags & M_VLANTAG) == 0) { 1302 /* 1303 * Extract vlan tag if hardware does not do it for us 1304 */ 1305 vlan_ether_decap(&m); 1306 if (m == NULL) 1307 goto failed; 1308 1309 eh = mtod(m, struct ether_header *); 1310 ether_type = ntohs(eh->ether_type); 1311 } 1312 1313 if (ether_type == ETHERTYPE_VLAN && (m->m_flags & M_VLANTAG)) { 1314 /* 1315 * To prevent possible dangerous recursion, 1316 * we don't do vlan-in-vlan 1317 */ 1318 IFNET_STAT_INC(m->m_pkthdr.rcvif, noproto, 1); 1319 goto failed; 1320 } 1321 KKASSERT(ether_type != ETHERTYPE_VLAN); 1322 1323 m->m_flags |= M_ETHER_VLANCHECKED; 1324 *m0 = m; 1325 return TRUE; 1326 failed: 1327 if (m != NULL) 1328 m_freem(m); 1329 *m0 = NULL; 1330 return FALSE; 1331 } 1332 1333 static void 1334 ether_input_handler(netmsg_t nmsg) 1335 { 1336 struct netmsg_packet *nmp = &nmsg->packet; /* actual size */ 1337 struct ether_header *eh; 1338 struct ifnet *ifp; 1339 struct mbuf *m; 1340 1341 m = nmp->nm_packet; 1342 M_ASSERTPKTHDR(m); 1343 1344 if ((m->m_flags & M_ETHER_VLANCHECKED) == 0) { 1345 if (!ether_vlancheck(&m)) { 1346 KKASSERT(m == NULL); 1347 return; 1348 } 1349 } 1350 if ((m->m_flags & (M_HASH | M_CKHASH)) == (M_HASH | M_CKHASH) || 1351 __predict_false(ether_input_ckhash)) { 1352 int isr; 1353 1354 /* 1355 * Need to verify the hash supplied by the hardware 1356 * which could be wrong. 1357 */ 1358 m->m_flags &= ~(M_HASH | M_CKHASH); 1359 isr = ether_characterize(&m); 1360 if (m == NULL) 1361 return; 1362 KKASSERT(m->m_flags & M_HASH); 1363 1364 if (netisr_hashcpu(m->m_pkthdr.hash) != mycpuid) { 1365 /* 1366 * Wrong hardware supplied hash; redispatch 1367 */ 1368 ether_dispatch(isr, m, -1); 1369 if (__predict_false(ether_input_ckhash)) 1370 atomic_add_long(ðer_input_wronghwhash, 1); 1371 return; 1372 } 1373 } 1374 ifp = m->m_pkthdr.rcvif; 1375 1376 eh = mtod(m, struct ether_header *); 1377 if (ETHER_IS_MULTICAST(eh->ether_dhost)) { 1378 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost, 1379 ifp->if_addrlen) == 0) 1380 m->m_flags |= M_BCAST; 1381 else 1382 m->m_flags |= M_MCAST; 1383 IFNET_STAT_INC(ifp, imcasts, 1); 1384 } 1385 1386 ether_input_oncpu(ifp, m); 1387 } 1388 1389 /* 1390 * Send the packet to the target netisr msgport 1391 * 1392 * At this point the packet must be characterized (M_HASH set), 1393 * so we know which netisr to send it to. 1394 */ 1395 static void 1396 ether_dispatch(int isr, struct mbuf *m, int cpuid) 1397 { 1398 struct netmsg_packet *pmsg; 1399 int target_cpuid; 1400 1401 KKASSERT(m->m_flags & M_HASH); 1402 target_cpuid = netisr_hashcpu(m->m_pkthdr.hash); 1403 1404 pmsg = &m->m_hdr.mh_netmsg; 1405 netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport, 1406 0, ether_input_handler); 1407 pmsg->nm_packet = m; 1408 pmsg->base.lmsg.u.ms_result = isr; 1409 1410 logether(disp_beg, NULL); 1411 if (target_cpuid == cpuid) { 1412 lwkt_sendmsg_oncpu(netisr_cpuport(target_cpuid), 1413 &pmsg->base.lmsg); 1414 } else { 1415 lwkt_sendmsg(netisr_cpuport(target_cpuid), 1416 &pmsg->base.lmsg); 1417 } 1418 logether(disp_end, NULL); 1419 } 1420 1421 /* 1422 * Process a received Ethernet packet. 1423 * 1424 * The ethernet header is assumed to be in the mbuf so the caller 1425 * MUST MAKE SURE that there are at least sizeof(struct ether_header) 1426 * bytes in the first mbuf. 1427 * 1428 * If the caller knows that the current thread is stick to the current 1429 * cpu, e.g. the interrupt thread or the netisr thread, the current cpuid 1430 * (mycpuid) should be passed through 'cpuid' argument. Else -1 should 1431 * be passed as 'cpuid' argument. 1432 */ 1433 void 1434 ether_input(struct ifnet *ifp, struct mbuf *m, const struct pktinfo *pi, 1435 int cpuid) 1436 { 1437 int isr; 1438 1439 M_ASSERTPKTHDR(m); 1440 1441 /* Discard packet if interface is not up */ 1442 if (!(ifp->if_flags & IFF_UP)) { 1443 m_freem(m); 1444 return; 1445 } 1446 1447 if (m->m_len < sizeof(struct ether_header)) { 1448 /* XXX error in the caller. */ 1449 m_freem(m); 1450 return; 1451 } 1452 1453 m->m_pkthdr.rcvif = ifp; 1454 1455 logether(pkt_beg, ifp); 1456 1457 ETHER_BPF_MTAP(ifp, m); 1458 1459 IFNET_STAT_INC(ifp, ibytes, m->m_pkthdr.len); 1460 1461 if (ifp->if_flags & IFF_MONITOR) { 1462 struct ether_header *eh; 1463 1464 eh = mtod(m, struct ether_header *); 1465 if (ETHER_IS_MULTICAST(eh->ether_dhost)) 1466 IFNET_STAT_INC(ifp, imcasts, 1); 1467 1468 /* 1469 * Interface marked for monitoring; discard packet. 1470 */ 1471 m_freem(m); 1472 1473 logether(pkt_end, ifp); 1474 return; 1475 } 1476 1477 /* 1478 * If the packet has been characterized (pi->pi_netisr / M_HASH) 1479 * we can dispatch it immediately with trivial checks. 1480 */ 1481 if (pi != NULL && (m->m_flags & M_HASH)) { 1482 #ifdef RSS_DEBUG 1483 atomic_add_long(ðer_pktinfo_try, 1); 1484 #endif 1485 netisr_hashcheck(pi->pi_netisr, m, pi); 1486 if (m->m_flags & M_HASH) { 1487 ether_dispatch(pi->pi_netisr, m, cpuid); 1488 #ifdef RSS_DEBUG 1489 atomic_add_long(ðer_pktinfo_hit, 1); 1490 #endif 1491 logether(pkt_end, ifp); 1492 return; 1493 } 1494 } 1495 #ifdef RSS_DEBUG 1496 else if (ifp->if_capenable & IFCAP_RSS) { 1497 if (pi == NULL) 1498 atomic_add_long(ðer_rss_nopi, 1); 1499 else 1500 atomic_add_long(ðer_rss_nohash, 1); 1501 } 1502 #endif 1503 1504 /* 1505 * Packet hash will be recalculated by software, so clear 1506 * the M_HASH and M_CKHASH flag set by the driver; the hash 1507 * value calculated by the hardware may not be exactly what 1508 * we want. 1509 */ 1510 m->m_flags &= ~(M_HASH | M_CKHASH); 1511 1512 if (!ether_vlancheck(&m)) { 1513 KKASSERT(m == NULL); 1514 logether(pkt_end, ifp); 1515 return; 1516 } 1517 1518 isr = ether_characterize(&m); 1519 if (m == NULL) { 1520 logether(pkt_end, ifp); 1521 return; 1522 } 1523 1524 /* 1525 * Finally dispatch it 1526 */ 1527 ether_dispatch(isr, m, cpuid); 1528 1529 logether(pkt_end, ifp); 1530 } 1531 1532 static int 1533 ether_characterize(struct mbuf **m0) 1534 { 1535 struct mbuf *m = *m0; 1536 struct ether_header *eh; 1537 uint16_t ether_type; 1538 int isr; 1539 1540 eh = mtod(m, struct ether_header *); 1541 ether_type = ntohs(eh->ether_type); 1542 1543 /* 1544 * Map ether type to netisr id. 1545 */ 1546 switch (ether_type) { 1547 #ifdef INET 1548 case ETHERTYPE_IP: 1549 isr = NETISR_IP; 1550 break; 1551 1552 case ETHERTYPE_ARP: 1553 isr = NETISR_ARP; 1554 break; 1555 #endif 1556 1557 #ifdef INET6 1558 case ETHERTYPE_IPV6: 1559 isr = NETISR_IPV6; 1560 break; 1561 #endif 1562 1563 #ifdef MPLS 1564 case ETHERTYPE_MPLS: 1565 case ETHERTYPE_MPLS_MCAST: 1566 m->m_flags |= M_MPLSLABELED; 1567 isr = NETISR_MPLS; 1568 break; 1569 #endif 1570 1571 default: 1572 /* 1573 * NETISR_MAX is an invalid value; it is chosen to let 1574 * netisr_characterize() know that we have no clear 1575 * idea where this packet should go. 1576 */ 1577 isr = NETISR_MAX; 1578 break; 1579 } 1580 1581 /* 1582 * Ask the isr to characterize the packet since we couldn't. 1583 * This is an attempt to optimally get us onto the correct protocol 1584 * thread. 1585 */ 1586 netisr_characterize(isr, &m, sizeof(struct ether_header)); 1587 1588 *m0 = m; 1589 return isr; 1590 } 1591 1592 static void 1593 ether_demux_handler(netmsg_t nmsg) 1594 { 1595 struct netmsg_packet *nmp = &nmsg->packet; /* actual size */ 1596 struct ifnet *ifp; 1597 struct mbuf *m; 1598 1599 m = nmp->nm_packet; 1600 M_ASSERTPKTHDR(m); 1601 ifp = m->m_pkthdr.rcvif; 1602 1603 ether_demux_oncpu(ifp, m); 1604 } 1605 1606 void 1607 ether_demux(struct mbuf *m) 1608 { 1609 struct netmsg_packet *pmsg; 1610 int isr; 1611 1612 isr = ether_characterize(&m); 1613 if (m == NULL) 1614 return; 1615 1616 KKASSERT(m->m_flags & M_HASH); 1617 pmsg = &m->m_hdr.mh_netmsg; 1618 netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport, 1619 0, ether_demux_handler); 1620 pmsg->nm_packet = m; 1621 pmsg->base.lmsg.u.ms_result = isr; 1622 1623 lwkt_sendmsg(netisr_hashport(m->m_pkthdr.hash), &pmsg->base.lmsg); 1624 } 1625 1626 u_char * 1627 kether_aton(const char *macstr, u_char *addr) 1628 { 1629 unsigned int o0, o1, o2, o3, o4, o5; 1630 int n; 1631 1632 if (macstr == NULL || addr == NULL) 1633 return NULL; 1634 1635 n = ksscanf(macstr, "%x:%x:%x:%x:%x:%x", &o0, &o1, &o2, 1636 &o3, &o4, &o5); 1637 if (n != 6) 1638 return NULL; 1639 1640 addr[0] = o0; 1641 addr[1] = o1; 1642 addr[2] = o2; 1643 addr[3] = o3; 1644 addr[4] = o4; 1645 addr[5] = o5; 1646 1647 return addr; 1648 } 1649 1650 char * 1651 kether_ntoa(const u_char *addr, char *buf) 1652 { 1653 int len = ETHER_ADDRSTRLEN + 1; 1654 int n; 1655 1656 n = ksnprintf(buf, len, "%02x:%02x:%02x:%02x:%02x:%02x", addr[0], 1657 addr[1], addr[2], addr[3], addr[4], addr[5]); 1658 1659 if (n < 17) 1660 return NULL; 1661 1662 return buf; 1663 } 1664 1665 MODULE_VERSION(ether, 1); 1666