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.cont = 0; 505 args.m = *m0; /* the packet we are looking at */ 506 args.oif = dst; /* destination, if any */ 507 args.rule = *rule; /* matching rule to restart */ 508 args.eh = &save_eh; /* MAC header for bridged/MAC packets */ 509 i = ip_fw_chk_ptr(&args); 510 *m0 = args.m; 511 *rule = args.rule; 512 513 if (*m0 == NULL) 514 return FALSE; 515 516 switch (i) { 517 case IP_FW_PASS: 518 return TRUE; 519 520 case IP_FW_DIVERT: 521 case IP_FW_TEE: 522 case IP_FW_DENY: 523 /* 524 * XXX at some point add support for divert/forward actions. 525 * If none of the above matches, we have to drop the pkt. 526 */ 527 return FALSE; 528 529 case IP_FW_DUMMYNET: 530 /* 531 * Pass the pkt to dummynet, which consumes it. 532 */ 533 m = *m0; /* pass the original to dummynet */ 534 *m0 = NULL; /* and nothing back to the caller */ 535 536 ether_restore_header(&m, eh, &save_eh); 537 if (m == NULL) 538 return FALSE; 539 540 m = ip_fw_dn_io_ptr(m, args.cookie, 541 dst ? DN_TO_ETH_OUT: DN_TO_ETH_DEMUX, &args); 542 if (m != NULL) 543 ip_dn_queue(m); 544 return FALSE; 545 546 default: 547 panic("unknown ipfw return value: %d", i); 548 } 549 } 550 551 /* 552 * Perform common duties while attaching to interface list 553 */ 554 void 555 ether_ifattach(struct ifnet *ifp, const uint8_t *lla, 556 lwkt_serialize_t serializer) 557 { 558 ether_ifattach_bpf(ifp, lla, DLT_EN10MB, sizeof(struct ether_header), 559 serializer); 560 } 561 562 void 563 ether_ifattach_bpf(struct ifnet *ifp, const uint8_t *lla, 564 u_int dlt, u_int hdrlen, lwkt_serialize_t serializer) 565 { 566 struct sockaddr_dl *sdl; 567 char ethstr[ETHER_ADDRSTRLEN + 1]; 568 struct ifaltq *ifq; 569 int i; 570 571 /* 572 * If driver does not configure # of mbuf clusters/jclusters 573 * that could sit on the device queues for quite some time, 574 * we then assume: 575 * - The device queues only consume mbuf clusters. 576 * - No more than ether_nmbclusters_default (by default 256) 577 * mbuf clusters will sit on the device queues for quite 578 * some time. 579 */ 580 if (ifp->if_nmbclusters <= 0 && ifp->if_nmbjclusters <= 0) { 581 if (ether_nmbclusters_default < ETHER_NMBCLUSTERS_DEFMIN) { 582 kprintf("ether nmbclusters %d -> %d\n", 583 ether_nmbclusters_default, 584 ETHER_NMBCLUSTERS_DEFAULT); 585 ether_nmbclusters_default = ETHER_NMBCLUSTERS_DEFAULT; 586 } 587 ifp->if_nmbclusters = ether_nmbclusters_default; 588 } 589 590 ifp->if_type = IFT_ETHER; 591 ifp->if_addrlen = ETHER_ADDR_LEN; 592 ifp->if_hdrlen = ETHER_HDR_LEN; 593 if_attach(ifp, serializer); 594 ifq = &ifp->if_snd; 595 for (i = 0; i < ifq->altq_subq_cnt; ++i) { 596 struct ifaltq_subque *ifsq = ifq_get_subq(ifq, i); 597 598 ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * 599 (ETHER_MAX_LEN - ETHER_CRC_LEN); 600 } 601 ifp->if_mtu = ETHERMTU; 602 if (ifp->if_tsolen <= 0) { 603 if ((ether_tsolen_default / ETHERMTU) < 2) { 604 kprintf("ether TSO maxlen %d -> %d\n", 605 ether_tsolen_default, ETHER_TSOLEN_DEFAULT); 606 ether_tsolen_default = ETHER_TSOLEN_DEFAULT; 607 } 608 ifp->if_tsolen = ether_tsolen_default; 609 } 610 if (ifp->if_baudrate == 0) 611 ifp->if_baudrate = 10000000; 612 ifp->if_output = ether_output; 613 ifp->if_input = ether_input; 614 ifp->if_resolvemulti = ether_resolvemulti; 615 ifp->if_broadcastaddr = etherbroadcastaddr; 616 sdl = IF_LLSOCKADDR(ifp); 617 sdl->sdl_type = IFT_ETHER; 618 sdl->sdl_alen = ifp->if_addrlen; 619 bcopy(lla, LLADDR(sdl), ifp->if_addrlen); 620 /* 621 * XXX Keep the current drivers happy. 622 * XXX Remove once all drivers have been cleaned up 623 */ 624 if (lla != IFP2AC(ifp)->ac_enaddr) 625 bcopy(lla, IFP2AC(ifp)->ac_enaddr, ifp->if_addrlen); 626 bpfattach(ifp, dlt, hdrlen); 627 if (ng_ether_attach_p != NULL) 628 (*ng_ether_attach_p)(ifp); 629 630 if_printf(ifp, "MAC address: %s\n", kether_ntoa(lla, ethstr)); 631 } 632 633 /* 634 * Perform common duties while detaching an Ethernet interface 635 */ 636 void 637 ether_ifdetach(struct ifnet *ifp) 638 { 639 if_down(ifp); 640 641 if (ng_ether_detach_p != NULL) 642 (*ng_ether_detach_p)(ifp); 643 bpfdetach(ifp); 644 if_detach(ifp); 645 } 646 647 int 648 ether_ioctl(struct ifnet *ifp, u_long command, caddr_t data) 649 { 650 struct ifaddr *ifa = (struct ifaddr *) data; 651 struct ifreq *ifr = (struct ifreq *) data; 652 int error = 0; 653 654 #define IF_INIT(ifp) \ 655 do { \ 656 if (((ifp)->if_flags & IFF_UP) == 0) { \ 657 (ifp)->if_flags |= IFF_UP; \ 658 (ifp)->if_init((ifp)->if_softc); \ 659 } \ 660 } while (0) 661 662 ASSERT_IFNET_SERIALIZED_ALL(ifp); 663 664 switch (command) { 665 case SIOCSIFADDR: 666 switch (ifa->ifa_addr->sa_family) { 667 #ifdef INET 668 case AF_INET: 669 IF_INIT(ifp); /* before arpwhohas */ 670 arp_ifinit(ifp, ifa); 671 break; 672 #endif 673 default: 674 IF_INIT(ifp); 675 break; 676 } 677 break; 678 679 case SIOCGIFADDR: 680 bcopy(IFP2AC(ifp)->ac_enaddr, 681 ((struct sockaddr *)ifr->ifr_data)->sa_data, 682 ETHER_ADDR_LEN); 683 break; 684 685 case SIOCSIFMTU: 686 /* 687 * Set the interface MTU. 688 */ 689 if (ifr->ifr_mtu > ETHERMTU) { 690 error = EINVAL; 691 } else { 692 ifp->if_mtu = ifr->ifr_mtu; 693 } 694 break; 695 default: 696 error = EINVAL; 697 break; 698 } 699 return (error); 700 701 #undef IF_INIT 702 } 703 704 static int 705 ether_resolvemulti( 706 struct ifnet *ifp, 707 struct sockaddr **llsa, 708 struct sockaddr *sa) 709 { 710 struct sockaddr_dl *sdl; 711 #ifdef INET 712 struct sockaddr_in *sin; 713 #endif 714 #ifdef INET6 715 struct sockaddr_in6 *sin6; 716 #endif 717 u_char *e_addr; 718 719 switch(sa->sa_family) { 720 case AF_LINK: 721 /* 722 * No mapping needed. Just check that it's a valid MC address. 723 */ 724 sdl = (struct sockaddr_dl *)sa; 725 e_addr = LLADDR(sdl); 726 if ((e_addr[0] & 1) != 1) 727 return EADDRNOTAVAIL; 728 *llsa = NULL; 729 return 0; 730 731 #ifdef INET 732 case AF_INET: 733 sin = (struct sockaddr_in *)sa; 734 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) 735 return EADDRNOTAVAIL; 736 sdl = kmalloc(sizeof *sdl, M_IFMADDR, M_WAITOK | M_ZERO); 737 sdl->sdl_len = sizeof *sdl; 738 sdl->sdl_family = AF_LINK; 739 sdl->sdl_index = ifp->if_index; 740 sdl->sdl_type = IFT_ETHER; 741 sdl->sdl_alen = ETHER_ADDR_LEN; 742 e_addr = LLADDR(sdl); 743 ETHER_MAP_IP_MULTICAST(&sin->sin_addr, e_addr); 744 *llsa = (struct sockaddr *)sdl; 745 return 0; 746 #endif 747 #ifdef INET6 748 case AF_INET6: 749 sin6 = (struct sockaddr_in6 *)sa; 750 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 751 /* 752 * An IP6 address of 0 means listen to all 753 * of the Ethernet multicast address used for IP6. 754 * (This is used for multicast routers.) 755 */ 756 ifp->if_flags |= IFF_ALLMULTI; 757 *llsa = NULL; 758 return 0; 759 } 760 if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 761 return EADDRNOTAVAIL; 762 sdl = kmalloc(sizeof *sdl, M_IFMADDR, M_WAITOK | M_ZERO); 763 sdl->sdl_len = sizeof *sdl; 764 sdl->sdl_family = AF_LINK; 765 sdl->sdl_index = ifp->if_index; 766 sdl->sdl_type = IFT_ETHER; 767 sdl->sdl_alen = ETHER_ADDR_LEN; 768 e_addr = LLADDR(sdl); 769 ETHER_MAP_IPV6_MULTICAST(&sin6->sin6_addr, e_addr); 770 *llsa = (struct sockaddr *)sdl; 771 return 0; 772 #endif 773 774 default: 775 /* 776 * Well, the text isn't quite right, but it's the name 777 * that counts... 778 */ 779 return EAFNOSUPPORT; 780 } 781 } 782 783 #if 0 784 /* 785 * This is for reference. We have a table-driven version 786 * of the little-endian crc32 generator, which is faster 787 * than the double-loop. 788 */ 789 uint32_t 790 ether_crc32_le(const uint8_t *buf, size_t len) 791 { 792 uint32_t c, crc, carry; 793 size_t i, j; 794 795 crc = 0xffffffffU; /* initial value */ 796 797 for (i = 0; i < len; i++) { 798 c = buf[i]; 799 for (j = 0; j < 8; j++) { 800 carry = ((crc & 0x01) ? 1 : 0) ^ (c & 0x01); 801 crc >>= 1; 802 c >>= 1; 803 if (carry) 804 crc = (crc ^ ETHER_CRC_POLY_LE); 805 } 806 } 807 808 return (crc); 809 } 810 #else 811 uint32_t 812 ether_crc32_le(const uint8_t *buf, size_t len) 813 { 814 static const uint32_t crctab[] = { 815 0x00000000, 0x1db71064, 0x3b6e20c8, 0x26d930ac, 816 0x76dc4190, 0x6b6b51f4, 0x4db26158, 0x5005713c, 817 0xedb88320, 0xf00f9344, 0xd6d6a3e8, 0xcb61b38c, 818 0x9b64c2b0, 0x86d3d2d4, 0xa00ae278, 0xbdbdf21c 819 }; 820 uint32_t crc; 821 size_t i; 822 823 crc = 0xffffffffU; /* initial value */ 824 825 for (i = 0; i < len; i++) { 826 crc ^= buf[i]; 827 crc = (crc >> 4) ^ crctab[crc & 0xf]; 828 crc = (crc >> 4) ^ crctab[crc & 0xf]; 829 } 830 831 return (crc); 832 } 833 #endif 834 835 uint32_t 836 ether_crc32_be(const uint8_t *buf, size_t len) 837 { 838 uint32_t c, crc, carry; 839 size_t i, j; 840 841 crc = 0xffffffffU; /* initial value */ 842 843 for (i = 0; i < len; i++) { 844 c = buf[i]; 845 for (j = 0; j < 8; j++) { 846 carry = ((crc & 0x80000000U) ? 1 : 0) ^ (c & 0x01); 847 crc <<= 1; 848 c >>= 1; 849 if (carry) 850 crc = (crc ^ ETHER_CRC_POLY_BE) | carry; 851 } 852 } 853 854 return (crc); 855 } 856 857 /* 858 * find the size of ethernet header, and call classifier 859 */ 860 void 861 altq_etherclassify(struct ifaltq *ifq, struct mbuf *m, 862 struct altq_pktattr *pktattr) 863 { 864 struct ether_header *eh; 865 uint16_t ether_type; 866 int hlen, af, hdrsize; 867 868 hlen = sizeof(struct ether_header); 869 eh = mtod(m, struct ether_header *); 870 871 ether_type = ntohs(eh->ether_type); 872 if (ether_type < ETHERMTU) { 873 /* ick! LLC/SNAP */ 874 struct llc *llc = (struct llc *)(eh + 1); 875 hlen += 8; 876 877 if (m->m_len < hlen || 878 llc->llc_dsap != LLC_SNAP_LSAP || 879 llc->llc_ssap != LLC_SNAP_LSAP || 880 llc->llc_control != LLC_UI) 881 goto bad; /* not snap! */ 882 883 ether_type = ntohs(llc->llc_un.type_snap.ether_type); 884 } 885 886 if (ether_type == ETHERTYPE_IP) { 887 af = AF_INET; 888 hdrsize = 20; /* sizeof(struct ip) */ 889 #ifdef INET6 890 } else if (ether_type == ETHERTYPE_IPV6) { 891 af = AF_INET6; 892 hdrsize = 40; /* sizeof(struct ip6_hdr) */ 893 #endif 894 } else 895 goto bad; 896 897 while (m->m_len <= hlen) { 898 hlen -= m->m_len; 899 m = m->m_next; 900 } 901 if (m->m_len < hlen + hdrsize) { 902 /* 903 * ip header is not in a single mbuf. this should not 904 * happen in the current code. 905 * (todo: use m_pulldown in the future) 906 */ 907 goto bad; 908 } 909 m->m_data += hlen; 910 m->m_len -= hlen; 911 ifq_classify(ifq, m, af, pktattr); 912 m->m_data -= hlen; 913 m->m_len += hlen; 914 915 return; 916 917 bad: 918 pktattr->pattr_class = NULL; 919 pktattr->pattr_hdr = NULL; 920 pktattr->pattr_af = AF_UNSPEC; 921 } 922 923 static void 924 ether_restore_header(struct mbuf **m0, const struct ether_header *eh, 925 const struct ether_header *save_eh) 926 { 927 struct mbuf *m = *m0; 928 929 ether_restore_hdr++; 930 931 /* 932 * Prepend the header, optimize for the common case of 933 * eh pointing into the mbuf. 934 */ 935 if ((const void *)(eh + 1) == (void *)m->m_data) { 936 m->m_data -= ETHER_HDR_LEN; 937 m->m_len += ETHER_HDR_LEN; 938 m->m_pkthdr.len += ETHER_HDR_LEN; 939 } else { 940 ether_prepend_hdr++; 941 942 M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT); 943 if (m != NULL) { 944 bcopy(save_eh, mtod(m, struct ether_header *), 945 ETHER_HDR_LEN); 946 } 947 } 948 *m0 = m; 949 } 950 951 /* 952 * Upper layer processing for a received Ethernet packet. 953 */ 954 void 955 ether_demux_oncpu(struct ifnet *ifp, struct mbuf *m) 956 { 957 struct ether_header *eh; 958 int isr, discard = 0; 959 u_short ether_type; 960 struct ip_fw *rule = NULL; 961 962 M_ASSERTPKTHDR(m); 963 KASSERT(m->m_len >= ETHER_HDR_LEN, 964 ("ether header is not contiguous!")); 965 966 eh = mtod(m, struct ether_header *); 967 968 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) { 969 struct m_tag *mtag; 970 971 /* Extract info from dummynet tag */ 972 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL); 973 KKASSERT(mtag != NULL); 974 rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv; 975 KKASSERT(rule != NULL); 976 977 m_tag_delete(m, mtag); 978 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED; 979 980 /* packet is passing the second time */ 981 goto post_stats; 982 } 983 984 /* 985 * We got a packet which was unicast to a different Ethernet 986 * address. If the driver is working properly, then this 987 * situation can only happen when the interface is in 988 * promiscuous mode. We defer the packet discarding until the 989 * vlan processing is done, so that vlan/bridge or vlan/netgraph 990 * could work. 991 */ 992 if (((ifp->if_flags & (IFF_PROMISC | IFF_PPROMISC)) == IFF_PROMISC) && 993 !ETHER_IS_MULTICAST(eh->ether_dhost) && 994 bcmp(eh->ether_dhost, IFP2AC(ifp)->ac_enaddr, ETHER_ADDR_LEN)) { 995 if (ether_debug & 1) { 996 kprintf("%02x:%02x:%02x:%02x:%02x:%02x " 997 "%02x:%02x:%02x:%02x:%02x:%02x " 998 "%04x vs %02x:%02x:%02x:%02x:%02x:%02x\n", 999 eh->ether_dhost[0], 1000 eh->ether_dhost[1], 1001 eh->ether_dhost[2], 1002 eh->ether_dhost[3], 1003 eh->ether_dhost[4], 1004 eh->ether_dhost[5], 1005 eh->ether_shost[0], 1006 eh->ether_shost[1], 1007 eh->ether_shost[2], 1008 eh->ether_shost[3], 1009 eh->ether_shost[4], 1010 eh->ether_shost[5], 1011 eh->ether_type, 1012 ((u_char *)IFP2AC(ifp)->ac_enaddr)[0], 1013 ((u_char *)IFP2AC(ifp)->ac_enaddr)[1], 1014 ((u_char *)IFP2AC(ifp)->ac_enaddr)[2], 1015 ((u_char *)IFP2AC(ifp)->ac_enaddr)[3], 1016 ((u_char *)IFP2AC(ifp)->ac_enaddr)[4], 1017 ((u_char *)IFP2AC(ifp)->ac_enaddr)[5] 1018 ); 1019 } 1020 if ((ether_debug & 2) == 0) 1021 discard = 1; 1022 } 1023 1024 post_stats: 1025 if ((IPFW_LOADED || IPFW3_LOADED) && ether_ipfw != 0 && !discard) { 1026 struct ether_header save_eh = *eh; 1027 1028 /* XXX old crufty stuff, needs to be removed */ 1029 m_adj(m, sizeof(struct ether_header)); 1030 1031 if (!ether_ipfw_chk(&m, NULL, &rule, eh)) { 1032 m_freem(m); 1033 return; 1034 } 1035 1036 ether_restore_header(&m, eh, &save_eh); 1037 if (m == NULL) 1038 return; 1039 eh = mtod(m, struct ether_header *); 1040 } 1041 1042 ether_type = ntohs(eh->ether_type); 1043 KKASSERT(ether_type != ETHERTYPE_VLAN); 1044 1045 /* Handle input from a lagg(4) port */ 1046 if (ifp->if_type == IFT_IEEE8023ADLAG) { 1047 KASSERT(lagg_input_p != NULL, 1048 ("%s: if_lagg not loaded!", __func__)); 1049 (*lagg_input_p)(ifp, m); 1050 return; 1051 } 1052 1053 if (m->m_flags & M_VLANTAG) { 1054 void (*vlan_input_func)(struct mbuf *); 1055 1056 vlan_input_func = vlan_input_p; 1057 /* Make sure 'vlan_input_func' is really used. */ 1058 cpu_ccfence(); 1059 if (vlan_input_func != NULL) { 1060 vlan_input_func(m); 1061 } else { 1062 IFNET_STAT_INC(m->m_pkthdr.rcvif, noproto, 1); 1063 m_freem(m); 1064 } 1065 return; 1066 } 1067 1068 /* 1069 * If we have been asked to discard this packet 1070 * (e.g. not for us), drop it before entering 1071 * the upper layer. 1072 */ 1073 if (discard) { 1074 m_freem(m); 1075 return; 1076 } 1077 1078 /* 1079 * Clear protocol specific flags, 1080 * before entering the upper layer. 1081 */ 1082 m->m_flags &= ~M_ETHER_FLAGS; 1083 1084 /* Strip ethernet header. */ 1085 m_adj(m, sizeof(struct ether_header)); 1086 1087 switch (ether_type) { 1088 #ifdef INET 1089 case ETHERTYPE_IP: 1090 if ((m->m_flags & M_LENCHECKED) == 0) { 1091 if (!ip_lengthcheck(&m, 0)) 1092 return; 1093 } 1094 if (ipflow_fastforward(m)) 1095 return; 1096 isr = NETISR_IP; 1097 break; 1098 1099 case ETHERTYPE_ARP: 1100 if (ifp->if_flags & IFF_NOARP) { 1101 /* Discard packet if ARP is disabled on interface */ 1102 m_freem(m); 1103 return; 1104 } 1105 isr = NETISR_ARP; 1106 break; 1107 #endif 1108 1109 #ifdef INET6 1110 case ETHERTYPE_IPV6: 1111 isr = NETISR_IPV6; 1112 break; 1113 #endif 1114 1115 #ifdef MPLS 1116 case ETHERTYPE_MPLS: 1117 case ETHERTYPE_MPLS_MCAST: 1118 /* Should have been set by ether_input(). */ 1119 KKASSERT(m->m_flags & M_MPLSLABELED); 1120 isr = NETISR_MPLS; 1121 break; 1122 #endif 1123 1124 default: 1125 /* 1126 * The accurate msgport is not determined before 1127 * we reach here, so recharacterize packet. 1128 */ 1129 m->m_flags &= ~M_HASH; 1130 if (ng_ether_input_orphan_p != NULL) { 1131 /* 1132 * Put back the ethernet header so netgraph has a 1133 * consistent view of inbound packets. 1134 */ 1135 M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT); 1136 if (m == NULL) { 1137 /* 1138 * M_PREPEND frees the mbuf in case of failure. 1139 */ 1140 return; 1141 } 1142 /* 1143 * Hold BGL and recheck ng_ether_input_orphan_p 1144 */ 1145 get_mplock(); 1146 if (ng_ether_input_orphan_p != NULL) { 1147 ng_ether_input_orphan_p(ifp, m); 1148 rel_mplock(); 1149 return; 1150 } 1151 rel_mplock(); 1152 } 1153 m_freem(m); 1154 return; 1155 } 1156 1157 if (m->m_flags & M_HASH) { 1158 if (&curthread->td_msgport == 1159 netisr_hashport(m->m_pkthdr.hash)) { 1160 netisr_handle(isr, m); 1161 return; 1162 } else { 1163 /* 1164 * XXX Something is wrong, 1165 * we probably should panic here! 1166 */ 1167 m->m_flags &= ~M_HASH; 1168 atomic_add_long(ðer_input_wronghash, 1); 1169 } 1170 } 1171 #ifdef RSS_DEBUG 1172 atomic_add_long(ðer_input_requeue, 1); 1173 #endif 1174 netisr_queue(isr, m); 1175 } 1176 1177 /* 1178 * First we perform any link layer operations, then continue to the 1179 * upper layers with ether_demux_oncpu(). 1180 */ 1181 static void 1182 ether_input_oncpu(struct ifnet *ifp, struct mbuf *m) 1183 { 1184 #ifdef CARP 1185 void *carp; 1186 #endif 1187 1188 if ((ifp->if_flags & (IFF_UP | IFF_MONITOR)) != IFF_UP) { 1189 /* 1190 * Receiving interface's flags are changed, when this 1191 * packet is waiting for processing; discard it. 1192 */ 1193 m_freem(m); 1194 return; 1195 } 1196 1197 /* 1198 * Tap the packet off here for a bridge. bridge_input() 1199 * will return NULL if it has consumed the packet, otherwise 1200 * it gets processed as normal. Note that bridge_input() 1201 * will always return the original packet if we need to 1202 * process it locally. 1203 */ 1204 if (ifp->if_bridge) { 1205 KASSERT(bridge_input_p != NULL, 1206 ("%s: if_bridge not loaded!", __func__)); 1207 1208 if(m->m_flags & M_ETHER_BRIDGED) { 1209 m->m_flags &= ~M_ETHER_BRIDGED; 1210 } else { 1211 m = bridge_input_p(ifp, m); 1212 if (m == NULL) 1213 return; 1214 1215 KASSERT(ifp == m->m_pkthdr.rcvif, 1216 ("bridge_input_p changed rcvif")); 1217 } 1218 } 1219 1220 #ifdef CARP 1221 carp = ifp->if_carp; 1222 if (carp) { 1223 m = carp_input(carp, m); 1224 if (m == NULL) 1225 return; 1226 KASSERT(ifp == m->m_pkthdr.rcvif, 1227 ("carp_input changed rcvif")); 1228 } 1229 #endif 1230 1231 /* Handle ng_ether(4) processing, if any */ 1232 if (ng_ether_input_p != NULL) { 1233 /* 1234 * Hold BGL and recheck ng_ether_input_p 1235 */ 1236 get_mplock(); 1237 if (ng_ether_input_p != NULL) 1238 ng_ether_input_p(ifp, &m); 1239 rel_mplock(); 1240 1241 if (m == NULL) 1242 return; 1243 } 1244 1245 /* Continue with upper layer processing */ 1246 ether_demux_oncpu(ifp, m); 1247 } 1248 1249 /* 1250 * Perform certain functions of ether_input(): 1251 * - Test IFF_UP 1252 * - Update statistics 1253 * - Run bpf(4) tap if requested 1254 * Then pass the packet to ether_input_oncpu(). 1255 * 1256 * This function should be used by pseudo interface (e.g. vlan(4)), 1257 * when it tries to claim that the packet is received by it. 1258 * 1259 * REINPUT_KEEPRCVIF 1260 * REINPUT_RUNBPF 1261 */ 1262 void 1263 ether_reinput_oncpu(struct ifnet *ifp, struct mbuf *m, int reinput_flags) 1264 { 1265 /* Discard packet if interface is not up */ 1266 if (!(ifp->if_flags & IFF_UP)) { 1267 m_freem(m); 1268 return; 1269 } 1270 1271 /* 1272 * Change receiving interface. The bridge will often pass a flag to 1273 * ask that this not be done so ARPs get applied to the correct 1274 * side. 1275 */ 1276 if ((reinput_flags & REINPUT_KEEPRCVIF) == 0 || 1277 m->m_pkthdr.rcvif == NULL) { 1278 m->m_pkthdr.rcvif = ifp; 1279 } 1280 1281 /* Update statistics */ 1282 IFNET_STAT_INC(ifp, ipackets, 1); 1283 IFNET_STAT_INC(ifp, ibytes, m->m_pkthdr.len); 1284 if (m->m_flags & (M_MCAST | M_BCAST)) 1285 IFNET_STAT_INC(ifp, imcasts, 1); 1286 1287 if (reinput_flags & REINPUT_RUNBPF) 1288 BPF_MTAP(ifp, m); 1289 1290 ether_input_oncpu(ifp, m); 1291 } 1292 1293 static __inline boolean_t 1294 ether_vlancheck(struct mbuf **m0) 1295 { 1296 struct mbuf *m = *m0; 1297 struct ether_header *eh; 1298 uint16_t ether_type; 1299 1300 eh = mtod(m, struct ether_header *); 1301 ether_type = ntohs(eh->ether_type); 1302 1303 if (ether_type == ETHERTYPE_VLAN && (m->m_flags & M_VLANTAG) == 0) { 1304 /* 1305 * Extract vlan tag if hardware does not do it for us 1306 */ 1307 vlan_ether_decap(&m); 1308 if (m == NULL) 1309 goto failed; 1310 1311 eh = mtod(m, struct ether_header *); 1312 ether_type = ntohs(eh->ether_type); 1313 } 1314 1315 if (ether_type == ETHERTYPE_VLAN && (m->m_flags & M_VLANTAG)) { 1316 /* 1317 * To prevent possible dangerous recursion, 1318 * we don't do vlan-in-vlan 1319 */ 1320 IFNET_STAT_INC(m->m_pkthdr.rcvif, noproto, 1); 1321 goto failed; 1322 } 1323 KKASSERT(ether_type != ETHERTYPE_VLAN); 1324 1325 m->m_flags |= M_ETHER_VLANCHECKED; 1326 *m0 = m; 1327 return TRUE; 1328 failed: 1329 if (m != NULL) 1330 m_freem(m); 1331 *m0 = NULL; 1332 return FALSE; 1333 } 1334 1335 static void 1336 ether_input_handler(netmsg_t nmsg) 1337 { 1338 struct netmsg_packet *nmp = &nmsg->packet; /* actual size */ 1339 struct ether_header *eh; 1340 struct ifnet *ifp; 1341 struct mbuf *m; 1342 1343 m = nmp->nm_packet; 1344 M_ASSERTPKTHDR(m); 1345 1346 if ((m->m_flags & M_ETHER_VLANCHECKED) == 0) { 1347 if (!ether_vlancheck(&m)) { 1348 KKASSERT(m == NULL); 1349 return; 1350 } 1351 } 1352 if ((m->m_flags & (M_HASH | M_CKHASH)) == (M_HASH | M_CKHASH) || 1353 __predict_false(ether_input_ckhash)) { 1354 int isr; 1355 1356 /* 1357 * Need to verify the hash supplied by the hardware 1358 * which could be wrong. 1359 */ 1360 m->m_flags &= ~(M_HASH | M_CKHASH); 1361 isr = ether_characterize(&m); 1362 if (m == NULL) 1363 return; 1364 KKASSERT(m->m_flags & M_HASH); 1365 1366 if (netisr_hashcpu(m->m_pkthdr.hash) != mycpuid) { 1367 /* 1368 * Wrong hardware supplied hash; redispatch 1369 */ 1370 ether_dispatch(isr, m, -1); 1371 if (__predict_false(ether_input_ckhash)) 1372 atomic_add_long(ðer_input_wronghwhash, 1); 1373 return; 1374 } 1375 } 1376 ifp = m->m_pkthdr.rcvif; 1377 1378 eh = mtod(m, struct ether_header *); 1379 if (ETHER_IS_MULTICAST(eh->ether_dhost)) { 1380 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost, 1381 ifp->if_addrlen) == 0) 1382 m->m_flags |= M_BCAST; 1383 else 1384 m->m_flags |= M_MCAST; 1385 IFNET_STAT_INC(ifp, imcasts, 1); 1386 } 1387 1388 ether_input_oncpu(ifp, m); 1389 } 1390 1391 /* 1392 * Send the packet to the target netisr msgport 1393 * 1394 * At this point the packet must be characterized (M_HASH set), 1395 * so we know which netisr to send it to. 1396 */ 1397 static void 1398 ether_dispatch(int isr, struct mbuf *m, int cpuid) 1399 { 1400 struct netmsg_packet *pmsg; 1401 int target_cpuid; 1402 1403 KKASSERT(m->m_flags & M_HASH); 1404 target_cpuid = netisr_hashcpu(m->m_pkthdr.hash); 1405 1406 pmsg = &m->m_hdr.mh_netmsg; 1407 netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport, 1408 0, ether_input_handler); 1409 pmsg->nm_packet = m; 1410 pmsg->base.lmsg.u.ms_result = isr; 1411 1412 logether(disp_beg, NULL); 1413 if (target_cpuid == cpuid) { 1414 lwkt_sendmsg_oncpu(netisr_cpuport(target_cpuid), 1415 &pmsg->base.lmsg); 1416 } else { 1417 lwkt_sendmsg(netisr_cpuport(target_cpuid), 1418 &pmsg->base.lmsg); 1419 } 1420 logether(disp_end, NULL); 1421 } 1422 1423 /* 1424 * Process a received Ethernet packet. 1425 * 1426 * The ethernet header is assumed to be in the mbuf so the caller 1427 * MUST MAKE SURE that there are at least sizeof(struct ether_header) 1428 * bytes in the first mbuf. 1429 * 1430 * If the caller knows that the current thread is stick to the current 1431 * cpu, e.g. the interrupt thread or the netisr thread, the current cpuid 1432 * (mycpuid) should be passed through 'cpuid' argument. Else -1 should 1433 * be passed as 'cpuid' argument. 1434 */ 1435 void 1436 ether_input(struct ifnet *ifp, struct mbuf *m, const struct pktinfo *pi, 1437 int cpuid) 1438 { 1439 int isr; 1440 1441 M_ASSERTPKTHDR(m); 1442 1443 /* Discard packet if interface is not up */ 1444 if (!(ifp->if_flags & IFF_UP)) { 1445 m_freem(m); 1446 return; 1447 } 1448 1449 if (m->m_len < sizeof(struct ether_header)) { 1450 /* XXX error in the caller. */ 1451 m_freem(m); 1452 return; 1453 } 1454 1455 m->m_pkthdr.rcvif = ifp; 1456 1457 logether(pkt_beg, ifp); 1458 1459 ETHER_BPF_MTAP(ifp, m); 1460 1461 IFNET_STAT_INC(ifp, ibytes, m->m_pkthdr.len); 1462 1463 if (ifp->if_flags & IFF_MONITOR) { 1464 struct ether_header *eh; 1465 1466 eh = mtod(m, struct ether_header *); 1467 if (ETHER_IS_MULTICAST(eh->ether_dhost)) 1468 IFNET_STAT_INC(ifp, imcasts, 1); 1469 1470 /* 1471 * Interface marked for monitoring; discard packet. 1472 */ 1473 m_freem(m); 1474 1475 logether(pkt_end, ifp); 1476 return; 1477 } 1478 1479 /* 1480 * If the packet has been characterized (pi->pi_netisr / M_HASH) 1481 * we can dispatch it immediately with trivial checks. 1482 */ 1483 if (pi != NULL && (m->m_flags & M_HASH)) { 1484 #ifdef RSS_DEBUG 1485 atomic_add_long(ðer_pktinfo_try, 1); 1486 #endif 1487 netisr_hashcheck(pi->pi_netisr, m, pi); 1488 if (m->m_flags & M_HASH) { 1489 ether_dispatch(pi->pi_netisr, m, cpuid); 1490 #ifdef RSS_DEBUG 1491 atomic_add_long(ðer_pktinfo_hit, 1); 1492 #endif 1493 logether(pkt_end, ifp); 1494 return; 1495 } 1496 } 1497 #ifdef RSS_DEBUG 1498 else if (ifp->if_capenable & IFCAP_RSS) { 1499 if (pi == NULL) 1500 atomic_add_long(ðer_rss_nopi, 1); 1501 else 1502 atomic_add_long(ðer_rss_nohash, 1); 1503 } 1504 #endif 1505 1506 /* 1507 * Packet hash will be recalculated by software, so clear 1508 * the M_HASH and M_CKHASH flag set by the driver; the hash 1509 * value calculated by the hardware may not be exactly what 1510 * we want. 1511 */ 1512 m->m_flags &= ~(M_HASH | M_CKHASH); 1513 1514 if (!ether_vlancheck(&m)) { 1515 KKASSERT(m == NULL); 1516 logether(pkt_end, ifp); 1517 return; 1518 } 1519 1520 isr = ether_characterize(&m); 1521 if (m == NULL) { 1522 logether(pkt_end, ifp); 1523 return; 1524 } 1525 1526 /* 1527 * Finally dispatch it 1528 */ 1529 ether_dispatch(isr, m, cpuid); 1530 1531 logether(pkt_end, ifp); 1532 } 1533 1534 static int 1535 ether_characterize(struct mbuf **m0) 1536 { 1537 struct mbuf *m = *m0; 1538 struct ether_header *eh; 1539 uint16_t ether_type; 1540 int isr; 1541 1542 eh = mtod(m, struct ether_header *); 1543 ether_type = ntohs(eh->ether_type); 1544 1545 /* 1546 * Map ether type to netisr id. 1547 */ 1548 switch (ether_type) { 1549 #ifdef INET 1550 case ETHERTYPE_IP: 1551 isr = NETISR_IP; 1552 break; 1553 1554 case ETHERTYPE_ARP: 1555 isr = NETISR_ARP; 1556 break; 1557 #endif 1558 1559 #ifdef INET6 1560 case ETHERTYPE_IPV6: 1561 isr = NETISR_IPV6; 1562 break; 1563 #endif 1564 1565 #ifdef MPLS 1566 case ETHERTYPE_MPLS: 1567 case ETHERTYPE_MPLS_MCAST: 1568 m->m_flags |= M_MPLSLABELED; 1569 isr = NETISR_MPLS; 1570 break; 1571 #endif 1572 1573 default: 1574 /* 1575 * NETISR_MAX is an invalid value; it is chosen to let 1576 * netisr_characterize() know that we have no clear 1577 * idea where this packet should go. 1578 */ 1579 isr = NETISR_MAX; 1580 break; 1581 } 1582 1583 /* 1584 * Ask the isr to characterize the packet since we couldn't. 1585 * This is an attempt to optimally get us onto the correct protocol 1586 * thread. 1587 */ 1588 netisr_characterize(isr, &m, sizeof(struct ether_header)); 1589 1590 *m0 = m; 1591 return isr; 1592 } 1593 1594 static void 1595 ether_demux_handler(netmsg_t nmsg) 1596 { 1597 struct netmsg_packet *nmp = &nmsg->packet; /* actual size */ 1598 struct ifnet *ifp; 1599 struct mbuf *m; 1600 1601 m = nmp->nm_packet; 1602 M_ASSERTPKTHDR(m); 1603 ifp = m->m_pkthdr.rcvif; 1604 1605 ether_demux_oncpu(ifp, m); 1606 } 1607 1608 void 1609 ether_demux(struct mbuf *m) 1610 { 1611 struct netmsg_packet *pmsg; 1612 int isr; 1613 1614 isr = ether_characterize(&m); 1615 if (m == NULL) 1616 return; 1617 1618 KKASSERT(m->m_flags & M_HASH); 1619 pmsg = &m->m_hdr.mh_netmsg; 1620 netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport, 1621 0, ether_demux_handler); 1622 pmsg->nm_packet = m; 1623 pmsg->base.lmsg.u.ms_result = isr; 1624 1625 lwkt_sendmsg(netisr_hashport(m->m_pkthdr.hash), &pmsg->base.lmsg); 1626 } 1627 1628 u_char * 1629 kether_aton(const char *macstr, u_char *addr) 1630 { 1631 unsigned int o0, o1, o2, o3, o4, o5; 1632 int n; 1633 1634 if (macstr == NULL || addr == NULL) 1635 return NULL; 1636 1637 n = ksscanf(macstr, "%x:%x:%x:%x:%x:%x", &o0, &o1, &o2, 1638 &o3, &o4, &o5); 1639 if (n != 6) 1640 return NULL; 1641 1642 addr[0] = o0; 1643 addr[1] = o1; 1644 addr[2] = o2; 1645 addr[3] = o3; 1646 addr[4] = o4; 1647 addr[5] = o5; 1648 1649 return addr; 1650 } 1651 1652 char * 1653 kether_ntoa(const u_char *addr, char *buf) 1654 { 1655 int len = ETHER_ADDRSTRLEN + 1; 1656 int n; 1657 1658 n = ksnprintf(buf, len, "%02x:%02x:%02x:%02x:%02x:%02x", addr[0], 1659 addr[1], addr[2], addr[3], addr[4], addr[5]); 1660 1661 if (n < 17) 1662 return NULL; 1663 1664 return buf; 1665 } 1666 1667 MODULE_VERSION(ether, 1); 1668