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(struct ifnet *, 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 error = arpresolve(ifp, rt, m, dst, edst); 251 if (error != 0) 252 return error == EWOULDBLOCK ? 0 : error; 253 #ifdef MPLS 254 if (m->m_flags & M_MPLSLABELED) 255 eh->ether_type = htons(ETHERTYPE_MPLS); 256 else 257 #endif 258 eh->ether_type = htons(ETHERTYPE_IP); 259 break; 260 case AF_ARP: 261 { 262 struct arphdr *ah; 263 264 ah = mtod(m, struct arphdr *); 265 ah->ar_hrd = htons(ARPHRD_ETHER); 266 267 loop_copy = -1; /* if this is for us, don't do it */ 268 269 switch(ntohs(ah->ar_op)) { 270 case ARPOP_REVREQUEST: 271 case ARPOP_REVREPLY: 272 eh->ether_type = htons(ETHERTYPE_REVARP); 273 break; 274 case ARPOP_REQUEST: 275 case ARPOP_REPLY: 276 default: 277 eh->ether_type = htons(ETHERTYPE_ARP); 278 break; 279 } 280 281 if (m->m_flags & M_BCAST) 282 bcopy(ifp->if_broadcastaddr, edst, ETHER_ADDR_LEN); 283 else 284 bcopy(ar_tha(ah), edst, ETHER_ADDR_LEN); 285 } 286 #endif 287 #ifdef INET6 288 case AF_INET6: 289 error = nd6_resolve(&ac->ac_if, rt, m, dst, edst); 290 if (error != 0) 291 return error == EWOULDBLOCK ? 0 : error; 292 eh->ether_type = htons(ETHERTYPE_IPV6); 293 break; 294 #endif 295 case pseudo_AF_HDRCMPLT: 296 case AF_UNSPEC: 297 loop_copy = -1; /* if this is for us, don't do it */ 298 deh = (struct ether_header *)dst->sa_data; 299 memcpy(edst, deh->ether_dhost, ETHER_ADDR_LEN); 300 eh->ether_type = deh->ether_type; 301 break; 302 303 default: 304 if_printf(ifp, "can't handle af%d\n", dst->sa_family); 305 gotoerr(EAFNOSUPPORT); 306 } 307 308 if (dst->sa_family == pseudo_AF_HDRCMPLT) /* unlikely */ 309 memcpy(eh->ether_shost, 310 ((struct ether_header *)dst->sa_data)->ether_shost, 311 ETHER_ADDR_LEN); 312 else 313 memcpy(eh->ether_shost, ac->ac_enaddr, ETHER_ADDR_LEN); 314 315 /* 316 * Bridges require special output handling. 317 */ 318 if (ifp->if_bridge) { 319 KASSERT(bridge_output_p != NULL, 320 ("%s: if_bridge not loaded!", __func__)); 321 return bridge_output_p(ifp, m); 322 } 323 #if 0 /* XXX */ 324 if (ifp->if_lagg) { 325 KASSERT(lagg_output_p != NULL, 326 ("%s: if_lagg not loaded!", __func__)); 327 return lagg_output_p(ifp, m); 328 } 329 #endif 330 331 /* 332 * If a simplex interface, and the packet is being sent to our 333 * Ethernet address or a broadcast address, loopback a copy. 334 * XXX To make a simplex device behave exactly like a duplex 335 * device, we should copy in the case of sending to our own 336 * ethernet address (thus letting the original actually appear 337 * on the wire). However, we don't do that here for security 338 * reasons and compatibility with the original behavior. 339 */ 340 if ((ifp->if_flags & IFF_SIMPLEX) && (loop_copy != -1)) { 341 int csum_flags = 0; 342 343 if (m->m_pkthdr.csum_flags & CSUM_IP) 344 csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID); 345 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) 346 csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 347 if ((m->m_flags & M_BCAST) || (loop_copy > 0)) { 348 struct mbuf *n; 349 350 if ((n = m_copypacket(m, M_NOWAIT)) != NULL) { 351 n->m_pkthdr.csum_flags |= csum_flags; 352 if (csum_flags & CSUM_DATA_VALID) 353 n->m_pkthdr.csum_data = 0xffff; 354 if_simloop(ifp, n, dst->sa_family, hlen); 355 } else 356 IFNET_STAT_INC(ifp, iqdrops, 1); 357 } else if (bcmp(eh->ether_dhost, eh->ether_shost, 358 ETHER_ADDR_LEN) == 0) { 359 m->m_pkthdr.csum_flags |= csum_flags; 360 if (csum_flags & CSUM_DATA_VALID) 361 m->m_pkthdr.csum_data = 0xffff; 362 if_simloop(ifp, m, dst->sa_family, hlen); 363 return (0); /* XXX */ 364 } 365 } 366 367 #ifdef CARP 368 if (ifp->if_type == IFT_CARP) { 369 ifp = carp_parent(ifp); 370 if (ifp == NULL) 371 gotoerr(ENETUNREACH); 372 373 ac = IFP2AC(ifp); 374 375 /* 376 * Check precondition again 377 */ 378 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp); 379 380 if (ifp->if_flags & IFF_MONITOR) 381 gotoerr(ENETDOWN); 382 if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) != 383 (IFF_UP | IFF_RUNNING)) 384 gotoerr(ENETDOWN); 385 } 386 #endif 387 388 /* Handle ng_ether(4) processing, if any */ 389 if (ng_ether_output_p != NULL) { 390 /* 391 * Hold BGL and recheck ng_ether_output_p 392 */ 393 get_mplock(); 394 if (ng_ether_output_p != NULL) { 395 if ((error = ng_ether_output_p(ifp, &m)) != 0) { 396 rel_mplock(); 397 goto bad; 398 } 399 if (m == NULL) { 400 rel_mplock(); 401 return (0); 402 } 403 } 404 rel_mplock(); 405 } 406 407 /* Continue with link-layer output */ 408 return ether_output_frame(ifp, m); 409 410 bad: 411 m_freem(m); 412 return (error); 413 } 414 415 /* 416 * Returns the bridge interface an ifp is associated 417 * with. 418 * 419 * Only call if ifp->if_bridge != NULL. 420 */ 421 struct ifnet * 422 ether_bridge_interface(struct ifnet *ifp) 423 { 424 if (bridge_interface_p) 425 return(bridge_interface_p(ifp->if_bridge)); 426 return (ifp); 427 } 428 429 /* 430 * Ethernet link layer output routine to send a raw frame to the device. 431 * 432 * This assumes that the 14 byte Ethernet header is present and contiguous 433 * in the first mbuf. 434 */ 435 int 436 ether_output_frame(struct ifnet *ifp, struct mbuf *m) 437 { 438 struct ip_fw *rule = NULL; 439 int error = 0; 440 struct altq_pktattr pktattr; 441 442 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp); 443 444 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) { 445 struct m_tag *mtag; 446 447 /* Extract info from dummynet tag */ 448 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL); 449 KKASSERT(mtag != NULL); 450 rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv; 451 KKASSERT(rule != NULL); 452 453 m_tag_delete(m, mtag); 454 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED; 455 } 456 457 if (ifq_is_enabled(&ifp->if_snd)) 458 altq_etherclassify(&ifp->if_snd, m, &pktattr); 459 crit_enter(); 460 if ((IPFW_LOADED || IPFW3_LOADED) && ether_ipfw != 0) { 461 struct ether_header save_eh, *eh; 462 463 eh = mtod(m, struct ether_header *); 464 save_eh = *eh; 465 m_adj(m, ETHER_HDR_LEN); 466 if (!ether_ipfw_chk(&m, ifp, &rule, eh)) { 467 crit_exit(); 468 if (m != NULL) { 469 m_freem(m); 470 return ENOBUFS; /* pkt dropped */ 471 } else 472 return 0; /* consumed e.g. in a pipe */ 473 } 474 475 /* packet was ok, restore the ethernet header */ 476 ether_restore_header(&m, eh, &save_eh); 477 if (m == NULL) { 478 crit_exit(); 479 return ENOBUFS; 480 } 481 } 482 crit_exit(); 483 484 /* 485 * Queue message on interface, update output statistics if 486 * successful, and start output if interface not yet active. 487 */ 488 error = ifq_dispatch(ifp, m, &pktattr); 489 return (error); 490 } 491 492 /* 493 * ipfw processing for ethernet packets (in and out). 494 * The second parameter is NULL from ether_demux(), and ifp from 495 * ether_output_frame(). 496 */ 497 static boolean_t 498 ether_ipfw_chk(struct mbuf **m0, struct ifnet *dst, struct ip_fw **rule, 499 const struct ether_header *eh) 500 { 501 struct ether_header save_eh = *eh; /* might be a ptr in *m0 */ 502 struct ip_fw_args args; 503 struct m_tag *mtag; 504 struct mbuf *m; 505 int i; 506 507 if (*rule != NULL && fw_one_pass) 508 return TRUE; /* dummynet packet, already partially processed */ 509 510 /* 511 * I need some amount of data to be contiguous. 512 */ 513 i = min((*m0)->m_pkthdr.len, max_protohdr); 514 if ((*m0)->m_len < i) { 515 *m0 = m_pullup(*m0, i); 516 if (*m0 == NULL) 517 return FALSE; 518 } 519 520 /* 521 * Clean up tags 522 */ 523 if ((mtag = m_tag_find(*m0, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL) 524 m_tag_delete(*m0, mtag); 525 if ((*m0)->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) { 526 mtag = m_tag_find(*m0, PACKET_TAG_IPFORWARD, NULL); 527 KKASSERT(mtag != NULL); 528 m_tag_delete(*m0, mtag); 529 (*m0)->m_pkthdr.fw_flags &= ~IPFORWARD_MBUF_TAGGED; 530 } 531 532 args.flags = 0; 533 args.xlat = NULL; 534 args.m = *m0; /* the packet we are looking at */ 535 args.oif = dst; /* destination, if any */ 536 args.rule = *rule; /* matching rule to restart */ 537 args.eh = &save_eh; /* MAC header for bridged/MAC packets */ 538 i = ip_fw_chk_ptr(&args); 539 *m0 = args.m; 540 *rule = args.rule; 541 542 if (*m0 == NULL) 543 return FALSE; 544 545 switch (i) { 546 case IP_FW_PASS: 547 return TRUE; 548 549 case IP_FW_DIVERT: 550 case IP_FW_TEE: 551 case IP_FW_DENY: 552 /* 553 * XXX at some point add support for divert/forward actions. 554 * If none of the above matches, we have to drop the pkt. 555 */ 556 return FALSE; 557 558 case IP_FW_DUMMYNET: 559 /* 560 * Pass the pkt to dummynet, which consumes it. 561 */ 562 m = *m0; /* pass the original to dummynet */ 563 *m0 = NULL; /* and nothing back to the caller */ 564 565 ether_restore_header(&m, eh, &save_eh); 566 if (m == NULL) 567 return FALSE; 568 569 m = ip_fw_dn_io_ptr(m, args.cookie, 570 dst ? DN_TO_ETH_OUT: DN_TO_ETH_DEMUX, &args); 571 if (m != NULL) 572 ip_dn_queue(m); 573 return FALSE; 574 575 default: 576 panic("unknown ipfw return value: %d", i); 577 } 578 } 579 580 /* 581 * Perform common duties while attaching to interface list 582 */ 583 void 584 ether_ifattach(struct ifnet *ifp, const uint8_t *lla, 585 lwkt_serialize_t serializer) 586 { 587 ether_ifattach_bpf(ifp, lla, DLT_EN10MB, sizeof(struct ether_header), 588 serializer); 589 } 590 591 void 592 ether_ifattach_bpf(struct ifnet *ifp, const uint8_t *lla, 593 u_int dlt, u_int hdrlen, lwkt_serialize_t serializer) 594 { 595 struct sockaddr_dl *sdl; 596 char ethstr[ETHER_ADDRSTRLEN + 1]; 597 struct ifaltq *ifq; 598 int i; 599 600 /* 601 * If driver does not configure # of mbuf clusters/jclusters 602 * that could sit on the device queues for quite some time, 603 * we then assume: 604 * - The device queues only consume mbuf clusters. 605 * - No more than ether_nmbclusters_default (by default 256) 606 * mbuf clusters will sit on the device queues for quite 607 * some time. 608 */ 609 if (ifp->if_nmbclusters <= 0 && ifp->if_nmbjclusters <= 0) { 610 if (ether_nmbclusters_default < ETHER_NMBCLUSTERS_DEFMIN) { 611 kprintf("ether nmbclusters %d -> %d\n", 612 ether_nmbclusters_default, 613 ETHER_NMBCLUSTERS_DEFAULT); 614 ether_nmbclusters_default = ETHER_NMBCLUSTERS_DEFAULT; 615 } 616 ifp->if_nmbclusters = ether_nmbclusters_default; 617 } 618 619 ifp->if_type = IFT_ETHER; 620 ifp->if_addrlen = ETHER_ADDR_LEN; 621 ifp->if_hdrlen = ETHER_HDR_LEN; 622 if_attach(ifp, serializer); 623 ifq = &ifp->if_snd; 624 for (i = 0; i < ifq->altq_subq_cnt; ++i) { 625 struct ifaltq_subque *ifsq = ifq_get_subq(ifq, i); 626 627 ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * 628 (ETHER_MAX_LEN - ETHER_CRC_LEN); 629 } 630 ifp->if_mtu = ETHERMTU; 631 if (ifp->if_tsolen <= 0) { 632 if ((ether_tsolen_default / ETHERMTU) < 2) { 633 kprintf("ether TSO maxlen %d -> %d\n", 634 ether_tsolen_default, ETHER_TSOLEN_DEFAULT); 635 ether_tsolen_default = ETHER_TSOLEN_DEFAULT; 636 } 637 ifp->if_tsolen = ether_tsolen_default; 638 } 639 if (ifp->if_baudrate == 0) 640 ifp->if_baudrate = 10000000; 641 ifp->if_output = ether_output; 642 ifp->if_input = ether_input; 643 ifp->if_resolvemulti = ether_resolvemulti; 644 ifp->if_broadcastaddr = etherbroadcastaddr; 645 sdl = IF_LLSOCKADDR(ifp); 646 sdl->sdl_type = IFT_ETHER; 647 sdl->sdl_alen = ifp->if_addrlen; 648 bcopy(lla, LLADDR(sdl), ifp->if_addrlen); 649 /* 650 * XXX Keep the current drivers happy. 651 * XXX Remove once all drivers have been cleaned up 652 */ 653 if (lla != IFP2AC(ifp)->ac_enaddr) 654 bcopy(lla, IFP2AC(ifp)->ac_enaddr, ifp->if_addrlen); 655 bpfattach(ifp, dlt, hdrlen); 656 if (ng_ether_attach_p != NULL) 657 (*ng_ether_attach_p)(ifp); 658 659 if_printf(ifp, "MAC address: %s\n", kether_ntoa(lla, ethstr)); 660 } 661 662 /* 663 * Perform common duties while detaching an Ethernet interface 664 */ 665 void 666 ether_ifdetach(struct ifnet *ifp) 667 { 668 if_down(ifp); 669 670 if (ng_ether_detach_p != NULL) 671 (*ng_ether_detach_p)(ifp); 672 bpfdetach(ifp); 673 if_detach(ifp); 674 } 675 676 int 677 ether_ioctl(struct ifnet *ifp, u_long command, caddr_t data) 678 { 679 struct ifaddr *ifa = (struct ifaddr *) data; 680 struct ifreq *ifr = (struct ifreq *) data; 681 int error = 0; 682 683 #define IF_INIT(ifp) \ 684 do { \ 685 if (((ifp)->if_flags & IFF_UP) == 0) { \ 686 (ifp)->if_flags |= IFF_UP; \ 687 (ifp)->if_init((ifp)->if_softc); \ 688 } \ 689 } while (0) 690 691 ASSERT_IFNET_SERIALIZED_ALL(ifp); 692 693 switch (command) { 694 case SIOCSIFADDR: 695 switch (ifa->ifa_addr->sa_family) { 696 #ifdef INET 697 case AF_INET: 698 IF_INIT(ifp); /* before arpwhohas */ 699 arp_ifinit(ifp, ifa); 700 break; 701 #endif 702 default: 703 IF_INIT(ifp); 704 break; 705 } 706 break; 707 708 case SIOCGIFADDR: 709 bcopy(IFP2AC(ifp)->ac_enaddr, 710 ((struct sockaddr *)ifr->ifr_data)->sa_data, 711 ETHER_ADDR_LEN); 712 break; 713 714 case SIOCSIFMTU: 715 /* 716 * Set the interface MTU. 717 */ 718 if (ifr->ifr_mtu > ETHERMTU) { 719 error = EINVAL; 720 } else { 721 ifp->if_mtu = ifr->ifr_mtu; 722 } 723 break; 724 default: 725 error = EINVAL; 726 break; 727 } 728 return (error); 729 730 #undef IF_INIT 731 } 732 733 static int 734 ether_resolvemulti( 735 struct ifnet *ifp, 736 struct sockaddr **llsa, 737 struct sockaddr *sa) 738 { 739 struct sockaddr_dl *sdl; 740 #ifdef INET 741 struct sockaddr_in *sin; 742 #endif 743 #ifdef INET6 744 struct sockaddr_in6 *sin6; 745 #endif 746 u_char *e_addr; 747 748 switch(sa->sa_family) { 749 case AF_LINK: 750 /* 751 * No mapping needed. Just check that it's a valid MC address. 752 */ 753 sdl = (struct sockaddr_dl *)sa; 754 e_addr = LLADDR(sdl); 755 if ((e_addr[0] & 1) != 1) 756 return EADDRNOTAVAIL; 757 *llsa = NULL; 758 return 0; 759 760 #ifdef INET 761 case AF_INET: 762 sin = (struct sockaddr_in *)sa; 763 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) 764 return EADDRNOTAVAIL; 765 sdl = kmalloc(sizeof *sdl, M_IFMADDR, M_WAITOK | M_ZERO); 766 sdl->sdl_len = sizeof *sdl; 767 sdl->sdl_family = AF_LINK; 768 sdl->sdl_index = ifp->if_index; 769 sdl->sdl_type = IFT_ETHER; 770 sdl->sdl_alen = ETHER_ADDR_LEN; 771 e_addr = LLADDR(sdl); 772 ETHER_MAP_IP_MULTICAST(&sin->sin_addr, e_addr); 773 *llsa = (struct sockaddr *)sdl; 774 return 0; 775 #endif 776 #ifdef INET6 777 case AF_INET6: 778 sin6 = (struct sockaddr_in6 *)sa; 779 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 780 /* 781 * An IP6 address of 0 means listen to all 782 * of the Ethernet multicast address used for IP6. 783 * (This is used for multicast routers.) 784 */ 785 ifp->if_flags |= IFF_ALLMULTI; 786 *llsa = NULL; 787 return 0; 788 } 789 if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 790 return EADDRNOTAVAIL; 791 sdl = kmalloc(sizeof *sdl, M_IFMADDR, M_WAITOK | M_ZERO); 792 sdl->sdl_len = sizeof *sdl; 793 sdl->sdl_family = AF_LINK; 794 sdl->sdl_index = ifp->if_index; 795 sdl->sdl_type = IFT_ETHER; 796 sdl->sdl_alen = ETHER_ADDR_LEN; 797 e_addr = LLADDR(sdl); 798 ETHER_MAP_IPV6_MULTICAST(&sin6->sin6_addr, e_addr); 799 *llsa = (struct sockaddr *)sdl; 800 return 0; 801 #endif 802 803 default: 804 /* 805 * Well, the text isn't quite right, but it's the name 806 * that counts... 807 */ 808 return EAFNOSUPPORT; 809 } 810 } 811 812 #if 0 813 /* 814 * This is for reference. We have a table-driven version 815 * of the little-endian crc32 generator, which is faster 816 * than the double-loop. 817 */ 818 uint32_t 819 ether_crc32_le(const uint8_t *buf, size_t len) 820 { 821 uint32_t c, crc, carry; 822 size_t i, j; 823 824 crc = 0xffffffffU; /* initial value */ 825 826 for (i = 0; i < len; i++) { 827 c = buf[i]; 828 for (j = 0; j < 8; j++) { 829 carry = ((crc & 0x01) ? 1 : 0) ^ (c & 0x01); 830 crc >>= 1; 831 c >>= 1; 832 if (carry) 833 crc = (crc ^ ETHER_CRC_POLY_LE); 834 } 835 } 836 837 return (crc); 838 } 839 #else 840 uint32_t 841 ether_crc32_le(const uint8_t *buf, size_t len) 842 { 843 static const uint32_t crctab[] = { 844 0x00000000, 0x1db71064, 0x3b6e20c8, 0x26d930ac, 845 0x76dc4190, 0x6b6b51f4, 0x4db26158, 0x5005713c, 846 0xedb88320, 0xf00f9344, 0xd6d6a3e8, 0xcb61b38c, 847 0x9b64c2b0, 0x86d3d2d4, 0xa00ae278, 0xbdbdf21c 848 }; 849 uint32_t crc; 850 size_t i; 851 852 crc = 0xffffffffU; /* initial value */ 853 854 for (i = 0; i < len; i++) { 855 crc ^= buf[i]; 856 crc = (crc >> 4) ^ crctab[crc & 0xf]; 857 crc = (crc >> 4) ^ crctab[crc & 0xf]; 858 } 859 860 return (crc); 861 } 862 #endif 863 864 uint32_t 865 ether_crc32_be(const uint8_t *buf, size_t len) 866 { 867 uint32_t c, crc, carry; 868 size_t i, j; 869 870 crc = 0xffffffffU; /* initial value */ 871 872 for (i = 0; i < len; i++) { 873 c = buf[i]; 874 for (j = 0; j < 8; j++) { 875 carry = ((crc & 0x80000000U) ? 1 : 0) ^ (c & 0x01); 876 crc <<= 1; 877 c >>= 1; 878 if (carry) 879 crc = (crc ^ ETHER_CRC_POLY_BE) | carry; 880 } 881 } 882 883 return (crc); 884 } 885 886 /* 887 * find the size of ethernet header, and call classifier 888 */ 889 void 890 altq_etherclassify(struct ifaltq *ifq, struct mbuf *m, 891 struct altq_pktattr *pktattr) 892 { 893 struct ether_header *eh; 894 uint16_t ether_type; 895 int hlen, af, hdrsize; 896 897 hlen = sizeof(struct ether_header); 898 eh = mtod(m, struct ether_header *); 899 900 ether_type = ntohs(eh->ether_type); 901 if (ether_type < ETHERMTU) { 902 /* ick! LLC/SNAP */ 903 struct llc *llc = (struct llc *)(eh + 1); 904 hlen += 8; 905 906 if (m->m_len < hlen || 907 llc->llc_dsap != LLC_SNAP_LSAP || 908 llc->llc_ssap != LLC_SNAP_LSAP || 909 llc->llc_control != LLC_UI) 910 goto bad; /* not snap! */ 911 912 ether_type = ntohs(llc->llc_un.type_snap.ether_type); 913 } 914 915 if (ether_type == ETHERTYPE_IP) { 916 af = AF_INET; 917 hdrsize = 20; /* sizeof(struct ip) */ 918 #ifdef INET6 919 } else if (ether_type == ETHERTYPE_IPV6) { 920 af = AF_INET6; 921 hdrsize = 40; /* sizeof(struct ip6_hdr) */ 922 #endif 923 } else 924 goto bad; 925 926 while (m->m_len <= hlen) { 927 hlen -= m->m_len; 928 m = m->m_next; 929 } 930 if (m->m_len < hlen + hdrsize) { 931 /* 932 * ip header is not in a single mbuf. this should not 933 * happen in the current code. 934 * (todo: use m_pulldown in the future) 935 */ 936 goto bad; 937 } 938 m->m_data += hlen; 939 m->m_len -= hlen; 940 ifq_classify(ifq, m, af, pktattr); 941 m->m_data -= hlen; 942 m->m_len += hlen; 943 944 return; 945 946 bad: 947 pktattr->pattr_class = NULL; 948 pktattr->pattr_hdr = NULL; 949 pktattr->pattr_af = AF_UNSPEC; 950 } 951 952 static void 953 ether_restore_header(struct mbuf **m0, const struct ether_header *eh, 954 const struct ether_header *save_eh) 955 { 956 struct mbuf *m = *m0; 957 958 ether_restore_hdr++; 959 960 /* 961 * Prepend the header, optimize for the common case of 962 * eh pointing into the mbuf. 963 */ 964 if ((const void *)(eh + 1) == (void *)m->m_data) { 965 m->m_data -= ETHER_HDR_LEN; 966 m->m_len += ETHER_HDR_LEN; 967 m->m_pkthdr.len += ETHER_HDR_LEN; 968 } else { 969 ether_prepend_hdr++; 970 971 M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT); 972 if (m != NULL) { 973 bcopy(save_eh, mtod(m, struct ether_header *), 974 ETHER_HDR_LEN); 975 } 976 } 977 *m0 = m; 978 } 979 980 /* 981 * Upper layer processing for a received Ethernet packet. 982 */ 983 void 984 ether_demux_oncpu(struct ifnet *ifp, struct mbuf *m) 985 { 986 struct ether_header *eh; 987 int isr, discard = 0; 988 u_short ether_type; 989 struct ip_fw *rule = NULL; 990 991 M_ASSERTPKTHDR(m); 992 KASSERT(m->m_len >= ETHER_HDR_LEN, 993 ("ether header is not contiguous!")); 994 995 eh = mtod(m, struct ether_header *); 996 997 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) { 998 struct m_tag *mtag; 999 1000 /* Extract info from dummynet tag */ 1001 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL); 1002 KKASSERT(mtag != NULL); 1003 rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv; 1004 KKASSERT(rule != NULL); 1005 1006 m_tag_delete(m, mtag); 1007 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED; 1008 1009 /* packet is passing the second time */ 1010 goto post_stats; 1011 } 1012 1013 /* 1014 * We got a packet which was unicast to a different Ethernet 1015 * address. If the driver is working properly, then this 1016 * situation can only happen when the interface is in 1017 * promiscuous mode. We defer the packet discarding until the 1018 * vlan processing is done, so that vlan/bridge or vlan/netgraph 1019 * could work. 1020 */ 1021 if (((ifp->if_flags & (IFF_PROMISC | IFF_PPROMISC)) == IFF_PROMISC) && 1022 !ETHER_IS_MULTICAST(eh->ether_dhost) && 1023 bcmp(eh->ether_dhost, IFP2AC(ifp)->ac_enaddr, ETHER_ADDR_LEN)) { 1024 if (ether_debug & 1) { 1025 kprintf("%02x:%02x:%02x:%02x:%02x:%02x " 1026 "%02x:%02x:%02x:%02x:%02x:%02x " 1027 "%04x vs %02x:%02x:%02x:%02x:%02x:%02x\n", 1028 eh->ether_dhost[0], 1029 eh->ether_dhost[1], 1030 eh->ether_dhost[2], 1031 eh->ether_dhost[3], 1032 eh->ether_dhost[4], 1033 eh->ether_dhost[5], 1034 eh->ether_shost[0], 1035 eh->ether_shost[1], 1036 eh->ether_shost[2], 1037 eh->ether_shost[3], 1038 eh->ether_shost[4], 1039 eh->ether_shost[5], 1040 eh->ether_type, 1041 ((u_char *)IFP2AC(ifp)->ac_enaddr)[0], 1042 ((u_char *)IFP2AC(ifp)->ac_enaddr)[1], 1043 ((u_char *)IFP2AC(ifp)->ac_enaddr)[2], 1044 ((u_char *)IFP2AC(ifp)->ac_enaddr)[3], 1045 ((u_char *)IFP2AC(ifp)->ac_enaddr)[4], 1046 ((u_char *)IFP2AC(ifp)->ac_enaddr)[5] 1047 ); 1048 } 1049 if ((ether_debug & 2) == 0) 1050 discard = 1; 1051 } 1052 1053 post_stats: 1054 if ((IPFW_LOADED || IPFW3_LOADED) && ether_ipfw != 0 && !discard) { 1055 struct ether_header save_eh = *eh; 1056 1057 /* XXX old crufty stuff, needs to be removed */ 1058 m_adj(m, sizeof(struct ether_header)); 1059 1060 if (!ether_ipfw_chk(&m, NULL, &rule, eh)) { 1061 m_freem(m); 1062 return; 1063 } 1064 1065 ether_restore_header(&m, eh, &save_eh); 1066 if (m == NULL) 1067 return; 1068 eh = mtod(m, struct ether_header *); 1069 } 1070 1071 ether_type = ntohs(eh->ether_type); 1072 KKASSERT(ether_type != ETHERTYPE_VLAN); 1073 1074 /* Handle input from a lagg(4) port */ 1075 if (ifp->if_type == IFT_IEEE8023ADLAG) { 1076 KASSERT(lagg_input_p != NULL, 1077 ("%s: if_lagg not loaded!", __func__)); 1078 (*lagg_input_p)(ifp, m); 1079 return; 1080 } 1081 1082 if (m->m_flags & M_VLANTAG) { 1083 void (*vlan_input_func)(struct mbuf *); 1084 1085 vlan_input_func = vlan_input_p; 1086 /* Make sure 'vlan_input_func' is really used. */ 1087 cpu_ccfence(); 1088 if (vlan_input_func != NULL) { 1089 vlan_input_func(m); 1090 } else { 1091 IFNET_STAT_INC(m->m_pkthdr.rcvif, noproto, 1); 1092 m_freem(m); 1093 } 1094 return; 1095 } 1096 1097 /* 1098 * If we have been asked to discard this packet 1099 * (e.g. not for us), drop it before entering 1100 * the upper layer. 1101 */ 1102 if (discard) { 1103 m_freem(m); 1104 return; 1105 } 1106 1107 /* 1108 * Clear protocol specific flags, 1109 * before entering the upper layer. 1110 */ 1111 m->m_flags &= ~M_ETHER_FLAGS; 1112 1113 /* Strip ethernet header. */ 1114 m_adj(m, sizeof(struct ether_header)); 1115 1116 switch (ether_type) { 1117 #ifdef INET 1118 case ETHERTYPE_IP: 1119 if ((m->m_flags & M_LENCHECKED) == 0) { 1120 if (!ip_lengthcheck(&m, 0)) 1121 return; 1122 } 1123 if (ipflow_fastforward(m)) 1124 return; 1125 isr = NETISR_IP; 1126 break; 1127 1128 case ETHERTYPE_ARP: 1129 if (ifp->if_flags & IFF_NOARP) { 1130 /* Discard packet if ARP is disabled on interface */ 1131 m_freem(m); 1132 return; 1133 } 1134 isr = NETISR_ARP; 1135 break; 1136 #endif 1137 1138 #ifdef INET6 1139 case ETHERTYPE_IPV6: 1140 isr = NETISR_IPV6; 1141 break; 1142 #endif 1143 1144 #ifdef MPLS 1145 case ETHERTYPE_MPLS: 1146 case ETHERTYPE_MPLS_MCAST: 1147 /* Should have been set by ether_input(). */ 1148 KKASSERT(m->m_flags & M_MPLSLABELED); 1149 isr = NETISR_MPLS; 1150 break; 1151 #endif 1152 1153 default: 1154 /* 1155 * The accurate msgport is not determined before 1156 * we reach here, so recharacterize packet. 1157 */ 1158 m->m_flags &= ~M_HASH; 1159 if (ng_ether_input_orphan_p != NULL) { 1160 /* 1161 * Put back the ethernet header so netgraph has a 1162 * consistent view of inbound packets. 1163 */ 1164 M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT); 1165 if (m == NULL) { 1166 /* 1167 * M_PREPEND frees the mbuf in case of failure. 1168 */ 1169 return; 1170 } 1171 /* 1172 * Hold BGL and recheck ng_ether_input_orphan_p 1173 */ 1174 get_mplock(); 1175 if (ng_ether_input_orphan_p != NULL) { 1176 ng_ether_input_orphan_p(ifp, m); 1177 rel_mplock(); 1178 return; 1179 } 1180 rel_mplock(); 1181 } 1182 m_freem(m); 1183 return; 1184 } 1185 1186 if (m->m_flags & M_HASH) { 1187 if (&curthread->td_msgport == 1188 netisr_hashport(m->m_pkthdr.hash)) { 1189 netisr_handle(isr, m); 1190 return; 1191 } else { 1192 /* 1193 * XXX Something is wrong, 1194 * we probably should panic here! 1195 */ 1196 m->m_flags &= ~M_HASH; 1197 atomic_add_long(ðer_input_wronghash, 1); 1198 } 1199 } 1200 #ifdef RSS_DEBUG 1201 atomic_add_long(ðer_input_requeue, 1); 1202 #endif 1203 netisr_queue(isr, m); 1204 } 1205 1206 /* 1207 * First we perform any link layer operations, then continue to the 1208 * upper layers with ether_demux_oncpu(). 1209 */ 1210 static void 1211 ether_input_oncpu(struct ifnet *ifp, struct mbuf *m) 1212 { 1213 #ifdef CARP 1214 void *carp; 1215 #endif 1216 1217 if ((ifp->if_flags & (IFF_UP | IFF_MONITOR)) != IFF_UP) { 1218 /* 1219 * Receiving interface's flags are changed, when this 1220 * packet is waiting for processing; discard it. 1221 */ 1222 m_freem(m); 1223 return; 1224 } 1225 1226 /* 1227 * Tap the packet off here for a bridge. bridge_input() 1228 * will return NULL if it has consumed the packet, otherwise 1229 * it gets processed as normal. Note that bridge_input() 1230 * will always return the original packet if we need to 1231 * process it locally. 1232 */ 1233 if (ifp->if_bridge) { 1234 KASSERT(bridge_input_p != NULL, 1235 ("%s: if_bridge not loaded!", __func__)); 1236 1237 if(m->m_flags & M_ETHER_BRIDGED) { 1238 m->m_flags &= ~M_ETHER_BRIDGED; 1239 } else { 1240 m = bridge_input_p(ifp, m); 1241 if (m == NULL) 1242 return; 1243 1244 KASSERT(ifp == m->m_pkthdr.rcvif, 1245 ("bridge_input_p changed rcvif")); 1246 } 1247 } 1248 1249 #ifdef CARP 1250 carp = ifp->if_carp; 1251 if (carp) { 1252 m = carp_input(carp, m); 1253 if (m == NULL) 1254 return; 1255 KASSERT(ifp == m->m_pkthdr.rcvif, 1256 ("carp_input changed rcvif")); 1257 } 1258 #endif 1259 1260 /* Handle ng_ether(4) processing, if any */ 1261 if (ng_ether_input_p != NULL) { 1262 /* 1263 * Hold BGL and recheck ng_ether_input_p 1264 */ 1265 get_mplock(); 1266 if (ng_ether_input_p != NULL) 1267 ng_ether_input_p(ifp, &m); 1268 rel_mplock(); 1269 1270 if (m == NULL) 1271 return; 1272 } 1273 1274 /* Continue with upper layer processing */ 1275 ether_demux_oncpu(ifp, m); 1276 } 1277 1278 /* 1279 * Perform certain functions of ether_input(): 1280 * - Test IFF_UP 1281 * - Update statistics 1282 * - Run bpf(4) tap if requested 1283 * Then pass the packet to ether_input_oncpu(). 1284 * 1285 * This function should be used by pseudo interface (e.g. vlan(4)), 1286 * when it tries to claim that the packet is received by it. 1287 * 1288 * REINPUT_KEEPRCVIF 1289 * REINPUT_RUNBPF 1290 */ 1291 void 1292 ether_reinput_oncpu(struct ifnet *ifp, struct mbuf *m, int reinput_flags) 1293 { 1294 /* Discard packet if interface is not up */ 1295 if (!(ifp->if_flags & IFF_UP)) { 1296 m_freem(m); 1297 return; 1298 } 1299 1300 /* 1301 * Change receiving interface. The bridge will often pass a flag to 1302 * ask that this not be done so ARPs get applied to the correct 1303 * side. 1304 */ 1305 if ((reinput_flags & REINPUT_KEEPRCVIF) == 0 || 1306 m->m_pkthdr.rcvif == NULL) { 1307 m->m_pkthdr.rcvif = ifp; 1308 } 1309 1310 /* Update statistics */ 1311 IFNET_STAT_INC(ifp, ipackets, 1); 1312 IFNET_STAT_INC(ifp, ibytes, m->m_pkthdr.len); 1313 if (m->m_flags & (M_MCAST | M_BCAST)) 1314 IFNET_STAT_INC(ifp, imcasts, 1); 1315 1316 if (reinput_flags & REINPUT_RUNBPF) 1317 BPF_MTAP(ifp, m); 1318 1319 ether_input_oncpu(ifp, m); 1320 } 1321 1322 static __inline boolean_t 1323 ether_vlancheck(struct mbuf **m0) 1324 { 1325 struct mbuf *m = *m0; 1326 struct ether_header *eh = mtod(m, struct ether_header *); 1327 uint16_t ether_type = ntohs(eh->ether_type); 1328 1329 if (ether_type == ETHERTYPE_VLAN) { 1330 if ((m->m_flags & M_VLANTAG) == 0) { 1331 /* 1332 * Extract vlan tag if hardware does not do 1333 * it for us. 1334 */ 1335 vlan_ether_decap(&m); 1336 if (m == NULL) 1337 goto failed; 1338 1339 eh = mtod(m, struct ether_header *); 1340 ether_type = ntohs(eh->ether_type); 1341 if (ether_type == ETHERTYPE_VLAN) { 1342 /* 1343 * To prevent possible dangerous recursion, 1344 * we don't do vlan-in-vlan. 1345 */ 1346 IFNET_STAT_INC(m->m_pkthdr.rcvif, noproto, 1); 1347 goto failed; 1348 } 1349 } else { 1350 /* 1351 * To prevent possible dangerous recursion, 1352 * we don't do vlan-in-vlan. 1353 */ 1354 IFNET_STAT_INC(m->m_pkthdr.rcvif, noproto, 1); 1355 goto failed; 1356 } 1357 KKASSERT(ether_type != ETHERTYPE_VLAN); 1358 } 1359 1360 m->m_flags |= M_ETHER_VLANCHECKED; 1361 *m0 = m; 1362 return TRUE; 1363 failed: 1364 if (m != NULL) 1365 m_freem(m); 1366 *m0 = NULL; 1367 return FALSE; 1368 } 1369 1370 static void 1371 ether_input_handler(netmsg_t nmsg) 1372 { 1373 struct netmsg_packet *nmp = &nmsg->packet; /* actual size */ 1374 struct ether_header *eh; 1375 struct ifnet *ifp; 1376 struct mbuf *m; 1377 1378 m = nmp->nm_packet; 1379 M_ASSERTPKTHDR(m); 1380 1381 if ((m->m_flags & M_ETHER_VLANCHECKED) == 0) { 1382 if (!ether_vlancheck(&m)) { 1383 KKASSERT(m == NULL); 1384 return; 1385 } 1386 } 1387 1388 ifp = m->m_pkthdr.rcvif; 1389 if ((m->m_flags & (M_HASH | M_CKHASH)) == (M_HASH | M_CKHASH) || 1390 __predict_false(ether_input_ckhash)) { 1391 int isr; 1392 1393 /* 1394 * Need to verify the hash supplied by the hardware 1395 * which could be wrong. 1396 */ 1397 m->m_flags &= ~(M_HASH | M_CKHASH); 1398 isr = ether_characterize(&m); 1399 if (m == NULL) 1400 return; 1401 KKASSERT(m->m_flags & M_HASH); 1402 1403 if (netisr_hashcpu(m->m_pkthdr.hash) != mycpuid) { 1404 /* 1405 * Wrong hardware supplied hash; redispatch 1406 */ 1407 ether_dispatch(ifp, isr, m, -1); 1408 if (__predict_false(ether_input_ckhash)) 1409 atomic_add_long(ðer_input_wronghwhash, 1); 1410 return; 1411 } 1412 } 1413 1414 eh = mtod(m, struct ether_header *); 1415 if (ETHER_IS_MULTICAST(eh->ether_dhost)) { 1416 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost, 1417 ifp->if_addrlen) == 0) 1418 m->m_flags |= M_BCAST; 1419 else 1420 m->m_flags |= M_MCAST; 1421 IFNET_STAT_INC(ifp, imcasts, 1); 1422 } 1423 1424 ether_input_oncpu(ifp, m); 1425 } 1426 1427 /* 1428 * Send the packet to the target netisr msgport 1429 * 1430 * At this point the packet must be characterized (M_HASH set), 1431 * so we know which netisr to send it to. 1432 */ 1433 static void 1434 ether_dispatch(struct ifnet *ifp, int isr, struct mbuf *m, int cpuid) 1435 { 1436 struct netmsg_packet *pmsg; 1437 int target_cpuid; 1438 1439 KKASSERT(m->m_flags & M_HASH); 1440 target_cpuid = netisr_hashcpu(m->m_pkthdr.hash); 1441 1442 pmsg = &m->m_hdr.mh_netmsg; 1443 netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport, 1444 0, ether_input_handler); 1445 pmsg->nm_packet = m; 1446 pmsg->base.lmsg.u.ms_result = isr; 1447 1448 logether(disp_beg, NULL); 1449 if (target_cpuid == cpuid) { 1450 if ((ifp->if_flags & IFF_IDIRECT) && IN_NETISR_NCPUS(cpuid)) { 1451 ether_input_handler((netmsg_t)pmsg); 1452 } else { 1453 lwkt_sendmsg_oncpu(netisr_cpuport(target_cpuid), 1454 &pmsg->base.lmsg); 1455 } 1456 } else { 1457 lwkt_sendmsg(netisr_cpuport(target_cpuid), 1458 &pmsg->base.lmsg); 1459 } 1460 logether(disp_end, NULL); 1461 } 1462 1463 /* 1464 * Process a received Ethernet packet. 1465 * 1466 * The ethernet header is assumed to be in the mbuf so the caller 1467 * MUST MAKE SURE that there are at least sizeof(struct ether_header) 1468 * bytes in the first mbuf. 1469 * 1470 * If the caller knows that the current thread is stick to the current 1471 * cpu, e.g. the interrupt thread or the netisr thread, the current cpuid 1472 * (mycpuid) should be passed through 'cpuid' argument. Else -1 should 1473 * be passed as 'cpuid' argument. 1474 */ 1475 void 1476 ether_input(struct ifnet *ifp, struct mbuf *m, const struct pktinfo *pi, 1477 int cpuid) 1478 { 1479 int isr; 1480 1481 M_ASSERTPKTHDR(m); 1482 1483 /* Discard packet if interface is not up */ 1484 if (!(ifp->if_flags & IFF_UP)) { 1485 m_freem(m); 1486 return; 1487 } 1488 1489 if (m->m_len < sizeof(struct ether_header)) { 1490 /* XXX error in the caller. */ 1491 m_freem(m); 1492 return; 1493 } 1494 1495 m->m_pkthdr.rcvif = ifp; 1496 1497 logether(pkt_beg, ifp); 1498 1499 ETHER_BPF_MTAP(ifp, m); 1500 1501 IFNET_STAT_INC(ifp, ibytes, m->m_pkthdr.len); 1502 1503 if (ifp->if_flags & IFF_MONITOR) { 1504 struct ether_header *eh; 1505 1506 eh = mtod(m, struct ether_header *); 1507 if (ETHER_IS_MULTICAST(eh->ether_dhost)) 1508 IFNET_STAT_INC(ifp, imcasts, 1); 1509 1510 /* 1511 * Interface marked for monitoring; discard packet. 1512 */ 1513 m_freem(m); 1514 1515 logether(pkt_end, ifp); 1516 return; 1517 } 1518 1519 /* 1520 * If the packet has been characterized (pi->pi_netisr / M_HASH) 1521 * we can dispatch it immediately with trivial checks. 1522 */ 1523 if (pi != NULL && (m->m_flags & M_HASH)) { 1524 #ifdef RSS_DEBUG 1525 atomic_add_long(ðer_pktinfo_try, 1); 1526 #endif 1527 netisr_hashcheck(pi->pi_netisr, m, pi); 1528 if (m->m_flags & M_HASH) { 1529 ether_dispatch(ifp, pi->pi_netisr, m, cpuid); 1530 #ifdef RSS_DEBUG 1531 atomic_add_long(ðer_pktinfo_hit, 1); 1532 #endif 1533 logether(pkt_end, ifp); 1534 return; 1535 } 1536 } 1537 #ifdef RSS_DEBUG 1538 else if (ifp->if_capenable & IFCAP_RSS) { 1539 if (pi == NULL) 1540 atomic_add_long(ðer_rss_nopi, 1); 1541 else 1542 atomic_add_long(ðer_rss_nohash, 1); 1543 } 1544 #endif 1545 1546 /* 1547 * Packet hash will be recalculated by software, so clear 1548 * the M_HASH and M_CKHASH flag set by the driver; the hash 1549 * value calculated by the hardware may not be exactly what 1550 * we want. 1551 */ 1552 m->m_flags &= ~(M_HASH | M_CKHASH); 1553 1554 if (!ether_vlancheck(&m)) { 1555 KKASSERT(m == NULL); 1556 logether(pkt_end, ifp); 1557 return; 1558 } 1559 1560 isr = ether_characterize(&m); 1561 if (m == NULL) { 1562 logether(pkt_end, ifp); 1563 return; 1564 } 1565 1566 /* 1567 * Finally dispatch it 1568 */ 1569 ether_dispatch(ifp, isr, m, cpuid); 1570 1571 logether(pkt_end, ifp); 1572 } 1573 1574 static int 1575 ether_characterize(struct mbuf **m0) 1576 { 1577 struct mbuf *m = *m0; 1578 struct ether_header *eh; 1579 uint16_t ether_type; 1580 int isr; 1581 1582 eh = mtod(m, struct ether_header *); 1583 ether_type = ntohs(eh->ether_type); 1584 1585 /* 1586 * Map ether type to netisr id. 1587 */ 1588 switch (ether_type) { 1589 #ifdef INET 1590 case ETHERTYPE_IP: 1591 isr = NETISR_IP; 1592 break; 1593 1594 case ETHERTYPE_ARP: 1595 isr = NETISR_ARP; 1596 break; 1597 #endif 1598 1599 #ifdef INET6 1600 case ETHERTYPE_IPV6: 1601 isr = NETISR_IPV6; 1602 break; 1603 #endif 1604 1605 #ifdef MPLS 1606 case ETHERTYPE_MPLS: 1607 case ETHERTYPE_MPLS_MCAST: 1608 m->m_flags |= M_MPLSLABELED; 1609 isr = NETISR_MPLS; 1610 break; 1611 #endif 1612 1613 default: 1614 /* 1615 * NETISR_MAX is an invalid value; it is chosen to let 1616 * netisr_characterize() know that we have no clear 1617 * idea where this packet should go. 1618 */ 1619 isr = NETISR_MAX; 1620 break; 1621 } 1622 1623 /* 1624 * Ask the isr to characterize the packet since we couldn't. 1625 * This is an attempt to optimally get us onto the correct protocol 1626 * thread. 1627 */ 1628 netisr_characterize(isr, &m, sizeof(struct ether_header)); 1629 1630 *m0 = m; 1631 return isr; 1632 } 1633 1634 static void 1635 ether_demux_handler(netmsg_t nmsg) 1636 { 1637 struct netmsg_packet *nmp = &nmsg->packet; /* actual size */ 1638 struct ifnet *ifp; 1639 struct mbuf *m; 1640 1641 m = nmp->nm_packet; 1642 M_ASSERTPKTHDR(m); 1643 ifp = m->m_pkthdr.rcvif; 1644 1645 ether_demux_oncpu(ifp, m); 1646 } 1647 1648 void 1649 ether_demux(struct mbuf *m) 1650 { 1651 struct netmsg_packet *pmsg; 1652 int isr; 1653 1654 isr = ether_characterize(&m); 1655 if (m == NULL) 1656 return; 1657 1658 KKASSERT(m->m_flags & M_HASH); 1659 pmsg = &m->m_hdr.mh_netmsg; 1660 netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport, 1661 0, ether_demux_handler); 1662 pmsg->nm_packet = m; 1663 pmsg->base.lmsg.u.ms_result = isr; 1664 1665 lwkt_sendmsg(netisr_hashport(m->m_pkthdr.hash), &pmsg->base.lmsg); 1666 } 1667 1668 u_char * 1669 kether_aton(const char *macstr, u_char *addr) 1670 { 1671 unsigned int o0, o1, o2, o3, o4, o5; 1672 int n; 1673 1674 if (macstr == NULL || addr == NULL) 1675 return NULL; 1676 1677 n = ksscanf(macstr, "%x:%x:%x:%x:%x:%x", &o0, &o1, &o2, 1678 &o3, &o4, &o5); 1679 if (n != 6) 1680 return NULL; 1681 1682 addr[0] = o0; 1683 addr[1] = o1; 1684 addr[2] = o2; 1685 addr[3] = o3; 1686 addr[4] = o4; 1687 addr[5] = o5; 1688 1689 return addr; 1690 } 1691 1692 char * 1693 kether_ntoa(const u_char *addr, char *buf) 1694 { 1695 int len = ETHER_ADDRSTRLEN + 1; 1696 int n; 1697 1698 n = ksnprintf(buf, len, "%02x:%02x:%02x:%02x:%02x:%02x", addr[0], 1699 addr[1], addr[2], addr[3], addr[4], addr[5]); 1700 1701 if (n < 17) 1702 return NULL; 1703 1704 return buf; 1705 } 1706 1707 MODULE_VERSION(ether, 1); 1708