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 #endif 261 #ifdef INET6 262 case AF_INET6: 263 error = nd6_resolve(&ac->ac_if, rt, m, dst, edst); 264 if (error != 0) 265 return error == EWOULDBLOCK ? 0 : error; 266 eh->ether_type = htons(ETHERTYPE_IPV6); 267 break; 268 #endif 269 case pseudo_AF_HDRCMPLT: 270 case AF_UNSPEC: 271 loop_copy = -1; /* if this is for us, don't do it */ 272 deh = (struct ether_header *)dst->sa_data; 273 memcpy(edst, deh->ether_dhost, ETHER_ADDR_LEN); 274 eh->ether_type = deh->ether_type; 275 break; 276 277 default: 278 if_printf(ifp, "can't handle af%d\n", dst->sa_family); 279 gotoerr(EAFNOSUPPORT); 280 } 281 282 if (dst->sa_family == pseudo_AF_HDRCMPLT) /* unlikely */ 283 memcpy(eh->ether_shost, 284 ((struct ether_header *)dst->sa_data)->ether_shost, 285 ETHER_ADDR_LEN); 286 else 287 memcpy(eh->ether_shost, ac->ac_enaddr, ETHER_ADDR_LEN); 288 289 /* 290 * Bridges require special output handling. 291 */ 292 if (ifp->if_bridge) { 293 KASSERT(bridge_output_p != NULL, 294 ("%s: if_bridge not loaded!", __func__)); 295 return bridge_output_p(ifp, m); 296 } 297 #if 0 /* XXX */ 298 if (ifp->if_lagg) { 299 KASSERT(lagg_output_p != NULL, 300 ("%s: if_lagg not loaded!", __func__)); 301 return lagg_output_p(ifp, m); 302 } 303 #endif 304 305 /* 306 * If a simplex interface, and the packet is being sent to our 307 * Ethernet address or a broadcast address, loopback a copy. 308 * XXX To make a simplex device behave exactly like a duplex 309 * device, we should copy in the case of sending to our own 310 * ethernet address (thus letting the original actually appear 311 * on the wire). However, we don't do that here for security 312 * reasons and compatibility with the original behavior. 313 */ 314 if ((ifp->if_flags & IFF_SIMPLEX) && (loop_copy != -1)) { 315 int csum_flags = 0; 316 317 if (m->m_pkthdr.csum_flags & CSUM_IP) 318 csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID); 319 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) 320 csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 321 if ((m->m_flags & M_BCAST) || (loop_copy > 0)) { 322 struct mbuf *n; 323 324 if ((n = m_copypacket(m, M_NOWAIT)) != NULL) { 325 n->m_pkthdr.csum_flags |= csum_flags; 326 if (csum_flags & CSUM_DATA_VALID) 327 n->m_pkthdr.csum_data = 0xffff; 328 if_simloop(ifp, n, dst->sa_family, hlen); 329 } else 330 IFNET_STAT_INC(ifp, iqdrops, 1); 331 } else if (bcmp(eh->ether_dhost, eh->ether_shost, 332 ETHER_ADDR_LEN) == 0) { 333 m->m_pkthdr.csum_flags |= csum_flags; 334 if (csum_flags & CSUM_DATA_VALID) 335 m->m_pkthdr.csum_data = 0xffff; 336 if_simloop(ifp, m, dst->sa_family, hlen); 337 return (0); /* XXX */ 338 } 339 } 340 341 #ifdef CARP 342 if (ifp->if_type == IFT_CARP) { 343 ifp = carp_parent(ifp); 344 if (ifp == NULL) 345 gotoerr(ENETUNREACH); 346 347 ac = IFP2AC(ifp); 348 349 /* 350 * Check precondition again 351 */ 352 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp); 353 354 if (ifp->if_flags & IFF_MONITOR) 355 gotoerr(ENETDOWN); 356 if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) != 357 (IFF_UP | IFF_RUNNING)) 358 gotoerr(ENETDOWN); 359 } 360 #endif 361 362 /* Handle ng_ether(4) processing, if any */ 363 if (ng_ether_output_p != NULL) { 364 /* 365 * Hold BGL and recheck ng_ether_output_p 366 */ 367 get_mplock(); 368 if (ng_ether_output_p != NULL) { 369 if ((error = ng_ether_output_p(ifp, &m)) != 0) { 370 rel_mplock(); 371 goto bad; 372 } 373 if (m == NULL) { 374 rel_mplock(); 375 return (0); 376 } 377 } 378 rel_mplock(); 379 } 380 381 /* Continue with link-layer output */ 382 return ether_output_frame(ifp, m); 383 384 bad: 385 m_freem(m); 386 return (error); 387 } 388 389 /* 390 * Returns the bridge interface an ifp is associated 391 * with. 392 * 393 * Only call if ifp->if_bridge != NULL. 394 */ 395 struct ifnet * 396 ether_bridge_interface(struct ifnet *ifp) 397 { 398 if (bridge_interface_p) 399 return(bridge_interface_p(ifp->if_bridge)); 400 return (ifp); 401 } 402 403 /* 404 * Ethernet link layer output routine to send a raw frame to the device. 405 * 406 * This assumes that the 14 byte Ethernet header is present and contiguous 407 * in the first mbuf. 408 */ 409 int 410 ether_output_frame(struct ifnet *ifp, struct mbuf *m) 411 { 412 struct ip_fw *rule = NULL; 413 int error = 0; 414 struct altq_pktattr pktattr; 415 416 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp); 417 418 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) { 419 struct m_tag *mtag; 420 421 /* Extract info from dummynet tag */ 422 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL); 423 KKASSERT(mtag != NULL); 424 rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv; 425 KKASSERT(rule != NULL); 426 427 m_tag_delete(m, mtag); 428 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED; 429 } 430 431 if (ifq_is_enabled(&ifp->if_snd)) 432 altq_etherclassify(&ifp->if_snd, m, &pktattr); 433 crit_enter(); 434 if ((IPFW_LOADED || IPFW3_LOADED) && ether_ipfw != 0) { 435 struct ether_header save_eh, *eh; 436 437 eh = mtod(m, struct ether_header *); 438 save_eh = *eh; 439 m_adj(m, ETHER_HDR_LEN); 440 if (!ether_ipfw_chk(&m, ifp, &rule, eh)) { 441 crit_exit(); 442 if (m != NULL) { 443 m_freem(m); 444 return ENOBUFS; /* pkt dropped */ 445 } else 446 return 0; /* consumed e.g. in a pipe */ 447 } 448 449 /* packet was ok, restore the ethernet header */ 450 ether_restore_header(&m, eh, &save_eh); 451 if (m == NULL) { 452 crit_exit(); 453 return ENOBUFS; 454 } 455 } 456 crit_exit(); 457 458 /* 459 * Queue message on interface, update output statistics if 460 * successful, and start output if interface not yet active. 461 */ 462 error = ifq_dispatch(ifp, m, &pktattr); 463 return (error); 464 } 465 466 /* 467 * ipfw processing for ethernet packets (in and out). 468 * The second parameter is NULL from ether_demux(), and ifp from 469 * ether_output_frame(). 470 */ 471 static boolean_t 472 ether_ipfw_chk(struct mbuf **m0, struct ifnet *dst, struct ip_fw **rule, 473 const struct ether_header *eh) 474 { 475 struct ether_header save_eh = *eh; /* might be a ptr in *m0 */ 476 struct ip_fw_args args; 477 struct m_tag *mtag; 478 struct mbuf *m; 479 int i; 480 481 if (*rule != NULL && fw_one_pass) 482 return TRUE; /* dummynet packet, already partially processed */ 483 484 /* 485 * I need some amount of data to be contiguous. 486 */ 487 i = min((*m0)->m_pkthdr.len, max_protohdr); 488 if ((*m0)->m_len < i) { 489 *m0 = m_pullup(*m0, i); 490 if (*m0 == NULL) 491 return FALSE; 492 } 493 494 /* 495 * Clean up tags 496 */ 497 if ((mtag = m_tag_find(*m0, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL) 498 m_tag_delete(*m0, mtag); 499 if ((*m0)->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) { 500 mtag = m_tag_find(*m0, PACKET_TAG_IPFORWARD, NULL); 501 KKASSERT(mtag != NULL); 502 m_tag_delete(*m0, mtag); 503 (*m0)->m_pkthdr.fw_flags &= ~IPFORWARD_MBUF_TAGGED; 504 } 505 506 args.flags = 0; 507 args.xlat = NULL; 508 args.m = *m0; /* the packet we are looking at */ 509 args.oif = dst; /* destination, if any */ 510 args.rule = *rule; /* matching rule to restart */ 511 args.eh = &save_eh; /* MAC header for bridged/MAC packets */ 512 i = ip_fw_chk_ptr(&args); 513 *m0 = args.m; 514 *rule = args.rule; 515 516 if (*m0 == NULL) 517 return FALSE; 518 519 switch (i) { 520 case IP_FW_PASS: 521 return TRUE; 522 523 case IP_FW_DIVERT: 524 case IP_FW_TEE: 525 case IP_FW_DENY: 526 /* 527 * XXX at some point add support for divert/forward actions. 528 * If none of the above matches, we have to drop the pkt. 529 */ 530 return FALSE; 531 532 case IP_FW_DUMMYNET: 533 /* 534 * Pass the pkt to dummynet, which consumes it. 535 */ 536 m = *m0; /* pass the original to dummynet */ 537 *m0 = NULL; /* and nothing back to the caller */ 538 539 ether_restore_header(&m, eh, &save_eh); 540 if (m == NULL) 541 return FALSE; 542 543 m = ip_fw_dn_io_ptr(m, args.cookie, 544 dst ? DN_TO_ETH_OUT: DN_TO_ETH_DEMUX, &args); 545 if (m != NULL) 546 ip_dn_queue(m); 547 return FALSE; 548 549 default: 550 panic("unknown ipfw return value: %d", i); 551 } 552 } 553 554 /* 555 * Perform common duties while attaching to interface list 556 */ 557 void 558 ether_ifattach(struct ifnet *ifp, const uint8_t *lla, 559 lwkt_serialize_t serializer) 560 { 561 ether_ifattach_bpf(ifp, lla, DLT_EN10MB, sizeof(struct ether_header), 562 serializer); 563 } 564 565 void 566 ether_ifattach_bpf(struct ifnet *ifp, const uint8_t *lla, 567 u_int dlt, u_int hdrlen, lwkt_serialize_t serializer) 568 { 569 struct sockaddr_dl *sdl; 570 char ethstr[ETHER_ADDRSTRLEN + 1]; 571 struct ifaltq *ifq; 572 int i; 573 574 /* 575 * If driver does not configure # of mbuf clusters/jclusters 576 * that could sit on the device queues for quite some time, 577 * we then assume: 578 * - The device queues only consume mbuf clusters. 579 * - No more than ether_nmbclusters_default (by default 256) 580 * mbuf clusters will sit on the device queues for quite 581 * some time. 582 */ 583 if (ifp->if_nmbclusters <= 0 && ifp->if_nmbjclusters <= 0) { 584 if (ether_nmbclusters_default < ETHER_NMBCLUSTERS_DEFMIN) { 585 kprintf("ether nmbclusters %d -> %d\n", 586 ether_nmbclusters_default, 587 ETHER_NMBCLUSTERS_DEFAULT); 588 ether_nmbclusters_default = ETHER_NMBCLUSTERS_DEFAULT; 589 } 590 ifp->if_nmbclusters = ether_nmbclusters_default; 591 } 592 593 ifp->if_type = IFT_ETHER; 594 ifp->if_addrlen = ETHER_ADDR_LEN; 595 ifp->if_hdrlen = ETHER_HDR_LEN; 596 if_attach(ifp, serializer); 597 ifq = &ifp->if_snd; 598 for (i = 0; i < ifq->altq_subq_cnt; ++i) { 599 struct ifaltq_subque *ifsq = ifq_get_subq(ifq, i); 600 601 ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * 602 (ETHER_MAX_LEN - ETHER_CRC_LEN); 603 } 604 ifp->if_mtu = ETHERMTU; 605 if (ifp->if_tsolen <= 0) { 606 if ((ether_tsolen_default / ETHERMTU) < 2) { 607 kprintf("ether TSO maxlen %d -> %d\n", 608 ether_tsolen_default, ETHER_TSOLEN_DEFAULT); 609 ether_tsolen_default = ETHER_TSOLEN_DEFAULT; 610 } 611 ifp->if_tsolen = ether_tsolen_default; 612 } 613 if (ifp->if_baudrate == 0) 614 ifp->if_baudrate = 10000000; 615 ifp->if_output = ether_output; 616 ifp->if_input = ether_input; 617 ifp->if_resolvemulti = ether_resolvemulti; 618 ifp->if_broadcastaddr = etherbroadcastaddr; 619 sdl = IF_LLSOCKADDR(ifp); 620 sdl->sdl_type = IFT_ETHER; 621 sdl->sdl_alen = ifp->if_addrlen; 622 bcopy(lla, LLADDR(sdl), ifp->if_addrlen); 623 /* 624 * XXX Keep the current drivers happy. 625 * XXX Remove once all drivers have been cleaned up 626 */ 627 if (lla != IFP2AC(ifp)->ac_enaddr) 628 bcopy(lla, IFP2AC(ifp)->ac_enaddr, ifp->if_addrlen); 629 bpfattach(ifp, dlt, hdrlen); 630 if (ng_ether_attach_p != NULL) 631 (*ng_ether_attach_p)(ifp); 632 633 if_printf(ifp, "MAC address: %s\n", kether_ntoa(lla, ethstr)); 634 } 635 636 /* 637 * Perform common duties while detaching an Ethernet interface 638 */ 639 void 640 ether_ifdetach(struct ifnet *ifp) 641 { 642 if_down(ifp); 643 644 if (ng_ether_detach_p != NULL) 645 (*ng_ether_detach_p)(ifp); 646 bpfdetach(ifp); 647 if_detach(ifp); 648 } 649 650 int 651 ether_ioctl(struct ifnet *ifp, u_long command, caddr_t data) 652 { 653 struct ifaddr *ifa = (struct ifaddr *) data; 654 struct ifreq *ifr = (struct ifreq *) data; 655 int error = 0; 656 657 #define IF_INIT(ifp) \ 658 do { \ 659 if (((ifp)->if_flags & IFF_UP) == 0) { \ 660 (ifp)->if_flags |= IFF_UP; \ 661 (ifp)->if_init((ifp)->if_softc); \ 662 } \ 663 } while (0) 664 665 ASSERT_IFNET_SERIALIZED_ALL(ifp); 666 667 switch (command) { 668 case SIOCSIFADDR: 669 switch (ifa->ifa_addr->sa_family) { 670 #ifdef INET 671 case AF_INET: 672 IF_INIT(ifp); /* before arpwhohas */ 673 arp_ifinit(ifp, ifa); 674 break; 675 #endif 676 default: 677 IF_INIT(ifp); 678 break; 679 } 680 break; 681 682 case SIOCGIFADDR: 683 bcopy(IFP2AC(ifp)->ac_enaddr, 684 ((struct sockaddr *)ifr->ifr_data)->sa_data, 685 ETHER_ADDR_LEN); 686 break; 687 688 case SIOCSIFMTU: 689 /* 690 * Set the interface MTU. 691 */ 692 if (ifr->ifr_mtu > ETHERMTU) { 693 error = EINVAL; 694 } else { 695 ifp->if_mtu = ifr->ifr_mtu; 696 } 697 break; 698 default: 699 error = EINVAL; 700 break; 701 } 702 return (error); 703 704 #undef IF_INIT 705 } 706 707 static int 708 ether_resolvemulti( 709 struct ifnet *ifp, 710 struct sockaddr **llsa, 711 struct sockaddr *sa) 712 { 713 struct sockaddr_dl *sdl; 714 #ifdef INET 715 struct sockaddr_in *sin; 716 #endif 717 #ifdef INET6 718 struct sockaddr_in6 *sin6; 719 #endif 720 u_char *e_addr; 721 722 switch(sa->sa_family) { 723 case AF_LINK: 724 /* 725 * No mapping needed. Just check that it's a valid MC address. 726 */ 727 sdl = (struct sockaddr_dl *)sa; 728 e_addr = LLADDR(sdl); 729 if ((e_addr[0] & 1) != 1) 730 return EADDRNOTAVAIL; 731 *llsa = NULL; 732 return 0; 733 734 #ifdef INET 735 case AF_INET: 736 sin = (struct sockaddr_in *)sa; 737 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) 738 return EADDRNOTAVAIL; 739 sdl = kmalloc(sizeof *sdl, M_IFMADDR, M_WAITOK | M_ZERO); 740 sdl->sdl_len = sizeof *sdl; 741 sdl->sdl_family = AF_LINK; 742 sdl->sdl_index = ifp->if_index; 743 sdl->sdl_type = IFT_ETHER; 744 sdl->sdl_alen = ETHER_ADDR_LEN; 745 e_addr = LLADDR(sdl); 746 ETHER_MAP_IP_MULTICAST(&sin->sin_addr, e_addr); 747 *llsa = (struct sockaddr *)sdl; 748 return 0; 749 #endif 750 #ifdef INET6 751 case AF_INET6: 752 sin6 = (struct sockaddr_in6 *)sa; 753 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 754 /* 755 * An IP6 address of 0 means listen to all 756 * of the Ethernet multicast address used for IP6. 757 * (This is used for multicast routers.) 758 */ 759 ifp->if_flags |= IFF_ALLMULTI; 760 *llsa = NULL; 761 return 0; 762 } 763 if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_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_IPV6_MULTICAST(&sin6->sin6_addr, e_addr); 773 *llsa = (struct sockaddr *)sdl; 774 return 0; 775 #endif 776 777 default: 778 /* 779 * Well, the text isn't quite right, but it's the name 780 * that counts... 781 */ 782 return EAFNOSUPPORT; 783 } 784 } 785 786 #if 0 787 /* 788 * This is for reference. We have a table-driven version 789 * of the little-endian crc32 generator, which is faster 790 * than the double-loop. 791 */ 792 uint32_t 793 ether_crc32_le(const uint8_t *buf, size_t len) 794 { 795 uint32_t c, crc, carry; 796 size_t i, j; 797 798 crc = 0xffffffffU; /* initial value */ 799 800 for (i = 0; i < len; i++) { 801 c = buf[i]; 802 for (j = 0; j < 8; j++) { 803 carry = ((crc & 0x01) ? 1 : 0) ^ (c & 0x01); 804 crc >>= 1; 805 c >>= 1; 806 if (carry) 807 crc = (crc ^ ETHER_CRC_POLY_LE); 808 } 809 } 810 811 return (crc); 812 } 813 #else 814 uint32_t 815 ether_crc32_le(const uint8_t *buf, size_t len) 816 { 817 static const uint32_t crctab[] = { 818 0x00000000, 0x1db71064, 0x3b6e20c8, 0x26d930ac, 819 0x76dc4190, 0x6b6b51f4, 0x4db26158, 0x5005713c, 820 0xedb88320, 0xf00f9344, 0xd6d6a3e8, 0xcb61b38c, 821 0x9b64c2b0, 0x86d3d2d4, 0xa00ae278, 0xbdbdf21c 822 }; 823 uint32_t crc; 824 size_t i; 825 826 crc = 0xffffffffU; /* initial value */ 827 828 for (i = 0; i < len; i++) { 829 crc ^= buf[i]; 830 crc = (crc >> 4) ^ crctab[crc & 0xf]; 831 crc = (crc >> 4) ^ crctab[crc & 0xf]; 832 } 833 834 return (crc); 835 } 836 #endif 837 838 uint32_t 839 ether_crc32_be(const uint8_t *buf, size_t len) 840 { 841 uint32_t c, crc, carry; 842 size_t i, j; 843 844 crc = 0xffffffffU; /* initial value */ 845 846 for (i = 0; i < len; i++) { 847 c = buf[i]; 848 for (j = 0; j < 8; j++) { 849 carry = ((crc & 0x80000000U) ? 1 : 0) ^ (c & 0x01); 850 crc <<= 1; 851 c >>= 1; 852 if (carry) 853 crc = (crc ^ ETHER_CRC_POLY_BE) | carry; 854 } 855 } 856 857 return (crc); 858 } 859 860 /* 861 * find the size of ethernet header, and call classifier 862 */ 863 void 864 altq_etherclassify(struct ifaltq *ifq, struct mbuf *m, 865 struct altq_pktattr *pktattr) 866 { 867 struct ether_header *eh; 868 uint16_t ether_type; 869 int hlen, af, hdrsize; 870 871 hlen = sizeof(struct ether_header); 872 eh = mtod(m, struct ether_header *); 873 874 ether_type = ntohs(eh->ether_type); 875 if (ether_type < ETHERMTU) { 876 /* ick! LLC/SNAP */ 877 struct llc *llc = (struct llc *)(eh + 1); 878 hlen += 8; 879 880 if (m->m_len < hlen || 881 llc->llc_dsap != LLC_SNAP_LSAP || 882 llc->llc_ssap != LLC_SNAP_LSAP || 883 llc->llc_control != LLC_UI) 884 goto bad; /* not snap! */ 885 886 ether_type = ntohs(llc->llc_un.type_snap.ether_type); 887 } 888 889 if (ether_type == ETHERTYPE_IP) { 890 af = AF_INET; 891 hdrsize = 20; /* sizeof(struct ip) */ 892 #ifdef INET6 893 } else if (ether_type == ETHERTYPE_IPV6) { 894 af = AF_INET6; 895 hdrsize = 40; /* sizeof(struct ip6_hdr) */ 896 #endif 897 } else 898 goto bad; 899 900 while (m->m_len <= hlen) { 901 hlen -= m->m_len; 902 m = m->m_next; 903 } 904 if (m->m_len < hlen + hdrsize) { 905 /* 906 * ip header is not in a single mbuf. this should not 907 * happen in the current code. 908 * (todo: use m_pulldown in the future) 909 */ 910 goto bad; 911 } 912 m->m_data += hlen; 913 m->m_len -= hlen; 914 ifq_classify(ifq, m, af, pktattr); 915 m->m_data -= hlen; 916 m->m_len += hlen; 917 918 return; 919 920 bad: 921 pktattr->pattr_class = NULL; 922 pktattr->pattr_hdr = NULL; 923 pktattr->pattr_af = AF_UNSPEC; 924 } 925 926 static void 927 ether_restore_header(struct mbuf **m0, const struct ether_header *eh, 928 const struct ether_header *save_eh) 929 { 930 struct mbuf *m = *m0; 931 932 ether_restore_hdr++; 933 934 /* 935 * Prepend the header, optimize for the common case of 936 * eh pointing into the mbuf. 937 */ 938 if ((const void *)(eh + 1) == (void *)m->m_data) { 939 m->m_data -= ETHER_HDR_LEN; 940 m->m_len += ETHER_HDR_LEN; 941 m->m_pkthdr.len += ETHER_HDR_LEN; 942 } else { 943 ether_prepend_hdr++; 944 945 M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT); 946 if (m != NULL) { 947 bcopy(save_eh, mtod(m, struct ether_header *), 948 ETHER_HDR_LEN); 949 } 950 } 951 *m0 = m; 952 } 953 954 /* 955 * Upper layer processing for a received Ethernet packet. 956 */ 957 void 958 ether_demux_oncpu(struct ifnet *ifp, struct mbuf *m) 959 { 960 struct ether_header *eh; 961 int isr, discard = 0; 962 u_short ether_type; 963 struct ip_fw *rule = NULL; 964 965 M_ASSERTPKTHDR(m); 966 KASSERT(m->m_len >= ETHER_HDR_LEN, 967 ("ether header is not contiguous!")); 968 969 eh = mtod(m, struct ether_header *); 970 971 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) { 972 struct m_tag *mtag; 973 974 /* Extract info from dummynet tag */ 975 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL); 976 KKASSERT(mtag != NULL); 977 rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv; 978 KKASSERT(rule != NULL); 979 980 m_tag_delete(m, mtag); 981 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED; 982 983 /* packet is passing the second time */ 984 goto post_stats; 985 } 986 987 /* 988 * We got a packet which was unicast to a different Ethernet 989 * address. If the driver is working properly, then this 990 * situation can only happen when the interface is in 991 * promiscuous mode. We defer the packet discarding until the 992 * vlan processing is done, so that vlan/bridge or vlan/netgraph 993 * could work. 994 */ 995 if (((ifp->if_flags & (IFF_PROMISC | IFF_PPROMISC)) == IFF_PROMISC) && 996 !ETHER_IS_MULTICAST(eh->ether_dhost) && 997 bcmp(eh->ether_dhost, IFP2AC(ifp)->ac_enaddr, ETHER_ADDR_LEN)) { 998 if (ether_debug & 1) { 999 kprintf("%02x:%02x:%02x:%02x:%02x:%02x " 1000 "%02x:%02x:%02x:%02x:%02x:%02x " 1001 "%04x vs %02x:%02x:%02x:%02x:%02x:%02x\n", 1002 eh->ether_dhost[0], 1003 eh->ether_dhost[1], 1004 eh->ether_dhost[2], 1005 eh->ether_dhost[3], 1006 eh->ether_dhost[4], 1007 eh->ether_dhost[5], 1008 eh->ether_shost[0], 1009 eh->ether_shost[1], 1010 eh->ether_shost[2], 1011 eh->ether_shost[3], 1012 eh->ether_shost[4], 1013 eh->ether_shost[5], 1014 eh->ether_type, 1015 ((u_char *)IFP2AC(ifp)->ac_enaddr)[0], 1016 ((u_char *)IFP2AC(ifp)->ac_enaddr)[1], 1017 ((u_char *)IFP2AC(ifp)->ac_enaddr)[2], 1018 ((u_char *)IFP2AC(ifp)->ac_enaddr)[3], 1019 ((u_char *)IFP2AC(ifp)->ac_enaddr)[4], 1020 ((u_char *)IFP2AC(ifp)->ac_enaddr)[5] 1021 ); 1022 } 1023 if ((ether_debug & 2) == 0) 1024 discard = 1; 1025 } 1026 1027 post_stats: 1028 if ((IPFW_LOADED || IPFW3_LOADED) && ether_ipfw != 0 && !discard) { 1029 struct ether_header save_eh = *eh; 1030 1031 /* XXX old crufty stuff, needs to be removed */ 1032 m_adj(m, sizeof(struct ether_header)); 1033 1034 if (!ether_ipfw_chk(&m, NULL, &rule, eh)) { 1035 m_freem(m); 1036 return; 1037 } 1038 1039 ether_restore_header(&m, eh, &save_eh); 1040 if (m == NULL) 1041 return; 1042 eh = mtod(m, struct ether_header *); 1043 } 1044 1045 ether_type = ntohs(eh->ether_type); 1046 KKASSERT(ether_type != ETHERTYPE_VLAN); 1047 1048 /* Handle input from a lagg(4) port */ 1049 if (ifp->if_type == IFT_IEEE8023ADLAG) { 1050 KASSERT(lagg_input_p != NULL, 1051 ("%s: if_lagg not loaded!", __func__)); 1052 (*lagg_input_p)(ifp, m); 1053 return; 1054 } 1055 1056 if (m->m_flags & M_VLANTAG) { 1057 void (*vlan_input_func)(struct mbuf *); 1058 1059 vlan_input_func = vlan_input_p; 1060 /* Make sure 'vlan_input_func' is really used. */ 1061 cpu_ccfence(); 1062 if (vlan_input_func != NULL) { 1063 vlan_input_func(m); 1064 } else { 1065 IFNET_STAT_INC(m->m_pkthdr.rcvif, noproto, 1); 1066 m_freem(m); 1067 } 1068 return; 1069 } 1070 1071 /* 1072 * If we have been asked to discard this packet 1073 * (e.g. not for us), drop it before entering 1074 * the upper layer. 1075 */ 1076 if (discard) { 1077 m_freem(m); 1078 return; 1079 } 1080 1081 /* 1082 * Clear protocol specific flags, 1083 * before entering the upper layer. 1084 */ 1085 m->m_flags &= ~M_ETHER_FLAGS; 1086 1087 /* Strip ethernet header. */ 1088 m_adj(m, sizeof(struct ether_header)); 1089 1090 switch (ether_type) { 1091 #ifdef INET 1092 case ETHERTYPE_IP: 1093 if ((m->m_flags & M_LENCHECKED) == 0) { 1094 if (!ip_lengthcheck(&m, 0)) 1095 return; 1096 } 1097 if (ipflow_fastforward(m)) 1098 return; 1099 isr = NETISR_IP; 1100 break; 1101 1102 case ETHERTYPE_ARP: 1103 if (ifp->if_flags & IFF_NOARP) { 1104 /* Discard packet if ARP is disabled on interface */ 1105 m_freem(m); 1106 return; 1107 } 1108 isr = NETISR_ARP; 1109 break; 1110 #endif 1111 1112 #ifdef INET6 1113 case ETHERTYPE_IPV6: 1114 isr = NETISR_IPV6; 1115 break; 1116 #endif 1117 1118 #ifdef MPLS 1119 case ETHERTYPE_MPLS: 1120 case ETHERTYPE_MPLS_MCAST: 1121 /* Should have been set by ether_input(). */ 1122 KKASSERT(m->m_flags & M_MPLSLABELED); 1123 isr = NETISR_MPLS; 1124 break; 1125 #endif 1126 1127 default: 1128 /* 1129 * The accurate msgport is not determined before 1130 * we reach here, so recharacterize packet. 1131 */ 1132 m->m_flags &= ~M_HASH; 1133 if (ng_ether_input_orphan_p != NULL) { 1134 /* 1135 * Put back the ethernet header so netgraph has a 1136 * consistent view of inbound packets. 1137 */ 1138 M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT); 1139 if (m == NULL) { 1140 /* 1141 * M_PREPEND frees the mbuf in case of failure. 1142 */ 1143 return; 1144 } 1145 /* 1146 * Hold BGL and recheck ng_ether_input_orphan_p 1147 */ 1148 get_mplock(); 1149 if (ng_ether_input_orphan_p != NULL) { 1150 ng_ether_input_orphan_p(ifp, m); 1151 rel_mplock(); 1152 return; 1153 } 1154 rel_mplock(); 1155 } 1156 m_freem(m); 1157 return; 1158 } 1159 1160 if (m->m_flags & M_HASH) { 1161 if (&curthread->td_msgport == 1162 netisr_hashport(m->m_pkthdr.hash)) { 1163 netisr_handle(isr, m); 1164 return; 1165 } else { 1166 /* 1167 * XXX Something is wrong, 1168 * we probably should panic here! 1169 */ 1170 m->m_flags &= ~M_HASH; 1171 atomic_add_long(ðer_input_wronghash, 1); 1172 } 1173 } 1174 #ifdef RSS_DEBUG 1175 atomic_add_long(ðer_input_requeue, 1); 1176 #endif 1177 netisr_queue(isr, m); 1178 } 1179 1180 /* 1181 * First we perform any link layer operations, then continue to the 1182 * upper layers with ether_demux_oncpu(). 1183 */ 1184 static void 1185 ether_input_oncpu(struct ifnet *ifp, struct mbuf *m) 1186 { 1187 #ifdef CARP 1188 void *carp; 1189 #endif 1190 1191 if ((ifp->if_flags & (IFF_UP | IFF_MONITOR)) != IFF_UP) { 1192 /* 1193 * Receiving interface's flags are changed, when this 1194 * packet is waiting for processing; discard it. 1195 */ 1196 m_freem(m); 1197 return; 1198 } 1199 1200 /* 1201 * Tap the packet off here for a bridge. bridge_input() 1202 * will return NULL if it has consumed the packet, otherwise 1203 * it gets processed as normal. Note that bridge_input() 1204 * will always return the original packet if we need to 1205 * process it locally. 1206 */ 1207 if (ifp->if_bridge) { 1208 KASSERT(bridge_input_p != NULL, 1209 ("%s: if_bridge not loaded!", __func__)); 1210 1211 if(m->m_flags & M_ETHER_BRIDGED) { 1212 m->m_flags &= ~M_ETHER_BRIDGED; 1213 } else { 1214 m = bridge_input_p(ifp, m); 1215 if (m == NULL) 1216 return; 1217 1218 KASSERT(ifp == m->m_pkthdr.rcvif, 1219 ("bridge_input_p changed rcvif")); 1220 } 1221 } 1222 1223 #ifdef CARP 1224 carp = ifp->if_carp; 1225 if (carp) { 1226 m = carp_input(carp, m); 1227 if (m == NULL) 1228 return; 1229 KASSERT(ifp == m->m_pkthdr.rcvif, 1230 ("carp_input changed rcvif")); 1231 } 1232 #endif 1233 1234 /* Handle ng_ether(4) processing, if any */ 1235 if (ng_ether_input_p != NULL) { 1236 /* 1237 * Hold BGL and recheck ng_ether_input_p 1238 */ 1239 get_mplock(); 1240 if (ng_ether_input_p != NULL) 1241 ng_ether_input_p(ifp, &m); 1242 rel_mplock(); 1243 1244 if (m == NULL) 1245 return; 1246 } 1247 1248 /* Continue with upper layer processing */ 1249 ether_demux_oncpu(ifp, m); 1250 } 1251 1252 /* 1253 * Perform certain functions of ether_input(): 1254 * - Test IFF_UP 1255 * - Update statistics 1256 * - Run bpf(4) tap if requested 1257 * Then pass the packet to ether_input_oncpu(). 1258 * 1259 * This function should be used by pseudo interface (e.g. vlan(4)), 1260 * when it tries to claim that the packet is received by it. 1261 * 1262 * REINPUT_KEEPRCVIF 1263 * REINPUT_RUNBPF 1264 */ 1265 void 1266 ether_reinput_oncpu(struct ifnet *ifp, struct mbuf *m, int reinput_flags) 1267 { 1268 /* Discard packet if interface is not up */ 1269 if (!(ifp->if_flags & IFF_UP)) { 1270 m_freem(m); 1271 return; 1272 } 1273 1274 /* 1275 * Change receiving interface. The bridge will often pass a flag to 1276 * ask that this not be done so ARPs get applied to the correct 1277 * side. 1278 */ 1279 if ((reinput_flags & REINPUT_KEEPRCVIF) == 0 || 1280 m->m_pkthdr.rcvif == NULL) { 1281 m->m_pkthdr.rcvif = ifp; 1282 } 1283 1284 /* Update statistics */ 1285 IFNET_STAT_INC(ifp, ipackets, 1); 1286 IFNET_STAT_INC(ifp, ibytes, m->m_pkthdr.len); 1287 if (m->m_flags & (M_MCAST | M_BCAST)) 1288 IFNET_STAT_INC(ifp, imcasts, 1); 1289 1290 if (reinput_flags & REINPUT_RUNBPF) 1291 BPF_MTAP(ifp, m); 1292 1293 ether_input_oncpu(ifp, m); 1294 } 1295 1296 static __inline boolean_t 1297 ether_vlancheck(struct mbuf **m0) 1298 { 1299 struct mbuf *m = *m0; 1300 struct ether_header *eh = mtod(m, struct ether_header *); 1301 uint16_t ether_type = ntohs(eh->ether_type); 1302 1303 if (ether_type == ETHERTYPE_VLAN) { 1304 if ((m->m_flags & M_VLANTAG) == 0) { 1305 /* 1306 * Extract vlan tag if hardware does not do 1307 * it for us. 1308 */ 1309 vlan_ether_decap(&m); 1310 if (m == NULL) 1311 goto failed; 1312 1313 eh = mtod(m, struct ether_header *); 1314 ether_type = ntohs(eh->ether_type); 1315 if (ether_type == ETHERTYPE_VLAN) { 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 } else { 1324 /* 1325 * To prevent possible dangerous recursion, 1326 * we don't do vlan-in-vlan. 1327 */ 1328 IFNET_STAT_INC(m->m_pkthdr.rcvif, noproto, 1); 1329 goto failed; 1330 } 1331 KKASSERT(ether_type != ETHERTYPE_VLAN); 1332 } 1333 1334 m->m_flags |= M_ETHER_VLANCHECKED; 1335 *m0 = m; 1336 return TRUE; 1337 failed: 1338 if (m != NULL) 1339 m_freem(m); 1340 *m0 = NULL; 1341 return FALSE; 1342 } 1343 1344 static void 1345 ether_input_handler(netmsg_t nmsg) 1346 { 1347 struct netmsg_packet *nmp = &nmsg->packet; /* actual size */ 1348 struct ether_header *eh; 1349 struct ifnet *ifp; 1350 struct mbuf *m; 1351 1352 m = nmp->nm_packet; 1353 M_ASSERTPKTHDR(m); 1354 1355 if ((m->m_flags & M_ETHER_VLANCHECKED) == 0) { 1356 if (!ether_vlancheck(&m)) { 1357 KKASSERT(m == NULL); 1358 return; 1359 } 1360 } 1361 1362 ifp = m->m_pkthdr.rcvif; 1363 if ((m->m_flags & (M_HASH | M_CKHASH)) == (M_HASH | M_CKHASH) || 1364 __predict_false(ether_input_ckhash)) { 1365 int isr; 1366 1367 /* 1368 * Need to verify the hash supplied by the hardware 1369 * which could be wrong. 1370 */ 1371 m->m_flags &= ~(M_HASH | M_CKHASH); 1372 isr = ether_characterize(&m); 1373 if (m == NULL) 1374 return; 1375 KKASSERT(m->m_flags & M_HASH); 1376 1377 if (netisr_hashcpu(m->m_pkthdr.hash) != mycpuid) { 1378 /* 1379 * Wrong hardware supplied hash; redispatch 1380 */ 1381 ether_dispatch(ifp, isr, m, -1); 1382 if (__predict_false(ether_input_ckhash)) 1383 atomic_add_long(ðer_input_wronghwhash, 1); 1384 return; 1385 } 1386 } 1387 1388 eh = mtod(m, struct ether_header *); 1389 if (ETHER_IS_MULTICAST(eh->ether_dhost)) { 1390 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost, 1391 ifp->if_addrlen) == 0) 1392 m->m_flags |= M_BCAST; 1393 else 1394 m->m_flags |= M_MCAST; 1395 IFNET_STAT_INC(ifp, imcasts, 1); 1396 } 1397 1398 ether_input_oncpu(ifp, m); 1399 } 1400 1401 /* 1402 * Send the packet to the target netisr msgport 1403 * 1404 * At this point the packet must be characterized (M_HASH set), 1405 * so we know which netisr to send it to. 1406 */ 1407 static void 1408 ether_dispatch(struct ifnet *ifp, int isr, struct mbuf *m, int cpuid) 1409 { 1410 struct netmsg_packet *pmsg; 1411 int target_cpuid; 1412 1413 KKASSERT(m->m_flags & M_HASH); 1414 target_cpuid = netisr_hashcpu(m->m_pkthdr.hash); 1415 1416 pmsg = &m->m_hdr.mh_netmsg; 1417 netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport, 1418 0, ether_input_handler); 1419 pmsg->nm_packet = m; 1420 pmsg->base.lmsg.u.ms_result = isr; 1421 1422 logether(disp_beg, NULL); 1423 if (target_cpuid == cpuid) { 1424 if ((ifp->if_flags & IFF_IDIRECT) && IN_NETISR_NCPUS(cpuid)) { 1425 ether_input_handler((netmsg_t)pmsg); 1426 } else { 1427 lwkt_sendmsg_oncpu(netisr_cpuport(target_cpuid), 1428 &pmsg->base.lmsg); 1429 } 1430 } else { 1431 lwkt_sendmsg(netisr_cpuport(target_cpuid), 1432 &pmsg->base.lmsg); 1433 } 1434 logether(disp_end, NULL); 1435 } 1436 1437 /* 1438 * Process a received Ethernet packet. 1439 * 1440 * The ethernet header is assumed to be in the mbuf so the caller 1441 * MUST MAKE SURE that there are at least sizeof(struct ether_header) 1442 * bytes in the first mbuf. 1443 * 1444 * If the caller knows that the current thread is stick to the current 1445 * cpu, e.g. the interrupt thread or the netisr thread, the current cpuid 1446 * (mycpuid) should be passed through 'cpuid' argument. Else -1 should 1447 * be passed as 'cpuid' argument. 1448 */ 1449 void 1450 ether_input(struct ifnet *ifp, struct mbuf *m, const struct pktinfo *pi, 1451 int cpuid) 1452 { 1453 int isr; 1454 1455 M_ASSERTPKTHDR(m); 1456 1457 /* Discard packet if interface is not up */ 1458 if (!(ifp->if_flags & IFF_UP)) { 1459 m_freem(m); 1460 return; 1461 } 1462 1463 if (m->m_len < sizeof(struct ether_header)) { 1464 /* XXX error in the caller. */ 1465 m_freem(m); 1466 return; 1467 } 1468 1469 m->m_pkthdr.rcvif = ifp; 1470 1471 logether(pkt_beg, ifp); 1472 1473 ETHER_BPF_MTAP(ifp, m); 1474 1475 IFNET_STAT_INC(ifp, ibytes, m->m_pkthdr.len); 1476 1477 if (ifp->if_flags & IFF_MONITOR) { 1478 struct ether_header *eh; 1479 1480 eh = mtod(m, struct ether_header *); 1481 if (ETHER_IS_MULTICAST(eh->ether_dhost)) 1482 IFNET_STAT_INC(ifp, imcasts, 1); 1483 1484 /* 1485 * Interface marked for monitoring; discard packet. 1486 */ 1487 m_freem(m); 1488 1489 logether(pkt_end, ifp); 1490 return; 1491 } 1492 1493 /* 1494 * If the packet has been characterized (pi->pi_netisr / M_HASH) 1495 * we can dispatch it immediately with trivial checks. 1496 */ 1497 if (pi != NULL && (m->m_flags & M_HASH)) { 1498 #ifdef RSS_DEBUG 1499 atomic_add_long(ðer_pktinfo_try, 1); 1500 #endif 1501 netisr_hashcheck(pi->pi_netisr, m, pi); 1502 if (m->m_flags & M_HASH) { 1503 ether_dispatch(ifp, pi->pi_netisr, m, cpuid); 1504 #ifdef RSS_DEBUG 1505 atomic_add_long(ðer_pktinfo_hit, 1); 1506 #endif 1507 logether(pkt_end, ifp); 1508 return; 1509 } 1510 } 1511 #ifdef RSS_DEBUG 1512 else if (ifp->if_capenable & IFCAP_RSS) { 1513 if (pi == NULL) 1514 atomic_add_long(ðer_rss_nopi, 1); 1515 else 1516 atomic_add_long(ðer_rss_nohash, 1); 1517 } 1518 #endif 1519 1520 /* 1521 * Packet hash will be recalculated by software, so clear 1522 * the M_HASH and M_CKHASH flag set by the driver; the hash 1523 * value calculated by the hardware may not be exactly what 1524 * we want. 1525 */ 1526 m->m_flags &= ~(M_HASH | M_CKHASH); 1527 1528 if (!ether_vlancheck(&m)) { 1529 KKASSERT(m == NULL); 1530 logether(pkt_end, ifp); 1531 return; 1532 } 1533 1534 isr = ether_characterize(&m); 1535 if (m == NULL) { 1536 logether(pkt_end, ifp); 1537 return; 1538 } 1539 1540 /* 1541 * Finally dispatch it 1542 */ 1543 ether_dispatch(ifp, isr, m, cpuid); 1544 1545 logether(pkt_end, ifp); 1546 } 1547 1548 static int 1549 ether_characterize(struct mbuf **m0) 1550 { 1551 struct mbuf *m = *m0; 1552 struct ether_header *eh; 1553 uint16_t ether_type; 1554 int isr; 1555 1556 eh = mtod(m, struct ether_header *); 1557 ether_type = ntohs(eh->ether_type); 1558 1559 /* 1560 * Map ether type to netisr id. 1561 */ 1562 switch (ether_type) { 1563 #ifdef INET 1564 case ETHERTYPE_IP: 1565 isr = NETISR_IP; 1566 break; 1567 1568 case ETHERTYPE_ARP: 1569 isr = NETISR_ARP; 1570 break; 1571 #endif 1572 1573 #ifdef INET6 1574 case ETHERTYPE_IPV6: 1575 isr = NETISR_IPV6; 1576 break; 1577 #endif 1578 1579 #ifdef MPLS 1580 case ETHERTYPE_MPLS: 1581 case ETHERTYPE_MPLS_MCAST: 1582 m->m_flags |= M_MPLSLABELED; 1583 isr = NETISR_MPLS; 1584 break; 1585 #endif 1586 1587 default: 1588 /* 1589 * NETISR_MAX is an invalid value; it is chosen to let 1590 * netisr_characterize() know that we have no clear 1591 * idea where this packet should go. 1592 */ 1593 isr = NETISR_MAX; 1594 break; 1595 } 1596 1597 /* 1598 * Ask the isr to characterize the packet since we couldn't. 1599 * This is an attempt to optimally get us onto the correct protocol 1600 * thread. 1601 */ 1602 netisr_characterize(isr, &m, sizeof(struct ether_header)); 1603 1604 *m0 = m; 1605 return isr; 1606 } 1607 1608 static void 1609 ether_demux_handler(netmsg_t nmsg) 1610 { 1611 struct netmsg_packet *nmp = &nmsg->packet; /* actual size */ 1612 struct ifnet *ifp; 1613 struct mbuf *m; 1614 1615 m = nmp->nm_packet; 1616 M_ASSERTPKTHDR(m); 1617 ifp = m->m_pkthdr.rcvif; 1618 1619 ether_demux_oncpu(ifp, m); 1620 } 1621 1622 void 1623 ether_demux(struct mbuf *m) 1624 { 1625 struct netmsg_packet *pmsg; 1626 int isr; 1627 1628 isr = ether_characterize(&m); 1629 if (m == NULL) 1630 return; 1631 1632 KKASSERT(m->m_flags & M_HASH); 1633 pmsg = &m->m_hdr.mh_netmsg; 1634 netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport, 1635 0, ether_demux_handler); 1636 pmsg->nm_packet = m; 1637 pmsg->base.lmsg.u.ms_result = isr; 1638 1639 lwkt_sendmsg(netisr_hashport(m->m_pkthdr.hash), &pmsg->base.lmsg); 1640 } 1641 1642 u_char * 1643 kether_aton(const char *macstr, u_char *addr) 1644 { 1645 unsigned int o0, o1, o2, o3, o4, o5; 1646 int n; 1647 1648 if (macstr == NULL || addr == NULL) 1649 return NULL; 1650 1651 n = ksscanf(macstr, "%x:%x:%x:%x:%x:%x", &o0, &o1, &o2, 1652 &o3, &o4, &o5); 1653 if (n != 6) 1654 return NULL; 1655 1656 addr[0] = o0; 1657 addr[1] = o1; 1658 addr[2] = o2; 1659 addr[3] = o3; 1660 addr[4] = o4; 1661 addr[5] = o5; 1662 1663 return addr; 1664 } 1665 1666 char * 1667 kether_ntoa(const u_char *addr, char *buf) 1668 { 1669 int len = ETHER_ADDRSTRLEN + 1; 1670 int n; 1671 1672 n = ksnprintf(buf, len, "%02x:%02x:%02x:%02x:%02x:%02x", addr[0], 1673 addr[1], addr[2], addr[3], addr[4], addr[5]); 1674 1675 if (n < 17) 1676 return NULL; 1677 1678 return buf; 1679 } 1680 1681 MODULE_VERSION(ether, 1); 1682