1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * The Internet Protocol (IP) module. 8 * 9 * Authors: Ross Biro 10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 11 * Donald Becker, <becker@super.org> 12 * Alan Cox, <alan@lxorguk.ukuu.org.uk> 13 * Richard Underwood 14 * Stefan Becker, <stefanb@yello.ping.de> 15 * Jorge Cwik, <jorge@laser.satlink.net> 16 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 17 * 18 * Fixes: 19 * Alan Cox : Commented a couple of minor bits of surplus code 20 * Alan Cox : Undefining IP_FORWARD doesn't include the code 21 * (just stops a compiler warning). 22 * Alan Cox : Frames with >=MAX_ROUTE record routes, strict routes or loose routes 23 * are junked rather than corrupting things. 24 * Alan Cox : Frames to bad broadcast subnets are dumped 25 * We used to process them non broadcast and 26 * boy could that cause havoc. 27 * Alan Cox : ip_forward sets the free flag on the 28 * new frame it queues. Still crap because 29 * it copies the frame but at least it 30 * doesn't eat memory too. 31 * Alan Cox : Generic queue code and memory fixes. 32 * Fred Van Kempen : IP fragment support (borrowed from NET2E) 33 * Gerhard Koerting: Forward fragmented frames correctly. 34 * Gerhard Koerting: Fixes to my fix of the above 8-). 35 * Gerhard Koerting: IP interface addressing fix. 36 * Linus Torvalds : More robustness checks 37 * Alan Cox : Even more checks: Still not as robust as it ought to be 38 * Alan Cox : Save IP header pointer for later 39 * Alan Cox : ip option setting 40 * Alan Cox : Use ip_tos/ip_ttl settings 41 * Alan Cox : Fragmentation bogosity removed 42 * (Thanks to Mark.Bush@prg.ox.ac.uk) 43 * Dmitry Gorodchanin : Send of a raw packet crash fix. 44 * Alan Cox : Silly ip bug when an overlength 45 * fragment turns up. Now frees the 46 * queue. 47 * Linus Torvalds/ : Memory leakage on fragmentation 48 * Alan Cox : handling. 49 * Gerhard Koerting: Forwarding uses IP priority hints 50 * Teemu Rantanen : Fragment problems. 51 * Alan Cox : General cleanup, comments and reformat 52 * Alan Cox : SNMP statistics 53 * Alan Cox : BSD address rule semantics. Also see 54 * UDP as there is a nasty checksum issue 55 * if you do things the wrong way. 56 * Alan Cox : Always defrag, moved IP_FORWARD to the config.in file 57 * Alan Cox : IP options adjust sk->priority. 58 * Pedro Roque : Fix mtu/length error in ip_forward. 59 * Alan Cox : Avoid ip_chk_addr when possible. 60 * Richard Underwood : IP multicasting. 61 * Alan Cox : Cleaned up multicast handlers. 62 * Alan Cox : RAW sockets demultiplex in the BSD style. 63 * Gunther Mayer : Fix the SNMP reporting typo 64 * Alan Cox : Always in group 224.0.0.1 65 * Pauline Middelink : Fast ip_checksum update when forwarding 66 * Masquerading support. 67 * Alan Cox : Multicast loopback error for 224.0.0.1 68 * Alan Cox : IP_MULTICAST_LOOP option. 69 * Alan Cox : Use notifiers. 70 * Bjorn Ekwall : Removed ip_csum (from slhc.c too) 71 * Bjorn Ekwall : Moved ip_fast_csum to ip.h (inline!) 72 * Stefan Becker : Send out ICMP HOST REDIRECT 73 * Arnt Gulbrandsen : ip_build_xmit 74 * Alan Cox : Per socket routing cache 75 * Alan Cox : Fixed routing cache, added header cache. 76 * Alan Cox : Loopback didn't work right in original ip_build_xmit - fixed it. 77 * Alan Cox : Only send ICMP_REDIRECT if src/dest are the same net. 78 * Alan Cox : Incoming IP option handling. 79 * Alan Cox : Set saddr on raw output frames as per BSD. 80 * Alan Cox : Stopped broadcast source route explosions. 81 * Alan Cox : Can disable source routing 82 * Takeshi Sone : Masquerading didn't work. 83 * Dave Bonn,Alan Cox : Faster IP forwarding whenever possible. 84 * Alan Cox : Memory leaks, tramples, misc debugging. 85 * Alan Cox : Fixed multicast (by popular demand 8)) 86 * Alan Cox : Fixed forwarding (by even more popular demand 8)) 87 * Alan Cox : Fixed SNMP statistics [I think] 88 * Gerhard Koerting : IP fragmentation forwarding fix 89 * Alan Cox : Device lock against page fault. 90 * Alan Cox : IP_HDRINCL facility. 91 * Werner Almesberger : Zero fragment bug 92 * Alan Cox : RAW IP frame length bug 93 * Alan Cox : Outgoing firewall on build_xmit 94 * A.N.Kuznetsov : IP_OPTIONS support throughout the kernel 95 * Alan Cox : Multicast routing hooks 96 * Jos Vos : Do accounting *before* call_in_firewall 97 * Willy Konynenberg : Transparent proxying support 98 * 99 * To Fix: 100 * IP fragmentation wants rewriting cleanly. The RFC815 algorithm is much more efficient 101 * and could be made very efficient with the addition of some virtual memory hacks to permit 102 * the allocation of a buffer that can then be 'grown' by twiddling page tables. 103 * Output fragmentation wants updating along with the buffer management to use a single 104 * interleaved copy algorithm so that fragmenting has a one copy overhead. Actual packet 105 * output should probably do its own fragmentation at the UDP/RAW layer. TCP shouldn't cause 106 * fragmentation anyway. 107 */ 108 109 #define pr_fmt(fmt) "IPv4: " fmt 110 111 #include <linux/module.h> 112 #include <linux/types.h> 113 #include <linux/kernel.h> 114 #include <linux/string.h> 115 #include <linux/errno.h> 116 #include <linux/slab.h> 117 118 #include <linux/net.h> 119 #include <linux/socket.h> 120 #include <linux/sockios.h> 121 #include <linux/in.h> 122 #include <linux/inet.h> 123 #include <linux/inetdevice.h> 124 #include <linux/netdevice.h> 125 #include <linux/etherdevice.h> 126 #include <linux/indirect_call_wrapper.h> 127 128 #include <net/snmp.h> 129 #include <net/ip.h> 130 #include <net/protocol.h> 131 #include <net/route.h> 132 #include <linux/skbuff.h> 133 #include <net/sock.h> 134 #include <net/arp.h> 135 #include <net/icmp.h> 136 #include <net/raw.h> 137 #include <net/checksum.h> 138 #include <net/inet_ecn.h> 139 #include <linux/netfilter_ipv4.h> 140 #include <net/xfrm.h> 141 #include <linux/mroute.h> 142 #include <linux/netlink.h> 143 #include <net/dst_metadata.h> 144 145 /* 146 * Process Router Attention IP option (RFC 2113) 147 */ 148 bool ip_call_ra_chain(struct sk_buff *skb) 149 { 150 struct ip_ra_chain *ra; 151 u8 protocol = ip_hdr(skb)->protocol; 152 struct sock *last = NULL; 153 struct net_device *dev = skb->dev; 154 struct net *net = dev_net(dev); 155 156 for (ra = rcu_dereference(net->ipv4.ra_chain); ra; ra = rcu_dereference(ra->next)) { 157 struct sock *sk = ra->sk; 158 159 /* If socket is bound to an interface, only report 160 * the packet if it came from that interface. 161 */ 162 if (sk && inet_sk(sk)->inet_num == protocol && 163 (!sk->sk_bound_dev_if || 164 sk->sk_bound_dev_if == dev->ifindex)) { 165 if (ip_is_fragment(ip_hdr(skb))) { 166 if (ip_defrag(net, skb, IP_DEFRAG_CALL_RA_CHAIN)) 167 return true; 168 } 169 if (last) { 170 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); 171 if (skb2) 172 raw_rcv(last, skb2); 173 } 174 last = sk; 175 } 176 } 177 178 if (last) { 179 raw_rcv(last, skb); 180 return true; 181 } 182 return false; 183 } 184 185 INDIRECT_CALLABLE_DECLARE(int udp_rcv(struct sk_buff *)); 186 INDIRECT_CALLABLE_DECLARE(int tcp_v4_rcv(struct sk_buff *)); 187 void ip_protocol_deliver_rcu(struct net *net, struct sk_buff *skb, int protocol) 188 { 189 const struct net_protocol *ipprot; 190 int raw, ret; 191 192 resubmit: 193 raw = raw_local_deliver(skb, protocol); 194 195 ipprot = rcu_dereference(inet_protos[protocol]); 196 if (ipprot) { 197 if (!ipprot->no_policy) { 198 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 199 kfree_skb_reason(skb, 200 SKB_DROP_REASON_XFRM_POLICY); 201 return; 202 } 203 nf_reset_ct(skb); 204 } 205 ret = INDIRECT_CALL_2(ipprot->handler, tcp_v4_rcv, udp_rcv, 206 skb); 207 if (ret < 0) { 208 protocol = -ret; 209 goto resubmit; 210 } 211 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS); 212 } else { 213 if (!raw) { 214 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 215 __IP_INC_STATS(net, IPSTATS_MIB_INUNKNOWNPROTOS); 216 icmp_send(skb, ICMP_DEST_UNREACH, 217 ICMP_PROT_UNREACH, 0); 218 } 219 kfree_skb_reason(skb, SKB_DROP_REASON_IP_NOPROTO); 220 } else { 221 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS); 222 consume_skb(skb); 223 } 224 } 225 } 226 227 static int ip_local_deliver_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 228 { 229 __skb_pull(skb, skb_network_header_len(skb)); 230 231 rcu_read_lock(); 232 ip_protocol_deliver_rcu(net, skb, ip_hdr(skb)->protocol); 233 rcu_read_unlock(); 234 235 return 0; 236 } 237 238 /* 239 * Deliver IP Packets to the higher protocol layers. 240 */ 241 int ip_local_deliver(struct sk_buff *skb) 242 { 243 /* 244 * Reassemble IP fragments. 245 */ 246 struct net *net = dev_net(skb->dev); 247 248 if (ip_is_fragment(ip_hdr(skb))) { 249 if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER)) 250 return 0; 251 } 252 253 return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN, 254 net, NULL, skb, skb->dev, NULL, 255 ip_local_deliver_finish); 256 } 257 EXPORT_SYMBOL(ip_local_deliver); 258 259 static inline bool ip_rcv_options(struct sk_buff *skb, struct net_device *dev) 260 { 261 struct ip_options *opt; 262 const struct iphdr *iph; 263 264 /* It looks as overkill, because not all 265 IP options require packet mangling. 266 But it is the easiest for now, especially taking 267 into account that combination of IP options 268 and running sniffer is extremely rare condition. 269 --ANK (980813) 270 */ 271 if (skb_cow(skb, skb_headroom(skb))) { 272 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS); 273 goto drop; 274 } 275 276 iph = ip_hdr(skb); 277 opt = &(IPCB(skb)->opt); 278 opt->optlen = iph->ihl*4 - sizeof(struct iphdr); 279 280 if (ip_options_compile(dev_net(dev), opt, skb)) { 281 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS); 282 goto drop; 283 } 284 285 if (unlikely(opt->srr)) { 286 struct in_device *in_dev = __in_dev_get_rcu(dev); 287 288 if (in_dev) { 289 if (!IN_DEV_SOURCE_ROUTE(in_dev)) { 290 if (IN_DEV_LOG_MARTIANS(in_dev)) 291 net_info_ratelimited("source route option %pI4 -> %pI4\n", 292 &iph->saddr, 293 &iph->daddr); 294 goto drop; 295 } 296 } 297 298 if (ip_options_rcv_srr(skb, dev)) 299 goto drop; 300 } 301 302 return false; 303 drop: 304 return true; 305 } 306 307 static bool ip_can_use_hint(const struct sk_buff *skb, const struct iphdr *iph, 308 const struct sk_buff *hint) 309 { 310 return hint && !skb_dst(skb) && ip_hdr(hint)->daddr == iph->daddr && 311 ip_hdr(hint)->tos == iph->tos; 312 } 313 314 INDIRECT_CALLABLE_DECLARE(int udp_v4_early_demux(struct sk_buff *)); 315 INDIRECT_CALLABLE_DECLARE(int tcp_v4_early_demux(struct sk_buff *)); 316 static int ip_rcv_finish_core(struct net *net, struct sock *sk, 317 struct sk_buff *skb, struct net_device *dev, 318 const struct sk_buff *hint) 319 { 320 const struct iphdr *iph = ip_hdr(skb); 321 int (*edemux)(struct sk_buff *skb); 322 int err, drop_reason; 323 struct rtable *rt; 324 325 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 326 327 if (ip_can_use_hint(skb, iph, hint)) { 328 err = ip_route_use_hint(skb, iph->daddr, iph->saddr, iph->tos, 329 dev, hint); 330 if (unlikely(err)) 331 goto drop_error; 332 } 333 334 if (net->ipv4.sysctl_ip_early_demux && 335 !skb_dst(skb) && 336 !skb->sk && 337 !ip_is_fragment(iph)) { 338 const struct net_protocol *ipprot; 339 int protocol = iph->protocol; 340 341 ipprot = rcu_dereference(inet_protos[protocol]); 342 if (ipprot && (edemux = READ_ONCE(ipprot->early_demux))) { 343 err = INDIRECT_CALL_2(edemux, tcp_v4_early_demux, 344 udp_v4_early_demux, skb); 345 if (unlikely(err)) 346 goto drop_error; 347 /* must reload iph, skb->head might have changed */ 348 iph = ip_hdr(skb); 349 } 350 } 351 352 /* 353 * Initialise the virtual path cache for the packet. It describes 354 * how the packet travels inside Linux networking. 355 */ 356 if (!skb_valid_dst(skb)) { 357 err = ip_route_input_noref(skb, iph->daddr, iph->saddr, 358 iph->tos, dev); 359 if (unlikely(err)) 360 goto drop_error; 361 } 362 363 #ifdef CONFIG_IP_ROUTE_CLASSID 364 if (unlikely(skb_dst(skb)->tclassid)) { 365 struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct); 366 u32 idx = skb_dst(skb)->tclassid; 367 st[idx&0xFF].o_packets++; 368 st[idx&0xFF].o_bytes += skb->len; 369 st[(idx>>16)&0xFF].i_packets++; 370 st[(idx>>16)&0xFF].i_bytes += skb->len; 371 } 372 #endif 373 374 if (iph->ihl > 5 && ip_rcv_options(skb, dev)) 375 goto drop; 376 377 rt = skb_rtable(skb); 378 if (rt->rt_type == RTN_MULTICAST) { 379 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len); 380 } else if (rt->rt_type == RTN_BROADCAST) { 381 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len); 382 } else if (skb->pkt_type == PACKET_BROADCAST || 383 skb->pkt_type == PACKET_MULTICAST) { 384 struct in_device *in_dev = __in_dev_get_rcu(dev); 385 386 /* RFC 1122 3.3.6: 387 * 388 * When a host sends a datagram to a link-layer broadcast 389 * address, the IP destination address MUST be a legal IP 390 * broadcast or IP multicast address. 391 * 392 * A host SHOULD silently discard a datagram that is received 393 * via a link-layer broadcast (see Section 2.4) but does not 394 * specify an IP multicast or broadcast destination address. 395 * 396 * This doesn't explicitly say L2 *broadcast*, but broadcast is 397 * in a way a form of multicast and the most common use case for 398 * this is 802.11 protecting against cross-station spoofing (the 399 * so-called "hole-196" attack) so do it for both. 400 */ 401 if (in_dev && 402 IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST)) { 403 drop_reason = SKB_DROP_REASON_UNICAST_IN_L2_MULTICAST; 404 goto drop; 405 } 406 } 407 408 return NET_RX_SUCCESS; 409 410 drop: 411 kfree_skb_reason(skb, drop_reason); 412 return NET_RX_DROP; 413 414 drop_error: 415 if (err == -EXDEV) { 416 drop_reason = SKB_DROP_REASON_IP_RPFILTER; 417 __NET_INC_STATS(net, LINUX_MIB_IPRPFILTER); 418 } 419 goto drop; 420 } 421 422 static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 423 { 424 struct net_device *dev = skb->dev; 425 int ret; 426 427 /* if ingress device is enslaved to an L3 master device pass the 428 * skb to its handler for processing 429 */ 430 skb = l3mdev_ip_rcv(skb); 431 if (!skb) 432 return NET_RX_SUCCESS; 433 434 ret = ip_rcv_finish_core(net, sk, skb, dev, NULL); 435 if (ret != NET_RX_DROP) 436 ret = dst_input(skb); 437 return ret; 438 } 439 440 /* 441 * Main IP Receive routine. 442 */ 443 static struct sk_buff *ip_rcv_core(struct sk_buff *skb, struct net *net) 444 { 445 const struct iphdr *iph; 446 int drop_reason; 447 u32 len; 448 449 /* When the interface is in promisc. mode, drop all the crap 450 * that it receives, do not try to analyse it. 451 */ 452 if (skb->pkt_type == PACKET_OTHERHOST) { 453 drop_reason = SKB_DROP_REASON_OTHERHOST; 454 goto drop; 455 } 456 457 __IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len); 458 459 skb = skb_share_check(skb, GFP_ATOMIC); 460 if (!skb) { 461 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS); 462 goto out; 463 } 464 465 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 466 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 467 goto inhdr_error; 468 469 iph = ip_hdr(skb); 470 471 /* 472 * RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum. 473 * 474 * Is the datagram acceptable? 475 * 476 * 1. Length at least the size of an ip header 477 * 2. Version of 4 478 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums] 479 * 4. Doesn't have a bogus length 480 */ 481 482 if (iph->ihl < 5 || iph->version != 4) 483 goto inhdr_error; 484 485 BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1); 486 BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0); 487 BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE); 488 __IP_ADD_STATS(net, 489 IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK), 490 max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs)); 491 492 if (!pskb_may_pull(skb, iph->ihl*4)) 493 goto inhdr_error; 494 495 iph = ip_hdr(skb); 496 497 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl))) 498 goto csum_error; 499 500 len = ntohs(iph->tot_len); 501 if (skb->len < len) { 502 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; 503 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS); 504 goto drop; 505 } else if (len < (iph->ihl*4)) 506 goto inhdr_error; 507 508 /* Our transport medium may have padded the buffer out. Now we know it 509 * is IP we can trim to the true length of the frame. 510 * Note this now means skb->len holds ntohs(iph->tot_len). 511 */ 512 if (pskb_trim_rcsum(skb, len)) { 513 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS); 514 goto drop; 515 } 516 517 iph = ip_hdr(skb); 518 skb->transport_header = skb->network_header + iph->ihl*4; 519 520 /* Remove any debris in the socket control block */ 521 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); 522 IPCB(skb)->iif = skb->skb_iif; 523 524 /* Must drop socket now because of tproxy. */ 525 if (!skb_sk_is_prefetched(skb)) 526 skb_orphan(skb); 527 528 return skb; 529 530 csum_error: 531 drop_reason = SKB_DROP_REASON_IP_CSUM; 532 __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS); 533 inhdr_error: 534 if (drop_reason == SKB_DROP_REASON_NOT_SPECIFIED) 535 drop_reason = SKB_DROP_REASON_IP_INHDR; 536 __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS); 537 drop: 538 kfree_skb_reason(skb, drop_reason); 539 out: 540 return NULL; 541 } 542 543 /* 544 * IP receive entry point 545 */ 546 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, 547 struct net_device *orig_dev) 548 { 549 struct net *net = dev_net(dev); 550 551 skb = ip_rcv_core(skb, net); 552 if (skb == NULL) 553 return NET_RX_DROP; 554 555 return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, 556 net, NULL, skb, dev, NULL, 557 ip_rcv_finish); 558 } 559 560 static void ip_sublist_rcv_finish(struct list_head *head) 561 { 562 struct sk_buff *skb, *next; 563 564 list_for_each_entry_safe(skb, next, head, list) { 565 skb_list_del_init(skb); 566 dst_input(skb); 567 } 568 } 569 570 static struct sk_buff *ip_extract_route_hint(const struct net *net, 571 struct sk_buff *skb, int rt_type) 572 { 573 if (fib4_has_custom_rules(net) || rt_type == RTN_BROADCAST) 574 return NULL; 575 576 return skb; 577 } 578 579 static void ip_list_rcv_finish(struct net *net, struct sock *sk, 580 struct list_head *head) 581 { 582 struct sk_buff *skb, *next, *hint = NULL; 583 struct dst_entry *curr_dst = NULL; 584 struct list_head sublist; 585 586 INIT_LIST_HEAD(&sublist); 587 list_for_each_entry_safe(skb, next, head, list) { 588 struct net_device *dev = skb->dev; 589 struct dst_entry *dst; 590 591 skb_list_del_init(skb); 592 /* if ingress device is enslaved to an L3 master device pass the 593 * skb to its handler for processing 594 */ 595 skb = l3mdev_ip_rcv(skb); 596 if (!skb) 597 continue; 598 if (ip_rcv_finish_core(net, sk, skb, dev, hint) == NET_RX_DROP) 599 continue; 600 601 dst = skb_dst(skb); 602 if (curr_dst != dst) { 603 hint = ip_extract_route_hint(net, skb, 604 ((struct rtable *)dst)->rt_type); 605 606 /* dispatch old sublist */ 607 if (!list_empty(&sublist)) 608 ip_sublist_rcv_finish(&sublist); 609 /* start new sublist */ 610 INIT_LIST_HEAD(&sublist); 611 curr_dst = dst; 612 } 613 list_add_tail(&skb->list, &sublist); 614 } 615 /* dispatch final sublist */ 616 ip_sublist_rcv_finish(&sublist); 617 } 618 619 static void ip_sublist_rcv(struct list_head *head, struct net_device *dev, 620 struct net *net) 621 { 622 NF_HOOK_LIST(NFPROTO_IPV4, NF_INET_PRE_ROUTING, net, NULL, 623 head, dev, NULL, ip_rcv_finish); 624 ip_list_rcv_finish(net, NULL, head); 625 } 626 627 /* Receive a list of IP packets */ 628 void ip_list_rcv(struct list_head *head, struct packet_type *pt, 629 struct net_device *orig_dev) 630 { 631 struct net_device *curr_dev = NULL; 632 struct net *curr_net = NULL; 633 struct sk_buff *skb, *next; 634 struct list_head sublist; 635 636 INIT_LIST_HEAD(&sublist); 637 list_for_each_entry_safe(skb, next, head, list) { 638 struct net_device *dev = skb->dev; 639 struct net *net = dev_net(dev); 640 641 skb_list_del_init(skb); 642 skb = ip_rcv_core(skb, net); 643 if (skb == NULL) 644 continue; 645 646 if (curr_dev != dev || curr_net != net) { 647 /* dispatch old sublist */ 648 if (!list_empty(&sublist)) 649 ip_sublist_rcv(&sublist, curr_dev, curr_net); 650 /* start new sublist */ 651 INIT_LIST_HEAD(&sublist); 652 curr_dev = dev; 653 curr_net = net; 654 } 655 list_add_tail(&skb->list, &sublist); 656 } 657 /* dispatch final sublist */ 658 if (!list_empty(&sublist)) 659 ip_sublist_rcv(&sublist, curr_dev, curr_net); 660 } 661