linux/net/ipv4/ip_input.c
<<
>>
Prefs
   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 */
 148bool 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
 185INDIRECT_CALLABLE_DECLARE(int udp_rcv(struct sk_buff *));
 186INDIRECT_CALLABLE_DECLARE(int tcp_v4_rcv(struct sk_buff *));
 187void 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
 192resubmit:
 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
 227static int ip_local_deliver_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 228{
 229        skb_clear_delivery_time(skb);
 230        __skb_pull(skb, skb_network_header_len(skb));
 231
 232        rcu_read_lock();
 233        ip_protocol_deliver_rcu(net, skb, ip_hdr(skb)->protocol);
 234        rcu_read_unlock();
 235
 236        return 0;
 237}
 238
 239/*
 240 *      Deliver IP Packets to the higher protocol layers.
 241 */
 242int ip_local_deliver(struct sk_buff *skb)
 243{
 244        /*
 245         *      Reassemble IP fragments.
 246         */
 247        struct net *net = dev_net(skb->dev);
 248
 249        if (ip_is_fragment(ip_hdr(skb))) {
 250                if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER))
 251                        return 0;
 252        }
 253
 254        return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN,
 255                       net, NULL, skb, skb->dev, NULL,
 256                       ip_local_deliver_finish);
 257}
 258EXPORT_SYMBOL(ip_local_deliver);
 259
 260static inline bool ip_rcv_options(struct sk_buff *skb, struct net_device *dev)
 261{
 262        struct ip_options *opt;
 263        const struct iphdr *iph;
 264
 265        /* It looks as overkill, because not all
 266           IP options require packet mangling.
 267           But it is the easiest for now, especially taking
 268           into account that combination of IP options
 269           and running sniffer is extremely rare condition.
 270                                              --ANK (980813)
 271        */
 272        if (skb_cow(skb, skb_headroom(skb))) {
 273                __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 274                goto drop;
 275        }
 276
 277        iph = ip_hdr(skb);
 278        opt = &(IPCB(skb)->opt);
 279        opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
 280
 281        if (ip_options_compile(dev_net(dev), opt, skb)) {
 282                __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 283                goto drop;
 284        }
 285
 286        if (unlikely(opt->srr)) {
 287                struct in_device *in_dev = __in_dev_get_rcu(dev);
 288
 289                if (in_dev) {
 290                        if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
 291                                if (IN_DEV_LOG_MARTIANS(in_dev))
 292                                        net_info_ratelimited("source route option %pI4 -> %pI4\n",
 293                                                             &iph->saddr,
 294                                                             &iph->daddr);
 295                                goto drop;
 296                        }
 297                }
 298
 299                if (ip_options_rcv_srr(skb, dev))
 300                        goto drop;
 301        }
 302
 303        return false;
 304drop:
 305        return true;
 306}
 307
 308static bool ip_can_use_hint(const struct sk_buff *skb, const struct iphdr *iph,
 309                            const struct sk_buff *hint)
 310{
 311        return hint && !skb_dst(skb) && ip_hdr(hint)->daddr == iph->daddr &&
 312               ip_hdr(hint)->tos == iph->tos;
 313}
 314
 315INDIRECT_CALLABLE_DECLARE(int udp_v4_early_demux(struct sk_buff *));
 316INDIRECT_CALLABLE_DECLARE(int tcp_v4_early_demux(struct sk_buff *));
 317static int ip_rcv_finish_core(struct net *net, struct sock *sk,
 318                              struct sk_buff *skb, struct net_device *dev,
 319                              const struct sk_buff *hint)
 320{
 321        const struct iphdr *iph = ip_hdr(skb);
 322        int (*edemux)(struct sk_buff *skb);
 323        int err, drop_reason;
 324        struct rtable *rt;
 325
 326        drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
 327
 328        if (ip_can_use_hint(skb, iph, hint)) {
 329                err = ip_route_use_hint(skb, iph->daddr, iph->saddr, iph->tos,
 330                                        dev, hint);
 331                if (unlikely(err))
 332                        goto drop_error;
 333        }
 334
 335        if (net->ipv4.sysctl_ip_early_demux &&
 336            !skb_dst(skb) &&
 337            !skb->sk &&
 338            !ip_is_fragment(iph)) {
 339                const struct net_protocol *ipprot;
 340                int protocol = iph->protocol;
 341
 342                ipprot = rcu_dereference(inet_protos[protocol]);
 343                if (ipprot && (edemux = READ_ONCE(ipprot->early_demux))) {
 344                        err = INDIRECT_CALL_2(edemux, tcp_v4_early_demux,
 345                                              udp_v4_early_demux, skb);
 346                        if (unlikely(err))
 347                                goto drop_error;
 348                        /* must reload iph, skb->head might have changed */
 349                        iph = ip_hdr(skb);
 350                }
 351        }
 352
 353        /*
 354         *      Initialise the virtual path cache for the packet. It describes
 355         *      how the packet travels inside Linux networking.
 356         */
 357        if (!skb_valid_dst(skb)) {
 358                err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
 359                                           iph->tos, dev);
 360                if (unlikely(err))
 361                        goto drop_error;
 362        }
 363
 364#ifdef CONFIG_IP_ROUTE_CLASSID
 365        if (unlikely(skb_dst(skb)->tclassid)) {
 366                struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
 367                u32 idx = skb_dst(skb)->tclassid;
 368                st[idx&0xFF].o_packets++;
 369                st[idx&0xFF].o_bytes += skb->len;
 370                st[(idx>>16)&0xFF].i_packets++;
 371                st[(idx>>16)&0xFF].i_bytes += skb->len;
 372        }
 373#endif
 374
 375        if (iph->ihl > 5 && ip_rcv_options(skb, dev))
 376                goto drop;
 377
 378        rt = skb_rtable(skb);
 379        if (rt->rt_type == RTN_MULTICAST) {
 380                __IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len);
 381        } else if (rt->rt_type == RTN_BROADCAST) {
 382                __IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len);
 383        } else if (skb->pkt_type == PACKET_BROADCAST ||
 384                   skb->pkt_type == PACKET_MULTICAST) {
 385                struct in_device *in_dev = __in_dev_get_rcu(dev);
 386
 387                /* RFC 1122 3.3.6:
 388                 *
 389                 *   When a host sends a datagram to a link-layer broadcast
 390                 *   address, the IP destination address MUST be a legal IP
 391                 *   broadcast or IP multicast address.
 392                 *
 393                 *   A host SHOULD silently discard a datagram that is received
 394                 *   via a link-layer broadcast (see Section 2.4) but does not
 395                 *   specify an IP multicast or broadcast destination address.
 396                 *
 397                 * This doesn't explicitly say L2 *broadcast*, but broadcast is
 398                 * in a way a form of multicast and the most common use case for
 399                 * this is 802.11 protecting against cross-station spoofing (the
 400                 * so-called "hole-196" attack) so do it for both.
 401                 */
 402                if (in_dev &&
 403                    IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST)) {
 404                        drop_reason = SKB_DROP_REASON_UNICAST_IN_L2_MULTICAST;
 405                        goto drop;
 406                }
 407        }
 408
 409        return NET_RX_SUCCESS;
 410
 411drop:
 412        kfree_skb_reason(skb, drop_reason);
 413        return NET_RX_DROP;
 414
 415drop_error:
 416        if (err == -EXDEV) {
 417                drop_reason = SKB_DROP_REASON_IP_RPFILTER;
 418                __NET_INC_STATS(net, LINUX_MIB_IPRPFILTER);
 419        }
 420        goto drop;
 421}
 422
 423static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 424{
 425        struct net_device *dev = skb->dev;
 426        int ret;
 427
 428        /* if ingress device is enslaved to an L3 master device pass the
 429         * skb to its handler for processing
 430         */
 431        skb = l3mdev_ip_rcv(skb);
 432        if (!skb)
 433                return NET_RX_SUCCESS;
 434
 435        ret = ip_rcv_finish_core(net, sk, skb, dev, NULL);
 436        if (ret != NET_RX_DROP)
 437                ret = dst_input(skb);
 438        return ret;
 439}
 440
 441/*
 442 *      Main IP Receive routine.
 443 */
 444static struct sk_buff *ip_rcv_core(struct sk_buff *skb, struct net *net)
 445{
 446        const struct iphdr *iph;
 447        int drop_reason;
 448        u32 len;
 449
 450        /* When the interface is in promisc. mode, drop all the crap
 451         * that it receives, do not try to analyse it.
 452         */
 453        if (skb->pkt_type == PACKET_OTHERHOST) {
 454                drop_reason = SKB_DROP_REASON_OTHERHOST;
 455                goto drop;
 456        }
 457
 458        __IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len);
 459
 460        skb = skb_share_check(skb, GFP_ATOMIC);
 461        if (!skb) {
 462                __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
 463                goto out;
 464        }
 465
 466        drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
 467        if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 468                goto inhdr_error;
 469
 470        iph = ip_hdr(skb);
 471
 472        /*
 473         *      RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
 474         *
 475         *      Is the datagram acceptable?
 476         *
 477         *      1.      Length at least the size of an ip header
 478         *      2.      Version of 4
 479         *      3.      Checksums correctly. [Speed optimisation for later, skip loopback checksums]
 480         *      4.      Doesn't have a bogus length
 481         */
 482
 483        if (iph->ihl < 5 || iph->version != 4)
 484                goto inhdr_error;
 485
 486        BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
 487        BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
 488        BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
 489        __IP_ADD_STATS(net,
 490                       IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
 491                       max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
 492
 493        if (!pskb_may_pull(skb, iph->ihl*4))
 494                goto inhdr_error;
 495
 496        iph = ip_hdr(skb);
 497
 498        if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 499                goto csum_error;
 500
 501        len = ntohs(iph->tot_len);
 502        if (skb->len < len) {
 503                drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
 504                __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
 505                goto drop;
 506        } else if (len < (iph->ihl*4))
 507                goto inhdr_error;
 508
 509        /* Our transport medium may have padded the buffer out. Now we know it
 510         * is IP we can trim to the true length of the frame.
 511         * Note this now means skb->len holds ntohs(iph->tot_len).
 512         */
 513        if (pskb_trim_rcsum(skb, len)) {
 514                __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
 515                goto drop;
 516        }
 517
 518        iph = ip_hdr(skb);
 519        skb->transport_header = skb->network_header + iph->ihl*4;
 520
 521        /* Remove any debris in the socket control block */
 522        memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 523        IPCB(skb)->iif = skb->skb_iif;
 524
 525        /* Must drop socket now because of tproxy. */
 526        if (!skb_sk_is_prefetched(skb))
 527                skb_orphan(skb);
 528
 529        return skb;
 530
 531csum_error:
 532        drop_reason = SKB_DROP_REASON_IP_CSUM;
 533        __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
 534inhdr_error:
 535        if (drop_reason == SKB_DROP_REASON_NOT_SPECIFIED)
 536                drop_reason = SKB_DROP_REASON_IP_INHDR;
 537        __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
 538drop:
 539        kfree_skb_reason(skb, drop_reason);
 540out:
 541        return NULL;
 542}
 543
 544/*
 545 * IP receive entry point
 546 */
 547int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
 548           struct net_device *orig_dev)
 549{
 550        struct net *net = dev_net(dev);
 551
 552        skb = ip_rcv_core(skb, net);
 553        if (skb == NULL)
 554                return NET_RX_DROP;
 555
 556        return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING,
 557                       net, NULL, skb, dev, NULL,
 558                       ip_rcv_finish);
 559}
 560
 561static void ip_sublist_rcv_finish(struct list_head *head)
 562{
 563        struct sk_buff *skb, *next;
 564
 565        list_for_each_entry_safe(skb, next, head, list) {
 566                skb_list_del_init(skb);
 567                dst_input(skb);
 568        }
 569}
 570
 571static struct sk_buff *ip_extract_route_hint(const struct net *net,
 572                                             struct sk_buff *skb, int rt_type)
 573{
 574        if (fib4_has_custom_rules(net) || rt_type == RTN_BROADCAST)
 575                return NULL;
 576
 577        return skb;
 578}
 579
 580static void ip_list_rcv_finish(struct net *net, struct sock *sk,
 581                               struct list_head *head)
 582{
 583        struct sk_buff *skb, *next, *hint = NULL;
 584        struct dst_entry *curr_dst = NULL;
 585        struct list_head sublist;
 586
 587        INIT_LIST_HEAD(&sublist);
 588        list_for_each_entry_safe(skb, next, head, list) {
 589                struct net_device *dev = skb->dev;
 590                struct dst_entry *dst;
 591
 592                skb_list_del_init(skb);
 593                /* if ingress device is enslaved to an L3 master device pass the
 594                 * skb to its handler for processing
 595                 */
 596                skb = l3mdev_ip_rcv(skb);
 597                if (!skb)
 598                        continue;
 599                if (ip_rcv_finish_core(net, sk, skb, dev, hint) == NET_RX_DROP)
 600                        continue;
 601
 602                dst = skb_dst(skb);
 603                if (curr_dst != dst) {
 604                        hint = ip_extract_route_hint(net, skb,
 605                                               ((struct rtable *)dst)->rt_type);
 606
 607                        /* dispatch old sublist */
 608                        if (!list_empty(&sublist))
 609                                ip_sublist_rcv_finish(&sublist);
 610                        /* start new sublist */
 611                        INIT_LIST_HEAD(&sublist);
 612                        curr_dst = dst;
 613                }
 614                list_add_tail(&skb->list, &sublist);
 615        }
 616        /* dispatch final sublist */
 617        ip_sublist_rcv_finish(&sublist);
 618}
 619
 620static void ip_sublist_rcv(struct list_head *head, struct net_device *dev,
 621                           struct net *net)
 622{
 623        NF_HOOK_LIST(NFPROTO_IPV4, NF_INET_PRE_ROUTING, net, NULL,
 624                     head, dev, NULL, ip_rcv_finish);
 625        ip_list_rcv_finish(net, NULL, head);
 626}
 627
 628/* Receive a list of IP packets */
 629void ip_list_rcv(struct list_head *head, struct packet_type *pt,
 630                 struct net_device *orig_dev)
 631{
 632        struct net_device *curr_dev = NULL;
 633        struct net *curr_net = NULL;
 634        struct sk_buff *skb, *next;
 635        struct list_head sublist;
 636
 637        INIT_LIST_HEAD(&sublist);
 638        list_for_each_entry_safe(skb, next, head, list) {
 639                struct net_device *dev = skb->dev;
 640                struct net *net = dev_net(dev);
 641
 642                skb_list_del_init(skb);
 643                skb = ip_rcv_core(skb, net);
 644                if (skb == NULL)
 645                        continue;
 646
 647                if (curr_dev != dev || curr_net != net) {
 648                        /* dispatch old sublist */
 649                        if (!list_empty(&sublist))
 650                                ip_sublist_rcv(&sublist, curr_dev, curr_net);
 651                        /* start new sublist */
 652                        INIT_LIST_HEAD(&sublist);
 653                        curr_dev = dev;
 654                        curr_net = net;
 655                }
 656                list_add_tail(&skb->list, &sublist);
 657        }
 658        /* dispatch final sublist */
 659        if (!list_empty(&sublist))
 660                ip_sublist_rcv(&sublist, curr_dev, curr_net);
 661}
 662