net/ipv4/ip_input.c

0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /*
0003  * INET     An implementation of the TCP/IP protocol suite for the LINUX
0004  *      operating system.  INET is implemented using the  BSD Socket
0005  *      interface as the means of communication with the user level.
0006  *
0007  *      The Internet Protocol (IP) module.
0008  *
0009  * Authors: Ross Biro
0010  *      Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
0011  *      Donald Becker, <becker@super.org>
0012  *      Alan Cox, <alan@lxorguk.ukuu.org.uk>
0013  *      Richard Underwood
0014  *      Stefan Becker, <stefanb@yello.ping.de>
0015  *      Jorge Cwik, <jorge@laser.satlink.net>
0016  *      Arnt Gulbrandsen, <agulbra@nvg.unit.no>
0017  *
0018  * Fixes:
0019  *      Alan Cox    :   Commented a couple of minor bits of surplus code
0020  *      Alan Cox    :   Undefining IP_FORWARD doesn't include the code
0021  *                  (just stops a compiler warning).
0022  *      Alan Cox    :   Frames with >=MAX_ROUTE record routes, strict routes or loose routes
0023  *                  are junked rather than corrupting things.
0024  *      Alan Cox    :   Frames to bad broadcast subnets are dumped
0025  *                  We used to process them non broadcast and
0026  *                  boy could that cause havoc.
0027  *      Alan Cox    :   ip_forward sets the free flag on the
0028  *                  new frame it queues. Still crap because
0029  *                  it copies the frame but at least it
0030  *                  doesn't eat memory too.
0031  *      Alan Cox    :   Generic queue code and memory fixes.
0032  *      Fred Van Kempen :   IP fragment support (borrowed from NET2E)
0033  *      Gerhard Koerting:   Forward fragmented frames correctly.
0034  *      Gerhard Koerting:   Fixes to my fix of the above 8-).
0035  *      Gerhard Koerting:   IP interface addressing fix.
0036  *      Linus Torvalds  :   More robustness checks
0037  *      Alan Cox    :   Even more checks: Still not as robust as it ought to be
0038  *      Alan Cox    :   Save IP header pointer for later
0039  *      Alan Cox    :   ip option setting
0040  *      Alan Cox    :   Use ip_tos/ip_ttl settings
0041  *      Alan Cox    :   Fragmentation bogosity removed
0042  *                  (Thanks to Mark.Bush@prg.ox.ac.uk)
0043  *      Dmitry Gorodchanin :    Send of a raw packet crash fix.
0044  *      Alan Cox    :   Silly ip bug when an overlength
0045  *                  fragment turns up. Now frees the
0046  *                  queue.
0047  *      Linus Torvalds/ :   Memory leakage on fragmentation
0048  *      Alan Cox    :   handling.
0049  *      Gerhard Koerting:   Forwarding uses IP priority hints
0050  *      Teemu Rantanen  :   Fragment problems.
0051  *      Alan Cox    :   General cleanup, comments and reformat
0052  *      Alan Cox    :   SNMP statistics
0053  *      Alan Cox    :   BSD address rule semantics. Also see
0054  *                  UDP as there is a nasty checksum issue
0055  *                  if you do things the wrong way.
0056  *      Alan Cox    :   Always defrag, moved IP_FORWARD to the config.in file
0057  *      Alan Cox    :   IP options adjust sk->priority.
0058  *      Pedro Roque :   Fix mtu/length error in ip_forward.
0059  *      Alan Cox    :   Avoid ip_chk_addr when possible.
0060  *  Richard Underwood   :   IP multicasting.
0061  *      Alan Cox    :   Cleaned up multicast handlers.
0062  *      Alan Cox    :   RAW sockets demultiplex in the BSD style.
0063  *      Gunther Mayer   :   Fix the SNMP reporting typo
0064  *      Alan Cox    :   Always in group 224.0.0.1
0065  *  Pauline Middelink   :   Fast ip_checksum update when forwarding
0066  *                  Masquerading support.
0067  *      Alan Cox    :   Multicast loopback error for 224.0.0.1
0068  *      Alan Cox    :   IP_MULTICAST_LOOP option.
0069  *      Alan Cox    :   Use notifiers.
0070  *      Bjorn Ekwall    :   Removed ip_csum (from slhc.c too)
0071  *      Bjorn Ekwall    :   Moved ip_fast_csum to ip.h (inline!)
0072  *      Stefan Becker   :       Send out ICMP HOST REDIRECT
0073  *  Arnt Gulbrandsen    :   ip_build_xmit
0074  *      Alan Cox    :   Per socket routing cache
0075  *      Alan Cox    :   Fixed routing cache, added header cache.
0076  *      Alan Cox    :   Loopback didn't work right in original ip_build_xmit - fixed it.
0077  *      Alan Cox    :   Only send ICMP_REDIRECT if src/dest are the same net.
0078  *      Alan Cox    :   Incoming IP option handling.
0079  *      Alan Cox    :   Set saddr on raw output frames as per BSD.
0080  *      Alan Cox    :   Stopped broadcast source route explosions.
0081  *      Alan Cox    :   Can disable source routing
0082  *      Takeshi Sone    :   Masquerading didn't work.
0083  *  Dave Bonn,Alan Cox  :   Faster IP forwarding whenever possible.
0084  *      Alan Cox    :   Memory leaks, tramples, misc debugging.
0085  *      Alan Cox    :   Fixed multicast (by popular demand 8))
0086  *      Alan Cox    :   Fixed forwarding (by even more popular demand 8))
0087  *      Alan Cox    :   Fixed SNMP statistics [I think]
0088  *  Gerhard Koerting    :   IP fragmentation forwarding fix
0089  *      Alan Cox    :   Device lock against page fault.
0090  *      Alan Cox    :   IP_HDRINCL facility.
0091  *  Werner Almesberger  :   Zero fragment bug
0092  *      Alan Cox    :   RAW IP frame length bug
0093  *      Alan Cox    :   Outgoing firewall on build_xmit
0094  *      A.N.Kuznetsov   :   IP_OPTIONS support throughout the kernel
0095  *      Alan Cox    :   Multicast routing hooks
0096  *      Jos Vos     :   Do accounting *before* call_in_firewall
0097  *  Willy Konynenberg   :   Transparent proxying support
0098  *
0099  * To Fix:
0100  *      IP fragmentation wants rewriting cleanly. The RFC815 algorithm is much more efficient
0101  *      and could be made very efficient with the addition of some virtual memory hacks to permit
0102  *      the allocation of a buffer that can then be 'grown' by twiddling page tables.
0103  *      Output fragmentation wants updating along with the buffer management to use a single
0104  *      interleaved copy algorithm so that fragmenting has a one copy overhead. Actual packet
0105  *      output should probably do its own fragmentation at the UDP/RAW layer. TCP shouldn't cause
0106  *      fragmentation anyway.
0107  */
0108
0109 #define pr_fmt(fmt) "IPv4: " fmt
0110
0111 #include <linux/module.h>
0112 #include <linux/types.h>
0113 #include <linux/kernel.h>
0114 #include <linux/string.h>
0115 #include <linux/errno.h>
0116 #include <linux/slab.h>
0117
0118 #include <linux/net.h>
0119 #include <linux/socket.h>
0120 #include <linux/sockios.h>
0121 #include <linux/in.h>
0122 #include <linux/inet.h>
0123 #include <linux/inetdevice.h>
0124 #include <linux/netdevice.h>
0125 #include <linux/etherdevice.h>
0126 #include <linux/indirect_call_wrapper.h>
0127
0128 #include <net/snmp.h>
0129 #include <net/ip.h>
0130 #include <net/protocol.h>
0131 #include <net/route.h>
0132 #include <linux/skbuff.h>
0133 #include <net/sock.h>
0134 #include <net/arp.h>
0135 #include <net/icmp.h>
0136 #include <net/raw.h>
0137 #include <net/checksum.h>
0138 #include <net/inet_ecn.h>
0139 #include <linux/netfilter_ipv4.h>
0140 #include <net/xfrm.h>
0141 #include <linux/mroute.h>
0142 #include <linux/netlink.h>
0143 #include <net/dst_metadata.h>
0144
0145 /*
0146  *  Process Router Attention IP option (RFC 2113)
0147  */
0148 bool ip_call_ra_chain(struct sk_buff *skb)
0149 {
0150     struct ip_ra_chain *ra;
0151     u8 protocol = ip_hdr(skb)->protocol;
0152     struct sock *last = NULL;
0153     struct net_device *dev = skb->dev;
0154     struct net *net = dev_net(dev);
0155
0156     for (ra = rcu_dereference(net->ipv4.ra_chain); ra; ra = rcu_dereference(ra->next)) {
0157         struct sock *sk = ra->sk;
0158
0159         /* If socket is bound to an interface, only report
0160          * the packet if it came  from that interface.
0161          */
0162         if (sk && inet_sk(sk)->inet_num == protocol &&
0163             (!sk->sk_bound_dev_if ||
0164              sk->sk_bound_dev_if == dev->ifindex)) {
0165             if (ip_is_fragment(ip_hdr(skb))) {
0166                 if (ip_defrag(net, skb, IP_DEFRAG_CALL_RA_CHAIN))
0167                     return true;
0168             }
0169             if (last) {
0170                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
0171                 if (skb2)
0172                     raw_rcv(last, skb2);
0173             }
0174             last = sk;
0175         }
0176     }
0177
0178     if (last) {
0179         raw_rcv(last, skb);
0180         return true;
0181     }
0182     return false;
0183 }
0184
0185 INDIRECT_CALLABLE_DECLARE(int udp_rcv(struct sk_buff *));
0186 INDIRECT_CALLABLE_DECLARE(int tcp_v4_rcv(struct sk_buff *));
0187 void ip_protocol_deliver_rcu(struct net *net, struct sk_buff *skb, int protocol)
0188 {
0189     const struct net_protocol *ipprot;
0190     int raw, ret;
0191
0192 resubmit:
0193     raw = raw_local_deliver(skb, protocol);
0194
0195     ipprot = rcu_dereference(inet_protos[protocol]);
0196     if (ipprot) {
0197         if (!ipprot->no_policy) {
0198             if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
0199                 kfree_skb_reason(skb,
0200                          SKB_DROP_REASON_XFRM_POLICY);
0201                 return;
0202             }
0203             nf_reset_ct(skb);
0204         }
0205         ret = INDIRECT_CALL_2(ipprot->handler, tcp_v4_rcv, udp_rcv,
0206                       skb);
0207         if (ret < 0) {
0208             protocol = -ret;
0209             goto resubmit;
0210         }
0211         __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
0212     } else {
0213         if (!raw) {
0214             if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
0215                 __IP_INC_STATS(net, IPSTATS_MIB_INUNKNOWNPROTOS);
0216                 icmp_send(skb, ICMP_DEST_UNREACH,
0217                       ICMP_PROT_UNREACH, 0);
0218             }
0219             kfree_skb_reason(skb, SKB_DROP_REASON_IP_NOPROTO);
0220         } else {
0221             __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS);
0222             consume_skb(skb);
0223         }
0224     }
0225 }
0226
0227 static int ip_local_deliver_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
0228 {
0229     skb_clear_delivery_time(skb);
0230     __skb_pull(skb, skb_network_header_len(skb));
0231
0232     rcu_read_lock();
0233     ip_protocol_deliver_rcu(net, skb, ip_hdr(skb)->protocol);
0234     rcu_read_unlock();
0235
0236     return 0;
0237 }
0238
0239 /*
0240  *  Deliver IP Packets to the higher protocol layers.
0241  */
0242 int ip_local_deliver(struct sk_buff *skb)
0243 {
0244     /*
0245      *  Reassemble IP fragments.
0246      */
0247     struct net *net = dev_net(skb->dev);
0248
0249     if (ip_is_fragment(ip_hdr(skb))) {
0250         if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER))
0251             return 0;
0252     }
0253
0254     return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN,
0255                net, NULL, skb, skb->dev, NULL,
0256                ip_local_deliver_finish);
0257 }
0258 EXPORT_SYMBOL(ip_local_deliver);
0259
0260 static inline bool ip_rcv_options(struct sk_buff *skb, struct net_device *dev)
0261 {
0262     struct ip_options *opt;
0263     const struct iphdr *iph;
0264
0265     /* It looks as overkill, because not all
0266        IP options require packet mangling.
0267        But it is the easiest for now, especially taking
0268        into account that combination of IP options
0269        and running sniffer is extremely rare condition.
0270                           --ANK (980813)
0271     */
0272     if (skb_cow(skb, skb_headroom(skb))) {
0273         __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS);
0274         goto drop;
0275     }
0276
0277     iph = ip_hdr(skb);
0278     opt = &(IPCB(skb)->opt);
0279     opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
0280
0281     if (ip_options_compile(dev_net(dev), opt, skb)) {
0282         __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
0283         goto drop;
0284     }
0285
0286     if (unlikely(opt->srr)) {
0287         struct in_device *in_dev = __in_dev_get_rcu(dev);
0288
0289         if (in_dev) {
0290             if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
0291                 if (IN_DEV_LOG_MARTIANS(in_dev))
0292                     net_info_ratelimited("source route option %pI4 -> %pI4\n",
0293                                  &iph->saddr,
0294                                  &iph->daddr);
0295                 goto drop;
0296             }
0297         }
0298
0299         if (ip_options_rcv_srr(skb, dev))
0300             goto drop;
0301     }
0302
0303     return false;
0304 drop:
0305     return true;
0306 }
0307
0308 static bool ip_can_use_hint(const struct sk_buff *skb, const struct iphdr *iph,
0309                 const struct sk_buff *hint)
0310 {
0311     return hint && !skb_dst(skb) && ip_hdr(hint)->daddr == iph->daddr &&
0312            ip_hdr(hint)->tos == iph->tos;
0313 }
0314
0315 int tcp_v4_early_demux(struct sk_buff *skb);
0316 int udp_v4_early_demux(struct sk_buff *skb);
0317 static int ip_rcv_finish_core(struct net *net, struct sock *sk,
0318                   struct sk_buff *skb, struct net_device *dev,
0319                   const struct sk_buff *hint)
0320 {
0321     const struct iphdr *iph = ip_hdr(skb);
0322     int err, drop_reason;
0323     struct rtable *rt;
0324
0325     drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
0326
0327     if (ip_can_use_hint(skb, iph, hint)) {
0328         err = ip_route_use_hint(skb, iph->daddr, iph->saddr, iph->tos,
0329                     dev, hint);
0330         if (unlikely(err))
0331             goto drop_error;
0332     }
0333
0334     if (READ_ONCE(net->ipv4.sysctl_ip_early_demux) &&
0335         !skb_dst(skb) &&
0336         !skb->sk &&
0337         !ip_is_fragment(iph)) {
0338         switch (iph->protocol) {
0339         case IPPROTO_TCP:
0340             if (READ_ONCE(net->ipv4.sysctl_tcp_early_demux)) {
0341                 tcp_v4_early_demux(skb);
0342
0343                 /* must reload iph, skb->head might have changed */
0344                 iph = ip_hdr(skb);
0345             }
0346             break;
0347         case IPPROTO_UDP:
0348             if (READ_ONCE(net->ipv4.sysctl_udp_early_demux)) {
0349                 err = udp_v4_early_demux(skb);
0350                 if (unlikely(err))
0351                     goto drop_error;
0352
0353                 /* must reload iph, skb->head might have changed */
0354                 iph = ip_hdr(skb);
0355             }
0356             break;
0357         }
0358     }
0359
0360     /*
0361      *  Initialise the virtual path cache for the packet. It describes
0362      *  how the packet travels inside Linux networking.
0363      */
0364     if (!skb_valid_dst(skb)) {
0365         err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
0366                        iph->tos, dev);
0367         if (unlikely(err))
0368             goto drop_error;
0369     }
0370
0371 #ifdef CONFIG_IP_ROUTE_CLASSID
0372     if (unlikely(skb_dst(skb)->tclassid)) {
0373         struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
0374         u32 idx = skb_dst(skb)->tclassid;
0375         st[idx&0xFF].o_packets++;
0376         st[idx&0xFF].o_bytes += skb->len;
0377         st[(idx>>16)&0xFF].i_packets++;
0378         st[(idx>>16)&0xFF].i_bytes += skb->len;
0379     }
0380 #endif
0381
0382     if (iph->ihl > 5 && ip_rcv_options(skb, dev))
0383         goto drop;
0384
0385     rt = skb_rtable(skb);
0386     if (rt->rt_type == RTN_MULTICAST) {
0387         __IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len);
0388     } else if (rt->rt_type == RTN_BROADCAST) {
0389         __IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len);
0390     } else if (skb->pkt_type == PACKET_BROADCAST ||
0391            skb->pkt_type == PACKET_MULTICAST) {
0392         struct in_device *in_dev = __in_dev_get_rcu(dev);
0393
0394         /* RFC 1122 3.3.6:
0395          *
0396          *   When a host sends a datagram to a link-layer broadcast
0397          *   address, the IP destination address MUST be a legal IP
0398          *   broadcast or IP multicast address.
0399          *
0400          *   A host SHOULD silently discard a datagram that is received
0401          *   via a link-layer broadcast (see Section 2.4) but does not
0402          *   specify an IP multicast or broadcast destination address.
0403          *
0404          * This doesn't explicitly say L2 *broadcast*, but broadcast is
0405          * in a way a form of multicast and the most common use case for
0406          * this is 802.11 protecting against cross-station spoofing (the
0407          * so-called "hole-196" attack) so do it for both.
0408          */
0409         if (in_dev &&
0410             IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST)) {
0411             drop_reason = SKB_DROP_REASON_UNICAST_IN_L2_MULTICAST;
0412             goto drop;
0413         }
0414     }
0415
0416     return NET_RX_SUCCESS;
0417
0418 drop:
0419     kfree_skb_reason(skb, drop_reason);
0420     return NET_RX_DROP;
0421
0422 drop_error:
0423     if (err == -EXDEV) {
0424         drop_reason = SKB_DROP_REASON_IP_RPFILTER;
0425         __NET_INC_STATS(net, LINUX_MIB_IPRPFILTER);
0426     }
0427     goto drop;
0428 }
0429
0430 static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
0431 {
0432     struct net_device *dev = skb->dev;
0433     int ret;
0434
0435     /* if ingress device is enslaved to an L3 master device pass the
0436      * skb to its handler for processing
0437      */
0438     skb = l3mdev_ip_rcv(skb);
0439     if (!skb)
0440         return NET_RX_SUCCESS;
0441
0442     ret = ip_rcv_finish_core(net, sk, skb, dev, NULL);
0443     if (ret != NET_RX_DROP)
0444         ret = dst_input(skb);
0445     return ret;
0446 }
0447
0448 /*
0449  *  Main IP Receive routine.
0450  */
0451 static struct sk_buff *ip_rcv_core(struct sk_buff *skb, struct net *net)
0452 {
0453     const struct iphdr *iph;
0454     int drop_reason;
0455     u32 len;
0456
0457     /* When the interface is in promisc. mode, drop all the crap
0458      * that it receives, do not try to analyse it.
0459      */
0460     if (skb->pkt_type == PACKET_OTHERHOST) {
0461         dev_core_stats_rx_otherhost_dropped_inc(skb->dev);
0462         drop_reason = SKB_DROP_REASON_OTHERHOST;
0463         goto drop;
0464     }
0465
0466     __IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len);
0467
0468     skb = skb_share_check(skb, GFP_ATOMIC);
0469     if (!skb) {
0470         __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
0471         goto out;
0472     }
0473
0474     drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
0475     if (!pskb_may_pull(skb, sizeof(struct iphdr)))
0476         goto inhdr_error;
0477
0478     iph = ip_hdr(skb);
0479
0480     /*
0481      *  RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
0482      *
0483      *  Is the datagram acceptable?
0484      *
0485      *  1.  Length at least the size of an ip header
0486      *  2.  Version of 4
0487      *  3.  Checksums correctly. [Speed optimisation for later, skip loopback checksums]
0488      *  4.  Doesn't have a bogus length
0489      */
0490
0491     if (iph->ihl < 5 || iph->version != 4)
0492         goto inhdr_error;
0493
0494     BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
0495     BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
0496     BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
0497     __IP_ADD_STATS(net,
0498                IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
0499                max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
0500
0501     if (!pskb_may_pull(skb, iph->ihl*4))
0502         goto inhdr_error;
0503
0504     iph = ip_hdr(skb);
0505
0506     if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
0507         goto csum_error;
0508
0509     len = ntohs(iph->tot_len);
0510     if (skb->len < len) {
0511         drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
0512         __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
0513         goto drop;
0514     } else if (len < (iph->ihl*4))
0515         goto inhdr_error;
0516
0517     /* Our transport medium may have padded the buffer out. Now we know it
0518      * is IP we can trim to the true length of the frame.
0519      * Note this now means skb->len holds ntohs(iph->tot_len).
0520      */
0521     if (pskb_trim_rcsum(skb, len)) {
0522         __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
0523         goto drop;
0524     }
0525
0526     iph = ip_hdr(skb);
0527     skb->transport_header = skb->network_header + iph->ihl*4;
0528
0529     /* Remove any debris in the socket control block */
0530     memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
0531     IPCB(skb)->iif = skb->skb_iif;
0532
0533     /* Must drop socket now because of tproxy. */
0534     if (!skb_sk_is_prefetched(skb))
0535         skb_orphan(skb);
0536
0537     return skb;
0538
0539 csum_error:
0540     drop_reason = SKB_DROP_REASON_IP_CSUM;
0541     __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
0542 inhdr_error:
0543     if (drop_reason == SKB_DROP_REASON_NOT_SPECIFIED)
0544         drop_reason = SKB_DROP_REASON_IP_INHDR;
0545     __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
0546 drop:
0547     kfree_skb_reason(skb, drop_reason);
0548 out:
0549     return NULL;
0550 }
0551
0552 /*
0553  * IP receive entry point
0554  */
0555 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
0556        struct net_device *orig_dev)
0557 {
0558     struct net *net = dev_net(dev);
0559
0560     skb = ip_rcv_core(skb, net);
0561     if (skb == NULL)
0562         return NET_RX_DROP;
0563
0564     return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING,
0565                net, NULL, skb, dev, NULL,
0566                ip_rcv_finish);
0567 }
0568
0569 static void ip_sublist_rcv_finish(struct list_head *head)
0570 {
0571     struct sk_buff *skb, *next;
0572
0573     list_for_each_entry_safe(skb, next, head, list) {
0574         skb_list_del_init(skb);
0575         dst_input(skb);
0576     }
0577 }
0578
0579 static struct sk_buff *ip_extract_route_hint(const struct net *net,
0580                          struct sk_buff *skb, int rt_type)
0581 {
0582     if (fib4_has_custom_rules(net) || rt_type == RTN_BROADCAST)
0583         return NULL;
0584
0585     return skb;
0586 }
0587
0588 static void ip_list_rcv_finish(struct net *net, struct sock *sk,
0589                    struct list_head *head)
0590 {
0591     struct sk_buff *skb, *next, *hint = NULL;
0592     struct dst_entry *curr_dst = NULL;
0593     struct list_head sublist;
0594
0595     INIT_LIST_HEAD(&sublist);
0596     list_for_each_entry_safe(skb, next, head, list) {
0597         struct net_device *dev = skb->dev;
0598         struct dst_entry *dst;
0599
0600         skb_list_del_init(skb);
0601         /* if ingress device is enslaved to an L3 master device pass the
0602          * skb to its handler for processing
0603          */
0604         skb = l3mdev_ip_rcv(skb);
0605         if (!skb)
0606             continue;
0607         if (ip_rcv_finish_core(net, sk, skb, dev, hint) == NET_RX_DROP)
0608             continue;
0609
0610         dst = skb_dst(skb);
0611         if (curr_dst != dst) {
0612             hint = ip_extract_route_hint(net, skb,
0613                            ((struct rtable *)dst)->rt_type);
0614
0615             /* dispatch old sublist */
0616             if (!list_empty(&sublist))
0617                 ip_sublist_rcv_finish(&sublist);
0618             /* start new sublist */
0619             INIT_LIST_HEAD(&sublist);
0620             curr_dst = dst;
0621         }
0622         list_add_tail(&skb->list, &sublist);
0623     }
0624     /* dispatch final sublist */
0625     ip_sublist_rcv_finish(&sublist);
0626 }
0627
0628 static void ip_sublist_rcv(struct list_head *head, struct net_device *dev,
0629                struct net *net)
0630 {
0631     NF_HOOK_LIST(NFPROTO_IPV4, NF_INET_PRE_ROUTING, net, NULL,
0632              head, dev, NULL, ip_rcv_finish);
0633     ip_list_rcv_finish(net, NULL, head);
0634 }
0635
0636 /* Receive a list of IP packets */
0637 void ip_list_rcv(struct list_head *head, struct packet_type *pt,
0638          struct net_device *orig_dev)
0639 {
0640     struct net_device *curr_dev = NULL;
0641     struct net *curr_net = NULL;
0642     struct sk_buff *skb, *next;
0643     struct list_head sublist;
0644
0645     INIT_LIST_HEAD(&sublist);
0646     list_for_each_entry_safe(skb, next, head, list) {
0647         struct net_device *dev = skb->dev;
0648         struct net *net = dev_net(dev);
0649
0650         skb_list_del_init(skb);
0651         skb = ip_rcv_core(skb, net);
0652         if (skb == NULL)
0653             continue;
0654
0655         if (curr_dev != dev || curr_net != net) {
0656             /* dispatch old sublist */
0657             if (!list_empty(&sublist))
0658                 ip_sublist_rcv(&sublist, curr_dev, curr_net);
0659             /* start new sublist */
0660             INIT_LIST_HEAD(&sublist);
0661             curr_dev = dev;
0662             curr_net = net;
0663         }
0664         list_add_tail(&skb->list, &sublist);
0665     }
0666     /* dispatch final sublist */
0667     if (!list_empty(&sublist))
0668         ip_sublist_rcv(&sublist, curr_dev, curr_net);
0669 }