LXR linux/net/ipv4/ip

   1/*
   2 * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3 *              operating system.  INET is implemented using the  BSD Socket
   4 *              interface as the means of communication with the user level.
   5 *
   6 *              The Internet Protocol (IP) module.
   7 *
   8 * Authors:     Ross Biro
   9 *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10 *              Donald Becker, <becker@super.org>
  11 *              Alan Cox, <alan@lxorguk.ukuu.org.uk>
  12 *              Richard Underwood
  13 *              Stefan Becker, <stefanb@yello.ping.de>
  14 *              Jorge Cwik, <jorge@laser.satlink.net>
  15 *              Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  16 *
  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 *
 100 *
 101 * To Fix:
 102 *              IP fragmentation wants rewriting cleanly. The RFC815 algorithm is much more efficient
 103 *              and could be made very efficient with the addition of some virtual memory hacks to permit
 104 *              the allocation of a buffer that can then be 'grown' by twiddling page tables.
 105 *              Output fragmentation wants updating along with the buffer management to use a single
 106 *              interleaved copy algorithm so that fragmenting has a one copy overhead. Actual packet
 107 *              output should probably do its own fragmentation at the UDP/RAW layer. TCP shouldn't cause
 108 *              fragmentation anyway.
 109 *
 110 *              This program is free software; you can redistribute it and/or
 111 *              modify it under the terms of the GNU General Public License
 112 *              as published by the Free Software Foundation; either version
 113 *              2 of the License, or (at your option) any later version.
 114 */
 115
 116#define pr_fmt(fmt) "IPv4: " fmt
 117
 118#include <linux/module.h>
 119#include <linux/types.h>
 120#include <linux/kernel.h>
 121#include <linux/string.h>
 122#include <linux/errno.h>
 123#include <linux/slab.h>
 124
 125#include <linux/net.h>
 126#include <linux/socket.h>
 127#include <linux/sockios.h>
 128#include <linux/in.h>
 129#include <linux/inet.h>
 130#include <linux/inetdevice.h>
 131#include <linux/netdevice.h>
 132#include <linux/etherdevice.h>
 133
 134#include <net/snmp.h>
 135#include <net/ip.h>
 136#include <net/protocol.h>
 137#include <net/route.h>
 138#include <linux/skbuff.h>
 139#include <net/sock.h>
 140#include <net/arp.h>
 141#include <net/icmp.h>
 142#include <net/raw.h>
 143#include <net/checksum.h>
 144#include <net/inet_ecn.h>
 145#include <linux/netfilter_ipv4.h>
 146#include <net/xfrm.h>
 147#include <linux/mroute.h>
 148#include <linux/netlink.h>
 149#include <net/dst_metadata.h>
 150
 151/*
 152 *      Process Router Attention IP option (RFC 2113)
 153 */
 154bool ip_call_ra_chain(struct sk_buff *skb)
 155{
 156        struct ip_ra_chain *ra;
 157        u8 protocol = ip_hdr(skb)->protocol;
 158        struct sock *last = NULL;
 159        struct net_device *dev = skb->dev;
 160
 161        for (ra = rcu_dereference(ip_ra_chain); ra; ra = rcu_dereference(ra->next)) {
 162                struct sock *sk = ra->sk;
 163
 164                /* If socket is bound to an interface, only report
 165                 * the packet if it came  from that interface.
 166                 */
 167                if (sk && inet_sk(sk)->inet_num == protocol &&
 168                    (!sk->sk_bound_dev_if ||
 169                     sk->sk_bound_dev_if == dev->ifindex) &&
 170                    net_eq(sock_net(sk), dev_net(dev))) {
 171                        if (ip_is_fragment(ip_hdr(skb))) {
 172                                if (ip_defrag(skb, IP_DEFRAG_CALL_RA_CHAIN))
 173                                        return true;
 174                        }
 175                        if (last) {
 176                                struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
 177                                if (skb2)
 178                                        raw_rcv(last, skb2);
 179                        }
 180                        last = sk;
 181                }
 182        }
 183
 184        if (last) {
 185                raw_rcv(last, skb);
 186                return true;
 187        }
 188        return false;
 189}
 190
 191static int ip_local_deliver_finish(struct sock *sk, struct sk_buff *skb)
 192{
 193        struct net *net = dev_net(skb->dev);
 194
 195        __skb_pull(skb, skb_network_header_len(skb));
 196
 197        rcu_read_lock();
 198        {
 199                int protocol = ip_hdr(skb)->protocol;
 200                const struct net_protocol *ipprot;
 201                int raw;
 202
 203        resubmit:
 204                raw = raw_local_deliver(skb, protocol);
 205
 206                ipprot = rcu_dereference(inet_protos[protocol]);
 207                if (ipprot != NULL) {
 208                        int ret;
 209
 210                        if (!ipprot->no_policy) {
 211                                if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 212                                        kfree_skb(skb);
 213                                        goto out;
 214                                }
 215                                nf_reset(skb);
 216                        }
 217                        ret = ipprot->handler(skb);
 218                        if (ret < 0) {
 219                                protocol = -ret;
 220                                goto resubmit;
 221                        }
 222                        IP_INC_STATS_BH(net, IPSTATS_MIB_INDELIVERS);
 223                } else {
 224                        if (!raw) {
 225                                if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 226                                        IP_INC_STATS_BH(net, IPSTATS_MIB_INUNKNOWNPROTOS);
 227                                        icmp_send(skb, ICMP_DEST_UNREACH,
 228                                                  ICMP_PROT_UNREACH, 0);
 229                                }
 230                                kfree_skb(skb);
 231                        } else {
 232                                IP_INC_STATS_BH(net, IPSTATS_MIB_INDELIVERS);
 233                                consume_skb(skb);
 234                        }
 235                }
 236        }
 237 out:
 238        rcu_read_unlock();
 239
 240        return 0;
 241}
 242
 243/*
 244 *      Deliver IP Packets to the higher protocol layers.
 245 */
 246int ip_local_deliver(struct sk_buff *skb)
 247{
 248        /*
 249         *      Reassemble IP fragments.
 250         */
 251
 252        if (ip_is_fragment(ip_hdr(skb))) {
 253                if (ip_defrag(skb, IP_DEFRAG_LOCAL_DELIVER))
 254                        return 0;
 255        }
 256
 257        return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN, NULL, skb,
 258                       skb->dev, NULL,
 259                       ip_local_deliver_finish);
 260}
 261
 262static inline bool ip_rcv_options(struct sk_buff *skb)
 263{
 264        struct ip_options *opt;
 265        const struct iphdr *iph;
 266        struct net_device *dev = skb->dev;
 267
 268        /* It looks as overkill, because not all
 269           IP options require packet mangling.
 270           But it is the easiest for now, especially taking
 271           into account that combination of IP options
 272           and running sniffer is extremely rare condition.
 273                                              --ANK (980813)
 274        */
 275        if (skb_cow(skb, skb_headroom(skb))) {
 276                IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 277                goto drop;
 278        }
 279
 280        iph = ip_hdr(skb);
 281        opt = &(IPCB(skb)->opt);
 282        opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
 283
 284        if (ip_options_compile(dev_net(dev), opt, skb)) {
 285                IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 286                goto drop;
 287        }
 288
 289        if (unlikely(opt->srr)) {
 290                struct in_device *in_dev = __in_dev_get_rcu(dev);
 291
 292                if (in_dev) {
 293                        if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
 294                                if (IN_DEV_LOG_MARTIANS(in_dev))
 295                                        net_info_ratelimited("source route option %pI4 -> %pI4\n",
 296                                                             &iph->saddr,
 297                                                             &iph->daddr);
 298                                goto drop;
 299                        }
 300                }
 301
 302                if (ip_options_rcv_srr(skb))
 303                        goto drop;
 304        }
 305
 306        return false;
 307drop:
 308        return true;
 309}
 310
 311int sysctl_ip_early_demux __read_mostly = 1;
 312EXPORT_SYMBOL(sysctl_ip_early_demux);
 313
 314static int ip_rcv_finish(struct sock *sk, struct sk_buff *skb)
 315{
 316        const struct iphdr *iph = ip_hdr(skb);
 317        struct rtable *rt;
 318        int err;
 319
 320        if (sysctl_ip_early_demux && !skb_dst(skb) && skb->sk == NULL) {
 321                const struct net_protocol *ipprot;
 322                int protocol = iph->protocol;
 323
 324                ipprot = rcu_dereference(inet_protos[protocol]);
 325                if (ipprot && ipprot->early_demux) {
 326                        err = ipprot->early_demux(skb);
 327                        if (unlikely(err))
 328                                goto drop_error;
 329                        /* must reload iph, skb->head might have changed */
 330                        iph = ip_hdr(skb);
 331                }
 332        }
 333
 334        /*
 335         *      Initialise the virtual path cache for the packet. It describes
 336         *      how the packet travels inside Linux networking.
 337         */
 338        if (!skb_valid_dst(skb)) {
 339                err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
 340                                           iph->tos, skb->dev);
 341                if (unlikely(err))
 342                        goto drop_error;
 343        }
 344
 345#ifdef CONFIG_IP_ROUTE_CLASSID
 346        if (unlikely(skb_dst(skb)->tclassid)) {
 347                struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
 348                u32 idx = skb_dst(skb)->tclassid;
 349                st[idx&0xFF].o_packets++;
 350                st[idx&0xFF].o_bytes += skb->len;
 351                st[(idx>>16)&0xFF].i_packets++;
 352                st[(idx>>16)&0xFF].i_bytes += skb->len;
 353        }
 354#endif
 355
 356        if (iph->ihl > 5 && ip_rcv_options(skb))
 357                goto drop;
 358
 359        rt = skb_rtable(skb);
 360        if (rt->rt_type == RTN_MULTICAST) {
 361                IP_UPD_PO_STATS_BH(dev_net(rt->dst.dev), IPSTATS_MIB_INMCAST,
 362                                skb->len);
 363        } else if (rt->rt_type == RTN_BROADCAST)
 364                IP_UPD_PO_STATS_BH(dev_net(rt->dst.dev), IPSTATS_MIB_INBCAST,
 365                                skb->len);
 366
 367        return dst_input(skb);
 368
 369drop:
 370        kfree_skb(skb);
 371        return NET_RX_DROP;
 372
 373drop_error:
 374        if (err == -EXDEV)
 375                NET_INC_STATS_BH(dev_net(skb->dev), LINUX_MIB_IPRPFILTER);
 376        goto drop;
 377}
 378
 379/*
 380 *      Main IP Receive routine.
 381 */
 382int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 383{
 384        const struct iphdr *iph;
 385        u32 len;
 386
 387        /* When the interface is in promisc. mode, drop all the crap
 388         * that it receives, do not try to analyse it.
 389         */
 390        if (skb->pkt_type == PACKET_OTHERHOST)
 391                goto drop;
 392
 393
 394        IP_UPD_PO_STATS_BH(dev_net(dev), IPSTATS_MIB_IN, skb->len);
 395
 396        if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL) {
 397                IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 398                goto out;
 399        }
 400
 401        if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 402                goto inhdr_error;
 403
 404        iph = ip_hdr(skb);
 405
 406        /*
 407         *      RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
 408         *
 409         *      Is the datagram acceptable?
 410         *
 411         *      1.      Length at least the size of an ip header
 412         *      2.      Version of 4
 413         *      3.      Checksums correctly. [Speed optimisation for later, skip loopback checksums]
 414         *      4.      Doesn't have a bogus length
 415         */
 416
 417        if (iph->ihl < 5 || iph->version != 4)
 418                goto inhdr_error;
 419
 420        BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1);
 421        BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0);
 422        BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE);
 423        IP_ADD_STATS_BH(dev_net(dev),
 424                        IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK),
 425                        max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
 426
 427        if (!pskb_may_pull(skb, iph->ihl*4))
 428                goto inhdr_error;
 429
 430        iph = ip_hdr(skb);
 431
 432        if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 433                goto csum_error;
 434
 435        len = ntohs(iph->tot_len);
 436        if (skb->len < len) {
 437                IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 438                goto drop;
 439        } else if (len < (iph->ihl*4))
 440                goto inhdr_error;
 441
 442        /* Our transport medium may have padded the buffer out. Now we know it
 443         * is IP we can trim to the true length of the frame.
 444         * Note this now means skb->len holds ntohs(iph->tot_len).
 445         */
 446        if (pskb_trim_rcsum(skb, len)) {
 447                IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 448                goto drop;
 449        }
 450
 451        skb->transport_header = skb->network_header + iph->ihl*4;
 452
 453        /* Remove any debris in the socket control block */
 454        memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 455
 456        /* Must drop socket now because of tproxy. */
 457        skb_orphan(skb);
 458
 459        return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, NULL, skb,
 460                       dev, NULL,
 461                       ip_rcv_finish);
 462
 463csum_error:
 464        IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_CSUMERRORS);
 465inhdr_error:
 466        IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 467drop:
 468        kfree_skb(skb);
 469out:
 470        return NET_RX_DROP;
 471}
 472