net/ipv4/ip_input.c

   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 <linux/netfilter_ipv4.h>
 145 #include <net/xfrm.h>
 146 #include <linux/mroute.h>
 147 #include <linux/netlink.h>
 148
 149 /*
 150  *      Process Router Attention IP option (RFC 2113)
 151  */
 152 bool ip_call_ra_chain(struct sk_buff *skb)
 153 {
 154         struct ip_ra_chain *ra;
 155         u8 protocol = ip_hdr(skb)->protocol;
 156         struct sock *last = NULL;
 157         struct net_device *dev = skb->dev;
 158
 159         for (ra = rcu_dereference(ip_ra_chain); ra; ra = rcu_dereference(ra->next)) {
 160                 struct sock *sk = ra->sk;
 161
 162                 /* If socket is bound to an interface, only report
 163                  * the packet if it came  from that interface.
 164                  */
 165                 if (sk && inet_sk(sk)->inet_num == protocol &&
 166                     (!sk->sk_bound_dev_if ||
 167                      sk->sk_bound_dev_if == dev->ifindex) &&
 168                     net_eq(sock_net(sk), dev_net(dev))) {
 169                         if (ip_is_fragment(ip_hdr(skb))) {
 170                                 if (ip_defrag(skb, IP_DEFRAG_CALL_RA_CHAIN))
 171                                         return true;
 172                         }
 173                         if (last) {
 174                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
 175                                 if (skb2)
 176                                         raw_rcv(last, skb2);
 177                         }
 178                         last = sk;
 179                 }
 180         }
 181
 182         if (last) {
 183                 raw_rcv(last, skb);
 184                 return true;
 185         }
 186         return false;
 187 }
 188
 189 static int ip_local_deliver_finish(struct sk_buff *skb)
 190 {
 191         struct net *net = dev_net(skb->dev);
 192
 193         __skb_pull(skb, ip_hdrlen(skb));
 194
 195         /* Point into the IP datagram, just past the header. */
 196         skb_reset_transport_header(skb);
 197
 198         rcu_read_lock();
 199         {
 200                 int protocol = ip_hdr(skb)->protocol;
 201                 const struct net_protocol *ipprot;
 202                 int raw;
 203
 204         resubmit:
 205                 raw = raw_local_deliver(skb, protocol);
 206
 207                 ipprot = rcu_dereference(inet_protos[protocol]);
 208                 if (ipprot != NULL) {
 209                         int ret;
 210
 211                         if (!net_eq(net, &init_net) && !ipprot->netns_ok) {
 212                                 net_info_ratelimited("%s: proto %d isn't netns-ready\n",
 213                                                      __func__, protocol);
 214                                 kfree_skb(skb);
 215                                 goto out;
 216                         }
 217
 218                         if (!ipprot->no_policy) {
 219                                 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 220                                         kfree_skb(skb);
 221                                         goto out;
 222                                 }
 223                                 nf_reset(skb);
 224                         }
 225                         ret = ipprot->handler(skb);
 226                         if (ret < 0) {
 227                                 protocol = -ret;
 228                                 goto resubmit;
 229                         }
 230                         IP_INC_STATS_BH(net, IPSTATS_MIB_INDELIVERS);
 231                 } else {
 232                         if (!raw) {
 233                                 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 234                                         IP_INC_STATS_BH(net, IPSTATS_MIB_INUNKNOWNPROTOS);
 235                                         icmp_send(skb, ICMP_DEST_UNREACH,
 236                                                   ICMP_PROT_UNREACH, 0);
 237                                 }
 238                         } else
 239                                 IP_INC_STATS_BH(net, IPSTATS_MIB_INDELIVERS);
 240                         kfree_skb(skb);
 241                 }
 242         }
 243  out:
 244         rcu_read_unlock();
 245
 246         return 0;
 247 }
 248
 249 /*
 250  *      Deliver IP Packets to the higher protocol layers.
 251  */
 252 int ip_local_deliver(struct sk_buff *skb)
 253 {
 254         /*
 255          *      Reassemble IP fragments.
 256          */
 257
 258         if (ip_is_fragment(ip_hdr(skb))) {
 259                 if (ip_defrag(skb, IP_DEFRAG_LOCAL_DELIVER))
 260                         return 0;
 261         }
 262
 263         return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN, skb, skb->dev, NULL,
 264                        ip_local_deliver_finish);
 265 }
 266
 267 static inline bool ip_rcv_options(struct sk_buff *skb)
 268 {
 269         struct ip_options *opt;
 270         const struct iphdr *iph;
 271         struct net_device *dev = skb->dev;
 272
 273         /* It looks as overkill, because not all
 274            IP options require packet mangling.
 275            But it is the easiest for now, especially taking
 276            into account that combination of IP options
 277            and running sniffer is extremely rare condition.
 278                                               --ANK (980813)
 279         */
 280         if (skb_cow(skb, skb_headroom(skb))) {
 281                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 282                 goto drop;
 283         }
 284
 285         iph = ip_hdr(skb);
 286         opt = &(IPCB(skb)->opt);
 287         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
 288
 289         if (ip_options_compile(dev_net(dev), opt, skb)) {
 290                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 291                 goto drop;
 292         }
 293
 294         if (unlikely(opt->srr)) {
 295                 struct in_device *in_dev = __in_dev_get_rcu(dev);
 296
 297                 if (in_dev) {
 298                         if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
 299                                 if (IN_DEV_LOG_MARTIANS(in_dev))
 300                                         net_info_ratelimited("source route option %pI4 -> %pI4\n",
 301                                                              &iph->saddr,
 302                                                              &iph->daddr);
 303                                 goto drop;
 304                         }
 305                 }
 306
 307                 if (ip_options_rcv_srr(skb))
 308                         goto drop;
 309         }
 310
 311         return false;
 312 drop:
 313         return true;
 314 }
 315
 316 int sysctl_ip_early_demux __read_mostly = 1;
 317
 318 static int ip_rcv_finish(struct sk_buff *skb)
 319 {
 320         const struct iphdr *iph = ip_hdr(skb);
 321         struct rtable *rt;
 322
 323         /*
 324          *      Initialise the virtual path cache for the packet. It describes
 325          *      how the packet travels inside Linux networking.
 326          */
 327         if (skb_dst(skb) == NULL) {
 328                 int err = -ENOENT;
 329                 bool nocache = false;
 330
 331                 if (sysctl_ip_early_demux) {
 332                         const struct net_protocol *ipprot;
 333                         int protocol = iph->protocol;
 334
 335                         rcu_read_lock();
 336                         ipprot = rcu_dereference(inet_protos[protocol]);
 337                         if (ipprot && ipprot->early_demux)
 338                                 err = ipprot->early_demux(skb, &nocache);
 339                         rcu_read_unlock();
 340                 }
 341
 342                 if (err) {
 343                         err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
 344                                                    iph->tos, skb->dev, nocache);
 345                         if (unlikely(err)) {
 346                                 if (err == -EXDEV)
 347                                         NET_INC_STATS_BH(dev_net(skb->dev),
 348                                                          LINUX_MIB_IPRPFILTER);
 349                                 goto drop;
 350                         }
 351                 }
 352         }
 353
 354 #ifdef CONFIG_IP_ROUTE_CLASSID
 355         if (unlikely(skb_dst(skb)->tclassid)) {
 356                 struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
 357                 u32 idx = skb_dst(skb)->tclassid;
 358                 st[idx&0xFF].o_packets++;
 359                 st[idx&0xFF].o_bytes += skb->len;
 360                 st[(idx>>16)&0xFF].i_packets++;
 361                 st[(idx>>16)&0xFF].i_bytes += skb->len;
 362         }
 363 #endif
 364
 365         if (iph->ihl > 5 && ip_rcv_options(skb))
 366                 goto drop;
 367
 368         rt = skb_rtable(skb);
 369         if (rt->rt_type == RTN_MULTICAST) {
 370                 IP_UPD_PO_STATS_BH(dev_net(rt->dst.dev), IPSTATS_MIB_INMCAST,
 371                                 skb->len);
 372         } else if (rt->rt_type == RTN_BROADCAST)
 373                 IP_UPD_PO_STATS_BH(dev_net(rt->dst.dev), IPSTATS_MIB_INBCAST,
 374                                 skb->len);
 375
 376         return dst_input(skb);
 377
 378 drop:
 379         kfree_skb(skb);
 380         return NET_RX_DROP;
 381 }
 382
 383 /*
 384  *      Main IP Receive routine.
 385  */
 386 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 387 {
 388         const struct iphdr *iph;
 389         u32 len;
 390
 391         /* When the interface is in promisc. mode, drop all the crap
 392          * that it receives, do not try to analyse it.
 393          */
 394         if (skb->pkt_type == PACKET_OTHERHOST)
 395                 goto drop;
 396
 397
 398         IP_UPD_PO_STATS_BH(dev_net(dev), IPSTATS_MIB_IN, skb->len);
 399
 400         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL) {
 401                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 402                 goto out;
 403         }
 404
 405         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 406                 goto inhdr_error;
 407
 408         iph = ip_hdr(skb);
 409
 410         /*
 411          *      RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
 412          *
 413          *      Is the datagram acceptable?
 414          *
 415          *      1.      Length at least the size of an ip header
 416          *      2.      Version of 4
 417          *      3.      Checksums correctly. [Speed optimisation for later, skip loopback checksums]
 418          *      4.      Doesn't have a bogus length
 419          */
 420
 421         if (iph->ihl < 5 || iph->version != 4)
 422                 goto inhdr_error;
 423
 424         if (!pskb_may_pull(skb, iph->ihl*4))
 425                 goto inhdr_error;
 426
 427         iph = ip_hdr(skb);
 428
 429         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 430                 goto inhdr_error;
 431
 432         len = ntohs(iph->tot_len);
 433         if (skb->len < len) {
 434                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 435                 goto drop;
 436         } else if (len < (iph->ihl*4))
 437                 goto inhdr_error;
 438
 439         /* Our transport medium may have padded the buffer out. Now we know it
 440          * is IP we can trim to the true length of the frame.
 441          * Note this now means skb->len holds ntohs(iph->tot_len).
 442          */
 443         if (pskb_trim_rcsum(skb, len)) {
 444                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 445                 goto drop;
 446         }
 447
 448         /* Remove any debris in the socket control block */
 449         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 450
 451         /* Must drop socket now because of tproxy. */
 452         skb_orphan(skb);
 453
 454         return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, dev, NULL,
 455                        ip_rcv_finish);
 456
 457 inhdr_error:
 458         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 459 drop:
 460         kfree_skb(skb);
 461 out:
 462         return NET_RX_DROP;
 463 }