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.Cox@linux.org>
  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 #include <asm/system.h>
 117 #include <linux/module.h>
 118 #include <linux/types.h>
 119 #include <linux/kernel.h>
 120 #include <linux/string.h>
 121 #include <linux/errno.h>
 122
 123 #include <linux/net.h>
 124 #include <linux/socket.h>
 125 #include <linux/sockios.h>
 126 #include <linux/in.h>
 127 #include <linux/inet.h>
 128 #include <linux/inetdevice.h>
 129 #include <linux/netdevice.h>
 130 #include <linux/etherdevice.h>
 131
 132 #include <net/snmp.h>
 133 #include <net/ip.h>
 134 #include <net/protocol.h>
 135 #include <net/route.h>
 136 #include <linux/skbuff.h>
 137 #include <net/sock.h>
 138 #include <net/arp.h>
 139 #include <net/icmp.h>
 140 #include <net/raw.h>
 141 #include <net/checksum.h>
 142 #include <linux/netfilter_ipv4.h>
 143 #include <net/xfrm.h>
 144 #include <linux/mroute.h>
 145 #include <linux/netlink.h>
 146
 147 /*
 148  *      SNMP management statistics
 149  */
 150
 151 DEFINE_SNMP_STAT(struct ipstats_mib, ip_statistics) __read_mostly;
 152
 153 /*
 154  *      Process Router Attention IP option
 155  */
 156 int ip_call_ra_chain(struct sk_buff *skb)
 157 {
 158         struct ip_ra_chain *ra;
 159         u8 protocol = ip_hdr(skb)->protocol;
 160         struct sock *last = NULL;
 161         struct net_device *dev = skb->dev;
 162
 163         read_lock(&ip_ra_lock);
 164         for (ra = ip_ra_chain; ra; ra = ra->next) {
 165                 struct sock *sk = ra->sk;
 166
 167                 /* If socket is bound to an interface, only report
 168                  * the packet if it came  from that interface.
 169                  */
 170                 if (sk && inet_sk(sk)->num == protocol &&
 171                     (!sk->sk_bound_dev_if ||
 172                      sk->sk_bound_dev_if == dev->ifindex) &&
 173                     sock_net(sk) == dev_net(dev)) {
 174                         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
 175                                 if (ip_defrag(skb, IP_DEFRAG_CALL_RA_CHAIN)) {
 176                                         read_unlock(&ip_ra_lock);
 177                                         return 1;
 178                                 }
 179                         }
 180                         if (last) {
 181                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
 182                                 if (skb2)
 183                                         raw_rcv(last, skb2);
 184                         }
 185                         last = sk;
 186                 }
 187         }
 188
 189         if (last) {
 190                 raw_rcv(last, skb);
 191                 read_unlock(&ip_ra_lock);
 192                 return 1;
 193         }
 194         read_unlock(&ip_ra_lock);
 195         return 0;
 196 }
 197
 198 static int ip_local_deliver_finish(struct sk_buff *skb)
 199 {
 200         struct net *net = dev_net(skb->dev);
 201
 202         __skb_pull(skb, ip_hdrlen(skb));
 203
 204         /* Point into the IP datagram, just past the header. */
 205         skb_reset_transport_header(skb);
 206
 207         rcu_read_lock();
 208         {
 209                 int protocol = ip_hdr(skb)->protocol;
 210                 int hash, raw;
 211                 struct net_protocol *ipprot;
 212
 213         resubmit:
 214                 raw = raw_local_deliver(skb, protocol);
 215
 216                 hash = protocol & (MAX_INET_PROTOS - 1);
 217                 ipprot = rcu_dereference(inet_protos[hash]);
 218                 if (ipprot != NULL && (net == &init_net || ipprot->netns_ok)) {
 219                         int ret;
 220
 221                         if (!ipprot->no_policy) {
 222                                 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 223                                         kfree_skb(skb);
 224                                         goto out;
 225                                 }
 226                                 nf_reset(skb);
 227                         }
 228                         ret = ipprot->handler(skb);
 229                         if (ret < 0) {
 230                                 protocol = -ret;
 231                                 goto resubmit;
 232                         }
 233                         IP_INC_STATS_BH(IPSTATS_MIB_INDELIVERS);
 234                 } else {
 235                         if (!raw) {
 236                                 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 237                                         IP_INC_STATS_BH(IPSTATS_MIB_INUNKNOWNPROTOS);
 238                                         icmp_send(skb, ICMP_DEST_UNREACH,
 239                                                   ICMP_PROT_UNREACH, 0);
 240                                 }
 241                         } else
 242                                 IP_INC_STATS_BH(IPSTATS_MIB_INDELIVERS);
 243                         kfree_skb(skb);
 244                 }
 245         }
 246  out:
 247         rcu_read_unlock();
 248
 249         return 0;
 250 }
 251
 252 /*
 253  *      Deliver IP Packets to the higher protocol layers.
 254  */
 255 int ip_local_deliver(struct sk_buff *skb)
 256 {
 257         /*
 258          *      Reassemble IP fragments.
 259          */
 260
 261         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
 262                 if (ip_defrag(skb, IP_DEFRAG_LOCAL_DELIVER))
 263                         return 0;
 264         }
 265
 266         return NF_HOOK(PF_INET, NF_INET_LOCAL_IN, skb, skb->dev, NULL,
 267                        ip_local_deliver_finish);
 268 }
 269
 270 static inline int ip_rcv_options(struct sk_buff *skb)
 271 {
 272         struct ip_options *opt;
 273         struct iphdr *iph;
 274         struct net_device *dev = skb->dev;
 275
 276         /* It looks as overkill, because not all
 277            IP options require packet mangling.
 278            But it is the easiest for now, especially taking
 279            into account that combination of IP options
 280            and running sniffer is extremely rare condition.
 281                                               --ANK (980813)
 282         */
 283         if (skb_cow(skb, skb_headroom(skb))) {
 284                 IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS);
 285                 goto drop;
 286         }
 287
 288         iph = ip_hdr(skb);
 289         opt = &(IPCB(skb)->opt);
 290         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
 291
 292         if (ip_options_compile(dev_net(dev), opt, skb)) {
 293                 IP_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
 294                 goto drop;
 295         }
 296
 297         if (unlikely(opt->srr)) {
 298                 struct in_device *in_dev = in_dev_get(dev);
 299                 if (in_dev) {
 300                         if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
 301                                 if (IN_DEV_LOG_MARTIANS(in_dev) &&
 302                                     net_ratelimit())
 303                                         printk(KERN_INFO "source route option "
 304                                                NIPQUAD_FMT " -> " NIPQUAD_FMT "\n",
 305                                                NIPQUAD(iph->saddr),
 306                                                NIPQUAD(iph->daddr));
 307                                 in_dev_put(in_dev);
 308                                 goto drop;
 309                         }
 310
 311                         in_dev_put(in_dev);
 312                 }
 313
 314                 if (ip_options_rcv_srr(skb))
 315                         goto drop;
 316         }
 317
 318         return 0;
 319 drop:
 320         return -1;
 321 }
 322
 323 static int ip_rcv_finish(struct sk_buff *skb)
 324 {
 325         const struct iphdr *iph = ip_hdr(skb);
 326         struct rtable *rt;
 327
 328         /*
 329          *      Initialise the virtual path cache for the packet. It describes
 330          *      how the packet travels inside Linux networking.
 331          */
 332         if (skb->dst == NULL) {
 333                 int err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos,
 334                                          skb->dev);
 335                 if (unlikely(err)) {
 336                         if (err == -EHOSTUNREACH)
 337                                 IP_INC_STATS_BH(IPSTATS_MIB_INADDRERRORS);
 338                         else if (err == -ENETUNREACH)
 339                                 IP_INC_STATS_BH(IPSTATS_MIB_INNOROUTES);
 340                         goto drop;
 341                 }
 342         }
 343
 344 #ifdef CONFIG_NET_CLS_ROUTE
 345         if (unlikely(skb->dst->tclassid)) {
 346                 struct ip_rt_acct *st = per_cpu_ptr(ip_rt_acct, smp_processor_id());
 347                 u32 idx = skb->dst->tclassid;
 348                 st[idx&0xFF].o_packets++;
 349                 st[idx&0xFF].o_bytes+=skb->len;
 350                 st[(idx>>16)&0xFF].i_packets++;
 351                 st[(idx>>16)&0xFF].i_bytes+=skb->len;
 352         }
 353 #endif
 354
 355         if (iph->ihl > 5 && ip_rcv_options(skb))
 356                 goto drop;
 357
 358         rt = skb->rtable;
 359         if (rt->rt_type == RTN_MULTICAST)
 360                 IP_INC_STATS_BH(IPSTATS_MIB_INMCASTPKTS);
 361         else if (rt->rt_type == RTN_BROADCAST)
 362                 IP_INC_STATS_BH(IPSTATS_MIB_INBCASTPKTS);
 363
 364         return dst_input(skb);
 365
 366 drop:
 367         kfree_skb(skb);
 368         return NET_RX_DROP;
 369 }
 370
 371 /*
 372  *      Main IP Receive routine.
 373  */
 374 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 375 {
 376         struct iphdr *iph;
 377         u32 len;
 378
 379         /* When the interface is in promisc. mode, drop all the crap
 380          * that it receives, do not try to analyse it.
 381          */
 382         if (skb->pkt_type == PACKET_OTHERHOST)
 383                 goto drop;
 384
 385         IP_INC_STATS_BH(IPSTATS_MIB_INRECEIVES);
 386
 387         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL) {
 388                 IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS);
 389                 goto out;
 390         }
 391
 392         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 393                 goto inhdr_error;
 394
 395         iph = ip_hdr(skb);
 396
 397         /*
 398          *      RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum.
 399          *
 400          *      Is the datagram acceptable?
 401          *
 402          *      1.      Length at least the size of an ip header
 403          *      2.      Version of 4
 404          *      3.      Checksums correctly. [Speed optimisation for later, skip loopback checksums]
 405          *      4.      Doesn't have a bogus length
 406          */
 407
 408         if (iph->ihl < 5 || iph->version != 4)
 409                 goto inhdr_error;
 410
 411         if (!pskb_may_pull(skb, iph->ihl*4))
 412                 goto inhdr_error;
 413
 414         iph = ip_hdr(skb);
 415
 416         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 417                 goto inhdr_error;
 418
 419         len = ntohs(iph->tot_len);
 420         if (skb->len < len) {
 421                 IP_INC_STATS_BH(IPSTATS_MIB_INTRUNCATEDPKTS);
 422                 goto drop;
 423         } else if (len < (iph->ihl*4))
 424                 goto inhdr_error;
 425
 426         /* Our transport medium may have padded the buffer out. Now we know it
 427          * is IP we can trim to the true length of the frame.
 428          * Note this now means skb->len holds ntohs(iph->tot_len).
 429          */
 430         if (pskb_trim_rcsum(skb, len)) {
 431                 IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS);
 432                 goto drop;
 433         }
 434
 435         /* Remove any debris in the socket control block */
 436         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 437
 438         return NF_HOOK(PF_INET, NF_INET_PRE_ROUTING, skb, dev, NULL,
 439                        ip_rcv_finish);
 440
 441 inhdr_error:
 442         IP_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
 443 drop:
 444         kfree_skb(skb);
 445 out:
 446         return NET_RX_DROP;
 447 }
 448
 449 EXPORT_SYMBOL(ip_statistics);