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  * Version:     $Id: ip_input.c,v 1.55 2002/01/12 07:39:45 davem Exp $
   9  *
  10  * Authors:     Ross Biro
  11  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  12  *              Donald Becker, <becker@super.org>
  13  *              Alan Cox, <Alan.Cox@linux.org>
  14  *              Richard Underwood
  15  *              Stefan Becker, <stefanb@yello.ping.de>
  16  *              Jorge Cwik, <jorge@laser.satlink.net>
  17  *              Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  18  *
  19  *
  20  * Fixes:
  21  *              Alan Cox        :       Commented a couple of minor bits of surplus code
  22  *              Alan Cox        :       Undefining IP_FORWARD doesn't include the code
  23  *                                      (just stops a compiler warning).
  24  *              Alan Cox        :       Frames with >=MAX_ROUTE record routes, strict routes or loose routes
  25  *                                      are junked rather than corrupting things.
  26  *              Alan Cox        :       Frames to bad broadcast subnets are dumped
  27  *                                      We used to process them non broadcast and
  28  *                                      boy could that cause havoc.
  29  *              Alan Cox        :       ip_forward sets the free flag on the
  30  *                                      new frame it queues. Still crap because
  31  *                                      it copies the frame but at least it
  32  *                                      doesn't eat memory too.
  33  *              Alan Cox        :       Generic queue code and memory fixes.
  34  *              Fred Van Kempen :       IP fragment support (borrowed from NET2E)
  35  *              Gerhard Koerting:       Forward fragmented frames correctly.
  36  *              Gerhard Koerting:       Fixes to my fix of the above 8-).
  37  *              Gerhard Koerting:       IP interface addressing fix.
  38  *              Linus Torvalds  :       More robustness checks
  39  *              Alan Cox        :       Even more checks: Still not as robust as it ought to be
  40  *              Alan Cox        :       Save IP header pointer for later
  41  *              Alan Cox        :       ip option setting
  42  *              Alan Cox        :       Use ip_tos/ip_ttl settings
  43  *              Alan Cox        :       Fragmentation bogosity removed
  44  *                                      (Thanks to Mark.Bush@prg.ox.ac.uk)
  45  *              Dmitry Gorodchanin :    Send of a raw packet crash fix.
  46  *              Alan Cox        :       Silly ip bug when an overlength
  47  *                                      fragment turns up. Now frees the
  48  *                                      queue.
  49  *              Linus Torvalds/ :       Memory leakage on fragmentation
  50  *              Alan Cox        :       handling.
  51  *              Gerhard Koerting:       Forwarding uses IP priority hints
  52  *              Teemu Rantanen  :       Fragment problems.
  53  *              Alan Cox        :       General cleanup, comments and reformat
  54  *              Alan Cox        :       SNMP statistics
  55  *              Alan Cox        :       BSD address rule semantics. Also see
  56  *                                      UDP as there is a nasty checksum issue
  57  *                                      if you do things the wrong way.
  58  *              Alan Cox        :       Always defrag, moved IP_FORWARD to the config.in file
  59  *              Alan Cox        :       IP options adjust sk->priority.
  60  *              Pedro Roque     :       Fix mtu/length error in ip_forward.
  61  *              Alan Cox        :       Avoid ip_chk_addr when possible.
  62  *      Richard Underwood       :       IP multicasting.
  63  *              Alan Cox        :       Cleaned up multicast handlers.
  64  *              Alan Cox        :       RAW sockets demultiplex in the BSD style.
  65  *              Gunther Mayer   :       Fix the SNMP reporting typo
  66  *              Alan Cox        :       Always in group 224.0.0.1
  67  *      Pauline Middelink       :       Fast ip_checksum update when forwarding
  68  *                                      Masquerading support.
  69  *              Alan Cox        :       Multicast loopback error for 224.0.0.1
  70  *              Alan Cox        :       IP_MULTICAST_LOOP option.
  71  *              Alan Cox        :       Use notifiers.
  72  *              Bjorn Ekwall    :       Removed ip_csum (from slhc.c too)
  73  *              Bjorn Ekwall    :       Moved ip_fast_csum to ip.h (inline!)
  74  *              Stefan Becker   :       Send out ICMP HOST REDIRECT
  75  *      Arnt Gulbrandsen        :       ip_build_xmit
  76  *              Alan Cox        :       Per socket routing cache
  77  *              Alan Cox        :       Fixed routing cache, added header cache.
  78  *              Alan Cox        :       Loopback didn't work right in original ip_build_xmit - fixed it.
  79  *              Alan Cox        :       Only send ICMP_REDIRECT if src/dest are the same net.
  80  *              Alan Cox        :       Incoming IP option handling.
  81  *              Alan Cox        :       Set saddr on raw output frames as per BSD.
  82  *              Alan Cox        :       Stopped broadcast source route explosions.
  83  *              Alan Cox        :       Can disable source routing
  84  *              Takeshi Sone    :       Masquerading didn't work.
  85  *      Dave Bonn,Alan Cox      :       Faster IP forwarding whenever possible.
  86  *              Alan Cox        :       Memory leaks, tramples, misc debugging.
  87  *              Alan Cox        :       Fixed multicast (by popular demand 8))
  88  *              Alan Cox        :       Fixed forwarding (by even more popular demand 8))
  89  *              Alan Cox        :       Fixed SNMP statistics [I think]
  90  *      Gerhard Koerting        :       IP fragmentation forwarding fix
  91  *              Alan Cox        :       Device lock against page fault.
  92  *              Alan Cox        :       IP_HDRINCL facility.
  93  *      Werner Almesberger      :       Zero fragment bug
  94  *              Alan Cox        :       RAW IP frame length bug
  95  *              Alan Cox        :       Outgoing firewall on build_xmit
  96  *              A.N.Kuznetsov   :       IP_OPTIONS support throughout the kernel
  97  *              Alan Cox        :       Multicast routing hooks
  98  *              Jos Vos         :       Do accounting *before* call_in_firewall
  99  *      Willy Konynenberg       :       Transparent proxying support
 100  *
 101  *
 102  *
 103  * To Fix:
 104  *              IP fragmentation wants rewriting cleanly. The RFC815 algorithm is much more efficient
 105  *              and could be made very efficient with the addition of some virtual memory hacks to permit
 106  *              the allocation of a buffer that can then be 'grown' by twiddling page tables.
 107  *              Output fragmentation wants updating along with the buffer management to use a single
 108  *              interleaved copy algorithm so that fragmenting has a one copy overhead. Actual packet
 109  *              output should probably do its own fragmentation at the UDP/RAW layer. TCP shouldn't cause
 110  *              fragmentation anyway.
 111  *
 112  *              This program is free software; you can redistribute it and/or
 113  *              modify it under the terms of the GNU General Public License
 114  *              as published by the Free Software Foundation; either version
 115  *              2 of the License, or (at your option) any later version.
 116  */
 117
 118 #include <asm/system.h>
 119 #include <linux/module.h>
 120 #include <linux/types.h>
 121 #include <linux/kernel.h>
 122 #include <linux/string.h>
 123 #include <linux/errno.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  *      SNMP management statistics
 151  */
 152
 153 DEFINE_SNMP_STAT(struct ipstats_mib, ip_statistics) __read_mostly;
 154
 155 /*
 156  *      Process Router Attention IP option
 157  */
 158 int ip_call_ra_chain(struct sk_buff *skb)
 159 {
 160         struct ip_ra_chain *ra;
 161         u8 protocol = ip_hdr(skb)->protocol;
 162         struct sock *last = NULL;
 163
 164         read_lock(&ip_ra_lock);
 165         for (ra = ip_ra_chain; ra; ra = ra->next) {
 166                 struct sock *sk = ra->sk;
 167
 168                 /* If socket is bound to an interface, only report
 169                  * the packet if it came  from that interface.
 170                  */
 171                 if (sk && inet_sk(sk)->num == protocol &&
 172                     (!sk->sk_bound_dev_if ||
 173                      sk->sk_bound_dev_if == skb->dev->ifindex)) {
 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         __skb_pull(skb, ip_hdrlen(skb));
 201
 202         /* Point into the IP datagram, just past the header. */
 203         skb_reset_transport_header(skb);
 204
 205         rcu_read_lock();
 206         {
 207                 int protocol = ip_hdr(skb)->protocol;
 208                 int hash, raw;
 209                 struct net_protocol *ipprot;
 210
 211         resubmit:
 212                 raw = raw_local_deliver(skb, protocol);
 213
 214                 hash = protocol & (MAX_INET_PROTOS - 1);
 215                 if ((ipprot = rcu_dereference(inet_protos[hash])) != NULL) {
 216                         int ret;
 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(IPSTATS_MIB_INDELIVERS);
 231                 } else {
 232                         if (!raw) {
 233                                 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 234                                         IP_INC_STATS_BH(IPSTATS_MIB_INUNKNOWNPROTOS);
 235                                         icmp_send(skb, ICMP_DEST_UNREACH,
 236                                                   ICMP_PROT_UNREACH, 0);
 237                                 }
 238                         } else
 239                                 IP_INC_STATS_BH(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_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
 259                 if (ip_defrag(skb, IP_DEFRAG_LOCAL_DELIVER))
 260                         return 0;
 261         }
 262
 263         return NF_HOOK(PF_INET, NF_INET_LOCAL_IN, skb, skb->dev, NULL,
 264                        ip_local_deliver_finish);
 265 }
 266
 267 static inline int ip_rcv_options(struct sk_buff *skb)
 268 {
 269         struct ip_options *opt;
 270         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(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(opt, skb)) {
 290                 IP_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
 291                 goto drop;
 292         }
 293
 294         if (unlikely(opt->srr)) {
 295                 struct in_device *in_dev = in_dev_get(dev);
 296                 if (in_dev) {
 297                         if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
 298                                 if (IN_DEV_LOG_MARTIANS(in_dev) &&
 299                                     net_ratelimit())
 300                                         printk(KERN_INFO "source route option "
 301                                                "%u.%u.%u.%u -> %u.%u.%u.%u\n",
 302                                                NIPQUAD(iph->saddr),
 303                                                NIPQUAD(iph->daddr));
 304                                 in_dev_put(in_dev);
 305                                 goto drop;
 306                         }
 307
 308                         in_dev_put(in_dev);
 309                 }
 310
 311                 if (ip_options_rcv_srr(skb))
 312                         goto drop;
 313         }
 314
 315         return 0;
 316 drop:
 317         return -1;
 318 }
 319
 320 static int ip_rcv_finish(struct sk_buff *skb)
 321 {
 322         const struct iphdr *iph = ip_hdr(skb);
 323         struct rtable *rt;
 324
 325         /*
 326          *      Initialise the virtual path cache for the packet. It describes
 327          *      how the packet travels inside Linux networking.
 328          */
 329         if (skb->dst == NULL) {
 330                 int err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos,
 331                                          skb->dev);
 332                 if (unlikely(err)) {
 333                         if (err == -EHOSTUNREACH)
 334                                 IP_INC_STATS_BH(IPSTATS_MIB_INADDRERRORS);
 335                         else if (err == -ENETUNREACH)
 336                                 IP_INC_STATS_BH(IPSTATS_MIB_INNOROUTES);
 337                         goto drop;
 338                 }
 339         }
 340
 341 #ifdef CONFIG_NET_CLS_ROUTE
 342         if (unlikely(skb->dst->tclassid)) {
 343                 struct ip_rt_acct *st = per_cpu_ptr(ip_rt_acct, smp_processor_id());
 344                 u32 idx = skb->dst->tclassid;
 345                 st[idx&0xFF].o_packets++;
 346                 st[idx&0xFF].o_bytes+=skb->len;
 347                 st[(idx>>16)&0xFF].i_packets++;
 348                 st[(idx>>16)&0xFF].i_bytes+=skb->len;
 349         }
 350 #endif
 351
 352         if (iph->ihl > 5 && ip_rcv_options(skb))
 353                 goto drop;
 354
 355         rt = skb->rtable;
 356         if (rt->rt_type == RTN_MULTICAST)
 357                 IP_INC_STATS_BH(IPSTATS_MIB_INMCASTPKTS);
 358         else if (rt->rt_type == RTN_BROADCAST)
 359                 IP_INC_STATS_BH(IPSTATS_MIB_INBCASTPKTS);
 360
 361         return dst_input(skb);
 362
 363 drop:
 364         kfree_skb(skb);
 365         return NET_RX_DROP;
 366 }
 367
 368 /*
 369  *      Main IP Receive routine.
 370  */
 371 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 372 {
 373         struct iphdr *iph;
 374         u32 len;
 375
 376         if (dev->nd_net != &init_net)
 377                 goto drop;
 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: 3.1.2.2 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);