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                                 skb = ip_defrag(skb, IP_DEFRAG_CALL_RA_CHAIN);
 176                                 if (skb == NULL) {
 177                                         read_unlock(&ip_ra_lock);
 178                                         return 1;
 179                                 }
 180                         }
 181                         if (last) {
 182                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
 183                                 if (skb2)
 184                                         raw_rcv(last, skb2);
 185                         }
 186                         last = sk;
 187                 }
 188         }
 189
 190         if (last) {
 191                 raw_rcv(last, skb);
 192                 read_unlock(&ip_ra_lock);
 193                 return 1;
 194         }
 195         read_unlock(&ip_ra_lock);
 196         return 0;
 197 }
 198
 199 static inline int ip_local_deliver_finish(struct sk_buff *skb)
 200 {
 201         __skb_pull(skb, ip_hdrlen(skb));
 202
 203         /* Point into the IP datagram, just past the header. */
 204         skb_reset_transport_header(skb);
 205
 206         rcu_read_lock();
 207         {
 208                 /* Note: See raw.c and net/raw.h, RAWV4_HTABLE_SIZE==MAX_INET_PROTOS */
 209                 int protocol = ip_hdr(skb)->protocol;
 210                 int hash;
 211                 struct sock *raw_sk;
 212                 struct net_protocol *ipprot;
 213
 214         resubmit:
 215                 hash = protocol & (MAX_INET_PROTOS - 1);
 216                 raw_sk = sk_head(&raw_v4_htable[hash]);
 217
 218                 /* If there maybe a raw socket we must check - if not we
 219                  * don't care less
 220                  */
 221                 if (raw_sk && !raw_v4_input(skb, ip_hdr(skb), hash))
 222                         raw_sk = NULL;
 223
 224                 if ((ipprot = rcu_dereference(inet_protos[hash])) != NULL) {
 225                         int ret;
 226
 227                         if (!ipprot->no_policy) {
 228                                 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 229                                         kfree_skb(skb);
 230                                         goto out;
 231                                 }
 232                                 nf_reset(skb);
 233                         }
 234                         ret = ipprot->handler(skb);
 235                         if (ret < 0) {
 236                                 protocol = -ret;
 237                                 goto resubmit;
 238                         }
 239                         IP_INC_STATS_BH(IPSTATS_MIB_INDELIVERS);
 240                 } else {
 241                         if (!raw_sk) {
 242                                 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 243                                         IP_INC_STATS_BH(IPSTATS_MIB_INUNKNOWNPROTOS);
 244                                         icmp_send(skb, ICMP_DEST_UNREACH,
 245                                                   ICMP_PROT_UNREACH, 0);
 246                                 }
 247                         } else
 248                                 IP_INC_STATS_BH(IPSTATS_MIB_INDELIVERS);
 249                         kfree_skb(skb);
 250                 }
 251         }
 252  out:
 253         rcu_read_unlock();
 254
 255         return 0;
 256 }
 257
 258 /*
 259  *      Deliver IP Packets to the higher protocol layers.
 260  */
 261 int ip_local_deliver(struct sk_buff *skb)
 262 {
 263         /*
 264          *      Reassemble IP fragments.
 265          */
 266
 267         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
 268                 skb = ip_defrag(skb, IP_DEFRAG_LOCAL_DELIVER);
 269                 if (!skb)
 270                         return 0;
 271         }
 272
 273         return NF_HOOK(PF_INET, NF_IP_LOCAL_IN, skb, skb->dev, NULL,
 274                        ip_local_deliver_finish);
 275 }
 276
 277 static inline int ip_rcv_options(struct sk_buff *skb)
 278 {
 279         struct ip_options *opt;
 280         struct iphdr *iph;
 281         struct net_device *dev = skb->dev;
 282
 283         /* It looks as overkill, because not all
 284            IP options require packet mangling.
 285            But it is the easiest for now, especially taking
 286            into account that combination of IP options
 287            and running sniffer is extremely rare condition.
 288                                               --ANK (980813)
 289         */
 290         if (skb_cow(skb, skb_headroom(skb))) {
 291                 IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS);
 292                 goto drop;
 293         }
 294
 295         iph = ip_hdr(skb);
 296
 297         if (ip_options_compile(NULL, skb)) {
 298                 IP_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
 299                 goto drop;
 300         }
 301
 302         opt = &(IPCB(skb)->opt);
 303         if (unlikely(opt->srr)) {
 304                 struct in_device *in_dev = in_dev_get(dev);
 305                 if (in_dev) {
 306                         if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
 307                                 if (IN_DEV_LOG_MARTIANS(in_dev) &&
 308                                     net_ratelimit())
 309                                         printk(KERN_INFO "source route option "
 310                                                "%u.%u.%u.%u -> %u.%u.%u.%u\n",
 311                                                NIPQUAD(iph->saddr),
 312                                                NIPQUAD(iph->daddr));
 313                                 in_dev_put(in_dev);
 314                                 goto drop;
 315                         }
 316
 317                         in_dev_put(in_dev);
 318                 }
 319
 320                 if (ip_options_rcv_srr(skb))
 321                         goto drop;
 322         }
 323
 324         return 0;
 325 drop:
 326         return -1;
 327 }
 328
 329 static inline int ip_rcv_finish(struct sk_buff *skb)
 330 {
 331         const struct iphdr *iph = ip_hdr(skb);
 332
 333         /*
 334          *      Initialise the virtual path cache for the packet. It describes
 335          *      how the packet travels inside Linux networking.
 336          */
 337         if (skb->dst == NULL) {
 338                 int err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos,
 339                                          skb->dev);
 340                 if (unlikely(err)) {
 341                         if (err == -EHOSTUNREACH)
 342                                 IP_INC_STATS_BH(IPSTATS_MIB_INADDRERRORS);
 343                         goto drop;
 344                 }
 345         }
 346
 347 #ifdef CONFIG_NET_CLS_ROUTE
 348         if (unlikely(skb->dst->tclassid)) {
 349                 struct ip_rt_acct *st = ip_rt_acct + 256*smp_processor_id();
 350                 u32 idx = skb->dst->tclassid;
 351                 st[idx&0xFF].o_packets++;
 352                 st[idx&0xFF].o_bytes+=skb->len;
 353                 st[(idx>>16)&0xFF].i_packets++;
 354                 st[(idx>>16)&0xFF].i_bytes+=skb->len;
 355         }
 356 #endif
 357
 358         if (iph->ihl > 5 && ip_rcv_options(skb))
 359                 goto drop;
 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         /* When the interface is in promisc. mode, drop all the crap
 377          * that it receives, do not try to analyse it.
 378          */
 379         if (skb->pkt_type == PACKET_OTHERHOST)
 380                 goto drop;
 381
 382         IP_INC_STATS_BH(IPSTATS_MIB_INRECEIVES);
 383
 384         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL) {
 385                 IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS);
 386                 goto out;
 387         }
 388
 389         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 390                 goto inhdr_error;
 391
 392         iph = ip_hdr(skb);
 393
 394         /*
 395          *      RFC1122: 3.1.2.2 MUST silently discard any IP frame that fails the checksum.
 396          *
 397          *      Is the datagram acceptable?
 398          *
 399          *      1.      Length at least the size of an ip header
 400          *      2.      Version of 4
 401          *      3.      Checksums correctly. [Speed optimisation for later, skip loopback checksums]
 402          *      4.      Doesn't have a bogus length
 403          */
 404
 405         if (iph->ihl < 5 || iph->version != 4)
 406                 goto inhdr_error;
 407
 408         if (!pskb_may_pull(skb, iph->ihl*4))
 409                 goto inhdr_error;
 410
 411         iph = ip_hdr(skb);
 412
 413         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 414                 goto inhdr_error;
 415
 416         len = ntohs(iph->tot_len);
 417         if (skb->len < len || len < (iph->ihl*4))
 418                 goto inhdr_error;
 419
 420         /* Our transport medium may have padded the buffer out. Now we know it
 421          * is IP we can trim to the true length of the frame.
 422          * Note this now means skb->len holds ntohs(iph->tot_len).
 423          */
 424         if (pskb_trim_rcsum(skb, len)) {
 425                 IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS);
 426                 goto drop;
 427         }
 428
 429         /* Remove any debris in the socket control block */
 430         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 431
 432         return NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, dev, NULL,
 433                        ip_rcv_finish);
 434
 435 inhdr_error:
 436         IP_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
 437 drop:
 438         kfree_skb(skb);
 439 out:
 440         return NET_RX_DROP;
 441 }
 442
 443 EXPORT_SYMBOL(ip_statistics);