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 #include <linux/config.h>
 125
 126 #include <linux/net.h>
 127 #include <linux/socket.h>
 128 #include <linux/sockios.h>
 129 #include <linux/in.h>
 130 #include <linux/inet.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);
 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 = skb->nh.iph->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 (skb->nh.iph->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                         nf_reset(skb);
 188                 }
 189         }
 190
 191         if (last) {
 192                 raw_rcv(last, skb);
 193                 read_unlock(&ip_ra_lock);
 194                 return 1;
 195         }
 196         read_unlock(&ip_ra_lock);
 197         return 0;
 198 }
 199
 200 static inline int ip_local_deliver_finish(struct sk_buff *skb)
 201 {
 202         int ihl = skb->nh.iph->ihl*4;
 203
 204         __skb_pull(skb, ihl);
 205
 206         /* Free reference early: we don't need it any more, and it may
 207            hold ip_conntrack module loaded indefinitely. */
 208         nf_reset(skb);
 209
 210         /* Point into the IP datagram, just past the header. */
 211         skb->h.raw = skb->data;
 212
 213         rcu_read_lock();
 214         {
 215                 /* Note: See raw.c and net/raw.h, RAWV4_HTABLE_SIZE==MAX_INET_PROTOS */
 216                 int protocol = skb->nh.iph->protocol;
 217                 int hash;
 218                 struct sock *raw_sk;
 219                 struct net_protocol *ipprot;
 220
 221         resubmit:
 222                 hash = protocol & (MAX_INET_PROTOS - 1);
 223                 raw_sk = sk_head(&raw_v4_htable[hash]);
 224
 225                 /* If there maybe a raw socket we must check - if not we
 226                  * don't care less
 227                  */
 228                 if (raw_sk)
 229                         raw_v4_input(skb, skb->nh.iph, hash);
 230
 231                 if ((ipprot = rcu_dereference(inet_protos[hash])) != NULL) {
 232                         int ret;
 233
 234                         if (!ipprot->no_policy &&
 235                             !xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 236                                 kfree_skb(skb);
 237                                 goto out;
 238                         }
 239                         ret = ipprot->handler(skb);
 240                         if (ret < 0) {
 241                                 protocol = -ret;
 242                                 goto resubmit;
 243                         }
 244                         IP_INC_STATS_BH(IPSTATS_MIB_INDELIVERS);
 245                 } else {
 246                         if (!raw_sk) {
 247                                 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 248                                         IP_INC_STATS_BH(IPSTATS_MIB_INUNKNOWNPROTOS);
 249                                         icmp_send(skb, ICMP_DEST_UNREACH,
 250                                                   ICMP_PROT_UNREACH, 0);
 251                                 }
 252                         } else
 253                                 IP_INC_STATS_BH(IPSTATS_MIB_INDELIVERS);
 254                         kfree_skb(skb);
 255                 }
 256         }
 257  out:
 258         rcu_read_unlock();
 259
 260         return 0;
 261 }
 262
 263 /*
 264  *      Deliver IP Packets to the higher protocol layers.
 265  */
 266 int ip_local_deliver(struct sk_buff *skb)
 267 {
 268         /*
 269          *      Reassemble IP fragments.
 270          */
 271
 272         if (skb->nh.iph->frag_off & htons(IP_MF|IP_OFFSET)) {
 273                 skb = ip_defrag(skb, IP_DEFRAG_LOCAL_DELIVER);
 274                 if (!skb)
 275                         return 0;
 276         }
 277
 278         return NF_HOOK(PF_INET, NF_IP_LOCAL_IN, skb, skb->dev, NULL,
 279                        ip_local_deliver_finish);
 280 }
 281
 282 static inline int ip_rcv_finish(struct sk_buff *skb)
 283 {
 284         struct net_device *dev = skb->dev;
 285         struct iphdr *iph = skb->nh.iph;
 286
 287         /*
 288          *      Initialise the virtual path cache for the packet. It describes
 289          *      how the packet travels inside Linux networking.
 290          */
 291         if (skb->dst == NULL) {
 292                 if (ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))
 293                         goto drop;
 294         }
 295
 296 #ifdef CONFIG_NET_CLS_ROUTE
 297         if (skb->dst->tclassid) {
 298                 struct ip_rt_acct *st = ip_rt_acct + 256*smp_processor_id();
 299                 u32 idx = skb->dst->tclassid;
 300                 st[idx&0xFF].o_packets++;
 301                 st[idx&0xFF].o_bytes+=skb->len;
 302                 st[(idx>>16)&0xFF].i_packets++;
 303                 st[(idx>>16)&0xFF].i_bytes+=skb->len;
 304         }
 305 #endif
 306
 307         if (iph->ihl > 5) {
 308                 struct ip_options *opt;
 309
 310                 /* It looks as overkill, because not all
 311                    IP options require packet mangling.
 312                    But it is the easiest for now, especially taking
 313                    into account that combination of IP options
 314                    and running sniffer is extremely rare condition.
 315                                                       --ANK (980813)
 316                 */
 317
 318                 if (skb_cow(skb, skb_headroom(skb))) {
 319                         IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS);
 320                         goto drop;
 321                 }
 322                 iph = skb->nh.iph;
 323
 324                 if (ip_options_compile(NULL, skb))
 325                         goto inhdr_error;
 326
 327                 opt = &(IPCB(skb)->opt);
 328                 if (opt->srr) {
 329                         struct in_device *in_dev = in_dev_get(dev);
 330                         if (in_dev) {
 331                                 if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
 332                                         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
 333                                                 printk(KERN_INFO "source route option %u.%u.%u.%u -> %u.%u.%u.%u\n",
 334                                                        NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
 335                                         in_dev_put(in_dev);
 336                                         goto drop;
 337                                 }
 338                                 in_dev_put(in_dev);
 339                         }
 340                         if (ip_options_rcv_srr(skb))
 341                                 goto drop;
 342                 }
 343         }
 344
 345         return dst_input(skb);
 346
 347 inhdr_error:
 348         IP_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
 349 drop:
 350         kfree_skb(skb);
 351         return NET_RX_DROP;
 352 }
 353
 354 /*
 355  *      Main IP Receive routine.
 356  */
 357 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt)
 358 {
 359         struct iphdr *iph;
 360
 361         /* When the interface is in promisc. mode, drop all the crap
 362          * that it receives, do not try to analyse it.
 363          */
 364         if (skb->pkt_type == PACKET_OTHERHOST)
 365                 goto drop;
 366
 367         IP_INC_STATS_BH(IPSTATS_MIB_INRECEIVES);
 368
 369         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL) {
 370                 IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS);
 371                 goto out;
 372         }
 373
 374         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 375                 goto inhdr_error;
 376
 377         iph = skb->nh.iph;
 378
 379         /*
 380          *      RFC1122: 3.1.2.2 MUST silently discard any IP frame that fails the checksum.
 381          *
 382          *      Is the datagram acceptable?
 383          *
 384          *      1.      Length at least the size of an ip header
 385          *      2.      Version of 4
 386          *      3.      Checksums correctly. [Speed optimisation for later, skip loopback checksums]
 387          *      4.      Doesn't have a bogus length
 388          */
 389
 390         if (iph->ihl < 5 || iph->version != 4)
 391                 goto inhdr_error;
 392
 393         if (!pskb_may_pull(skb, iph->ihl*4))
 394                 goto inhdr_error;
 395
 396         iph = skb->nh.iph;
 397
 398         if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
 399                 goto inhdr_error;
 400
 401         {
 402                 __u32 len = ntohs(iph->tot_len);
 403                 if (skb->len < len || len < (iph->ihl<<2))
 404                         goto inhdr_error;
 405
 406                 /* Our transport medium may have padded the buffer out. Now we know it
 407                  * is IP we can trim to the true length of the frame.
 408                  * Note this now means skb->len holds ntohs(iph->tot_len).
 409                  */
 410                 if (pskb_trim_rcsum(skb, len)) {
 411                         IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS);
 412                         goto drop;
 413                 }
 414         }
 415
 416         return NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, dev, NULL,
 417                        ip_rcv_finish);
 418
 419 inhdr_error:
 420         IP_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
 421 drop:
 422         kfree_skb(skb);
 423 out:
 424         return NET_RX_DROP;
 425 }
 426
 427 EXPORT_SYMBOL(ip_rcv);
 428 EXPORT_SYMBOL(ip_statistics);