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         struct net_device *dev = skb->dev;
 164
 165         read_lock(&ip_ra_lock);
 166         for (ra = ip_ra_chain; ra; ra = ra->next) {
 167                 struct sock *sk = ra->sk;
 168
 169                 /* If socket is bound to an interface, only report
 170                  * the packet if it came  from that interface.
 171                  */
 172                 if (sk && inet_sk(sk)->num == protocol &&
 173                     (!sk->sk_bound_dev_if ||
 174                      sk->sk_bound_dev_if == dev->ifindex) &&
 175                     sk->sk_net == dev_net(dev)) {
 176                         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
 177                                 if (ip_defrag(skb, IP_DEFRAG_CALL_RA_CHAIN)) {
 178                                         read_unlock(&ip_ra_lock);
 179                                         return 1;
 180                                 }
 181                         }
 182                         if (last) {
 183                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
 184                                 if (skb2)
 185                                         raw_rcv(last, skb2);
 186                         }
 187                         last = sk;
 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 int ip_local_deliver_finish(struct sk_buff *skb)
 201 {
 202         struct net *net = dev_net(skb->dev);
 203
 204         __skb_pull(skb, ip_hdrlen(skb));
 205
 206         /* Point into the IP datagram, just past the header. */
 207         skb_reset_transport_header(skb);
 208
 209         rcu_read_lock();
 210         {
 211                 int protocol = ip_hdr(skb)->protocol;
 212                 int hash, raw;
 213                 struct net_protocol *ipprot;
 214
 215         resubmit:
 216                 raw = raw_local_deliver(skb, protocol);
 217
 218                 hash = protocol & (MAX_INET_PROTOS - 1);
 219                 ipprot = rcu_dereference(inet_protos[hash]);
 220                 if (ipprot != NULL && (net == &init_net || ipprot->netns_ok)) {
 221                         int ret;
 222
 223                         if (!ipprot->no_policy) {
 224                                 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 225                                         kfree_skb(skb);
 226                                         goto out;
 227                                 }
 228                                 nf_reset(skb);
 229                         }
 230                         ret = ipprot->handler(skb);
 231                         if (ret < 0) {
 232                                 protocol = -ret;
 233                                 goto resubmit;
 234                         }
 235                         IP_INC_STATS_BH(IPSTATS_MIB_INDELIVERS);
 236                 } else {
 237                         if (!raw) {
 238                                 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 239                                         IP_INC_STATS_BH(IPSTATS_MIB_INUNKNOWNPROTOS);
 240                                         icmp_send(skb, ICMP_DEST_UNREACH,
 241                                                   ICMP_PROT_UNREACH, 0);
 242                                 }
 243                         } else
 244                                 IP_INC_STATS_BH(IPSTATS_MIB_INDELIVERS);
 245                         kfree_skb(skb);
 246                 }
 247         }
 248  out:
 249         rcu_read_unlock();
 250
 251         return 0;
 252 }
 253
 254 /*
 255  *      Deliver IP Packets to the higher protocol layers.
 256  */
 257 int ip_local_deliver(struct sk_buff *skb)
 258 {
 259         /*
 260          *      Reassemble IP fragments.
 261          */
 262
 263         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
 264                 if (ip_defrag(skb, IP_DEFRAG_LOCAL_DELIVER))
 265                         return 0;
 266         }
 267
 268         return NF_HOOK(PF_INET, NF_INET_LOCAL_IN, skb, skb->dev, NULL,
 269                        ip_local_deliver_finish);
 270 }
 271
 272 static inline int ip_rcv_options(struct sk_buff *skb)
 273 {
 274         struct ip_options *opt;
 275         struct iphdr *iph;
 276         struct net_device *dev = skb->dev;
 277
 278         /* It looks as overkill, because not all
 279            IP options require packet mangling.
 280            But it is the easiest for now, especially taking
 281            into account that combination of IP options
 282            and running sniffer is extremely rare condition.
 283                                               --ANK (980813)
 284         */
 285         if (skb_cow(skb, skb_headroom(skb))) {
 286                 IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS);
 287                 goto drop;
 288         }
 289
 290         iph = ip_hdr(skb);
 291         opt = &(IPCB(skb)->opt);
 292         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
 293
 294         if (ip_options_compile(dev_net(dev), opt, skb)) {
 295                 IP_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
 296                 goto drop;
 297         }
 298
 299         if (unlikely(opt->srr)) {
 300                 struct in_device *in_dev = in_dev_get(dev);
 301                 if (in_dev) {
 302                         if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
 303                                 if (IN_DEV_LOG_MARTIANS(in_dev) &&
 304                                     net_ratelimit())
 305                                         printk(KERN_INFO "source route option "
 306                                                "%u.%u.%u.%u -> %u.%u.%u.%u\n",
 307                                                NIPQUAD(iph->saddr),
 308                                                NIPQUAD(iph->daddr));
 309                                 in_dev_put(in_dev);
 310                                 goto drop;
 311                         }
 312
 313                         in_dev_put(in_dev);
 314                 }
 315
 316                 if (ip_options_rcv_srr(skb))
 317                         goto drop;
 318         }
 319
 320         return 0;
 321 drop:
 322         return -1;
 323 }
 324
 325 static int ip_rcv_finish(struct sk_buff *skb)
 326 {
 327         const struct iphdr *iph = ip_hdr(skb);
 328         struct rtable *rt;
 329
 330         /*
 331          *      Initialise the virtual path cache for the packet. It describes
 332          *      how the packet travels inside Linux networking.
 333          */
 334         if (skb->dst == NULL) {
 335                 int err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos,
 336                                          skb->dev);
 337                 if (unlikely(err)) {
 338                         if (err == -EHOSTUNREACH)
 339                                 IP_INC_STATS_BH(IPSTATS_MIB_INADDRERRORS);
 340                         else if (err == -ENETUNREACH)
 341                                 IP_INC_STATS_BH(IPSTATS_MIB_INNOROUTES);
 342                         goto drop;
 343                 }
 344         }
 345
 346 #ifdef CONFIG_NET_CLS_ROUTE
 347         if (unlikely(skb->dst->tclassid)) {
 348                 struct ip_rt_acct *st = per_cpu_ptr(ip_rt_acct, smp_processor_id());
 349                 u32 idx = skb->dst->tclassid;
 350                 st[idx&0xFF].o_packets++;
 351                 st[idx&0xFF].o_bytes+=skb->len;
 352                 st[(idx>>16)&0xFF].i_packets++;
 353                 st[(idx>>16)&0xFF].i_bytes+=skb->len;
 354         }
 355 #endif
 356
 357         if (iph->ihl > 5 && ip_rcv_options(skb))
 358                 goto drop;
 359
 360         rt = skb->rtable;
 361         if (rt->rt_type == RTN_MULTICAST)
 362                 IP_INC_STATS_BH(IPSTATS_MIB_INMCASTPKTS);
 363         else if (rt->rt_type == RTN_BROADCAST)
 364                 IP_INC_STATS_BH(IPSTATS_MIB_INBCASTPKTS);
 365
 366         return dst_input(skb);
 367
 368 drop:
 369         kfree_skb(skb);
 370         return NET_RX_DROP;
 371 }
 372
 373 /*
 374  *      Main IP Receive routine.
 375  */
 376 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 377 {
 378         struct iphdr *iph;
 379         u32 len;
 380
 381         /* When the interface is in promisc. mode, drop all the crap
 382          * that it receives, do not try to analyse it.
 383          */
 384         if (skb->pkt_type == PACKET_OTHERHOST)
 385                 goto drop;
 386
 387         IP_INC_STATS_BH(IPSTATS_MIB_INRECEIVES);
 388
 389         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL) {
 390                 IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS);
 391                 goto out;
 392         }
 393
 394         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 395                 goto inhdr_error;
 396
 397         iph = ip_hdr(skb);
 398
 399         /*
 400          *      RFC1122: 3.1.2.2 MUST silently discard any IP frame that fails the checksum.
 401          *
 402          *      Is the datagram acceptable?
 403          *
 404          *      1.      Length at least the size of an ip header
 405          *      2.      Version of 4
 406          *      3.      Checksums correctly. [Speed optimisation for later, skip loopback checksums]
 407          *      4.      Doesn't have a bogus length
 408          */
 409
 410         if (iph->ihl < 5 || iph->version != 4)
 411                 goto inhdr_error;
 412
 413         if (!pskb_may_pull(skb, iph->ihl*4))
 414                 goto inhdr_error;
 415
 416         iph = ip_hdr(skb);
 417
 418         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 419                 goto inhdr_error;
 420
 421         len = ntohs(iph->tot_len);
 422         if (skb->len < len) {
 423                 IP_INC_STATS_BH(IPSTATS_MIB_INTRUNCATEDPKTS);
 424                 goto drop;
 425         } else if (len < (iph->ihl*4))
 426                 goto inhdr_error;
 427
 428         /* Our transport medium may have padded the buffer out. Now we know it
 429          * is IP we can trim to the true length of the frame.
 430          * Note this now means skb->len holds ntohs(iph->tot_len).
 431          */
 432         if (pskb_trim_rcsum(skb, len)) {
 433                 IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS);
 434                 goto drop;
 435         }
 436
 437         /* Remove any debris in the socket control block */
 438         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 439
 440         return NF_HOOK(PF_INET, NF_INET_PRE_ROUTING, skb, dev, NULL,
 441                        ip_rcv_finish);
 442
 443 inhdr_error:
 444         IP_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
 445 drop:
 446         kfree_skb(skb);
 447 out:
 448         return NET_RX_DROP;
 449 }
 450
 451 EXPORT_SYMBOL(ip_statistics);