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 1.40 1999/06/09 10:10:55 davem Exp $
   9  *
  10  * Authors:     Ross Biro, <bir7@leland.Stanford.Edu>
  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/types.h>
 120 #include <linux/kernel.h>
 121 #include <linux/string.h>
 122 #include <linux/errno.h>
 123 #include <linux/config.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/netdevice.h>
 131 #include <linux/etherdevice.h>
 132
 133 #include <net/snmp.h>
 134 #include <net/ip.h>
 135 #include <net/protocol.h>
 136 #include <net/route.h>
 137 #include <linux/skbuff.h>
 138 #include <net/sock.h>
 139 #include <net/arp.h>
 140 #include <net/icmp.h>
 141 #include <net/raw.h>
 142 #include <net/checksum.h>
 143 #include <linux/ip_fw.h>
 144 #ifdef CONFIG_IP_MASQUERADE
 145 #include <net/ip_masq.h>
 146 #endif
 147 #include <linux/firewall.h>
 148 #include <linux/mroute.h>
 149 #include <linux/netlink.h>
 150
 151 /*
 152  *      SNMP management statistics
 153  */
 154
 155 struct ip_mib ip_statistics={2,IPDEFTTL,};      /* Forwarding=No, Default TTL=64 */
 156
 157 #if defined(CONFIG_IP_TRANSPARENT_PROXY) && !defined(CONFIG_IP_ALWAYS_DEFRAG)
 158 #define CONFIG_IP_ALWAYS_DEFRAG 1
 159 #endif
 160
 161 /*
 162  *      Process Router Attention IP option
 163  */
 164 int ip_call_ra_chain(struct sk_buff *skb)
 165 {
 166         struct ip_ra_chain *ra;
 167         u8 protocol = skb->nh.iph->protocol;
 168         struct sock *last = NULL;
 169
 170         for (ra = ip_ra_chain; ra; ra = ra->next) {
 171                 struct sock *sk = ra->sk;
 172                 if (sk && sk->num == protocol) {
 173                         if (skb->nh.iph->frag_off & htons(IP_MF|IP_OFFSET)) {
 174                                 skb = ip_defrag(skb);
 175                                 if (skb == NULL)
 176                                         return 1;
 177                         }
 178                         if (last) {
 179                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
 180                                 if (skb2)
 181                                         raw_rcv(last, skb2);
 182                         }
 183                         last = sk;
 184                 }
 185         }
 186
 187         if (last) {
 188                 raw_rcv(last, skb);
 189                 return 1;
 190         }
 191         return 0;
 192 }
 193
 194 /* Handle this out of line, it is rare. */
 195 static int ip_run_ipprot(struct sk_buff *skb, struct iphdr *iph,
 196                          struct inet_protocol *ipprot, int force_copy)
 197 {
 198         int ret = 0;
 199
 200         do {
 201                 if (ipprot->protocol == iph->protocol) {
 202                         struct sk_buff *skb2 = skb;
 203                         if (ipprot->copy || force_copy)
 204                                 skb2 = skb_clone(skb, GFP_ATOMIC);
 205                         if(skb2 != NULL) {
 206                                 ret = 1;
 207                                 ipprot->handler(skb2,
 208                                                 ntohs(iph->tot_len) - (iph->ihl * 4));
 209                         }
 210                 }
 211                 ipprot = (struct inet_protocol *) ipprot->next;
 212         } while(ipprot != NULL);
 213
 214         return ret;
 215 }
 216
 217 extern struct sock *raw_v4_input(struct sk_buff *, struct iphdr *, int);
 218
 219 /*
 220  *      Deliver IP Packets to the higher protocol layers.
 221  */
 222 int ip_local_deliver(struct sk_buff *skb)
 223 {
 224         struct iphdr *iph = skb->nh.iph;
 225
 226 #ifndef CONFIG_IP_ALWAYS_DEFRAG
 227         /*
 228          *      Reassemble IP fragments.
 229          */
 230
 231         if (iph->frag_off & htons(IP_MF|IP_OFFSET)) {
 232                 skb = ip_defrag(skb);
 233                 if (!skb)
 234                         return 0;
 235                 iph = skb->nh.iph;
 236         }
 237 #endif
 238
 239 #ifdef CONFIG_IP_MASQUERADE
 240         /* Do we need to de-masquerade this packet? */
 241         if((IPCB(skb)->flags&IPSKB_MASQUERADED)) {
 242                 /* Some masq modules can re-inject packets if
 243                  * bad configured.
 244                  */
 245                 printk(KERN_DEBUG "ip_input(): demasq recursion detected. "
 246                        "Check masq modules configuration\n");
 247                 kfree_skb(skb);
 248                 return 0;
 249         } else {
 250                 int ret = ip_fw_demasquerade(&skb);
 251
 252                 if (ret < 0) {
 253                         kfree_skb(skb);
 254                         return 0;
 255                 }
 256                 if (ret) {
 257                         iph = skb->nh.iph;
 258                         IPCB(skb)->flags |= IPSKB_MASQUERADED;
 259                         dst_release(skb->dst);
 260                         skb->dst = NULL;
 261                         if (ip_route_input(skb, iph->daddr, iph->saddr,
 262                                            iph->tos, skb->dev)) {
 263                                 kfree_skb(skb);
 264                                 return 0;
 265                         }
 266                         return skb->dst->input(skb);
 267                 }
 268         }
 269 #endif
 270
 271         /* Point into the IP datagram, just past the header. */
 272         skb->h.raw = skb->nh.raw + iph->ihl*4;
 273
 274         {
 275                 /* Note: See raw.c and net/raw.h, RAWV4_HTABLE_SIZE==MAX_INET_PROTOS */
 276                 int hash = iph->protocol & (MAX_INET_PROTOS - 1);
 277                 struct sock *raw_sk = raw_v4_htable[hash];
 278                 struct inet_protocol *ipprot;
 279                 int flag;
 280
 281                 /* If there maybe a raw socket we must check - if not we
 282                  * don't care less
 283                  */
 284                 if(raw_sk != NULL)
 285                         raw_sk = raw_v4_input(skb, iph, hash);
 286
 287                 ipprot = (struct inet_protocol *) inet_protos[hash];
 288                 flag = 0;
 289                 if(ipprot != NULL) {
 290                         if(raw_sk == NULL &&
 291                            ipprot->next == NULL &&
 292                            ipprot->protocol == iph->protocol) {
 293                                 /* Fast path... */
 294                                 return ipprot->handler(skb, (ntohs(iph->tot_len) -
 295                                                              (iph->ihl * 4)));
 296                         } else {
 297                                 flag = ip_run_ipprot(skb, iph, ipprot, (raw_sk != NULL));
 298                         }
 299                 }
 300
 301                 /* All protocols checked.
 302                  * If this packet was a broadcast, we may *not* reply to it, since that
 303                  * causes (proven, grin) ARP storms and a leakage of memory (i.e. all
 304                  * ICMP reply messages get queued up for transmission...)
 305                  */
 306                 if(raw_sk != NULL) {    /* Shift to last raw user */
 307                         raw_rcv(raw_sk, skb);
 308                 } else if (!flag) {             /* Free and report errors */
 309                         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PROT_UNREACH, 0);
 310                         kfree_skb(skb);
 311                 }
 312         }
 313
 314         return 0;
 315 }
 316
 317 /*
 318  *      Main IP Receive routine.
 319  */
 320 int ip_rcv(struct sk_buff *skb, struct device *dev, struct packet_type *pt)
 321 {
 322         struct iphdr *iph = skb->nh.iph;
 323 #ifdef  CONFIG_FIREWALL
 324         int fwres;
 325         u16 rport;
 326 #endif /* CONFIG_FIREWALL */
 327
 328         /* When the interface is in promisc. mode, drop all the crap
 329          * that it receives, do not try to analyse it.
 330          */
 331         if (skb->pkt_type == PACKET_OTHERHOST)
 332                 goto drop;
 333
 334         ip_statistics.IpInReceives++;
 335
 336         /*
 337          *      RFC1122: 3.1.2.2 MUST silently discard any IP frame that fails the checksum.
 338          *
 339          *      Is the datagram acceptable?
 340          *
 341          *      1.      Length at least the size of an ip header
 342          *      2.      Version of 4
 343          *      3.      Checksums correctly. [Speed optimisation for later, skip loopback checksums]
 344          *      4.      Doesn't have a bogus length
 345          */
 346
 347         if (skb->len < sizeof(struct iphdr))
 348                 goto inhdr_error;
 349         if (iph->ihl < 5 || iph->version != 4 || ip_fast_csum((u8 *)iph, iph->ihl) != 0)
 350                 goto inhdr_error;
 351
 352         {
 353                 __u32 len = ntohs(iph->tot_len);
 354                 if (skb->len < len)
 355                         goto inhdr_error;
 356
 357                 /* Our transport medium may have padded the buffer out. Now we know it
 358                  * is IP we can trim to the true length of the frame.
 359                  * Note this now means skb->len holds ntohs(iph->tot_len).
 360                  */
 361                 __skb_trim(skb, len);
 362         }
 363
 364 #ifdef CONFIG_IP_ALWAYS_DEFRAG
 365         /* Won't send ICMP reply, since skb->dst == NULL. --RR */
 366         if (iph->frag_off & htons(IP_MF|IP_OFFSET)) {
 367                 skb = ip_defrag(skb);
 368                 if (!skb)
 369                         return 0;
 370                 iph = skb->nh.iph;
 371                 ip_send_check(iph);
 372         }
 373 #endif
 374
 375 #ifdef CONFIG_FIREWALL
 376         /*
 377          *      See if the firewall wants to dispose of the packet.
 378          *
 379          * We can't do ICMP reply or local delivery before routing,
 380          * so we delay those decisions until after route. --RR
 381          */
 382         fwres = call_in_firewall(PF_INET, dev, iph, &rport, &skb);
 383         if (fwres < FW_ACCEPT && fwres != FW_REJECT)
 384                 goto drop;
 385         iph = skb->nh.iph;
 386 #endif /* CONFIG_FIREWALL */
 387
 388         /*
 389          *      Initialise the virtual path cache for the packet. It describes
 390          *      how the packet travels inside Linux networking.
 391          */
 392         if (skb->dst == NULL) {
 393                 if (ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))
 394                         goto drop;
 395         }
 396
 397 #ifdef CONFIG_NET_CLS_ROUTE
 398         if (skb->dst->tclassid) {
 399                 u32 idx = skb->dst->tclassid;
 400                 write_lock(&ip_rt_acct_lock);
 401                 ip_rt_acct[idx&0xFF].o_packets++;
 402                 ip_rt_acct[idx&0xFF].o_bytes+=skb->len;
 403                 ip_rt_acct[(idx>>16)&0xFF].i_packets++;
 404                 ip_rt_acct[(idx>>16)&0xFF].i_bytes+=skb->len;
 405                 write_unlock(&ip_rt_acct_lock);
 406         }
 407 #endif
 408
 409         if (iph->ihl > 5) {
 410                 struct ip_options *opt;
 411
 412                 /* It looks as overkill, because not all
 413                    IP options require packet mangling.
 414                    But it is the easiest for now, especially taking
 415                    into account that combination of IP options
 416                    and running sniffer is extremely rare condition.
 417                                                       --ANK (980813)
 418                 */
 419
 420                 skb = skb_cow(skb, skb_headroom(skb));
 421                 if (skb == NULL)
 422                         return 0;
 423                 iph = skb->nh.iph;
 424
 425                 skb->ip_summed = 0;
 426                 if (ip_options_compile(NULL, skb))
 427                         goto inhdr_error;
 428
 429                 opt = &(IPCB(skb)->opt);
 430                 if (opt->srr) {
 431                         struct in_device *in_dev = dev->ip_ptr;
 432                         if (in_dev && !IN_DEV_SOURCE_ROUTE(in_dev)) {
 433                                 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
 434                                         printk(KERN_INFO "source route option %d.%d.%d.%d -> %d.%d.%d.%d\n",
 435                                                NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
 436                                 goto drop;
 437                         }
 438                         if (ip_options_rcv_srr(skb))
 439                                 goto drop;
 440                 }
 441         }
 442
 443 #ifdef CONFIG_FIREWALL
 444 #ifdef  CONFIG_IP_TRANSPARENT_PROXY
 445         if (fwres == FW_REDIRECT && (IPCB(skb)->redirport = rport) != 0)
 446                 return ip_local_deliver(skb);
 447 #endif /* CONFIG_IP_TRANSPARENT_PROXY */
 448
 449         if (fwres == FW_REJECT) {
 450                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 451                 goto drop;
 452         }
 453 #endif /* CONFIG_FIREWALL */
 454
 455         return skb->dst->input(skb);
 456
 457 inhdr_error:
 458         ip_statistics.IpInHdrErrors++;
 459 drop:
 460         kfree_skb(skb);
 461         return(0);
 462 }
 463