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  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Donald Becker, <becker@super.org>
  11  *              Alan Cox, <alan@lxorguk.ukuu.org.uk>
  12  *              Richard Underwood
  13  *              Stefan Becker, <stefanb@yello.ping.de>
  14  *              Jorge Cwik, <jorge@laser.satlink.net>
  15  *              Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  16  *
  17  *
  18  * Fixes:
  19  *              Alan Cox        :       Commented a couple of minor bits of surplus code
  20  *              Alan Cox        :       Undefining IP_FORWARD doesn't include the code
  21  *                                      (just stops a compiler warning).
  22  *              Alan Cox        :       Frames with >=MAX_ROUTE record routes, strict routes or loose routes
  23  *                                      are junked rather than corrupting things.
  24  *              Alan Cox        :       Frames to bad broadcast subnets are dumped
  25  *                                      We used to process them non broadcast and
  26  *                                      boy could that cause havoc.
  27  *              Alan Cox        :       ip_forward sets the free flag on the
  28  *                                      new frame it queues. Still crap because
  29  *                                      it copies the frame but at least it
  30  *                                      doesn't eat memory too.
  31  *              Alan Cox        :       Generic queue code and memory fixes.
  32  *              Fred Van Kempen :       IP fragment support (borrowed from NET2E)
  33  *              Gerhard Koerting:       Forward fragmented frames correctly.
  34  *              Gerhard Koerting:       Fixes to my fix of the above 8-).
  35  *              Gerhard Koerting:       IP interface addressing fix.
  36  *              Linus Torvalds  :       More robustness checks
  37  *              Alan Cox        :       Even more checks: Still not as robust as it ought to be
  38  *              Alan Cox        :       Save IP header pointer for later
  39  *              Alan Cox        :       ip option setting
  40  *              Alan Cox        :       Use ip_tos/ip_ttl settings
  41  *              Alan Cox        :       Fragmentation bogosity removed
  42  *                                      (Thanks to Mark.Bush@prg.ox.ac.uk)
  43  *              Dmitry Gorodchanin :    Send of a raw packet crash fix.
  44  *              Alan Cox        :       Silly ip bug when an overlength
  45  *                                      fragment turns up. Now frees the
  46  *                                      queue.
  47  *              Linus Torvalds/ :       Memory leakage on fragmentation
  48  *              Alan Cox        :       handling.
  49  *              Gerhard Koerting:       Forwarding uses IP priority hints
  50  *              Teemu Rantanen  :       Fragment problems.
  51  *              Alan Cox        :       General cleanup, comments and reformat
  52  *              Alan Cox        :       SNMP statistics
  53  *              Alan Cox        :       BSD address rule semantics. Also see
  54  *                                      UDP as there is a nasty checksum issue
  55  *                                      if you do things the wrong way.
  56  *              Alan Cox        :       Always defrag, moved IP_FORWARD to the config.in file
  57  *              Alan Cox        :       IP options adjust sk->priority.
  58  *              Pedro Roque     :       Fix mtu/length error in ip_forward.
  59  *              Alan Cox        :       Avoid ip_chk_addr when possible.
  60  *      Richard Underwood       :       IP multicasting.
  61  *              Alan Cox        :       Cleaned up multicast handlers.
  62  *              Alan Cox        :       RAW sockets demultiplex in the BSD style.
  63  *              Gunther Mayer   :       Fix the SNMP reporting typo
  64  *              Alan Cox        :       Always in group 224.0.0.1
  65  *      Pauline Middelink       :       Fast ip_checksum update when forwarding
  66  *                                      Masquerading support.
  67  *              Alan Cox        :       Multicast loopback error for 224.0.0.1
  68  *              Alan Cox        :       IP_MULTICAST_LOOP option.
  69  *              Alan Cox        :       Use notifiers.
  70  *              Bjorn Ekwall    :       Removed ip_csum (from slhc.c too)
  71  *              Bjorn Ekwall    :       Moved ip_fast_csum to ip.h (inline!)
  72  *              Stefan Becker   :       Send out ICMP HOST REDIRECT
  73  *      Arnt Gulbrandsen        :       ip_build_xmit
  74  *              Alan Cox        :       Per socket routing cache
  75  *              Alan Cox        :       Fixed routing cache, added header cache.
  76  *              Alan Cox        :       Loopback didn't work right in original ip_build_xmit - fixed it.
  77  *              Alan Cox        :       Only send ICMP_REDIRECT if src/dest are the same net.
  78  *              Alan Cox        :       Incoming IP option handling.
  79  *              Alan Cox        :       Set saddr on raw output frames as per BSD.
  80  *              Alan Cox        :       Stopped broadcast source route explosions.
  81  *              Alan Cox        :       Can disable source routing
  82  *              Takeshi Sone    :       Masquerading didn't work.
  83  *      Dave Bonn,Alan Cox      :       Faster IP forwarding whenever possible.
  84  *              Alan Cox        :       Memory leaks, tramples, misc debugging.
  85  *              Alan Cox        :       Fixed multicast (by popular demand 8))
  86  *              Alan Cox        :       Fixed forwarding (by even more popular demand 8))
  87  *              Alan Cox        :       Fixed SNMP statistics [I think]
  88  *      Gerhard Koerting        :       IP fragmentation forwarding fix
  89  *              Alan Cox        :       Device lock against page fault.
  90  *              Alan Cox        :       IP_HDRINCL facility.
  91  *      Werner Almesberger      :       Zero fragment bug
  92  *              Alan Cox        :       RAW IP frame length bug
  93  *              Alan Cox        :       Outgoing firewall on build_xmit
  94  *              A.N.Kuznetsov   :       IP_OPTIONS support throughout the kernel
  95  *              Alan Cox        :       Multicast routing hooks
  96  *              Jos Vos         :       Do accounting *before* call_in_firewall
  97  *      Willy Konynenberg       :       Transparent proxying support
  98  *
  99  *
 100  *
 101  * To Fix:
 102  *              IP fragmentation wants rewriting cleanly. The RFC815 algorithm is much more efficient
 103  *              and could be made very efficient with the addition of some virtual memory hacks to permit
 104  *              the allocation of a buffer that can then be 'grown' by twiddling page tables.
 105  *              Output fragmentation wants updating along with the buffer management to use a single
 106  *              interleaved copy algorithm so that fragmenting has a one copy overhead. Actual packet
 107  *              output should probably do its own fragmentation at the UDP/RAW layer. TCP shouldn't cause
 108  *              fragmentation anyway.
 109  *
 110  *              This program is free software; you can redistribute it and/or
 111  *              modify it under the terms of the GNU General Public License
 112  *              as published by the Free Software Foundation; either version
 113  *              2 of the License, or (at your option) any later version.
 114  */
 115
 116 #define pr_fmt(fmt) "IPv4: " fmt
 117
 118 #include <linux/module.h>
 119 #include <linux/types.h>
 120 #include <linux/kernel.h>
 121 #include <linux/string.h>
 122 #include <linux/errno.h>
 123 #include <linux/slab.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  *      Process Router Attention IP option (RFC 2113)
 151  */
 152 bool ip_call_ra_chain(struct sk_buff *skb)
 153 {
 154         struct ip_ra_chain *ra;
 155         u8 protocol = ip_hdr(skb)->protocol;
 156         struct sock *last = NULL;
 157         struct net_device *dev = skb->dev;
 158
 159         for (ra = rcu_dereference(ip_ra_chain); ra; ra = rcu_dereference(ra->next)) {
 160                 struct sock *sk = ra->sk;
 161
 162                 /* If socket is bound to an interface, only report
 163                  * the packet if it came  from that interface.
 164                  */
 165                 if (sk && inet_sk(sk)->inet_num == protocol &&
 166                     (!sk->sk_bound_dev_if ||
 167                      sk->sk_bound_dev_if == dev->ifindex) &&
 168                     net_eq(sock_net(sk), dev_net(dev))) {
 169                         if (ip_is_fragment(ip_hdr(skb))) {
 170                                 if (ip_defrag(skb, IP_DEFRAG_CALL_RA_CHAIN))
 171                                         return true;
 172                         }
 173                         if (last) {
 174                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
 175                                 if (skb2)
 176                                         raw_rcv(last, skb2);
 177                         }
 178                         last = sk;
 179                 }
 180         }
 181
 182         if (last) {
 183                 raw_rcv(last, skb);
 184                 return true;
 185         }
 186         return false;
 187 }
 188
 189 static int ip_local_deliver_finish(struct sk_buff *skb)
 190 {
 191         struct net *net = dev_net(skb->dev);
 192
 193         __skb_pull(skb, skb_network_header_len(skb));
 194
 195         rcu_read_lock();
 196         {
 197                 int protocol = ip_hdr(skb)->protocol;
 198                 const struct net_protocol *ipprot;
 199                 int raw;
 200
 201         resubmit:
 202                 raw = raw_local_deliver(skb, protocol);
 203
 204                 ipprot = rcu_dereference(inet_protos[protocol]);
 205                 if (ipprot != NULL) {
 206                         int ret;
 207
 208                         if (!ipprot->no_policy) {
 209                                 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 210                                         kfree_skb(skb);
 211                                         goto out;
 212                                 }
 213                                 nf_reset(skb);
 214                         }
 215                         ret = ipprot->handler(skb);
 216                         if (ret < 0) {
 217                                 protocol = -ret;
 218                                 goto resubmit;
 219                         }
 220                         IP_INC_STATS_BH(net, IPSTATS_MIB_INDELIVERS);
 221                 } else {
 222                         if (!raw) {
 223                                 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 224                                         IP_INC_STATS_BH(net, IPSTATS_MIB_INUNKNOWNPROTOS);
 225                                         icmp_send(skb, ICMP_DEST_UNREACH,
 226                                                   ICMP_PROT_UNREACH, 0);
 227                                 }
 228                                 kfree_skb(skb);
 229                         } else {
 230                                 IP_INC_STATS_BH(net, IPSTATS_MIB_INDELIVERS);
 231                                 consume_skb(skb);
 232                         }
 233                 }
 234         }
 235  out:
 236         rcu_read_unlock();
 237
 238         return 0;
 239 }
 240
 241 /*
 242  *      Deliver IP Packets to the higher protocol layers.
 243  */
 244 int ip_local_deliver(struct sk_buff *skb)
 245 {
 246         /*
 247          *      Reassemble IP fragments.
 248          */
 249
 250         if (ip_is_fragment(ip_hdr(skb))) {
 251                 if (ip_defrag(skb, IP_DEFRAG_LOCAL_DELIVER))
 252                         return 0;
 253         }
 254
 255         return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN, skb, skb->dev, NULL,
 256                        ip_local_deliver_finish);
 257 }
 258
 259 static inline bool ip_rcv_options(struct sk_buff *skb)
 260 {
 261         struct ip_options *opt;
 262         const struct iphdr *iph;
 263         struct net_device *dev = skb->dev;
 264
 265         /* It looks as overkill, because not all
 266            IP options require packet mangling.
 267            But it is the easiest for now, especially taking
 268            into account that combination of IP options
 269            and running sniffer is extremely rare condition.
 270                                               --ANK (980813)
 271         */
 272         if (skb_cow(skb, skb_headroom(skb))) {
 273                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 274                 goto drop;
 275         }
 276
 277         iph = ip_hdr(skb);
 278         opt = &(IPCB(skb)->opt);
 279         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
 280
 281         if (ip_options_compile(dev_net(dev), opt, skb)) {
 282                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 283                 goto drop;
 284         }
 285
 286         if (unlikely(opt->srr)) {
 287                 struct in_device *in_dev = __in_dev_get_rcu(dev);
 288
 289                 if (in_dev) {
 290                         if (!IN_DEV_SOURCE_ROUTE(in_dev)) {
 291                                 if (IN_DEV_LOG_MARTIANS(in_dev))
 292                                         net_info_ratelimited("source route option %pI4 -> %pI4\n",
 293                                                              &iph->saddr,
 294                                                              &iph->daddr);
 295                                 goto drop;
 296                         }
 297                 }
 298
 299                 if (ip_options_rcv_srr(skb))
 300                         goto drop;
 301         }
 302
 303         return false;
 304 drop:
 305         return true;
 306 }
 307
 308 int sysctl_ip_early_demux __read_mostly = 1;
 309 EXPORT_SYMBOL(sysctl_ip_early_demux);
 310
 311 static int ip_rcv_finish(struct sk_buff *skb)
 312 {
 313         const struct iphdr *iph = ip_hdr(skb);
 314         struct rtable *rt;
 315
 316         if (sysctl_ip_early_demux && !skb_dst(skb) && skb->sk == NULL) {
 317                 const struct net_protocol *ipprot;
 318                 int protocol = iph->protocol;
 319
 320                 ipprot = rcu_dereference(inet_protos[protocol]);
 321                 if (ipprot && ipprot->early_demux) {
 322                         ipprot->early_demux(skb);
 323                         /* must reload iph, skb->head might have changed */
 324                         iph = ip_hdr(skb);
 325                 }
 326         }
 327
 328         /*
 329          *      Initialise the virtual path cache for the packet. It describes
 330          *      how the packet travels inside Linux networking.
 331          */
 332         if (!skb_dst(skb)) {
 333                 int err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
 334                                                iph->tos, skb->dev);
 335                 if (unlikely(err)) {
 336                         if (err == -EXDEV)
 337                                 NET_INC_STATS_BH(dev_net(skb->dev),
 338                                                  LINUX_MIB_IPRPFILTER);
 339                         goto drop;
 340                 }
 341         }
 342
 343 #ifdef CONFIG_IP_ROUTE_CLASSID
 344         if (unlikely(skb_dst(skb)->tclassid)) {
 345                 struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct);
 346                 u32 idx = skb_dst(skb)->tclassid;
 347                 st[idx&0xFF].o_packets++;
 348                 st[idx&0xFF].o_bytes += skb->len;
 349                 st[(idx>>16)&0xFF].i_packets++;
 350                 st[(idx>>16)&0xFF].i_bytes += skb->len;
 351         }
 352 #endif
 353
 354         if (iph->ihl > 5 && ip_rcv_options(skb))
 355                 goto drop;
 356
 357         rt = skb_rtable(skb);
 358         if (rt->rt_type == RTN_MULTICAST) {
 359                 IP_UPD_PO_STATS_BH(dev_net(rt->dst.dev), IPSTATS_MIB_INMCAST,
 360                                 skb->len);
 361         } else if (rt->rt_type == RTN_BROADCAST)
 362                 IP_UPD_PO_STATS_BH(dev_net(rt->dst.dev), IPSTATS_MIB_INBCAST,
 363                                 skb->len);
 364
 365         return dst_input(skb);
 366
 367 drop:
 368         kfree_skb(skb);
 369         return NET_RX_DROP;
 370 }
 371
 372 /*
 373  *      Main IP Receive routine.
 374  */
 375 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 376 {
 377         const struct iphdr *iph;
 378         u32 len;
 379
 380         /* When the interface is in promisc. mode, drop all the crap
 381          * that it receives, do not try to analyse it.
 382          */
 383         if (skb->pkt_type == PACKET_OTHERHOST)
 384                 goto drop;
 385
 386
 387         IP_UPD_PO_STATS_BH(dev_net(dev), IPSTATS_MIB_IN, skb->len);
 388
 389         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL) {
 390                 IP_INC_STATS_BH(dev_net(dev), 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: 3.2.1.2 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 csum_error;
 420
 421         len = ntohs(iph->tot_len);
 422         if (skb->len < len) {
 423                 IP_INC_STATS_BH(dev_net(dev), 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(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 434                 goto drop;
 435         }
 436
 437         skb->transport_header = skb->network_header + iph->ihl*4;
 438
 439         /* Remove any debris in the socket control block */
 440         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 441
 442         /* Must drop socket now because of tproxy. */
 443         skb_orphan(skb);
 444
 445         return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, dev, NULL,
 446                        ip_rcv_finish);
 447
 448 csum_error:
 449         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_CSUMERRORS);
 450 inhdr_error:
 451         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 452 drop:
 453         kfree_skb(skb);
 454 out:
 455         return NET_RX_DROP;
 456 }