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 <net/inet_ecn.h> 145 #include <linux/netfilter_ipv4.h> 146 #include <net/xfrm.h> 147 #include <linux/mroute.h> 148 #include <linux/netlink.h> 149 #include <net/dst_metadata.h> 150 151 /* 152 * Process Router Attention IP option (RFC 2113) 153 */ 154 bool ip_call_ra_chain(struct sk_buff *skb) 155 { 156 struct ip_ra_chain *ra; 157 u8 protocol = ip_hdr(skb)->protocol; 158 struct sock *last = NULL; 159 struct net_device *dev = skb->dev; 160 struct net *net = dev_net(dev); 161 162 for (ra = rcu_dereference(net->ipv4.ra_chain); ra; ra = rcu_dereference(ra->next)) { 163 struct sock *sk = ra->sk; 164 165 /* If socket is bound to an interface, only report 166 * the packet if it came from that interface. 167 */ 168 if (sk && inet_sk(sk)->inet_num == protocol && 169 (!sk->sk_bound_dev_if || 170 sk->sk_bound_dev_if == dev->ifindex)) { 171 if (ip_is_fragment(ip_hdr(skb))) { 172 if (ip_defrag(net, skb, IP_DEFRAG_CALL_RA_CHAIN)) 173 return true; 174 } 175 if (last) { 176 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); 177 if (skb2) 178 raw_rcv(last, skb2); 179 } 180 last = sk; 181 } 182 } 183 184 if (last) { 185 raw_rcv(last, skb); 186 return true; 187 } 188 return false; 189 } 190 191 static int ip_local_deliver_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 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) { 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(net, IPSTATS_MIB_INDELIVERS); 221 } else { 222 if (!raw) { 223 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 224 __IP_INC_STATS(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(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 struct net *net = dev_net(skb->dev); 250 251 if (ip_is_fragment(ip_hdr(skb))) { 252 if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER)) 253 return 0; 254 } 255 256 return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN, 257 net, NULL, skb, skb->dev, NULL, 258 ip_local_deliver_finish); 259 } 260 261 static inline bool ip_rcv_options(struct sk_buff *skb) 262 { 263 struct ip_options *opt; 264 const struct iphdr *iph; 265 struct net_device *dev = skb->dev; 266 267 /* It looks as overkill, because not all 268 IP options require packet mangling. 269 But it is the easiest for now, especially taking 270 into account that combination of IP options 271 and running sniffer is extremely rare condition. 272 --ANK (980813) 273 */ 274 if (skb_cow(skb, skb_headroom(skb))) { 275 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS); 276 goto drop; 277 } 278 279 iph = ip_hdr(skb); 280 opt = &(IPCB(skb)->opt); 281 opt->optlen = iph->ihl*4 - sizeof(struct iphdr); 282 283 if (ip_options_compile(dev_net(dev), opt, skb)) { 284 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS); 285 goto drop; 286 } 287 288 if (unlikely(opt->srr)) { 289 struct in_device *in_dev = __in_dev_get_rcu(dev); 290 291 if (in_dev) { 292 if (!IN_DEV_SOURCE_ROUTE(in_dev)) { 293 if (IN_DEV_LOG_MARTIANS(in_dev)) 294 net_info_ratelimited("source route option %pI4 -> %pI4\n", 295 &iph->saddr, 296 &iph->daddr); 297 goto drop; 298 } 299 } 300 301 if (ip_options_rcv_srr(skb)) 302 goto drop; 303 } 304 305 return false; 306 drop: 307 return true; 308 } 309 310 static int ip_rcv_finish_core(struct net *net, struct sock *sk, 311 struct sk_buff *skb) 312 { 313 const struct iphdr *iph = ip_hdr(skb); 314 int (*edemux)(struct sk_buff *skb); 315 struct net_device *dev = skb->dev; 316 struct rtable *rt; 317 int err; 318 319 if (net->ipv4.sysctl_ip_early_demux && 320 !skb_dst(skb) && 321 !skb->sk && 322 !ip_is_fragment(iph)) { 323 const struct net_protocol *ipprot; 324 int protocol = iph->protocol; 325 326 ipprot = rcu_dereference(inet_protos[protocol]); 327 if (ipprot && (edemux = READ_ONCE(ipprot->early_demux))) { 328 err = edemux(skb); 329 if (unlikely(err)) 330 goto drop_error; 331 /* must reload iph, skb->head might have changed */ 332 iph = ip_hdr(skb); 333 } 334 } 335 336 /* 337 * Initialise the virtual path cache for the packet. It describes 338 * how the packet travels inside Linux networking. 339 */ 340 if (!skb_valid_dst(skb)) { 341 err = ip_route_input_noref(skb, iph->daddr, iph->saddr, 342 iph->tos, dev); 343 if (unlikely(err)) 344 goto drop_error; 345 } 346 347 #ifdef CONFIG_IP_ROUTE_CLASSID 348 if (unlikely(skb_dst(skb)->tclassid)) { 349 struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct); 350 u32 idx = skb_dst(skb)->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 rt = skb_rtable(skb); 362 if (rt->rt_type == RTN_MULTICAST) { 363 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len); 364 } else if (rt->rt_type == RTN_BROADCAST) { 365 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len); 366 } else if (skb->pkt_type == PACKET_BROADCAST || 367 skb->pkt_type == PACKET_MULTICAST) { 368 struct in_device *in_dev = __in_dev_get_rcu(dev); 369 370 /* RFC 1122 3.3.6: 371 * 372 * When a host sends a datagram to a link-layer broadcast 373 * address, the IP destination address MUST be a legal IP 374 * broadcast or IP multicast address. 375 * 376 * A host SHOULD silently discard a datagram that is received 377 * via a link-layer broadcast (see Section 2.4) but does not 378 * specify an IP multicast or broadcast destination address. 379 * 380 * This doesn't explicitly say L2 *broadcast*, but broadcast is 381 * in a way a form of multicast and the most common use case for 382 * this is 802.11 protecting against cross-station spoofing (the 383 * so-called "hole-196" attack) so do it for both. 384 */ 385 if (in_dev && 386 IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST)) 387 goto drop; 388 } 389 390 return NET_RX_SUCCESS; 391 392 drop: 393 kfree_skb(skb); 394 return NET_RX_DROP; 395 396 drop_error: 397 if (err == -EXDEV) 398 __NET_INC_STATS(net, LINUX_MIB_IPRPFILTER); 399 goto drop; 400 } 401 402 static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 403 { 404 int ret; 405 406 /* if ingress device is enslaved to an L3 master device pass the 407 * skb to its handler for processing 408 */ 409 skb = l3mdev_ip_rcv(skb); 410 if (!skb) 411 return NET_RX_SUCCESS; 412 413 ret = ip_rcv_finish_core(net, sk, skb); 414 if (ret != NET_RX_DROP) 415 ret = dst_input(skb); 416 return ret; 417 } 418 419 /* 420 * Main IP Receive routine. 421 */ 422 static struct sk_buff *ip_rcv_core(struct sk_buff *skb, struct net *net) 423 { 424 const struct iphdr *iph; 425 u32 len; 426 427 /* When the interface is in promisc. mode, drop all the crap 428 * that it receives, do not try to analyse it. 429 */ 430 if (skb->pkt_type == PACKET_OTHERHOST) 431 goto drop; 432 433 434 __IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len); 435 436 skb = skb_share_check(skb, GFP_ATOMIC); 437 if (!skb) { 438 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS); 439 goto out; 440 } 441 442 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 443 goto inhdr_error; 444 445 iph = ip_hdr(skb); 446 447 /* 448 * RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum. 449 * 450 * Is the datagram acceptable? 451 * 452 * 1. Length at least the size of an ip header 453 * 2. Version of 4 454 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums] 455 * 4. Doesn't have a bogus length 456 */ 457 458 if (iph->ihl < 5 || iph->version != 4) 459 goto inhdr_error; 460 461 BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1); 462 BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0); 463 BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE); 464 __IP_ADD_STATS(net, 465 IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK), 466 max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs)); 467 468 if (!pskb_may_pull(skb, iph->ihl*4)) 469 goto inhdr_error; 470 471 iph = ip_hdr(skb); 472 473 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl))) 474 goto csum_error; 475 476 len = ntohs(iph->tot_len); 477 if (skb->len < len) { 478 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS); 479 goto drop; 480 } else if (len < (iph->ihl*4)) 481 goto inhdr_error; 482 483 /* Our transport medium may have padded the buffer out. Now we know it 484 * is IP we can trim to the true length of the frame. 485 * Note this now means skb->len holds ntohs(iph->tot_len). 486 */ 487 if (pskb_trim_rcsum(skb, len)) { 488 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS); 489 goto drop; 490 } 491 492 skb->transport_header = skb->network_header + iph->ihl*4; 493 494 /* Remove any debris in the socket control block */ 495 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); 496 IPCB(skb)->iif = skb->skb_iif; 497 498 /* Must drop socket now because of tproxy. */ 499 skb_orphan(skb); 500 501 return skb; 502 503 csum_error: 504 __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS); 505 inhdr_error: 506 __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS); 507 drop: 508 kfree_skb(skb); 509 out: 510 return NULL; 511 } 512 513 /* 514 * IP receive entry point 515 */ 516 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, 517 struct net_device *orig_dev) 518 { 519 struct net *net = dev_net(dev); 520 521 skb = ip_rcv_core(skb, net); 522 if (skb == NULL) 523 return NET_RX_DROP; 524 return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, 525 net, NULL, skb, dev, NULL, 526 ip_rcv_finish); 527 } 528 529 static void ip_sublist_rcv_finish(struct list_head *head) 530 { 531 struct sk_buff *skb, *next; 532 533 list_for_each_entry_safe(skb, next, head, list) { 534 list_del(&skb->list); 535 /* Handle ip{6}_forward case, as sch_direct_xmit have 536 * another kind of SKB-list usage (see validate_xmit_skb_list) 537 */ 538 skb->next = NULL; 539 dst_input(skb); 540 } 541 } 542 543 static void ip_list_rcv_finish(struct net *net, struct sock *sk, 544 struct list_head *head) 545 { 546 struct dst_entry *curr_dst = NULL; 547 struct sk_buff *skb, *next; 548 struct list_head sublist; 549 550 INIT_LIST_HEAD(&sublist); 551 list_for_each_entry_safe(skb, next, head, list) { 552 struct dst_entry *dst; 553 554 list_del(&skb->list); 555 /* if ingress device is enslaved to an L3 master device pass the 556 * skb to its handler for processing 557 */ 558 skb = l3mdev_ip_rcv(skb); 559 if (!skb) 560 continue; 561 if (ip_rcv_finish_core(net, sk, skb) == NET_RX_DROP) 562 continue; 563 564 dst = skb_dst(skb); 565 if (curr_dst != dst) { 566 /* dispatch old sublist */ 567 if (!list_empty(&sublist)) 568 ip_sublist_rcv_finish(&sublist); 569 /* start new sublist */ 570 INIT_LIST_HEAD(&sublist); 571 curr_dst = dst; 572 } 573 list_add_tail(&skb->list, &sublist); 574 } 575 /* dispatch final sublist */ 576 ip_sublist_rcv_finish(&sublist); 577 } 578 579 static void ip_sublist_rcv(struct list_head *head, struct net_device *dev, 580 struct net *net) 581 { 582 NF_HOOK_LIST(NFPROTO_IPV4, NF_INET_PRE_ROUTING, net, NULL, 583 head, dev, NULL, ip_rcv_finish); 584 ip_list_rcv_finish(net, NULL, head); 585 } 586 587 /* Receive a list of IP packets */ 588 void ip_list_rcv(struct list_head *head, struct packet_type *pt, 589 struct net_device *orig_dev) 590 { 591 struct net_device *curr_dev = NULL; 592 struct net *curr_net = NULL; 593 struct sk_buff *skb, *next; 594 struct list_head sublist; 595 596 INIT_LIST_HEAD(&sublist); 597 list_for_each_entry_safe(skb, next, head, list) { 598 struct net_device *dev = skb->dev; 599 struct net *net = dev_net(dev); 600 601 list_del(&skb->list); 602 skb = ip_rcv_core(skb, net); 603 if (skb == NULL) 604 continue; 605 606 if (curr_dev != dev || curr_net != net) { 607 /* dispatch old sublist */ 608 if (!list_empty(&sublist)) 609 ip_sublist_rcv(&sublist, curr_dev, curr_net); 610 /* start new sublist */ 611 INIT_LIST_HEAD(&sublist); 612 curr_dev = dev; 613 curr_net = net; 614 } 615 list_add_tail(&skb->list, &sublist); 616 } 617 /* dispatch final sublist */ 618 ip_sublist_rcv(&sublist, curr_dev, curr_net); 619 } 620