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 #include <linux/indirect_call_wrapper.h> 134 135 #include <net/snmp.h> 136 #include <net/ip.h> 137 #include <net/protocol.h> 138 #include <net/route.h> 139 #include <linux/skbuff.h> 140 #include <net/sock.h> 141 #include <net/arp.h> 142 #include <net/icmp.h> 143 #include <net/raw.h> 144 #include <net/checksum.h> 145 #include <net/inet_ecn.h> 146 #include <linux/netfilter_ipv4.h> 147 #include <net/xfrm.h> 148 #include <linux/mroute.h> 149 #include <linux/netlink.h> 150 #include <net/dst_metadata.h> 151 152 /* 153 * Process Router Attention IP option (RFC 2113) 154 */ 155 bool ip_call_ra_chain(struct sk_buff *skb) 156 { 157 struct ip_ra_chain *ra; 158 u8 protocol = ip_hdr(skb)->protocol; 159 struct sock *last = NULL; 160 struct net_device *dev = skb->dev; 161 struct net *net = dev_net(dev); 162 163 for (ra = rcu_dereference(net->ipv4.ra_chain); ra; ra = rcu_dereference(ra->next)) { 164 struct sock *sk = ra->sk; 165 166 /* If socket is bound to an interface, only report 167 * the packet if it came from that interface. 168 */ 169 if (sk && inet_sk(sk)->inet_num == protocol && 170 (!sk->sk_bound_dev_if || 171 sk->sk_bound_dev_if == dev->ifindex)) { 172 if (ip_is_fragment(ip_hdr(skb))) { 173 if (ip_defrag(net, skb, IP_DEFRAG_CALL_RA_CHAIN)) 174 return true; 175 } 176 if (last) { 177 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); 178 if (skb2) 179 raw_rcv(last, skb2); 180 } 181 last = sk; 182 } 183 } 184 185 if (last) { 186 raw_rcv(last, skb); 187 return true; 188 } 189 return false; 190 } 191 192 INDIRECT_CALLABLE_DECLARE(int udp_rcv(struct sk_buff *)); 193 INDIRECT_CALLABLE_DECLARE(int tcp_v4_rcv(struct sk_buff *)); 194 void ip_protocol_deliver_rcu(struct net *net, struct sk_buff *skb, int protocol) 195 { 196 const struct net_protocol *ipprot; 197 int raw, ret; 198 199 resubmit: 200 raw = raw_local_deliver(skb, protocol); 201 202 ipprot = rcu_dereference(inet_protos[protocol]); 203 if (ipprot) { 204 if (!ipprot->no_policy) { 205 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 206 kfree_skb(skb); 207 return; 208 } 209 nf_reset(skb); 210 } 211 ret = INDIRECT_CALL_2(ipprot->handler, tcp_v4_rcv, udp_rcv, 212 skb); 213 if (ret < 0) { 214 protocol = -ret; 215 goto resubmit; 216 } 217 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS); 218 } else { 219 if (!raw) { 220 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 221 __IP_INC_STATS(net, IPSTATS_MIB_INUNKNOWNPROTOS); 222 icmp_send(skb, ICMP_DEST_UNREACH, 223 ICMP_PROT_UNREACH, 0); 224 } 225 kfree_skb(skb); 226 } else { 227 __IP_INC_STATS(net, IPSTATS_MIB_INDELIVERS); 228 consume_skb(skb); 229 } 230 } 231 } 232 233 static int ip_local_deliver_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 234 { 235 __skb_pull(skb, skb_network_header_len(skb)); 236 237 rcu_read_lock(); 238 ip_protocol_deliver_rcu(net, skb, ip_hdr(skb)->protocol); 239 rcu_read_unlock(); 240 241 return 0; 242 } 243 244 /* 245 * Deliver IP Packets to the higher protocol layers. 246 */ 247 int ip_local_deliver(struct sk_buff *skb) 248 { 249 /* 250 * Reassemble IP fragments. 251 */ 252 struct net *net = dev_net(skb->dev); 253 254 if (ip_is_fragment(ip_hdr(skb))) { 255 if (ip_defrag(net, skb, IP_DEFRAG_LOCAL_DELIVER)) 256 return 0; 257 } 258 259 return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN, 260 net, NULL, skb, skb->dev, NULL, 261 ip_local_deliver_finish); 262 } 263 264 static inline bool ip_rcv_options(struct sk_buff *skb, struct net_device *dev) 265 { 266 struct ip_options *opt; 267 const struct iphdr *iph; 268 269 /* It looks as overkill, because not all 270 IP options require packet mangling. 271 But it is the easiest for now, especially taking 272 into account that combination of IP options 273 and running sniffer is extremely rare condition. 274 --ANK (980813) 275 */ 276 if (skb_cow(skb, skb_headroom(skb))) { 277 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INDISCARDS); 278 goto drop; 279 } 280 281 iph = ip_hdr(skb); 282 opt = &(IPCB(skb)->opt); 283 opt->optlen = iph->ihl*4 - sizeof(struct iphdr); 284 285 if (ip_options_compile(dev_net(dev), opt, skb)) { 286 __IP_INC_STATS(dev_net(dev), IPSTATS_MIB_INHDRERRORS); 287 goto drop; 288 } 289 290 if (unlikely(opt->srr)) { 291 struct in_device *in_dev = __in_dev_get_rcu(dev); 292 293 if (in_dev) { 294 if (!IN_DEV_SOURCE_ROUTE(in_dev)) { 295 if (IN_DEV_LOG_MARTIANS(in_dev)) 296 net_info_ratelimited("source route option %pI4 -> %pI4\n", 297 &iph->saddr, 298 &iph->daddr); 299 goto drop; 300 } 301 } 302 303 if (ip_options_rcv_srr(skb, dev)) 304 goto drop; 305 } 306 307 return false; 308 drop: 309 return true; 310 } 311 312 INDIRECT_CALLABLE_DECLARE(int udp_v4_early_demux(struct sk_buff *)); 313 INDIRECT_CALLABLE_DECLARE(int tcp_v4_early_demux(struct sk_buff *)); 314 static int ip_rcv_finish_core(struct net *net, struct sock *sk, 315 struct sk_buff *skb, struct net_device *dev) 316 { 317 const struct iphdr *iph = ip_hdr(skb); 318 int (*edemux)(struct sk_buff *skb); 319 struct rtable *rt; 320 int err; 321 322 if (net->ipv4.sysctl_ip_early_demux && 323 !skb_dst(skb) && 324 !skb->sk && 325 !ip_is_fragment(iph)) { 326 const struct net_protocol *ipprot; 327 int protocol = iph->protocol; 328 329 ipprot = rcu_dereference(inet_protos[protocol]); 330 if (ipprot && (edemux = READ_ONCE(ipprot->early_demux))) { 331 err = INDIRECT_CALL_2(edemux, tcp_v4_early_demux, 332 udp_v4_early_demux, skb); 333 if (unlikely(err)) 334 goto drop_error; 335 /* must reload iph, skb->head might have changed */ 336 iph = ip_hdr(skb); 337 } 338 } 339 340 /* 341 * Initialise the virtual path cache for the packet. It describes 342 * how the packet travels inside Linux networking. 343 */ 344 if (!skb_valid_dst(skb)) { 345 err = ip_route_input_noref(skb, iph->daddr, iph->saddr, 346 iph->tos, dev); 347 if (unlikely(err)) 348 goto drop_error; 349 } 350 351 #ifdef CONFIG_IP_ROUTE_CLASSID 352 if (unlikely(skb_dst(skb)->tclassid)) { 353 struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct); 354 u32 idx = skb_dst(skb)->tclassid; 355 st[idx&0xFF].o_packets++; 356 st[idx&0xFF].o_bytes += skb->len; 357 st[(idx>>16)&0xFF].i_packets++; 358 st[(idx>>16)&0xFF].i_bytes += skb->len; 359 } 360 #endif 361 362 if (iph->ihl > 5 && ip_rcv_options(skb, dev)) 363 goto drop; 364 365 rt = skb_rtable(skb); 366 if (rt->rt_type == RTN_MULTICAST) { 367 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INMCAST, skb->len); 368 } else if (rt->rt_type == RTN_BROADCAST) { 369 __IP_UPD_PO_STATS(net, IPSTATS_MIB_INBCAST, skb->len); 370 } else if (skb->pkt_type == PACKET_BROADCAST || 371 skb->pkt_type == PACKET_MULTICAST) { 372 struct in_device *in_dev = __in_dev_get_rcu(dev); 373 374 /* RFC 1122 3.3.6: 375 * 376 * When a host sends a datagram to a link-layer broadcast 377 * address, the IP destination address MUST be a legal IP 378 * broadcast or IP multicast address. 379 * 380 * A host SHOULD silently discard a datagram that is received 381 * via a link-layer broadcast (see Section 2.4) but does not 382 * specify an IP multicast or broadcast destination address. 383 * 384 * This doesn't explicitly say L2 *broadcast*, but broadcast is 385 * in a way a form of multicast and the most common use case for 386 * this is 802.11 protecting against cross-station spoofing (the 387 * so-called "hole-196" attack) so do it for both. 388 */ 389 if (in_dev && 390 IN_DEV_ORCONF(in_dev, DROP_UNICAST_IN_L2_MULTICAST)) 391 goto drop; 392 } 393 394 return NET_RX_SUCCESS; 395 396 drop: 397 kfree_skb(skb); 398 return NET_RX_DROP; 399 400 drop_error: 401 if (err == -EXDEV) 402 __NET_INC_STATS(net, LINUX_MIB_IPRPFILTER); 403 goto drop; 404 } 405 406 static int ip_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 407 { 408 struct net_device *dev = skb->dev; 409 int ret; 410 411 /* if ingress device is enslaved to an L3 master device pass the 412 * skb to its handler for processing 413 */ 414 skb = l3mdev_ip_rcv(skb); 415 if (!skb) 416 return NET_RX_SUCCESS; 417 418 ret = ip_rcv_finish_core(net, sk, skb, dev); 419 if (ret != NET_RX_DROP) 420 ret = dst_input(skb); 421 return ret; 422 } 423 424 /* 425 * Main IP Receive routine. 426 */ 427 static struct sk_buff *ip_rcv_core(struct sk_buff *skb, struct net *net) 428 { 429 const struct iphdr *iph; 430 u32 len; 431 432 /* When the interface is in promisc. mode, drop all the crap 433 * that it receives, do not try to analyse it. 434 */ 435 if (skb->pkt_type == PACKET_OTHERHOST) 436 goto drop; 437 438 __IP_UPD_PO_STATS(net, IPSTATS_MIB_IN, skb->len); 439 440 skb = skb_share_check(skb, GFP_ATOMIC); 441 if (!skb) { 442 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS); 443 goto out; 444 } 445 446 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 447 goto inhdr_error; 448 449 iph = ip_hdr(skb); 450 451 /* 452 * RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum. 453 * 454 * Is the datagram acceptable? 455 * 456 * 1. Length at least the size of an ip header 457 * 2. Version of 4 458 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums] 459 * 4. Doesn't have a bogus length 460 */ 461 462 if (iph->ihl < 5 || iph->version != 4) 463 goto inhdr_error; 464 465 BUILD_BUG_ON(IPSTATS_MIB_ECT1PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_1); 466 BUILD_BUG_ON(IPSTATS_MIB_ECT0PKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_ECT_0); 467 BUILD_BUG_ON(IPSTATS_MIB_CEPKTS != IPSTATS_MIB_NOECTPKTS + INET_ECN_CE); 468 __IP_ADD_STATS(net, 469 IPSTATS_MIB_NOECTPKTS + (iph->tos & INET_ECN_MASK), 470 max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs)); 471 472 if (!pskb_may_pull(skb, iph->ihl*4)) 473 goto inhdr_error; 474 475 iph = ip_hdr(skb); 476 477 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl))) 478 goto csum_error; 479 480 len = ntohs(iph->tot_len); 481 if (skb->len < len) { 482 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS); 483 goto drop; 484 } else if (len < (iph->ihl*4)) 485 goto inhdr_error; 486 487 /* Our transport medium may have padded the buffer out. Now we know it 488 * is IP we can trim to the true length of the frame. 489 * Note this now means skb->len holds ntohs(iph->tot_len). 490 */ 491 if (pskb_trim_rcsum(skb, len)) { 492 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS); 493 goto drop; 494 } 495 496 iph = ip_hdr(skb); 497 skb->transport_header = skb->network_header + iph->ihl*4; 498 499 /* Remove any debris in the socket control block */ 500 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); 501 IPCB(skb)->iif = skb->skb_iif; 502 503 /* Must drop socket now because of tproxy. */ 504 skb_orphan(skb); 505 506 return skb; 507 508 csum_error: 509 __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS); 510 inhdr_error: 511 __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS); 512 drop: 513 kfree_skb(skb); 514 out: 515 return NULL; 516 } 517 518 /* 519 * IP receive entry point 520 */ 521 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, 522 struct net_device *orig_dev) 523 { 524 struct net *net = dev_net(dev); 525 526 skb = ip_rcv_core(skb, net); 527 if (skb == NULL) 528 return NET_RX_DROP; 529 530 return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, 531 net, NULL, skb, dev, NULL, 532 ip_rcv_finish); 533 } 534 535 static void ip_sublist_rcv_finish(struct list_head *head) 536 { 537 struct sk_buff *skb, *next; 538 539 list_for_each_entry_safe(skb, next, head, list) { 540 skb_list_del_init(skb); 541 dst_input(skb); 542 } 543 } 544 545 static void ip_list_rcv_finish(struct net *net, struct sock *sk, 546 struct list_head *head) 547 { 548 struct dst_entry *curr_dst = NULL; 549 struct sk_buff *skb, *next; 550 struct list_head sublist; 551 552 INIT_LIST_HEAD(&sublist); 553 list_for_each_entry_safe(skb, next, head, list) { 554 struct net_device *dev = skb->dev; 555 struct dst_entry *dst; 556 557 skb_list_del_init(skb); 558 /* if ingress device is enslaved to an L3 master device pass the 559 * skb to its handler for processing 560 */ 561 skb = l3mdev_ip_rcv(skb); 562 if (!skb) 563 continue; 564 if (ip_rcv_finish_core(net, sk, skb, dev) == NET_RX_DROP) 565 continue; 566 567 dst = skb_dst(skb); 568 if (curr_dst != dst) { 569 /* dispatch old sublist */ 570 if (!list_empty(&sublist)) 571 ip_sublist_rcv_finish(&sublist); 572 /* start new sublist */ 573 INIT_LIST_HEAD(&sublist); 574 curr_dst = dst; 575 } 576 list_add_tail(&skb->list, &sublist); 577 } 578 /* dispatch final sublist */ 579 ip_sublist_rcv_finish(&sublist); 580 } 581 582 static void ip_sublist_rcv(struct list_head *head, struct net_device *dev, 583 struct net *net) 584 { 585 NF_HOOK_LIST(NFPROTO_IPV4, NF_INET_PRE_ROUTING, net, NULL, 586 head, dev, NULL, ip_rcv_finish); 587 ip_list_rcv_finish(net, NULL, head); 588 } 589 590 /* Receive a list of IP packets */ 591 void ip_list_rcv(struct list_head *head, struct packet_type *pt, 592 struct net_device *orig_dev) 593 { 594 struct net_device *curr_dev = NULL; 595 struct net *curr_net = NULL; 596 struct sk_buff *skb, *next; 597 struct list_head sublist; 598 599 INIT_LIST_HEAD(&sublist); 600 list_for_each_entry_safe(skb, next, head, list) { 601 struct net_device *dev = skb->dev; 602 struct net *net = dev_net(dev); 603 604 skb_list_del_init(skb); 605 skb = ip_rcv_core(skb, net); 606 if (skb == NULL) 607 continue; 608 609 if (curr_dev != dev || curr_net != net) { 610 /* dispatch old sublist */ 611 if (!list_empty(&sublist)) 612 ip_sublist_rcv(&sublist, curr_dev, curr_net); 613 /* start new sublist */ 614 INIT_LIST_HEAD(&sublist); 615 curr_dev = dev; 616 curr_net = net; 617 } 618 list_add_tail(&skb->list, &sublist); 619 } 620 /* dispatch final sublist */ 621 ip_sublist_rcv(&sublist, curr_dev, curr_net); 622 } 623