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#include <asm/system.h> 117#include <linux/module.h> 118#include <linux/types.h> 119#include <linux/kernel.h> 120#include <linux/string.h> 121#include <linux/errno.h> 122#include <linux/slab.h> 123 124#include <linux/net.h> 125#include <linux/socket.h> 126#include <linux/sockios.h> 127#include <linux/in.h> 128#include <linux/inet.h> 129#include <linux/inetdevice.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/netfilter_ipv4.h> 144#include <net/xfrm.h> 145#include <linux/mroute.h> 146#include <linux/netlink.h> 147 148#include <typedefs.h> 149#include <bcmdefs.h> 150 151/* Fxcn port-S Wins, 0714-09 */ 152extern int (*ip_pre_insert_hook)(struct sk_buff *skb);//Foxconn add , Lewis Min, for UBD, 04/18/2008 153/* Fxcn port-E Wins, 0714-09 */ 154 155/* 156 * Process Router Attention IP option (RFC 2113) 157 */ 158int 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 for (ra = rcu_dereference(ip_ra_chain); ra; ra = rcu_dereference(ra->next)) { 166 struct sock *sk = ra->sk; 167 168 /* If socket is bound to an interface, only report 169 * the packet if it came from that interface. 170 */ 171 if (sk && inet_sk(sk)->inet_num == protocol && 172 (!sk->sk_bound_dev_if || 173 sk->sk_bound_dev_if == dev->ifindex) && 174 net_eq(sock_net(sk), dev_net(dev))) { 175 if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) { 176 if (ip_defrag(skb, IP_DEFRAG_CALL_RA_CHAIN)) 177 return 1; 178 } 179 if (last) { 180 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); 181 if (skb2) 182 raw_rcv(last, skb2); 183 } 184 last = sk; 185 } 186 } 187 188 if (last) { 189 raw_rcv(last, skb); 190 return 1; 191 } 192 return 0; 193} 194 195static int ip_local_deliver_finish(struct sk_buff *skb) 196{ 197 struct net *net = dev_net(skb->dev); 198 199 __skb_pull(skb, ip_hdrlen(skb)); 200 201 /* Point into the IP datagram, just past the header. */ 202 skb_reset_transport_header(skb); 203 204 rcu_read_lock(); 205 { 206 int protocol = ip_hdr(skb)->protocol; 207 int hash, raw; 208 const struct net_protocol *ipprot; 209 210 resubmit: 211 raw = raw_local_deliver(skb, protocol); 212 213 hash = protocol & (MAX_INET_PROTOS - 1); 214 ipprot = rcu_dereference(inet_protos[hash]); 215 if (ipprot != NULL) { 216 int ret; 217 218 if (!net_eq(net, &init_net) && !ipprot->netns_ok) { 219 if (net_ratelimit()) 220 printk("%s: proto %d isn't netns-ready\n", 221 __func__, protocol); 222 kfree_skb(skb); 223 goto out; 224 } 225 226 if (!ipprot->no_policy) { 227 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 228 kfree_skb(skb); 229 goto out; 230 } 231 nf_reset(skb); 232 } 233 ret = ipprot->handler(skb); 234 if (ret < 0) { 235 protocol = -ret; 236 goto resubmit; 237 } 238 IP_INC_STATS_BH(net, IPSTATS_MIB_INDELIVERS); 239 } else { 240 if (!raw) { 241 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 242 IP_INC_STATS_BH(net, IPSTATS_MIB_INUNKNOWNPROTOS); 243 icmp_send(skb, ICMP_DEST_UNREACH, 244 ICMP_PROT_UNREACH, 0); 245 } 246 } else 247 IP_INC_STATS_BH(net, IPSTATS_MIB_INDELIVERS); 248 kfree_skb(skb); 249 } 250 } 251 out: 252 rcu_read_unlock(); 253 254 return 0; 255} 256 257/* 258 * Deliver IP Packets to the higher protocol layers. 259 */ 260int ip_local_deliver(struct sk_buff *skb) 261{ 262 /* 263 * Reassemble IP fragments. 264 */ 265 266 if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) { 267 if (ip_defrag(skb, IP_DEFRAG_LOCAL_DELIVER)) 268 return 0; 269 } 270 271 return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN, skb, skb->dev, NULL, 272 ip_local_deliver_finish); 273} 274 275static inline int ip_rcv_options(struct sk_buff *skb) 276{ 277 struct ip_options *opt; 278 struct iphdr *iph; 279 struct net_device *dev = skb->dev; 280 281 /* It looks as overkill, because not all 282 IP options require packet mangling. 283 But it is the easiest for now, especially taking 284 into account that combination of IP options 285 and running sniffer is extremely rare condition. 286 --ANK (980813) 287 */ 288 if (skb_cow(skb, skb_headroom(skb))) { 289 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS); 290 goto drop; 291 } 292 293 iph = ip_hdr(skb); 294 opt = &(IPCB(skb)->opt); 295 opt->optlen = iph->ihl*4 - sizeof(struct iphdr); 296 297 if (ip_options_compile(dev_net(dev), opt, skb)) { 298 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS); 299 goto drop; 300 } 301 302 if (unlikely(opt->srr)) { 303 struct in_device *in_dev = __in_dev_get_rcu(dev); 304 305 if (in_dev) { 306 if (!IN_DEV_SOURCE_ROUTE(in_dev)) { 307 if (IN_DEV_LOG_MARTIANS(in_dev) && 308 net_ratelimit()) 309 printk(KERN_INFO "source route option %pI4 -> %pI4\n", 310 &iph->saddr, &iph->daddr); 311 goto drop; 312 } 313 } 314 315 if (ip_options_rcv_srr(skb)) 316 goto drop; 317 } 318 319 return 0; 320drop: 321 return -1; 322} 323 324static int BCMFASTPATH_HOST ip_rcv_finish(struct sk_buff *skb) 325{ 326 const struct iphdr *iph = ip_hdr(skb); 327 struct rtable *rt; 328 329 /* 330 * Initialise the virtual path cache for the packet. It describes 331 * how the packet travels inside Linux networking. 332 */ 333 if (skb_dst(skb) == NULL) { 334 int err = ip_route_input_noref(skb, iph->daddr, iph->saddr, 335 iph->tos, skb->dev); 336 if (unlikely(err)) { 337 if (err == -EHOSTUNREACH) 338 IP_INC_STATS_BH(dev_net(skb->dev), 339 IPSTATS_MIB_INADDRERRORS); 340 else if (err == -ENETUNREACH) 341 IP_INC_STATS_BH(dev_net(skb->dev), 342 IPSTATS_MIB_INNOROUTES); 343 else if (err == -EXDEV) 344 NET_INC_STATS_BH(dev_net(skb->dev), 345 LINUX_MIB_IPRPFILTER); 346 goto drop; 347 } 348 } 349 350#ifdef CONFIG_NET_CLS_ROUTE 351 if (unlikely(skb_dst(skb)->tclassid)) { 352 struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct); 353 u32 idx = skb_dst(skb)->tclassid; 354 st[idx&0xFF].o_packets++; 355 st[idx&0xFF].o_bytes += skb->len; 356 st[(idx>>16)&0xFF].i_packets++; 357 st[(idx>>16)&0xFF].i_bytes += skb->len; 358 } 359#endif 360 361 if (iph->ihl > 5 && ip_rcv_options(skb)) 362 goto drop; 363 364 rt = skb_rtable(skb); 365 if (rt->rt_type == RTN_MULTICAST) { 366 IP_UPD_PO_STATS_BH(dev_net(rt->dst.dev), IPSTATS_MIB_INMCAST, 367 skb->len); 368 } else if (rt->rt_type == RTN_BROADCAST) 369 IP_UPD_PO_STATS_BH(dev_net(rt->dst.dev), IPSTATS_MIB_INBCAST, 370 skb->len); 371 372 return dst_input(skb); 373 374drop: 375 kfree_skb(skb); 376 return NET_RX_DROP; 377} 378 379/* Fxcn port-S Wins, 0714-09 */ 380int (*br_pre_insert_hook)(struct sk_buff *skb);//Foxconn add , Lewis Min, for UBD, 04/18/2008 381/* Fxcn port-E Wins, 0714-09 */ 382/* 383 * Main IP Receive routine. 384 */ 385int BCMFASTPATH_HOST ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev) 386{ 387 struct iphdr *iph; 388 u32 len; 389 390 /* When the interface is in promisc. mode, drop all the crap 391 * that it receives, do not try to analyse it. 392 */ 393 if (skb->pkt_type == PACKET_OTHERHOST) 394 goto drop; 395 396 397 IP_UPD_PO_STATS_BH(dev_net(dev), IPSTATS_MIB_IN, skb->len); 398 399 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL) { 400 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS); 401 goto out; 402 } 403 404 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 405 goto inhdr_error; 406 407 iph = ip_hdr(skb); 408 409 /* 410 * RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum. 411 * 412 * Is the datagram acceptable? 413 * 414 * 1. Length at least the size of an ip header 415 * 2. Version of 4 416 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums] 417 * 4. Doesn't have a bogus length 418 */ 419 420 if (iph->ihl < 5 || iph->version != 4) 421 goto inhdr_error; 422 423 if (!pskb_may_pull(skb, iph->ihl*4)) 424 goto inhdr_error; 425 426/* Fxcn port-S Wins, 0714-09 */ 427 //Foxconn add start, Lewis Min, for OpenDNS, 12/08/2008 428 if(NULL!=br_pre_insert_hook) 429 { 430 int ret; 431 432 ret=br_pre_insert_hook(skb); 433 434 if((ret==NF_DROP)||(ret==NF_STOLEN)) 435 return 0; 436 } 437 //Foxconn add end, Lewis Min, for OpenDNS, 12/08/2008 438 439 //Foxconn add start, Lewis Min, for UBD, 04/18/2008 440 if(NULL!=ip_pre_insert_hook) 441 { 442 int ret; 443 444 ret=ip_pre_insert_hook(skb); 445 446 if((ret==NF_DROP)||(ret==NF_STOLEN)) 447 return 0; 448 } 449 //Foxconn add end, Lewis Min, for UBD, 04/18/2008 450/* Fxcn port-E Wins, 0714-09 */ 451 452 iph = ip_hdr(skb); 453 454 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl))) 455 goto inhdr_error; 456 457 len = ntohs(iph->tot_len); 458 if (skb->len < len) { 459 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS); 460 goto drop; 461 } else if (len < (iph->ihl*4)) 462 goto inhdr_error; 463 464 /* Our transport medium may have padded the buffer out. Now we know it 465 * is IP we can trim to the true length of the frame. 466 * Note this now means skb->len holds ntohs(iph->tot_len). 467 */ 468 if (pskb_trim_rcsum(skb, len)) { 469 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS); 470 goto drop; 471 } 472 473 /* Remove any debris in the socket control block */ 474 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); 475 476 /* Must drop socket now because of tproxy. */ 477 skb_orphan(skb); 478 479 /* Skip NF lookup of TCP ACKs for SMB data packets */ 480#if 0 /* foxconn removed , to wait BRCM's solutoin. It will cause WAN site port scan failed. */ 481 if (ip_hdr(skb)->protocol == IPPROTO_TCP) { 482 __be16 *th = (__be16 *)(((__u8 *)ip_hdr(skb)) + (ip_hdr(skb)->ihl << 2)); 483 skb->tcpf_smb = (th[1] == htons(0x01bd)); /* SMB data */ 484 } 485#endif 486 487 return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, dev, NULL, 488 ip_rcv_finish); 489 490inhdr_error: 491 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS); 492drop: 493 kfree_skb(skb); 494out: 495 return NET_RX_DROP; 496} 497 498 499/* Fxcn port-S Wins, 0714-09 */ 500//Foxconn add start, Lewis Min, for OpenDNS, 12/12/2008 501void insert_func_to_BR_PRE_ROUTE(void *FUNC) 502{ 503 br_pre_insert_hook= FUNC; 504} 505 506 507void remove_func_from_BR_PRE_ROUTE(void) 508{ 509 br_pre_insert_hook= NULL; 510} 511//Foxconn add end, Lewis Min, for OpenDNS, 12/12/2008 512/* Fxcn port-E Wins, 0714-09 */ 513 514