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 1.55 2002/01/12 07:39:45 davem Exp $ 9 * 10 * Authors: Ross Biro 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/module.h> 120 #include <linux/types.h> 121 #include <linux/kernel.h> 122 #include <linux/string.h> 123 #include <linux/errno.h> 124 #include <linux/config.h> 125 126 #include <linux/net.h> 127 #include <linux/socket.h> 128 #include <linux/sockios.h> 129 #include <linux/in.h> 130 #include <linux/inet.h> 131 #include <linux/inetdevice.h> 132 #include <linux/netdevice.h> 133 #include <linux/etherdevice.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 <linux/netfilter_ipv4.h> 146 #include <net/xfrm.h> 147 #include <linux/mroute.h> 148 #include <linux/netlink.h> 149 150 /* 151 * SNMP management statistics 152 */ 153 154 DEFINE_SNMP_STAT(struct ipstats_mib, ip_statistics) __read_mostly; 155 156 /* 157 * Process Router Attention IP option 158 */ 159 int ip_call_ra_chain(struct sk_buff *skb) 160 { 161 struct ip_ra_chain *ra; 162 u8 protocol = skb->nh.iph->protocol; 163 struct sock *last = NULL; 164 165 read_lock(&ip_ra_lock); 166 for (ra = ip_ra_chain; ra; ra = ra->next) { 167 struct sock *sk = ra->sk; 168 169 /* If socket is bound to an interface, only report 170 * the packet if it came from that interface. 171 */ 172 if (sk && inet_sk(sk)->num == protocol && 173 (!sk->sk_bound_dev_if || 174 sk->sk_bound_dev_if == skb->dev->ifindex)) { 175 if (skb->nh.iph->frag_off & htons(IP_MF|IP_OFFSET)) { 176 skb = ip_defrag(skb, IP_DEFRAG_CALL_RA_CHAIN); 177 if (skb == NULL) { 178 read_unlock(&ip_ra_lock); 179 return 1; 180 } 181 } 182 if (last) { 183 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); 184 if (skb2) 185 raw_rcv(last, skb2); 186 } 187 last = sk; 188 } 189 } 190 191 if (last) { 192 raw_rcv(last, skb); 193 read_unlock(&ip_ra_lock); 194 return 1; 195 } 196 read_unlock(&ip_ra_lock); 197 return 0; 198 } 199 200 static inline int ip_local_deliver_finish(struct sk_buff *skb) 201 { 202 int ihl = skb->nh.iph->ihl*4; 203 204 __skb_pull(skb, ihl); 205 206 /* Point into the IP datagram, just past the header. */ 207 skb->h.raw = skb->data; 208 209 rcu_read_lock(); 210 { 211 /* Note: See raw.c and net/raw.h, RAWV4_HTABLE_SIZE==MAX_INET_PROTOS */ 212 int protocol = skb->nh.iph->protocol; 213 int hash; 214 struct sock *raw_sk; 215 struct net_protocol *ipprot; 216 217 resubmit: 218 hash = protocol & (MAX_INET_PROTOS - 1); 219 raw_sk = sk_head(&raw_v4_htable[hash]); 220 221 /* If there maybe a raw socket we must check - if not we 222 * don't care less 223 */ 224 if (raw_sk && !raw_v4_input(skb, skb->nh.iph, hash)) 225 raw_sk = NULL; 226 227 if ((ipprot = rcu_dereference(inet_protos[hash])) != NULL) { 228 int ret; 229 230 if (!ipprot->no_policy) { 231 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 232 kfree_skb(skb); 233 goto out; 234 } 235 nf_reset(skb); 236 } 237 ret = ipprot->handler(skb); 238 if (ret < 0) { 239 protocol = -ret; 240 goto resubmit; 241 } 242 IP_INC_STATS_BH(IPSTATS_MIB_INDELIVERS); 243 } else { 244 if (!raw_sk) { 245 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 246 IP_INC_STATS_BH(IPSTATS_MIB_INUNKNOWNPROTOS); 247 icmp_send(skb, ICMP_DEST_UNREACH, 248 ICMP_PROT_UNREACH, 0); 249 } 250 } else 251 IP_INC_STATS_BH(IPSTATS_MIB_INDELIVERS); 252 kfree_skb(skb); 253 } 254 } 255 out: 256 rcu_read_unlock(); 257 258 return 0; 259 } 260 261 /* 262 * Deliver IP Packets to the higher protocol layers. 263 */ 264 int ip_local_deliver(struct sk_buff *skb) 265 { 266 /* 267 * Reassemble IP fragments. 268 */ 269 270 if (skb->nh.iph->frag_off & htons(IP_MF|IP_OFFSET)) { 271 skb = ip_defrag(skb, IP_DEFRAG_LOCAL_DELIVER); 272 if (!skb) 273 return 0; 274 } 275 276 return NF_HOOK(PF_INET, NF_IP_LOCAL_IN, skb, skb->dev, NULL, 277 ip_local_deliver_finish); 278 } 279 280 static inline int ip_rcv_options(struct sk_buff *skb) 281 { 282 struct ip_options *opt; 283 struct iphdr *iph; 284 struct net_device *dev = skb->dev; 285 286 /* It looks as overkill, because not all 287 IP options require packet mangling. 288 But it is the easiest for now, especially taking 289 into account that combination of IP options 290 and running sniffer is extremely rare condition. 291 --ANK (980813) 292 */ 293 if (skb_cow(skb, skb_headroom(skb))) { 294 IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS); 295 goto drop; 296 } 297 298 iph = skb->nh.iph; 299 300 if (ip_options_compile(NULL, skb)) { 301 IP_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS); 302 goto drop; 303 } 304 305 opt = &(IPCB(skb)->opt); 306 if (unlikely(opt->srr)) { 307 struct in_device *in_dev = in_dev_get(dev); 308 if (in_dev) { 309 if (!IN_DEV_SOURCE_ROUTE(in_dev)) { 310 if (IN_DEV_LOG_MARTIANS(in_dev) && 311 net_ratelimit()) 312 printk(KERN_INFO "source route option " 313 "%u.%u.%u.%u -> %u.%u.%u.%u\n", 314 NIPQUAD(iph->saddr), 315 NIPQUAD(iph->daddr)); 316 in_dev_put(in_dev); 317 goto drop; 318 } 319 320 in_dev_put(in_dev); 321 } 322 323 if (ip_options_rcv_srr(skb)) 324 goto drop; 325 } 326 327 return 0; 328 drop: 329 return -1; 330 } 331 332 static inline int ip_rcv_finish(struct sk_buff *skb) 333 { 334 struct iphdr *iph = skb->nh.iph; 335 336 /* 337 * Initialise the virtual path cache for the packet. It describes 338 * how the packet travels inside Linux networking. 339 */ 340 if (likely(skb->dst == NULL)) { 341 int err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, 342 skb->dev); 343 if (unlikely(err)) { 344 if (err == -EHOSTUNREACH) 345 IP_INC_STATS_BH(IPSTATS_MIB_INADDRERRORS); 346 goto drop; 347 } 348 } 349 350 #ifdef CONFIG_NET_CLS_ROUTE 351 if (unlikely(skb->dst->tclassid)) { 352 struct ip_rt_acct *st = ip_rt_acct + 256*smp_processor_id(); 353 u32 idx = skb->dst->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 return dst_input(skb); 365 366 drop: 367 kfree_skb(skb); 368 return NET_RX_DROP; 369 } 370 371 /* 372 * Main IP Receive routine. 373 */ 374 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev) 375 { 376 struct iphdr *iph; 377 u32 len; 378 379 /* When the interface is in promisc. mode, drop all the crap 380 * that it receives, do not try to analyse it. 381 */ 382 if (skb->pkt_type == PACKET_OTHERHOST) 383 goto drop; 384 385 IP_INC_STATS_BH(IPSTATS_MIB_INRECEIVES); 386 387 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL) { 388 IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS); 389 goto out; 390 } 391 392 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 393 goto inhdr_error; 394 395 iph = skb->nh.iph; 396 397 /* 398 * RFC1122: 3.1.2.2 MUST silently discard any IP frame that fails the checksum. 399 * 400 * Is the datagram acceptable? 401 * 402 * 1. Length at least the size of an ip header 403 * 2. Version of 4 404 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums] 405 * 4. Doesn't have a bogus length 406 */ 407 408 if (iph->ihl < 5 || iph->version != 4) 409 goto inhdr_error; 410 411 if (!pskb_may_pull(skb, iph->ihl*4)) 412 goto inhdr_error; 413 414 iph = skb->nh.iph; 415 416 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl))) 417 goto inhdr_error; 418 419 len = ntohs(iph->tot_len); 420 if (skb->len < len || len < (iph->ihl*4)) 421 goto inhdr_error; 422 423 /* Our transport medium may have padded the buffer out. Now we know it 424 * is IP we can trim to the true length of the frame. 425 * Note this now means skb->len holds ntohs(iph->tot_len). 426 */ 427 if (pskb_trim_rcsum(skb, len)) { 428 IP_INC_STATS_BH(IPSTATS_MIB_INDISCARDS); 429 goto drop; 430 } 431 432 return NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, dev, NULL, 433 ip_rcv_finish); 434 435 inhdr_error: 436 IP_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS); 437 drop: 438 kfree_skb(skb); 439 out: 440 return NET_RX_DROP; 441 } 442 443 EXPORT_SYMBOL(ip_statistics); 444