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, ip_hdrlen(skb)); 194 195 /* Point into the IP datagram, just past the header. */ 196 skb_reset_transport_header(skb); 197 198 rcu_read_lock(); 199 { 200 int protocol = ip_hdr(skb)->protocol; 201 int hash, raw; 202 const struct net_protocol *ipprot; 203 204 resubmit: 205 raw = raw_local_deliver(skb, protocol); 206 207 hash = protocol & (MAX_INET_PROTOS - 1); 208 ipprot = rcu_dereference(inet_protos[hash]); 209 if (ipprot != NULL) { 210 int ret; 211 212 if (!net_eq(net, &init_net) && !ipprot->netns_ok) { 213 if (net_ratelimit()) 214 printk("%s: proto %d isn't netns-ready\n", 215 __func__, protocol); 216 kfree_skb(skb); 217 goto out; 218 } 219 220 if (!ipprot->no_policy) { 221 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 222 kfree_skb(skb); 223 goto out; 224 } 225 nf_reset(skb); 226 } 227 ret = ipprot->handler(skb); 228 if (ret < 0) { 229 protocol = -ret; 230 goto resubmit; 231 } 232 IP_INC_STATS_BH(net, IPSTATS_MIB_INDELIVERS); 233 } else { 234 if (!raw) { 235 if (xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 236 IP_INC_STATS_BH(net, IPSTATS_MIB_INUNKNOWNPROTOS); 237 icmp_send(skb, ICMP_DEST_UNREACH, 238 ICMP_PROT_UNREACH, 0); 239 } 240 } else 241 IP_INC_STATS_BH(net, IPSTATS_MIB_INDELIVERS); 242 kfree_skb(skb); 243 } 244 } 245 out: 246 rcu_read_unlock(); 247 248 return 0; 249 } 250 251 /* 252 * Deliver IP Packets to the higher protocol layers. 253 */ 254 int ip_local_deliver(struct sk_buff *skb) 255 { 256 /* 257 * Reassemble IP fragments. 258 */ 259 260 if (ip_is_fragment(ip_hdr(skb))) { 261 if (ip_defrag(skb, IP_DEFRAG_LOCAL_DELIVER)) 262 return 0; 263 } 264 265 return NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_IN, skb, skb->dev, NULL, 266 ip_local_deliver_finish); 267 } 268 269 static inline bool ip_rcv_options(struct sk_buff *skb) 270 { 271 struct ip_options *opt; 272 const struct iphdr *iph; 273 struct net_device *dev = skb->dev; 274 275 /* It looks as overkill, because not all 276 IP options require packet mangling. 277 But it is the easiest for now, especially taking 278 into account that combination of IP options 279 and running sniffer is extremely rare condition. 280 --ANK (980813) 281 */ 282 if (skb_cow(skb, skb_headroom(skb))) { 283 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS); 284 goto drop; 285 } 286 287 iph = ip_hdr(skb); 288 opt = &(IPCB(skb)->opt); 289 opt->optlen = iph->ihl*4 - sizeof(struct iphdr); 290 291 if (ip_options_compile(dev_net(dev), opt, skb)) { 292 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS); 293 goto drop; 294 } 295 296 if (unlikely(opt->srr)) { 297 struct in_device *in_dev = __in_dev_get_rcu(dev); 298 299 if (in_dev) { 300 if (!IN_DEV_SOURCE_ROUTE(in_dev)) { 301 if (IN_DEV_LOG_MARTIANS(in_dev) && 302 net_ratelimit()) 303 pr_info("source route option %pI4 -> %pI4\n", 304 &iph->saddr, &iph->daddr); 305 goto drop; 306 } 307 } 308 309 if (ip_options_rcv_srr(skb)) 310 goto drop; 311 } 312 313 return false; 314 drop: 315 return true; 316 } 317 318 static int ip_rcv_finish(struct sk_buff *skb) 319 { 320 const struct iphdr *iph = ip_hdr(skb); 321 struct rtable *rt; 322 323 /* 324 * Initialise the virtual path cache for the packet. It describes 325 * how the packet travels inside Linux networking. 326 */ 327 if (skb_dst(skb) == NULL) { 328 int err = ip_route_input_noref(skb, iph->daddr, iph->saddr, 329 iph->tos, skb->dev); 330 if (unlikely(err)) { 331 if (err == -EHOSTUNREACH) 332 IP_INC_STATS_BH(dev_net(skb->dev), 333 IPSTATS_MIB_INADDRERRORS); 334 else if (err == -ENETUNREACH) 335 IP_INC_STATS_BH(dev_net(skb->dev), 336 IPSTATS_MIB_INNOROUTES); 337 else if (err == -EXDEV) 338 NET_INC_STATS_BH(dev_net(skb->dev), 339 LINUX_MIB_IPRPFILTER); 340 goto drop; 341 } 342 } 343 344 #ifdef CONFIG_IP_ROUTE_CLASSID 345 if (unlikely(skb_dst(skb)->tclassid)) { 346 struct ip_rt_acct *st = this_cpu_ptr(ip_rt_acct); 347 u32 idx = skb_dst(skb)->tclassid; 348 st[idx&0xFF].o_packets++; 349 st[idx&0xFF].o_bytes += skb->len; 350 st[(idx>>16)&0xFF].i_packets++; 351 st[(idx>>16)&0xFF].i_bytes += skb->len; 352 } 353 #endif 354 355 if (iph->ihl > 5 && ip_rcv_options(skb)) 356 goto drop; 357 358 rt = skb_rtable(skb); 359 if (rt->rt_type == RTN_MULTICAST) { 360 IP_UPD_PO_STATS_BH(dev_net(rt->dst.dev), IPSTATS_MIB_INMCAST, 361 skb->len); 362 } else if (rt->rt_type == RTN_BROADCAST) 363 IP_UPD_PO_STATS_BH(dev_net(rt->dst.dev), IPSTATS_MIB_INBCAST, 364 skb->len); 365 366 return dst_input(skb); 367 368 drop: 369 kfree_skb(skb); 370 return NET_RX_DROP; 371 } 372 373 /* 374 * Main IP Receive routine. 375 */ 376 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev) 377 { 378 const struct iphdr *iph; 379 u32 len; 380 381 /* When the interface is in promisc. mode, drop all the crap 382 * that it receives, do not try to analyse it. 383 */ 384 if (skb->pkt_type == PACKET_OTHERHOST) 385 goto drop; 386 387 388 IP_UPD_PO_STATS_BH(dev_net(dev), IPSTATS_MIB_IN, skb->len); 389 390 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL) { 391 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS); 392 goto out; 393 } 394 395 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 396 goto inhdr_error; 397 398 iph = ip_hdr(skb); 399 400 /* 401 * RFC1122: 3.2.1.2 MUST silently discard any IP frame that fails the checksum. 402 * 403 * Is the datagram acceptable? 404 * 405 * 1. Length at least the size of an ip header 406 * 2. Version of 4 407 * 3. Checksums correctly. [Speed optimisation for later, skip loopback checksums] 408 * 4. Doesn't have a bogus length 409 */ 410 411 if (iph->ihl < 5 || iph->version != 4) 412 goto inhdr_error; 413 414 if (!pskb_may_pull(skb, iph->ihl*4)) 415 goto inhdr_error; 416 417 iph = ip_hdr(skb); 418 419 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl))) 420 goto inhdr_error; 421 422 len = ntohs(iph->tot_len); 423 if (skb->len < len) { 424 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS); 425 goto drop; 426 } else if (len < (iph->ihl*4)) 427 goto inhdr_error; 428 429 /* Our transport medium may have padded the buffer out. Now we know it 430 * is IP we can trim to the true length of the frame. 431 * Note this now means skb->len holds ntohs(iph->tot_len). 432 */ 433 if (pskb_trim_rcsum(skb, len)) { 434 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS); 435 goto drop; 436 } 437 438 /* Remove any debris in the socket control block */ 439 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); 440 441 /* Must drop socket now because of tproxy. */ 442 skb_orphan(skb); 443 444 return NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, dev, NULL, 445 ip_rcv_finish); 446 447 inhdr_error: 448 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS); 449 drop: 450 kfree_skb(skb); 451 out: 452 return NET_RX_DROP; 453 } 454