1 /* 2 * NET3: Implementation of the ICMP protocol layer. 3 * 4 * Alan Cox, <alan@lxorguk.ukuu.org.uk> 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License 8 * as published by the Free Software Foundation; either version 9 * 2 of the License, or (at your option) any later version. 10 * 11 * Some of the function names and the icmp unreach table for this 12 * module were derived from [icmp.c 1.0.11 06/02/93] by 13 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting. 14 * Other than that this module is a complete rewrite. 15 * 16 * Fixes: 17 * Clemens Fruhwirth : introduce global icmp rate limiting 18 * with icmp type masking ability instead 19 * of broken per type icmp timeouts. 20 * Mike Shaver : RFC1122 checks. 21 * Alan Cox : Multicast ping reply as self. 22 * Alan Cox : Fix atomicity lockup in ip_build_xmit 23 * call. 24 * Alan Cox : Added 216,128 byte paths to the MTU 25 * code. 26 * Martin Mares : RFC1812 checks. 27 * Martin Mares : Can be configured to follow redirects 28 * if acting as a router _without_ a 29 * routing protocol (RFC 1812). 30 * Martin Mares : Echo requests may be configured to 31 * be ignored (RFC 1812). 32 * Martin Mares : Limitation of ICMP error message 33 * transmit rate (RFC 1812). 34 * Martin Mares : TOS and Precedence set correctly 35 * (RFC 1812). 36 * Martin Mares : Now copying as much data from the 37 * original packet as we can without 38 * exceeding 576 bytes (RFC 1812). 39 * Willy Konynenberg : Transparent proxying support. 40 * Keith Owens : RFC1191 correction for 4.2BSD based 41 * path MTU bug. 42 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are 43 * valid (RFC 1812). 44 * Andi Kleen : Check all packet lengths properly 45 * and moved all kfree_skb() up to 46 * icmp_rcv. 47 * Andi Kleen : Move the rate limit bookkeeping 48 * into the dest entry and use a token 49 * bucket filter (thanks to ANK). Make 50 * the rates sysctl configurable. 51 * Yu Tianli : Fixed two ugly bugs in icmp_send 52 * - IP option length was accounted wrongly 53 * - ICMP header length was not accounted 54 * at all. 55 * Tristan Greaves : Added sysctl option to ignore bogus 56 * broadcast responses from broken routers. 57 * 58 * To Fix: 59 * 60 * - Should use skb_pull() instead of all the manual checking. 61 * This would also greatly simply some upper layer error handlers. --AK 62 * 63 */ 64 65 #include <linux/module.h> 66 #include <linux/types.h> 67 #include <linux/jiffies.h> 68 #include <linux/kernel.h> 69 #include <linux/fcntl.h> 70 #include <linux/socket.h> 71 #include <linux/in.h> 72 #include <linux/inet.h> 73 #include <linux/inetdevice.h> 74 #include <linux/netdevice.h> 75 #include <linux/string.h> 76 #include <linux/netfilter_ipv4.h> 77 #include <linux/slab.h> 78 #include <net/snmp.h> 79 #include <net/ip.h> 80 #include <net/route.h> 81 #include <net/protocol.h> 82 #include <net/icmp.h> 83 #include <net/tcp.h> 84 #include <net/udp.h> 85 #include <net/raw.h> 86 #include <net/ping.h> 87 #include <linux/skbuff.h> 88 #include <net/sock.h> 89 #include <linux/errno.h> 90 #include <linux/timer.h> 91 #include <linux/init.h> 92 #include <asm/system.h> 93 #include <asm/uaccess.h> 94 #include <net/checksum.h> 95 #include <net/xfrm.h> 96 #include <net/inet_common.h> 97 98 /* 99 * Build xmit assembly blocks 100 */ 101 102 struct icmp_bxm { 103 struct sk_buff *skb; 104 int offset; 105 int data_len; 106 107 struct { 108 struct icmphdr icmph; 109 __be32 times[3]; 110 } data; 111 int head_len; 112 struct ip_options_data replyopts; 113 }; 114 115 /* An array of errno for error messages from dest unreach. */ 116 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */ 117 118 const struct icmp_err icmp_err_convert[] = { 119 { 120 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */ 121 .fatal = 0, 122 }, 123 { 124 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */ 125 .fatal = 0, 126 }, 127 { 128 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */, 129 .fatal = 1, 130 }, 131 { 132 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */ 133 .fatal = 1, 134 }, 135 { 136 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */ 137 .fatal = 0, 138 }, 139 { 140 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */ 141 .fatal = 0, 142 }, 143 { 144 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */ 145 .fatal = 1, 146 }, 147 { 148 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */ 149 .fatal = 1, 150 }, 151 { 152 .errno = ENONET, /* ICMP_HOST_ISOLATED */ 153 .fatal = 1, 154 }, 155 { 156 .errno = ENETUNREACH, /* ICMP_NET_ANO */ 157 .fatal = 1, 158 }, 159 { 160 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */ 161 .fatal = 1, 162 }, 163 { 164 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */ 165 .fatal = 0, 166 }, 167 { 168 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */ 169 .fatal = 0, 170 }, 171 { 172 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */ 173 .fatal = 1, 174 }, 175 { 176 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */ 177 .fatal = 1, 178 }, 179 { 180 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */ 181 .fatal = 1, 182 }, 183 }; 184 EXPORT_SYMBOL(icmp_err_convert); 185 186 /* 187 * ICMP control array. This specifies what to do with each ICMP. 188 */ 189 190 struct icmp_control { 191 void (*handler)(struct sk_buff *skb); 192 short error; /* This ICMP is classed as an error message */ 193 }; 194 195 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1]; 196 197 /* 198 * The ICMP socket(s). This is the most convenient way to flow control 199 * our ICMP output as well as maintain a clean interface throughout 200 * all layers. All Socketless IP sends will soon be gone. 201 * 202 * On SMP we have one ICMP socket per-cpu. 203 */ 204 static struct sock *icmp_sk(struct net *net) 205 { 206 return net->ipv4.icmp_sk[smp_processor_id()]; 207 } 208 209 static inline struct sock *icmp_xmit_lock(struct net *net) 210 { 211 struct sock *sk; 212 213 local_bh_disable(); 214 215 sk = icmp_sk(net); 216 217 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) { 218 /* This can happen if the output path signals a 219 * dst_link_failure() for an outgoing ICMP packet. 220 */ 221 local_bh_enable(); 222 return NULL; 223 } 224 return sk; 225 } 226 227 static inline void icmp_xmit_unlock(struct sock *sk) 228 { 229 spin_unlock_bh(&sk->sk_lock.slock); 230 } 231 232 /* 233 * Send an ICMP frame. 234 */ 235 236 static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt, 237 struct flowi4 *fl4, int type, int code) 238 { 239 struct dst_entry *dst = &rt->dst; 240 bool rc = true; 241 242 if (type > NR_ICMP_TYPES) 243 goto out; 244 245 /* Don't limit PMTU discovery. */ 246 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 247 goto out; 248 249 /* No rate limit on loopback */ 250 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK)) 251 goto out; 252 253 /* Limit if icmp type is enabled in ratemask. */ 254 if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) { 255 if (!rt->peer) 256 rt_bind_peer(rt, fl4->daddr, 1); 257 rc = inet_peer_xrlim_allow(rt->peer, 258 net->ipv4.sysctl_icmp_ratelimit); 259 } 260 out: 261 return rc; 262 } 263 264 /* 265 * Maintain the counters used in the SNMP statistics for outgoing ICMP 266 */ 267 void icmp_out_count(struct net *net, unsigned char type) 268 { 269 ICMPMSGOUT_INC_STATS(net, type); 270 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS); 271 } 272 273 /* 274 * Checksum each fragment, and on the first include the headers and final 275 * checksum. 276 */ 277 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd, 278 struct sk_buff *skb) 279 { 280 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from; 281 __wsum csum; 282 283 csum = skb_copy_and_csum_bits(icmp_param->skb, 284 icmp_param->offset + offset, 285 to, len, 0); 286 287 skb->csum = csum_block_add(skb->csum, csum, odd); 288 if (icmp_pointers[icmp_param->data.icmph.type].error) 289 nf_ct_attach(skb, icmp_param->skb); 290 return 0; 291 } 292 293 static void icmp_push_reply(struct icmp_bxm *icmp_param, 294 struct flowi4 *fl4, 295 struct ipcm_cookie *ipc, struct rtable **rt) 296 { 297 struct sock *sk; 298 struct sk_buff *skb; 299 300 sk = icmp_sk(dev_net((*rt)->dst.dev)); 301 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param, 302 icmp_param->data_len+icmp_param->head_len, 303 icmp_param->head_len, 304 ipc, rt, MSG_DONTWAIT) < 0) { 305 ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS); 306 ip_flush_pending_frames(sk); 307 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) { 308 struct icmphdr *icmph = icmp_hdr(skb); 309 __wsum csum = 0; 310 struct sk_buff *skb1; 311 312 skb_queue_walk(&sk->sk_write_queue, skb1) { 313 csum = csum_add(csum, skb1->csum); 314 } 315 csum = csum_partial_copy_nocheck((void *)&icmp_param->data, 316 (char *)icmph, 317 icmp_param->head_len, csum); 318 icmph->checksum = csum_fold(csum); 319 skb->ip_summed = CHECKSUM_NONE; 320 ip_push_pending_frames(sk, fl4); 321 } 322 } 323 324 /* 325 * Driving logic for building and sending ICMP messages. 326 */ 327 328 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb) 329 { 330 struct ipcm_cookie ipc; 331 struct rtable *rt = skb_rtable(skb); 332 struct net *net = dev_net(rt->dst.dev); 333 struct flowi4 fl4; 334 struct sock *sk; 335 struct inet_sock *inet; 336 __be32 daddr; 337 338 if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb)) 339 return; 340 341 sk = icmp_xmit_lock(net); 342 if (sk == NULL) 343 return; 344 inet = inet_sk(sk); 345 346 icmp_param->data.icmph.checksum = 0; 347 348 inet->tos = ip_hdr(skb)->tos; 349 daddr = ipc.addr = ip_hdr(skb)->saddr; 350 ipc.opt = NULL; 351 ipc.tx_flags = 0; 352 if (icmp_param->replyopts.opt.opt.optlen) { 353 ipc.opt = &icmp_param->replyopts.opt; 354 if (ipc.opt->opt.srr) 355 daddr = icmp_param->replyopts.opt.opt.faddr; 356 } 357 memset(&fl4, 0, sizeof(fl4)); 358 fl4.daddr = daddr; 359 fl4.saddr = rt->rt_spec_dst; 360 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos); 361 fl4.flowi4_proto = IPPROTO_ICMP; 362 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); 363 rt = ip_route_output_key(net, &fl4); 364 if (IS_ERR(rt)) 365 goto out_unlock; 366 if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type, 367 icmp_param->data.icmph.code)) 368 icmp_push_reply(icmp_param, &fl4, &ipc, &rt); 369 ip_rt_put(rt); 370 out_unlock: 371 icmp_xmit_unlock(sk); 372 } 373 374 static struct rtable *icmp_route_lookup(struct net *net, 375 struct flowi4 *fl4, 376 struct sk_buff *skb_in, 377 const struct iphdr *iph, 378 __be32 saddr, u8 tos, 379 int type, int code, 380 struct icmp_bxm *param) 381 { 382 struct rtable *rt, *rt2; 383 int err; 384 385 memset(fl4, 0, sizeof(*fl4)); 386 fl4->daddr = (param->replyopts.opt.opt.srr ? 387 param->replyopts.opt.opt.faddr : iph->saddr); 388 fl4->saddr = saddr; 389 fl4->flowi4_tos = RT_TOS(tos); 390 fl4->flowi4_proto = IPPROTO_ICMP; 391 fl4->fl4_icmp_type = type; 392 fl4->fl4_icmp_code = code; 393 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4)); 394 rt = __ip_route_output_key(net, fl4); 395 if (IS_ERR(rt)) 396 return rt; 397 398 /* No need to clone since we're just using its address. */ 399 rt2 = rt; 400 401 rt = (struct rtable *) xfrm_lookup(net, &rt->dst, 402 flowi4_to_flowi(fl4), NULL, 0); 403 if (!IS_ERR(rt)) { 404 if (rt != rt2) 405 return rt; 406 } else if (PTR_ERR(rt) == -EPERM) { 407 rt = NULL; 408 } else 409 return rt; 410 411 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(fl4), AF_INET); 412 if (err) 413 goto relookup_failed; 414 415 if (inet_addr_type(net, fl4->saddr) == RTN_LOCAL) { 416 rt2 = __ip_route_output_key(net, fl4); 417 if (IS_ERR(rt2)) 418 err = PTR_ERR(rt2); 419 } else { 420 struct flowi4 fl4_2 = {}; 421 unsigned long orefdst; 422 423 fl4_2.daddr = fl4->saddr; 424 rt2 = ip_route_output_key(net, &fl4_2); 425 if (IS_ERR(rt2)) { 426 err = PTR_ERR(rt2); 427 goto relookup_failed; 428 } 429 /* Ugh! */ 430 orefdst = skb_in->_skb_refdst; /* save old refdst */ 431 err = ip_route_input(skb_in, fl4->daddr, fl4->saddr, 432 RT_TOS(tos), rt2->dst.dev); 433 434 dst_release(&rt2->dst); 435 rt2 = skb_rtable(skb_in); 436 skb_in->_skb_refdst = orefdst; /* restore old refdst */ 437 } 438 439 if (err) 440 goto relookup_failed; 441 442 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst, 443 flowi4_to_flowi(fl4), NULL, 444 XFRM_LOOKUP_ICMP); 445 if (!IS_ERR(rt2)) { 446 dst_release(&rt->dst); 447 rt = rt2; 448 } else if (PTR_ERR(rt2) == -EPERM) { 449 if (rt) 450 dst_release(&rt->dst); 451 return rt2; 452 } else { 453 err = PTR_ERR(rt2); 454 goto relookup_failed; 455 } 456 return rt; 457 458 relookup_failed: 459 if (rt) 460 return rt; 461 return ERR_PTR(err); 462 } 463 464 /* 465 * Send an ICMP message in response to a situation 466 * 467 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header. 468 * MAY send more (we do). 469 * MUST NOT change this header information. 470 * MUST NOT reply to a multicast/broadcast IP address. 471 * MUST NOT reply to a multicast/broadcast MAC address. 472 * MUST reply to only the first fragment. 473 */ 474 475 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info) 476 { 477 struct iphdr *iph; 478 int room; 479 struct icmp_bxm icmp_param; 480 struct rtable *rt = skb_rtable(skb_in); 481 struct ipcm_cookie ipc; 482 struct flowi4 fl4; 483 __be32 saddr; 484 u8 tos; 485 struct net *net; 486 struct sock *sk; 487 488 if (!rt) 489 goto out; 490 net = dev_net(rt->dst.dev); 491 492 /* 493 * Find the original header. It is expected to be valid, of course. 494 * Check this, icmp_send is called from the most obscure devices 495 * sometimes. 496 */ 497 iph = ip_hdr(skb_in); 498 499 if ((u8 *)iph < skb_in->head || 500 (skb_in->network_header + sizeof(*iph)) > skb_in->tail) 501 goto out; 502 503 /* 504 * No replies to physical multicast/broadcast 505 */ 506 if (skb_in->pkt_type != PACKET_HOST) 507 goto out; 508 509 /* 510 * Now check at the protocol level 511 */ 512 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 513 goto out; 514 515 /* 516 * Only reply to fragment 0. We byte re-order the constant 517 * mask for efficiency. 518 */ 519 if (iph->frag_off & htons(IP_OFFSET)) 520 goto out; 521 522 /* 523 * If we send an ICMP error to an ICMP error a mess would result.. 524 */ 525 if (icmp_pointers[type].error) { 526 /* 527 * We are an error, check if we are replying to an 528 * ICMP error 529 */ 530 if (iph->protocol == IPPROTO_ICMP) { 531 u8 _inner_type, *itp; 532 533 itp = skb_header_pointer(skb_in, 534 skb_network_header(skb_in) + 535 (iph->ihl << 2) + 536 offsetof(struct icmphdr, 537 type) - 538 skb_in->data, 539 sizeof(_inner_type), 540 &_inner_type); 541 if (itp == NULL) 542 goto out; 543 544 /* 545 * Assume any unknown ICMP type is an error. This 546 * isn't specified by the RFC, but think about it.. 547 */ 548 if (*itp > NR_ICMP_TYPES || 549 icmp_pointers[*itp].error) 550 goto out; 551 } 552 } 553 554 sk = icmp_xmit_lock(net); 555 if (sk == NULL) 556 return; 557 558 /* 559 * Construct source address and options. 560 */ 561 562 saddr = iph->daddr; 563 if (!(rt->rt_flags & RTCF_LOCAL)) { 564 struct net_device *dev = NULL; 565 566 rcu_read_lock(); 567 if (rt_is_input_route(rt) && 568 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr) 569 dev = dev_get_by_index_rcu(net, rt->rt_iif); 570 571 if (dev) 572 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK); 573 else 574 saddr = 0; 575 rcu_read_unlock(); 576 } 577 578 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) | 579 IPTOS_PREC_INTERNETCONTROL) : 580 iph->tos; 581 582 if (ip_options_echo(&icmp_param.replyopts.opt.opt, skb_in)) 583 goto out_unlock; 584 585 586 /* 587 * Prepare data for ICMP header. 588 */ 589 590 icmp_param.data.icmph.type = type; 591 icmp_param.data.icmph.code = code; 592 icmp_param.data.icmph.un.gateway = info; 593 icmp_param.data.icmph.checksum = 0; 594 icmp_param.skb = skb_in; 595 icmp_param.offset = skb_network_offset(skb_in); 596 inet_sk(sk)->tos = tos; 597 ipc.addr = iph->saddr; 598 ipc.opt = &icmp_param.replyopts.opt; 599 ipc.tx_flags = 0; 600 601 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, 602 type, code, &icmp_param); 603 if (IS_ERR(rt)) 604 goto out_unlock; 605 606 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code)) 607 goto ende; 608 609 /* RFC says return as much as we can without exceeding 576 bytes. */ 610 611 room = dst_mtu(&rt->dst); 612 if (room > 576) 613 room = 576; 614 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen; 615 room -= sizeof(struct icmphdr); 616 617 icmp_param.data_len = skb_in->len - icmp_param.offset; 618 if (icmp_param.data_len > room) 619 icmp_param.data_len = room; 620 icmp_param.head_len = sizeof(struct icmphdr); 621 622 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt); 623 ende: 624 ip_rt_put(rt); 625 out_unlock: 626 icmp_xmit_unlock(sk); 627 out:; 628 } 629 EXPORT_SYMBOL(icmp_send); 630 631 632 /* 633 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH. 634 */ 635 636 static void icmp_unreach(struct sk_buff *skb) 637 { 638 const struct iphdr *iph; 639 struct icmphdr *icmph; 640 int hash, protocol; 641 const struct net_protocol *ipprot; 642 u32 info = 0; 643 struct net *net; 644 645 net = dev_net(skb_dst(skb)->dev); 646 647 /* 648 * Incomplete header ? 649 * Only checks for the IP header, there should be an 650 * additional check for longer headers in upper levels. 651 */ 652 653 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 654 goto out_err; 655 656 icmph = icmp_hdr(skb); 657 iph = (const struct iphdr *)skb->data; 658 659 if (iph->ihl < 5) /* Mangled header, drop. */ 660 goto out_err; 661 662 if (icmph->type == ICMP_DEST_UNREACH) { 663 switch (icmph->code & 15) { 664 case ICMP_NET_UNREACH: 665 case ICMP_HOST_UNREACH: 666 case ICMP_PROT_UNREACH: 667 case ICMP_PORT_UNREACH: 668 break; 669 case ICMP_FRAG_NEEDED: 670 if (ipv4_config.no_pmtu_disc) { 671 LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: fragmentation needed and DF set.\n", 672 &iph->daddr); 673 } else { 674 info = ip_rt_frag_needed(net, iph, 675 ntohs(icmph->un.frag.mtu), 676 skb->dev); 677 if (!info) 678 goto out; 679 } 680 break; 681 case ICMP_SR_FAILED: 682 LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: Source Route Failed.\n", 683 &iph->daddr); 684 break; 685 default: 686 break; 687 } 688 if (icmph->code > NR_ICMP_UNREACH) 689 goto out; 690 } else if (icmph->type == ICMP_PARAMETERPROB) 691 info = ntohl(icmph->un.gateway) >> 24; 692 693 /* 694 * Throw it at our lower layers 695 * 696 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed 697 * header. 698 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the 699 * transport layer. 700 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to 701 * transport layer. 702 */ 703 704 /* 705 * Check the other end isn't violating RFC 1122. Some routers send 706 * bogus responses to broadcast frames. If you see this message 707 * first check your netmask matches at both ends, if it does then 708 * get the other vendor to fix their kit. 709 */ 710 711 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses && 712 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) { 713 if (net_ratelimit()) 714 printk(KERN_WARNING "%pI4 sent an invalid ICMP " 715 "type %u, code %u " 716 "error to a broadcast: %pI4 on %s\n", 717 &ip_hdr(skb)->saddr, 718 icmph->type, icmph->code, 719 &iph->daddr, 720 skb->dev->name); 721 goto out; 722 } 723 724 /* Checkin full IP header plus 8 bytes of protocol to 725 * avoid additional coding at protocol handlers. 726 */ 727 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) 728 goto out; 729 730 iph = (const struct iphdr *)skb->data; 731 protocol = iph->protocol; 732 733 /* 734 * Deliver ICMP message to raw sockets. Pretty useless feature? 735 */ 736 raw_icmp_error(skb, protocol, info); 737 738 hash = protocol & (MAX_INET_PROTOS - 1); 739 rcu_read_lock(); 740 ipprot = rcu_dereference(inet_protos[hash]); 741 if (ipprot && ipprot->err_handler) 742 ipprot->err_handler(skb, info); 743 rcu_read_unlock(); 744 745 out: 746 return; 747 out_err: 748 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 749 goto out; 750 } 751 752 753 /* 754 * Handle ICMP_REDIRECT. 755 */ 756 757 static void icmp_redirect(struct sk_buff *skb) 758 { 759 const struct iphdr *iph; 760 761 if (skb->len < sizeof(struct iphdr)) 762 goto out_err; 763 764 /* 765 * Get the copied header of the packet that caused the redirect 766 */ 767 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 768 goto out; 769 770 iph = (const struct iphdr *)skb->data; 771 772 switch (icmp_hdr(skb)->code & 7) { 773 case ICMP_REDIR_NET: 774 case ICMP_REDIR_NETTOS: 775 /* 776 * As per RFC recommendations now handle it as a host redirect. 777 */ 778 case ICMP_REDIR_HOST: 779 case ICMP_REDIR_HOSTTOS: 780 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr, 781 icmp_hdr(skb)->un.gateway, 782 iph->saddr, skb->dev); 783 break; 784 } 785 786 /* Ping wants to see redirects. 787 * Let's pretend they are errors of sorts... */ 788 if (iph->protocol == IPPROTO_ICMP && 789 iph->ihl >= 5 && 790 pskb_may_pull(skb, (iph->ihl<<2)+8)) { 791 ping_err(skb, icmp_hdr(skb)->un.gateway); 792 } 793 794 out: 795 return; 796 out_err: 797 ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS); 798 goto out; 799 } 800 801 /* 802 * Handle ICMP_ECHO ("ping") requests. 803 * 804 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo 805 * requests. 806 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be 807 * included in the reply. 808 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring 809 * echo requests, MUST have default=NOT. 810 * See also WRT handling of options once they are done and working. 811 */ 812 813 static void icmp_echo(struct sk_buff *skb) 814 { 815 struct net *net; 816 817 net = dev_net(skb_dst(skb)->dev); 818 if (!net->ipv4.sysctl_icmp_echo_ignore_all) { 819 struct icmp_bxm icmp_param; 820 821 icmp_param.data.icmph = *icmp_hdr(skb); 822 icmp_param.data.icmph.type = ICMP_ECHOREPLY; 823 icmp_param.skb = skb; 824 icmp_param.offset = 0; 825 icmp_param.data_len = skb->len; 826 icmp_param.head_len = sizeof(struct icmphdr); 827 icmp_reply(&icmp_param, skb); 828 } 829 } 830 831 /* 832 * Handle ICMP Timestamp requests. 833 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests. 834 * SHOULD be in the kernel for minimum random latency. 835 * MUST be accurate to a few minutes. 836 * MUST be updated at least at 15Hz. 837 */ 838 static void icmp_timestamp(struct sk_buff *skb) 839 { 840 struct timespec tv; 841 struct icmp_bxm icmp_param; 842 /* 843 * Too short. 844 */ 845 if (skb->len < 4) 846 goto out_err; 847 848 /* 849 * Fill in the current time as ms since midnight UT: 850 */ 851 getnstimeofday(&tv); 852 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC + 853 tv.tv_nsec / NSEC_PER_MSEC); 854 icmp_param.data.times[2] = icmp_param.data.times[1]; 855 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4)) 856 BUG(); 857 icmp_param.data.icmph = *icmp_hdr(skb); 858 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY; 859 icmp_param.data.icmph.code = 0; 860 icmp_param.skb = skb; 861 icmp_param.offset = 0; 862 icmp_param.data_len = 0; 863 icmp_param.head_len = sizeof(struct icmphdr) + 12; 864 icmp_reply(&icmp_param, skb); 865 out: 866 return; 867 out_err: 868 ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS); 869 goto out; 870 } 871 872 873 /* 874 * Handle ICMP_ADDRESS_MASK requests. (RFC950) 875 * 876 * RFC1122 (3.2.2.9). A host MUST only send replies to 877 * ADDRESS_MASK requests if it's been configured as an address mask 878 * agent. Receiving a request doesn't constitute implicit permission to 879 * act as one. Of course, implementing this correctly requires (SHOULD) 880 * a way to turn the functionality on and off. Another one for sysctl(), 881 * I guess. -- MS 882 * 883 * RFC1812 (4.3.3.9). A router MUST implement it. 884 * A router SHOULD have switch turning it on/off. 885 * This switch MUST be ON by default. 886 * 887 * Gratuitous replies, zero-source replies are not implemented, 888 * that complies with RFC. DO NOT implement them!!! All the idea 889 * of broadcast addrmask replies as specified in RFC950 is broken. 890 * The problem is that it is not uncommon to have several prefixes 891 * on one physical interface. Moreover, addrmask agent can even be 892 * not aware of existing another prefixes. 893 * If source is zero, addrmask agent cannot choose correct prefix. 894 * Gratuitous mask announcements suffer from the same problem. 895 * RFC1812 explains it, but still allows to use ADDRMASK, 896 * that is pretty silly. --ANK 897 * 898 * All these rules are so bizarre, that I removed kernel addrmask 899 * support at all. It is wrong, it is obsolete, nobody uses it in 900 * any case. --ANK 901 * 902 * Furthermore you can do it with a usermode address agent program 903 * anyway... 904 */ 905 906 static void icmp_address(struct sk_buff *skb) 907 { 908 #if 0 909 if (net_ratelimit()) 910 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n"); 911 #endif 912 } 913 914 /* 915 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain 916 * loudly if an inconsistency is found. 917 * called with rcu_read_lock() 918 */ 919 920 static void icmp_address_reply(struct sk_buff *skb) 921 { 922 struct rtable *rt = skb_rtable(skb); 923 struct net_device *dev = skb->dev; 924 struct in_device *in_dev; 925 struct in_ifaddr *ifa; 926 927 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC)) 928 return; 929 930 in_dev = __in_dev_get_rcu(dev); 931 if (!in_dev) 932 return; 933 934 if (in_dev->ifa_list && 935 IN_DEV_LOG_MARTIANS(in_dev) && 936 IN_DEV_FORWARD(in_dev)) { 937 __be32 _mask, *mp; 938 939 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask); 940 BUG_ON(mp == NULL); 941 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) { 942 if (*mp == ifa->ifa_mask && 943 inet_ifa_match(ip_hdr(skb)->saddr, ifa)) 944 break; 945 } 946 if (!ifa && net_ratelimit()) { 947 printk(KERN_INFO "Wrong address mask %pI4 from %s/%pI4\n", 948 mp, dev->name, &ip_hdr(skb)->saddr); 949 } 950 } 951 } 952 953 static void icmp_discard(struct sk_buff *skb) 954 { 955 } 956 957 /* 958 * Deal with incoming ICMP packets. 959 */ 960 int icmp_rcv(struct sk_buff *skb) 961 { 962 struct icmphdr *icmph; 963 struct rtable *rt = skb_rtable(skb); 964 struct net *net = dev_net(rt->dst.dev); 965 966 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 967 struct sec_path *sp = skb_sec_path(skb); 968 int nh; 969 970 if (!(sp && sp->xvec[sp->len - 1]->props.flags & 971 XFRM_STATE_ICMP)) 972 goto drop; 973 974 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr))) 975 goto drop; 976 977 nh = skb_network_offset(skb); 978 skb_set_network_header(skb, sizeof(*icmph)); 979 980 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb)) 981 goto drop; 982 983 skb_set_network_header(skb, nh); 984 } 985 986 ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS); 987 988 switch (skb->ip_summed) { 989 case CHECKSUM_COMPLETE: 990 if (!csum_fold(skb->csum)) 991 break; 992 /* fall through */ 993 case CHECKSUM_NONE: 994 skb->csum = 0; 995 if (__skb_checksum_complete(skb)) 996 goto error; 997 } 998 999 if (!pskb_pull(skb, sizeof(*icmph))) 1000 goto error; 1001 1002 icmph = icmp_hdr(skb); 1003 1004 ICMPMSGIN_INC_STATS_BH(net, icmph->type); 1005 /* 1006 * 18 is the highest 'known' ICMP type. Anything else is a mystery 1007 * 1008 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently 1009 * discarded. 1010 */ 1011 if (icmph->type > NR_ICMP_TYPES) 1012 goto error; 1013 1014 1015 /* 1016 * Parse the ICMP message 1017 */ 1018 1019 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 1020 /* 1021 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be 1022 * silently ignored (we let user decide with a sysctl). 1023 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently 1024 * discarded if to broadcast/multicast. 1025 */ 1026 if ((icmph->type == ICMP_ECHO || 1027 icmph->type == ICMP_TIMESTAMP) && 1028 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) { 1029 goto error; 1030 } 1031 if (icmph->type != ICMP_ECHO && 1032 icmph->type != ICMP_TIMESTAMP && 1033 icmph->type != ICMP_ADDRESS && 1034 icmph->type != ICMP_ADDRESSREPLY) { 1035 goto error; 1036 } 1037 } 1038 1039 icmp_pointers[icmph->type].handler(skb); 1040 1041 drop: 1042 kfree_skb(skb); 1043 return 0; 1044 error: 1045 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 1046 goto drop; 1047 } 1048 1049 /* 1050 * This table is the definition of how we handle ICMP. 1051 */ 1052 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = { 1053 [ICMP_ECHOREPLY] = { 1054 .handler = ping_rcv, 1055 }, 1056 [1] = { 1057 .handler = icmp_discard, 1058 .error = 1, 1059 }, 1060 [2] = { 1061 .handler = icmp_discard, 1062 .error = 1, 1063 }, 1064 [ICMP_DEST_UNREACH] = { 1065 .handler = icmp_unreach, 1066 .error = 1, 1067 }, 1068 [ICMP_SOURCE_QUENCH] = { 1069 .handler = icmp_unreach, 1070 .error = 1, 1071 }, 1072 [ICMP_REDIRECT] = { 1073 .handler = icmp_redirect, 1074 .error = 1, 1075 }, 1076 [6] = { 1077 .handler = icmp_discard, 1078 .error = 1, 1079 }, 1080 [7] = { 1081 .handler = icmp_discard, 1082 .error = 1, 1083 }, 1084 [ICMP_ECHO] = { 1085 .handler = icmp_echo, 1086 }, 1087 [9] = { 1088 .handler = icmp_discard, 1089 .error = 1, 1090 }, 1091 [10] = { 1092 .handler = icmp_discard, 1093 .error = 1, 1094 }, 1095 [ICMP_TIME_EXCEEDED] = { 1096 .handler = icmp_unreach, 1097 .error = 1, 1098 }, 1099 [ICMP_PARAMETERPROB] = { 1100 .handler = icmp_unreach, 1101 .error = 1, 1102 }, 1103 [ICMP_TIMESTAMP] = { 1104 .handler = icmp_timestamp, 1105 }, 1106 [ICMP_TIMESTAMPREPLY] = { 1107 .handler = icmp_discard, 1108 }, 1109 [ICMP_INFO_REQUEST] = { 1110 .handler = icmp_discard, 1111 }, 1112 [ICMP_INFO_REPLY] = { 1113 .handler = icmp_discard, 1114 }, 1115 [ICMP_ADDRESS] = { 1116 .handler = icmp_address, 1117 }, 1118 [ICMP_ADDRESSREPLY] = { 1119 .handler = icmp_address_reply, 1120 }, 1121 }; 1122 1123 static void __net_exit icmp_sk_exit(struct net *net) 1124 { 1125 int i; 1126 1127 for_each_possible_cpu(i) 1128 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]); 1129 kfree(net->ipv4.icmp_sk); 1130 net->ipv4.icmp_sk = NULL; 1131 } 1132 1133 static int __net_init icmp_sk_init(struct net *net) 1134 { 1135 int i, err; 1136 1137 net->ipv4.icmp_sk = 1138 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL); 1139 if (net->ipv4.icmp_sk == NULL) 1140 return -ENOMEM; 1141 1142 for_each_possible_cpu(i) { 1143 struct sock *sk; 1144 1145 err = inet_ctl_sock_create(&sk, PF_INET, 1146 SOCK_RAW, IPPROTO_ICMP, net); 1147 if (err < 0) 1148 goto fail; 1149 1150 net->ipv4.icmp_sk[i] = sk; 1151 1152 /* Enough space for 2 64K ICMP packets, including 1153 * sk_buff struct overhead. 1154 */ 1155 sk->sk_sndbuf = 1156 (2 * ((64 * 1024) + sizeof(struct sk_buff))); 1157 1158 /* 1159 * Speedup sock_wfree() 1160 */ 1161 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); 1162 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT; 1163 } 1164 1165 /* Control parameters for ECHO replies. */ 1166 net->ipv4.sysctl_icmp_echo_ignore_all = 0; 1167 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1; 1168 1169 /* Control parameter - ignore bogus broadcast responses? */ 1170 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1; 1171 1172 /* 1173 * Configurable global rate limit. 1174 * 1175 * ratelimit defines tokens/packet consumed for dst->rate_token 1176 * bucket ratemask defines which icmp types are ratelimited by 1177 * setting it's bit position. 1178 * 1179 * default: 1180 * dest unreachable (3), source quench (4), 1181 * time exceeded (11), parameter problem (12) 1182 */ 1183 1184 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ; 1185 net->ipv4.sysctl_icmp_ratemask = 0x1818; 1186 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0; 1187 1188 return 0; 1189 1190 fail: 1191 for_each_possible_cpu(i) 1192 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]); 1193 kfree(net->ipv4.icmp_sk); 1194 return err; 1195 } 1196 1197 static struct pernet_operations __net_initdata icmp_sk_ops = { 1198 .init = icmp_sk_init, 1199 .exit = icmp_sk_exit, 1200 }; 1201 1202 int __init icmp_init(void) 1203 { 1204 return register_pernet_subsys(&icmp_sk_ops); 1205 } 1206