1 /* 2 * NET3: Implementation of the ICMP protocol layer. 3 * 4 * Alan Cox, <alan@redhat.com> 5 * 6 * Version: $Id: icmp.c,v 1.85 2002/02/01 22:01:03 davem Exp $ 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 * 13 * Some of the function names and the icmp unreach table for this 14 * module were derived from [icmp.c 1.0.11 06/02/93] by 15 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting. 16 * Other than that this module is a complete rewrite. 17 * 18 * Fixes: 19 * Clemens Fruhwirth : introduce global icmp rate limiting 20 * with icmp type masking ability instead 21 * of broken per type icmp timeouts. 22 * Mike Shaver : RFC1122 checks. 23 * Alan Cox : Multicast ping reply as self. 24 * Alan Cox : Fix atomicity lockup in ip_build_xmit 25 * call. 26 * Alan Cox : Added 216,128 byte paths to the MTU 27 * code. 28 * Martin Mares : RFC1812 checks. 29 * Martin Mares : Can be configured to follow redirects 30 * if acting as a router _without_ a 31 * routing protocol (RFC 1812). 32 * Martin Mares : Echo requests may be configured to 33 * be ignored (RFC 1812). 34 * Martin Mares : Limitation of ICMP error message 35 * transmit rate (RFC 1812). 36 * Martin Mares : TOS and Precedence set correctly 37 * (RFC 1812). 38 * Martin Mares : Now copying as much data from the 39 * original packet as we can without 40 * exceeding 576 bytes (RFC 1812). 41 * Willy Konynenberg : Transparent proxying support. 42 * Keith Owens : RFC1191 correction for 4.2BSD based 43 * path MTU bug. 44 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are 45 * valid (RFC 1812). 46 * Andi Kleen : Check all packet lengths properly 47 * and moved all kfree_skb() up to 48 * icmp_rcv. 49 * Andi Kleen : Move the rate limit bookkeeping 50 * into the dest entry and use a token 51 * bucket filter (thanks to ANK). Make 52 * the rates sysctl configurable. 53 * Yu Tianli : Fixed two ugly bugs in icmp_send 54 * - IP option length was accounted wrongly 55 * - ICMP header length was not accounted 56 * at all. 57 * Tristan Greaves : Added sysctl option to ignore bogus 58 * broadcast responses from broken routers. 59 * 60 * To Fix: 61 * 62 * - Should use skb_pull() instead of all the manual checking. 63 * This would also greatly simply some upper layer error handlers. --AK 64 * 65 */ 66 67 #include <linux/module.h> 68 #include <linux/types.h> 69 #include <linux/jiffies.h> 70 #include <linux/kernel.h> 71 #include <linux/fcntl.h> 72 #include <linux/socket.h> 73 #include <linux/in.h> 74 #include <linux/inet.h> 75 #include <linux/inetdevice.h> 76 #include <linux/netdevice.h> 77 #include <linux/string.h> 78 #include <linux/netfilter_ipv4.h> 79 #include <net/snmp.h> 80 #include <net/ip.h> 81 #include <net/route.h> 82 #include <net/protocol.h> 83 #include <net/icmp.h> 84 #include <net/tcp.h> 85 #include <net/udp.h> 86 #include <net/raw.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 replyopts; 113 unsigned char optbuf[40]; 114 }; 115 116 /* 117 * Statistics 118 */ 119 DEFINE_SNMP_STAT(struct icmp_mib, icmp_statistics) __read_mostly; 120 DEFINE_SNMP_STAT(struct icmpmsg_mib, icmpmsg_statistics) __read_mostly; 121 122 /* An array of errno for error messages from dest unreach. */ 123 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */ 124 125 struct icmp_err icmp_err_convert[] = { 126 { 127 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */ 128 .fatal = 0, 129 }, 130 { 131 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */ 132 .fatal = 0, 133 }, 134 { 135 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */, 136 .fatal = 1, 137 }, 138 { 139 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */ 140 .fatal = 1, 141 }, 142 { 143 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */ 144 .fatal = 0, 145 }, 146 { 147 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */ 148 .fatal = 0, 149 }, 150 { 151 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */ 152 .fatal = 1, 153 }, 154 { 155 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */ 156 .fatal = 1, 157 }, 158 { 159 .errno = ENONET, /* ICMP_HOST_ISOLATED */ 160 .fatal = 1, 161 }, 162 { 163 .errno = ENETUNREACH, /* ICMP_NET_ANO */ 164 .fatal = 1, 165 }, 166 { 167 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */ 168 .fatal = 1, 169 }, 170 { 171 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */ 172 .fatal = 0, 173 }, 174 { 175 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */ 176 .fatal = 0, 177 }, 178 { 179 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */ 180 .fatal = 1, 181 }, 182 { 183 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */ 184 .fatal = 1, 185 }, 186 { 187 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */ 188 .fatal = 1, 189 }, 190 }; 191 192 /* 193 * ICMP control array. This specifies what to do with each ICMP. 194 */ 195 196 struct icmp_control { 197 void (*handler)(struct sk_buff *skb); 198 short error; /* This ICMP is classed as an error message */ 199 }; 200 201 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1]; 202 203 /* 204 * The ICMP socket(s). This is the most convenient way to flow control 205 * our ICMP output as well as maintain a clean interface throughout 206 * all layers. All Socketless IP sends will soon be gone. 207 * 208 * On SMP we have one ICMP socket per-cpu. 209 */ 210 static struct sock *icmp_sk(struct net *net) 211 { 212 return net->ipv4.icmp_sk[smp_processor_id()]; 213 } 214 215 static inline int icmp_xmit_lock(struct sock *sk) 216 { 217 local_bh_disable(); 218 219 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) { 220 /* This can happen if the output path signals a 221 * dst_link_failure() for an outgoing ICMP packet. 222 */ 223 local_bh_enable(); 224 return 1; 225 } 226 return 0; 227 } 228 229 static inline void icmp_xmit_unlock(struct sock *sk) 230 { 231 spin_unlock_bh(&sk->sk_lock.slock); 232 } 233 234 /* 235 * Send an ICMP frame. 236 */ 237 238 /* 239 * Check transmit rate limitation for given message. 240 * The rate information is held in the destination cache now. 241 * This function is generic and could be used for other purposes 242 * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov. 243 * 244 * Note that the same dst_entry fields are modified by functions in 245 * route.c too, but these work for packet destinations while xrlim_allow 246 * works for icmp destinations. This means the rate limiting information 247 * for one "ip object" is shared - and these ICMPs are twice limited: 248 * by source and by destination. 249 * 250 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate 251 * SHOULD allow setting of rate limits 252 * 253 * Shared between ICMPv4 and ICMPv6. 254 */ 255 #define XRLIM_BURST_FACTOR 6 256 int xrlim_allow(struct dst_entry *dst, int timeout) 257 { 258 unsigned long now, token = dst->rate_tokens; 259 int rc = 0; 260 261 now = jiffies; 262 token += now - dst->rate_last; 263 dst->rate_last = now; 264 if (token > XRLIM_BURST_FACTOR * timeout) 265 token = XRLIM_BURST_FACTOR * timeout; 266 if (token >= timeout) { 267 token -= timeout; 268 rc = 1; 269 } 270 dst->rate_tokens = token; 271 return rc; 272 } 273 274 static inline int icmpv4_xrlim_allow(struct net *net, struct rtable *rt, 275 int type, int code) 276 { 277 struct dst_entry *dst = &rt->u.dst; 278 int rc = 1; 279 280 if (type > NR_ICMP_TYPES) 281 goto out; 282 283 /* Don't limit PMTU discovery. */ 284 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 285 goto out; 286 287 /* No rate limit on loopback */ 288 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK)) 289 goto out; 290 291 /* Limit if icmp type is enabled in ratemask. */ 292 if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) 293 rc = xrlim_allow(dst, net->ipv4.sysctl_icmp_ratelimit); 294 out: 295 return rc; 296 } 297 298 /* 299 * Maintain the counters used in the SNMP statistics for outgoing ICMP 300 */ 301 void icmp_out_count(unsigned char type) 302 { 303 ICMPMSGOUT_INC_STATS(type); 304 ICMP_INC_STATS(ICMP_MIB_OUTMSGS); 305 } 306 307 /* 308 * Checksum each fragment, and on the first include the headers and final 309 * checksum. 310 */ 311 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd, 312 struct sk_buff *skb) 313 { 314 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from; 315 __wsum csum; 316 317 csum = skb_copy_and_csum_bits(icmp_param->skb, 318 icmp_param->offset + offset, 319 to, len, 0); 320 321 skb->csum = csum_block_add(skb->csum, csum, odd); 322 if (icmp_pointers[icmp_param->data.icmph.type].error) 323 nf_ct_attach(skb, icmp_param->skb); 324 return 0; 325 } 326 327 static void icmp_push_reply(struct icmp_bxm *icmp_param, 328 struct ipcm_cookie *ipc, struct rtable *rt) 329 { 330 struct sock *sk; 331 struct sk_buff *skb; 332 333 sk = icmp_sk(dev_net(rt->u.dst.dev)); 334 if (ip_append_data(sk, icmp_glue_bits, icmp_param, 335 icmp_param->data_len+icmp_param->head_len, 336 icmp_param->head_len, 337 ipc, rt, MSG_DONTWAIT) < 0) 338 ip_flush_pending_frames(sk); 339 else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) { 340 struct icmphdr *icmph = icmp_hdr(skb); 341 __wsum csum = 0; 342 struct sk_buff *skb1; 343 344 skb_queue_walk(&sk->sk_write_queue, skb1) { 345 csum = csum_add(csum, skb1->csum); 346 } 347 csum = csum_partial_copy_nocheck((void *)&icmp_param->data, 348 (char *)icmph, 349 icmp_param->head_len, csum); 350 icmph->checksum = csum_fold(csum); 351 skb->ip_summed = CHECKSUM_NONE; 352 ip_push_pending_frames(sk); 353 } 354 } 355 356 /* 357 * Driving logic for building and sending ICMP messages. 358 */ 359 360 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb) 361 { 362 struct ipcm_cookie ipc; 363 struct rtable *rt = skb->rtable; 364 struct net *net = dev_net(rt->u.dst.dev); 365 struct sock *sk = icmp_sk(net); 366 struct inet_sock *inet = inet_sk(sk); 367 __be32 daddr; 368 369 if (ip_options_echo(&icmp_param->replyopts, skb)) 370 return; 371 372 if (icmp_xmit_lock(sk)) 373 return; 374 375 icmp_param->data.icmph.checksum = 0; 376 377 inet->tos = ip_hdr(skb)->tos; 378 daddr = ipc.addr = rt->rt_src; 379 ipc.opt = NULL; 380 if (icmp_param->replyopts.optlen) { 381 ipc.opt = &icmp_param->replyopts; 382 if (ipc.opt->srr) 383 daddr = icmp_param->replyopts.faddr; 384 } 385 { 386 struct flowi fl = { .nl_u = { .ip4_u = 387 { .daddr = daddr, 388 .saddr = rt->rt_spec_dst, 389 .tos = RT_TOS(ip_hdr(skb)->tos) } }, 390 .proto = IPPROTO_ICMP }; 391 security_skb_classify_flow(skb, &fl); 392 if (ip_route_output_key(net, &rt, &fl)) 393 goto out_unlock; 394 } 395 if (icmpv4_xrlim_allow(net, rt, icmp_param->data.icmph.type, 396 icmp_param->data.icmph.code)) 397 icmp_push_reply(icmp_param, &ipc, rt); 398 ip_rt_put(rt); 399 out_unlock: 400 icmp_xmit_unlock(sk); 401 } 402 403 404 /* 405 * Send an ICMP message in response to a situation 406 * 407 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header. 408 * MAY send more (we do). 409 * MUST NOT change this header information. 410 * MUST NOT reply to a multicast/broadcast IP address. 411 * MUST NOT reply to a multicast/broadcast MAC address. 412 * MUST reply to only the first fragment. 413 */ 414 415 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info) 416 { 417 struct iphdr *iph; 418 int room; 419 struct icmp_bxm icmp_param; 420 struct rtable *rt = skb_in->rtable; 421 struct ipcm_cookie ipc; 422 __be32 saddr; 423 u8 tos; 424 struct net *net; 425 struct sock *sk; 426 427 if (!rt) 428 goto out; 429 net = dev_net(rt->u.dst.dev); 430 sk = icmp_sk(net); 431 432 /* 433 * Find the original header. It is expected to be valid, of course. 434 * Check this, icmp_send is called from the most obscure devices 435 * sometimes. 436 */ 437 iph = ip_hdr(skb_in); 438 439 if ((u8 *)iph < skb_in->head || 440 (skb_in->network_header + sizeof(*iph)) > skb_in->tail) 441 goto out; 442 443 /* 444 * No replies to physical multicast/broadcast 445 */ 446 if (skb_in->pkt_type != PACKET_HOST) 447 goto out; 448 449 /* 450 * Now check at the protocol level 451 */ 452 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 453 goto out; 454 455 /* 456 * Only reply to fragment 0. We byte re-order the constant 457 * mask for efficiency. 458 */ 459 if (iph->frag_off & htons(IP_OFFSET)) 460 goto out; 461 462 /* 463 * If we send an ICMP error to an ICMP error a mess would result.. 464 */ 465 if (icmp_pointers[type].error) { 466 /* 467 * We are an error, check if we are replying to an 468 * ICMP error 469 */ 470 if (iph->protocol == IPPROTO_ICMP) { 471 u8 _inner_type, *itp; 472 473 itp = skb_header_pointer(skb_in, 474 skb_network_header(skb_in) + 475 (iph->ihl << 2) + 476 offsetof(struct icmphdr, 477 type) - 478 skb_in->data, 479 sizeof(_inner_type), 480 &_inner_type); 481 if (itp == NULL) 482 goto out; 483 484 /* 485 * Assume any unknown ICMP type is an error. This 486 * isn't specified by the RFC, but think about it.. 487 */ 488 if (*itp > NR_ICMP_TYPES || 489 icmp_pointers[*itp].error) 490 goto out; 491 } 492 } 493 494 if (icmp_xmit_lock(sk)) 495 return; 496 497 /* 498 * Construct source address and options. 499 */ 500 501 saddr = iph->daddr; 502 if (!(rt->rt_flags & RTCF_LOCAL)) { 503 struct net_device *dev = NULL; 504 505 if (rt->fl.iif && 506 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr) 507 dev = dev_get_by_index(net, rt->fl.iif); 508 509 if (dev) { 510 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK); 511 dev_put(dev); 512 } else 513 saddr = 0; 514 } 515 516 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) | 517 IPTOS_PREC_INTERNETCONTROL) : 518 iph->tos; 519 520 if (ip_options_echo(&icmp_param.replyopts, skb_in)) 521 goto out_unlock; 522 523 524 /* 525 * Prepare data for ICMP header. 526 */ 527 528 icmp_param.data.icmph.type = type; 529 icmp_param.data.icmph.code = code; 530 icmp_param.data.icmph.un.gateway = info; 531 icmp_param.data.icmph.checksum = 0; 532 icmp_param.skb = skb_in; 533 icmp_param.offset = skb_network_offset(skb_in); 534 inet_sk(sk)->tos = tos; 535 ipc.addr = iph->saddr; 536 ipc.opt = &icmp_param.replyopts; 537 538 { 539 struct flowi fl = { 540 .nl_u = { 541 .ip4_u = { 542 .daddr = icmp_param.replyopts.srr ? 543 icmp_param.replyopts.faddr : 544 iph->saddr, 545 .saddr = saddr, 546 .tos = RT_TOS(tos) 547 } 548 }, 549 .proto = IPPROTO_ICMP, 550 .uli_u = { 551 .icmpt = { 552 .type = type, 553 .code = code 554 } 555 } 556 }; 557 int err; 558 struct rtable *rt2; 559 560 security_skb_classify_flow(skb_in, &fl); 561 if (__ip_route_output_key(net, &rt, &fl)) 562 goto out_unlock; 563 564 /* No need to clone since we're just using its address. */ 565 rt2 = rt; 566 567 err = xfrm_lookup((struct dst_entry **)&rt, &fl, NULL, 0); 568 switch (err) { 569 case 0: 570 if (rt != rt2) 571 goto route_done; 572 break; 573 case -EPERM: 574 rt = NULL; 575 break; 576 default: 577 goto out_unlock; 578 } 579 580 if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET)) 581 goto relookup_failed; 582 583 if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL) 584 err = __ip_route_output_key(net, &rt2, &fl); 585 else { 586 struct flowi fl2 = {}; 587 struct dst_entry *odst; 588 589 fl2.fl4_dst = fl.fl4_src; 590 if (ip_route_output_key(net, &rt2, &fl2)) 591 goto relookup_failed; 592 593 /* Ugh! */ 594 odst = skb_in->dst; 595 err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src, 596 RT_TOS(tos), rt2->u.dst.dev); 597 598 dst_release(&rt2->u.dst); 599 rt2 = skb_in->rtable; 600 skb_in->dst = odst; 601 } 602 603 if (err) 604 goto relookup_failed; 605 606 err = xfrm_lookup((struct dst_entry **)&rt2, &fl, NULL, 607 XFRM_LOOKUP_ICMP); 608 switch (err) { 609 case 0: 610 dst_release(&rt->u.dst); 611 rt = rt2; 612 break; 613 case -EPERM: 614 goto ende; 615 default: 616 relookup_failed: 617 if (!rt) 618 goto out_unlock; 619 break; 620 } 621 } 622 623 route_done: 624 if (!icmpv4_xrlim_allow(net, rt, type, code)) 625 goto ende; 626 627 /* RFC says return as much as we can without exceeding 576 bytes. */ 628 629 room = dst_mtu(&rt->u.dst); 630 if (room > 576) 631 room = 576; 632 room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen; 633 room -= sizeof(struct icmphdr); 634 635 icmp_param.data_len = skb_in->len - icmp_param.offset; 636 if (icmp_param.data_len > room) 637 icmp_param.data_len = room; 638 icmp_param.head_len = sizeof(struct icmphdr); 639 640 icmp_push_reply(&icmp_param, &ipc, rt); 641 ende: 642 ip_rt_put(rt); 643 out_unlock: 644 icmp_xmit_unlock(sk); 645 out:; 646 } 647 648 649 /* 650 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH. 651 */ 652 653 static void icmp_unreach(struct sk_buff *skb) 654 { 655 struct iphdr *iph; 656 struct icmphdr *icmph; 657 int hash, protocol; 658 struct net_protocol *ipprot; 659 u32 info = 0; 660 struct net *net; 661 662 net = dev_net(skb->dst->dev); 663 664 /* 665 * Incomplete header ? 666 * Only checks for the IP header, there should be an 667 * additional check for longer headers in upper levels. 668 */ 669 670 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 671 goto out_err; 672 673 icmph = icmp_hdr(skb); 674 iph = (struct iphdr *)skb->data; 675 676 if (iph->ihl < 5) /* Mangled header, drop. */ 677 goto out_err; 678 679 if (icmph->type == ICMP_DEST_UNREACH) { 680 switch (icmph->code & 15) { 681 case ICMP_NET_UNREACH: 682 case ICMP_HOST_UNREACH: 683 case ICMP_PROT_UNREACH: 684 case ICMP_PORT_UNREACH: 685 break; 686 case ICMP_FRAG_NEEDED: 687 if (ipv4_config.no_pmtu_disc) { 688 LIMIT_NETDEBUG(KERN_INFO "ICMP: " NIPQUAD_FMT ": " 689 "fragmentation needed " 690 "and DF set.\n", 691 NIPQUAD(iph->daddr)); 692 } else { 693 info = ip_rt_frag_needed(net, iph, 694 ntohs(icmph->un.frag.mtu)); 695 if (!info) 696 goto out; 697 } 698 break; 699 case ICMP_SR_FAILED: 700 LIMIT_NETDEBUG(KERN_INFO "ICMP: " NIPQUAD_FMT ": Source " 701 "Route Failed.\n", 702 NIPQUAD(iph->daddr)); 703 break; 704 default: 705 break; 706 } 707 if (icmph->code > NR_ICMP_UNREACH) 708 goto out; 709 } else if (icmph->type == ICMP_PARAMETERPROB) 710 info = ntohl(icmph->un.gateway) >> 24; 711 712 /* 713 * Throw it at our lower layers 714 * 715 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed 716 * header. 717 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the 718 * transport layer. 719 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to 720 * transport layer. 721 */ 722 723 /* 724 * Check the other end isnt violating RFC 1122. Some routers send 725 * bogus responses to broadcast frames. If you see this message 726 * first check your netmask matches at both ends, if it does then 727 * get the other vendor to fix their kit. 728 */ 729 730 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses && 731 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) { 732 if (net_ratelimit()) 733 printk(KERN_WARNING NIPQUAD_FMT " sent an invalid ICMP " 734 "type %u, code %u " 735 "error to a broadcast: " NIPQUAD_FMT " on %s\n", 736 NIPQUAD(ip_hdr(skb)->saddr), 737 icmph->type, icmph->code, 738 NIPQUAD(iph->daddr), 739 skb->dev->name); 740 goto out; 741 } 742 743 /* Checkin full IP header plus 8 bytes of protocol to 744 * avoid additional coding at protocol handlers. 745 */ 746 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) 747 goto out; 748 749 iph = (struct iphdr *)skb->data; 750 protocol = iph->protocol; 751 752 /* 753 * Deliver ICMP message to raw sockets. Pretty useless feature? 754 */ 755 raw_icmp_error(skb, protocol, info); 756 757 hash = protocol & (MAX_INET_PROTOS - 1); 758 rcu_read_lock(); 759 ipprot = rcu_dereference(inet_protos[hash]); 760 if (ipprot && ipprot->err_handler) 761 ipprot->err_handler(skb, info); 762 rcu_read_unlock(); 763 764 out: 765 return; 766 out_err: 767 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 768 goto out; 769 } 770 771 772 /* 773 * Handle ICMP_REDIRECT. 774 */ 775 776 static void icmp_redirect(struct sk_buff *skb) 777 { 778 struct iphdr *iph; 779 780 if (skb->len < sizeof(struct iphdr)) 781 goto out_err; 782 783 /* 784 * Get the copied header of the packet that caused the redirect 785 */ 786 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 787 goto out; 788 789 iph = (struct iphdr *)skb->data; 790 791 switch (icmp_hdr(skb)->code & 7) { 792 case ICMP_REDIR_NET: 793 case ICMP_REDIR_NETTOS: 794 /* 795 * As per RFC recommendations now handle it as a host redirect. 796 */ 797 case ICMP_REDIR_HOST: 798 case ICMP_REDIR_HOSTTOS: 799 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr, 800 icmp_hdr(skb)->un.gateway, 801 iph->saddr, skb->dev); 802 break; 803 } 804 out: 805 return; 806 out_err: 807 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 808 goto out; 809 } 810 811 /* 812 * Handle ICMP_ECHO ("ping") requests. 813 * 814 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo 815 * requests. 816 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be 817 * included in the reply. 818 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring 819 * echo requests, MUST have default=NOT. 820 * See also WRT handling of options once they are done and working. 821 */ 822 823 static void icmp_echo(struct sk_buff *skb) 824 { 825 struct net *net; 826 827 net = dev_net(skb->dst->dev); 828 if (!net->ipv4.sysctl_icmp_echo_ignore_all) { 829 struct icmp_bxm icmp_param; 830 831 icmp_param.data.icmph = *icmp_hdr(skb); 832 icmp_param.data.icmph.type = ICMP_ECHOREPLY; 833 icmp_param.skb = skb; 834 icmp_param.offset = 0; 835 icmp_param.data_len = skb->len; 836 icmp_param.head_len = sizeof(struct icmphdr); 837 icmp_reply(&icmp_param, skb); 838 } 839 } 840 841 /* 842 * Handle ICMP Timestamp requests. 843 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests. 844 * SHOULD be in the kernel for minimum random latency. 845 * MUST be accurate to a few minutes. 846 * MUST be updated at least at 15Hz. 847 */ 848 static void icmp_timestamp(struct sk_buff *skb) 849 { 850 struct timeval tv; 851 struct icmp_bxm icmp_param; 852 /* 853 * Too short. 854 */ 855 if (skb->len < 4) 856 goto out_err; 857 858 /* 859 * Fill in the current time as ms since midnight UT: 860 */ 861 do_gettimeofday(&tv); 862 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 + 863 tv.tv_usec / 1000); 864 icmp_param.data.times[2] = icmp_param.data.times[1]; 865 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4)) 866 BUG(); 867 icmp_param.data.icmph = *icmp_hdr(skb); 868 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY; 869 icmp_param.data.icmph.code = 0; 870 icmp_param.skb = skb; 871 icmp_param.offset = 0; 872 icmp_param.data_len = 0; 873 icmp_param.head_len = sizeof(struct icmphdr) + 12; 874 icmp_reply(&icmp_param, skb); 875 out: 876 return; 877 out_err: 878 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 879 goto out; 880 } 881 882 883 /* 884 * Handle ICMP_ADDRESS_MASK requests. (RFC950) 885 * 886 * RFC1122 (3.2.2.9). A host MUST only send replies to 887 * ADDRESS_MASK requests if it's been configured as an address mask 888 * agent. Receiving a request doesn't constitute implicit permission to 889 * act as one. Of course, implementing this correctly requires (SHOULD) 890 * a way to turn the functionality on and off. Another one for sysctl(), 891 * I guess. -- MS 892 * 893 * RFC1812 (4.3.3.9). A router MUST implement it. 894 * A router SHOULD have switch turning it on/off. 895 * This switch MUST be ON by default. 896 * 897 * Gratuitous replies, zero-source replies are not implemented, 898 * that complies with RFC. DO NOT implement them!!! All the idea 899 * of broadcast addrmask replies as specified in RFC950 is broken. 900 * The problem is that it is not uncommon to have several prefixes 901 * on one physical interface. Moreover, addrmask agent can even be 902 * not aware of existing another prefixes. 903 * If source is zero, addrmask agent cannot choose correct prefix. 904 * Gratuitous mask announcements suffer from the same problem. 905 * RFC1812 explains it, but still allows to use ADDRMASK, 906 * that is pretty silly. --ANK 907 * 908 * All these rules are so bizarre, that I removed kernel addrmask 909 * support at all. It is wrong, it is obsolete, nobody uses it in 910 * any case. --ANK 911 * 912 * Furthermore you can do it with a usermode address agent program 913 * anyway... 914 */ 915 916 static void icmp_address(struct sk_buff *skb) 917 { 918 #if 0 919 if (net_ratelimit()) 920 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n"); 921 #endif 922 } 923 924 /* 925 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain 926 * loudly if an inconsistency is found. 927 */ 928 929 static void icmp_address_reply(struct sk_buff *skb) 930 { 931 struct rtable *rt = skb->rtable; 932 struct net_device *dev = skb->dev; 933 struct in_device *in_dev; 934 struct in_ifaddr *ifa; 935 936 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC)) 937 goto out; 938 939 in_dev = in_dev_get(dev); 940 if (!in_dev) 941 goto out; 942 rcu_read_lock(); 943 if (in_dev->ifa_list && 944 IN_DEV_LOG_MARTIANS(in_dev) && 945 IN_DEV_FORWARD(in_dev)) { 946 __be32 _mask, *mp; 947 948 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask); 949 BUG_ON(mp == NULL); 950 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) { 951 if (*mp == ifa->ifa_mask && 952 inet_ifa_match(rt->rt_src, ifa)) 953 break; 954 } 955 if (!ifa && net_ratelimit()) { 956 printk(KERN_INFO "Wrong address mask " NIPQUAD_FMT " from " 957 "%s/" NIPQUAD_FMT "\n", 958 NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src)); 959 } 960 } 961 rcu_read_unlock(); 962 in_dev_put(in_dev); 963 out:; 964 } 965 966 static void icmp_discard(struct sk_buff *skb) 967 { 968 } 969 970 /* 971 * Deal with incoming ICMP packets. 972 */ 973 int icmp_rcv(struct sk_buff *skb) 974 { 975 struct icmphdr *icmph; 976 struct rtable *rt = skb->rtable; 977 978 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 979 int nh; 980 981 if (!(skb->sp && skb->sp->xvec[skb->sp->len - 1]->props.flags & 982 XFRM_STATE_ICMP)) 983 goto drop; 984 985 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr))) 986 goto drop; 987 988 nh = skb_network_offset(skb); 989 skb_set_network_header(skb, sizeof(*icmph)); 990 991 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb)) 992 goto drop; 993 994 skb_set_network_header(skb, nh); 995 } 996 997 ICMP_INC_STATS_BH(ICMP_MIB_INMSGS); 998 999 switch (skb->ip_summed) { 1000 case CHECKSUM_COMPLETE: 1001 if (!csum_fold(skb->csum)) 1002 break; 1003 /* fall through */ 1004 case CHECKSUM_NONE: 1005 skb->csum = 0; 1006 if (__skb_checksum_complete(skb)) 1007 goto error; 1008 } 1009 1010 if (!pskb_pull(skb, sizeof(*icmph))) 1011 goto error; 1012 1013 icmph = icmp_hdr(skb); 1014 1015 ICMPMSGIN_INC_STATS_BH(icmph->type); 1016 /* 1017 * 18 is the highest 'known' ICMP type. Anything else is a mystery 1018 * 1019 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently 1020 * discarded. 1021 */ 1022 if (icmph->type > NR_ICMP_TYPES) 1023 goto error; 1024 1025 1026 /* 1027 * Parse the ICMP message 1028 */ 1029 1030 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 1031 struct net *net; 1032 1033 net = dev_net(rt->u.dst.dev); 1034 /* 1035 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be 1036 * silently ignored (we let user decide with a sysctl). 1037 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently 1038 * discarded if to broadcast/multicast. 1039 */ 1040 if ((icmph->type == ICMP_ECHO || 1041 icmph->type == ICMP_TIMESTAMP) && 1042 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) { 1043 goto error; 1044 } 1045 if (icmph->type != ICMP_ECHO && 1046 icmph->type != ICMP_TIMESTAMP && 1047 icmph->type != ICMP_ADDRESS && 1048 icmph->type != ICMP_ADDRESSREPLY) { 1049 goto error; 1050 } 1051 } 1052 1053 icmp_pointers[icmph->type].handler(skb); 1054 1055 drop: 1056 kfree_skb(skb); 1057 return 0; 1058 error: 1059 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 1060 goto drop; 1061 } 1062 1063 /* 1064 * This table is the definition of how we handle ICMP. 1065 */ 1066 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = { 1067 [ICMP_ECHOREPLY] = { 1068 .handler = icmp_discard, 1069 }, 1070 [1] = { 1071 .handler = icmp_discard, 1072 .error = 1, 1073 }, 1074 [2] = { 1075 .handler = icmp_discard, 1076 .error = 1, 1077 }, 1078 [ICMP_DEST_UNREACH] = { 1079 .handler = icmp_unreach, 1080 .error = 1, 1081 }, 1082 [ICMP_SOURCE_QUENCH] = { 1083 .handler = icmp_unreach, 1084 .error = 1, 1085 }, 1086 [ICMP_REDIRECT] = { 1087 .handler = icmp_redirect, 1088 .error = 1, 1089 }, 1090 [6] = { 1091 .handler = icmp_discard, 1092 .error = 1, 1093 }, 1094 [7] = { 1095 .handler = icmp_discard, 1096 .error = 1, 1097 }, 1098 [ICMP_ECHO] = { 1099 .handler = icmp_echo, 1100 }, 1101 [9] = { 1102 .handler = icmp_discard, 1103 .error = 1, 1104 }, 1105 [10] = { 1106 .handler = icmp_discard, 1107 .error = 1, 1108 }, 1109 [ICMP_TIME_EXCEEDED] = { 1110 .handler = icmp_unreach, 1111 .error = 1, 1112 }, 1113 [ICMP_PARAMETERPROB] = { 1114 .handler = icmp_unreach, 1115 .error = 1, 1116 }, 1117 [ICMP_TIMESTAMP] = { 1118 .handler = icmp_timestamp, 1119 }, 1120 [ICMP_TIMESTAMPREPLY] = { 1121 .handler = icmp_discard, 1122 }, 1123 [ICMP_INFO_REQUEST] = { 1124 .handler = icmp_discard, 1125 }, 1126 [ICMP_INFO_REPLY] = { 1127 .handler = icmp_discard, 1128 }, 1129 [ICMP_ADDRESS] = { 1130 .handler = icmp_address, 1131 }, 1132 [ICMP_ADDRESSREPLY] = { 1133 .handler = icmp_address_reply, 1134 }, 1135 }; 1136 1137 static void __net_exit icmp_sk_exit(struct net *net) 1138 { 1139 int i; 1140 1141 for_each_possible_cpu(i) 1142 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]); 1143 kfree(net->ipv4.icmp_sk); 1144 net->ipv4.icmp_sk = NULL; 1145 } 1146 1147 int __net_init icmp_sk_init(struct net *net) 1148 { 1149 int i, err; 1150 1151 net->ipv4.icmp_sk = 1152 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL); 1153 if (net->ipv4.icmp_sk == NULL) 1154 return -ENOMEM; 1155 1156 for_each_possible_cpu(i) { 1157 struct sock *sk; 1158 1159 err = inet_ctl_sock_create(&sk, PF_INET, 1160 SOCK_RAW, IPPROTO_ICMP, net); 1161 if (err < 0) 1162 goto fail; 1163 1164 net->ipv4.icmp_sk[i] = sk; 1165 1166 /* Enough space for 2 64K ICMP packets, including 1167 * sk_buff struct overhead. 1168 */ 1169 sk->sk_sndbuf = 1170 (2 * ((64 * 1024) + sizeof(struct sk_buff))); 1171 1172 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT; 1173 } 1174 1175 /* Control parameters for ECHO replies. */ 1176 net->ipv4.sysctl_icmp_echo_ignore_all = 0; 1177 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1; 1178 1179 /* Control parameter - ignore bogus broadcast responses? */ 1180 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1; 1181 1182 /* 1183 * Configurable global rate limit. 1184 * 1185 * ratelimit defines tokens/packet consumed for dst->rate_token 1186 * bucket ratemask defines which icmp types are ratelimited by 1187 * setting it's bit position. 1188 * 1189 * default: 1190 * dest unreachable (3), source quench (4), 1191 * time exceeded (11), parameter problem (12) 1192 */ 1193 1194 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ; 1195 net->ipv4.sysctl_icmp_ratemask = 0x1818; 1196 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0; 1197 1198 return 0; 1199 1200 fail: 1201 for_each_possible_cpu(i) 1202 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]); 1203 kfree(net->ipv4.icmp_sk); 1204 return err; 1205 } 1206 1207 static struct pernet_operations __net_initdata icmp_sk_ops = { 1208 .init = icmp_sk_init, 1209 .exit = icmp_sk_exit, 1210 }; 1211 1212 int __init icmp_init(void) 1213 { 1214 return register_pernet_device(&icmp_sk_ops); 1215 } 1216 1217 EXPORT_SYMBOL(icmp_err_convert); 1218 EXPORT_SYMBOL(icmp_send); 1219 EXPORT_SYMBOL(icmp_statistics); 1220 EXPORT_SYMBOL(xrlim_allow); 1221